{ "models": [ { "id": "o3-mini-high", "name": "o3-mini (high)", "provider": "OpenAI", "release_date": "2025-01-31", "params_total_M": null, "params_active_M": null, "architecture": null, "is_reasoning": true, "open_weights": false, "canonical_setting": { "mode": "thinking", "effort": "high", "tools": "none (per-bench override)", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default", "context": "default", "notes": "Per OpenAI GPT-4.1 blog table." } }, { "id": "gpt-4.5", "name": "GPT-4.5", "provider": "OpenAI", "release_date": "2025-02-27", "params_total_M": null, "params_active_M": null, "architecture": null, "is_reasoning": false, "open_weights": false, "canonical_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none (per-bench override)", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default", "context": "default", "notes": "Per OpenAI GPT-4.1 blog table." } }, { "id": "gpt-4.1", "name": "GPT-4.1", "provider": "OpenAI", "release_date": "2025-04-14", "params_total_M": null, "params_active_M": null, "architecture": null, "is_reasoning": false, "open_weights": false, "canonical_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none (per-bench override)", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default", "context": "default", "notes": "Per OpenAI GPT-5 dev blog. non-reasoning model." } }, { "id": "gpt-4.1-mini", "name": "GPT-4.1 mini", "provider": "OpenAI", "release_date": "2025-04-14", "params_total_M": null, "params_active_M": null, "architecture": null, "is_reasoning": false, "open_weights": false, "canonical_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none (per-bench override)", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default", "context": "default", "notes": "Per OpenAI GPT-5 dev blog. non-reasoning model." } }, { "id": "gpt-4.1-nano", "name": "GPT-4.1 nano", "provider": "OpenAI", "release_date": "2025-04-14", "params_total_M": null, "params_active_M": null, "architecture": null, "is_reasoning": false, "open_weights": false, "canonical_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none (per-bench override)", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default", "context": "default", "notes": "Per OpenAI GPT-5 dev blog. non-reasoning model." } }, { "id": "o3-high", "name": "o3 (high)", "provider": "OpenAI", "release_date": "2025-04-16", "params_total_M": null, "params_active_M": null, "architecture": null, "is_reasoning": true, "open_weights": false, "canonical_setting": { "mode": "thinking", "effort": "high", "tools": "none (per-bench override)", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default", "context": "default", "notes": "Per OpenAI GPT-5 dev blog. high reasoning effort." } }, { "id": "o4-mini-high", "name": "o4-mini (high)", "provider": "OpenAI", "release_date": "2025-04-16", "params_total_M": null, "params_active_M": null, "architecture": null, "is_reasoning": true, "open_weights": false, "canonical_setting": { "mode": "thinking", "effort": "high", "tools": "none (per-bench override)", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default", "context": "default", "notes": "Per OpenAI GPT-5 dev blog. high reasoning effort." } }, { "id": "gpt-5", "name": "GPT-5", "provider": "OpenAI", "release_date": "2025-08-01", "params_total_M": null, "params_active_M": null, "architecture": null, "is_reasoning": true, "open_weights": false, "canonical_setting": { "mode": "thinking", "effort": "high", "tools": "none (per-bench override)", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default", "context": "default", "notes": "Per OpenAI GPT-5 dev blog. high reasoning effort." } }, { "id": "gpt-oss-120b", "name": "gpt-oss-120B", "provider": "OpenAI", "release_date": "2025-08-05", "params_total_M": 116800, "params_active_M": 5100, "architecture": "MoE", "is_reasoning": true, "open_weights": true, "canonical_setting": { "mode": "thinking", "effort": "high", "tools": "with tools (per-bench override; harmony chat format default)", "sampling": "pass@1", "judge": "rule-based", "harness": "official (harmony chat format)", "prompt_style": "default", "temperature": "default", "context": "default (128K)", "notes": "Per OpenAI gpt-oss model card (arxiv:2508.10925) Table 3, high reasoning effort. Default with-tools for agentic; AIME/GPQA also has no-tools variant in same table." } }, { "id": "gpt-oss-20b", "name": "gpt-oss-20B", "provider": "OpenAI", "release_date": "2025-08-05", "params_total_M": 20900, "params_active_M": 3600, "architecture": "MoE", "is_reasoning": true, "open_weights": true, "canonical_setting": { "mode": "thinking", "effort": "high", "tools": "with tools (per-bench override; harmony chat format default)", "sampling": "pass@1", "judge": "rule-based", "harness": "official (harmony chat format)", "prompt_style": "default", "temperature": "default", "context": "default (128K)", "notes": "Per OpenAI gpt-oss model card (arxiv:2508.10925) Table 3, high reasoning effort. Default with-tools for agentic; AIME/GPQA also has no-tools variant in same table." } }, { "id": "gpt-5.1", "name": "GPT-5.1", "provider": "OpenAI", "release_date": "2025-11-13", "params_total_M": null, "params_active_M": null, "architecture": null, "is_reasoning": true, "open_weights": false, "canonical_setting": { "mode": "thinking", "effort": "high", "tools": "none (per-bench override)", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default (with prompt adjustment for τ²-bench)", "temperature": "default", "context": "default", "notes": "Per OpenAI GPT-5.1 dev blog: \"high\" reasoning effort (not xhigh like 5.2/5.4/5.5)." } }, { "id": "gpt-5.2", "name": "GPT-5.2", "provider": "OpenAI", "release_date": "2025-12-11", "params_total_M": null, "params_active_M": null, "architecture": null, "is_reasoning": true, "open_weights": false, "canonical_setting": { "mode": "thinking", "effort": "xhigh", "tools": "none (per-bench override)", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default (with prompt adjustment for τ²-bench)", "temperature": "default", "context": "default", "notes": "Per OpenAI gpt-5.2 blog: reasoning effort=xhigh, research environment." } }, { "id": "gpt-5.3-codex", "name": "GPT-5.3-Codex", "provider": "OpenAI", "release_date": "2026-02-05", "params_total_M": null, "params_active_M": null, "architecture": null, "is_reasoning": true, "open_weights": false, "canonical_setting": { "mode": "thinking", "effort": "xhigh", "tools": "none (per-bench override)", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default (with prompt adjustment for τ²-bench)", "temperature": "default", "context": "default", "notes": "Per OpenAI gpt-5.3-codex blog: reasoning effort=xhigh, research environment." } }, { "id": "gpt-5.4", "name": "GPT-5.4", "provider": "OpenAI", "release_date": "2026-03-01", "params_total_M": null, "params_active_M": null, "architecture": null, "is_reasoning": true, "open_weights": false, "canonical_setting": { "mode": "thinking", "effort": "xhigh", "tools": "none (per-bench override)", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default", "context": "default", "notes": "Per OpenAI gpt-5.4 blog: reasoning effort=xhigh, research environment (may differ from production ChatGPT)." } }, { "id": "gpt-5.5", "name": "GPT-5.5", "provider": "OpenAI", "release_date": "2026-04-22", "params_total_M": null, "params_active_M": null, "architecture": null, "is_reasoning": true, "open_weights": false, "canonical_setting": { "mode": "thinking", "effort": "xhigh", "tools": "none (per-bench override)", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default", "context": "default", "notes": "Per OpenAI gpt-5.5 blog: reasoning effort=xhigh, research environment (may differ from production ChatGPT)." } }, { "id": "claude-3.7-sonnet", "name": "Claude 3.7 Sonnet", "provider": "Anthropic", "release_date": "2025-02-24", "params_total_M": null, "params_active_M": null, "architecture": null, "is_reasoning": true, "open_weights": false, "canonical_setting": { "mode": "thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "zeroshot-cot", "temperature": 0.0, "context": "default", "notes": "Per Claude 4 blog cross-model table: with extended thinking for GPQA/MMMU/AIME/TAU; no extended thinking for SWE/Terminal/MMMLU." } }, { "id": "claude-sonnet-4", "name": "Claude Sonnet 4", "provider": "Anthropic", "release_date": "2025-05-22", "params_total_M": null, "params_active_M": null, "architecture": null, "is_reasoning": false, "open_weights": false, "canonical_setting": { "mode": "thinking", "effort": "default", "tools": "none (per-bench override)", "sampling": "pass@1 (single-pass)", "judge": "rule-based", "harness": "Claude Code agent framework (for agentic benchmarks)", "prompt_style": "default", "temperature": "top_p=0.95", "context": "default", "notes": "Per Anthropic Claude 4 announcement table: single-pass values; no extended thinking for SWE/Terminal/MMMLU; extended thinking up to 64K for GPQA/MMMU/AIME/TAU; tools=agentic for SWE/Terminal/TAU/OSWorld. Agentic harness=Claude Code per Claude 4 blog." } }, { "id": "claude-opus-4", "name": "Claude Opus 4", "provider": "Anthropic", "release_date": "2025-05-22", "params_total_M": null, "params_active_M": null, "architecture": null, "is_reasoning": true, "open_weights": false, "canonical_setting": { "mode": "thinking", "effort": "default", "tools": "none (per-bench override)", "sampling": "pass@1 (single-pass)", "judge": "rule-based", "harness": "Claude Code agent framework (for agentic benchmarks)", "prompt_style": "default", "temperature": "top_p=0.95", "context": "default", "notes": "Per Anthropic Claude 4 announcement table: single-pass values; no extended thinking for SWE/Terminal/MMMLU; extended thinking up to 64K for GPQA/MMMU/AIME/TAU; tools=agentic for SWE/Terminal/TAU/OSWorld. Agentic harness=Claude Code per Claude 4 blog." } }, { "id": "claude-opus-4.1", "name": "Claude Opus 4.1", "provider": "Anthropic", "release_date": "2025-08-05", "params_total_M": null, "params_active_M": null, "architecture": null, "is_reasoning": true, "open_weights": false, "canonical_setting": { "mode": "thinking", "effort": "default", "tools": "none (per-bench override)", "sampling": "pass@1 (single-pass)", "judge": "rule-based", "harness": "Terminus 1 (for agentic benchmarks)", "prompt_style": "default", "temperature": "top_p=0.95", "context": "default", "notes": "Per Anthropic Opus 4.1 announcement table. Single-pass values; harness=Terminus 1 for Terminal-Bench (note: differs from Claude 4 blog which used Claude Code framework)." } }, { "id": "claude-sonnet-4.5", "name": "Claude Sonnet 4.5", "provider": "Anthropic", "release_date": "2025-09-29", "params_total_M": null, "params_active_M": null, "architecture": null, "is_reasoning": true, "open_weights": false, "canonical_setting": { "mode": "thinking", "effort": "high (default)", "tools": "none (per-bench override)", "sampling": "pass@1 (avg 5 trials)", "judge": "rule-based", "harness": "Terminus-2 (for terminal-bench); official otherwise", "prompt_style": "interleaved scratchpads", "temperature": "default (top_p, temperature)", "context": "200K (default)", "notes": "Per Anthropic Opus 4.5 announcement: 64K thinking budget, interleaved scratchpads, 200K context, default effort=high, default sampling, avg 5 trials. Exceptions: SWE-bench Verified=no thinking; Terminal-Bench=128K thinking budget." } }, { "id": "claude-haiku-4.5", "name": "Claude Haiku 4.5", "provider": "Anthropic", "release_date": "2025-10-15", "params_total_M": null, "params_active_M": null, "architecture": null, "is_reasoning": true, "open_weights": false, "canonical_setting": { "mode": "thinking", "effort": "128K thinking budget", "tools": "none (per-bench override for agentic benchmarks)", "sampling": "pass@1 (avg over ≥10 runs unless noted)", "judge": "official harness", "harness": "official", "prompt_style": "default", "temperature": "default (Anthropic default)" } }, { "id": "claude-opus-4.5", "name": "Claude Opus 4.5", "provider": "Anthropic", "release_date": "2025-11-24", "params_total_M": null, "params_active_M": null, "architecture": null, "is_reasoning": true, "open_weights": false, "canonical_setting": { "mode": "thinking", "effort": "high (default)", "tools": "none (per-bench override)", "sampling": "pass@1 (avg 5 trials)", "judge": "rule-based", "harness": "Terminus-2 (for terminal-bench); official otherwise", "prompt_style": "interleaved scratchpads", "temperature": "default (top_p, temperature)", "context": "200K (default)", "notes": "Per Anthropic Opus 4.5 announcement: 64K thinking budget, interleaved scratchpads, 200K context, default effort=high, default sampling, avg 5 trials. Exceptions: SWE-bench Verified=no thinking; Terminal-Bench=128K thinking budget." } }, { "id": "claude-opus-4.6", "name": "Claude Opus 4.6", "provider": "Anthropic", "release_date": "2026-02-05", "params_total_M": null, "params_active_M": null, "architecture": null, "is_reasoning": true, "open_weights": false, "canonical_setting": { "mode": "thinking", "effort": "max", "tools": "none (per-bench override)", "sampling": "pass@1 (avg 5 trials)", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default", "context": "up to 1M", "notes": "Per Mythos System Card Table 6.3.A: standard config = adaptive thinking max effort, default sampling, avg 5 trials, context up to 1M." } }, { "id": "claude-sonnet-4.6", "name": "Claude Sonnet 4.6", "provider": "Anthropic", "release_date": "2026-02-17", "params_total_M": null, "params_active_M": null, "architecture": null, "is_reasoning": true, "open_weights": false, "canonical_setting": { "mode": "thinking", "effort": "max", "tools": "none (per-bench override)", "sampling": "pass@1 (avg 5-15 trials)", "judge": "rule-based", "harness": "Terminus-2 (terminal-bench, thinking off); official otherwise", "prompt_style": "default", "temperature": "default", "context": "default", "notes": "Per Anthropic Sonnet 4.6 announcement: max thinking effort default; Terminal-Bench=thinking off, Terminus-2; SWE-bench=avg 10 trials; BrowseComp/HLE-with-tools have specific tool configs (web search/fetch + 50k context compaction)." } }, { "id": "claude-mythos", "name": "Claude Mythos Preview", "provider": "Anthropic", "release_date": "2026-04-07", "params_total_M": null, "params_active_M": null, "architecture": null, "is_reasoning": true, "open_weights": false, "canonical_setting": { "mode": "thinking", "effort": "adaptive thinking at max effort", "tools": "none (per-bench override)", "sampling": "pass@1 (avg 5 trials)", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default", "context": "up to 1M", "notes": "Per Mythos System Card Table 6.3.A: standard config is adaptive thinking max effort, default sampling (temperature, top_p), avg 5 trials, context up to 1M." } }, { "id": "claude-opus-4.7", "name": "Claude Opus 4.7", "provider": "Anthropic", "release_date": "2026-04-22", "params_total_M": null, "params_active_M": null, "architecture": null, "is_reasoning": true, "open_weights": false, "canonical_setting": { "mode": "thinking", "effort": "default", "tools": "none (per-bench override)", "sampling": "pass@1 (avg 5 trials)", "judge": "rule-based", "harness": "Terminus-2 (for terminal-bench, thinking disabled)", "prompt_style": "default", "temperature": "default", "context": "default", "notes": "Per Anthropic Opus 4.7 announcement table. Default thinking mode with extended thinking; Terminal-Bench uses Terminus-2 harness with thinking disabled (1× guaranteed/3× ceiling, avg 5 attempts)." } }, { "id": "gemini-2.0-flash", "name": "Gemini 2.0 Flash", "provider": "Google", "release_date": "2025-02-01", "params_total_M": null, "params_active_M": null, "architecture": null, "is_reasoning": false, "open_weights": false, "canonical_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default", "context": "default", "notes": "Per Gemini 2.0 Flash Model Card (April 15 2025): non-thinking generation, default sampling. Same canonical setting applies to 1.5 / 2.0 family." } }, { "id": "gemini-2.5-pro", "name": "Gemini 2.5 Pro (GA)", "provider": "Google", "release_date": "2025-06-27", "params_total_M": null, "params_active_M": null, "architecture": null, "is_reasoning": true, "open_weights": false, "canonical_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1 (single attempt)", "judge": "rule-based", "harness": "AI Studio API", "prompt_style": "default", "temperature": "default", "context": "default", "notes": "Per Gemini 2.5 Pro Model Card (GA, June 27 2025): pass@1 single attempt, AI Studio API model-id gemini-2.5-pro-preview-06-05 / gemini-2.5-pro GA, default sampling. Release date set to GA date 2025-06-27." } }, { "id": "gemini-2.5-flash", "name": "Gemini 2.5 Flash", "provider": "Google", "release_date": "2025-05-20", "params_total_M": null, "params_active_M": null, "architecture": null, "is_reasoning": true, "open_weights": false, "canonical_setting": { "mode": "thinking", "effort": "default", "tools": "none (per-bench override)", "sampling": "pass@1 (single attempt; multi-trial avg for smaller bench)", "judge": "rule-based", "harness": "AI Studio API (gemini-2.5-flash GA)", "prompt_style": "default", "temperature": "default", "context": "default (1M)", "notes": "Per Gemini 2.5 Flash Model Card (Dec 2025) GA Thinking column." } }, { "id": "gemma-3-27b", "name": "Gemma 3 27B", "provider": "Google", "release_date": "2025-03-12", "params_total_M": 27000, "params_active_M": 27000, "architecture": "Dense", "is_reasoning": false, "open_weights": true, "canonical_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default", "context": "default (128K)", "notes": "Per Gemma 3 tech report (arxiv:2503.19786) Table 6 IT column for 27B." } }, { "id": "gemma-4-31b", "name": "Gemma 4 31B", "provider": "Google", "release_date": "2026-04-02", "params_total_M": 30700, "params_active_M": 30700, "architecture": "Dense multimodal decoder-only", "is_reasoning": true, "open_weights": true, "canonical_setting": { "mode": "thinking", "effort": "default", "tools": "none (per-benchmark override)", "sampling": "provider-reported (per-benchmark override)", "judge": "benchmark-specified", "harness": "official provider evaluation", "prompt_style": "official chat template", "temperature": "1.0; top_p=0.95; top_k=64", "context": "262,144", "notes": "Official checkpoint google/gemma-4-31B-it; 60 layers, 1,024-token sliding window, 262K vocabulary." } }, { "id": "gemma-4-26b-a4b", "name": "Gemma 4 26B A4B", "provider": "Google", "release_date": "2026-04-02", "params_total_M": 25200, "params_active_M": 3800, "architecture": "MoE multimodal decoder-only", "is_reasoning": true, "open_weights": true, "canonical_setting": { "mode": "thinking", "effort": "default", "tools": "none (per-benchmark override)", "sampling": "provider-reported (per-benchmark override)", "judge": "benchmark-specified", "harness": "official provider evaluation", "prompt_style": "official chat template", "temperature": "1.0; top_p=0.95; top_k=64", "context": "262,144", "notes": "Official checkpoint google/gemma-4-26B-A4B-it; 30 layers, 8 active of 128 routed experts plus one shared expert." } }, { "id": "gemma-4-e4b", "name": "Gemma 4 E4B", "provider": "Google", "release_date": "2026-04-02", "params_total_M": 8000, "params_active_M": 8000, "architecture": "Dense multimodal decoder-only", "is_reasoning": true, "open_weights": true, "canonical_setting": { "mode": "thinking", "effort": "default", "tools": "none (per-benchmark override)", "sampling": "provider-reported (per-benchmark override)", "judge": "benchmark-specified", "harness": "official provider evaluation", "prompt_style": "official chat template", "temperature": "1.0; top_p=0.95; top_k=64", "context": "131,072", "notes": "Official checkpoint google/gemma-4-E4B-it; 4.5B effective parameters, 8B including embeddings, 42 layers." } }, { "id": "gemma-4-e2b", "name": "Gemma 4 E2B", "provider": "Google", "release_date": "2026-04-02", "params_total_M": 5100, "params_active_M": 5100, "architecture": "Dense multimodal decoder-only", "is_reasoning": true, "open_weights": true, "canonical_setting": { "mode": "thinking", "effort": "default", "tools": "none (per-benchmark override)", "sampling": "provider-reported (per-benchmark override)", "judge": "benchmark-specified", "harness": "official provider evaluation", "prompt_style": "official chat template", "temperature": "1.0; top_p=0.95; top_k=64", "context": "131,072", "notes": "Official checkpoint google/gemma-4-E2B-it; 2.3B effective parameters, 5.1B including embeddings, 35 layers." } }, { "id": "gemini-3-pro", "name": "Gemini 3 Pro", "provider": "Google", "release_date": "2025-11-18", "params_total_M": null, "params_active_M": null, "architecture": null, "is_reasoning": true, "open_weights": false, "canonical_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default", "context": "default", "notes": "Per Google DeepMind /models/gemini/pro/ table: Gemini 3 Pro Thinking (High), no tools by default." } }, { "id": "gemini-3-flash", "name": "Gemini 3 Flash", "provider": "Google", "release_date": "2025-11-18", "params_total_M": null, "params_active_M": null, "architecture": null, "is_reasoning": true, "open_weights": false, "canonical_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default", "context": "default", "notes": "Per Google DeepMind /models/gemini/flash/ table: Gemini 3 Flash Thinking, no tools by default." } }, { "id": "gemini-3.1-pro", "name": "Gemini 3.1 Pro", "provider": "Google", "release_date": "2026-02-19", "params_total_M": null, "params_active_M": null, "architecture": null, "is_reasoning": true, "open_weights": false, "canonical_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default", "context": "default", "notes": "Per Google DeepMind /models/gemini/pro/ table: Gemini 3.1 Pro Thinking (High), no tools by default." } }, { "id": "llama-4-scout", "name": "Llama 4 Scout", "provider": "Meta", "release_date": "2025-04-05", "params_total_M": 109000, "params_active_M": 17000, "architecture": "MoE", "is_reasoning": false, "open_weights": true, "canonical_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official-meta", "prompt_style": "zeroshot", "temperature": "0.0", "context": "default", "multimodal_input": true } }, { "id": "llama-4-maverick", "name": "Llama 4 Maverick", "provider": "Meta", "release_date": "2025-04-05", "params_total_M": 402000, "params_active_M": 17000, "architecture": "MoE", "is_reasoning": false, "open_weights": true, "canonical_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official-meta", "prompt_style": "zeroshot", "temperature": "0.0", "context": "default", "multimodal_input": true } }, { "id": "llama-4-behemoth", "name": "Llama 4 Behemoth", "provider": "Meta", "release_date": "2025-04-05", "params_total_M": 2000000, "params_active_M": 288000, "architecture": "MoE", "is_reasoning": false, "open_weights": true, "canonical_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official-meta", "prompt_style": "zeroshot", "temperature": "0.0", "context": "default", "multimodal_input": true } }, { "id": "muse-spark", "name": "Muse Spark", "provider": "Meta", "release_date": "2026-04-01", "params_total_M": null, "params_active_M": null, "architecture": null, "is_reasoning": true, "open_weights": false, "canonical_setting": { "mode": "thinking", "effort": "default", "tools": "none (per-benchmark override where tools are required)", "sampling": "pass@1 (single-pass)", "judge": "rule-based or gpt-oss-120b (per benchmark)", "harness": "internal (Meta MSL)", "prompt_style": "default", "temperature": "default", "notes": "Per Meta MSL blog https://ai.meta.com/blog/introducing-muse-spark-msl/. Standard Thinking mode. Contemplating mode results flagged for review." } }, { "id": "grok-3-beta", "name": "Grok 3 Beta", "provider": "xAI", "release_date": "2025-02-19", "params_total_M": null, "params_active_M": null, "architecture": null, "is_reasoning": false, "open_weights": false, "canonical_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none (per-bench override)", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default", "context": "default", "notes": "Per xAI Grok 3 blog: canonical = non-reasoning mode (the cross-model table without 'Think' label). Reasoning ON values not used." } }, { "id": "grok-4", "name": "Grok 4", "provider": "xAI", "release_date": "2025-07-09", "params_total_M": null, "params_active_M": null, "architecture": null, "is_reasoning": true, "open_weights": false, "canonical_setting": { "mode": "thinking", "effort": "default", "tools": "none (per-bench override)", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default", "context": "default", "notes": "Per xAI Grok 4 blog: standard chart values (no Heavy, no with-Python tool unless noted)." } }, { "id": "grok-4.1", "name": "Grok 4.1", "provider": "xAI", "release_date": "2025-11-17", "params_total_M": null, "params_active_M": null, "architecture": null, "is_reasoning": true, "open_weights": false, "canonical_setting": { "mode": "thinking", "effort": "default", "tools": "none (per-bench override)", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default", "context": "default", "notes": "Per xAI Grok 4.1 blog: standard chart values. Hallucination/FActScore use with-web-search but those are not stored in BP." } }, { "id": "grok-4.20", "name": "Grok 4.20", "provider": "xAI", "release_date": "2026-03-09", "params_total_M": null, "params_active_M": null, "architecture": null, "is_reasoning": true, "open_weights": false }, { "id": "deepseek-r1", "name": "DeepSeek-R1", "provider": "DeepSeek", "release_date": "2025-01-20", "params_total_M": 671000, "params_active_M": 37000, "architecture": "MoE", "is_reasoning": true, "open_weights": true, "canonical_setting": { "mode": "thinking", "effort": "default", "tools": "none (per-bench override)", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default", "context": "default", "notes": "Per DeepSeek R1-0528 model card." } }, { "id": "deepseek-v3", "name": "DeepSeek-V3", "provider": "DeepSeek", "release_date": "2025-01-01", "params_total_M": 671000, "params_active_M": 37000, "architecture": "MoE", "is_reasoning": false, "open_weights": true }, { "id": "deepseek-v3-0324", "name": "DeepSeek-V3-0324", "provider": "DeepSeek", "release_date": "2025-03-24", "params_total_M": 671000, "params_active_M": 37000, "architecture": "MoE", "is_reasoning": false, "open_weights": true }, { "id": "deepseek-r1-0528", "name": "DeepSeek-R1-0528", "provider": "DeepSeek", "release_date": "2025-05-28", "params_total_M": 671000, "params_active_M": 37000, "architecture": "MoE", "is_reasoning": true, "open_weights": true, "canonical_setting": { "mode": "thinking", "effort": "default", "tools": "none (per-bench override)", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default", "context": "default", "notes": "Per DeepSeek R1-0528 model card." } }, { "id": "deepseek-r1-distill-qwen-32b", "name": "DeepSeek-R1-Distill-Qwen-32B", "provider": "DeepSeek", "release_date": "2025-01-20", "params_total_M": 32000, "params_active_M": 32000, "architecture": "Dense", "is_reasoning": true, "open_weights": true, "canonical_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default", "context": "default", "notes": "Per DeepSeek R1 paper Table 15 (distilled model)." } }, { "id": "deepseek-r1-distill-qwen-14b", "name": "DeepSeek-R1-Distill-Qwen-14B", "provider": "DeepSeek", "release_date": "2025-01-20", "params_total_M": 14000, "params_active_M": 14000, "architecture": "Dense", "is_reasoning": true, "open_weights": true, "canonical_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default", "context": "default", "notes": "Per DeepSeek R1 paper Table 15 (distilled model)." } }, { "id": "deepseek-r1-distill-qwen-7b", "name": "DeepSeek-R1-Distill-Qwen-7B", "provider": "DeepSeek", "release_date": "2025-01-20", "params_total_M": 7000, "params_active_M": 7000, "architecture": "Dense", "is_reasoning": true, "open_weights": true, "canonical_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default", "context": "default", "notes": "Per DeepSeek R1 paper Table 15 (distilled model)." } }, { "id": "deepseek-r1-distill-qwen-1.5b", "name": "DeepSeek-R1-Distill-Qwen-1.5B", "provider": "DeepSeek", "release_date": "2025-01-20", "params_total_M": 1500, "params_active_M": 1500, "architecture": "Dense", "is_reasoning": true, "open_weights": true, "canonical_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default", "context": "default", "notes": "Per DeepSeek R1 paper Table 15 (distilled model)." } }, { "id": "deepseek-r1-distill-llama-8b", "name": "DeepSeek-R1-Distill-Llama-8B", "provider": "DeepSeek", "release_date": "2025-01-20", "params_total_M": 8000, "params_active_M": 8000, "architecture": "Dense", "is_reasoning": true, "open_weights": true, "canonical_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default", "context": "default", "notes": "Per DeepSeek R1 paper Table 15 (distilled model)." } }, { "id": "deepseek-r1-distill-llama-70b", "name": "DeepSeek-R1-Distill-Llama-70B", "provider": "DeepSeek", "release_date": "2025-01-20", "params_total_M": 70000, "params_active_M": 70000, "architecture": "Dense", "is_reasoning": true, "open_weights": true, "canonical_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official (DeepSeek SGLang/vLLM)", "prompt_style": "default", "temperature": "default", "context": "default", "notes": "Distill of DeepSeek-R1 onto Llama-70B base. Cells in BP currently primarily sourced from Qwen3 tech report Table 13 third-party self-test (matches_canonical=false). DeepSeek own paper audit pending — canonical_setting may need refinement." } }, { "id": "deepseek-v3.2", "name": "DeepSeek-V3.2", "provider": "DeepSeek", "release_date": "2025-12-01", "params_total_M": 671000, "params_active_M": 37000, "architecture": "MoE", "is_reasoning": true, "open_weights": true, "canonical_setting": { "mode": "thinking", "effort": "default", "tools": "none (per-bench override)", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "1.0", "context": "128K", "notes": "Per DeepSeek V3.2 tech report: temperature=1.0, context=128K, thinking mode for tool-use." } }, { "id": "deepseek-v3.2-speciale", "name": "DeepSeek-V3.2-Speciale", "provider": "DeepSeek", "release_date": "2025-12-01", "params_total_M": 671000, "params_active_M": 37000, "architecture": "MoE", "is_reasoning": true, "open_weights": true, "canonical_setting": { "mode": "thinking", "effort": "max", "tools": "none (per-bench override)", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "1.0", "context": "128K", "notes": "DS V3.2-Speciale: high-compute reasoning variant of V3.2. Per V3.2 tech report Table 3." } }, { "id": "deepseek-v4-pro", "name": "DeepSeek-V4-Pro (Preview)", "provider": "DeepSeek", "release_date": "2026-04-22", "params_total_M": 1600000, "params_active_M": 49000, "architecture": "MoE", "is_reasoning": true, "open_weights": true, "canonical_setting": { "mode": "thinking", "effort": "Max", "tools": "none (per-bench override)", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default", "context": "default", "notes": "Per DeepSeek V4-Pro model card. Three reasoning modes available (Non-Think/High/Max); BP canonical = Max mode (most powerful). Other modes' values stored as candidates if needed." } }, { "id": "deepseek-v4-flash", "name": "DeepSeek-V4-Flash (Preview)", "provider": "DeepSeek", "release_date": "2026-04-22", "params_total_M": 284000, "params_active_M": 13000, "architecture": "MoE", "is_reasoning": true, "open_weights": true, "canonical_setting": { "mode": "thinking", "effort": "Max", "tools": "none (per-bench override)", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default", "context": "default", "notes": "Per DeepSeek V4-Pro model card. Three reasoning modes available (Non-Think/High/Max); BP canonical = Max mode (most powerful). Other modes' values stored as candidates if needed." } }, { "id": "qwen3-235b", "name": "Qwen3-235B-A22B", "provider": "Alibaba", "release_date": "2025-05-15", "params_total_M": 235000, "params_active_M": 22000, "architecture": "MoE", "is_reasoning": true, "open_weights": true, "canonical_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official (Qwen-Agent / sglang)", "prompt_style": "default (enable_thinking=True)", "temperature": "default", "context": "default", "notes": "Per Qwen3 tech report (arxiv:2505.09388) Table 11 Thinking column. Output length 32768 (38912 for AIME)." } }, { "id": "qwen3-32b", "name": "Qwen3-32B", "provider": "Alibaba", "release_date": "2025-05-15", "params_total_M": 32000, "params_active_M": 32000, "architecture": "Dense", "is_reasoning": true, "open_weights": true, "canonical_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official (Qwen-Agent / sglang)", "prompt_style": "default (enable_thinking=True)", "temperature": "default", "context": "default", "notes": "Per Qwen3 tech report (arxiv:2505.09388) Table 13/15/17/19 Thinking column." } }, { "id": "qwen3-4b", "name": "Qwen3-4B", "provider": "Alibaba", "release_date": "2025-05-15", "params_total_M": 4000, "params_active_M": 4000, "architecture": "Dense", "is_reasoning": true, "open_weights": true, "canonical_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official (Qwen-Agent / sglang)", "prompt_style": "default (enable_thinking=True)", "temperature": "default", "context": "default", "notes": "Per Qwen3 tech report (arxiv:2505.09388) Table 13/15/17/19 Thinking column." } }, { "id": "qwen3-0.6b", "name": "Qwen3-0.6B", "provider": "Alibaba", "release_date": "2025-05-15", "params_total_M": 600, "params_active_M": 600, "architecture": "Dense", "is_reasoning": true, "open_weights": true, "canonical_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official (Qwen-Agent / sglang)", "prompt_style": "default (enable_thinking=True)", "temperature": "default", "context": "default", "notes": "Per Qwen3 tech report (arxiv:2505.09388) Table 13/15/17/19 Thinking column." } }, { "id": "qwen3-1.7b", "name": "Qwen3-1.7B", "provider": "Alibaba", "release_date": "2025-05-15", "params_total_M": 1700, "params_active_M": 1700, "architecture": "Dense", "is_reasoning": true, "open_weights": true, "canonical_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official (Qwen-Agent / sglang)", "prompt_style": "default (enable_thinking=True)", "temperature": "default", "context": "default", "notes": "Per Qwen3 tech report (arxiv:2505.09388) Table 13/15/17/19 Thinking column." } }, { "id": "qwen3-8b", "name": "Qwen3-8B", "provider": "Alibaba", "release_date": "2025-05-15", "params_total_M": 8000, "params_active_M": 8000, "architecture": "Dense", "is_reasoning": true, "open_weights": true, "canonical_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official (Qwen-Agent / sglang)", "prompt_style": "default (enable_thinking=True)", "temperature": "default", "context": "default", "notes": "Per Qwen3 tech report (arxiv:2505.09388) Table 13/15/17/19 Thinking column." } }, { "id": "qwen3-14b", "name": "Qwen3-14B", "provider": "Alibaba", "release_date": "2025-05-15", "params_total_M": 14000, "params_active_M": 14000, "architecture": "Dense", "is_reasoning": true, "open_weights": true, "canonical_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official (Qwen-Agent / sglang)", "prompt_style": "default (enable_thinking=True)", "temperature": "default", "context": "default", "notes": "Per Qwen3 tech report (arxiv:2505.09388) Table 13/15/17/19 Thinking column." } }, { "id": "qwen3-30b-a3b", "name": "Qwen3-30B-A3B", "provider": "Alibaba", "release_date": "2025-05-15", "params_total_M": 30000, "params_active_M": 3000, "architecture": "MoE", "is_reasoning": true, "open_weights": true, "canonical_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official (Qwen-Agent / sglang)", "prompt_style": "default (enable_thinking=True)", "temperature": "default", "context": "default", "notes": "Per Qwen3 tech report (arxiv:2505.09388) Table 13/15/17/19 Thinking column." } }, { "id": "qwq-32b", "name": "QwQ-32B", "provider": "Alibaba", "release_date": "2025-03-05", "params_total_M": 32800, "params_active_M": 32800, "architecture": "Dense", "is_reasoning": true, "open_weights": true, "canonical_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default", "context": "default (32K)", "notes": "Per QwQ-32B release blog (qwenlm.github.io/blog/qwq-32b/) + Qwen3 tech report Table 13 baseline column." } }, { "id": "qwen3.5-397b", "name": "Qwen3.5-397B-A17B", "provider": "Alibaba", "release_date": "2026-02-01", "params_total_M": 397000, "params_active_M": 17000, "architecture": "MoE", "is_reasoning": true, "open_weights": true, "canonical_setting": { "mode": "thinking", "effort": "default (96k thinking budget)", "tools": "none (per-bench override)", "sampling": "pass@1", "judge": "rule-based", "harness": "official (sglang/vllm with --reasoning-parser qwen3)", "prompt_style": "default (thinking by default)", "temperature": "default", "context": "default (262K)", "notes": "Per Qwen3.5 HF model card. Vision-language MoE 397B/17B." } }, { "id": "qwen3.6-plus", "name": "Qwen3.6-Plus", "provider": "Alibaba", "release_date": "2026-03-31", "params_total_M": null, "params_active_M": null, "architecture": null, "is_reasoning": true, "open_weights": false }, { "id": "kimi-k2", "name": "Kimi K2", "provider": "Moonshot AI", "release_date": "2025-07-16", "params_total_M": null, "params_active_M": null, "architecture": "MoE", "is_reasoning": false, "open_weights": true, "canonical_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none (per-bench override)", "sampling": "pass@1 (avg of 4-64 trials per benchmark)", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default", "context": "default (256K)", "notes": "Per Kimi K2 tech report (arxiv:2507.20534) Table 3 Kimi-K2-Instruct column. Non-thinking model." } }, { "id": "kimi-k2-thinking", "name": "Kimi K2 Thinking", "provider": "Moonshot AI", "release_date": "2025-11-01", "params_total_M": null, "params_active_M": null, "architecture": "MoE", "is_reasoning": true, "open_weights": true, "canonical_setting": { "mode": "thinking", "effort": "default (96k thinking-token budget for HLE/AIME/HMMT/GPQA; 128k for IMO/LCB/OJ)", "tools": "none (per-bench override)", "sampling": "pass@1 (avg@32 for AIME/HMMT no-tools)", "judge": "rule-based (HLE uses o3-mini)", "harness": "official", "prompt_style": "default", "temperature": "default", "context": "default (256K)", "notes": "Per Kimi K2 Thinking blog (moonshotai.github.io/Kimi-K2/thinking.html). Thinking model." } }, { "id": "kimi-k2.5", "name": "Kimi K2.5", "provider": "Moonshot AI", "release_date": "2026-01-01", "params_total_M": null, "params_active_M": null, "architecture": "MoE", "is_reasoning": true, "open_weights": true, "canonical_setting": { "mode": "thinking", "effort": "default (96k for reasoning, 64k for vision)", "tools": "none (per-bench override)", "sampling": "pass@1 (avg@32 AIME/HMMT, avg@8 GPQA, avg@3 vision, avg@4 Seal/WideSearch)", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "1.0", "context": "256K", "notes": "Per Kimi K2.5 model card (HF moonshotai/Kimi-K2.5). Thinking mode, t=1.0 top_p=0.95." } }, { "id": "kimi-k2.6", "name": "Kimi K2.6", "provider": "Moonshot AI", "release_date": "2026-04-22", "params_total_M": null, "params_active_M": null, "architecture": "MoE", "is_reasoning": true, "open_weights": true, "canonical_setting": { "mode": "thinking", "effort": "max (98304 generation tokens)", "tools": "none (per-bench override)", "sampling": "pass@1 (avg@10 coding, avg@3 vision)", "judge": "rule-based", "harness": "official (in-house SWE-agent-derived for SWE-Bench)", "prompt_style": "default", "temperature": "1.0", "context": "262144 (256K)", "notes": "Per Kimi K2.6 model card (HF moonshotai/Kimi-K2.6). Thinking mode, max effort, t=1.0 top_p=1.0." } }, { "id": "glm-4.6", "name": "GLM-4.6", "provider": "Zhipu AI", "release_date": "2025-09-01", "params_total_M": null, "params_active_M": null, "architecture": null, "is_reasoning": true, "open_weights": false, "canonical_setting": { "mode": "thinking", "effort": "default", "tools": "none (per-bench override)", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default", "context": "default", "notes": "Per GLM-4.7 blog." } }, { "id": "glm-4.7", "name": "GLM-4.7", "provider": "Zhipu AI", "release_date": "2025-12-01", "params_total_M": null, "params_active_M": null, "architecture": null, "is_reasoning": true, "open_weights": false, "canonical_setting": { "mode": "thinking", "effort": "default", "tools": "none (per-bench override)", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default", "context": "default", "notes": "Per GLM-5 blog." } }, { "id": "glm-5", "name": "GLM-5", "provider": "Zhipu AI", "release_date": "2026-03-15", "params_total_M": null, "params_active_M": null, "architecture": null, "is_reasoning": true, "open_weights": true, "canonical_setting": { "mode": "thinking", "effort": "default", "tools": "none (per-bench override)", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default", "context": "default", "notes": "Per GLM-5 blog." } }, { "id": "glm-5.1", "name": "GLM-5.1", "provider": "Z.ai", "release_date": "2026-04-07", "params_total_M": null, "params_active_M": null, "architecture": null, "is_reasoning": true, "open_weights": true, "canonical_setting": { "mode": "thinking", "effort": "default", "tools": "none (per-bench override)", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default", "context": "default", "notes": "Per GLM-5.1 blog." } }, { "id": "doubao-seed-2.0-pro", "name": "Doubao Seed 2.0 Pro", "provider": "ByteDance", "release_date": "2026-02-14", "params_total_M": null, "params_active_M": null, "architecture": null, "is_reasoning": true, "open_weights": false, "canonical_setting": { "mode": "thinking", "effort": "high", "tools": "none (per-bench override)", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default", "context": "default", "notes": "Per Doubao Seed 2.0 Pro model card. High thinking config." } }, { "id": "mistral-small-3.1", "name": "Mistral Small 3.1", "provider": "Mistral", "release_date": "2025-03-01", "params_total_M": 24000, "params_active_M": 24000, "architecture": "Dense", "is_reasoning": false, "open_weights": true, "canonical_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "mistral-eval", "prompt_style": "mixed (5-shot CoT for MMLU/Pro/GPQA per HF MC; default for others)", "temperature": "0.0", "context": "default", "multimodal_input": true } }, { "id": "mistral-medium-3", "name": "Mistral Medium 3", "provider": "Mistral", "release_date": "2025-05-01", "params_total_M": null, "params_active_M": null, "architecture": null, "is_reasoning": false, "open_weights": false, "canonical_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "mistral-eval", "prompt_style": "mixed (5-shot CoT for GPQA Diamond / MMLU Pro per Mistral blog; 0-shot for others)", "temperature": "0.0", "context": "default", "multimodal_input": true, "notes": "Per Mistral Medium 3 blog: https://mistral.ai/news/mistral-medium-3. \"Same internal evaluation pipeline\" across all benchmarks." } }, { "id": "mistral-large-3", "name": "Mistral Large 3", "provider": "Mistral", "release_date": "2025-12-01", "params_total_M": 675000, "params_active_M": null, "architecture": "MoE", "is_reasoning": false, "open_weights": true }, { "id": "codestral-25.01", "name": "Codestral 25.01", "provider": "Mistral", "release_date": "2025-01-15", "params_total_M": 22000, "params_active_M": 22000, "architecture": "Dense", "is_reasoning": false, "open_weights": true }, { "id": "devstral-2", "name": "Devstral 2", "provider": "Mistral", "release_date": "2025-12-01", "params_total_M": 123000, "params_active_M": null, "architecture": "Dense", "is_reasoning": false, "open_weights": true }, { "id": "phi-4", "name": "Phi-4", "provider": "Microsoft", "release_date": "2025-01-01", "params_total_M": 14000, "params_active_M": 14000, "architecture": "Dense", "is_reasoning": false, "open_weights": true, "canonical_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "OpenAI simple-evals", "prompt_style": "default", "temperature": "0.5", "context": "default (16K)", "notes": "Per Phi-4 paper (arxiv:2412.08905) Table 1, simple-evals framework, temperature 0.5." } }, { "id": "phi-4-reasoning", "name": "Phi-4-reasoning", "provider": "Microsoft", "release_date": "2025-04-30", "params_total_M": 14000, "params_active_M": 14000, "architecture": "Dense", "is_reasoning": true, "open_weights": true, "canonical_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1 (avg of 5)", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "0.8", "context": "default (32K)", "notes": "Per Phi-4-reasoning paper (arxiv:2504.21318) Tables 1-2. Thinking model, temp 0.8, max_seq_len 31k." } }, { "id": "phi-4-reasoning-plus", "name": "Phi-4-reasoning-plus", "provider": "Microsoft", "release_date": "2025-04-30", "params_total_M": 14000, "params_active_M": 14000, "architecture": "Dense", "is_reasoning": true, "open_weights": true, "canonical_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1 (avg of 5)", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "0.8", "context": "default (32K)", "notes": "Per Phi-4-reasoning paper. Same as phi-4-reasoning + GRPO RL on top." } }, { "id": "nemotron-ultra-253b", "name": "Nemotron Ultra 253B", "provider": "NVIDIA", "release_date": "2025-04-10", "params_total_M": 253000, "params_active_M": null, "architecture": "MoE", "is_reasoning": true, "open_weights": true, "canonical_setting": { "mode": "thinking", "effort": "default", "tools": "none (benchmark-specified)", "sampling": "pass@1 (avg of up to 16 benchmark runs)", "judge": "rule-based", "harness": "NVIDIA eval pipeline (transformers/vLLM)", "prompt_style": "custom prompt templates per benchmark", "temperature": "0.6 (thinking mode)", "context": "32k", "notes": "Per NVIDIA HF model card https://huggingface.co/nvidia/Llama-3_1-Nemotron-Ultra-253B-v1: Reasoning On mode, temperature=0.6, top_p=0.95, 32k context, benchmarks run up to 16x and averaged." } }, { "id": "amazon-nova-pro", "name": "Amazon Nova Pro", "provider": "Amazon", "release_date": "2025-01-01", "params_total_M": null, "params_active_M": null, "architecture": null, "is_reasoning": false, "open_weights": false, "canonical_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "zeroshot-cot", "temperature": 0.0, "context": "default", "notes": "Per Amazon Nova Premier tech report (cross-reports Nova Pro) and Nova family arxiv 2506.12103. SimpleQA cell reported with SerpApi web tool — see cell-level reported_setting." } }, { "id": "amazon-nova-premier", "name": "Amazon Nova Premier", "provider": "Amazon", "release_date": "2025-04-01", "params_total_M": null, "params_active_M": null, "architecture": null, "is_reasoning": false, "open_weights": false, "canonical_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "zeroshot-cot", "temperature": 0.0, "context": "default", "notes": "Per Amazon Nova Premier tech report (Apr 2025). SimpleQA cell reported with SerpApi web tool — see cell-level reported_setting." } }, { "id": "command-a", "name": "Command A", "provider": "Cohere", "release_date": "2025-03-13", "params_total_M": 111000, "params_active_M": null, "architecture": "Dense", "is_reasoning": false, "open_weights": true, "canonical_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "simple-evals (for MMLU/MMLU-Pro/GPQA); internal reproductions for code/SQL", "prompt_style": "default", "temperature": "default", "context": "default (256K)", "notes": "Per Command A tech report (cohere.com/research/papers/command-a-technical-report.pdf)." } }, { "id": "exaone-4.0-32b", "name": "EXAONE 4.0 32B", "provider": "LG AI Research", "release_date": "2025-07-15", "params_total_M": 32000, "params_active_M": 32000, "architecture": "Hybrid (Local+Global Attention 3:1, QK-Reorder-LN)", "is_reasoning": true, "open_weights": true, "canonical_setting": { "mode": "thinking", "effort": "default", "tools": "none (per-bench override)", "sampling": "avg-of-N (n=32 AIME/HMMT, n=8 GPQA, n=4 LCB/Tau-bench, n=1 others)", "judge": "rule-based", "harness": "LG AI Research internal", "prompt_style": "default", "temperature": "0.6, top_p=0.95, presence_penalty=1.5 (Reasoning mode)", "context": "128k", "notes": "Per LG AI Research arxiv:2507.11407 (Table 3): Reasoning mode is canonical (lead results). 32B max context 131,072 tokens." } }, { "id": "minimax-m2", "name": "MiniMax-M2", "provider": "MiniMax", "release_date": "2025-10-01", "params_total_M": null, "params_active_M": null, "architecture": null, "is_reasoning": true, "open_weights": false, "canonical_setting": { "mode": "thinking", "effort": "default", "tools": "none (per-bench override)", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default", "context": "default", "notes": "Per MiniMax M2 model card." } }, { "id": "minimax-m2.5", "name": "MiniMax-M2.5", "provider": "MiniMax", "release_date": "2026-04-22", "params_total_M": null, "params_active_M": null, "architecture": null, "is_reasoning": true, "open_weights": false, "canonical_setting": { "mode": "thinking", "effort": "default", "tools": "none (per-bench override)", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default", "context": "default", "notes": "Per MiniMax M2.5 model card." } }, { "id": "olmo-2-13b", "name": "OLMo 2 13B", "provider": "Allen AI", "release_date": "2025-01-01", "params_total_M": 13000, "params_active_M": 13000, "architecture": "Dense", "is_reasoning": false, "open_weights": true }, { "id": "lfm2.5-1.2b-thinking", "name": "LFM2.5-1.2B-Thinking", "provider": "Liquid AI", "release_date": "2026-01-06", "params_total_M": 1200, "params_active_M": 1200, "architecture": "Hybrid", "is_reasoning": true, "open_weights": true }, { "id": "phi-4-mini", "name": "Phi-4-mini", "provider": "Microsoft", "release_date": "2025-03-01", "params_total_M": 3800, "params_active_M": 3800, "architecture": "Dense", "is_reasoning": false, "open_weights": true, "canonical_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default", "context": "default (128K)", "notes": "Per Phi-4-mini-instruct HF model card." } }, { "id": "falcon3-10b", "name": "Falcon3-10B-Instruct", "provider": "TII", "release_date": "2025-01-01", "params_total_M": 10000, "params_active_M": 10000, "architecture": "Dense", "is_reasoning": false, "open_weights": true }, { "id": "internlm3-8b", "name": "InternLM3-8B-Instruct", "provider": "Shanghai AI Lab", "release_date": "2025-01-15", "params_total_M": 8000, "params_active_M": 8000, "architecture": "Dense", "is_reasoning": false, "open_weights": true, "canonical_setting": { "mode": "non-thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "OpenCompass", "prompt_style": "0-shot (default per benchmark)", "temperature": "default", "notes": "Per https://github.com/InternLM/InternLM README: evaluated using OpenCompass. Values marked * = Thinking Mode (excluded from non-thinking canonical). Most benchmarks 0-shot." } }, { "id": "seed-thinking-v1.5", "name": "Seed-Thinking-v1.5", "provider": "ByteDance", "release_date": "2025-04-01", "params_total_M": null, "params_active_M": null, "architecture": null, "is_reasoning": true, "open_weights": false, "canonical_setting": { "mode": "thinking", "effort": "default", "tools": "none (per-bench override)", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default", "context": "default", "notes": "Per Seed-Thinking-v1.5 paper." } }, { "id": "gpt-4o", "name": "GPT-4o (2024-11-20)", "provider": "OpenAI", "release_date": "2024-11-20", "params_total_M": null, "params_active_M": null, "architecture": null, "is_reasoning": false, "open_weights": false, "canonical_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none (per-bench override)", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default", "context": "default", "notes": "Per OpenAI GPT-4.1 blog table." } }, { "id": "gpt-4o-mini", "name": "GPT-4o mini", "provider": "OpenAI", "release_date": "2024-07-18", "params_total_M": null, "params_active_M": null, "architecture": null, "is_reasoning": false, "open_weights": false, "canonical_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none (per-bench override)", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default", "context": "default", "notes": "Per OpenAI GPT-4.1 blog table." } }, { "id": "o1-high", "name": "OpenAI o1 (high)", "provider": "OpenAI", "release_date": "2024-12-17", "params_total_M": null, "params_active_M": null, "architecture": null, "is_reasoning": true, "open_weights": false, "canonical_setting": { "mode": "thinking", "effort": "high", "tools": "none (per-bench override)", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default", "context": "default", "notes": "Per OpenAI GPT-4.1 blog table." } }, { "id": "gemini-1.5-flash", "name": "Gemini 1.5 Flash", "provider": "Google", "release_date": "2024-05-14", "params_total_M": null, "params_active_M": null, "architecture": "moe", "is_reasoning": false, "open_weights": false, "canonical_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default", "context": "default", "notes": "Per Gemini 2.0 Flash Model Card (April 15 2025): non-thinking generation, default sampling. Same canonical setting applies to 1.5 / 2.0 family." } }, { "id": "gemini-1.5-pro", "name": "Gemini 1.5 Pro", "provider": "Google", "release_date": "2024-05-14", "params_total_M": null, "params_active_M": null, "architecture": "moe", "is_reasoning": false, "open_weights": false, "canonical_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default", "context": "default", "notes": "Per Gemini 2.0 Flash Model Card (April 15 2025): non-thinking generation, default sampling. Same canonical setting applies to 1.5 / 2.0 family." } }, { "id": "gemini-2.0-flash-lite", "name": "Gemini 2.0 Flash-Lite", "provider": "Google", "release_date": "2025-02-25", "params_total_M": null, "params_active_M": null, "architecture": "moe", "is_reasoning": false, "open_weights": false, "canonical_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default", "context": "default", "notes": "Per Gemini 2.0 Flash Model Card (April 15 2025): non-thinking generation, default sampling. Same canonical setting applies to 1.5 / 2.0 family." } }, { "id": "deepseek-v2-0506", "name": "DeepSeek-V2-0506", "provider": "DeepSeek", "release_date": "2024-05-06", "params_total_M": null, "params_active_M": null, "architecture": null, "is_reasoning": false, "open_weights": true, "canonical_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default", "context": "default", "notes": "Per DeepSeek V3 paper Table 6." } }, { "id": "deepseek-v2.5-0905", "name": "DeepSeek-V2.5-0905", "provider": "DeepSeek", "release_date": "2024-09-05", "params_total_M": null, "params_active_M": null, "architecture": null, "is_reasoning": false, "open_weights": true, "canonical_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default", "context": "default", "notes": "Per DeepSeek V3 paper Table 6." } }, { "id": "qwen2.5-72b-instruct", "name": "Qwen2.5 72B Instruct", "provider": "Alibaba", "release_date": "2024-09-19", "params_total_M": null, "params_active_M": null, "architecture": null, "is_reasoning": false, "open_weights": true, "canonical_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default", "context": "default", "notes": "Per DeepSeek V3 paper Table 6." } }, { "id": "llama-3.1-405b-instruct", "name": "LLaMA-3.1 405B Instruct", "provider": "Meta", "release_date": "2024-07-23", "params_total_M": null, "params_active_M": null, "architecture": null, "is_reasoning": false, "open_weights": true, "canonical_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default", "context": "default", "notes": "Per DeepSeek V3 paper Table 6." } }, { "id": "claude-3.5-sonnet", "name": "Claude 3.5 Sonnet (1022)", "provider": "Anthropic", "release_date": "2024-10-22", "params_total_M": null, "params_active_M": null, "architecture": null, "is_reasoning": false, "open_weights": false, "canonical_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default", "context": "default", "notes": "Per DeepSeek V3 paper Table 6." } }, { "id": "gpt-4o-0513", "name": "GPT-4o (2024-05-13)", "provider": "OpenAI", "release_date": "2024-05-13", "params_total_M": null, "params_active_M": null, "architecture": null, "is_reasoning": false, "open_weights": false, "canonical_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default", "context": "default", "notes": "Per DeepSeek V3 paper Table 6." } }, { "id": "minimax-m2.7", "name": "MiniMax M2.7", "provider": "MiniMax", "release_date": "2026-04-01", "params_total_M": null, "params_active_M": null, "architecture": null, "is_reasoning": true, "open_weights": true, "canonical_setting": { "mode": "thinking", "effort": "default", "tools": "none (per-bench override)", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default", "context": "default", "notes": "Per GLM-5.1 blog (cross-model)." } }, { "id": "minimax-m2.1", "name": "MiniMax M2.1", "provider": "MiniMax", "release_date": "2025-12-01", "params_total_M": null, "params_active_M": null, "architecture": null, "is_reasoning": true, "open_weights": true, "canonical_setting": { "mode": "thinking", "effort": "default", "tools": "none (per-bench override)", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default", "context": "default", "notes": "Per MiniMax M2.5 model card (referenced as predecessor)." } }, { "id": "gemini-2.0-pro", "name": "Gemini 2.0 Pro", "provider": "Google", "release_date": "2025-02-05", "params_total_M": null, "params_active_M": null, "architecture": "moe", "is_reasoning": false, "open_weights": false, "canonical_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default", "context": "default (1M)", "notes": "Per Gemma 3 tech report Table 6 IT (Google self-report). Gemini 2.0 Pro experimental." } }, { "id": "gemma-2-2b", "name": "Gemma 2 2B", "provider": "Google", "release_date": "2024-06-27", "params_total_M": 2000, "params_active_M": 2000, "architecture": "dense", "is_reasoning": false, "open_weights": true, "canonical_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default", "context": "default", "notes": "Per Gemma 3 tech report Table 6 IT instruction-tuned column." } }, { "id": "gemma-2-9b", "name": "Gemma 2 9B", "provider": "Google", "release_date": "2024-06-27", "params_total_M": 9000, "params_active_M": 9000, "architecture": "dense", "is_reasoning": false, "open_weights": true, "canonical_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default", "context": "default", "notes": "Per Gemma 3 tech report Table 6 IT instruction-tuned column." } }, { "id": "gemma-2-27b", "name": "Gemma 2 27B", "provider": "Google", "release_date": "2024-06-27", "params_total_M": 27000, "params_active_M": 27000, "architecture": "dense", "is_reasoning": false, "open_weights": true, "canonical_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default", "context": "default", "notes": "Per Gemma 3 tech report Table 6 IT instruction-tuned column." } }, { "id": "gemma-3-1b", "name": "Gemma 3 1B", "provider": "Google", "release_date": "2025-03-12", "params_total_M": 1000, "params_active_M": 1000, "architecture": "dense", "is_reasoning": false, "open_weights": true, "canonical_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default", "context": "default", "notes": "Per Gemma 3 tech report Table 6 IT instruction-tuned column." } }, { "id": "gemma-3-4b", "name": "Gemma 3 4B", "provider": "Google", "release_date": "2025-03-12", "params_total_M": 4000, "params_active_M": 4000, "architecture": "dense", "is_reasoning": false, "open_weights": true, "canonical_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default", "context": "default", "notes": "Per Gemma 3 tech report Table 6 IT instruction-tuned column." } }, { "id": "gemma-3-12b", "name": "Gemma 3 12B", "provider": "Google", "release_date": "2025-03-12", "params_total_M": 12000, "params_active_M": 12000, "architecture": "dense", "is_reasoning": false, "open_weights": true, "canonical_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default", "context": "default", "notes": "Per Gemma 3 tech report Table 6 IT instruction-tuned column." } }, { "id": "phi-3-14b", "name": "Phi-3 Medium 14B", "provider": "Microsoft", "release_date": "2024-05-21", "params_total_M": 14000, "params_active_M": 14000, "architecture": "dense", "is_reasoning": false, "open_weights": true, "canonical_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "simple-evals", "prompt_style": "default", "temperature": "0.5", "context": "128K", "notes": "Per Phi-4 paper Table 1 (Microsoft self-test baseline)." } }, { "id": "qwen-2.5-14b-instruct", "name": "Qwen2.5-14B", "provider": "Alibaba", "release_date": "2024-09-19", "params_total_M": 14000, "params_active_M": 14000, "architecture": "dense", "is_reasoning": false, "open_weights": true, "canonical_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default", "context": "default (128K)", "notes": "Per Phi-4 paper Table 1 baseline." } }, { "id": "llama-3.3-70b-instruct", "name": "Llama-3.3-70B-Instruct", "provider": "Meta", "release_date": "2024-12-06", "params_total_M": 70000, "params_active_M": 70000, "architecture": "dense", "is_reasoning": false, "open_weights": true, "canonical_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default", "context": "default (128K)", "notes": "Per Phi-4 paper Table 1 baseline." } }, { "id": "openthinker2-32b", "name": "OpenThinker2-32B", "provider": "Open Thoughts", "release_date": "2025-04-04", "params_total_M": 32000, "params_active_M": 32000, "architecture": "dense", "is_reasoning": true, "open_weights": true, "canonical_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default", "context": "default", "notes": "Per Phi-4-reasoning paper Table 1 baseline (Microsoft self-test)." } }, { "id": "exaone-deep-32b", "name": "EXAONE-Deep-32B", "provider": "LG AI Research", "release_date": "2025-03-18", "params_total_M": 32000, "params_active_M": 32000, "architecture": "dense", "is_reasoning": true, "open_weights": true, "canonical_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default", "context": "default", "notes": "Per Phi-4-reasoning paper Table 1 baseline (Microsoft self-test). Distinct from BP exaone-4.0-32b." } }, { "id": "o1-mini", "name": "o1-mini", "provider": "OpenAI", "release_date": "2024-09-12", "params_total_M": null, "params_active_M": null, "architecture": null, "is_reasoning": true, "open_weights": false, "canonical_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default", "context": "default", "notes": "Per Phi-4-reasoning paper Table 1 baseline (Microsoft self-test)." } }, { "id": "gpt-3.5-turbo", "name": "GPT-3.5 Turbo (0125)", "provider": "OpenAI", "release_date": "2024-01-25", "params_total_M": null, "params_active_M": null, "architecture": null, "is_reasoning": false, "open_weights": false, "canonical_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "OLMES (Allen AI)", "prompt_style": "default", "temperature": "default", "context": "default", "notes": "Per OLMo 2 paper Table 7 Instruct (instruction-tuned variant)." } }, { "id": "llama-3.2-1b", "name": "Llama 3.2 1B Instruct", "provider": "Meta", "release_date": "2024-09-25", "params_total_M": 1000, "params_active_M": 1000, "architecture": "dense", "is_reasoning": false, "open_weights": true, "canonical_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "OLMES (Allen AI)", "prompt_style": "default", "temperature": "default", "context": "default", "notes": "Per OLMo 2 paper Table 7 Instruct (instruction-tuned variant)." } }, { "id": "qwen-2.5-1.5b", "name": "Qwen2.5-1.5B-Instruct", "provider": "Alibaba", "release_date": "2024-09-19", "params_total_M": 1500, "params_active_M": 1500, "architecture": "dense", "is_reasoning": false, "open_weights": true, "canonical_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "OLMES (Allen AI)", "prompt_style": "default", "temperature": "default", "context": "default", "notes": "Per OLMo 2 paper Table 7 Instruct (instruction-tuned variant)." } }, { "id": "ministral-8b", "name": "Ministral 8B Instruct 2410", "provider": "Mistral", "release_date": "2024-10-16", "params_total_M": 8000, "params_active_M": 8000, "architecture": "dense", "is_reasoning": false, "open_weights": true, "canonical_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "OLMES (Allen AI)", "prompt_style": "default", "temperature": "default", "context": "default", "notes": "Per OLMo 2 paper Table 7 Instruct (instruction-tuned variant)." } }, { "id": "llama-3.1-8b", "name": "Llama 3.1 8B Instruct", "provider": "Meta", "release_date": "2024-07-23", "params_total_M": 8000, "params_active_M": 8000, "architecture": "dense", "is_reasoning": false, "open_weights": true, "canonical_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "OLMES (Allen AI)", "prompt_style": "default", "temperature": "default", "context": "default", "notes": "Per OLMo 2 paper Table 7 Instruct (instruction-tuned variant)." } }, { "id": "tulu-3-8b", "name": "Tulu 3 8B", "provider": "Allen AI", "release_date": "2024-11-22", "params_total_M": 8000, "params_active_M": 8000, "architecture": "dense", "is_reasoning": false, "open_weights": true, "canonical_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "OLMES (Allen AI)", "prompt_style": "default", "temperature": "default", "context": "default", "notes": "Per OLMo 2 paper Table 7 Instruct (instruction-tuned variant)." } }, { "id": "qwen-2.5-7b", "name": "Qwen2.5-7B-Instruct", "provider": "Alibaba", "release_date": "2024-09-19", "params_total_M": 7000, "params_active_M": 7000, "architecture": "dense", "is_reasoning": false, "open_weights": true, "canonical_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "OLMES (Allen AI)", "prompt_style": "default", "temperature": "default", "context": "default", "notes": "Per OLMo 2 paper Table 7 Instruct (instruction-tuned variant)." } }, { "id": "qwen-2.5-14b", "name": "Qwen2.5-14B-Instruct", "provider": "Alibaba", "release_date": "2024-09-19", "params_total_M": 14000, "params_active_M": 14000, "architecture": "dense", "is_reasoning": false, "open_weights": true, "canonical_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "OLMES (Allen AI)", "prompt_style": "default", "temperature": "default", "context": "default", "notes": "Per OLMo 2 paper Table 7 Instruct (instruction-tuned variant)." } }, { "id": "qwen-2.5-32b", "name": "Qwen2.5-32B-Instruct", "provider": "Alibaba", "release_date": "2024-09-19", "params_total_M": 32000, "params_active_M": 32000, "architecture": "dense", "is_reasoning": false, "open_weights": true, "canonical_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "OLMES (Allen AI)", "prompt_style": "default", "temperature": "default", "context": "default", "notes": "Per OLMo 2 paper Table 7 Instruct (instruction-tuned variant)." } }, { "id": "mistral-small-24b", "name": "Mistral Small 24B Instruct", "provider": "Mistral", "release_date": "2025-01-30", "params_total_M": 24000, "params_active_M": 24000, "architecture": "dense", "is_reasoning": false, "open_weights": true, "canonical_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "OLMES (Allen AI)", "prompt_style": "default", "temperature": "default", "context": "default", "notes": "Per OLMo 2 paper Table 7 Instruct (instruction-tuned variant)." } }, { "id": "llama-3.1-70b", "name": "Llama 3.1 70B Instruct", "provider": "Meta", "release_date": "2024-07-23", "params_total_M": 70000, "params_active_M": 70000, "architecture": "dense", "is_reasoning": false, "open_weights": true, "canonical_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "OLMES (Allen AI)", "prompt_style": "default", "temperature": "default", "context": "default", "notes": "Per OLMo 2 paper Table 7 Instruct (instruction-tuned variant)." } }, { "id": "olmo-1b", "name": "OLMo 1B 0724", "provider": "Allen AI", "release_date": "2024-07-01", "params_total_M": 1000, "params_active_M": 1000, "architecture": "dense", "is_reasoning": false, "open_weights": true, "canonical_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "OLMES (Allen AI)", "prompt_style": "default", "temperature": "default", "context": "default", "notes": "Per OLMo 2 paper Table 7 Instruct (instruction-tuned variant)." } }, { "id": "smollm2-1.7b", "name": "SmolLM2 1.7B Instruct", "provider": "Hugging Face", "release_date": "2024-11-01", "params_total_M": 1700, "params_active_M": 1700, "architecture": "dense", "is_reasoning": false, "open_weights": true, "canonical_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "OLMES (Allen AI)", "prompt_style": "default", "temperature": "default", "context": "default", "notes": "Per OLMo 2 paper Table 7 Instruct (instruction-tuned variant)." } }, { "id": "olmo-7b-0424", "name": "OLMo 7B 0424 Instruct", "provider": "Allen AI", "release_date": "2024-04-17", "params_total_M": 7000, "params_active_M": 7000, "architecture": "dense", "is_reasoning": false, "open_weights": true, "canonical_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "OLMES (Allen AI)", "prompt_style": "default", "temperature": "default", "context": "default", "notes": "Per OLMo 2 paper Table 7 Instruct (instruction-tuned variant)." } }, { "id": "olmo-2-1b", "name": "OLMo 2 1B Instruct", "provider": "Allen AI", "release_date": "2025-01-03", "params_total_M": 1000, "params_active_M": 1000, "architecture": "dense", "is_reasoning": false, "open_weights": true, "canonical_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "OLMES (Allen AI)", "prompt_style": "default", "temperature": "default", "context": "default", "notes": "Per OLMo 2 paper Table 7 Instruct (instruction-tuned variant)." } }, { "id": "olmo-2-7b", "name": "OLMo 2 7B Instruct", "provider": "Allen AI", "release_date": "2025-01-03", "params_total_M": 7000, "params_active_M": 7000, "architecture": "dense", "is_reasoning": false, "open_weights": true, "canonical_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "OLMES (Allen AI)", "prompt_style": "default", "temperature": "default", "context": "default", "notes": "Per OLMo 2 paper Table 7 Instruct (instruction-tuned variant)." } }, { "id": "olmo-2-32b", "name": "OLMo 2 32B Instruct", "provider": "Allen AI", "release_date": "2025-01-03", "params_total_M": 32000, "params_active_M": 32000, "architecture": "dense", "is_reasoning": false, "open_weights": true, "canonical_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "OLMES (Allen AI)", "prompt_style": "default", "temperature": "default", "context": "default", "notes": "Per OLMo 2 paper Table 7 Instruct (instruction-tuned variant)." } }, { "id": "lfm2.5-1.2b-instruct", "name": "LFM2.5-1.2B-Instruct", "provider": "Liquid AI", "release_date": "2025-11", "params_b": 1.2, "canonical_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default", "context": "32k", "notes": "Per Liquid AI LFM2.5 introduction blog. Instruct (non-thinking) variant." } }, { "id": "granite-4.0-h-1b", "name": "Granite-4.0-H-1B", "provider": "IBM", "release_date": "2025", "params_b": 1.0, "canonical_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default", "context": "default", "notes": "IBM Granite 4.0 Hybrid 1B instruct." } }, { "id": "granite-4.0-1b", "name": "Granite-4.0-1B", "provider": "IBM", "release_date": "2025", "params_b": 1.0, "canonical_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default", "context": "default", "notes": "IBM Granite 4.0 1B instruct." } }, { "id": "gemma-3-1b-it", "name": "Gemma 3 1B", "provider": "Google", "release_date": "2025", "params_b": 1.0, "canonical_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default", "context": "32k", "notes": "Google Gemma 3 1B instruction-tuned." } }, { "id": "llama-3.2-1b-instruct", "name": "Llama 3.2 1B", "provider": "Meta", "release_date": "2024-09", "params_b": 1.0, "canonical_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default", "context": "128k", "notes": "Meta Llama 3.2 1B Instruct." } }, { "id": "claude-sonnet-3.7", "name": "Claude Sonnet 3.7", "provider": "Anthropic", "release_date": "2025-02-24", "params_total_M": null, "params_active_M": null, "architecture": null, "is_reasoning": false, "open_weights": false }, { "id": "deepseek-3.1", "name": "DeepSeek 3.1", "provider": "DeepSeek", "release_date": "2025-08-21", "params_total_M": 685000, "params_active_M": 37000, "architecture": "MoE", "is_reasoning": false, "open_weights": true }, { "id": "mistral-large-2", "name": "Mistral Large 2", "provider": "Mistral", "release_date": "2024-07-24", "params_total_M": 123000, "params_active_M": 123000, "architecture": "dense", "is_reasoning": false, "open_weights": true, "canonical_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default", "context": "128K", "notes": "Per Command A paper (third-party Cohere reproduction baseline)." } }, { "id": "command-r7b", "name": "Command R7B", "provider": "Cohere", "release_date": "2024-12-13", "params_total_M": 7000, "params_active_M": 7000, "architecture": "dense", "is_reasoning": false, "open_weights": true, "canonical_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "simple-evals", "prompt_style": "default", "temperature": "default", "context": "128K", "notes": "Per Command A tech report (Cohere first-party)." } }, { "id": "gemini-1.5-flash-8b", "name": "Gemini 1.5 Flash 8B", "provider": "Google", "release_date": "2024-10-03", "params_total_M": 8000, "params_active_M": 8000, "architecture": "dense", "is_reasoning": false, "open_weights": false, "canonical_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default", "context": "1M", "notes": "Per Command A paper baseline." } }, { "id": "gpt-5-mini", "name": "GPT-5 mini", "provider": "OpenAI", "release_date": "2025-07-10", "params_total_M": null, "params_active_M": null, "architecture": null, "is_reasoning": true, "open_weights": false, "canonical_setting": { "mode": "thinking", "effort": "high", "tools": "none (per-bench override)", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default", "context": "default", "notes": "Per OpenAI GPT-5 developer blog https://openai.com/index/introducing-gpt-5-for-developers/" }, "notes": "GPT-5 mini (high effort thinking mode). Launch data from OpenAI dev blog." }, { "id": "gpt-5-nano", "name": "GPT-5 nano", "provider": "OpenAI", "release_date": "2025-07-10", "params_total_M": null, "params_active_M": null, "architecture": null, "is_reasoning": true, "open_weights": false, "canonical_setting": { "mode": "thinking", "effort": "high", "tools": "none (per-bench override)", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default", "context": "default", "notes": "Per OpenAI GPT-5 developer blog https://openai.com/index/introducing-gpt-5-for-developers/" }, "notes": "GPT-5 nano (high effort thinking mode). Launch data from OpenAI dev blog." }, { "id": "gpt-5.2-codex", "name": "GPT-5.2-Codex", "provider": "OpenAI", "release_date": "2025-07-10", "params_total_M": null, "params_active_M": null, "architecture": null, "is_reasoning": true, "open_weights": false, "canonical_setting": { "mode": "thinking", "effort": "xhigh", "tools": "none (per-bench override)", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default", "context": "default", "notes": "Per OpenAI GPT-5.3-Codex blog https://openai.com/index/introducing-gpt-5-3-codex/ — shows GPT-5.2-Codex (xhigh) as cross-model baseline" }, "notes": "GPT-5.2-Codex (xhigh thinking mode). Seen as baseline in GPT-5.3-Codex blog." }, { "id": "minicpm-sala", "name": "MiniCPM-SALA", "provider": "OpenBMB", "release_date": null, "params_total_M": null, "params_active_M": null, "architecture": null, "is_reasoning": false, "open_weights": true, "canonical_setting": { "mode": "unknown", "effort": "n/a", "tools": "unknown", "sampling": "unknown", "judge": "unknown", "harness": "unknown", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown", "notes": "Model added from third-party aggregator (llm-stats.com); canonical setting not established." }, "notes": "Added from llm-stats.com 2026-04-29; only third-party scores available." }, { "id": "qwen2.5-coder-32b-instruct", "name": "Qwen2.5-Coder 32B Instruct", "provider": "Alibaba", "release_date": null, "params_total_M": null, "params_active_M": null, "architecture": null, "is_reasoning": false, "open_weights": true, "canonical_setting": { "mode": "unknown", "effort": "n/a", "tools": "unknown", "sampling": "unknown", "judge": "unknown", "harness": "unknown", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown", "notes": "Model added from third-party aggregator (llm-stats.com); canonical setting not established." }, "notes": "Added from llm-stats.com 2026-04-29; only third-party scores available." }, { "id": "sarvam-30b", "name": "Sarvam-30B", "provider": "Sarvam AI", "release_date": null, "params_total_M": null, "params_active_M": null, "architecture": null, "is_reasoning": false, "open_weights": true, "canonical_setting": { "mode": "unknown", "effort": "n/a", "tools": "unknown", "sampling": "unknown", "judge": "unknown", "harness": "unknown", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown", "notes": "Model added from third-party aggregator (llm-stats.com); canonical setting not established." }, "notes": "Added from llm-stats.com 2026-04-29; only third-party scores available." }, { "id": "qwen2.5-vl-32b-instruct", "name": "Qwen2.5 VL 32B Instruct", "provider": "Alibaba", "release_date": null, "params_total_M": null, "params_active_M": null, "architecture": null, "is_reasoning": false, "open_weights": true, "canonical_setting": { "mode": "unknown", "effort": "n/a", "tools": "unknown", "sampling": "unknown", "judge": "unknown", "harness": "unknown", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown", "notes": "Model added from third-party aggregator (llm-stats.com); canonical setting not established." }, "notes": "Added from llm-stats.com 2026-04-29; only third-party scores available." }, { "id": "granite-3.3-8b-instruct", "name": "Granite 3.3 8B", "provider": "IBM", "release_date": null, "params_total_M": null, "params_active_M": null, "architecture": null, "is_reasoning": false, "open_weights": true, "canonical_setting": { "mode": "unknown", "effort": "n/a", "tools": "unknown", "sampling": "unknown", "judge": "unknown", "harness": "unknown", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown", "notes": "Model added from third-party aggregator (llm-stats.com); canonical setting not established." }, "notes": "Added from llm-stats.com 2026-04-29; only third-party scores available." }, { "id": "granite-3.3-8b-base", "name": "Granite 3.3 8B Base", "provider": "IBM", "release_date": null, "params_total_M": null, "params_active_M": null, "architecture": null, "is_reasoning": false, "open_weights": true, "canonical_setting": { "mode": "unknown", "effort": "n/a", "tools": "unknown", "sampling": "unknown", "judge": "unknown", "harness": "unknown", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown", "notes": "Model added from third-party aggregator (llm-stats.com); canonical setting not established." }, "notes": "Added from llm-stats.com 2026-04-29; only third-party scores available." }, { "id": "gemini-diffusion", "name": "Gemini Diffusion", "provider": "Google", "release_date": null, "params_total_M": null, "params_active_M": null, "architecture": null, "is_reasoning": false, "open_weights": false, "canonical_setting": { "mode": "unknown", "effort": "n/a", "tools": "unknown", "sampling": "unknown", "judge": "unknown", "harness": "unknown", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown", "notes": "Model added from third-party aggregator (llm-stats.com); canonical setting not established." }, "notes": "Added from llm-stats.com 2026-04-29; only third-party scores available." }, { "id": "longcat-flash-chat", "name": "LongCat-Flash-Chat", "provider": "Meituan", "release_date": null, "params_total_M": null, "params_active_M": null, "architecture": null, "is_reasoning": false, "open_weights": true, "canonical_setting": { "mode": "unknown", "effort": "n/a", "tools": "unknown", "sampling": "unknown", "judge": "unknown", "harness": "unknown", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown", "notes": "Model added from third-party aggregator (llm-stats.com); canonical setting not established." }, "notes": "Added from llm-stats.com 2026-04-29; only third-party scores available." }, { "id": "qwen2.5-coder-7b-instruct", "name": "Qwen2.5-Coder 7B Instruct", "provider": "Alibaba", "release_date": null, "params_total_M": null, "params_active_M": null, "architecture": null, "is_reasoning": false, "open_weights": true, "canonical_setting": { "mode": "unknown", "effort": "n/a", "tools": "unknown", "sampling": "unknown", "judge": "unknown", "harness": "unknown", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown", "notes": "Model added from third-party aggregator (llm-stats.com); canonical setting not established." }, "notes": "Added from llm-stats.com 2026-04-29; only third-party scores available." }, { "id": "grok-2", "name": "Grok-2", "provider": "xAI", "release_date": null, "params_total_M": null, "params_active_M": null, "architecture": null, "is_reasoning": false, "open_weights": false, "canonical_setting": { "mode": "unknown", "effort": "n/a", "tools": "unknown", "sampling": "unknown", "judge": "unknown", "harness": "unknown", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown", "notes": "Model added from third-party aggregator (llm-stats.com); canonical setting not established." }, "notes": "Added from llm-stats.com 2026-04-29; only third-party scores available." }, { "id": "claude-3.5-haiku", "name": "Claude 3.5 Haiku", "provider": "Anthropic", "release_date": null, "params_total_M": null, "params_active_M": null, "architecture": null, "is_reasoning": false, "open_weights": false, "canonical_setting": { "mode": "unknown", "effort": "n/a", "tools": "unknown", "sampling": "unknown", "judge": "unknown", "harness": "unknown", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown", "notes": "Model added from third-party aggregator (llm-stats.com); canonical setting not established." }, "notes": "Added from llm-stats.com 2026-04-29; only third-party scores available." }, { "id": "gpt-4-turbo", "name": "GPT-4 Turbo", "provider": "OpenAI", "release_date": null, "params_total_M": null, "params_active_M": null, "architecture": null, "is_reasoning": false, "open_weights": false, "canonical_setting": { "mode": "unknown", "effort": "n/a", "tools": "unknown", "sampling": "unknown", "judge": "unknown", "harness": "unknown", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown", "notes": "Model added from third-party aggregator (llm-stats.com); canonical setting not established." }, "notes": "Added from llm-stats.com 2026-04-29; only third-party scores available." }, { "id": "qwen2-72b-instruct", "name": "Qwen2 72B Instruct", "provider": "Alibaba", "release_date": null, "params_total_M": null, "params_active_M": null, "architecture": null, "is_reasoning": false, "open_weights": true, "canonical_setting": { "mode": "unknown", "effort": "n/a", "tools": "unknown", "sampling": "unknown", "judge": "unknown", "harness": "unknown", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown", "notes": "Model added from third-party aggregator (llm-stats.com); canonical setting not established." }, "notes": "Added from llm-stats.com 2026-04-29; only third-party scores available." }, { "id": "grok-2-mini", "name": "Grok-2 mini", "provider": "xAI", "release_date": null, "params_total_M": null, "params_active_M": null, "architecture": null, "is_reasoning": false, "open_weights": false, "canonical_setting": { "mode": "unknown", "effort": "n/a", "tools": "unknown", "sampling": "unknown", "judge": "unknown", "harness": "unknown", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown", "notes": "Model added from third-party aggregator (llm-stats.com); canonical setting not established." }, "notes": "Added from llm-stats.com 2026-04-29; only third-party scores available." }, { "id": "nova-lite", "name": "Nova Lite", "provider": "Amazon", "release_date": null, "params_total_M": null, "params_active_M": null, "architecture": null, "is_reasoning": false, "open_weights": false, "canonical_setting": { "mode": "unknown", "effort": "n/a", "tools": "unknown", "sampling": "unknown", "judge": "unknown", "harness": "unknown", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown", "notes": "Model added from third-party aggregator (llm-stats.com); canonical setting not established." }, "notes": "Added from llm-stats.com 2026-04-29; only third-party scores available." }, { "id": "claude-3-opus", "name": "Claude 3 Opus", "provider": "Anthropic", "release_date": null, "params_total_M": null, "params_active_M": null, "architecture": null, "is_reasoning": false, "open_weights": false, "canonical_setting": { "mode": "unknown", "effort": "n/a", "tools": "unknown", "sampling": "unknown", "judge": "unknown", "harness": "unknown", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown", "notes": "Model added from third-party aggregator (llm-stats.com); canonical setting not established." }, "notes": "Added from llm-stats.com 2026-04-29; only third-party scores available." }, { "id": "mistral-small-3-24b-instruct", "name": "Mistral Small 3 24B Instruct", "provider": "Mistral AI", "release_date": null, "params_total_M": null, "params_active_M": null, "architecture": null, "is_reasoning": false, "open_weights": true, "canonical_setting": { "mode": "unknown", "effort": "n/a", "tools": "unknown", "sampling": "unknown", "judge": "unknown", "harness": "unknown", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown", "notes": "Model added from third-party aggregator (llm-stats.com); canonical setting not established." }, "notes": "Added from llm-stats.com 2026-04-29; only third-party scores available." }, { "id": "ibm-granite-4.0-tiny-preview", "name": "IBM Granite 4.0 Tiny Preview", "provider": "IBM", "release_date": null, "params_total_M": null, "params_active_M": null, "architecture": null, "is_reasoning": false, "open_weights": true, "canonical_setting": { "mode": "unknown", "effort": "n/a", "tools": "unknown", "sampling": "unknown", "judge": "unknown", "harness": "unknown", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown", "notes": "Model added from third-party aggregator (llm-stats.com); canonical setting not established." }, "notes": "Added from llm-stats.com 2026-04-29; only third-party scores available." }, { "id": "nova-micro", "name": "Nova Micro", "provider": "Amazon", "release_date": null, "params_total_M": null, "params_active_M": null, "architecture": null, "is_reasoning": false, "open_weights": false, "canonical_setting": { "mode": "unknown", "effort": "n/a", "tools": "unknown", "sampling": "unknown", "judge": "unknown", "harness": "unknown", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown", "notes": "Model added from third-party aggregator (llm-stats.com); canonical setting not established." }, "notes": "Added from llm-stats.com 2026-04-29; only third-party scores available." }, { "id": "codestral-22b", "name": "Codestral-22B", "provider": "Mistral AI", "release_date": null, "params_total_M": null, "params_active_M": null, "architecture": null, "is_reasoning": false, "open_weights": true, "canonical_setting": { "mode": "unknown", "effort": "n/a", "tools": "unknown", "sampling": "unknown", "judge": "unknown", "harness": "unknown", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown", "notes": "Model added from third-party aggregator (llm-stats.com); canonical setting not established." }, "notes": "Added from llm-stats.com 2026-04-29; only third-party scores available." }, { "id": "qwen2-7b-instruct", "name": "Qwen2 7B Instruct", "provider": "Alibaba", "release_date": null, "params_total_M": null, "params_active_M": null, "architecture": null, "is_reasoning": false, "open_weights": true, "canonical_setting": { "mode": "unknown", "effort": "n/a", "tools": "unknown", "sampling": "unknown", "judge": "unknown", "harness": "unknown", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown", "notes": "Model added from third-party aggregator (llm-stats.com); canonical setting not established." }, "notes": "Added from llm-stats.com 2026-04-29; only third-party scores available." }, { "id": "qwen2.5-omni-7b", "name": "Qwen2.5-Omni-7B", "provider": "Alibaba", "release_date": null, "params_total_M": null, "params_active_M": null, "architecture": null, "is_reasoning": false, "open_weights": true, "canonical_setting": { "mode": "unknown", "effort": "n/a", "tools": "unknown", "sampling": "unknown", "judge": "unknown", "harness": "unknown", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown", "notes": "Model added from third-party aggregator (llm-stats.com); canonical setting not established." }, "notes": "Added from llm-stats.com 2026-04-29; only third-party scores available." }, { "id": "claude-3-haiku", "name": "Claude 3 Haiku", "provider": "Anthropic", "release_date": null, "params_total_M": null, "params_active_M": null, "architecture": null, "is_reasoning": false, "open_weights": false, "canonical_setting": { "mode": "unknown", "effort": "n/a", "tools": "unknown", "sampling": "unknown", "judge": "unknown", "harness": "unknown", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown", "notes": "Model added from third-party aggregator (llm-stats.com); canonical setting not established." }, "notes": "Added from llm-stats.com 2026-04-29; only third-party scores available." }, { "id": "gemma-3n-e4b-instructed-litert-preview", "name": "Gemma 3n E4B Instructed LiteRT Preview", "provider": "Google", "release_date": null, "params_total_M": null, "params_active_M": null, "architecture": null, "is_reasoning": false, "open_weights": true, "canonical_setting": { "mode": "unknown", "effort": "n/a", "tools": "unknown", "sampling": "unknown", "judge": "unknown", "harness": "unknown", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown", "notes": "Model added from third-party aggregator (llm-stats.com); canonical setting not established." }, "notes": "Added from llm-stats.com 2026-04-29; only third-party scores available." }, { "id": "gemma-3n-e4b-instructed", "name": "Gemma 3n E4B Instructed", "provider": "Google", "release_date": null, "params_total_M": null, "params_active_M": null, "architecture": null, "is_reasoning": false, "open_weights": true, "canonical_setting": { "mode": "unknown", "effort": "n/a", "tools": "unknown", "sampling": "unknown", "judge": "unknown", "harness": "unknown", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown", "notes": "Model added from third-party aggregator (llm-stats.com); canonical setting not established." }, "notes": "Added from llm-stats.com 2026-04-29; only third-party scores available." }, { "id": "grok-1.5", "name": "Grok-1.5", "provider": "xAI", "release_date": null, "params_total_M": null, "params_active_M": null, "architecture": null, "is_reasoning": false, "open_weights": false, "canonical_setting": { "mode": "unknown", "effort": "n/a", "tools": "unknown", "sampling": "unknown", "judge": "unknown", "harness": "unknown", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown", "notes": "Model added from third-party aggregator (llm-stats.com); canonical setting not established." }, "notes": "Added from llm-stats.com 2026-04-29; only third-party scores available." }, { "id": "claude-3-sonnet", "name": "Claude 3 Sonnet", "provider": "Anthropic", "release_date": null, "params_total_M": null, "params_active_M": null, "architecture": null, "is_reasoning": false, "open_weights": false, "canonical_setting": { "mode": "unknown", "effort": "n/a", "tools": "unknown", "sampling": "unknown", "judge": "unknown", "harness": "unknown", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown", "notes": "Model added from third-party aggregator (llm-stats.com); canonical setting not established." }, "notes": "Added from llm-stats.com 2026-04-29; only third-party scores available." }, { "id": "pixtral-12b", "name": "Pixtral-12B", "provider": "Mistral AI", "release_date": null, "params_total_M": null, "params_active_M": null, "architecture": null, "is_reasoning": false, "open_weights": true, "canonical_setting": { "mode": "unknown", "effort": "n/a", "tools": "unknown", "sampling": "unknown", "judge": "unknown", "harness": "unknown", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown", "notes": "Model added from third-party aggregator (llm-stats.com); canonical setting not established." }, "notes": "Added from llm-stats.com 2026-04-29; only third-party scores available." }, { "id": "phi-3.5-moe-instruct", "name": "Phi-3.5-MoE-instruct", "provider": "Microsoft", "release_date": null, "params_total_M": null, "params_active_M": null, "architecture": null, "is_reasoning": false, "open_weights": true, "canonical_setting": { "mode": "unknown", "effort": "n/a", "tools": "unknown", "sampling": "unknown", "judge": "unknown", "harness": "unknown", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown", "notes": "Model added from third-party aggregator (llm-stats.com); canonical setting not established." }, "notes": "Added from llm-stats.com 2026-04-29; only third-party scores available." }, { "id": "gemma-3n-e2b-instructed-litert-preview", "name": "Gemma 3n E2B Instructed LiteRT (Preview)", "provider": "Google", "release_date": null, "params_total_M": null, "params_active_M": null, "architecture": null, "is_reasoning": false, "open_weights": true, "canonical_setting": { "mode": "unknown", "effort": "n/a", "tools": "unknown", "sampling": "unknown", "judge": "unknown", "harness": "unknown", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown", "notes": "Model added from third-party aggregator (llm-stats.com); canonical setting not established." }, "notes": "Added from llm-stats.com 2026-04-29; only third-party scores available." }, { "id": "gemma-3n-e2b-instructed", "name": "Gemma 3n E2B Instructed", "provider": "Google", "release_date": null, "params_total_M": null, "params_active_M": null, "architecture": null, "is_reasoning": false, "open_weights": true, "canonical_setting": { "mode": "unknown", "effort": "n/a", "tools": "unknown", "sampling": "unknown", "judge": "unknown", "harness": "unknown", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown", "notes": "Model added from third-party aggregator (llm-stats.com); canonical setting not established." }, "notes": "Added from llm-stats.com 2026-04-29; only third-party scores available." }, { "id": "phi-3.5-mini-instruct", "name": "Phi-3.5-mini-instruct", "provider": "Microsoft", "release_date": null, "params_total_M": null, "params_active_M": null, "architecture": null, "is_reasoning": false, "open_weights": true, "canonical_setting": { "mode": "unknown", "effort": "n/a", "tools": "unknown", "sampling": "unknown", "judge": "unknown", "harness": "unknown", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown", "notes": "Model added from third-party aggregator (llm-stats.com); canonical setting not established." }, "notes": "Added from llm-stats.com 2026-04-29; only third-party scores available." }, { "id": "kimi-k2.7-code", "name": "Kimi K2.7 Code", "provider": "Moonshot AI", "release_date": "2026-06-11", "params_total_M": 1000000, "params_active_M": 32000, "architecture": "MoE", "is_reasoning": true, "open_weights": true, "canonical_setting": { "mode": "thinking", "effort": "default", "tools": "benchmark-specified", "sampling": "pass@1 (avg-of-3 for long-horizon/tool-use benchmarks)", "judge": "benchmark-specified", "harness": "Kimi Code CLI + official benchmark harness", "prompt_style": "default", "temperature": "1.0; top_p=0.95", "context": "262144 (256K)", "notes": "Per the Kimi K2.7 Code model card. Thinking mode in Kimi Code CLI with temperature=1.0, top_p=0.95, and 262,144-token context." } }, { "id": "claude-opus-4.8", "name": "Claude Opus 4.8", "provider": "Anthropic", "release_date": "2026-05-28", "params_total_M": null, "params_active_M": null, "architecture": null, "is_reasoning": true, "open_weights": false, "canonical_setting": { "mode": "thinking", "effort": "high", "tools": "none (per-bench override)", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default", "context": "default", "notes": "Per Anthropic's Claude Opus 4.8 release post. Opus 4.8 defaults to high effort." } }, { "id": "kimi-k3", "name": "Kimi K3", "provider": "Moonshot AI", "release_date": "2026-07-17", "params_total_M": 2800000, "params_active_M": 104000, "architecture": "MoE", "is_reasoning": true, "open_weights": true, "canonical_setting": { "mode": "thinking", "effort": "max", "tools": "none (per-benchmark override)", "sampling": "pass@1; benchmark-specific multi-run sampling where explicitly reported", "judge": "benchmark-specified", "harness": "official or benchmark-specified", "prompt_style": "default", "temperature": "1.0; top_p=0.95 for reasoning/no-tool vision, top_p=1.0 for coding/agentic", "context": "1048576 (1M)", "notes": "Per the official Kimi K3 model card and technical report." } }, { "id": "claude-fable-5", "name": "Claude Fable 5", "provider": "Anthropic", "release_date": "2026-06-09", "params_total_M": null, "params_active_M": null, "architecture": null, "is_reasoning": true, "open_weights": false, "canonical_setting": { "mode": "thinking", "effort": "max", "tools": "none (per-benchmark override)", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official; product fallback enabled", "prompt_style": "default", "temperature": "default", "context": "default", "notes": "Per the Kimi K3 report's Claude Fable 5 max-with-fallback column. Fable 5 is distinct from Claude Mythos Preview." } }, { "id": "gpt-5.6-sol", "name": "GPT-5.6 Sol", "provider": "OpenAI", "release_date": "2026-07-09", "params_total_M": null, "params_active_M": null, "architecture": null, "is_reasoning": true, "open_weights": false, "canonical_setting": { "mode": "thinking", "effort": "max", "tools": "none (per-benchmark override)", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official or benchmark-specified", "prompt_style": "default", "temperature": "default", "context": "default", "notes": "Per the Kimi K3 report's GPT-5.6 Sol max column; cyber safeguards may apply to cyber evaluations." } }, { "id": "glm-5.2", "name": "GLM-5.2", "provider": "Z.ai", "release_date": "2026-06-16", "params_total_M": 753330, "params_active_M": null, "architecture": "MoE (DSA + IndexShare)", "is_reasoning": true, "open_weights": true, "canonical_setting": { "mode": "thinking", "effort": "max", "tools": "none (per-benchmark override)", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official or benchmark-specified", "prompt_style": "default", "temperature": "default", "context": "1048576 (1M)", "notes": "Per the official GLM-5.2 release blog/model card. High and Max effort are available; BP canonical uses Max. HF metadata reports 753.33B parameters." } }, { "id": "deepseek-v4-flash-0731", "name": "DeepSeek-V4-Flash-0731", "provider": "DeepSeek", "release_date": "2026-07-31", "params_total_M": 284000, "params_active_M": 13000, "architecture": "MoE", "is_reasoning": true, "open_weights": true, "canonical_setting": { "mode": "thinking", "effort": "max", "tools": "none (per-benchmark override)", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "DeepSeek Harness minimal mode for Code Agent tasks", "prompt_style": "default", "temperature": "1.0; top_p=0.95 for agentic scenarios, top_p=1.0 otherwise", "context": "1048576 (1M)", "notes": "Official DeepSeek-V4-Flash release dated 2026-07-31, superseding DeepSeek-V4-Flash Preview. The model card reports substantially enhanced agentic capabilities." } }, { "id": "qwen3.7-max", "name": "Qwen3.7-Max", "provider": "Alibaba", "release_date": "2026-05-21", "params_total_M": null, "params_active_M": null, "architecture": null, "is_reasoning": true, "open_weights": false, "canonical_setting": { "mode": "thinking", "effort": "xhigh", "tools": "none (per-benchmark override)", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official or benchmark-specified", "prompt_style": "recommended xhigh reasoning prompt", "temperature": "default", "context": "1000000 (1M)", "notes": "Official Alibaba release calls Qwen3.7-Max proprietary and recommends xhigh reasoning for reasoning scenarios." } }, { "id": "minimax-m3", "name": "MiniMax-M3", "provider": "MiniMax", "release_date": "2026-06-01", "params_total_M": 428000, "params_active_M": 23000, "architecture": "MoE (MiniMax Sparse Attention)", "is_reasoning": true, "open_weights": true, "canonical_setting": { "mode": "thinking", "effort": "adaptive", "tools": "none (per-benchmark override)", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official or benchmark-specified", "prompt_style": "default", "temperature": "1.0", "context": "1000000 (1M)", "notes": "OpenAI-compatible API enables thinking by default; Anthropic-compatible API requires adaptive. BP canonical records the enabled/adaptive reasoning setting." } }, { "id": "hy3-preview", "name": "Hy3 Preview", "provider": "Tencent", "release_date": "2026-04-23", "params_total_M": 295000, "params_active_M": 21000, "architecture": "MoE (192 experts, top-8; 3.8B MTP layer)", "open_weights": true, "is_reasoning": true, "canonical_setting": { "mode": "thinking", "effort": "high", "tools": "none (per-benchmark override)", "harness": "official or benchmark-specified", "prompt_style": "default", "sampling": "pass@1", "temperature": "0.9", "context": "256000", "judge": "benchmark-specified", "notes": "Hy3 official card recommends temperature=0.9, top_p=1.0 and high reasoning for complex tasks." } }, { "id": "hy3", "name": "Hy3", "provider": "Tencent", "release_date": "2026-07-06", "params_total_M": 295000, "params_active_M": 21000, "architecture": "MoE (192 experts, top-8; 3.8B MTP layer)", "open_weights": true, "is_reasoning": true, "canonical_setting": { "mode": "thinking", "effort": "high", "tools": "none (per-benchmark override)", "harness": "official or benchmark-specified", "prompt_style": "default", "sampling": "pass@1", "temperature": "0.9", "context": "256000", "judge": "benchmark-specified", "notes": "Hy3 official card recommends temperature=0.9, top_p=1.0 and high reasoning for complex tasks." } }, { "id": "doubao-seed-2.1-pro", "name": "Doubao Seed 2.1 Pro", "provider": "ByteDance", "release_date": "2026-06-23", "params_total_M": null, "params_active_M": null, "architecture": null, "is_reasoning": true, "open_weights": false, "canonical_setting": { "mode": "thinking", "effort": "unknown", "tools": "none (per-benchmark override)", "sampling": "pass@1 unless benchmark specifies otherwise", "judge": "benchmark-specified", "harness": "official or benchmark-specified", "prompt_style": "official default", "temperature": "unknown", "context": "unknown", "notes": "Per the official Seed2.1 release and model card. Pro is the stronger reasoning/agentic family member; reasoning effort, topology, parameter count, and API context are undisclosed." } }, { "id": "longcat-2.0", "name": "LongCat-2.0", "provider": "Meituan LongCat", "release_date": "2026-06-30", "params_total_M": 1600000, "params_active_M": 48000, "architecture": "MoE (768 routed experts, top-12; LongCat Sparse Attention; 3-step MTP; 135B N-gram embedding)", "is_reasoning": true, "open_weights": true, "canonical_setting": { "mode": "thinking", "effort": "default", "tools": "none (per-benchmark override)", "sampling": "pass@1 (benchmark-specific repeats)", "judge": "benchmark-specified", "harness": "official or benchmark-specified", "prompt_style": "default", "temperature": "benchmark-specific", "context": "262144 released config; trained on 1M-context data", "notes": "Official chat template defaults to thinking enabled. The release blog reports benchmark-specific temperatures and agent harnesses." } }, { "id": "intern-s2-preview-397b", "name": "Intern-S2-Preview-397B", "provider": "Shanghai AI Laboratory", "release_date": "2026-07-20", "params_total_M": 397000, "params_active_M": null, "architecture": "Multimodal MoE (Qwen3.5-MoE text backbone; 512 experts, top-10; hybrid linear/full attention; vision and time-series encoders)", "is_reasoning": true, "open_weights": true, "canonical_setting": { "mode": "thinking", "effort": "default", "tools": "none (per-benchmark override)", "sampling": "pass@1 unless benchmark-specific", "judge": "benchmark-specified", "harness": "OpenCompass, VLMEvalKit, or AgentCompass", "prompt_style": "default", "temperature": "0.8; top_p=0.95; top_k=50; min_p=0.0 recommended", "context": "262144; source evaluation max 256K text / 64K multimodal", "notes": "Official chat template enables thinking by default. The 397B model name is first-party; an active-parameter count is not disclosed." } }, { "id": "nemotron-3-ultra-550b-a55b", "name": "NVIDIA Nemotron 3 Ultra 550B-A55B", "provider": "NVIDIA", "release_date": "2026-06-04", "params_total_M": 550000, "params_active_M": 55000, "architecture": "Hybrid Mamba-2 + MoE + Attention", "is_reasoning": true, "open_weights": true, "canonical_setting": { "mode": "thinking", "effort": "default", "tools": "none (per-bench override)", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "1.0; top_p=0.95", "context": "1M", "notes": "Per NVIDIA Nemotron 3 Ultra technical report and BF16 model card. Reasoning is enabled by default; benchmark-specific agent tools and harnesses are recorded per score." } }, { "id": "glm-4.5", "name": "GLM-4.5", "provider": "Z.ai", "release_date": "2025-07-28", "params_total_M": 355000, "params_active_M": 32000, "architecture": "MoE", "is_reasoning": true, "open_weights": true, "canonical_setting": { "mode": "thinking", "effort": "default", "tools": "none (per-bench override)", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default", "context": "128K", "notes": "Per zai-org/GLM-4.5 model card and Z.ai release blog. The post-trained checkpoint supports hybrid thinking and non-thinking modes; BenchPress canonical is thinking." } }, { "id": "gpt-5.6-terra", "name": "GPT-5.6 Terra", "provider": "OpenAI", "release_date": "2026-07-09", "params_total_M": null, "params_active_M": null, "architecture": null, "is_reasoning": true, "open_weights": false, "canonical_setting": { "mode": "thinking", "effort": "max", "tools": "none (per-benchmark override)", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official or benchmark-specified", "prompt_style": "default", "temperature": "default", "context": "default", "notes": "Durable GPT-5.6 balanced capability tier. The API default is medium; BenchPress canonicalizes the maximum stable reasoning setting. Scores without a disclosed effort remain noncanonical." } }, { "id": "gpt-5.6-luna", "name": "GPT-5.6 Luna", "provider": "OpenAI", "release_date": "2026-07-09", "params_total_M": null, "params_active_M": null, "architecture": null, "is_reasoning": true, "open_weights": false, "canonical_setting": { "mode": "thinking", "effort": "max", "tools": "none (per-benchmark override)", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official or benchmark-specified", "prompt_style": "default", "temperature": "default", "context": "default", "notes": "Durable GPT-5.6 cost-sensitive capability tier. The API default is medium; BenchPress canonicalizes the maximum stable reasoning setting. Scores without a disclosed effort remain noncanonical." } }, { "id": "gemini-3.5-flash", "name": "Gemini 3.5 Flash", "provider": "Google", "release_date": "2026-05-19", "params_total_M": null, "params_active_M": null, "architecture": null, "is_reasoning": true, "open_weights": false, "canonical_setting": { "mode": "thinking", "effort": "medium", "tools": "none (per-benchmark override)", "sampling": "pass@1 (per-benchmark override)", "judge": "benchmark-specified", "harness": "official", "prompt_style": "default", "temperature": "default", "context": "1,048,576 input; 65,536 output", "notes": "Official API model ID gemini-3.5-flash. Runtime default thinking level is medium; the May benchmark evaluation uses default settings. Natively multimodal text/image/audio/video/PDF input with text output." } }, { "id": "claude-sonnet-5", "name": "Claude Sonnet 5", "provider": "Anthropic", "release_date": "2026-06-30", "params_total_M": null, "params_active_M": null, "architecture": null, "is_reasoning": true, "open_weights": false, "canonical_setting": { "mode": "thinking", "effort": "high", "tools": "none (per-bench override)", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official", "prompt_style": "default", "temperature": "default", "context": "1m", "notes": "Per the Anthropic Claude Sonnet 5 release and System Card: adaptive thinking is enabled by default; API effort defaults to high; supported efforts are low, medium, high, xhigh, and max; context is 1M with up to 128k synchronous output." } }, { "id": "claude-opus-5", "name": "Claude Opus 5", "provider": "Anthropic", "release_date": "2026-07-24", "params_total_M": null, "params_active_M": null, "architecture": null, "is_reasoning": true, "open_weights": false, "canonical_setting": { "mode": "thinking", "effort": "high", "tools": "none (per-benchmark override)", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official", "prompt_style": "default", "temperature": "default", "context": "1m", "notes": "Adaptive thinking is enabled by default; API and Claude Code default effort is high. Supported efforts are low, medium, high, xhigh, and max." } }, { "id": "composer-2.5", "name": "Composer 2.5", "provider": "Cursor", "release_date": "2026-05-18", "params_total_M": null, "params_active_M": null, "architecture": "MoE", "is_reasoning": true, "open_weights": false, "canonical_setting": { "mode": "unknown", "effort": "default", "tools": "Cursor agent tools", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "Cursor", "prompt_style": "default", "temperature": "default", "context": "default", "notes": "Per Cursor's 2026-05-18 release and model docs: Composer 2.5 is Cursor's agentic model, built on the Kimi K2.5 checkpoint with proprietary continued pretraining and RL. Fast is the default product tier and has the same intelligence as standard." } }, { "id": "grok-4.5", "name": "Grok 4.5", "provider": "xAI / SpaceXAI", "release_date": "2026-07-16", "params_total_M": null, "params_active_M": null, "architecture": null, "is_reasoning": true, "open_weights": false, "canonical_setting": { "mode": "thinking", "effort": "high", "tools": "none (per-benchmark override)", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official SpaceXAI API or benchmark-specified", "prompt_style": "unknown", "temperature": "unknown", "context": "500,000", "notes": "Grok 4.5 was publicly released on 2026-07-16; its model card is dated 2026-07-14. Official API identity grok-4.5; aliases grok-4.5-latest and grok-build-latest. Reasoning efforts are low, medium, and high (default); xhigh is treated as high by current docs but is not a distinct Grok 4.5 setting. Official prompt and temperature are undisclosed." } }, { "id": "gemini-3.6-flash", "name": "Gemini 3.6 Flash", "provider": "Google", "release_date": "2026-07-21", "params_total_M": null, "params_active_M": null, "architecture": null, "is_reasoning": true, "open_weights": false, "canonical_setting": { "mode": "thinking", "effort": "medium", "tools": "none (per-benchmark override)", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official", "prompt_style": "default", "temperature": "default", "context": "1,048,576 input; 65,536 output", "notes": "Stable API ID gemini-3.6-flash; runtime default is medium thinking. Text/image/video/audio/PDF input and text output. Custom temperature, top-p and top-k are unsupported/ignored. DeepSWE is an explicit high-thinking benchmark override." } }, { "id": "muse-spark-1.1", "name": "Muse Spark 1.1", "provider": "Meta", "release_date": "2026-07-09", "params_total_M": null, "params_active_M": null, "architecture": null, "is_reasoning": true, "open_weights": false, "canonical_setting": { "mode": "thinking", "effort": "xhigh", "tools": "none (per-benchmark override)", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "Meta Model API", "prompt_style": "default", "temperature": "1.0; top_p=1.0", "context": "1048576", "notes": "Official Meta API identity muse-spark-1.1. Inputs: text, image, video, audio, and PDF; text output. Reasoning is always on and supports minimal/low/medium/high/xhigh; omitted effort is model-determined and none is unsupported. The general-capability report uses xhigh, which remains the score-matrix canonical evaluation setting; safety/preparedness Sections 1-4 use high. Architecture, total parameters, active parameters, and the 1.1-specific maximum output length are undisclosed." } }, { "id": "composer-1", "name": "Composer 1", "provider": "Cursor", "release_date": "2025-10-29", "params_total_M": null, "params_active_M": null, "architecture": "MoE", "is_reasoning": false, "open_weights": false, "canonical_setting": { "mode": "agentic", "effort": "default", "tools": "Cursor agent tools", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "Cursor", "prompt_style": "default", "temperature": "default", "context": "long context; exact limit undisclosed", "notes": "Per Cursor's 2025-10-29 release: first Composer agent model, a proprietary MoE trained with RL for software engineering. No named reasoning-effort control or parameter count is stated." } }, { "id": "composer-1.5", "name": "Composer 1.5", "provider": "Cursor", "release_date": "2026-02-09", "params_total_M": null, "params_active_M": null, "architecture": null, "is_reasoning": true, "open_weights": false, "canonical_setting": { "mode": "thinking", "effort": "adaptive; no user-facing tier", "tools": "Cursor agent tools", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "Cursor", "prompt_style": "default", "temperature": "default", "context": "self-summarizing; exact limit undisclosed", "notes": "Per Cursor's 2026-02-09 release: a thinking model trained with 20x more RL on the same pretrained model as Composer 1. It adapts thinking length and can recursively self-summarize." } }, { "id": "composer-2", "name": "Composer 2", "provider": "Cursor", "release_date": "2026-03-19", "params_total_M": 1040000, "params_active_M": 32000, "architecture": "MoE", "is_reasoning": true, "open_weights": false, "canonical_setting": { "mode": "thinking", "effort": "default", "tools": "Cursor agent tools", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "Cursor", "prompt_style": "default", "temperature": "default", "context": "default", "notes": "Per Cursor's 2026-03-19 release and Composer 2 technical report: continued pretraining and RL on the Kimi K2.5 checkpoint. The final model is 1.04T total / 32B active. Fast is the default serving tier and has the same intelligence as standard; no effort tier is named." } }, { "id": "grok-4.6", "name": "Grok 4.6", "provider": "xAI / SpaceXAI", "release_date": "2026-08-12", "params_total_M": null, "params_active_M": null, "architecture": null, "is_reasoning": true, "open_weights": false, "canonical_setting": { "mode": "thinking", "effort": "high", "tools": "none (per-benchmark override)", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official SpaceXAI API or benchmark-specified", "prompt_style": "unknown", "temperature": "unknown", "context": "500,000", "notes": "Official reasoning efforts are low, medium, high (default), and xhigh. Grok 4.6 received supplemental training on anonymized Cursor workflow data; the card describes collaboration with Cursor, not joint training. The fused MoE optimization example concerns an agent-authored inference optimization and does not establish the model's full architecture. Official prompt and temperature are undisclosed." } }, { "id": "gemini-3.7-flash", "name": "Gemini 3.7 Flash", "provider": "Google", "release_date": "2026-08-13", "params_total_M": null, "params_active_M": null, "architecture": "Undisclosed; model card states that it is based on Gemini 3.6 Flash", "is_reasoning": true, "open_weights": false, "canonical_setting": { "mode": "thinking", "effort": "medium", "tools": "none (per-benchmark override)", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official", "prompt_style": "default", "temperature": "default", "context": "1,048,576 input; 65,536 output", "input_modalities": "text, image, video, audio, PDF", "output_modalities": "text", "thinking_levels": "low, medium, high; minimal unsupported", "api_model_ids": "gemini-3.7-flash (stable)", "knowledge_cutoff": "March 2026; some domains may remain limited to January 2025", "version": "GA 2026-08-13; stable model; latest update August 2026", "parameter_disclosure": "not disclosed", "notes": "Google API model page and changelog identify the stable gemini-3.7-flash endpoint, default medium thinking, low/medium/high support, 1,048,576 input tokens, 65,536 output tokens, natively multimodal inputs, and text output. The DeepMind model card states that the model is based on Gemini 3.6 Flash and gives the March 2026 / January 2025 cutoff qualification." } }, { "id": "gemini-3-pro-deep-think-2025-12", "name": "Gemini 3 Deep Think (Dec 2025)", "provider": "Google", "release_date": "2025-12-04", "params_total_M": null, "params_active_M": null, "architecture": null, "is_reasoning": true, "open_weights": false, "canonical_setting": { "mode": "deep-think", "effort": "vendor-specialized", "tools": "none (per-benchmark override)", "sampling": "pass@1 (per-benchmark override)", "judge": "benchmark-specified", "harness": "Google Deep Think", "prompt_style": "default", "temperature": "default", "context": "1m base model; Deep Think-specific limit undisclosed", "notes": "Specialized parallel-reasoning mode built on Gemini 3 Pro. The same Nov 18 scores were publicly released to Ultra users on Dec 4, 2025. No standalone API model ID or checkpoint hash was published." } }, { "id": "gemini-3.1-pro-deep-think-2026-02", "name": "Gemini 3.1 Deep Think (Feb 2026)", "provider": "Google", "release_date": "2026-02-12", "params_total_M": null, "params_active_M": null, "architecture": null, "is_reasoning": true, "open_weights": false, "canonical_setting": { "mode": "deep-think", "effort": "vendor-specialized", "tools": "none (per-benchmark override)", "sampling": "pass@1 (per-benchmark override)", "judge": "benchmark-specified", "harness": "Google Deep Think", "prompt_style": "default", "temperature": "default", "context": "1m base model; Deep Think-specific limit undisclosed", "notes": "Specialized parallel-reasoning mode built on Gemini 3.1 Pro. Official score column is explicitly dated Feb 2026. Originally announced as an upgrade to Gemini 3 Deep Think; no public stable API model ID or checkpoint hash was published." } }, { "id": "gemini-3.5-flash-lite", "name": "Gemini 3.5 Flash-Lite", "provider": "Google", "release_date": "2026-07-21", "params_total_M": null, "params_active_M": null, "architecture": null, "is_reasoning": true, "open_weights": false, "canonical_setting": { "mode": "thinking", "effort": "minimal", "tools": "none (per-benchmark override)", "sampling": "pass@1 (per-benchmark override)", "judge": "benchmark-specified", "harness": "official", "prompt_style": "default", "temperature": "default", "context": "1,048,576 input; 65,536 output", "notes": "Stable API ID gemini-3.5-flash-lite. Production default is minimal thinking; the July 2026 official benchmark table uses high thinking and therefore records noncanonical per-cell settings." } }, { "id": "gemini-3.1-flash-lite", "name": "Gemini 3.1 Flash-Lite", "provider": "Google", "release_date": "2026-03-03", "params_total_M": null, "params_active_M": null, "architecture": null, "is_reasoning": true, "open_weights": false, "canonical_setting": { "mode": "thinking", "effort": "minimal", "tools": "none (per-benchmark override)", "sampling": "pass@1 (per-benchmark override)", "judge": "benchmark-specified", "harness": "official", "prompt_style": "default", "temperature": "default", "context": "1,048,576 input; 65,536 output", "notes": "Stable API ID gemini-3.1-flash-lite; natively multimodal and based on Gemini 3 Pro. Official comparison scores use high thinking." } }, { "id": "gpt-5.4-mini", "name": "GPT-5.4 mini", "provider": "OpenAI", "release_date": "2026-03-17", "params_total_M": null, "params_active_M": null, "architecture": null, "is_reasoning": true, "open_weights": false, "canonical_setting": { "mode": "reasoning", "effort": "none", "tools": "none (per-benchmark override)", "sampling": "pass@1 (per-benchmark override)", "judge": "benchmark-specified", "harness": "official", "prompt_style": "default", "temperature": "default", "context": "400,000 context; 272,000 max input; 128,000 max output", "notes": "Stable ID gpt-5.4-mini; default snapshot gpt-5.4-mini-2026-03-17. Reasoning effort supports none (default), low, medium, high and xhigh. Text/image input." } }, { "id": "gemma-4-12b", "name": "Gemma 4 12B", "provider": "Google", "release_date": "2026-06-03", "params_total_M": 11950, "params_active_M": 11950, "architecture": "Dense unified multimodal decoder-only", "open_weights": true, "is_reasoning": true, "canonical_setting": { "mode": "thinking", "effort": "default", "tools": "none (per-benchmark override)", "sampling": "provider-reported (per-benchmark override)", "judge": "benchmark-specified", "harness": "official provider evaluation", "prompt_style": "official chat template", "temperature": "1.0; top_p=0.95; top_k=64", "context": "262,144", "notes": "Canonical checkpoint google/gemma-4-12B-it; base checkpoint google/gemma-4-12B has no separately reported benchmark scores. 48 layers, 1,024-token sliding window, 262K vocabulary; direct text/image/audio input and text output." } }, { "id": "mimo-v2.5-pro", "name": "MiMo-V2.5-Pro", "provider": "Xiaomi", "release_date": "2026-04-27", "params_total_M": 1023244.718976, "params_active_M": 42000, "architecture": "Sparse MoE with hybrid sliding/global attention; 70 layers, 384 routed experts, and 3 MTP layers", "is_reasoning": true, "open_weights": true, "canonical_setting": { "mode": "thinking", "effort": "default", "tools": "none (per-benchmark override)", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official Xiaomi MiMo / benchmark-specific", "prompt_style": "official model chat template", "temperature": "1.0; top_p=0.95", "context": "1,048,576", "input_modalities": "text", "trials": "benchmark-specific", "notes": "Official XiaomiMiMo/MiMo-V2.5-Pro checkpoint at revision 21d1ecfecd7bd70f31be25ca49d7edd21f003659. Exact total parameter count is 1,023,244,718,976; 42B are active." } }, { "id": "deepseek-v4-pro-non-thinking", "name": "DeepSeek-V4-Pro Non-Thinking", "provider": "DeepSeek", "release_date": "2026-04-22", "params_total_M": 1600000, "params_active_M": 49000, "architecture": "MoE hybrid CSA/HCA with mHC", "open_weights": true, "is_reasoning": false, "canonical_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none (per-benchmark override)", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official DeepSeek", "prompt_style": "official non-think mode", "temperature": "default", "context": "1,048,576", "notes": "Same released deepseek-ai/DeepSeek-V4-Pro weights as the thinking row, fixed to the official Non-think mode." } }, { "id": "deepseek-v3.2-exp-thinking", "name": "DeepSeek-V3.2-Exp Thinking", "provider": "DeepSeek", "release_date": "2025-09-29", "params_total_M": 671000, "params_active_M": 37000, "architecture": "DSA MoE", "open_weights": true, "is_reasoning": true, "canonical_setting": { "mode": "thinking", "effort": "default", "tools": "none (per-benchmark override)", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official DeepSeek", "prompt_style": "official reasoner", "temperature": "0.6; top_p=0.95", "context": "163,840", "notes": "Distinct experimental checkpoint deepseek-ai/DeepSeek-V3.2-Exp, not the later DeepSeek-V3.2 release." } }, { "id": "qwen3.5-122b-a10b", "name": "Qwen3.5-122B-A10B", "provider": "Alibaba", "release_date": "2026-02-24", "params_total_M": 122000, "params_active_M": 10000, "architecture": "Unified multimodal Gated DeltaNet/attention MoE", "open_weights": true, "is_reasoning": true, "canonical_setting": { "mode": "thinking", "effort": "default", "tools": "none (per-benchmark override)", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official Qwen qwen3 reasoning parser", "prompt_style": "official", "temperature": "1.0; top_p=0.95; top_k=20; presence_penalty=1.5", "context": "262,144 native", "notes": "Official checkpoint Qwen/Qwen3.5-122B-A10B; YaRN extension supports about 1M tokens." } }, { "id": "qwen3.5-27b", "name": "Qwen3.5-27B", "provider": "Alibaba", "release_date": "2026-02-24", "params_total_M": 27000, "params_active_M": 27000, "architecture": "Dense unified multimodal Gated DeltaNet/attention", "open_weights": true, "is_reasoning": true, "canonical_setting": { "mode": "thinking", "effort": "default", "tools": "none (per-benchmark override)", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official Qwen qwen3 reasoning parser", "prompt_style": "official", "temperature": "1.0; top_p=0.95; top_k=20; presence_penalty=1.5", "context": "262,144 native", "notes": "Official checkpoint Qwen/Qwen3.5-27B; YaRN extension supports about 1M tokens." } }, { "id": "diffusiongemma-26b-a4b", "name": "DiffusionGemma 26B A4B", "provider": "Google", "release_date": "2026-06-10", "params_total_M": 25200, "params_active_M": 3850, "architecture": "MoE encoder-decoder block text-diffusion with autoregressive fallback", "open_weights": true, "is_reasoning": true, "canonical_setting": { "mode": "thinking", "effort": "default", "decoding": "TD", "sampler": "entropy-bounded adaptive sampler", "max_denoising_steps": 48, "temperature": "0.8 -> 0.4 linear", "entropy_bound": 0.1, "stop_threshold": 0.005, "canvas": 256, "context": "262,144", "tools": "none (per-benchmark override)", "sampling": "provider-reported (replicate count per benchmark)", "judge": "benchmark-specified", "harness": "official Google evaluation", "prompt_style": "official chat template", "notes": "Canonical checkpoint google/diffusiongemma-26B-A4B-it. No separate base checkpoint registration; the released weights retain AR fallback." } }, { "id": "llada-2.1-flash-100b", "name": "LLaDA2.1-Flash 100B", "provider": "inclusionAI", "release_date": "2026-02-09", "params_total_M": 102890, "params_active_M": null, "architecture": "MoE block-diffusion", "open_weights": true, "is_reasoning": false, "canonical_setting": { "mode": "non-thinking", "effort": "n/a", "decoding": "TD S Mode", "threshold": 0.5, "editing_threshold": 0.0, "block_length": 32, "temperature": 0.0, "max_post_steps": 16, "context": "32,768", "tools": "none (per-benchmark override)", "sampling": "provider-reported", "judge": "benchmark-specified", "harness": "official provider implementation", "prompt_style": "official chat template", "notes": "Official checkpoint inclusionAI/LLaDA2.1-flash. Exact safetensor parameter total is 102,889.705216M; integer metadata rounds to 102,890M. Official chat template defaults detailed thinking off." } }, { "id": "nemotron-diffusion-14b", "name": "Nemotron-Labs-Diffusion 14B", "provider": "NVIDIA", "release_date": "2026-05-19", "params_total_M": 13506, "params_active_M": 13506, "architecture": "Dense tri-mode Transformer", "open_weights": true, "is_reasoning": false, "canonical_setting": { "mode": "non-thinking", "effort": "n/a", "decoding": "TD Diffusion Mode", "block_length": 32, "threshold": 0.9, "context": "262,144", "tools": "none (per-benchmark override)", "sampling": "provider-reported", "judge": "benchmark-specified", "harness": "official provider implementation", "prompt_style": "official chat template", "temperature": "provider default", "notes": "Official checkpoint nvidia/Nemotron-Labs-Diffusion-14B. One row covers AR, dLM, and self-speculation; the official chat template defaults thinking off." } }, { "id": "mercury-2", "name": "Mercury 2", "provider": "Inception Labs", "release_date": "2026-02-24", "params_total_M": null, "params_active_M": null, "architecture": "Proprietary diffusion Transformer", "open_weights": false, "is_reasoning": true, "canonical_setting": { "mode": "thinking", "decoding": "diffusion", "effort": "medium", "temperature": 0.75, "context": "128,000", "tools": "none (per-benchmark override)", "sampling": "provider-reported", "judge": "benchmark-specified", "harness": "official Inception API", "prompt_style": "official chat template", "notes": "One model row. Medium is the provider default/canonical effort; High is an alternate effort setting." } }, { "id": "grok-4.3", "name": "Grok 4.3", "provider": "xAI / SpaceXAI", "release_date": null, "params_total_M": null, "params_active_M": null, "architecture": null, "is_reasoning": true, "open_weights": false, "canonical_setting": { "mode": "thinking", "effort": "low", "tools": "none (per-benchmark override)", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official SpaceXAI API or benchmark-specified", "prompt_style": "unknown", "temperature": "unknown", "context": "1,000,000", "notes": "Official API identity grok-4.3 at https://docs.x.ai/developers/models/grok-4.3; alias grok-4.3-latest. The official API docs expose defaultEffort=low. The xAI score sources report high or omit effort; those settings are noncanonical." } }, { "id": "swe-1.7", "name": "SWE-1.7", "provider": "Cognition", "release_date": "2026-07-08", "params_total_M": null, "params_active_M": null, "architecture": null, "is_reasoning": true, "open_weights": false, "canonical_setting": { "mode": "agentic/reasoning", "effort": "unknown", "tools": "Devin coding tools", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "Devin", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown; self-compaction", "notes": "Cognition release source: https://cognition.com/blog/swe-1-7. The model is trained from a Kimi K2.7 base and used through Devin. Context length is not disclosed; Devin self-compacts long trajectories. The xAI FrontierCode score omits effort and is therefore noncanonical." } }, { "id": "doubao-seed-2.1-turbo", "name": "Doubao Seed 2.1 Turbo", "provider": "ByteDance", "release_date": "2026-06-23", "params_total_M": null, "params_active_M": null, "architecture": null, "is_reasoning": true, "open_weights": false, "canonical_setting": { "mode": "thinking", "effort": "unknown", "tools": "none (per-benchmark override)", "sampling": "pass@1 unless benchmark specifies otherwise", "judge": "benchmark-specified", "harness": "official or benchmark-specified", "prompt_style": "official default", "temperature": "unknown", "context": "unknown", "notes": "Per the official Seed2.1 release and model card. Turbo is the distinct efficient/high-throughput family member; reasoning effort, topology, parameter count, and API context are undisclosed." } }, { "id": "doubao-seed-2.1-deep-think", "name": "Doubao Seed 2.1 Deep Think", "provider": "ByteDance", "release_date": "2026-06-23", "params_total_M": null, "params_active_M": null, "architecture": null, "is_reasoning": true, "open_weights": false, "canonical_setting": { "mode": "deep-think", "effort": "unknown", "tools": "web search + sandboxed code execution", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "reason -> verify -> revise -> select loop", "prompt_style": "official default", "temperature": "unknown", "context": "unknown", "notes": "The official model card presents Seed2.1 Deep Think as a separate inference-time configuration and comparison column." } }, { "id": "doubao-seed-2.1-pro-preview", "name": "Doubao Seed 2.1 Pro Preview", "provider": "ByteDance", "release_date": "2026-06-23", "params_total_M": null, "params_active_M": null, "architecture": null, "is_reasoning": true, "open_weights": false, "canonical_setting": { "mode": "thinking", "effort": "unknown", "tools": "none", "sampling": "Arena leaderboard aggregation", "judge": "human preference", "harness": "Code Arena: Frontend", "prompt_style": "Arena default", "temperature": "unknown", "context": "unknown", "notes": "The release blog labels the leaderboard row Seed-2.1-Pro (Preview). No official mapping to the released Pro or Turbo checkpoint is published; 2026-06-23 is the public disclosure date, not a claimed availability date." } }, { "id": "longcat-flash-lite-sparse", "name": "LongCat-Flash-Lite-Sparse", "provider": "Meituan", "release_date": "2026-07-31", "params_total_M": 69000, "params_active_M": 3000, "architecture": "MoE with LongCat Sparse Attention, cross-layer indexing, hierarchical indexing, 256 routed experts and top-12 routing", "is_reasoning": false, "open_weights": true, "canonical_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none (per-benchmark override)", "sampling": "pass@1 unless benchmark specifies otherwise", "judge": "benchmark-specified", "harness": "official or benchmark-specified", "prompt_style": "official default", "temperature": "official default", "context": "983040 config; advertised up to 1M", "notes": "Official 69B-A3B non-thinking checkpoint. Hierarchical Indexing (HI) is an inference setting of this same checkpoint, not a separate model. Exact safetensor count is 69,127,158,912. The card advertises 1M while config max_position_embeddings is 983,040." } }, { "id": "longcat-flash-lite", "name": "LongCat-Flash-Lite", "provider": "Meituan", "release_date": "2026-01-27", "params_total_M": 69000, "params_active_M": 3000, "architecture": "MoE with dense MLA and N-gram embedding; 256 routed experts and top-12 routing", "is_reasoning": false, "open_weights": true, "canonical_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none (per-benchmark override)", "sampling": "pass@1 unless benchmark specifies otherwise", "judge": "benchmark-specified", "harness": "official or benchmark-specified", "prompt_style": "official default", "temperature": "official default", "context": "327680 config; evaluated through 512K in the sparse-attention paper", "notes": "Dense Lite comparator in the official sparse-attention paper. Exact safetensor count is 69,073,335,552. It is distinct from the existing 560B LongCat-Flash-Chat model." } }, { "id": "intern-s2-preview-35b", "name": "Intern-S2-Preview-35B", "provider": "Shanghai AI Laboratory", "release_date": "2026-05-15", "params_total_M": 35000, "params_active_M": 3000, "architecture": "Multimodal MoE derived from Qwen3.5-MoE: 40 text layers, 256 routed experts, top-8 plus a shared expert, hybrid linear/full attention, vision and time-series encoders", "is_reasoning": true, "open_weights": true, "canonical_setting": { "mode": "thinking", "effort": "default", "tools": "none (per-benchmark override; SWE is agentic)", "sampling": "pass@1 unless benchmark-specific", "judge": "benchmark-specified", "harness": "OpenCompass and VLMEvalKit per model card; AgentCompass/Mini-SWE-Agent for identified unified SWE alternatives", "prompt_style": "official chat template", "temperature": "recommended 0.8; top_p=0.95; top_k=50; min_p=0.0", "context": "262144; source evaluation max 128K text / 64K multimodal", "notes": "Header reports 35B-A3B; HF safetensors total is 36,098,254,656. Thinking is enabled by default. Exact per-row sampling/repeats are not stated." } }, { "id": "intern-s1-pro", "name": "Intern-S1-Pro", "provider": "Shanghai AI Laboratory", "release_date": "2026-02-02", "params_total_M": 1000000, "params_active_M": 22000, "architecture": "Multimodal MoE: 94 text layers, 512 experts, top-8, vision and time-series encoders; FP8 open checkpoint", "is_reasoning": true, "open_weights": true, "canonical_setting": { "mode": "thinking", "effort": "default", "tools": "none (per-benchmark override)", "sampling": "source-reported; exact repeats not stated in matrix", "judge": "benchmark-specified", "harness": "OpenCompass/VLMEvalKit in Intern-S2 comparison", "prompt_style": "official chat template", "temperature": "not stated for matrix", "context": "262144", "notes": "Header reports 1T-A22B. Release date uses official HF repository creation date." } }, { "id": "qwen3.6-35b-a3b", "name": "Qwen3.6-35B-A3B", "provider": "Alibaba", "release_date": "2026-04-15", "params_total_M": 35000, "params_active_M": 3000, "architecture": "Vision-capable MoE: 40 text layers, 256 routed experts, top-8 plus one shared expert, hybrid linear/full attention", "is_reasoning": true, "open_weights": true, "canonical_setting": { "mode": "thinking", "effort": "default", "tools": "none (per-benchmark override; SWE is agentic)", "sampling": "pass@1 unless benchmark-specific", "judge": "benchmark-specified", "harness": "official provider or source-table framework", "prompt_style": "official chat template", "temperature": "general thinking: 1.0; top_p=0.95; top_k=20; min_p=0.0; presence_penalty=1.5", "context": "262144 native; extensible to 1010000", "notes": "Header reports 35B-A3B; HF safetensors total is 35,951,822,704. SWE 73.4 is present in the official model card; exact SWE run count is not stated there." } }, { "id": "step-3.5-flash", "name": "Step 3.5 Flash", "provider": "StepFun", "release_date": "2026-02-01", "params_total_M": 199384.301376, "params_active_M": 11000, "architecture": "Decoder-only MoE: 45 layers, 288 experts, top-8 plus shared expert, hybrid attention", "is_reasoning": true, "open_weights": true, "canonical_setting": { "mode": "thinking", "effort": "default", "tools": "none (per-benchmark override; SWE is agentic)", "sampling": "pass@1; benchmark-specific repeated generations", "judge": "benchmark-specified", "harness": "official provider evaluation", "prompt_style": "official chat template", "temperature": "benchmark-specific; SWE official report uses temperature=1, top_p=0.95", "context": "262144", "notes": "Header reports 196B-A11B; HF safetensors total is 199,384,301,376. Paper averages 8 generations/problem for IMO-AnswerBench and 4 runs for SWE." } }, { "id": "gpt-5.4-nano", "name": "GPT-5.4 nano", "provider": "OpenAI", "release_date": "2026-03-17", "params_total_M": null, "params_active_M": null, "architecture": null, "is_reasoning": true, "open_weights": false, "canonical_setting": { "mode": "reasoning", "effort": "none (default); low, medium, high, xhigh supported", "tools": "none (per-benchmark override)", "sampling": "pass@1 (per-benchmark override)", "judge": "benchmark-specified", "harness": "official/source-table framework", "prompt_style": "default", "temperature": "default", "context": "400000; max input 272000; max output 128000", "notes": "Stable ID gpt-5.4-nano; immutable default snapshot gpt-5.4-nano-2026-03-17; text and image input." } }, { "id": "ernie-5.1", "name": "ERNIE 5.1", "provider": "Baidu", "release_date": "2026-05-09", "params_total_M": null, "params_active_M": null, "architecture": "MoE with elastic depth, expert-pool width, and variable Top-k sparsity; exact parameter counts undisclosed", "is_reasoning": true, "open_weights": false, "canonical_setting": { "mode": "thinking", "effort": "default", "tools": "none (per-benchmark override)", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "Baidu official or benchmark-specified", "prompt_style": "default", "temperature": "default", "context": "default", "notes": "Final ERNIE 5.1 release. Baidu does not disclose a public model card, exact checkpoint identifier, parameter counts, context window, or effort control. Keep distinct from ERNIE-5.1-Preview." } }, { "id": "ernie-5.1-preview", "name": "ERNIE-5.1-Preview", "provider": "Baidu", "release_date": "2026-04-30", "params_total_M": null, "params_active_M": null, "architecture": "Derived from ERNIE 5.0; public preview prose reports asynchronous reinforcement learning and scaled agentic post-training", "is_reasoning": true, "open_weights": false, "canonical_setting": { "mode": "thinking", "effort": "default", "tools": "none (per-benchmark override)", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "Baidu official or benchmark-specified", "prompt_style": "default", "temperature": "default", "context": "default", "notes": "Preview identity released on LMArena. Baidu never states that this checkpoint is identical to final ERNIE 5.1." } }, { "id": "mistral-medium-3.5-128b", "name": "Mistral Medium 3.5 128B", "provider": "Mistral", "release_date": "2026-04-28", "params_total_M": 128000, "params_active_M": 128000, "architecture": "Dense multimodal", "is_reasoning": true, "open_weights": true, "canonical_setting": { "mode": "thinking", "effort": "high", "tools": "none (per-benchmark override)", "sampling": "pass@1 (per-benchmark override)", "judge": "benchmark-specified", "harness": "official Mistral / benchmark-specific", "prompt_style": "official/default", "temperature": "0.7; top_p=0.95", "context": "256k", "notes": "Canonical complex-task profile uses reasoning_effort=high. The same checkpoint also supports reasoning_effort=none." } }, { "id": "mistral-small-4-119b-2603", "name": "Mistral Small 4 119B 2603", "provider": "Mistral", "release_date": "2026-03-16", "params_total_M": 119000, "params_active_M": 6500, "architecture": "MoE multimodal; 128 routed experts, 4 active", "is_reasoning": true, "open_weights": true, "canonical_setting": { "mode": "thinking", "effort": "high", "tools": "none (per-benchmark override)", "sampling": "pass@1 (per-benchmark override)", "judge": "benchmark-specified", "harness": "official Mistral / benchmark-specific", "prompt_style": "official/default", "temperature": "0.7", "context": "256k", "notes": "Canonical complex-task profile uses reasoning_effort=high. The same checkpoint also supports reasoning_effort=none. Official parameter descriptions vary between A6B, 6.5B active, 6B active plus embeddings/output, and chart A7B; 6.5B activated per token is the immutable docs value." } }, { "id": "magistral-medium-1.2", "name": "Magistral Medium 1.2", "provider": "Mistral", "release_date": "2025-09-18", "params_total_M": null, "params_active_M": null, "architecture": null, "is_reasoning": true, "open_weights": false, "canonical_setting": { "mode": "thinking", "effort": "default", "tools": "none (per-benchmark override)", "sampling": "pass@1 (benchmark repeat count preserved separately)", "judge": "benchmark-specified", "harness": "official Mistral API", "prompt_style": "official Magistral reasoning prompt", "temperature": "0.7; top_p=0.95 (Magistral 1.2 family recommendation)", "context": "128k", "notes": "Medium architecture/parameter count and a Medium-specific sampling sheet are undisclosed; sampling values are published on the official Magistral Small 1.2 card whose benchmark table includes Medium 1.2." } }, { "id": "magistral-small-1.2", "name": "Magistral Small 1.2", "provider": "Mistral", "release_date": "2025-09-18", "params_total_M": 24000, "params_active_M": 24000, "architecture": "Dense multimodal Mistral3", "is_reasoning": true, "open_weights": true, "canonical_setting": { "mode": "thinking", "effort": "default", "tools": "none (per-benchmark override)", "sampling": "pass@1 (benchmark repeat count preserved separately)", "judge": "benchmark-specified", "harness": "official Mistral reasoning parser / mistral-common", "prompt_style": "official Magistral reasoning system prompt", "temperature": "0.7; top_p=0.95; max_tokens=131072", "context": "128k", "notes": "Official model-specific configuration from locked follow-up sources." } }, { "id": "mistral-medium-3.1", "name": "Mistral Medium 3.1", "provider": "Mistral", "release_date": "2025-08-12", "params_total_M": null, "params_active_M": null, "architecture": null, "is_reasoning": false, "open_weights": false, "canonical_setting": { "mode": "non-thinking", "effort": "none", "tools": "none (per-benchmark override)", "sampling": "pass@1 (benchmark repeat count preserved separately)", "judge": "benchmark-specified", "harness": "official Mistral API", "prompt_style": "official/default", "temperature": "undisclosed Mistral API default", "context": "128k", "notes": "Proprietary parameter count and architecture are not disclosed in the immutable official model registry." } }, { "id": "mistral-small-3.2", "name": "Mistral Small 3.2", "provider": "Mistral", "release_date": "2025-06-20", "params_total_M": 24000, "params_active_M": 24000, "architecture": "Dense multimodal Mistral3", "is_reasoning": false, "open_weights": true, "canonical_setting": { "mode": "non-thinking", "effort": "none", "tools": "none (per-benchmark override)", "sampling": "pass@1 (benchmark repeat count preserved separately)", "judge": "benchmark-specified", "harness": "official mistral-common / vLLM", "prompt_style": "official SYSTEM_PROMPT.txt", "temperature": "0.15 recommended", "context": "128k", "notes": "Official model-specific configuration from locked follow-up sources." } }, { "id": "devstral-small-2", "name": "Devstral Small 2", "provider": "Mistral", "release_date": "2025-12-09", "params_total_M": 24000, "params_active_M": 24000, "architecture": "Dense multimodal Ministral-3-derived", "is_reasoning": false, "open_weights": true, "canonical_setting": { "mode": "non-thinking", "effort": "none", "tools": "agentic coding tools via benchmark scaffold", "sampling": "pass@1", "judge": "official benchmark verifier", "harness": "Mistral Vibe or benchmark-specified coding scaffold", "prompt_style": "official CHAT_SYSTEM_PROMPT.txt", "temperature": "0.2 release recommendation; immutable HF examples use 0.15", "context": "256k", "notes": "Record the 0.2 versus 0.15 official-source conflict; do not collapse agentic SWE-bench results into a tools=none setting." } }, { "id": "qwen3-next-80b-a3b-instruct", "name": "Qwen3-Next-80B-A3B-Instruct", "provider": "Alibaba", "release_date": "2025-09-11", "params_total_M": 80000, "params_active_M": 3000, "architecture": "MoE hybrid Gated DeltaNet / Gated Attention", "is_reasoning": false, "open_weights": true, "canonical_setting": { "mode": "non-thinking", "effort": "none", "tools": "none (per-benchmark override)", "sampling": "pass@1 (benchmark repeat count preserved separately)", "judge": "benchmark-specified", "harness": "official Qwen chat template", "prompt_style": "official; benchmark output-format prompts", "temperature": "temperature=0.7; top_p=0.8; top_k=20; min_p=0", "context": "262,144 native; approximately 1,010,000 with official extension", "notes": "This checkpoint supports non-thinking mode only; do not represent it as a switchable thinking checkpoint." } }, { "id": "qwen3-next-80b-a3b-thinking", "name": "Qwen3-Next-80B-A3B-Thinking", "provider": "Alibaba", "release_date": "2025-09-11", "params_total_M": 80000, "params_active_M": 3000, "architecture": "MoE hybrid Gated DeltaNet / Gated Attention", "is_reasoning": true, "open_weights": true, "canonical_setting": { "mode": "thinking", "effort": "default", "tools": "none (per-benchmark override)", "sampling": "pass@1 (benchmark repeat count preserved separately)", "judge": "benchmark-specified", "harness": "official Qwen thinking parser / chat template", "prompt_style": "official; benchmark output-format prompts", "temperature": "temperature=0.6; top_p=0.95; top_k=20; min_p=0", "context": "262,144 native; approximately 1,010,000 with official extension", "notes": "This checkpoint supports thinking mode only; recommended max output is 32,768 normally and 81,920 for complex math/programming benchmarks." } }, { "id": "lfm2.5-8b-a1b", "name": "LFM2.5-8B-A1B", "provider": "Liquid AI", "release_date": "2026-05-28", "params_total_M": 8300, "params_active_M": 1500, "architecture": "Hybrid MoE; 18 gated convolutions and 6 GQA layers", "is_reasoning": true, "open_weights": true, "canonical_setting": { "mode": "thinking", "effort": "default", "tools": "none (per-benchmark override)", "sampling": "benchmark-specific", "judge": "benchmark-specified", "harness": "official Liquid AI / benchmark-specific", "prompt_style": "official chat template", "temperature": "0.2; top_k=80; repetition_penalty=1.05", "context": "128,000", "notes": "Reasoning-only checkpoint; 38T-token training budget." } }, { "id": "lfm2.5-2.6b", "name": "LFM2.5-2.6B", "provider": "Liquid AI", "release_date": "2026-08-04", "params_total_M": 2690, "params_active_M": 2690, "architecture": "Hybrid dense; 22 gated convolutions and 8 GQA layers", "is_reasoning": true, "open_weights": true, "canonical_setting": { "mode": "thinking", "effort": "default", "tools": "none (per-benchmark override)", "sampling": "benchmark-specific", "judge": "benchmark-specified", "harness": "official Liquid AI / benchmark-specific", "prompt_style": "official chat template", "temperature": "0.1; top_k=50; repetition_penalty=1.1", "context": "131,072", "notes": "Pure reasoning checkpoint that always emits a think segment." } }, { "id": "lfm2.5-230m", "name": "LFM2.5-230M", "provider": "Liquid AI", "release_date": "2026-06-25", "params_total_M": 230, "params_active_M": 230, "architecture": "Hybrid dense; 8 gated convolutions and 6 GQA layers", "is_reasoning": false, "open_weights": true, "canonical_setting": { "mode": "non-thinking", "effort": "default", "tools": "none (per-benchmark override)", "sampling": "benchmark-specific", "judge": "benchmark-specified", "harness": "official Liquid AI / benchmark-specific", "prompt_style": "official chat template", "temperature": "0.1; top_k=50; repetition_penalty=1.05", "context": "32,768", "notes": "General-purpose instruction-tuned checkpoint." } }, { "id": "lfm2-8b-a1b", "name": "LFM2-8B-A1B", "provider": "Liquid AI", "release_date": "2025-10", "params_total_M": 8300, "params_active_M": 1500, "architecture": "Hybrid MoE", "is_reasoning": false, "open_weights": true, "canonical_setting": { "mode": "non-thinking", "effort": "default", "tools": "none (per-benchmark override)", "sampling": "benchmark-specific", "judge": "benchmark-specified", "harness": "official Liquid AI / benchmark-specific", "prompt_style": "official chat template", "temperature": "source does not state", "context": "32,768", "notes": "Predecessor comparison checkpoint from the official Liquid table." } }, { "id": "granite-4.0-h-tiny", "name": "Granite 4.0 H Tiny", "provider": "IBM", "release_date": null, "params_total_M": 7000, "params_active_M": 1000, "architecture": "Hybrid MoE", "is_reasoning": false, "open_weights": true, "canonical_setting": { "mode": "source/model default", "effort": "default", "tools": "none (per-benchmark override)", "sampling": "benchmark-specific", "judge": "benchmark-specified", "harness": "official Liquid AI / benchmark-specific", "prompt_style": "official chat template", "temperature": "source does not state", "context": "source does not state", "notes": "Comparator identity established by the official Liquid table." } }, { "id": "qwen3.5-4b", "name": "Qwen3.5-4B", "provider": "Alibaba", "release_date": null, "params_total_M": 4700, "params_active_M": 4700, "architecture": "Dense", "is_reasoning": true, "open_weights": true, "canonical_setting": { "mode": "thinking", "effort": "default", "tools": "none (per-benchmark override)", "sampling": "benchmark-specific", "judge": "benchmark-specified", "harness": "official Liquid AI / benchmark-specific", "prompt_style": "official chat template", "temperature": "source does not state", "context": "source does not state", "notes": "Comparator identity established by the official Liquid table." } }, { "id": "qwen3.5-9b", "name": "Qwen3.5-9B", "provider": "Alibaba", "release_date": null, "params_total_M": 9700, "params_active_M": 9700, "architecture": "Dense", "is_reasoning": true, "open_weights": true, "canonical_setting": { "mode": "thinking", "effort": "default", "tools": "none (per-benchmark override)", "sampling": "benchmark-specific", "judge": "benchmark-specified", "harness": "official Liquid AI / benchmark-specific", "prompt_style": "official chat template", "temperature": "source does not state", "context": "source does not state", "notes": "Comparator identity established by the official Liquid table." } }, { "id": "qwen3.5-0.8b-instruct", "name": "Qwen3.5-0.8B Instruct", "provider": "Alibaba", "release_date": null, "params_total_M": 800, "params_active_M": 800, "architecture": "Dense", "is_reasoning": false, "open_weights": true, "canonical_setting": { "mode": "non-thinking", "effort": "default", "tools": "none (per-benchmark override)", "sampling": "benchmark-specific", "judge": "benchmark-specified", "harness": "official Liquid AI / benchmark-specific", "prompt_style": "official chat template", "temperature": "source does not state", "context": "source does not state", "notes": "Explicit Instruct comparator in the official Liquid table." } }, { "id": "qwen3-30b-a3b-thinking-2507", "name": "Qwen3-30B-A3B-Thinking-2507", "provider": "Alibaba", "release_date": "2025-07", "params_total_M": 30500, "params_active_M": 3300, "architecture": "MoE", "is_reasoning": true, "open_weights": true, "canonical_setting": { "mode": "thinking", "effort": "default", "tools": "none (per-benchmark override)", "sampling": "benchmark-specific", "judge": "benchmark-specified", "harness": "official Liquid AI / benchmark-specific", "prompt_style": "official chat template", "temperature": "source does not state", "context": "source does not state", "notes": "Dated Thinking checkpoint named by the official Liquid table." } }, { "id": "lfm2.5-350m", "name": "LFM2.5-350M", "provider": "Liquid AI", "release_date": null, "params_total_M": 350, "params_active_M": 350, "architecture": "Hybrid dense", "is_reasoning": false, "open_weights": true, "canonical_setting": { "mode": "source/model default", "effort": "default", "tools": "none (per-benchmark override)", "sampling": "benchmark-specific", "judge": "benchmark-specified", "harness": "official Liquid AI / benchmark-specific", "prompt_style": "official chat template", "temperature": "source does not state", "context": "source does not state", "notes": "Teacher/comparator checkpoint named by the official Liquid table." } }, { "id": "lfm2-350m", "name": "LFM2-350M", "provider": "Liquid AI", "release_date": null, "params_total_M": 350, "params_active_M": 350, "architecture": "Hybrid dense", "is_reasoning": false, "open_weights": true, "canonical_setting": { "mode": "non-thinking", "effort": "default", "tools": "none (per-benchmark override)", "sampling": "benchmark-specific", "judge": "benchmark-specified", "harness": "official Liquid AI / benchmark-specific", "prompt_style": "official chat template", "temperature": "source does not state", "context": "source does not state", "notes": "Predecessor comparator named by the official Liquid table." } }, { "id": "granite-4.0-h-350m", "name": "Granite 4.0 H 350M", "provider": "IBM", "release_date": null, "params_total_M": 350, "params_active_M": 350, "architecture": "Hybrid dense", "is_reasoning": false, "open_weights": true, "canonical_setting": { "mode": "source/model default", "effort": "default", "tools": "none (per-benchmark override)", "sampling": "benchmark-specific", "judge": "benchmark-specified", "harness": "official Liquid AI / benchmark-specific", "prompt_style": "official chat template", "temperature": "source does not state", "context": "source does not state", "notes": "Comparator identity established by the official Liquid table." } }, { "id": "granite-4.0-350m", "name": "Granite 4.0 350M", "provider": "IBM", "release_date": null, "params_total_M": 350, "params_active_M": 350, "architecture": "Dense", "is_reasoning": false, "open_weights": true, "canonical_setting": { "mode": "source/model default", "effort": "default", "tools": "none (per-benchmark override)", "sampling": "benchmark-specific", "judge": "benchmark-specified", "harness": "official Liquid AI / benchmark-specific", "prompt_style": "official chat template", "temperature": "source does not state", "context": "source does not state", "notes": "Comparator identity established by the official Liquid table." } }, { "id": "mai-thinking-1", "name": "MAI-Thinking-1", "provider": "Microsoft AI", "release_date": "2026-06-02", "params_total_M": 1000000, "params_active_M": 35000, "architecture": "Sparse Mixture-of-Experts transformer", "is_reasoning": true, "open_weights": false, "canonical_setting": { "mode": "thinking", "effort": "maximum available reasoning", "tools": "none; ReAct tools for agentic coding benchmarks", "sampling": "pass@1; benchmark-specific where stated", "judge": "benchmark-specified", "harness": "Microsoft official evaluation suite", "prompt_style": "official Microsoft evaluation prompts", "temperature": "source does not state globally", "context": "256K total context for agentic coding; 256K maximum output tokens for other headline evaluations", "notes": "35B active / approximately 1T total parameters. Headline score-table configuration from the technical report and official model/release pages." } }, { "id": "mai-code-1-flash", "name": "MAI-Code-1-Flash", "provider": "Microsoft AI", "release_date": "2026-06-02", "params_total_M": 137000, "params_active_M": 5000, "architecture": "Sparse Mixture-of-Experts transformer", "is_reasoning": true, "open_weights": false, "canonical_setting": { "mode": "adaptive reasoning", "effort": "adaptive solution length control", "tools": "GitHub Copilot / VS Code production tools", "sampling": "pass@1; benchmark-specific where stated", "judge": "benchmark-specified", "harness": "GitHub Copilot VS Code production harness", "prompt_style": "official Microsoft evaluation prompts", "temperature": "source does not state globally", "context": "256K", "notes": "Coding-focused 137B-total / 5B-active model. Core coding scores use the same production harness for both models." } }, { "id": "mimo-v2.5", "name": "MiMo-V2.5", "provider": "Xiaomi", "release_date": "2026-04-22", "params_total_M": 310775.04, "params_active_M": 15000, "architecture": "Sparse MoE with hybrid sliding/global attention, native vision and audio encoders, and 3 MTP layers", "is_reasoning": true, "open_weights": true, "canonical_setting": { "mode": "thinking", "effort": "default", "tools": "none (per-benchmark override)", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official Xiaomi MiMo / benchmark-specific", "prompt_style": "official model chat template", "temperature": "1.0; top_p=0.95", "context": "1,048,576", "input_modalities": "text, image, video, audio", "trials": "benchmark-specific", "notes": "Official XiaomiMiMo/MiMo-V2.5 checkpoint at revision 63651580ca774f8504f676040460aed3e1244ac1. Exact total parameter count is 310,775,040,000; 15B are active." } }, { "id": "mimo-v2-pro", "name": "MiMo-V2-Pro", "provider": "Xiaomi", "release_date": "2026-03-18", "params_total_M": null, "params_active_M": 42000, "architecture": "Trillion-parameter sparse model with 42B active parameters, 7:1 hybrid attention, and lightweight multi-token prediction", "is_reasoning": true, "open_weights": false, "canonical_setting": { "mode": "source does not state", "effort": "source does not state", "tools": "none (per-benchmark override)", "sampling": "source does not state", "judge": "benchmark-specified", "harness": "source-specific official evaluation", "prompt_style": "source does not state", "temperature": "source does not state", "context": "up to 1M", "input_modalities": "text", "trials": "source does not state", "notes": "Official Xiaomi MiMo-V2-Pro release page dated 2026-03-18. The page states more than 1T total parameters, 42B active parameters, and up to 1M-token context; it does not publish an exact total parameter count or one universal evaluation mode/effort." } }, { "id": "mimo-v2-omni", "name": "MiMo-V2-Omni", "provider": "Xiaomi", "release_date": "2026-03-18", "params_total_M": null, "params_active_M": null, "architecture": "Unified image, video, and audio encoders feeding a shared multimodal backbone with native tool calling and UI grounding", "is_reasoning": true, "open_weights": false, "canonical_setting": { "mode": "source does not state", "effort": "source does not state", "tools": "none (per-benchmark override)", "sampling": "source does not state", "judge": "benchmark-specified", "harness": "source-specific official evaluation", "prompt_style": "source does not state", "temperature": "source does not state", "context": "source does not state", "input_modalities": "text, image, video, audio", "trials": "source does not state", "notes": "Official Xiaomi MiMo-V2-Omni release page dated 2026-03-18. The page establishes the unified modalities and agentic interface but does not publish exact parameter counts, a context limit, or one universal evaluation mode/effort." } }, { "id": "ling-2.6-flash", "name": "Ling-2.6-flash", "provider": "InclusionAI", "release_date": "2026-04-28", "params_total_M": 107494.409216, "params_active_M": 7400, "architecture": "32-layer hybrid 1:7 MLA/Lightning-Attention sparse MoE; 256 routed experts plus 1 shared expert, 8 active per token", "is_reasoning": false, "open_weights": true, "canonical_setting": { "checkpoint": "post-trained", "mode": "instant/non-reasoning", "effort": "default", "shots": "benchmark-specific", "samples": "benchmark-specific", "trials": "benchmark-specific", "aggregation": "benchmark-specific", "tools": "none; benchmark-specific override", "search": "none; benchmark-specific override", "judge": "benchmark-specific", "harness": "official InclusionAI / benchmark-specific", "dataset_version": "benchmark-specific", "split": "benchmark-specific", "context_length": "131,072 tokens in locked post-trained HF config", "context_claim": "256K appears in the official architecture/report family evidence; it is preserved separately from the 131,072 config", "multimodal_inputs": false, "temperature": "benchmark-specific", "top_p": "benchmark-specific", "top_k": "benchmark-specific", "prompt_style": "official model chat template", "evaluation_provenance": "inclusionAI/Ling-2.6-flash revision 11236968749136a78d4b7cbfb786c4460f2f353e", "notes": "Locked official model identity: inclusionAI/Ling-2.6-flash revision 11236968749136a78d4b7cbfb786c4460f2f353e. Context boundary: 256K appears in the official architecture/report family evidence; it is preserved separately from the 131,072 config." } }, { "id": "ling-2.6-1t", "name": "Ling-2.6-1T", "provider": "InclusionAI", "release_date": "2026-04-29", "params_total_M": 1025657.871744, "params_active_M": 63000, "architecture": "80-layer hybrid 7:1 Lightning-Attention/MLA sparse MoE; 256 routed experts plus 1 shared expert, 8 active per token", "is_reasoning": false, "open_weights": true, "canonical_setting": { "checkpoint": "post-trained", "mode": "instant/fast-thinking", "effort": "default", "shots": "benchmark-specific", "samples": "benchmark-specific", "trials": "benchmark-specific", "aggregation": "benchmark-specific", "tools": "none; benchmark-specific override", "search": "none; benchmark-specific override", "judge": "benchmark-specific", "harness": "official InclusionAI / benchmark-specific", "dataset_version": "benchmark-specific", "split": "benchmark-specific", "context_length": "262,144 tokens in locked HF config", "context_claim": "256K native evaluation and deployment context", "multimodal_inputs": false, "temperature": "benchmark-specific", "top_p": "benchmark-specific", "top_k": "benchmark-specific", "prompt_style": "official model chat template", "evaluation_provenance": "inclusionAI/Ling-2.6-1T revision 07b9e26bb1f97494b4fd62eed83f22722ffd560a", "notes": "Locked official model identity: inclusionAI/Ling-2.6-1T revision 07b9e26bb1f97494b4fd62eed83f22722ffd560a. Context boundary: 256K native evaluation and deployment context." } }, { "id": "ring-2.6-1t", "name": "Ring-2.6-1T", "provider": "InclusionAI", "release_date": "2026-05-14", "params_total_M": 1025657.871744, "params_active_M": 63000, "architecture": "80-layer hybrid sparse MoE shared with the Ling-2.6-1T backbone; reasoning post-training with high/xhigh effort", "is_reasoning": true, "open_weights": true, "canonical_setting": { "checkpoint": "post-trained", "mode": "thinking", "effort": "high (production default); xhigh optional", "shots": "benchmark-specific", "samples": "benchmark-specific", "trials": "benchmark-specific", "aggregation": "benchmark-specific", "tools": "none; benchmark-specific override", "search": "none; benchmark-specific override", "judge": "benchmark-specific", "harness": "official InclusionAI / benchmark-specific", "dataset_version": "benchmark-specific", "split": "benchmark-specific", "context_length": "131,072 tokens in locked HF config", "context_claim": "128K native to 256K via YaRN, as stated in the official card", "multimodal_inputs": false, "temperature": "benchmark-specific", "top_p": "benchmark-specific", "top_k": "benchmark-specific", "prompt_style": "official model chat template", "evaluation_provenance": "inclusionAI/Ring-2.6-1T revision 1e58be9318352541575130d4dbbdcc735fca7a03", "notes": "Locked official model identity: inclusionAI/Ring-2.6-1T revision 1e58be9318352541575130d4dbbdcc735fca7a03. Context boundary: 128K native to 256K via YaRN, as stated in the official card." } }, { "id": "ling-3.0-flash", "name": "Ling-3.0-flash", "provider": "InclusionAI", "release_date": "2026-08-02", "params_total_M": 127486.4056, "params_active_M": 5100, "architecture": "42-layer native hybrid MoE: 35 KDA plus 7 gated MLA layers; 512 routed experts plus 1 shared expert, 8 active per token", "is_reasoning": true, "open_weights": true, "canonical_setting": { "checkpoint": "post-trained", "mode": "thinking enabled by default", "effort": "default", "shots": "benchmark-specific", "samples": "benchmark-specific", "trials": "benchmark-specific", "aggregation": "benchmark-specific", "tools": "none; benchmark-specific override", "search": "none; benchmark-specific override", "judge": "benchmark-specific", "harness": "official InclusionAI / benchmark-specific", "dataset_version": "benchmark-specific", "split": "benchmark-specific", "context_length": "262,144 tokens in locked HF config and 256K native evaluation", "context_claim": "official architecture image states 1M supported content length; this architecture claim is not a deployment recipe and is not collapsed into the 256K config/eval identity", "multimodal_inputs": false, "temperature": "benchmark-specific", "top_p": "benchmark-specific", "top_k": "benchmark-specific", "prompt_style": "official model chat template", "evaluation_provenance": "inclusionAI/Ling-3.0-flash revision ecde16176a497adaff7419ff4de59da603c4edaa", "notes": "Locked official model identity: inclusionAI/Ling-3.0-flash revision ecde16176a497adaff7419ff4de59da603c4edaa. Context boundary: official architecture image states 1M supported content length; this architecture claim is not a deployment recipe and is not collapsed into the 256K config/eval identity." } }, { "id": "llada2.2-flash", "name": "LLaDA2.2-flash", "provider": "InclusionAI", "release_date": "2026-07-16", "params_total_M": 102889.705216, "params_active_M": null, "architecture": "32-layer MoE diffusion language model with block routing and Levenshtein DELETE/INSERT editing", "is_reasoning": false, "open_weights": true, "canonical_setting": { "checkpoint": "post-trained", "mode": "block-diffusion with Levenshtein editing", "effort": "default", "shots": "benchmark-specific", "samples": "benchmark-specific", "trials": "benchmark-specific", "aggregation": "benchmark-specific", "tools": "none; benchmark-specific override", "search": "none; benchmark-specific override", "judge": "benchmark-specific", "harness": "official InclusionAI / benchmark-specific", "dataset_version": "benchmark-specific", "split": "benchmark-specific", "context_length": "131,072 tokens in locked HF config", "context_claim": "official card states 128K; provider also reports 100B non-embedding parameters separately from physical count", "multimodal_inputs": false, "temperature": "benchmark-specific", "top_p": "benchmark-specific", "top_k": "benchmark-specific", "prompt_style": "official model chat template", "evaluation_provenance": "inclusionAI/LLaDA2.2-flash revision ee56264534721014d8e651293543f6dc3fcb1f9c", "notes": "Locked official model identity: inclusionAI/LLaDA2.2-flash revision ee56264534721014d8e651293543f6dc3fcb1f9c. Context boundary: official card states 128K; provider also reports 100B non-embedding parameters separately from physical count." } }, { "id": "step-3.7-flash", "name": "Step 3.7 Flash", "provider": "StepFun", "release_date": "2026-05-29", "params_total_M": 201365.31616, "params_active_M": 11000, "architecture": "Multimodal sparse MoE: 45-layer 196B language backbone, 288 routed experts with top-8 plus shared expert, 3:1 sliding/full attention, 3 MTP layers, and a 47-layer 1.8B vision encoder", "is_reasoning": true, "open_weights": true, "canonical_setting": { "mode": "reasoning level selectable; benchmark-specific", "effort": "benchmark-specific or source does not state", "tools": "none unless a benchmark-specific tool setting is reported", "sampling": "benchmark-specific", "judge": "benchmark-specified", "harness": "official StepFun or benchmark-specific", "prompt_style": "official Step 3.7 Flash chat template", "temperature": "benchmark-specific or source does not state", "top_p": "benchmark-specific or source does not state", "context": "262144", "output_cap": "source does not state globally", "input_modalities": "text and image", "notes": "Official BF16 checkpoint revision 5f6244077ac62e04eec3f320501ff8c2b293373a. Hugging Face safetensors reports exactly 201,365,316,160 physical parameters. The release rounds this to 198B: 196B language plus 1.8B vision, with approximately 11B active per token. The model supports low, medium, and high reasoning, native image input, tool calling, 262,144-token context, and three MTP draft layers. FP8 revision b3d7916fccac844cca050d7520f2aaa513f9a84f, NVFP4 revision 4275532ffd9a9496ff36b7a2dc4a9db1048da438, and GGUF revision 0b69336d2fd2adfdef9c66e425f7778196c31482 are deployment configurations of this same model identity." } }, { "id": "glm-5v-turbo", "name": "GLM-5V-Turbo", "provider": "Z.ai", "release_date": "2026-04-01", "params_total_M": null, "params_active_M": null, "architecture": "Proprietary multimodal coding foundation model with a CogViT vision encoder and inference-oriented MTP; parameter counts are not disclosed by the official release", "is_reasoning": true, "open_weights": false, "canonical_setting": { "mode": "thinking mode available; exact default not stated", "effort": "multiple thinking modes; benchmark-specific", "tools": "function calling and multimodal agent tools supported", "sampling": "benchmark-specific", "judge": "benchmark-specified", "harness": "official Z.ai or benchmark-specific", "prompt_style": "official Z.ai API", "temperature": "source does not state globally", "top_p": "source does not state globally", "context": "200000", "output_cap": "128000", "input_modalities": "text, image, video, and file", "notes": "Official Z.ai release notes date GLM-5V-Turbo to 2026-04-01. The official model documentation defines a 200K context, 128K maximum output, native multimodal input, thinking modes, function calling, and a CogViT vision encoder. No public parameter count or open-weight checkpoint is claimed." } }, { "id": "muse-spark-1.2", "name": "Muse Spark 1.2", "provider": "Meta", "release_date": "2026-08-05", "params_total_M": null, "params_active_M": null, "architecture": null, "is_reasoning": true, "open_weights": false, "canonical_setting": { "mode": "thinking", "effort": "xhigh", "tools": "benchmark-specified", "sampling": "pass@1 with source-defined repeated attempts", "judge": "benchmark-specified", "harness": "benchmark-specific; Muse Code for release coding evaluations", "prompt_style": "not disclosed", "temperature": "not disclosed", "context": "not disclosed", "notes": "Official release evaluations use xhigh when Meta controls effort. Muse Spark always reasons; minimal, low, medium, high, and xhigh are supported, omitted effort is model-determined, and none returns HTTP 400. The 1,048,576-token figure is model capacity, not a disclosed score-level run context." }, "notes": "Official API IDs: muse-spark-1.2 (Standard) and muse-spark-1.2-contributor (same checkpoint, Contributor tier). The model accepts text, image, video, audio, and PDF and emits text. The docs table omits audio while adjacent official prose includes it; the conflict is preserved. Architecture, total parameters, and active parameters are not disclosed.", "model_capacity_tokens": 1048576 } ], "benchmarks": [ { "id": "aa_intelligence_index", "name": "AA Intelligence Index", "category": "Composite", "metric": "index score", "num_problems": 12826, "source_url": "https://artificialanalysis.ai/methodology/intelligence-benchmarking", "canonical_setting": { "version": "Artificial Analysis Intelligence Index v4.0.4 (March 2026)", "metric_type": "index", "range": null, "higher_is_better": true, "multimodal_input": false, "tools": "composite", "sampling": "included in num_problems", "judge": "mixed scoring protocols", "notes": "Composite weighted index over 10 evaluations. Count is actual model generations across official questions/tasks and repeats: GDPval-AA 220*1, tau2-Bench Telecom 114*3, Terminal-Bench Hard 44*3, SciCode 288*3, AA-LCR 100*3, AA-Omniscience 6000*1, IFBench 294*5, HLE text-only 2158*1, GPQA Diamond 198*5, CritPt 70*5 = 12826. Cost burden is heterogeneous; tools=composite intentionally avoids applying one agentic multiplier to every component." }, "cost": { "source_id": "artificial_analysis_eval_token_cost_gemini_2_5_pro_anchor", "source_name": "Artificial Analysis per-evaluation token usage and cost with Gemini 2.5 Pro anchor", "source_url": "https://artificialanalysis.ai/evaluations/artificial-analysis-intelligence-index", "source_benchmark_name": "AA Intelligence Index", "source_model_name": "Gemini 2.5 Pro", "source_model_slug": "gemini-2-5-pro", "evidence_scope": "observed model-inference totals for the source evaluation run", "tokens": { "prompt_tokens": null, "completion_tokens": null, "input_tokens": 81649130, "output_tokens": 54625405, "reasoning_tokens": 49162377, "answer_tokens": 5463027, "total_tokens": 136274535 }, "total_cost_usd": 648.3154625, "reported_items": null, "reported_samples": null, "relative_tokens_to_source_anchor": 84.1, "relative_cost_to_source_anchor": 41.097, "notes": "Inference-side token/cost evidence only; source_model_* identifies the reported model row. Dollar cost is run/model specific, not benchmark-intrinsic. primary anchor model: Gemini 2.5 Pro" } }, { "id": "aa_lcr", "name": "AA Long Context Reasoning", "category": "Long Context", "metric": "% correct", "num_problems": 300, "source_url": "https://artificialanalysis.ai/methodology/intelligence-benchmarking", "canonical_setting": { "version": "Artificial Analysis Long Context Reasoning, 100 questions", "metric_type": "pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": false, "tools": "none", "sampling": "3 repeats per question; pass@1 aggregated", "judge": "official AA equality checker", "harness": "official Artificial Analysis LCR", "notes": "Official AA-LCR has 100 open-answer questions over roughly 100k-token document contexts and runs three repeats, so num_problems records 300 physical model generations. StepFun's avg@16 observation is a score-level repeated-sampling setting." }, "cost": { "source_id": "artificial_analysis_eval_token_cost_gemini_2_5_pro_anchor", "source_name": "Artificial Analysis per-evaluation token usage and cost with Gemini 2.5 Pro anchor", "source_url": "https://artificialanalysis.ai/evaluations/artificial-analysis-long-context-reasoning", "source_benchmark_name": "Artificial Analysis Long Context Reasoning", "source_model_name": "Gemini 2.5 Pro", "source_model_slug": "gemini-2-5-pro", "evidence_scope": "observed model-inference totals for the source evaluation run", "tokens": { "prompt_tokens": null, "completion_tokens": null, "input_tokens": 9489998, "output_tokens": 179014, "reasoning_tokens": 0, "answer_tokens": 179014, "total_tokens": 9669012 }, "total_cost_usd": 13.6526375, "reported_items": null, "reported_samples": null, "relative_tokens_to_source_anchor": 5.967, "relative_cost_to_source_anchor": 0.865, "notes": "Inference-side token/cost evidence only; source_model_* identifies the reported model row. Dollar cost is run/model specific, not benchmark-intrinsic. primary anchor model: Gemini 2.5 Pro" } }, { "id": "aethercode", "name": "AetherCode", "category": "Coding", "metric": "%", "num_problems": null, "source_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "canonical_setting": { "version": "AetherCode", "metric_type": "pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": false, "tools": "none", "notes": "Per Doubao Seed 2.0 Pro model card." } }, { "id": "agentcompany", "name": "AgentCompany", "category": "Agentic", "metric": "%", "num_problems": null, "source_url": "https://huggingface.co/MiniMaxAI/MiniMax-M2", "canonical_setting": { "version": "AgentCompany", "metric_type": "pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": false, "tools": "agentic", "notes": "Per MiniMax M2 model card." }, "cost": { "source_id": "theagentcompany_paper_cost_per_instance", "source_name": "TheAgentCompany paper reported cost per instance", "source_url": "https://arxiv.org/abs/2412.14161", "source_data_url": "https://arxiv.org/html/2412.14161", "source_benchmark_name": "TheAgentCompany", "source_model_name": "OpenHands 0.28.1 + Gemini-2.5-Pro", "source_model_slug": "openhands-0.28.1; gemini-2.5-pro", "evidence_scope": "paper-reported average dollar cost per instance; derived full-pass total over 175 tasks", "tokens": { "prompt_tokens": null, "completion_tokens": null, "input_tokens": null, "output_tokens": null, "reasoning_tokens": null, "answer_tokens": null, "total_tokens": null }, "total_cost_usd": 735.0, "reported_items": 175, "reported_samples": 1, "relative_tokens_to_source_anchor": null, "relative_cost_to_source_anchor": null, "notes": "TheAgentCompany paper reports $4.2 average cost per instance for OpenHands 0.28.1 + Gemini-2.5-Pro. Full-pass total is 175 tasks * $4.2 = $735. Dollar-only; raw prompt/completion token counts are not public." } }, { "id": "ai2d", "name": "AI2D", "category": "Multimodal", "metric": "%", "num_problems": null, "source_url": "https://mistral.ai/news/mistral-medium-3", "canonical_setting": { "version": "AI2D", "metric_type": "pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": true, "tools": "none", "notes": "Per Mistral Medium 3 blog: AI2 Diagram understanding benchmark, 0-shot." } }, { "id": "aider_polyglot_diff", "name": "Aider Polyglot (diff mode)", "category": "Coding", "metric": "%", "num_problems": 450, "source_url": "https://aider.chat/2024/12/21/polyglot.html", "canonical_setting": { "version": "Aider Polyglot benchmark; diff edit format", "metric_type": "pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": false, "tools": "agentic", "sampling": "included in num_problems", "notes": "Aider Polyglot uses 225 selected Exercism coding tasks across C++, Go, Java, JavaScript, Python, and Rust. The displayed leaderboard score corresponds to the second-try/pass_rate_2 setting, so cost count records actual model generations: 225 tasks times two tries = 450. Diff mode is selected by edit_format=diff." }, "cost": { "source_id": "aider_polyglot_leaderboard_gpt5_high_diff", "source_name": "Aider Polyglot leaderboard reported token usage and cost", "source_url": "https://github.com/Aider-AI/aider/blob/main/aider/website/_data/polyglot_leaderboard.yml#L1715-L1742", "source_benchmark_name": "Aider Polyglot (diff mode)", "source_model_name": "gpt-5 (high)", "source_model_slug": "openai/gpt-5; reasoning_effort=high", "evidence_scope": "reported leaderboard run totals for one model row", "tokens": { "prompt_tokens": 2675561, "completion_tokens": 2623429, "input_tokens": 2675561, "output_tokens": 2623429, "reasoning_tokens": null, "answer_tokens": null, "total_tokens": 5298990 }, "total_cost_usd": 29.0829, "reported_items": 225, "reported_samples": null, "relative_tokens_to_source_anchor": null, "relative_cost_to_source_anchor": null, "notes": "Aider leaderboard row reports prompt_tokens and completion_tokens for the gpt-5 (high) diff-mode run. The displayed pass_rate_2 score uses the second-try setting; source row also reports total_tests=225 and user_asks=96. Dollar cost is model/run specific." } }, { "id": "aider_polyglot_whole", "name": "Aider Polyglot (whole mode)", "category": "Coding", "metric": "%", "num_problems": 450, "source_url": "https://aider.chat/2024/12/21/polyglot.html", "canonical_setting": { "version": "Aider Polyglot benchmark; whole edit format", "metric_type": "pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": false, "tools": "agentic", "sampling": "included in num_problems", "notes": "Aider Polyglot uses 225 selected Exercism coding tasks across C++, Go, Java, JavaScript, Python, and Rust. The displayed leaderboard score corresponds to the second-try/pass_rate_2 setting, so cost count records actual model generations: 225 tasks times two tries = 450. Whole mode is selected by edit_format=whole." }, "cost": { "source_id": "aider_polyglot_leaderboard_o1mini_whole_cost", "source_name": "Aider Polyglot leaderboard reported whole-mode cost", "source_url": "https://aider.chat/docs/leaderboards/", "source_data_url": "https://github.com/Aider-AI/aider/blob/main/aider/website/_data/polyglot_leaderboard.yml#L235-L259", "source_benchmark_name": "Aider Polyglot (whole mode)", "source_model_name": "o1-mini-2024-09-12", "source_model_slug": "openai/o1-mini; edit_format=whole", "evidence_scope": "reported leaderboard run cost for one whole-mode model row", "tokens": { "prompt_tokens": null, "completion_tokens": null, "input_tokens": null, "output_tokens": null, "reasoning_tokens": null, "answer_tokens": null, "total_tokens": null }, "total_cost_usd": 18.577, "reported_items": 225, "reported_samples": 2, "relative_tokens_to_source_anchor": null, "relative_cost_to_source_anchor": null, "notes": "Dollar-only evidence. Aider row reports total_cost=18.5770 for the o1-mini whole-mode run; token fields are not present for this older row. Score is pass_rate_2 over 225 tests, so BenchPress cost count treats this as the two-try setting." } }, { "id": "aime_2024", "name": "AIME 2024", "category": "Math", "metric": "% correct (pass@1)", "num_problems": 30, "source_url": "https://artofproblemsolving.com/wiki/index.php/2024_AIME", "canonical_setting": { "version": "AIME-2024-I+II (30 problems)", "metric_type": "pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": false, "notes": "tools=none preferred (pure-reasoning eval). If only with-tool scores (python/web/RAG) are available, accept and mark cell matches_canonical=false.", "tools": "none" } }, { "id": "aime_2025", "name": "AIME 2025", "category": "Math", "metric": "% correct (pass@1)", "num_problems": 30, "source_url": "https://artofproblemsolving.com/wiki/index.php/2025_AIME", "canonical_setting": { "version": "AIME-2025-I+II (30 problems)", "metric_type": "pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": false, "notes": "tools=none preferred (pure-reasoning eval). If only with-tool scores (python/web/RAG) are available, accept and mark cell matches_canonical=false.", "tools": "none" }, "cost": { "source_id": "artificial_analysis_eval_token_cost_gemini_2_5_pro_anchor", "source_name": "Artificial Analysis per-evaluation token usage and cost with Gemini 2.5 Pro anchor", "source_url": "https://artificialanalysis.ai/evaluations/aime-2025", "source_benchmark_name": "AIME 2025", "source_model_name": "Gemini 2.5 Pro", "source_model_slug": "gemini-2-5-pro", "evidence_scope": "observed model-inference totals for the source evaluation run", "tokens": { "prompt_tokens": null, "completion_tokens": null, "input_tokens": 4347, "output_tokens": 590235, "reasoning_tokens": 525629, "answer_tokens": 64606, "total_tokens": 594582 }, "total_cost_usd": 5.907783749999999, "reported_items": null, "reported_samples": null, "relative_tokens_to_source_anchor": 0.367, "relative_cost_to_source_anchor": 0.374, "notes": "Inference-side token/cost evidence only; source_model_* identifies the reported model row. Dollar cost is run/model specific, not benchmark-intrinsic. primary anchor model: Gemini 2.5 Pro" } }, { "id": "aime_2026", "name": "AIME 2026", "category": "Math", "metric": "% correct (pass@1)", "num_problems": 30, "source_url": "https://huggingface.co/datasets/MathArena/aime_2026", "canonical_setting": { "version": "AIME-2026-I+II (30 problems)", "metric_type": "pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": false, "notes": "tools=none preferred (pure-reasoning eval). If only with-tool scores (python/web/RAG) are available, accept and mark cell matches_canonical=false. Canonical row combines AIME 2026 I and II: 30 problems total.", "tools": "none" }, "cost": { "source_id": "matharena_aime_2026_outputs_deepseek_v3_2_think", "source_name": "MathArena AIME 2026 model outputs", "source_url": "https://huggingface.co/datasets/MathArena/aime_2026_outputs", "source_data_url": "https://huggingface.co/datasets/MathArena/aime_2026_outputs", "source_benchmark_name": "AIME 2026", "source_model_name": "DeepSeek-v3.2 (Think)", "source_model_slug": "deepseek-v3.2-think", "evidence_scope": "observed model-inference totals over public MathArena output rows", "tokens": { "prompt_tokens": null, "completion_tokens": null, "input_tokens": 17637, "output_tokens": 1782506, "reasoning_tokens": null, "answer_tokens": null, "total_tokens": 1800143 }, "total_cost_usd": 0.75359088, "reported_items": 30, "reported_samples": 4, "relative_tokens_to_source_anchor": null, "relative_cost_to_source_anchor": null, "notes": "MathArena output logs cover 120 attempts = 30 problems x 4 runs. Cost uses published per-token rates: 17,637 input tokens at $0.28/M plus 1,782,506 output tokens at $0.42/M = $0.75359088." } }, { "id": "ainstein_bench", "name": "AInsteinBench", "category": "Science Discovery", "metric": "%", "num_problems": null, "source_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "canonical_setting": { "version": "AInsteinBench", "metric_type": "pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": false, "tools": "none", "notes": "Per Doubao Seed 2.0 Pro model card." } }, { "id": "all_angles", "name": "All-Angles", "category": "Vision Spatial", "metric": "%", "num_problems": null, "source_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "canonical_setting": { "version": "All-Angles", "metric_type": "pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": true, "tools": "none", "notes": "Per Doubao Seed 2.0 Pro model card." } }, { "id": "alpacaeval_2", "name": "AlpacaEval 2.0 (LC-winrate)", "category": "Chat", "metric": "%", "num_problems": null, "source_url": "https://arxiv.org/abs/2501.12948", "canonical_setting": { "version": "AlpacaEval 2.0 (LC-winrate)", "metric_type": "pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": false, "tools": "none", "notes": "Per DS R1 paper." }, "cost": { "source_id": "alpacaeval_2_readme_weighted_gpt4_turbo_price", "source_name": "AlpacaEval README evaluator price table", "source_url": "https://raw.githubusercontent.com/tatsu-lab/alpaca_eval/main/README.md", "source_data_url": "https://huggingface.co/datasets/tatsu-lab/alpaca_eval/raw/main/alpaca_eval.json", "source_benchmark_name": "AlpacaEval 2.0", "source_model_name": "weighted_alpaca_eval_gpt4_turbo auto-annotator", "source_model_slug": "weighted_alpaca_eval_gpt4_turbo; baseline=gpt4_turbo", "evidence_scope": "automatic judge/annotation cost for AlpacaEval 2.0 evaluation set; excludes candidate model output generation", "tokens": { "prompt_tokens": null, "completion_tokens": null, "input_tokens": null, "output_tokens": null, "reasoning_tokens": null, "answer_tokens": null, "total_tokens": null }, "total_cost_usd": 4.4275, "reported_items": 805, "reported_samples": null, "relative_tokens_to_source_anchor": null, "relative_cost_to_source_anchor": null, "notes": "README reports evaluator price as $5.5 per 1000 examples and states AlpacaEval 2.0 costs less than $10/run. The public dataset has 805 examples; total_cost_usd is mechanically derived as 5.5 * 805 / 1000. This is judge-only cost, not candidate model generation cost." } }, { "id": "apex_agents", "name": "APEX-Agents", "category": "Agentic", "source_url": "https://deepmind.google/models/evals-methodology/gemini-3-pro", "num_problems": null, "canonical_setting": { "version": "APEX-Agents (long-horizon professional tasks)", "metric_type": "pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": false, "notes": "DeepMind APEX-Agents long-horizon professional benchmark. Distinct from MathArena Apex 2025." }, "cost": { "source_id": "artificial_analysis_eval_token_cost_gemini_2_5_pro_anchor", "source_name": "Artificial Analysis per-evaluation token usage and cost with Gemini 2.5 Pro anchor", "source_url": "https://artificialanalysis.ai/evaluations/apex-agents-aa", "source_benchmark_name": "APEX-Agents", "source_model_name": "Gemini 3 Flash Preview (Reasoning)", "source_model_slug": "gemini-3-flash-reasoning", "evidence_scope": "observed model-inference totals for the source evaluation run", "tokens": { "prompt_tokens": null, "completion_tokens": null, "input_tokens": 1125772887.3333333, "output_tokens": 11152637, "reasoning_tokens": 9216933.333333334, "answer_tokens": 1935703.6666666667, "total_tokens": 1136925524.3333333 }, "total_cost_usd": 596.3443546666666, "reported_items": null, "reported_samples": null, "relative_tokens_to_source_anchor": 701.641, "relative_cost_to_source_anchor": 37.802, "notes": "Inference-side token/cost evidence only; source_model_* identifies the reported model row. Dollar cost is run/model specific, not benchmark-intrinsic. fallback: Gemini 2.5 Pro has no APEX-Agents token row; Gemini 3 Flash Reasoning is present on this page" } }, { "id": "apex_shortlist", "name": "Apex Shortlist", "category": "Math", "metric": "% correct (pass@1)", "num_problems": null, "source_url": "https://matharena.ai/apex/", "canonical_setting": { "version": "Apex shortlist", "metric_type": "pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": false, "notes": "tools=none preferred (pure-reasoning eval). If only with-tool scores (python/web/RAG) are available, accept and mark cell matches_canonical=false.", "tools": "none" }, "cost": { "source_id": "matharena_apex_shortlist_outputs_gpt_5_2_high", "source_name": "MathArena Apex Shortlist 2025 model outputs", "source_url": "https://matharena.ai/apex/", "source_data_url": "https://huggingface.co/datasets/MathArena/apex_shortlist_outputs", "source_benchmark_name": "Apex Shortlist 2025", "source_model_name": "GPT-5.2 (high)", "source_model_slug": "gpt-5.2-high", "evidence_scope": "observed model-inference totals over public MathArena output rows", "tokens": { "prompt_tokens": null, "completion_tokens": null, "input_tokens": 198248, "output_tokens": 10502349, "reasoning_tokens": null, "answer_tokens": null, "total_tokens": 10700597 }, "total_cost_usd": 147.37982, "reported_items": 48, "reported_samples": 4, "relative_tokens_to_source_anchor": null, "relative_cost_to_source_anchor": null, "notes": "MathArena output logs cover 192 attempts = 48 problems x 4 runs for GPT-5.2 (high). Cost uses 198,248 input tokens at $1.75/M plus 10,502,349 output tokens at $14/M = $147.37982." } }, { "id": "arc_agi_1", "name": "ARC-AGI-1", "category": "Reasoning", "metric": "% correct", "num_problems": 400, "source_url": "https://arcprize.org/arc-agi/1/", "canonical_setting": { "version": "ARC-AGI-1 (semi-private 400)", "metric_type": "pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": false, "notes": "tools=none preferred (pure-reasoning eval). If only with-tool scores (python/web/RAG) are available, accept and mark cell matches_canonical=false.", "tools": "none" }, "cost": { "source_id": "arc_prize_o3_preview_arc_agi_1_compute_table", "source_name": "ARC Prize o3-preview ARC-AGI-1 compute table", "source_url": "https://arcprize.org/blog/oai-o3-pub-breakthrough", "source_benchmark_name": "ARC-AGI-1 Public / High compute", "evidence_scope": "observed model-inference totals for the reported o3-preview public high-compute run", "tokens": { "total_tokens": 111000000 }, "total_cost_usd": 66772.0, "reported_items": 400, "reported_samples": 6, "notes": "ARC Prize reports 400 public tasks, 6 samples, 111M tokens, and $66,772 total retail cost for this o3-preview high-compute run. This is not a benchmark-required fixed cost." } }, { "id": "arc_agi_2", "name": "ARC-AGI-2", "category": "Reasoning", "metric": "% correct", "num_problems": 120, "source_url": "https://arcprize.org/arc-agi/2/", "canonical_setting": { "version": "ARC-AGI-2 v2 semi-private evaluation set (120 tasks)", "metric_type": "pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": false, "notes": "ARC-AGI-2 v2 semi-private evaluation tier contains 120 calibrated tasks. Each task passes only when all test grids are exact; up to two outputs per test input are allowed. Tool/scaffold differences remain cell settings.", "tools": "none" }, "cost": { "source_id": "arcprize_arc_agi_2_leaderboard_cost", "source_name": "ARC Prize leaderboard data", "source_url": "https://arcprize.org/leaderboard", "source_data_url": "https://arcprize.org/media/data/leaderboard/v2.json", "source_benchmark_name": "ARC-AGI-2", "source_model_name": "o3 (High)", "source_model_slug": "o3-2025-04-16-high", "evidence_scope": "official leaderboard costPerTask for the source model row; derived full-benchmark total over 120 tasks", "tokens": { "prompt_tokens": null, "completion_tokens": null, "input_tokens": null, "output_tokens": null, "reasoning_tokens": null, "answer_tokens": null, "total_tokens": null }, "total_cost_usd": 100.068, "reported_items": 120, "reported_samples": 1, "relative_tokens_to_source_anchor": null, "relative_cost_to_source_anchor": null, "notes": "Official v2 leaderboard reports costPerTask=$0.8339 for o3 (High). Derived 120-task semi-private total is $0.8339*120=$100.068. Cost is model/run specific." } }, { "id": "arc_challenge", "name": "ARC Challenge", "category": "Reasoning", "source_url": "https://huggingface.co/datasets/allenai/ai2_arc/resolve/210d026faf9955653af8916fad021475a3f00453/README.md", "num_problems": 1172, "canonical_setting": { "version": "AI2 ARC-Challenge test split", "metric_type": "pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": false, "notes": "Official immutable AI2 dataset card reports 1,172 test questions. Few-shot count is observation-specific.", "tools": "none", "sampling": "one multiple-choice response per question", "judge": "answer-key accuracy" }, "metric": "% accuracy" }, { "id": "arcagi1_image", "name": "ArcAGI1-Image", "category": "Vision Puzzles", "metric": "%", "num_problems": null, "source_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "canonical_setting": { "version": "ArcAGI1-Image", "metric_type": "pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": true, "tools": "none", "notes": "Per Doubao Seed 2.0 Pro model card." } }, { "id": "arcagi2_image", "name": "ArcAGI2-Image", "category": "Vision Puzzles", "metric": "%", "num_problems": null, "source_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "canonical_setting": { "version": "ArcAGI2-Image", "metric_type": "pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": true, "tools": "none", "notes": "Per Doubao Seed 2.0 Pro model card." } }, { "id": "arena_hard", "name": "Arena-Hard Auto", "category": "Instruction Following", "metric": "% win rate", "num_problems": 500, "source_url": "https://lmarena.ai/", "canonical_setting": { "version": "Arena-Hard-Auto", "metric_type": "pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": false, "notes": "tools=none preferred (pure-reasoning eval). If only with-tool scores (python/web/RAG) are available, accept and mark cell matches_canonical=false.", "tools": "none" }, "cost": { "source_id": "arena_hard_auto_paper_20usd", "source_name": "Arena-Hard-Auto paper reported benchmark cost", "source_url": "https://arxiv.org/abs/2406.11939", "source_data_url": "https://arxiv.org/html/2406.11939v2", "source_benchmark_name": "Arena-Hard-Auto", "source_model_name": "LLM-as-judge evaluation pipeline", "source_model_slug": "arena-hard-auto-judge-pipeline", "evidence_scope": "paper-reported total benchmark cost for 500 prompts", "tokens": { "prompt_tokens": null, "completion_tokens": null, "input_tokens": null, "output_tokens": null, "reasoning_tokens": null, "answer_tokens": null, "total_tokens": null }, "total_cost_usd": 20.0, "reported_items": 500, "reported_samples": 1, "relative_tokens_to_source_anchor": null, "relative_cost_to_source_anchor": null, "notes": "Arena-Hard-Auto / BenchBuilder paper reports the benchmark reaches high human-correlation at a cost of $20 over 500 prompts, or $0.04 per prompt. Paper-level dollar evidence only; raw token logs are not public." } }, { "id": "artifactsbench", "name": "ArtifactsBench", "category": "Coding", "metric": "%", "num_problems": 5475, "source_url": "https://github.com/Tencent-Hunyuan/ArtifactsBenchmark", "canonical_setting": { "version": "ArtifactsBench full benchmark; MiniMax-M2 reported setting", "metric_type": "pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": true, "tools": "none for model; evaluator renders generated artifacts and captures screenshots", "sampling": "included in num_problems", "judge": "Gemini-2.5-Pro MLLM-as-Judge with checklist-guided scoring", "notes": "Official ArtifactsBench contains 1825 diverse tasks / HF rows. The MiniMax-M2 score source reports scores averaged over three runs with the official implementation and stable Gemini-2.5-Pro judge. Count records actual model generations for the BenchPress row: 1825 tasks times three runs = 5475. Evaluation renders generated artifacts, captures dynamic behavior, and scores visual/interactivity quality with a multimodal judge." } }, { "id": "babe", "name": "BABE", "category": "Reasoning", "metric": "%", "num_problems": null, "source_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "canonical_setting": { "version": "BABE", "metric_type": "pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": false, "tools": "none", "notes": "Per Doubao Seed 2.0 Pro model card." } }, { "id": "babyvision", "name": "BabyVision", "category": "Multimodal", "source_url": "https://huggingface.co/datasets/UnipatAI/BabyVision", "num_problems": 388, "canonical_setting": { "version": "BabyVision MLLM evaluation", "metric_type": "pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": true, "judge": "LLM judge compares model output to ground truth answer", "sampling": "pass@1", "notes": "Official BabyVision MLLM evaluation has 388 visual reasoning tasks; BabyVision-Gen is a separate generation-track benchmark." }, "metric": "% accuracy" }, { "id": "beyond_aime", "name": "Beyond AIME", "category": "Math", "metric": "%", "num_problems": 100, "source_url": "https://huggingface.co/datasets/ByteDance-Seed/BeyondAIME", "canonical_setting": { "version": "Beyond AIME", "metric_type": "pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": false, "tools": "none", "notes": "HF dataset card reports one test split with 100 problems; answers are positive integers with automated exact verification. Per Seed-Thinking-v1.5 paper." } }, { "id": "bfcl", "name": "BFCL", "category": "Tool use", "source_url": "https://cohere.com/research/papers/command-a-technical-report.pdf", "num_problems": null, "canonical_setting": { "version": "Berkeley Function Calling Leaderboard (Tau-bench predecessor)", "metric_type": "pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": false, "notes": "Function calling benchmark. Distinct from bfcl_v3." } }, { "id": "bfcl_v3", "name": "BFCL v3", "category": "Tool use", "source_url": "https://gorilla.cs.berkeley.edu/leaderboard.html", "num_problems": null, "canonical_setting": { "version": "BFCL v3 (Berkeley Function Calling Leaderboard)", "metric_type": "pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": false, "notes": "Function-calling benchmark, FC format" } }, { "id": "bfcl_v3_multiturn", "name": "BFCL v3 (Multi-Turn)", "category": "Tool Use", "metric": "%", "num_problems": null, "source_url": "https://huggingface.co/deepseek-ai/DeepSeek-R1-0528", "canonical_setting": { "version": "BFCL v3 (Multi-Turn)", "metric_type": "pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": false, "tools": "agentic", "notes": "Per DeepSeek R1-0528 model card." } }, { "id": "bfcl_v4", "name": "BFCL v4", "category": "Tool Use", "metric": "%", "num_problems": null, "source_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "canonical_setting": { "version": "BFCL v4", "metric_type": "pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": false, "tools": "tool calls", "notes": "Per Doubao Seed 2.0 Pro model card." }, "cost": { "source_id": "bfcl_v4_leaderboard_gemini_3_pro_fc", "source_name": "Berkeley Function Calling Leaderboard v4 cost table", "source_url": "https://gorilla.cs.berkeley.edu/leaderboard.html", "source_data_url": "https://gorilla.cs.berkeley.edu/data_overall.csv", "source_benchmark_name": "BFCL v4", "source_model_name": "Gemini-3-Pro-Preview (FC)", "source_model_slug": "gemini-3-pro-preview-fc", "evidence_scope": "reported leaderboard total dollar cost for the source model row on the entire benchmark", "tokens": { "prompt_tokens": null, "completion_tokens": null, "input_tokens": null, "output_tokens": null, "reasoning_tokens": null, "answer_tokens": null, "total_tokens": null }, "total_cost_usd": 224.69, "reported_items": null, "reported_samples": null, "relative_tokens_to_source_anchor": null, "relative_cost_to_source_anchor": null, "notes": "Dollar-only evidence: BFCL reports Total Cost ($) as an estimate for the entire benchmark, but the CSV does not report token counts. This row matches the BenchPress bfcl_v4 score cell for gemini-3-pro." } }, { "id": "bigbench_extra_hard", "name": "BigBench Extra Hard", "category": "Reasoning", "source_url": "https://github.com/google-deepmind/bbeh", "num_problems": 4520, "canonical_setting": { "version": "Big-Bench Extra Hard full benchmark", "metric_type": "pct", "higher_is_better": true, "range": [ 0, 100 ], "multimodal_input": false, "tools": "none", "notes": "Full 4,520-example benchmark; Gemma reports example-weighted micro-average accuracy." }, "metric": "micro accuracy (%)" }, { "id": "bigbench_hard", "name": "BigBench Hard (BBH)", "category": "Reasoning", "source_url": "https://github.com/suzgunmirac/BIG-Bench-Hard/tree/9ee07bd481feebf959a6b59d61ea57bdcf30964d", "num_problems": 6511, "canonical_setting": { "version": "BIG-Bench Hard immutable official release", "metric_type": "pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": false, "notes": "The official paper calls BBH 23 challenging tasks; the locked release contains 27 JSON task files and exactly 6,511 prompt examples. Count is actual model generations.", "tools": "none", "sampling": "one response per released prompt", "judge": "task-specific exact-match normalization" }, "metric": "% exact-match accuracy" }, { "id": "bigcodebench", "name": "BigCodeBench", "category": "Coding", "metric": "pass@1 %", "num_problems": 1140, "source_url": "https://bigcode-bench.github.io/", "canonical_setting": { "version": "BigCodeBench (1140 full set)", "metric_type": "pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": false, "notes": "Official BigCodeBench complete/instruct evaluation uses generated code executed by the benchmark sandbox and unit tests; no agentic tools. Score observations must identify complete versus instruct split.", "tools": "none", "judge": "sandboxed unit-test evaluator" } }, { "id": "biobench", "name": "BIObench", "category": "Science Discovery", "metric": "%", "num_problems": null, "source_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "canonical_setting": { "version": "BIObench", "metric_type": "pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": false, "tools": "none", "notes": "Per Doubao Seed 2.0 Pro model card." } }, { "id": "bird_sql", "name": "Bird-SQL (Dev)", "category": "Coding", "source_url": "https://bird-bench.github.io/", "num_problems": null, "canonical_setting": { "version": "Bird-SQL Dev split (NL→SQL)", "metric_type": "pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": false, "notes": "Natural language to executable SQL on Bird-SQL dev split." } }, { "id": "bixbench", "name": "BixBench Zero-Shot MCQ", "category": "Science", "metric": "accuracy (%)", "num_problems": 205, "source_url": "https://github.com/Future-House/BixBench", "canonical_setting": { "version": "BixBench zero-shot MCQ", "metric_type": "pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": false, "tools": "analysis tools available", "sampling": "pass@1", "judge": "zero-shot multiple-choice accuracy", "harness": "BixBench official zero-shot MCQ; score-level agent harness", "notes": "Public benchmark contains 205 computational-biology questions. The xAI score uses Grok Build with analysis tools enabled by default." } }, { "id": "blink", "name": "BLINK", "category": "Vision Spatial", "metric": "%", "num_problems": null, "source_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "canonical_setting": { "version": "BLINK", "metric_type": "pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": true, "tools": "none", "notes": "Per Doubao Seed 2.0 Pro model card." } }, { "id": "browsecomp", "name": "BrowseComp", "category": "Agentic", "metric": "accuracy (%)", "num_problems": 1266, "source_url": "https://raw.githubusercontent.com/openai/simple-evals/652c89d0ca9df547706735883097e9537d40dc47/browsecomp_eval.py", "canonical_setting": { "version": "BrowseComp official 1,266-question release", "metric_type": "pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": false, "tools": "web browsing agent", "sampling": "one answer per question", "judge": "official BrowseComp grading protocol", "harness": "source-reported browser scaffold", "notes": "The locked official dataset contains 1,266 questions. Browser scaffold and context management remain score-level settings." } }, { "id": "browsecomp_cm", "name": "BrowseComp (w/ Context Manage)", "category": "Agentic", "metric": "accuracy (%)", "num_problems": null, "source_url": "https://z.ai/blog/glm-4.7", "notes": "BrowseComp evaluated with context management (discard-all strategy). Distinct from plain BrowseComp. First reported in GLM-4.7 blog; GLM-5.1 footnote: \"without context management we retain details from the most recent 5 turns; with context management we use the same discard-all strategy as GLM-5 and DeepSeek-v3.2.\"", "canonical_setting": { "version": "BrowseComp with discard-all context management", "metric_type": "pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": false, "tools": "agentic", "notes": "Context management: discard-all strategy (not retain-5-turns). Per z.ai/blog/glm-4.7 and GLM-5.1 blog footnote." } }, { "id": "browsecomp_long_context_128k", "name": "BrowseComp Long Context 128k", "category": "Long Context", "metric": "% accuracy", "num_problems": 1266, "source_url": "https://openai.com/index/gpt-5-1-for-developers/", "canonical_setting": { "version": "BrowseComp Long Context 128k", "metric_type": "pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": false, "tools": "none specified for the 128k long-context row", "sampling": "pass@1", "notes": "OpenAI GPT-5.1 appendix reports BrowseComp Long Context 128k but does not publish a separate count. Use the official BrowseComp 1,266-row test set as the source-backed count unless a 128k-specific slice is found." } }, { "id": "browsecomp_long_context_256k", "name": "BrowseComp Long Context 256k", "description": "BrowseComp long-context retrieval benchmark, 256k context", "canonical_setting": { "judge": "rule-based", "notes": "Per OpenAI GPT-5 developer blog https://openai.com/index/introducing-gpt-5-for-developers/" }, "category": "Long Context" }, { "id": "browsecomp_zh", "name": "BrowseComp-ZH", "category": "Agentic search", "source_url": "https://github.com/PALIN2018/BrowseComp-ZH", "num_problems": 1156, "canonical_setting": { "version": "BrowseComp-ZH official 289-question benchmark; Moonshot avg@4 setting", "metric_type": "pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": false, "tools": "web browsing and search tools", "sampling": "included in num_problems", "judge": "LLM-assisted answer extraction / grading", "notes": "BrowseComp-ZH official paper and repository define 289 native-Chinese multi-hop web-browsing questions across 11 domains. The Moonshot/Kimi score source reports BrowseComp-ZH with avg@4, so the cost count records actual model generations: 289 questions times 4 independent runs = 1,156. Do not use the parent English BrowseComp count." } }, { "id": "brumo_2025", "name": "BRUMO 2025", "category": "Math", "metric": "% correct (pass@1)", "num_problems": 30, "source_url": "https://huggingface.co/datasets/MathArena/brumo_2025", "canonical_setting": { "version": "BRUMO 2025", "metric_type": "pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": false, "sampling": "samples=4", "notes": "tools=none preferred (pure-reasoning eval). If only with-tool scores (python/web/RAG) are available, accept and mark cell matches_canonical=false.", "tools": "none" }, "cost": { "source_id": "matharena_brumo_2025_outputs_kimi_k25", "source_name": "MathArena BRUMO 2025 output logs", "source_url": "https://huggingface.co/datasets/MathArena/brumo_2025_outputs", "source_data_url": "https://huggingface.co/datasets/MathArena/brumo_2025_outputs/resolve/main/data/train-00000-of-00001.parquet", "source_benchmark_name": "BRUMO 2025", "source_model_name": "Kimi K2.5 (Think)", "source_model_slug": "moonshot/k25", "evidence_scope": "observed model-inference totals for 30 problems x 4 sampled answers", "tokens": { "prompt_tokens": null, "completion_tokens": null, "input_tokens": 12849, "output_tokens": 2429415, "reasoning_tokens": null, "answer_tokens": null, "total_tokens": 2442264 }, "total_cost_usd": 7.2959544, "reported_items": 30, "reported_samples": 4, "relative_tokens_to_source_anchor": null, "relative_cost_to_source_anchor": null, "notes": "MathArena public output parquet contains per-answer input_tokens, output_tokens, and cost. Totals are mechanically summed over rows with model_config='moonshot/k25'. Dollar cost is model/run specific." } }, { "id": "bullshit_pushback", "name": "Bullshit-Bench (Clear Pushback)", "category": "Behavior", "metric": "% clear pushback", "num_problems": 55, "source_url": "https://github.com/petergpt/bullshit-benchmark", "canonical_setting": { "version": "Bullshit-pushback (55)", "metric_type": "pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": false, "notes": "tools=none preferred (pure-reasoning eval). If only with-tool scores (python/web/RAG) are available, accept and mark cell matches_canonical=false.", "tools": "none" }, "cost": { "source_id": "bullshitbench_v1_latest_responses_openai_gpt54_none", "source_name": "BullshitBench v1 published response usage logs", "source_url": "https://github.com/petergpt/bullshit-benchmark", "source_data_url": "https://raw.githubusercontent.com/petergpt/bullshit-benchmark/main/data/latest/responses.jsonl", "source_benchmark_name": "BullshitBench v1 / Clear Pushback", "source_model_name": "openai/gpt-5.4@reasoning=none", "source_model_slug": "openai_gpt-5.4_reasoning_none", "evidence_scope": "sum of published response usage over 55 v1 BullshitBench prompts for the tested model row; model-answer collection only", "tokens": { "prompt_tokens": 2008, "completion_tokens": 36600, "input_tokens": null, "output_tokens": null, "reasoning_tokens": 0, "answer_tokens": null, "total_tokens": 38608 }, "total_cost_usd": 0.55402, "reported_items": 55, "reported_samples": 55, "relative_tokens_to_source_anchor": null, "relative_cost_to_source_anchor": null, "notes": "Raw JSON stream exposes per-sample prompt/completion/total/reasoning tokens and cost. This sums the model response collection rows only; separate judge-panel costs, if any, are not included." } }, { "id": "c_eval", "name": "C-Eval (Chinese)", "category": "Knowledge", "metric": "%", "num_problems": 12342, "source_url": "https://huggingface.co/datasets/ceval/ceval-exam", "canonical_setting": { "version": "C-Eval (Chinese)", "metric_type": "pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": false, "tools": "none", "notes": "HF dataset card reports 13,948 total questions across splits; the test split has 12,342 scored multiple-choice questions across 52 subjects." } }, { "id": "cgbench", "name": "CGBench", "category": "Video", "metric": "%", "num_problems": null, "source_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "canonical_setting": { "version": "CGBench", "metric_type": "pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": true, "tools": "none", "notes": "Per Doubao Seed 2.0 Pro model card." } }, { "id": "chartqa", "name": "ChartQA", "category": "Multimodal", "metric": "%", "num_problems": null, "source_url": "https://mistral.ai/news/mistral-medium-3", "canonical_setting": { "version": "ChartQA", "metric_type": "pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": true, "tools": "none", "notes": "Per Mistral Medium 3 blog: Chart visual question answering, 0-shot." } }, { "id": "chartqapro", "name": "ChartQAPro", "category": "Multimodal", "metric": "overall answer accuracy (%)", "num_problems": 1948, "source_url": "https://arxiv.org/abs/2504.05506", "canonical_setting": { "version": "ChartQAPro", "metric_type": "pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": true, "tools": "none", "sampling": "pass@1", "judge": "answer-type-aware official parser/evaluator", "harness": "official", "notes": "1,948 questions over 1,341 charts." } }, { "id": "charxiv_descriptive", "name": "CharXiv Descriptive", "category": "Vision", "metric": "% accuracy", "num_problems": 4000, "source_url": "https://charxiv.github.io/", "canonical_setting": { "version": "CharXiv validation descriptive questions", "metric_type": "pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": true, "tools": "none", "sampling": "pass@1", "notes": "Official leaderboard validation set has 1,000 charts and 5,000 questions; HF schema has four descriptive question fields per chart, so descriptive evaluation is 4,000 model answers." } }, { "id": "charxiv_reasoning", "name": "CharXiv Reasoning", "category": "Multimodal", "metric": "% accuracy", "num_problems": 1000, "source_url": "https://charxiv.github.io/", "canonical_setting": { "version": "CharXiv validation reasoning questions", "metric_type": "pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": true, "tools": "none", "sampling": "pass@1", "notes": "CharXiv v1.0 validation reasoning subset has 1,000 charts and one reasoning answer per chart. Official evaluator uses gpt-4o-2024-05-13 at temperature 0 and seed 42.", "judge": "gpt-4o-2024-05-13, temperature=0, seed=42, binary answer-key judge" } }, { "id": "chatbot_arena_elo", "name": "Chatbot Arena Elo", "category": "Human Preference", "metric": "Elo rating", "num_problems": 8000, "source_url": "https://arxiv.org/abs/2403.04132", "canonical_setting": { "version": "LMArena Chatbot Arena live Elo, text-only general leaderboard", "metric_type": "elo", "range": null, "higher_is_better": true, "multimodal_input": false, "tools": "none", "judge": "human pairwise preference votes", "notes": "Live crowdsourced pairwise comparison benchmark. The paper reports over 240K votes total and about 8K votes per model on average as of Jan 2024; use 8K battles as the source-backed per-model cost proxy. No fixed static item set." } }, { "id": "chinese_simpleqa", "name": "Chinese-SimpleQA", "category": "Knowledge", "metric": "%", "num_problems": 3000, "source_url": "https://huggingface.co/datasets/OpenStellarTeam/Chinese-SimpleQA", "canonical_setting": { "version": "Chinese-SimpleQA", "metric_type": "pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": false, "tools": "none", "judge": "LLM grader", "sampling": "single-pass; no repeated sampling specified", "notes": "Protocol audit: short Chinese factual QA. Each item asks a short-answer factual question; model output is judged for correctness against reference answers. HF dataset card reports 3,000 questions across 6 topics and says grading is run via existing LLMs. No tools, multimodal input, long context, multi-turn interaction, or repeated sampling is specified." } }, { "id": "cl_bench", "name": "CL-Bench", "category": "Long Context", "metric": "%", "num_problems": null, "source_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "canonical_setting": { "version": "CL-Bench", "metric_type": "pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": false, "tools": "none", "notes": "Per Doubao Seed 2.0 Pro model card." } }, { "id": "claw_eval_pass3", "name": "Claw Eval (pass^3)", "category": "Agentic", "source_url": "https://raw.githubusercontent.com/claw-eval/claw-eval/5680b8b11ff2ee5dd2b07b89086a29a5c5c984d7/README.md", "num_problems": 597, "canonical_setting": { "version": "Claw-Eval v1.1 non-multimodal Pass^3", "metric_type": "pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": false, "tools": "official Claw-Eval agent environment", "sampling": "3 independent successful trajectories per task", "judge": "full-trajectory completion/safety/robustness grading", "harness": "official Claw-Eval v1.1", "notes": "This campaign identity is the non-multimodal aggregate: 161 general plus 38 multi-turn tasks. Pass^3 requires all three trials, so num_problems records exactly 597 physical trajectories. The separate 101-task multimodal section is not included in this benchmark identity." }, "metric": "all-three-pass rate (%)" }, { "id": "cluewsc", "name": "CLUEWSC", "category": "Chinese", "metric": "%", "num_problems": 2574, "source_url": "https://huggingface.co/datasets/clue/clue", "canonical_setting": { "version": "CLUEWSC", "metric_type": "pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": false, "tools": "none", "judge": "rule-based", "sampling": "single-pass", "notes": "Protocol audit: Chinese Winograd/coreference-style binary classification. Each item contains a Chinese text and two target spans; the model predicts true/false and scoring is exact match/accuracy against the class label. HF clue/clue dataset card reports cluewsc2020 splits with 2,574 test examples, 1,244 train examples, and 304 validation examples. No LLM judge, tools, multimodal input, long context, multi-turn interaction, or repeated sampling is used." } }, { "id": "cmimc_2025", "name": "CMIMC 2025", "category": "Math", "metric": "% correct (pass@1)", "num_problems": 40, "source_url": "https://huggingface.co/datasets/MathArena/cmimc_2025", "canonical_setting": { "version": "CMIMC 2025", "metric_type": "pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": false, "sampling": "samples=4", "notes": "tools=none preferred (pure-reasoning eval). If only with-tool scores (python/web/RAG) are available, accept and mark cell matches_canonical=false.", "tools": "none" }, "cost": { "source_id": "matharena_cmimc_2025_outputs_kimi_k2_thinking_cost", "source_name": "MathArena CMIMC 2025 output logs for Kimi K2 Thinking", "source_url": "https://huggingface.co/datasets/MathArena/cmimc_2025_outputs", "source_data_url": "https://huggingface.co/datasets/MathArena/cmimc_2025_outputs/resolve/refs%2Fconvert%2Fparquet/default/train/0000.parquet", "source_benchmark_name": "CMIMC 2025", "source_model_name": "Kimi K2 Thinking", "source_model_slug": "moonshot/k2-thinking", "evidence_scope": "observed model-inference totals over public MathArena output rows", "tokens": { "prompt_tokens": null, "completion_tokens": null, "input_tokens": 24541, "output_tokens": 4190329, "reasoning_tokens": null, "answer_tokens": null, "total_tokens": 4214870 }, "total_cost_usd": 10.490547, "reported_items": 40, "reported_samples": 4, "relative_tokens_to_source_anchor": null, "relative_cost_to_source_anchor": null, "notes": "MathArena public outputs contain per-attempt input_tokens, output_tokens, and estimated API cost. Aggregate is over 160 attempts = 40 problems x 4 samples for Kimi K2 Thinking." } }, { "id": "cmmlu", "name": "CMMLU (Chinese)", "category": "Knowledge", "metric": "% accuracy", "num_problems": 11582, "source_url": "https://huggingface.co/datasets/haonan-li/cmmlu/resolve/efcc940752ea4a1ea94d2727f11f83858d64fc8e/README.md", "canonical_setting": { "version": "CMMLU v1.0.1 test split, 67 subjects", "metric_type": "pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": false, "tools": "none", "notes": "The locked official v1.0.1 archive contains 11,582 test questions across 67 subjects; the 5-question dev split is used for the source's reported 5-shot prompting.", "sampling": "one multiple-choice response per question", "judge": "answer-key accuracy" } }, { "id": "cnmo_2024", "name": "CNMO 2024", "category": "Math", "metric": "%", "num_problems": 6, "source_url": "https://www.cms.org.cn/Home/comp/comp_details/id/1253.html", "canonical_setting": { "version": "CNMO 2024", "metric_type": "pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": false, "tools": "none", "judge": "rule-based", "sampling": "samples=16", "notes": "Protocol audit: Chinese National High School Mathematics Olympiad 2024 finals, pure text olympiad math. The official CMS page identifies the 2024 national final / 40th winter camp; the standard CMO format is two days with 3 problems per day (format source: https://zh.wikipedia.org/wiki/中国数学奥林匹克), so the scored set has 6 proof-style math problems. DeepSeek-R1-0528 reports CNMO 2024 as Pass@1 and states that benchmarks requiring sampling use temperature 0.6, top-p 0.95, and 16 responses per query to estimate pass@1. Scoring is rule-based/manual exact mathematical correctness; no LLM judge, tools, multimodal input, long context, or multi-turn interaction." } }, { "id": "codeforces_avg8", "name": "Codeforces (avg@8)", "category": "Coding", "metric": "%", "num_problems": null, "source_url": "https://arxiv.org/abs/2504.13914", "canonical_setting": { "version": "Codeforces (avg@8)", "metric_type": "pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": false, "tools": "none", "notes": "Per Seed-Thinking-v1.5 paper." } }, { "id": "codeforces_pass8", "name": "Codeforces (pass@8)", "category": "Coding", "metric": "%", "num_problems": null, "source_url": "https://arxiv.org/abs/2504.13914", "canonical_setting": { "version": "Codeforces (pass@8)", "metric_type": "pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": false, "tools": "none", "notes": "Per Seed-Thinking-v1.5 paper." } }, { "id": "codeforces_rating", "name": "Codeforces Rating", "category": "Coding", "metric": "Elo rating", "num_problems": null, "source_url": "https://codeforces.com/", "canonical_setting": { "version": "Codeforces live rating", "metric_type": "rating", "range": null, "higher_is_better": true, "multimodal_input": false, "notes": "tools=agentic (scaffold-defined). Preferred = official harness tools (bash/editor for SWE-bench, browser for OSWorld/BrowseComp, official APIs for tau-bench/MCPAtlas/Toolathlon, terminal for Terminal-Bench). Non-official scaffolds → matches_canonical=false.", "tools": "agentic" } }, { "id": "codesimpleqa", "name": "CodeSimpleQA", "category": "Coding", "metric": "%", "num_problems": null, "source_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "canonical_setting": { "version": "CodeSimpleQA", "metric_type": "pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": false, "tools": "none", "notes": "Per Doubao Seed 2.0 Pro model card." } }, { "id": "collie", "name": "COLLIE", "category": "Instruction Following", "metric": "%", "num_problems": 2080, "source_url": "https://arxiv.org/abs/2307.08689", "canonical_setting": { "version": "COLLIE", "metric_type": "pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": false, "tools": "none", "judge": "rule-based", "sampling": "pass@1; single response", "notes": "Protocol audit: constrained text generation benchmark. Each item renders a natural-language instruction from a formal COLLIE constraint structure; the model outputs free-form text, and scoring checks whether the generated text satisfies the target constraint. The COLLIE paper reports COLLIE-v1 has 2,080 instances across 13 constraint structures. The official repo documents evaluation via the constraint checker, so scoring is rule-based/programmatic rather than LLM-judged. BenchPress cells from OpenAI/Doubao reports use pass@1/single-response settings. No tools, multimodal input, long context, or multi-turn interaction is used." } }, { "id": "complexfuncbench", "name": "ComplexFuncBench", "category": "Tool Use", "metric": "%", "num_problems": 1000, "source_url": "https://github.com/THUDM/ComplexFuncBench", "canonical_setting": { "version": "ComplexFuncBench 128k long-context function calling", "metric_type": "pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": false, "tools": "function calling", "judge": "ComplexEval automatic matching plus final-response LLM evaluation", "notes": "Official paper/repo define 1,000 samples: 600 single-domain and 400 cross-domain. Each sample is a multi-step function-calling dialogue; average 3.26 steps and 5.07 calls per sample. Includes real API responses and 128k long-context cases." } }, { "id": "contphy", "name": "ContPhy", "category": "Video", "metric": "%", "num_problems": null, "source_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "canonical_setting": { "version": "ContPhy", "metric_type": "pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": true, "tools": "none", "notes": "Per Doubao Seed 2.0 Pro model card." } }, { "id": "corpusqa_1m", "name": "CorpusQA 1M", "category": "Long Context", "metric": "%", "num_problems": null, "source_url": "https://huggingface.co/deepseek-ai/DeepSeek-V4-Pro", "canonical_setting": { "version": "CorpusQA 1M", "metric_type": "pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": false, "tools": "none", "notes": "Per DeepSeek V4-Pro model card." } }, { "id": "countbench", "name": "CountBench", "category": "Vision Counting", "metric": "%", "num_problems": null, "source_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "canonical_setting": { "version": "CountBench", "metric_type": "pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": true, "tools": "none", "notes": "Per Doubao Seed 2.0 Pro model card." } }, { "id": "covost2", "name": "CoVoST2 (21 lang)", "category": "Audio", "source_url": "https://github.com/facebookresearch/covost", "num_problems": null, "canonical_setting": { "version": "CoVoST2 21-language speech translation (BLEU)", "metric_type": "bleu", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": true, "notes": "Automatic speech translation across 21 languages (BLEU score)." } }, { "id": "creative_writing_v3", "name": "Creative Writing v3 (Elo Normalized)", "category": "Creative", "metric": "elo", "num_problems": null, "source_url": "https://x.ai/news/grok-4-1", "canonical_setting": { "version": "Creative Writing v3 (Elo Normalized)", "metric_type": "elo", "range": [ 1000, 2000 ], "higher_is_better": true, "multimodal_input": false, "tools": "none", "notes": "Creative Writing v3: 32 prompts × 3 iterations. LLM-judged with rubrics + pairwise battles. Elo normalized. Per xAI Grok 4.1 blog." }, "cost": { "source_id": "eqbench_creative_writing_v3_official_judge_cost", "source_name": "EQ-Bench Creative Writing v3 official benchmark documentation", "source_url": "https://eqbench.com/about.html#creative-writing-v3", "source_data_url": "https://eqbench.com/creative_writing.html", "source_benchmark_name": "Creative Writing v3", "source_model_name": "Claude judge for Creative Writing v3 scoring", "source_model_slug": "judge=claude-sonnet; benchmark=creative-writing-v3", "evidence_scope": "official approximate judge/evaluation API cost for scoring one model", "tokens": { "prompt_tokens": null, "completion_tokens": null, "input_tokens": null, "output_tokens": null, "reasoning_tokens": null, "answer_tokens": null, "total_tokens": null }, "total_cost_usd": 10.0, "reported_items": 32, "reported_samples": 3, "relative_tokens_to_source_anchor": null, "relative_cost_to_source_anchor": null, "notes": "Official EQ-Bench documentation states Creative Writing v3 runs 32 writing prompts for 3 iterations and scoring a model costs around $10 in API fees. Judge/evaluation cost only; evaluated-model inference cost and token counts are not reported." } }, { "id": "critpt", "name": "CritPt", "category": "Science", "metric": "% correct", "num_problems": 70, "source_url": "https://huggingface.co/datasets/CritPt-Benchmark/CritPt", "canonical_setting": { "version": "CRITPT", "metric_type": "pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": false, "notes": "Protocol audit: frontier research-level physics benchmark. The public test set has 70 challenges; the broader benchmark has 71 composite research challenges plus an example and 190 checkpoint tasks. Each challenge is a text-only, unpublished physics research problem spanning modern physics subfields, with guess-resistant, machine-verifiable answers. Primary leaderboard metric is average challenge accuracy over 5 runs x 70 test challenges. The official pipeline submits complete batches to an automated grading server customized for advanced physics-specific output formats. Canonical BenchPress setting is no tools; with-code/web-tool variants are non-canonical. The official repo's no-tool config disables Python and web search; reasoning-model examples use large reasoning budgets (e.g. 27k reasoning tokens).", "tools": "none", "judge": "automated rule-based scoring server", "sampling": "trials=5" }, "cost": { "source_id": "artificial_analysis_eval_token_cost_gemini_2_5_pro_anchor", "source_name": "Artificial Analysis per-evaluation token usage and cost with Gemini 2.5 Pro anchor", "source_url": "https://artificialanalysis.ai/evaluations/critpt", "source_benchmark_name": "CritPt", "source_model_name": "Gemini 2.5 Pro", "source_model_slug": "gemini-2-5-pro", "evidence_scope": "observed model-inference totals for the source evaluation run", "tokens": { "prompt_tokens": null, "completion_tokens": null, "input_tokens": 373931, "output_tokens": 1815626, "reasoning_tokens": 1557371, "answer_tokens": 258255, "total_tokens": 2189557 }, "total_cost_usd": 18.62367375, "reported_items": null, "reported_samples": null, "relative_tokens_to_source_anchor": 1.351, "relative_cost_to_source_anchor": 1.181, "notes": "Inference-side token/cost evidence only; source_model_* identifies the reported model row. Dollar cost is run/model specific, not benchmark-intrinsic. primary anchor model: Gemini 2.5 Pro" } }, { "id": "crossvid", "name": "CrossVid", "category": "Video", "metric": "%", "num_problems": null, "source_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "canonical_setting": { "version": "CrossVid", "metric_type": "pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": true, "tools": "none", "notes": "Per Doubao Seed 2.0 Pro model card." } }, { "id": "ctf_internal", "name": "Capture-the-Flags challenge tasks (Internal)", "category": "Cyber", "metric": "%", "num_problems": null, "source_url": "https://openai.com/index/introducing-gpt-5-5/", "canonical_setting": { "version": "Capture-the-Flags challenge tasks (Internal)", "metric_type": "pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": false, "tools": "agentic", "notes": "Hardest CTF challenges from system cards plus additional hard challenges." } }, { "id": "cybench", "name": "Cybench", "category": "Cyber", "metric": "%", "num_problems": 40, "source_url": "https://arxiv.org/abs/2408.08926", "canonical_setting": { "version": "Cybench (public)", "metric_type": "pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": false, "tools": "agentic", "notes": "Public CTF benchmark: 40 challenges from 4 competitions (Zhang et al., 2024). Anthropic evaluated 39/40 (1 skipped due to infra/timing). Score = % of 39 attempted. Pass@30 trials." }, "cost": { "source_id": "cybench_paper_gpt4o_baseline_unguided_tokens", "source_name": "Cybench paper baseline unguided token usage tables", "source_url": "https://arxiv.org/abs/2408.08926", "source_data_url": "https://arxiv.org/e-print/2408.08926", "source_benchmark_name": "Cybench", "source_model_name": "GPT-4o + structured bash baseline agent", "source_model_slug": "gpt-4o; agent=structured-bash-baseline; mode=unguided", "evidence_scope": "reported input/output token totals for one baseline unguided run over all 40 Cybench tasks", "tokens": { "prompt_tokens": null, "completion_tokens": null, "input_tokens": 1722210, "output_tokens": 292420, "reasoning_tokens": null, "answer_tokens": null, "total_tokens": 2014630 }, "total_cost_usd": null, "reported_items": 40, "reported_samples": 1, "relative_tokens_to_source_anchor": null, "relative_cost_to_source_anchor": null, "notes": "Token-only evidence from the Cybench paper source tables. Values are printed in thousands and converted to raw tokens. This is a GPT-4o structured-bash baseline unguided run, not Anthropic's later 39/40 pass@30 setup." } }, { "id": "cybergym", "name": "CyberGym", "category": "Agentic", "metric": "% solved", "num_problems": 1507, "source_url": "https://www.cybergym.io/", "canonical_setting": { "version": "CyberGym Level 1 vulnerability reproduction", "metric_type": "pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": false, "tools": "agentic code execution environment", "judge": "PoC reproduced on vulnerable version and not on fixed version", "notes": "Official benchmark has 1,507 historical vulnerability instances from 188 projects. Agents receive vulnerability description and unpatched codebase, generate PoCs, and are scored by execution against vulnerable/fixed program versions. The 10-task subset is not canonical." } }, { "id": "cybersecurity_ctf", "name": "Cybersecurity Capture The Flag Challenges", "category": "Cyber", "metric": "%", "num_problems": null, "source_url": "https://openai.com/index/introducing-gpt-5-3-codex/", "canonical_setting": { "version": "Cybersecurity Capture The Flag Challenges", "metric_type": "pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": false, "tools": "agentic", "notes": "Cybersecurity CTF benchmark per OpenAI GPT-5.3-Codex blog. Note: distinct from ctf_internal (GPT-5.5 blog uses different problem set)." } }, { "id": "da_2k", "name": "DA-2K", "category": "Vision Spatial", "metric": "%", "num_problems": null, "source_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "canonical_setting": { "version": "DA-2K", "metric_type": "pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": true, "tools": "none", "notes": "Per Doubao Seed 2.0 Pro model card." } }, { "id": "deepconsult", "name": "DeepConsult", "category": "Deep Research", "metric": "%", "num_problems": null, "source_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "canonical_setting": { "version": "DeepConsult", "metric_type": "pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": false, "tools": "research tools", "notes": "Per Doubao Seed 2.0 Pro model card." } }, { "id": "deepresearchbench", "name": "DeepResearchBench", "category": "Deep Research", "metric": "%", "num_problems": null, "source_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "canonical_setting": { "version": "DeepResearchBench", "metric_type": "pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": false, "tools": "research tools", "notes": "Per Doubao Seed 2.0 Pro model card." }, "cost": { "source_id": "deepresearchbench_official_race_judge_cost", "source_name": "DeepResearch Bench official judge-cost table", "source_url": "https://deepresearch-bench.github.io/", "source_benchmark_name": "DeepResearch Bench", "source_model_name": "Gemini 2.5 Pro Preview judge in RACE(Full)", "source_model_slug": "gemini-2.5-pro-preview; role=judge; framework=RACE", "evidence_scope": "official benchmark judge/evaluator cost per query; derived total over 100 tasks", "tokens": { "prompt_tokens": null, "completion_tokens": null, "input_tokens": null, "output_tokens": null, "reasoning_tokens": null, "answer_tokens": null, "total_tokens": null }, "total_cost_usd": 13.0, "reported_items": 100, "reported_samples": 1, "relative_tokens_to_source_anchor": null, "relative_cost_to_source_anchor": null, "notes": "DeepResearch Bench official page Table 3 reports Avg. Cost=$0.13 for Gemini 2.5 Pro Preview as judge LLM within RACE(Full). The benchmark has 100 research tasks, so derived judge cost is 100*$0.13=$13.00. This is evaluator cost, not deep-research agent generation cost." } }, { "id": "deepsearchqa_acc", "name": "DeepSearchQA (Accuracy)", "category": "Search Agent", "metric": "accuracy (%)", "num_problems": 900, "source_url": "https://huggingface.co/datasets/google/deepsearchqa/tree/b2623f8653065c2672de6d941fc5434cd652376c", "canonical_setting": { "version": "DeepSearchQA official 900-prompt accuracy (%)", "metric_type": "pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": false, "tools": "web search agent", "sampling": "one answer per prompt", "judge": "Gemini 2.5 Flash with the official Kaggle starter grading prompt", "harness": "DeepSearchQA official evaluation", "notes": "The pinned official dataset has 900 prompts across 17 fields; changing autorater or prompt can significantly change results." } }, { "id": "deepsearchqa_f1", "name": "DeepSearchQA (F1)", "category": "Search Agent", "metric": "F1 (%)", "num_problems": 900, "source_url": "https://huggingface.co/datasets/google/deepsearchqa/tree/b2623f8653065c2672de6d941fc5434cd652376c", "canonical_setting": { "version": "DeepSearchQA official 900-prompt F1 (%)", "metric_type": "pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": false, "tools": "web search agent", "sampling": "one answer per prompt", "judge": "Gemini 2.5 Flash with the official Kaggle starter grading prompt", "harness": "DeepSearchQA official evaluation", "notes": "The pinned official dataset has 900 prompts across 17 fields; changing autorater or prompt can significantly change results." } }, { "id": "der2_bench", "name": "DeR2 Bench", "category": "Reasoning", "metric": "%", "num_problems": null, "source_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "canonical_setting": { "version": "DeR2 Bench", "metric_type": "pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": false, "tools": "none", "notes": "Per Doubao Seed 2.0 Pro model card." } }, { "id": "disco_x", "name": "Disco-X", "category": "Multilingual", "metric": "%", "num_problems": null, "source_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "canonical_setting": { "version": "Disco-X", "metric_type": "pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": false, "tools": "none", "notes": "Per Doubao Seed 2.0 Pro model card." } }, { "id": "docvqa", "name": "DocVQA", "category": "Multimodal", "metric": "%", "num_problems": null, "source_url": "https://mistral.ai/news/mistral-medium-3", "canonical_setting": { "version": "DocVQA", "metric_type": "pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": true, "tools": "none", "notes": "Per Mistral Medium 3 blog: Document visual question answering, 0-shot." } }, { "id": "drop", "name": "DROP", "category": "Reasoning", "metric": "%", "num_problems": 9536, "source_url": "https://huggingface.co/datasets/EleutherAI/drop", "canonical_setting": { "version": "DROP validation split", "metric_type": "pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": false, "tools": "none", "notes": "DROP is passage-question reading comprehension requiring discrete reasoning. HF EleutherAI/drop reports 77,409 train rows and 9,536 validation rows; use validation as the scored evaluation split. HF ucinlp/drop reports 9,535 validation rows, so the one-row discrepancy is noted and EleutherAI/drop is used because it matches common lm-eval-style benchmark packaging. The official paper describes DROP as a 96k-question benchmark. Official evaluation uses normalized exact match and F1 over number/date/span answers; no LLM judge or tool use.", "judge": "rule-based", "sampling": "single-pass" } }, { "id": "dude", "name": "DUDE", "category": "Document/Chart", "metric": "%", "num_problems": null, "source_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "canonical_setting": { "version": "DUDE", "metric_type": "pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": true, "tools": "none", "notes": "Per Doubao Seed 2.0 Pro model card." } }, { "id": "dynamath", "name": "DynaMath", "category": "Math", "metric": "%", "num_problems": null, "source_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "canonical_setting": { "version": "DynaMath", "metric_type": "pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": true, "tools": "none", "notes": "Per Doubao Seed 2.0 Pro model card." } }, { "id": "egoschema", "name": "EgoSchema (test)", "category": "Video", "source_url": "https://egoschema.github.io/", "num_problems": null, "canonical_setting": { "version": "EgoSchema test split", "metric_type": "pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": true, "notes": "Long-form egocentric video QA across multiple domains." } }, { "id": "egotempo", "name": "EgoTempo", "category": "Video", "metric": "%", "num_problems": null, "source_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "canonical_setting": { "version": "EgoTempo", "metric_type": "pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": true, "tools": "none", "notes": "Per Doubao Seed 2.0 Pro model card." } }, { "id": "emma", "name": "EMMA", "category": "Vision STEM", "metric": "%", "num_problems": null, "source_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "canonical_setting": { "version": "EMMA", "metric_type": "pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": true, "tools": "none", "notes": "Per Doubao Seed 2.0 Pro model card." } }, { "id": "encyclo_k", "name": "Encyclo-K", "category": "Knowledge", "metric": "%", "num_problems": null, "source_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "canonical_setting": { "version": "Encyclo-K", "metric_type": "pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": false, "tools": "none", "notes": "Per Doubao Seed 2.0 Pro model card." } }, { "id": "eq_bench3", "name": "EQ-Bench3 (Emotional Intelligence, Elo Normalized)", "category": "EQ", "metric": "elo", "num_problems": null, "source_url": "https://x.ai/news/grok-4-1", "canonical_setting": { "version": "EQ-Bench3 (Emotional Intelligence, Elo Normalized)", "metric_type": "elo", "range": [ 1000, 2000 ], "higher_is_better": true, "multimodal_input": false, "tools": "none", "notes": "EQ-Bench3: 45 roleplay scenarios × 3 turns. LLM-judged with rubrics + pairwise battles. Elo normalized. Per xAI Grok 4.1 blog." }, "cost": { "source_id": "eqbench3_official_judge_cost", "source_name": "EQ-Bench 3 official about page judge cost", "source_url": "https://eqbench.com/about.html", "source_benchmark_name": "EQ-Bench 3", "source_model_name": "Claude Opus 4.6 judge via OpenRouter", "source_model_slug": "claude-opus-4.6; role=judge; provider=openrouter", "evidence_scope": "official approximate judge/evaluation API cost range for one full benchmark run; midpoint stored", "tokens": { "prompt_tokens": null, "completion_tokens": null, "input_tokens": null, "output_tokens": null, "reasoning_tokens": null, "answer_tokens": null, "total_tokens": null }, "total_cost_usd": 12.5, "reported_items": 45, "reported_samples": 1, "relative_tokens_to_source_anchor": null, "relative_cost_to_source_anchor": null, "notes": "Official EQ-Bench about page says one full EQ-Bench 3 run (rubric plus pairwise) costs approximately $10-$15 on OpenRouter; midpoint $12.50 stored. Rubric-only cost is approximately $1.5 per iteration. Judge cost only; evaluated-model inference cost varies and is excluded." } }, { "id": "erqa", "name": "ERQA", "category": "Vision", "metric": "%", "num_problems": 400, "source_url": "https://github.com/embodiedreasoning/ERQA", "canonical_setting": { "version": "ERQA full benchmark", "metric_type": "pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": true, "tools": "none", "notes": "Official ERQA GitHub README says the full benchmark consists of 400 examples. Questions are multimodal interleaved images and text, phrased as multiple-choice questions, with answers provided as a single letter (A, B, C, D). The dataset covers embodied/spatial/trajectory/action/state-estimation reasoning for real-world robotics scenarios. HF mirrors GeorgeBredis/ERQA and FlagEval/ERQA both report 400 rows. Use exact letter accuracy; no LLM judge, tools, or agentic environment are part of the canonical benchmark.", "judge": "rule-based", "sampling": "single-pass" } }, { "id": "expert_swe", "name": "Expert-SWE (Internal)", "category": "Coding", "metric": "%", "num_problems": null, "source_url": "https://openai.com/index/introducing-gpt-5-5/", "canonical_setting": { "version": "Expert-SWE (Internal)", "metric_type": "pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": false, "tools": "agentic", "notes": "Internal OpenAI software engineering benchmark." } }, { "id": "facts_benchmark", "name": "FACTS Benchmark Suite", "category": "Factuality", "source_url": "https://deepmind.google/models/gemini/flash/", "num_problems": null, "canonical_setting": { "version": "FACTS Benchmark Suite (grounding/parametric/search/MM)", "metric_type": "pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": true, "notes": "Factuality across grounding, parametric, search, and multimodal." } }, { "id": "facts_grounding", "name": "FACTS Grounding", "category": "Factuality", "source_url": "https://arxiv.org/abs/2501.03200", "num_problems": 1719, "canonical_setting": { "version": "FACTS Grounding long-context factuality benchmark", "metric_type": "pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": false, "notes": "FACTS Grounding evaluates whether long-form model responses are factually accurate and grounded in a provided context document. The paper reports 1,719 total examples split into Open N=860 and Blind N=859. Each prompt includes a user request and a full document, with context up to 32k tokens. Models generate long-form responses; scoring uses prompted LLM judges in two phases: responses are first disqualified if they do not fulfill the user request, then judged accurate if fully grounded in the document. The factuality score aggregates three judge models (Gemini 1.5 Pro, GPT-4o, Claude 3.5 Sonnet) to mitigate judge bias.", "tools": "none", "judge": "LLM judge ensemble (Gemini 1.5 Pro, GPT-4o, Claude 3.5 Sonnet)", "sampling": "single-pass model response; scored by three prompted judge models plus eligibility filter" } }, { "id": "factscore", "name": "FActScore (hallucination rate)", "category": "Hallucination", "metric": "%", "num_problems": 500, "source_url": "https://github.com/shmsw25/FActScore", "canonical_setting": { "version": "FActScore unlabeled 500-entity biography set; benchmark tables may use either FActScore or hallucination rate", "metric_type": "pct", "range": [ 0, 100 ], "higher_is_better": false, "multimodal_input": false, "tools": "none", "judge": "retrieval+LLM judge/factuality estimator", "sampling": "single-pass generation; each biography is decomposed into atomic facts and verified against Wikipedia", "notes": "Official FActScore evaluates long-form biography generation for factual precision. The README defines two prompt-entity sets: 183 labeled entities for human-annotated sections and 500 unlabeled entities for broad model evaluation; use the 500-entity unlabeled set as the scored benchmark count. Each model generates a biography for a person entity, then FActScore decomposes the generation into atomic facts and verifies each fact against a Wikipedia knowledge source using retrieval+ChatGPT or retrieval+LLAMA+NP. The official README estimates API cost at about $1 per 100 sentences and reports that 6,500 generations from 13 LMs would have cost $26K if evaluated by humans. Some provider tables report hallucination rate (lower is better) rather than FActScore factual precision (higher is better); preserve source-level score semantics in score-cell notes." } }, { "id": "finance_agent", "name": "Finance Agent v1.1", "category": "Agentic", "metric": "% solved", "num_problems": 537, "source_url": "https://arxiv.org/abs/2508.00828", "canonical_setting": { "version": "Finance Agent Benchmark v1.1; full 537-sample evaluation", "metric_type": "pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": false, "tools": "agentic financial-analysis harness with GoogleSearch, EdgarSearch, ParseHTML, and RetrieveInformation tools", "notes": "Finance Agent Benchmark evaluates autonomous finance agents on expert-authored real-world financial analysis questions requiring recent SEC filings and open-web information. The paper reports 537 expert-authored questions across nine task categories; each entry includes a question, ground-truth answer, source documents, and step-by-step solution approach, and all reported metrics were calculated on the complete 537 samples. The public/private/test split is 50/150/337, but the paper's benchmark results use all 537 samples. The harness gives models Google Search, EDGAR search, HTML parsing, and retrieved-document tools. Scoring uses an LLM-as-judge rubric system: GPT-4o-generated rubrics are manually reviewed, contradiction rubrics check conflicts with the expert answer, and reported metrics include class-balanced accuracy and naive accuracy; figures default to class-balanced accuracy unless otherwise specified. Anthropic Opus 4.7 blog reports Finance Agent v1.1 scores, while the arXiv benchmark paper supplies the source-backed task count and protocol.", "judge": "LLM-as-judge rubric and contradiction grader", "sampling": "single evaluated agent run per question; paper reports class-balanced accuracy and naive accuracy" }, "cost": { "source_id": "finance_agent_paper_o3_cost_per_query", "source_name": "Finance Agent Benchmark paper cost table", "source_url": "https://arxiv.org/abs/2508.00828", "source_data_url": "https://arxiv.org/e-print/2508.00828", "source_benchmark_name": "Finance Agent Benchmark v1.1", "source_model_name": "o3", "source_model_slug": "openai/o3", "evidence_scope": "paper-reported average USD cost per query; full benchmark cost derived using 537 questions", "tokens": { "prompt_tokens": null, "completion_tokens": null, "input_tokens": null, "output_tokens": null, "reasoning_tokens": null, "answer_tokens": null, "total_tokens": null }, "total_cost_usd": 2033.1365825181026, "reported_items": 537, "reported_samples": 1, "relative_tokens_to_source_anchor": null, "relative_cost_to_source_anchor": null, "notes": "Dollar-only evidence. Paper reports o3 average cost per query as $3.7861; TikZ source gives 3.78610164342291. Full-run cost is mechanically derived from the source-backed 537-question benchmark size. Cost is model/run specific." } }, { "id": "finsearchcomp", "name": "FinSearchComp", "category": "Search Agent", "metric": "%", "num_problems": null, "source_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "canonical_setting": { "version": "FinSearchComp", "metric_type": "pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": false, "tools": "search", "notes": "Per Doubao Seed 2.0 Pro model card." } }, { "id": "finsearchcomp_global", "name": "FinSearchComp-Global", "category": "Search Agent", "metric": "%", "num_problems": 317, "source_url": "https://arxiv.org/abs/2509.13160", "canonical_setting": { "version": "FinSearchComp Global subset", "metric_type": "pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": false, "tools": "open-domain search agent; MiniMax-M2 reports use the open-source FinSearchComp framework with search and Python tools", "judge": "LLM-as-a-Judge with task-specific rubrics", "sampling": "single evaluated answer per question; 0-1 correctness", "notes": "FinSearchComp is an open-domain financial search and reasoning benchmark. The paper reports 635 total expert-curated questions across Global and Greater China subsets; Figure 4 gives the Global subset task counts as T1=110, T2=119, and T3=88, so FinSearchComp-Global has 317 scored questions. Each question requires external search/tool use and has a single objective answer. Scoring uses LLM-as-a-Judge with task-specific rubrics and a binary 0-1 judgment; the paper reports roughly 95% agreement with human-verified labels on a representative validation sample. MiniMax-M2 reports FinSearchComp-global scores from the open-source FinSearchComp framework using both search and Python tools." } }, { "id": "finsearchcompt23", "name": "FinSearchComp T2&T3", "category": "Search Agent", "metric": "%", "num_problems": null, "source_url": "https://huggingface.co/moonshotai/Kimi-K2.5", "canonical_setting": { "version": "FinSearchComp T2&T3", "metric_type": "pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": false, "tools": "agentic", "notes": "Per Kimi K2.5 model card." } }, { "id": "flenqa_3k", "name": "FlenQA (3K-token)", "category": "Long Context", "source_url": "https://arxiv.org/abs/2402.14848", "num_problems": null, "canonical_setting": { "version": "FlenQA 3K-token subset", "metric_type": "pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": false, "notes": "Long-context QA at 3K tokens." } }, { "id": "fleurs", "name": "FLEURS", "category": "Audio", "source_url": "https://huggingface.co/datasets/google/fleurs", "num_problems": null, "canonical_setting": { "version": "FLEURS ASR reported language aggregate", "metric_type": "wer", "higher_is_better": false, "range": [ 0, 1 ], "multimodal_input": true, "tools": "none", "notes": "Locale scope varies by source and must be recorded per score. Gemma Table 7 uses a simple macro over 12 locales, with CER for Korean, Japanese and Chinese." }, "metric": "reported aggregate WER/CER (lower=better)" }, { "id": "frames", "name": "Frames", "category": "Agentic search", "metric": "%", "source_url": "https://arxiv.org/abs/2409.12941", "num_problems": 824, "canonical_setting": { "version": "FRAMES test set", "metric_type": "pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": false, "tools": "retrieval/search tools over Wikipedia; official baselines include naive prompting, BM25 retrieval, oracle retrieval, and multi-step retrieval", "judge": "LLM judge/autorater", "sampling": "single answer per question; multi-step retrieval variants iteratively generate search queries", "notes": "FRAMES (Factuality, Retrieval, And reasoning MEasurement Set) evaluates end-to-end RAG systems on 824 multi-hop questions requiring information from 2-15 Wikipedia articles. The official HF dataset google/frames-benchmark has one test split with 824 rows and provides prompt, gold answer, required Wikipedia links, and reasoning-type labels. The paper evaluates single-step settings (naive prompt, BM25-retrieved prompt, oracle prompt) and a multi-step retrieval pipeline where the model generates search queries, retrieves Wikipedia documents, and answers after iterative retrieval. Answers are free-form, so scoring uses an LLM autorater to check whether the candidate answer matches the gold answer; the paper reports 0.96 accuracy and Cohen's kappa 0.889 against human annotations for Gemini-Pro-1.5-0514 as autorating LLM. Kimi K2 Thinking reports Frames under its Agentic Search section; this metadata uses the official FRAMES paper/dataset for count and protocol." } }, { "id": "frontier_science_research", "name": "FrontierScience-Research", "category": "Science", "metric": "reported score (%)", "num_problems": null, "source_url": "https://arxiv.org/abs/2601.21165", "canonical_setting": { "version": "FrontierScience-Research", "metric_type": "pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": false, "tools": "research-oriented environment", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official", "notes": "Official FrontierScience Research track; exact scored task count pending." } }, { "id": "frontiermath", "name": "FrontierMath", "category": "Math", "metric": "% correct T1-3", "num_problems": 300, "source_url": "https://epoch.ai/benchmarks/frontiermath", "canonical_setting": { "version": "FrontierMath Tier 1-3 (300)", "metric_type": "pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": false, "notes": "tools=none preferred (pure-reasoning eval). If only with-tool scores (python/web/RAG) are available, accept and mark cell matches_canonical=false.", "tools": "none" } }, { "id": "frontiermath_tier4", "name": "FrontierMath Tier 4", "category": "Math", "metric": "%", "num_problems": 48, "source_url": "https://epoch.ai/benchmarks/frontiermath", "canonical_setting": { "version": "FrontierMath Tier 4 private set (48 problems)", "metric_type": "pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": false, "tools": "Python tool and submit_answer tool; code execution allowed during reasoning and answer grading", "judge": "rule-based answer-function grader", "sampling": "single evaluated run per problem; 1,000,000-token hard limit with forced submission after 660,000 tokens", "notes": "FrontierMath Tier 4 is the hardest tier of FrontierMath. Epoch’s official FrontierMath page documents the current Inspect-based evaluation: each question asks the model to solve a challenging mathematics problem, may use a Python tool, and must submit a Python function answer() through submit_answer. Correct answers receive 1 point and incorrect/no-submission answers receive 0. The answer function is executed with a 30-second runtime limit on typical 2025 commodity hardware. The page reports FrontierMath-Tier-4-2025-02-28-Private and FrontierMath-Tier-4-2025-07-01-Private evaluations with 48 samples/problems (e.g. Gemini 3 Pro: 3/48 API failures; Grok 4: 8/48 API errors). Use 48 as the scored Tier-4 item count. This protocol differs from OpenAI internal FrontierMath evaluations; OpenAI score pages remain score sources, while Epoch is the benchmark-definition source." } }, { "id": "frontiersci_olympiad", "name": "FrontierScience-Olympiad", "category": "Science", "metric": "reported score (%)", "num_problems": null, "source_url": "https://arxiv.org/abs/2601.21165", "canonical_setting": { "version": "FrontierScience-Olympiad", "metric_type": "pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": false, "tools": "none", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official", "notes": "Official FrontierScience Olympiad track; exact scored task count pending." } }, { "id": "frontiersci_research", "name": "FrontierSci-research", "category": "Science", "metric": "%", "num_problems": null, "source_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "canonical_setting": { "version": "FrontierSci-research", "metric_type": "pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": false, "tools": "none", "notes": "Per Doubao Seed 2.0 Pro model card." } }, { "id": "fsc_147", "name": "FSC-147 (lower=better)", "category": "Vision Counting", "metric": "metric (lower=better)", "num_problems": null, "source_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "canonical_setting": { "version": "FSC-147 (lower=better)", "metric_type": "pct", "range": null, "higher_is_better": false, "multimodal_input": true, "tools": "none", "notes": "Per Doubao Seed 2.0 Pro model card." } }, { "id": "gaia", "name": "GAIA (text only)", "category": "Agentic", "metric": "%", "num_problems": 103, "source_url": "https://arxiv.org/abs/2509.06501", "canonical_setting": { "version": "GAIA 103-sample text-only validation subset", "metric_type": "pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": false, "tools": "agentic web-search and browse tools", "judge": "LLM-as-Judge for WebExplorer-style reported scores; GAIA original answers are unambiguous final-answer tasks", "sampling": "single reported run not specified in MiniMax-M2 card; WebExplorer reports its own benchmark scores as Avg@4", "notes": "GAIA is a benchmark for general AI assistants with 466 total questions, requiring reasoning, tool use, web browsing, and sometimes multimodality. This BenchPress row is specifically GAIA (text only), not full GAIA. MiniMax-M2 reports GAIA (text only) using the same agent framework as WebExplorer and states that it uses the 103-sample text-only GAIA validation subset following WebExplorer. WebExplorer describes the GAIA setting as a widely adopted benchmark for General AI Assistants and uses a search/browse web-agent scaffold; it reports scores on information-seeking benchmarks using LLM-as-Judge, while the original GAIA benchmark defines unambiguous final-answer questions. Use 103 as the scored item count for this text-only subset." }, "cost": { "source_id": "hal_gaia_leaderboard_gpt5_medium_api_cost", "source_name": "HAL GAIA leaderboard reported API cost", "source_url": "https://hal.cs.princeton.edu/gaia", "source_benchmark_name": "GAIA", "source_model_name": "GPT-5 Medium (August 2025)", "source_model_slug": "hal-generalist-agent_gpt-5-medium-august-2025", "evidence_scope": "official leaderboard total API cost for a verified agent run", "tokens": { "prompt_tokens": null, "completion_tokens": null, "input_tokens": null, "output_tokens": null, "reasoning_tokens": null, "answer_tokens": null, "total_tokens": null }, "total_cost_usd": 104.75, "reported_items": null, "reported_samples": 1, "relative_tokens_to_source_anchor": null, "relative_cost_to_source_anchor": null, "notes": "Dollar-only source-backed cost. HAL page reports total API cost for all GAIA tasks; BenchPress row is GAIA text-only 103-sample subset, so mapping is not exact." } }, { "id": "gdpval_aa_elo", "name": "GDPval (Artificial Analysis ELO)", "category": "Knowledge", "metric": "score", "num_problems": 220, "source_url": "https://huggingface.co/datasets/openai/gdpval", "canonical_setting": { "version": "GDPval public 220-task set; Artificial Analysis Elo aggregation", "metric_type": "index", "range": null, "higher_is_better": true, "multimodal_input": false, "tools": "office/document/spreadsheet workflow; reference files and deliverable files vary by task", "judge": "rubric-based grader / pairwise Elo aggregation in Artificial Analysis", "sampling": "single deliverable per task; AA reports Elo-style score rather than raw percent", "notes": "GDPval evaluates AI model performance on real-world economically valuable tasks. The official OpenAI HF dataset reports 220 tasks across 44 occupations; each task consists of a text prompt and supporting reference files, with expected deliverable files such as Excel workbooks, Word documents, PDFs, or other work products. Rows include human-authored rubric criteria with point values. This BenchPress row is the Artificial Analysis GDPval Elo/index view, so the score source is Artificial Analysis, but the benchmark-definition source and item count are the official OpenAI GDPval dataset. Use 220 as the scored task count; do not use the Knowledge category fallback." }, "cost": { "source_id": "artificial_analysis_eval_token_cost_gemini_2_5_pro_anchor", "source_name": "Artificial Analysis per-evaluation token usage and cost with Gemini 2.5 Pro anchor", "source_url": "https://artificialanalysis.ai/evaluations/gdpval-aa", "source_benchmark_name": "GDPval-AA", "source_model_name": "Gemini 2.5 Pro", "source_model_slug": "gemini-2-5-pro", "evidence_scope": "observed model-inference totals for the source evaluation run", "tokens": { "prompt_tokens": null, "completion_tokens": null, "input_tokens": 34985674, "output_tokens": 3563478, "reasoning_tokens": 2514384, "answer_tokens": 1049093, "total_tokens": 38549152 }, "total_cost_usd": 79.3668725, "reported_items": null, "reported_samples": null, "relative_tokens_to_source_anchor": 23.79, "relative_cost_to_source_anchor": 5.031, "notes": "Inference-side token/cost evidence only; source_model_* identifies the reported model row. Dollar cost is run/model specific, not benchmark-intrinsic. primary anchor model: Gemini 2.5 Pro" } }, { "id": "gdpval_diamond", "name": "GDPVal-Diamond", "category": "Economic", "metric": "%", "num_problems": null, "source_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "canonical_setting": { "version": "GDPVal-Diamond", "metric_type": "pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": false, "tools": "none", "notes": "Per Doubao Seed 2.0 Pro model card." } }, { "id": "gdpval_oai_woe", "name": "GDPval (OpenAI wins-or-ties)", "category": "Office", "metric": "%", "num_problems": null, "source_url": "https://openai.com/index/introducing-gpt-5-5/", "canonical_setting": { "version": "GDPval (OpenAI wins-or-ties)", "metric_type": "pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": false, "tools": "agentic", "notes": "OpenAI GDPval head-to-head wins or ties %, baseline against reference. Single-source: OpenAI GPT-5.5 blog." }, "cost": { "source_id": "artificial_analysis_eval_token_cost_gemini_2_5_pro_anchor", "source_name": "Artificial Analysis per-evaluation token usage and cost with Gemini 2.5 Pro anchor", "source_url": "https://artificialanalysis.ai/evaluations/gdpval-aa", "source_benchmark_name": "GDPval-AA", "source_model_name": "Gemini 2.5 Pro", "source_model_slug": "gemini-2-5-pro", "evidence_scope": "observed model-inference totals for the source GDPval-AA agentic evaluation run", "tokens": { "prompt_tokens": null, "completion_tokens": null, "input_tokens": 34985674, "output_tokens": 3563478, "reasoning_tokens": 2514384, "answer_tokens": 1049093, "total_tokens": 38549152 }, "total_cost_usd": 79.3668725, "reported_items": null, "reported_samples": null, "relative_tokens_to_source_anchor": 23.79, "relative_cost_to_source_anchor": 5.031, "notes": "Candidate/proxy. Same GDPval dataset family, but Artificial Analysis reports GDPval-AA Elo/agentic framework, not OpenAI's wins-or-ties metric." } }, { "id": "genebench", "name": "GeneBench", "category": "Science", "metric": "%", "num_problems": null, "source_url": "https://openai.com/index/introducing-gpt-5-5/", "canonical_setting": { "version": "GeneBench", "metric_type": "pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": false, "tools": "none", "notes": "Genomics benchmark." } }, { "id": "global_mmlu_lite", "name": "Global MMLU Lite", "category": "Knowledge", "source_url": "https://huggingface.co/datasets/CohereForAI/Global-MMLU-Lite", "num_problems": 7200, "canonical_setting": { "version": "Global MMLU Lite (multilingual MMLU subset, Cohere)", "metric_type": "pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": false, "notes": "Multilingual MMLU subset by Cohere. HF dataset card reports 18 languages with 400 test examples each. Distinct from MMLU (5-shot 14k EN) and MMMLU (multilingual MMLU full)." }, "cost": { "source_id": "artificial_analysis_eval_token_cost_gemini_2_5_pro_anchor", "source_name": "Artificial Analysis per-evaluation token usage and cost with Gemini 2.5 Pro anchor", "source_url": "https://artificialanalysis.ai/evaluations/global-mmlu-lite", "source_benchmark_name": "Global MMLU Lite", "source_model_name": "Gemini 2.5 Pro", "source_model_slug": "gemini-2-5-pro", "evidence_scope": "observed model-inference totals for the source evaluation run", "tokens": { "prompt_tokens": null, "completion_tokens": null, "input_tokens": 3383385, "output_tokens": 30989139, "reasoning_tokens": 27435968, "answer_tokens": 3553171, "total_tokens": 34372524 }, "total_cost_usd": 314.12062125, "reported_items": null, "reported_samples": null, "relative_tokens_to_source_anchor": 21.213, "relative_cost_to_source_anchor": 19.912, "notes": "Inference-side token/cost evidence only; source_model_* identifies the reported model row. Dollar cost is run/model specific, not benchmark-intrinsic. primary anchor model: Gemini 2.5 Pro" } }, { "id": "global_piqa", "name": "Global PIQA", "category": "Reasoning", "source_url": "https://huggingface.co/datasets/mrlbenchmarks/global-piqa-parallel", "num_problems": 6283, "canonical_setting": { "version": "Global PIQA parallel test set", "metric_type": "pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": false, "tools": "none", "judge": "exact-match multiple-choice answer key", "sampling": "single answer per multiple-choice question; prompted format selects A/B/C/D or completion format ranks candidate likelihoods", "notes": "Global PIQA is a participatory commonsense-reasoning benchmark for 100+ languages and cultures. The arXiv preprint describes 116 language varieties constructed by 335 researchers from 65 countries. The HF Global PIQA Parallel dataset card says each example has a question prompt and four candidate solutions, one correct and three incorrect; evaluation can use either prompted multiple-choice selection or completion likelihood ranking. The HF dataset-server size endpoint reports 6,283 total rows across the parallel test configurations, with 103 examples in each language-variety config that is populated in the dataset server. Google Gemini reports Global PIQA as commonsense reasoning across 100 languages and cultures; use the HF dataset row count as the scored item count." } }, { "id": "gpqa_diamond", "name": "GPQA Diamond", "category": "Science", "metric": "multiple-choice accuracy (%)", "num_problems": 198, "source_url": "https://huggingface.co/datasets/Idavidrein/gpqa/tree/633f5ee89ab8ad4522a9f850766b73f62147ffdd", "canonical_setting": { "version": "GPQA Diamond official 198-item split", "metric_type": "pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": false, "tools": "none", "sampling": "pass@1", "judge": "multiple-choice answer-key grading", "harness": "official GPQA", "notes": "The pinned official dataset exposes 198 Diamond items. Repeated sampling and quantized decoding remain score-level settings." }, "cost": { "source_id": "artificial_analysis_eval_token_cost_gemini_2_5_pro_anchor", "source_name": "Artificial Analysis per-evaluation token usage and cost with Gemini 2.5 Pro anchor", "source_url": "https://artificialanalysis.ai/evaluations/gpqa-diamond", "source_benchmark_name": "GPQA Diamond", "source_model_name": "Gemini 2.5 Pro", "source_model_slug": "gemini-2-5-pro", "evidence_scope": "observed model-inference totals for the source evaluation run", "tokens": { "prompt_tokens": null, "completion_tokens": null, "input_tokens": 48975, "output_tokens": 1571405, "reasoning_tokens": 1372733, "answer_tokens": 198672, "total_tokens": 1620380 }, "total_cost_usd": 15.77526875, "reported_items": null, "reported_samples": null, "relative_tokens_to_source_anchor": 1.0, "relative_cost_to_source_anchor": 1.0, "notes": "Inference-side token/cost evidence only; source_model_* identifies the reported model row. Dollar cost is run/model specific, not benchmark-intrinsic. primary anchor model: Gemini 2.5 Pro" } }, { "id": "gpqa_main", "name": "GPQA Main (full set)", "category": "Science", "source_url": "https://arxiv.org/abs/2311.12022", "num_problems": 448, "canonical_setting": { "version": "GPQA full set (Rein et al. 2023, 448 questions)", "metric_type": "pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": false, "notes": "Full GPQA. Distinct from gpqa_diamond (198 hardest subset)." }, "cost": { "source_id": "helm_capabilities_v1_15_0", "source_name": "HELM Capabilities v1.15.0", "source_url": "https://raw.githubusercontent.com/stanford-crfm/helm/main/docs/benchmark.md", "source_benchmark_name": "GPQA", "evidence_scope": "observed model-inference totals for the source evaluation run", "tokens": { "prompt_tokens": 53239.17, "completion_tokens": 148568.469, "total_tokens": 201807.639 }, "total_cost_usd": null, "relative_tokens_to_source_anchor": 2.278, "relative_cost_to_source_anchor": null, "notes": "Inference-side token/cost evidence only; judge, human-review, tool/environment, and infrastructure costs are separate protocol factors." } }, { "id": "graphwalks_bfs_0k_128k", "name": "GraphWalks BFS 0-128K", "category": "Long Context", "metric": "%", "num_problems": 300, "source_url": "https://huggingface.co/datasets/openai/graphwalks", "canonical_setting": { "version": "GraphWalks BFS prompts with prompt_chars <=128K", "metric_type": "pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": false, "tools": "none", "judge": "deterministic set-overlap F1 against answer node list", "sampling": "single response per graph operation; prompt includes 3-shot examples plus directed edge list and BFS operation", "notes": "OpenAI GraphWalks is a multi-hop reasoning long-context benchmark. Each prompt gives a directed graph as an edge list and asks for either a BFS result set or a parent-node result set. The official HF dataset has 1,150 rows total, with columns prompt, answer_nodes, prompt_chars, problem_type, and date_added. Counting the official parquet files gives 550 BFS rows total; filtering to problem_type=bfs and prompt_chars<=128000 gives 300 rows for this BenchPress row. Outputs are parsed from a final \"Final Answer: [...]\" list and scored by precision/recall/F1 set overlap against the answer nodes." } }, { "id": "graphwalks_bfs_128k_plus", "name": "GraphWalks BFS >128k", "category": "Long Context", "metric": "%", "num_problems": null, "source_url": "https://openai.com/index/gpt-4-1/", "canonical_setting": { "version": "GraphWalks BFS >128k", "metric_type": "pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": false, "tools": "none", "notes": "Per OpenAI GPT-4.1 blog." } }, { "id": "graphwalks_bfs_256k_1m", "name": "GraphWalks BFS 256K-1M", "category": "Long Context", "metric": "% f1 (avg 256K-1M)", "num_problems": null, "source_url": "https://openai.com/index/introducing-gpt-5-5/", "canonical_setting": { "version": "GraphWalks BFS 256K-1M", "metric_type": "pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": false, "notes": "tools=none preferred (pure-reasoning eval). If only with-tool scores (python/web/RAG) are available, accept and mark cell matches_canonical=false.", "tools": "none" } }, { "id": "graphwalks_parents_0k_128k", "name": "GraphWalks parents 0-128K", "category": "Long Context", "metric": "%", "num_problems": 350, "source_url": "https://huggingface.co/datasets/openai/graphwalks", "canonical_setting": { "version": "GraphWalks parent-node prompts with prompt_chars <=128K", "metric_type": "pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": false, "tools": "none", "judge": "deterministic set-overlap F1 against answer node list", "sampling": "single response per graph operation; prompt includes 3-shot examples plus directed edge list and parent-node operation", "notes": "OpenAI GraphWalks is a multi-hop reasoning long-context benchmark. Each prompt gives a directed graph as an edge list and asks for either a BFS result set or a parent-node result set. The official HF dataset has 1,150 rows total, with columns prompt, answer_nodes, prompt_chars, problem_type, and date_added. Counting the official parquet files gives 600 parent-node rows total; filtering to problem_type=parents and prompt_chars<=128000 gives 350 rows for this BenchPress row. Outputs are parsed from a final \"Final Answer: [...]\" list and scored by precision/recall/F1 set overlap against the answer nodes." } }, { "id": "graphwalks_parents_128k_plus", "name": "GraphWalks parents >128k", "category": "Long Context", "metric": "%", "num_problems": null, "source_url": "https://openai.com/index/gpt-4-1/", "canonical_setting": { "version": "GraphWalks parents >128k", "metric_type": "pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": false, "tools": "none", "notes": "Per OpenAI GPT-4.1 blog." } }, { "id": "graphwalks_parents_256k_1m", "name": "GraphWalks parents 256K-1M", "category": "Long Context", "metric": "%", "num_problems": null, "source_url": "https://openai.com/index/introducing-gpt-5-4/", "canonical_setting": { "version": "GraphWalks parents 256K-1M", "metric_type": "pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": false, "tools": "none", "notes": "Per OpenAI GPT-5.4 blog." } }, { "id": "gsm8k", "name": "GSM8K", "category": "Math", "metric": "% correct", "num_problems": 1319, "source_url": "https://arxiv.org/abs/2110.14168", "canonical_setting": { "version": "GSM8K (test, 1319 problems)", "metric_type": "pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": false, "notes": "tools=none preferred (pure-reasoning eval). If only with-tool scores (python/web/RAG) are available, accept and mark cell matches_canonical=false.", "tools": "none" }, "cost": { "source_id": "helm_lite_v1_13_0", "source_name": "HELM Lite v1.13.0", "source_url": "https://storage.googleapis.com/crfm-helm-public/lite/benchmark_output/releases/v1.13.0/groups.json", "source_benchmark_name": "GSM8K (Grade School Math)", "evidence_scope": "reported HELM prompt/completion token totals for the source benchmark scenario", "tokens": { "prompt_tokens": 286535.82300000085, "completion_tokens": 24419.654999999995, "input_tokens": null, "output_tokens": null, "reasoning_tokens": null, "answer_tokens": null, "total_tokens": 310955.4780000008 }, "total_cost_usd": null, "reported_items": 1000.0, "reported_samples": null, "relative_tokens_to_source_anchor": null, "relative_cost_to_source_anchor": null, "notes": "HELM token evidence only; no dollar cost or hidden reasoning tokens are reported." } }, { "id": "hallusionbench", "name": "HallusionBench", "category": "Vision VQA", "metric": "%", "num_problems": null, "source_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "canonical_setting": { "version": "HallusionBench", "metric_type": "pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": true, "tools": "none", "notes": "Per Doubao Seed 2.0 Pro model card." } }, { "id": "healthbench", "name": "HealthBench", "category": "Knowledge", "metric": "%", "num_problems": 5000, "source_url": "https://huggingface.co/datasets/openai/healthbench", "canonical_setting": { "version": "HealthBench full OSS eval set", "metric_type": "pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": false, "tools": "none", "judge": "LLM judge over physician-written rubric criteria", "sampling": "single assistant completion per medical conversation prompt; scored against rubric items with point values", "notes": "HealthBench evaluates model responses to health and medical conversation prompts. The official OpenAI HF repository points to the HealthBench eval and OpenAI simple-evals reference implementation. The main OSS eval file 2025-05-07-06-14-12_oss_eval.jsonl has 5,000 rows. Each row contains a prompt conversation, example tags, and physician-written rubric criteria with point values; the public preview shows rubrics and prompt_id/canary fields. Separate official files exist for HealthBench Consensus (3,671 rows) and HealthBench Hard (1,000 rows), but this BenchPress row is the full HealthBench score, so use 5,000 items." } }, { "id": "healthbench_consensus", "name": "HealthBench Consensus", "category": "Health", "source_url": "https://arxiv.org/abs/2508.10925", "num_problems": null, "canonical_setting": { "version": "HealthBench Consensus", "metric_type": "pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": false, "notes": "Consensus subset of HealthBench." } }, { "id": "healthbench_hard", "name": "HealthBench Hard", "category": "Health", "source_url": "https://arxiv.org/abs/2508.10925", "num_problems": null, "canonical_setting": { "version": "HealthBench Hard subset", "metric_type": "pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": false, "notes": "Harder subset of HealthBench." } }, { "id": "hellaswag", "name": "HellaSwag", "category": "Reasoning", "metric": "% normalized multiple-choice accuracy", "num_problems": 10042, "source_url": "https://huggingface.co/datasets/Rowan/hellaswag/resolve/218ec52e09a7e7462a5400043bb9a69a41d06b76/README.md", "canonical_setting": { "version": "HellaSwag validation split", "metric_type": "pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": false, "tools": "none", "notes": "Official immutable dataset card reports 10,042 validation examples. The Xiaomi observation reports 10-shot prompting.", "sampling": "one four-choice response per validation example", "judge": "length-normalized answer-choice accuracy" }, "cost": { "source_id": "helm_classic_v0_4_0", "source_name": "HELM Classic v0.4.0", "source_url": "https://storage.googleapis.com/crfm-helm-public/classic/benchmark_output/releases/v0.4.0/groups.json", "source_benchmark_name": "HellaSwag", "evidence_scope": "reported HELM prompt/completion token totals for the source benchmark scenario", "tokens": { "prompt_tokens": 36808.06637073439, "completion_tokens": 36812.809153846036, "input_tokens": null, "output_tokens": null, "reasoning_tokens": null, "answer_tokens": null, "total_tokens": 73620.87552458042 }, "total_cost_usd": null, "reported_items": 615.2, "reported_samples": null, "relative_tokens_to_source_anchor": 0.011, "relative_cost_to_source_anchor": null, "notes": "HELM token evidence only; no dollar cost or hidden reasoning tokens are reported." } }, { "id": "hiddenmath", "name": "HiddenMath", "category": "Math", "source_url": "https://storage.googleapis.com/deepmind-media/Model-Cards/Gemini-2-0-Flash-Model-Card.pdf", "num_problems": null, "canonical_setting": { "version": "HiddenMath (held-out AIME/AMC-like, Google internal)", "metric_type": "pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": false, "notes": "Google internal held-out math benchmark, AIME/AMC-style, not leaked online." } }, { "id": "hipho", "name": "HiPhO", "category": "Vision STEM", "metric": "%", "num_problems": null, "source_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "canonical_setting": { "version": "HiPhO", "metric_type": "pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": true, "tools": "none", "notes": "Per Doubao Seed 2.0 Pro model card." } }, { "id": "hle", "name": "HLE (Humanity's Last Exam)", "category": "Reasoning", "metric": "% correct", "num_problems": 2500, "source_url": "https://lastexam.ai/", "canonical_setting": { "version": "Humanity's Last Exam (2500)", "metric_type": "pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": true, "notes": "Full finalized HLE contains 2,500 text and multimodal questions. Canonical setting is no tools; tool-enabled observations use the existing hle_tools row. Text-only observations use hle_text.", "tools": "none", "judge": "official answer-key rubric with o3-mini judge for non-exact answers" }, "cost": { "source_id": "artificial_analysis_eval_token_cost_gemini_2_5_pro_anchor", "source_name": "Artificial Analysis per-evaluation token usage and cost with Gemini 2.5 Pro anchor", "source_url": "https://artificialanalysis.ai/evaluations/humanitys-last-exam", "source_benchmark_name": "Humanity's Last Exam", "source_model_name": "Gemini 2.5 Pro", "source_model_slug": "gemini-2-5-pro", "evidence_scope": "observed model-inference totals for the source evaluation run", "tokens": { "prompt_tokens": null, "completion_tokens": null, "input_tokens": 659476, "output_tokens": 31433801, "reasoning_tokens": 28616051, "answer_tokens": 2817750, "total_tokens": 32093277 }, "total_cost_usd": 315.162355, "reported_items": null, "reported_samples": null, "relative_tokens_to_source_anchor": 19.806, "relative_cost_to_source_anchor": 19.978, "notes": "Inference-side token/cost evidence only; source_model_* identifies the reported model row. Dollar cost is run/model specific, not benchmark-intrinsic. primary anchor model: Gemini 2.5 Pro" } }, { "id": "hle_text", "name": "HLE Text", "category": "Reasoning", "metric": "%", "num_problems": 2158, "source_url": "https://labs.scale.com/leaderboard/humanitys_last_exam_text_only", "canonical_setting": { "version": "Humanity's Last Exam text-only subset of finalized 2,500-question HLE", "metric_type": "pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": false, "tools": "none", "judge": "answer-key scoring for closed-ended answers", "notes": "Scale Labs states the text-only leaderboard evaluates text-based HLE questions excluding multimodal content and represents 86% of the finalized 2,500-question HLE. Public text-only mirror DongfuJiang/hle_text_only reports 2,158 rows, consistent with that 86% subset. HLE consists of multiple-choice and short-answer questions with unambiguous answers suitable for automated grading; Doubao Seed is retained only as a score source, not as the benchmark-definition source." }, "cost": { "source_id": "artificial_analysis_eval_token_cost_gemini_2_5_pro_anchor", "source_name": "Artificial Analysis per-evaluation token usage and cost with Gemini 2.5 Pro anchor", "source_url": "https://artificialanalysis.ai/evaluations/humanitys-last-exam", "source_benchmark_name": "Humanity's Last Exam", "source_model_name": "Gemini 2.5 Pro", "source_model_slug": "gemini-2-5-pro", "evidence_scope": "observed model-inference totals for the source HLE evaluation run", "tokens": { "prompt_tokens": null, "completion_tokens": null, "input_tokens": 659476, "output_tokens": 31433801, "reasoning_tokens": 28616051, "answer_tokens": 2817750, "total_tokens": 32093277 }, "total_cost_usd": 315.162355, "reported_items": null, "reported_samples": null, "relative_tokens_to_source_anchor": 19.806, "relative_cost_to_source_anchor": 19.978, "notes": "Family-level candidate only. Artificial Analysis reports full HLE; BenchPress hle_text is the text-only subset." } }, { "id": "hle_tools", "name": "HLE (w/ tools)", "category": "Reasoning & Knowledge", "metric": "accuracy (%)", "num_problems": 2500, "higher_is_better": true, "source_url": "https://raw.githubusercontent.com/centerforaisafety/hle/73ae974b1844c3ffa64c3f4343d9f1f259575700/README.md", "notes": "Kimi K2.5 model card reports HLE-Full (w/ tools) on the full text+image HLE set. HLE paper and official HF metadata define the finalized full HLE set as 2,500 questions.", "canonical_setting": { "version": "HLE finalized full 2,500-question text+image set", "metric_type": "pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": true, "tools": "source-reported search/code/web tools", "sampling": "one scored answer per question", "judge": "official answer-key and HLE grading protocol", "harness": "source-specific tool harness", "notes": "The pinned official HLE repository defines the finalized 2,500-question release. Tool suites and context management remain score-level settings." } }, { "id": "hle_verified", "name": "HLE Verified", "category": "Knowledge", "metric": "accuracy (%)", "num_problems": 1811, "source_url": "https://raw.githubusercontent.com/SKYLENAGE-AI/HLE-Verified/b705e0fb541c025a1532ce0d60d70ae2f53b00e0/README.md", "canonical_setting": { "version": "HLE-Verified full verified set: Gold 668 + Revision 1,143; Uncertain 689 excluded", "metric_type": "accuracy_pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": true, "tools": "none", "notes": "Google explicitly reports accuracy over all 1,811 verified/revised items and excludes the 689 Uncertain items.", "sampling": "pass@1", "judge": "HLE-Verified answer evaluator", "harness": "official/self-computed", "dataset_split": "Gold + Revision" } }, { "id": "hle_vl", "name": "HLE-VL", "category": "Vision Agent", "metric": "%", "num_problems": null, "source_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "canonical_setting": { "version": "HLE-VL", "metric_type": "pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": false, "tools": "search", "notes": "Per Doubao Seed 2.0 Pro model card." } }, { "id": "hmmt_feb_2025", "name": "HMMT Feb 2025", "category": "Math", "metric": "%", "num_problems": 30, "source_url": "https://huggingface.co/datasets/MathArena/hmmt_feb_2025", "canonical_setting": { "version": "HMMT Feb 2025", "metric_type": "pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": false, "tools": "none", "sampling": "samples=4", "notes": "MathArena dataset has 30 questions; MathArena evaluates each model 4 times per problem." }, "cost": { "source_id": "matharena_hmmt_feb_2025_gpt52_xhigh_per_problem_cost", "source_name": "MathArena HMMT Feb 2025 per-problem evaluation cost", "source_url": "https://matharena.ai/", "source_data_url": "https://matharena.ai/competition_tables/hmmt--hmmt_feb_2025", "source_benchmark_name": "HMMT Feb 2025", "source_model_name": "GPT-5.2 (xhigh)", "source_model_slug": "gpt-5.2-xhigh", "evidence_scope": "reported dollar cost per one model run on one problem; full official setting multiplies by 30 problems and 4 runs", "tokens": { "prompt_tokens": null, "completion_tokens": null, "input_tokens": null, "output_tokens": null, "reasoning_tokens": null, "answer_tokens": null, "total_tokens": null }, "total_cost_usd": 28.8, "reported_items": 30, "reported_samples": 4, "relative_tokens_to_source_anchor": null, "relative_cost_to_source_anchor": null, "notes": "MathArena table reports $0.24 as average cost of one GPT-5.2 (xhigh) run on one HMMT Feb 2025 problem. MathArena states models are run 4 times per problem; HMMT Feb 2025 has 30 questions, so full official evaluation cost is 30*4*0.24=$28.80. Dollar-only; no token breakdown." } }, { "id": "hmmt_feb_2026", "name": "HMMT Feb 2026", "category": "Math", "metric": "% correct (pass@1)", "num_problems": 33, "source_url": "https://huggingface.co/datasets/MathArena/hmmt_feb_2026", "canonical_setting": { "version": "HMMT Feb 2026", "metric_type": "pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": false, "sampling": "samples=4", "notes": "tools=none preferred (pure-reasoning eval). If only with-tool scores (python/web/RAG) are available, accept and mark cell matches_canonical=false.", "tools": "none" }, "cost": { "source_id": "matharena_hmmt_feb_2026_outputs_kimi_k25", "source_name": "MathArena HMMT Feb 2026 output logs", "source_url": "https://huggingface.co/datasets/MathArena/hmmt_feb_2026_outputs", "source_data_url": "https://huggingface.co/datasets/MathArena/hmmt_feb_2026_outputs/resolve/main/data/train-00000-of-00001.parquet", "source_benchmark_name": "HMMT Feb 2026", "source_model_name": "Kimi K2.5 (Think)", "source_model_slug": "moonshot/k25", "evidence_scope": "observed model-inference totals for 33 problems x 4 sampled answers", "tokens": { "prompt_tokens": null, "completion_tokens": null, "input_tokens": 14612, "output_tokens": 3797023, "reasoning_tokens": null, "answer_tokens": null, "total_tokens": 3811635 }, "total_cost_usd": 11.3964827, "reported_items": 33, "reported_samples": 4, "relative_tokens_to_source_anchor": null, "relative_cost_to_source_anchor": null, "notes": "MathArena public output parquet contains per-answer input_tokens, output_tokens, and cost. Totals are mechanically summed over rows with model_config='moonshot/k25'. Dollar cost is model/run specific." } }, { "id": "hmmt_nov_2025", "name": "HMMT Nov 2025", "category": "Math", "metric": "% correct", "num_problems": 30, "source_url": "https://huggingface.co/datasets/MathArena/hmmt_nov_2025", "canonical_setting": { "version": "HMMT Nov 2025", "metric_type": "pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": false, "sampling": "samples=4", "notes": "tools=none preferred (pure-reasoning eval). If only with-tool scores (python/web/RAG) are available, accept and mark cell matches_canonical=false.", "tools": "none" }, "cost": { "source_id": "matharena_hmmt_nov_2025_outputs_kimi_k2_thinking_cost", "source_name": "MathArena HMMT Nov 2025 output logs for Kimi K2 Thinking", "source_url": "https://huggingface.co/datasets/MathArena/hmmt_nov_2025_outputs", "source_data_url": "https://huggingface.co/datasets/MathArena/hmmt_nov_2025_outputs/resolve/refs%2Fconvert%2Fparquet/default/train/0000.parquet", "source_benchmark_name": "HMMT November 2025", "source_model_name": "Kimi K2 Thinking", "source_model_slug": "moonshot/k2-thinking", "evidence_scope": "observed model-inference totals over public MathArena output rows", "tokens": { "prompt_tokens": null, "completion_tokens": null, "input_tokens": 14100, "output_tokens": 3450229, "reasoning_tokens": null, "answer_tokens": null, "total_tokens": 3464329 }, "total_cost_usd": 8.634032, "reported_items": 30, "reported_samples": 4, "relative_tokens_to_source_anchor": null, "relative_cost_to_source_anchor": null, "notes": "MathArena public outputs contain per-attempt input_tokens, output_tokens, and estimated API cost. Aggregate is over 120 attempts = 30 problems x 4 samples for Kimi K2 Thinking." } }, { "id": "humaneval", "name": "HumanEval", "category": "Coding", "metric": "pass@1 %", "num_problems": 164, "source_url": "https://github.com/openai/human-eval", "canonical_setting": { "version": "HumanEval (Chen et al. 2021)", "metric_type": "pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": false, "notes": "tools=none preferred (pure-reasoning eval). If only with-tool scores (python/web/RAG) are available, accept and mark cell matches_canonical=false.", "tools": "none" }, "cost": { "source_id": "helm_classic_v0_4_0", "source_name": "HELM Classic v0.4.0", "source_url": "https://storage.googleapis.com/crfm-helm-public/classic/benchmark_output/releases/v0.4.0/groups.json", "source_benchmark_name": "HumanEval (Code)", "evidence_scope": "reported HELM prompt/completion token totals for the source benchmark scenario", "tokens": { "prompt_tokens": 1022.0853658536586, "completion_tokens": 490.75609756097566, "input_tokens": null, "output_tokens": null, "reasoning_tokens": null, "answer_tokens": null, "total_tokens": 1512.8414634146343 }, "total_cost_usd": null, "reported_items": 164.0, "reported_samples": null, "relative_tokens_to_source_anchor": 0.000216, "relative_cost_to_source_anchor": null, "notes": "HELM token evidence only; no dollar cost or hidden reasoning tokens are reported." } }, { "id": "humaneval_plus", "name": "HumanEval+", "category": "Coding", "source_url": "https://huggingface.co/datasets/evalplus/humanevalplus/resolve/d32357cf319e50e9c8d8dab5ea876c72b0fd321b/README.md", "num_problems": 164, "canonical_setting": { "version": "EvalPlus HumanEval+ immutable test split", "metric_type": "pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": false, "notes": "Official immutable dataset card reports 164 problems.", "tools": "code execution for tests", "sampling": "pass@1", "judge": "EvalPlus expanded unit-test execution" }, "metric": "pass@1 (%)" }, { "id": "ib_modeling", "name": "Investment Banking Modeling Tasks (Internal)", "category": "Finance", "metric": "%", "num_problems": null, "source_url": "https://openai.com/index/introducing-gpt-5-5/", "canonical_setting": { "version": "Investment Banking Modeling Tasks (Internal)", "metric_type": "pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": false, "tools": "none", "notes": "Internal OpenAI IB modeling benchmark." } }, { "id": "ifbench", "name": "IFBench", "category": "Instruction Following", "metric": "% correct", "num_problems": 300, "source_url": "https://github.com/allenai/IFBench", "canonical_setting": { "version": "IFBench single-turn test set", "metric_type": "pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": false, "judge": "rule-based verification functions", "notes": "Correct benchmark source is AllenAI IFBench / arXiv 2507.02833, not the previously listed arXiv 2502.09980 V2V-QA paper. IFBench has 58 out-of-domain verifiable constraints; the final single-turn benchmark has 300 prompts, matching the allenai/IFBench_test HF dataset size endpoint. The paper generally reports prompt-level loose accuracy with automatic verifier functions.", "tools": "none" }, "cost": { "source_id": "artificial_analysis_eval_token_cost_gemini_2_5_pro_anchor", "source_name": "Artificial Analysis per-evaluation token usage and cost with Gemini 2.5 Pro anchor", "source_url": "https://artificialanalysis.ai/evaluations/ifbench", "source_benchmark_name": "IFBench", "source_model_name": "Gemini 2.5 Pro", "source_model_slug": "gemini-2-5-pro", "evidence_scope": "observed model-inference totals for the source evaluation run", "tokens": { "prompt_tokens": null, "completion_tokens": null, "input_tokens": 24896, "output_tokens": 881559, "reasoning_tokens": 714678, "answer_tokens": 166881, "total_tokens": 906455 }, "total_cost_usd": 8.84671, "reported_items": null, "reported_samples": null, "relative_tokens_to_source_anchor": 0.559, "relative_cost_to_source_anchor": 0.561, "notes": "Inference-side token/cost evidence only; source_model_* identifies the reported model row. Dollar cost is run/model specific, not benchmark-intrinsic. primary anchor model: Gemini 2.5 Pro" } }, { "id": "ifeval", "name": "IFEval", "category": "Instruction Following", "metric": "% correct (prompt strict)", "num_problems": 541, "source_url": "https://arxiv.org/abs/2311.07911", "canonical_setting": { "version": "IFEval prompt-strict (541)", "metric_type": "pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": false, "notes": "tools=none preferred (pure-reasoning eval). If only with-tool scores (python/web/RAG) are available, accept and mark cell matches_canonical=false.", "tools": "none" }, "cost": { "source_id": "helm_capabilities_v1_15_0", "source_name": "HELM Capabilities v1.15.0", "source_url": "https://raw.githubusercontent.com/stanford-crfm/helm/main/docs/benchmark.md", "source_benchmark_name": "IFEval", "evidence_scope": "observed model-inference totals for the source evaluation run", "tokens": { "prompt_tokens": 9577.93, "completion_tokens": 79005.488, "total_tokens": 88583.418 }, "total_cost_usd": null, "relative_tokens_to_source_anchor": 1.0, "relative_cost_to_source_anchor": null, "notes": "Inference-side token/cost evidence only; judge, human-review, tool/environment, and infrastructure costs are separate protocol factors." } }, { "id": "imo_2025", "name": "IMO 2025", "category": "Math", "metric": "% of 42 points", "num_problems": 6, "source_url": "https://matharena.ai/imo/", "canonical_setting": { "version": "IMO 2025", "metric_type": "pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": false, "notes": "tools=none preferred (pure-reasoning eval). If only with-tool scores (python/web/RAG) are available, accept and mark cell matches_canonical=false.", "tools": "none" }, "cost": { "source_id": "matharena_imo_2025_outputs_all_models_cost", "source_name": "MathArena IMO 2025 output logs", "source_url": "https://matharena.ai/imo/", "source_data_url": "https://datasets-server.huggingface.co/rows?dataset=MathArena/imo_2025_outputs&config=default&split=train", "source_benchmark_name": "IMO 2025", "source_model_name": "All 7 MathArena-reported models", "source_model_slug": "deepseek_r1_0528; gemini-pro-2.5; gpt-5; o3; o4-mini--high; grok-4; grok-4-new", "evidence_scope": "official output-log rows with per-row input tokens, output tokens, and cost for all public IMO 2025 runs", "tokens": { "prompt_tokens": null, "completion_tokens": null, "input_tokens": 79575208, "output_tokens": 212147743, "reasoning_tokens": null, "answer_tokens": null, "total_tokens": 291722951 }, "total_cost_usd": 2282.18689648, "reported_items": 6, "reported_samples": 4, "relative_tokens_to_source_anchor": null, "relative_cost_to_source_anchor": null, "notes": "Summed all 168 public MathArena output rows: 6 problems, 4 answers per model/problem, 7 models. Cost is the exact sum of row cost values using prices embedded in the dataset rows." } }, { "id": "imo_answerbench", "name": "IMO-AnswerBench", "category": "Math", "source_url": "https://imobench.github.io/", "num_problems": 400, "canonical_setting": { "version": "IMO-AnswerBench", "metric_type": "pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": false, "judge": "Gemini 2.5 Pro AnswerAutoGrader", "sampling": "samples=8 (reported as avg@8)", "notes": "IMO-Bench official site and arXiv 2511.01846 define IMO-AnswerBench as 400 Olympiad short-answer problems. The paper uses AnswerAutoGrader, built with Gemini 2.5 Pro, to extract final answers and assess correctness against ground truth. Kimi K2.5 reports IMO-AnswerBench with avg@8." } }, { "id": "infobench", "name": "InFoBench", "category": "Instruction following", "source_url": "https://github.com/qinyiwei/InfoBench", "num_problems": 2250, "canonical_setting": { "version": "InFoBench: 500 instructions with 2,250 decomposed requirement-level scoring units", "metric_type": "pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": false, "judge": "GPT-4-0314 judge for decomposed yes/no requirement questions", "tools": "none", "sampling": "greedy decoding; judge temperature=0", "notes": "Official InfoBench repo, arXiv 2401.03601, and HF kqsong/InFoBench define 500 instructions and 2,250 decomposed questions. The DRFR metric scores whether each decomposed requirement is satisfied, and the official evaluation script uses GPT-4-0314 by default to answer each decomposed yes/no question at temperature 0. Cohere Command A remains only a score source for existing cells." } }, { "id": "infovqa", "name": "InfoVQA (val)", "category": "Vision", "metric": "%", "num_problems": null, "source_url": "https://huggingface.co/moonshotai/Kimi-K2.5", "canonical_setting": { "version": "InfoVQA (val)", "metric_type": "pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": true, "tools": "none", "notes": "Per Kimi K2.5 model card." } }, { "id": "internal_api_if_hard", "name": "Internal API IF Hard", "category": "Instruction Following", "metric": "%", "num_problems": null, "source_url": "https://openai.com/index/introducing-gpt-5-for-developers/", "canonical_setting": { "version": "OpenAI internal API instruction-following eval, hard prompts", "metric_type": "pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": false, "tools": "none", "sampling": "pass@1; reasoning models run with high reasoning effort", "notes": "OpenAI GPT-5 developer blog states that the internal OpenAI API instruction-following eval uses difficult instructions derived from real developer feedback and that reasoning models were run with high reasoning effort. The GPT-4.1 API blog describes the same internal instruction-following eval as covering format following, negative instructions, ordered instructions, content requirements, ranking, and overconfidence, split into easy, medium, and hard prompts. OpenAI does not disclose item count or scoring implementation, so keep num_problems null rather than converting the 500 category fallback into a source-backed count." } }, { "id": "inverse_ifeval", "name": "Inverse IFEval", "category": "Instruction Following", "metric": "%", "num_problems": null, "source_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "canonical_setting": { "version": "Inverse IFEval", "metric_type": "pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": false, "tools": "none", "notes": "Per Doubao Seed 2.0 Pro model card." } }, { "id": "korbench", "name": "KORBench", "category": "Reasoning", "metric": "%", "num_problems": null, "source_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "canonical_setting": { "version": "KORBench", "metric_type": "pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": false, "tools": "none", "notes": "Per Doubao Seed 2.0 Pro model card." } }, { "id": "livebench", "name": "LiveBench", "category": "Composite", "metric": "overall score", "num_problems": 1000, "source_url": "https://github.com/LiveBench/LiveBench", "canonical_setting": { "version": "LiveBench 2024-11-25 full public release", "metric_type": "index", "range": null, "higher_is_better": true, "multimodal_input": false, "judge": "objective ground-truth scoring without LLM evaluators", "tools": "none", "sampling": "pass@1", "notes": "Official LiveBench README defines 18 tasks across 6 categories and states that each question has verifiable objective ground-truth answers, scored automatically without an LLM judge. The README says the current 2025-04-25 release is not fully public on Hugging Face and recommends --livebench-release-option 2024-11-25 for the most recent public full-category evaluation. Applying the official HF release/removal filter to the six livebench category datasets gives 1,000 active questions for 2024-11-25: coding 128, data_analysis 150, instruction_following 200, math 232, reasoning 150, language 140. Later public HF rows are incomplete for full-category evaluation." } }, { "id": "livecodebench", "name": "LiveCodeBench", "category": "Coding", "metric": "pass@1 %", "num_problems": 1055, "source_url": "https://livecodebench.github.io/", "canonical_setting": { "version": "LiveCodeBench (1055)", "metric_type": "pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": false, "notes": "tools=agentic (scaffold-defined). Preferred = official harness tools (bash/editor for SWE-bench, browser for OSWorld/BrowseComp, official APIs for tau-bench/MCPAtlas/Toolathlon, terminal for Terminal-Bench). Non-official scaffolds → matches_canonical=false.", "tools": "agentic" }, "cost": { "source_id": "artificial_analysis_eval_token_cost_gemini_2_5_pro_anchor", "source_name": "Artificial Analysis per-evaluation token usage and cost with Gemini 2.5 Pro anchor", "source_url": "https://artificialanalysis.ai/evaluations/livecodebench", "source_benchmark_name": "LiveCodeBench", "source_model_name": "Gemini 2.5 Pro", "source_model_slug": "gemini-2-5-pro", "evidence_scope": "observed model-inference totals for the source evaluation run", "tokens": { "prompt_tokens": null, "completion_tokens": null, "input_tokens": 506736, "output_tokens": 18305104, "reasoning_tokens": 17031151, "answer_tokens": 1273953, "total_tokens": 18811840 }, "total_cost_usd": 183.68446, "reported_items": null, "reported_samples": null, "relative_tokens_to_source_anchor": 11.61, "relative_cost_to_source_anchor": 11.644, "notes": "Inference-side token/cost evidence only; source_model_* identifies the reported model row. Dollar cost is run/model specific, not benchmark-intrinsic. primary anchor model: Gemini 2.5 Pro" } }, { "id": "livecodebench_pro", "name": "LiveCodeBench Pro (Elo)", "category": "Coding", "source_url": "https://livecodebench.github.io/pro.html", "num_problems": null, "canonical_setting": { "version": "LiveCodeBench Pro — Codeforces/ICPC/IOI competitive set", "metric_type": "elo", "range": [ 0, 4000 ], "higher_is_better": true, "multimodal_input": false, "notes": "Elo rating against competitive programming pool." } }, { "id": "livecodebench_v5", "name": "LiveCodeBench v5", "category": "Coding", "metric": "%", "num_problems": null, "source_url": "https://arxiv.org/abs/2504.13914", "canonical_setting": { "version": "LiveCodeBench v5", "metric_type": "pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": false, "tools": "none", "notes": "Per Seed-Thinking-v1.5 paper." }, "cost": { "source_id": "artificial_analysis_eval_token_cost_gpt_oss_120b_low", "source_name": "Artificial Analysis per-evaluation token usage and cost for gpt-oss-120B (low)", "source_url": "https://artificialanalysis.ai/evaluations/livecodebench", "source_benchmark_name": "LiveCodeBench", "source_model_name": "gpt-oss-120B (low)", "source_model_slug": "gpt-oss-120b", "evidence_scope": "observed model-inference totals for the source evaluation run", "tokens": { "prompt_tokens": null, "completion_tokens": null, "input_tokens": 509589, "output_tokens": 2759527, "reasoning_tokens": 1023415, "answer_tokens": 1736112, "total_tokens": 3269116 }, "total_cost_usd": 1.73215455, "reported_items": null, "reported_samples": null, "relative_tokens_to_source_anchor": null, "relative_cost_to_source_anchor": null, "notes": "Existing Artificial Analysis evidence row maps to generic LiveCodeBench, not explicitly LiveCodeBench v5; use as LiveCodeBench-family cost evidence with variant caveat." } }, { "id": "livecodebench_v6", "name": "LiveCodeBench v6", "category": "Coding", "metric": "pass@1 (%)", "num_problems": 1055, "source_url": "https://github.com/LiveCodeBench/LiveCodeBench", "canonical_setting": { "version": "LiveCodeBench release_v6", "metric_type": "pct", "higher_is_better": true, "range": [ 0, 100 ], "multimodal_input": false, "tools": "none", "notes": "1,055 code-generation problems from May 2023 through April 2025; generated code is executed by the judge." }, "cost": { "source_id": "artificial_analysis_eval_token_cost_gemini_2_5_pro_anchor", "source_name": "Artificial Analysis per-evaluation token usage and cost with Gemini 2.5 Pro anchor", "source_url": "https://artificialanalysis.ai/evaluations/livecodebench", "source_benchmark_name": "LiveCodeBench", "source_model_name": "Gemini 2.5 Pro", "source_model_slug": "gemini-2-5-pro", "evidence_scope": "observed model-inference totals for the source evaluation run", "tokens": { "prompt_tokens": null, "completion_tokens": null, "input_tokens": 506736, "output_tokens": 18305104, "reasoning_tokens": 17031151, "answer_tokens": 1273953, "total_tokens": 18811840 }, "total_cost_usd": 183.68446, "reported_items": null, "reported_samples": null, "relative_tokens_to_source_anchor": 11.61, "relative_cost_to_source_anchor": 11.644, "notes": "Candidate for LiveCodeBench v6. Artificial Analysis names the benchmark LiveCodeBench but does not expose the v6 variant label on the public page." } }, { "id": "livesports_3k", "name": "LiveSports-3K", "category": "Video", "metric": "%", "num_problems": null, "source_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "canonical_setting": { "version": "LiveSports-3K", "metric_type": "pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": true, "tools": "none", "notes": "Per Doubao Seed 2.0 Pro model card." } }, { "id": "loft_128k", "name": "LOFT (128k)", "category": "Long Context", "metric": "%", "num_problems": null, "source_url": "https://x.ai/news/grok-3", "canonical_setting": { "version": "LOFT (128k)", "metric_type": "pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": false, "tools": "none", "notes": "Per xAI Grok 3 blog." } }, { "id": "logicvista", "name": "LogicVista", "category": "Vision Reasoning", "metric": "%", "num_problems": null, "source_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "canonical_setting": { "version": "LogicVista", "metric_type": "pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": true, "tools": "none", "notes": "Per Doubao Seed 2.0 Pro model card." } }, { "id": "longbench_v2", "name": "LongBench-V2", "category": "Long Context", "metric": "%", "num_problems": 503, "source_url": "https://huggingface.co/datasets/THUDM/LongBench-v2", "canonical_setting": { "version": "LongBench-V2", "metric_type": "pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": false, "judge": "exact multiple-choice answer matching", "tools": "none", "sampling": "pass@1; temperature=0.1 in official script", "notes": "Official LongBench v2 sources are the THUDM LongBench repo, HF dataset, and arXiv:2412.15204, not the prior DeepSeek model card. The paper/README/HF card state 503 challenging multiple-choice questions with contexts from 8k to 2M words across six task categories. Loading THUDM/LongBench-v2 split=train returns 503 rows with answer keys A-D. The official pred.py default runs one direct-answer generation per item; --cot, --rag, and --no_context are optional settings. The official result.py reports overall accuracy as exact match between extracted option and the answer." } }, { "id": "longdocurl", "name": "LongDocURL", "category": "Vision Long Context", "metric": "%", "num_problems": null, "source_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "canonical_setting": { "version": "LongDocURL", "metric_type": "pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": true, "tools": "none", "notes": "Per Doubao Seed 2.0 Pro model card." } }, { "id": "longfact_concepts", "name": "LongFact-Concepts (hallucination rate)", "category": "Hallucination", "metric": "%", "num_problems": 1140, "source_url": "https://github.com/google-deepmind/long-form-factuality/tree/main/longfact", "canonical_setting": { "version": "LongFact-Concepts (hallucination rate)", "metric_type": "pct", "range": [ 0, 100 ], "higher_is_better": false, "multimodal_input": false, "judge": "SAFE LLM-as-a-judge factuality evaluator", "tools": "Google Search via Serper in SAFE evaluation", "sampling": "pass@1 model response; SAFE max_steps=5 and num_searches=3 by default", "notes": "Official LongFact sources are the google-deepmind/long-form-factuality repo and arXiv:2403.18802, not the prior OpenAI GPT-5 model blog. The longfact README states that LongFact-Concepts has the same 38 topics as LongFact-Objects and 30 unique prompts per topic, giving 1,140 prompts for the Concepts subtask and 2,280 prompts for the full LongFact benchmark. The SAFE README describes evaluation as an LLM-based pipeline that decomposes each long-form response into atomic facts, revises facts to be self-contained, classifies relevance, and checks support using Google Search calls; reported hallucination rate is therefore search-augmented posthoc factuality scoring. LOWER IS BETTER." }, "cost": { "source_id": "longfact_safe_readme_gpt35_turbo_cost_per_pair", "source_name": "Google DeepMind SAFE README cost estimate", "source_url": "https://github.com/google-deepmind/long-form-factuality/tree/main/eval/safe", "source_data_url": "https://raw.githubusercontent.com/google-deepmind/long-form-factuality/main/eval/safe/README.md", "source_benchmark_name": "LongFact-Concepts", "source_model_name": "SAFE evaluator with GPT-3.5-Turbo-0125 and Serper", "source_model_slug": "gpt_35_turbo; search=serper; num_searches=3; max_steps=5", "evidence_scope": "official evaluator cost per prompt-response pair; derived total over 1,140 LongFact-Concepts prompts", "tokens": { "prompt_tokens": null, "completion_tokens": null, "input_tokens": null, "output_tokens": null, "reasoning_tokens": null, "answer_tokens": null, "total_tokens": null }, "total_cost_usd": 228.0, "reported_items": 1140, "reported_samples": 1, "relative_tokens_to_source_anchor": null, "relative_cost_to_source_anchor": null, "notes": "SAFE README estimates about $0.20 per prompt-response pair using GPT-3.5-Turbo-0125 and Serper, up to $0.40 for longer responses. Derived standard evaluator cost: 1,140*$0.20=$228. Evaluator/search cost only; model response generation cost is excluded." } }, { "id": "longfact_objects", "name": "LongFact-Objects (hallucination rate)", "category": "Hallucination", "metric": "%", "num_problems": 1140, "source_url": "https://github.com/google-deepmind/long-form-factuality/tree/main/longfact", "canonical_setting": { "version": "LongFact-Objects (hallucination rate)", "metric_type": "pct", "range": [ 0, 100 ], "higher_is_better": false, "multimodal_input": false, "judge": "SAFE LLM-as-a-judge factuality evaluator", "tools": "Google Search via Serper in SAFE evaluation", "sampling": "pass@1 model response; SAFE max_steps=5 and num_searches=3 by default", "notes": "Official LongFact sources are the google-deepmind/long-form-factuality repo and arXiv:2403.18802, not the prior OpenAI GPT-5 model blog. The longfact README states that LongFact-Objects has the same 38 topics as LongFact-Concepts and 30 unique prompts per topic, giving 1,140 prompts for the Objects main task and 2,280 prompts for the full LongFact benchmark. The SAFE README describes evaluation as an LLM-based pipeline that decomposes each long-form response into atomic facts, revises facts to be self-contained, classifies relevance, and checks support using Google Search calls; reported hallucination rate is therefore search-augmented posthoc factuality scoring. LOWER IS BETTER." }, "cost": { "source_id": "longfact_safe_readme_gpt35_turbo_cost_per_pair", "source_name": "Google DeepMind SAFE README cost estimate", "source_url": "https://github.com/google-deepmind/long-form-factuality/tree/main/eval/safe", "source_data_url": "https://raw.githubusercontent.com/google-deepmind/long-form-factuality/main/eval/safe/README.md", "source_benchmark_name": "LongFact-Objects", "source_model_name": "SAFE evaluator with GPT-3.5-Turbo-0125 and Serper", "source_model_slug": "gpt_35_turbo; search=serper; num_searches=3; max_steps=5", "evidence_scope": "official evaluator cost per prompt-response pair; derived total over 1,140 LongFact-Objects prompts", "tokens": { "prompt_tokens": null, "completion_tokens": null, "input_tokens": null, "output_tokens": null, "reasoning_tokens": null, "answer_tokens": null, "total_tokens": null }, "total_cost_usd": 228.0, "reported_items": 1140, "reported_samples": 1, "relative_tokens_to_source_anchor": null, "relative_cost_to_source_anchor": null, "notes": "SAFE README estimates about $0.20 per prompt-response pair using GPT-3.5-Turbo-0125 and Serper, up to $0.40 for longer responses. Derived standard evaluator cost: 1,140*$0.20=$228. Evaluator/search cost only; model response generation cost is excluded." } }, { "id": "longform_writing", "name": "Longform Writing", "category": "Writing", "metric": "%", "num_problems": null, "source_url": "https://huggingface.co/moonshotai/Kimi-K2-Thinking", "canonical_setting": { "version": "Longform Writing", "metric_type": "pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": false, "tools": "none", "notes": "Per Kimi K2-Thinking model card." } }, { "id": "longvideobench", "name": "LongVideoBench", "category": "Vision", "metric": "%", "num_problems": null, "source_url": "https://huggingface.co/moonshotai/Kimi-K2.5", "canonical_setting": { "version": "LongVideoBench", "metric_type": "pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": true, "tools": "none", "notes": "Per Kimi K2.5 model card." } }, { "id": "lpfqa", "name": "LPFQA", "category": "Knowledge", "metric": "%", "num_problems": null, "source_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "canonical_setting": { "version": "LPFQA", "metric_type": "pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": false, "tools": "none", "notes": "Per Doubao Seed 2.0 Pro model card." } }, { "id": "lvbench", "name": "LVBench", "category": "Multimodal", "metric": "multiple-choice accuracy (%)", "num_problems": 1549, "source_url": "https://raw.githubusercontent.com/zai-org/LVBench/518df47219862534dad39fa1373b4e7c862a4cd5/README.md", "canonical_setting": { "version": "LVBench extreme long-video benchmark", "metric_type": "accuracy_pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": true, "tools": "none", "notes": "Official LVBench release: 103 videos and 1,549 multiple-choice question-answer pairs." } }, { "id": "mars_bench", "name": "MARS-Bench", "category": "Instruction Following", "metric": "%", "num_problems": null, "source_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "canonical_setting": { "version": "MARS-Bench", "metric_type": "pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": false, "tools": "none", "notes": "Per Doubao Seed 2.0 Pro model card." } }, { "id": "math", "name": "MATH", "category": "Math", "source_url": "https://arxiv.org/pdf/2103.03874v2", "num_problems": 5000, "canonical_setting": { "version": "MATH competition benchmark test split", "metric_type": "pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": false, "notes": "The official paper defines 12,500 total problems: 7,500 train and 5,000 test. Count is the 5,000 evaluated test problems, not the full corpus.", "tools": "none", "sampling": "one response per test problem", "judge": "normalized final-answer exact match" }, "cost": { "source_id": "helm_classic_v0_4_0", "source_name": "HELM Classic v0.4.0", "source_url": "https://storage.googleapis.com/crfm-helm-public/classic/benchmark_output/releases/v0.4.0/groups.json", "source_benchmark_name": "MATH", "evidence_scope": "reported HELM prompt/completion token totals for the source benchmark scenario", "tokens": { "prompt_tokens": 1306373.685494141, "completion_tokens": 15380.627375011733, "input_tokens": null, "output_tokens": null, "reasoning_tokens": null, "answer_tokens": null, "total_tokens": 1321754.3128691528 }, "total_cost_usd": null, "reported_items": 62.42857142857143, "reported_samples": null, "relative_tokens_to_source_anchor": 0.189, "relative_cost_to_source_anchor": null, "notes": "HELM token evidence only; no dollar cost or hidden reasoning tokens are reported." }, "metric": "% exact-answer accuracy" }, { "id": "math_500", "name": "MATH-500", "category": "Math", "metric": "% correct", "num_problems": 500, "source_url": "https://arxiv.org/abs/2103.03874", "canonical_setting": { "version": "MATH-500 subset (Hendrycks)", "metric_type": "pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": false, "notes": "tools=none preferred (pure-reasoning eval). If only with-tool scores (python/web/RAG) are available, accept and mark cell matches_canonical=false.", "tools": "none" }, "cost": { "source_id": "artificial_analysis_eval_token_cost_gemini_2_5_pro_anchor", "source_name": "Artificial Analysis per-evaluation token usage and cost with Gemini 2.5 Pro anchor", "source_url": "https://artificialanalysis.ai/evaluations/math-500", "source_benchmark_name": "MATH-500", "source_model_name": "Gemini 2.5 Pro", "source_model_slug": "gemini-2-5-pro", "evidence_scope": "observed model-inference totals for the source evaluation run", "tokens": { "prompt_tokens": null, "completion_tokens": null, "input_tokens": 47911, "output_tokens": 2367565, "reasoning_tokens": 1971303, "answer_tokens": 396262, "total_tokens": 2415476 }, "total_cost_usd": 23.735538749999996, "reported_items": null, "reported_samples": null, "relative_tokens_to_source_anchor": 1.491, "relative_cost_to_source_anchor": 1.505, "notes": "Inference-side token/cost evidence only; source_model_* identifies the reported model row. Dollar cost is run/model specific, not benchmark-intrinsic. primary anchor model: Gemini 2.5 Pro" } }, { "id": "matharena_apex_2025", "name": "MathArena Apex 2025", "category": "Math", "metric": "% correct", "num_problems": 12, "source_url": "https://matharena.ai/apex/", "canonical_setting": { "version": "MathArena Apex 2025", "metric_type": "pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": false, "judge": "exact final-answer auto-verification", "sampling": "samples=16 independent runs per problem in the main Apex table", "notes": "Official MathArena Apex page defines the current benchmark as 12 hard final-answer problems selected from 2025 competitions. Problem filtering used 4 attempts with frontier models, but the reported aggregate results evaluate 9 models with 16 independent runs per problem and report the average success rate over all problems and attempts. GPT-5 with scaffolding is a separate elicitation condition using 4 runs and should not define the canonical direct-prompt setting. Tools are none for the direct reasoning models; final answers are automatically verified.", "tools": "none" }, "cost": { "source_id": "matharena_apex_2025_official_cost_tokens", "source_name": "MathArena official Apex 2025 leaderboard", "source_url": "https://matharena.ai/?comp=apex--apex_2025", "source_data_url": "https://matharena.ai/competition_tables/apex--apex_2025", "source_benchmark_name": "MathArena Apex 2025", "source_model_name": "Qwen3-235B-2507-Think", "source_model_slug": "qwen3-235b-2507-think", "evidence_scope": "official leaderboard per-problem-run cost/token row; derived full direct-prompt evaluation total", "tokens": { "prompt_tokens": null, "completion_tokens": null, "input_tokens": 27072, "output_tokens": 8229312, "reasoning_tokens": null, "answer_tokens": null, "total_tokens": 8256384 }, "total_cost_usd": 9.984, "reported_items": 12, "reported_samples": 16, "relative_tokens_to_source_anchor": null, "relative_cost_to_source_anchor": null, "notes": "MathArena table row reports $0.052 cost per one model run on one problem, 141 average input tokens, and 42,861 average output tokens for Qwen3-235B-2507-Think. Apex page reports 12 problems and 16 attempts per model/problem, so total cost is $0.052*12*16=$9.984; token totals multiply averages by 12*16." } }, { "id": "mathcanvas", "name": "MathCanvas", "category": "Math", "metric": "%", "num_problems": null, "source_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "canonical_setting": { "version": "MathCanvas", "metric_type": "pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": true, "tools": "none", "notes": "Per Doubao Seed 2.0 Pro model card." } }, { "id": "mathkangaroo", "name": "MathKangaroo", "category": "Math", "metric": "%", "num_problems": null, "source_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "canonical_setting": { "version": "MathKangaroo", "metric_type": "pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": true, "tools": "none", "notes": "Per Doubao Seed 2.0 Pro model card." } }, { "id": "mathvision", "name": "MathVision", "category": "Math", "metric": "% correct", "num_problems": 3040, "source_url": "https://huggingface.co/datasets/MathLLMs/MathVision", "canonical_setting": { "version": "MATH-Vision full test split", "metric_type": "pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": true, "judge": "rule-based answer extraction and symbolic/option matching", "sampling": "pass@1", "notes": "Official MATH-Vision sources are the mathllm/MATH-V repo, MathLLMs/MathVision HF dataset, project page, and arXiv:2402.14804. The paper/project page define MATH-Vision as 3,040 mathematical problems with visual contexts across 16 subjects and 5 difficulty levels. The HF dataset has test=3,040 and testmini=304; the project page main leaderboard is on the full 3,040-example test set, while testmini is used for human performance and some smaller evaluations. Loading MathLLMs/MathVision confirms test has 3,040 rows and testmini has 304 rows. Official evaluation/evaluate.py computes overall accuracy by extracting a model answer and checking it against the answer key or option text with symbolic/equivalence rules. VLMEvalKit later supports LLM-based answer extraction, but this is extraction, not LLM-as-a-judge scoring.", "tools": "none" } }, { "id": "mathvista", "name": "MathVista", "category": "Math/Vision", "metric": "%", "num_problems": 1000, "source_url": "https://huggingface.co/datasets/AI4Math/MathVista", "canonical_setting": { "version": "MathVista public testmini subset", "metric_type": "pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": true, "judge": "rule-based answer extraction and answer-key matching", "sampling": "pass@1", "tools": "none", "notes": "Official MathVista sources are the project page, AI4Math/MathVista HF dataset, lupantech/MathVista repo, and arXiv:2310.02255, not the prior OpenAI GPT-4.1 model blog. The paper/project page define the full dataset as 6,141 examples from 31 datasets. HF splits are testmini=1,000 with public answer labels and test=5,141 for private standard evaluation; the HF card says the available leaderboard is testmini, while test labels are not public. Because the OpenAI GPT-4.1 source table reports only 'MathVista' without specifying private test, this canonical setting uses the public testmini subset. Evaluation uses image+question inputs; outputs are scored by extracting an answer and matching the answer key, not by LLM-as-a-judge." } }, { "id": "mathvista_mini", "name": "MathVista (mini)", "category": "Multimodal Math", "source_url": "https://huggingface.co/moonshotai/Kimi-K2.5", "num_problems": null, "canonical_setting": { "version": "MathVista mini", "metric_type": "pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": true, "notes": "avg@3." } }, { "id": "mbpp_plus", "name": "MBPP+", "category": "Coding", "source_url": "https://huggingface.co/datasets/evalplus/mbppplus/resolve/b2d74c91837c3f2a20c1299ae98133cbe7cfa077/README.md", "num_problems": 378, "canonical_setting": { "version": "EvalPlus MBPP+ immutable test split", "metric_type": "pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": false, "notes": "Official immutable dataset card reports 378 problems.", "tools": "code execution for tests", "sampling": "pass@1", "judge": "EvalPlus expanded unit-test execution" }, "metric": "pass@1 (%)" }, { "id": "mcpatlas", "name": "MCPAtlas Public", "category": "Agentic", "metric": "% correct (pass@1)", "num_problems": 500, "source_url": "https://huggingface.co/datasets/ScaleAI/MCP-Atlas", "canonical_setting": { "version": "MCP-Atlas public 500-task May 2026 leaderboard release", "metric_type": "pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": false, "notes": "May 2026 public release contains 500 tasks, 36 MCP servers and 220 tools, scored by claim coverage >=0.75. The current repo later expanded to 307 tools and a different default judge/turn cap; score-level settings must preserve the historical Scale leaderboard snapshot.", "tools": "agentic MCP servers in the official containerized harness", "judge": "Gemini 2.5 Pro claims-based coverage judge; pass if coverage >= 0.75", "sampling": "pass@1; public harness default maxTurns=20" } }, { "id": "mcpmark", "name": "MCPMark", "category": "Agentic", "source_url": "https://github.com/eval-sys/mcpmark", "num_problems": 127, "canonical_setting": { "version": "MCPMark standard task suite", "metric_type": "pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": false, "notes": "Official sources are arXiv:2509.24002, mcpmark.ai, and the eval-sys/mcpmark GitHub repo. MCPMark standard contains 127 tasks with curated initial states and verify.py scripts: 30 Filesystem, 28 Notion, 23 GitHub, 21 PostgreSQL, and 25 Playwright/Playwright-WebArena tasks. GitHub tree count confirms 127 meta.json files under tasks/*/standard; the additional 50 easy tasks are a later lightweight smoke-test suite and are not part of the canonical standard benchmark. Evaluation uses MCPMark-Agent in a tool-calling loop, max 100 turns and 3600-second timeout, then checks final environment state with programmatic verification.", "tools": "agentic MCP tool-calling loop over Notion, GitHub, Filesystem, PostgreSQL, and Playwright/WebArena", "judge": "programmatic verification scripts", "sampling": "pass@1; trials=4 independent runs in the official paper pass@1 mean; pass@4 and pass^4 also reported" }, "metric": "% success (pass@1)", "cost": { "source_id": "mcpmark_experiments_summary_gemini_3_pro_high", "source_name": "MCPMark experiments summary reported token usage and per-run cost", "source_url": "https://github.com/eval-sys/mcpmark-experiments/blob/main/summary.json", "source_benchmark_name": "MCPMark standard task suite", "source_model_name": "gemini-3-pro-high", "source_model_slug": "gemini-3-pro; setting=high", "evidence_scope": "reported per-run token usage and cost for one model row over the 127-task standard suite", "tokens": { "prompt_tokens": null, "completion_tokens": null, "input_tokens": 126003558.5, "output_tokens": 1131699.75, "reasoning_tokens": null, "answer_tokens": null, "total_tokens": 127135258.25 }, "total_cost_usd": 265.587514, "reported_items": 127, "reported_samples": 4, "relative_tokens_to_source_anchor": null, "relative_cost_to_source_anchor": null, "notes": "MCPMark release summary reports k=4. This dictionary stores per_run_input_tokens/per_run_output_tokens/per_run_cost for the gemini-3-pro-high row, i.e. one complete 127-task run rather than the four-run aggregate. Source row reports pass@1.avg=0.5394, pass@4=0.6693, actual_model_name=gemini-3-pro. Dollar cost is model/run specific." } }, { "id": "medxpertqa_mm", "name": "MedXpertQA MM", "category": "Medical", "source_url": "https://github.com/TsinghuaC3I/MedXpertQA", "num_problems": 2000, "canonical_setting": { "version": "MedXpertQA MM test split", "metric_type": "pct", "higher_is_better": true, "range": [ 0, 100 ], "multimodal_input": true, "tools": "none", "notes": "2,000 multimodal medical multiple-choice test questions." }, "metric": "multiple-choice accuracy (%)" }, { "id": "medxpertqa_text", "name": "MedXpertQA (Text)", "canonical_setting": { "version": "MedXpertQA Text (2,450 prompts, 10-choice A–J)", "metric_type": "pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": false, "judge": "gpt-oss-120b", "notes": "Per Meta MSL eval methodology: 2,450 prompts spanning medical specialties; 10 answer choices; graded with gpt-oss-120b. Source: https://ai.meta.com/blog/introducing-muse-spark-msl/" }, "category": "Medical" }, { "id": "mgsm", "name": "MGSM", "category": "Math", "metric": "exact match (%)", "num_problems": 2500, "source_url": "https://github.com/google-research/url-nlp/tree/main/mgsm", "canonical_setting": { "version": "MGSM 10-language benchmark", "metric_type": "pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": false, "tools": "none", "judge": "rule-based exact match on numeric answer", "sampling": "pass@1", "notes": "Official source is google-research/url-nlp MGSM, introduced by arXiv:2210.03057. The benchmark manually translates the same 250 GSM8K test problems into 10 languages (Spanish, French, German, Russian, Chinese, Japanese, Thai, Swahili, Bengali, Telugu), giving 2,500 multilingual scored prompts. The official repo also includes an English TSV with 250 rows; harnesses such as OpenAI simple-evals may include English for 2,750 total prompts, but the MGSM benchmark definition is the 10-language translation set." } }, { "id": "minedojo_verified", "name": "Minedojo Verified", "category": "Vision Agent", "metric": "%", "num_problems": null, "source_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "canonical_setting": { "version": "Minedojo Verified", "metric_type": "pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": false, "tools": "env", "notes": "Per Doubao Seed 2.0 Pro model card." } }, { "id": "minerva", "name": "MINERVA", "category": "Video", "metric": "five-choice accuracy (%)", "num_problems": 1515, "source_url": "https://arxiv.org/abs/2505.00681", "canonical_setting": { "version": "MINERVA", "metric_type": "pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": true, "tools": "none", "sampling": "pass@1", "judge": "exact multiple-choice grader", "harness": "official", "notes": "Original MINERVA, not Cultural or Ego variants." } }, { "id": "mm_browsecomp", "name": "MM-BrowseComp", "category": "Vision Agent", "metric": "%", "num_problems": null, "source_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "canonical_setting": { "version": "MM-BrowseComp", "metric_type": "pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": true, "tools": "search", "notes": "Per Doubao Seed 2.0 Pro model card." } }, { "id": "mme_cc", "name": "MME-CC", "category": "Vision VQA", "metric": "%", "num_problems": null, "source_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "canonical_setting": { "version": "MME-CC", "metric_type": "pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": true, "tools": "none", "notes": "Per Doubao Seed 2.0 Pro model card." } }, { "id": "mmlongbench", "name": "MMLongBench-128K v1.1", "category": "Vision Long Context", "metric": "aggregate score (%)", "num_problems": 13331, "source_url": "https://github.com/EdinburghNLP/MMLongBench", "canonical_setting": { "version": "MMLongBench-128K v1.1", "metric_type": "pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": true, "tools": "none", "sampling": "pass@1", "judge": "mixed rule-based and LLM-based task metrics", "harness": "official", "notes": "v1.1 128K evaluation across five task families." } }, { "id": "mmlongbench_doc", "name": "MMLongBench-Doc", "category": "Vision Long Context", "metric": "%", "num_problems": null, "source_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "canonical_setting": { "version": "MMLongBench-Doc", "metric_type": "pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": true, "tools": "none", "notes": "Per Doubao Seed 2.0 Pro model card." } }, { "id": "mmlu", "name": "MMLU", "category": "Knowledge", "metric": "% correct", "num_problems": 14042, "source_url": "https://arxiv.org/abs/2009.03300", "canonical_setting": { "version": "MMLU (5-shot, 14042 questions)", "metric_type": "pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": false, "notes": "tools=none preferred (pure-reasoning eval). If only with-tool scores (python/web/RAG) are available, accept and mark cell matches_canonical=false.", "tools": "none" }, "cost": { "source_id": "helm_classic_v0_4_0", "source_name": "HELM Classic v0.4.0", "source_url": "https://storage.googleapis.com/crfm-helm-public/classic/benchmark_output/releases/v0.4.0/groups.json", "source_benchmark_name": "MMLU (Massive Multitask Language Understanding)", "evidence_scope": "reported HELM prompt/completion token totals for the source benchmark scenario", "tokens": { "prompt_tokens": 6982477.176177128, "completion_tokens": 14985.454363539857, "input_tokens": null, "output_tokens": null, "reasoning_tokens": null, "answer_tokens": null, "total_tokens": 6997462.630540668 }, "total_cost_usd": null, "reported_items": 49.72727272727273, "reported_samples": null, "relative_tokens_to_source_anchor": 1.0, "relative_cost_to_source_anchor": null, "notes": "HELM token evidence only; no dollar cost or hidden reasoning tokens are reported." } }, { "id": "mmlu_pro", "name": "MMLU-Pro", "category": "Knowledge", "metric": "% correct", "num_problems": 12032, "source_url": "https://arxiv.org/abs/2406.01574", "canonical_setting": { "version": "MMLU-Pro", "metric_type": "pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": false, "notes": "tools=none preferred (pure-reasoning eval). If only with-tool scores (python/web/RAG) are available, accept and mark cell matches_canonical=false.", "tools": "none" }, "cost": { "source_id": "artificial_analysis_eval_token_cost_gemini_2_5_pro_anchor", "source_name": "Artificial Analysis per-evaluation token usage and cost with Gemini 2.5 Pro anchor", "source_url": "https://artificialanalysis.ai/evaluations/mmlu-pro", "source_benchmark_name": "MMLU-Pro", "source_model_name": "Gemini 2.5 Pro", "source_model_slug": "gemini-2-5-pro", "evidence_scope": "observed model-inference totals for the source evaluation run", "tokens": { "prompt_tokens": null, "completion_tokens": null, "input_tokens": 2723851, "output_tokens": 42639100, "reasoning_tokens": 36317607, "answer_tokens": 6321493, "total_tokens": 45362951 }, "total_cost_usd": 429.79581375, "reported_items": null, "reported_samples": null, "relative_tokens_to_source_anchor": 27.995, "relative_cost_to_source_anchor": 27.245, "notes": "Inference-side token/cost evidence only; source_model_* identifies the reported model row. Dollar cost is run/model specific, not benchmark-intrinsic. primary anchor model: Gemini 2.5 Pro" } }, { "id": "mmlu_redux", "name": "MMLU-Redux", "category": "Knowledge", "source_url": "https://huggingface.co/datasets/edinburgh-dawg/mmlu-redux/resolve/3720db6aeb3d019de48bf37916c1a54074ff4997/README.md", "num_problems": 3000, "canonical_setting": { "version": "MMLU-Redux original 30-subject test release", "metric_type": "pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": false, "notes": "The immutable original MMLU-Redux release has 30 subjects with 100 test questions each, totaling 3,000. It is distinct from MMLU-Redux 2.0.", "tools": "none", "sampling": "one multiple-choice response per question", "judge": "answer-key accuracy" }, "metric": "% accuracy" }, { "id": "mmmlu", "name": "MMMLU", "category": "Knowledge", "metric": "% correct", "num_problems": 196588, "source_url": "https://huggingface.co/datasets/openai/MMMLU", "canonical_setting": { "version": "MMMLU 14 translated MMLU test locales", "metric_type": "pct", "higher_is_better": true, "range": [ 0, 100 ], "multimodal_input": false, "tools": "none", "notes": "14 locales times 14,042 MMLU test questions = 196,588 scored prompts." } }, { "id": "mmmu", "name": "MMMU", "category": "Multimodal", "metric": "% correct", "num_problems": 900, "source_url": "https://mmmu-benchmark.github.io/", "canonical_setting": { "version": "MMMU validation (900 questions)", "metric_type": "pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": true, "notes": "tools=none preferred (pure-reasoning eval). If only with-tool scores (python/web/RAG) are available, accept and mark cell matches_canonical=false.", "tools": "none" } }, { "id": "mmmu_pro", "name": "MMMU-Pro", "category": "Multimodal", "metric": "% correct", "num_problems": 3460, "source_url": "https://huggingface.co/datasets/MMMU/MMMU_Pro", "canonical_setting": { "version": "MMMU-Pro overall: standard (10 options) + vision", "metric_type": "pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": true, "notes": "Official sources are arXiv:2409.02813 and the MMMU/MMMU_Pro HuggingFace dataset. The paper filters MMMU to 1,730 questions, augments them to the standard 10-option setting, and creates a matching vision-only version; official overall MMMU-Pro is the average of standard (10 options) and vision, so the evaluated prompt count is 1,730 + 1,730 = 3,460. The standard (4 options) split is a comparison setting and is not counted in the canonical overall. tools=none preferred (pure-reasoning eval). If only with-tool scores (python/web/RAG) are available, accept and mark cell matches_canonical=false.", "tools": "none", "judge": "rule-based multiple-choice accuracy", "sampling": "pass@1" }, "cost": { "source_id": "artificial_analysis_eval_token_cost_gemini_2_5_pro_anchor", "source_name": "Artificial Analysis per-evaluation token usage and cost with Gemini 2.5 Pro anchor", "source_url": "https://artificialanalysis.ai/evaluations/mmmu-pro", "source_benchmark_name": "MMMU-Pro", "source_model_name": "Gemini 2.5 Pro", "source_model_slug": "gemini-2-5-pro", "evidence_scope": "observed model-inference totals for the source evaluation run", "tokens": { "prompt_tokens": null, "completion_tokens": null, "input_tokens": 891495, "output_tokens": 7766561, "reasoning_tokens": 6677746, "answer_tokens": 1088815, "total_tokens": 8658056 }, "total_cost_usd": 78.77997875, "reported_items": null, "reported_samples": null, "relative_tokens_to_source_anchor": 5.343, "relative_cost_to_source_anchor": 4.994, "notes": "Inference-side token/cost evidence only; source_model_* identifies the reported model row. Dollar cost is run/model specific, not benchmark-intrinsic. primary anchor model: Gemini 2.5 Pro" } }, { "id": "mmsibench_circular", "name": "MMSIBench (circular)", "category": "Vision Spatial", "metric": "%", "num_problems": null, "source_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "canonical_setting": { "version": "MMSIBench (circular)", "metric_type": "pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": true, "tools": "none", "notes": "Per Doubao Seed 2.0 Pro model card." } }, { "id": "mmstar", "name": "MMStar", "category": "Vision VQA", "metric": "%", "num_problems": null, "source_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "canonical_setting": { "version": "MMStar", "metric_type": "pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": true, "tools": "none", "notes": "Per Doubao Seed 2.0 Pro model card." } }, { "id": "mmvu", "name": "MMVU", "category": "Vision", "metric": "%", "num_problems": null, "source_url": "https://huggingface.co/moonshotai/Kimi-K2.5", "canonical_setting": { "version": "MMVU", "metric_type": "pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": true, "tools": "none", "notes": "Per Kimi K2.5 model card." } }, { "id": "morse_500", "name": "Morse-500", "category": "Video", "metric": "%", "num_problems": null, "source_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "canonical_setting": { "version": "Morse-500", "metric_type": "pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": true, "tools": "none", "notes": "Per Doubao Seed 2.0 Pro model card." } }, { "id": "motionbench", "name": "MotionBench", "category": "Vision", "metric": "%", "num_problems": null, "source_url": "https://huggingface.co/moonshotai/Kimi-K2.5", "canonical_setting": { "version": "MotionBench", "metric_type": "pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": true, "tools": "none", "notes": "Per Kimi K2.5 model card." } }, { "id": "mrcr_v1", "name": "MRCR v1", "category": "Long-context", "source_url": "https://storage.googleapis.com/deepmind-media/gemini/gemini_v1_5_report.pdf", "num_problems": 2000, "canonical_setting": { "version": "MRCR v1 at up to 1M context", "metric_type": "pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": false, "tools": "none", "judge": "difflib SequenceMatcher string-similarity ratio", "sampling": "pass@1", "notes": "Official source is the Gemini 1.5 technical report. MRCR presents a long user-model conversation with two adversarially similar writing requests and asks the model to reproduce the response associated with a target request. Figure 12 reports cumulative average string-similarity score as a function of context length over 2,000 MRCR instances, up to 1M tokens. This v1 1M-context setting is distinct from OpenAI MRCR v2 128k / 2-needle / 8-needle variants." } }, { "id": "mrcr_v2", "name": "MRCR v2", "category": "Long Context", "metric": "% correct", "num_problems": 2400, "source_url": "https://huggingface.co/datasets/openai/mrcr", "canonical_setting": { "version": "OpenAI MRCR v2 full dataset", "metric_type": "pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": false, "notes": "Official source is the openai/mrcr HuggingFace dataset. OpenAI MRCR v2 expands Gemini MRCR into an open long-context multiple-needle benchmark with 2, 4, or 8 identical asks hidden in a synthetic conversation. The dataset has 100 samples per bin, 8 token-length bins from 4k through 1M, and three needle settings, giving 3 * 8 * 100 = 2,400 rows. Many model cards report the 128k slice; the full released dataset also includes 262k, 524k, and 1M bins. tools=none preferred (pure-reasoning eval). If only with-tool scores (python/web/RAG) are available, accept and mark cell matches_canonical=false.", "tools": "none", "judge": "difflib SequenceMatcher string-similarity ratio with required hash prefix", "sampling": "pass@1" }, "cost": { "source_id": "helm_long_context_v1_0_0", "source_name": "HELM Long Context v1.0.0", "source_url": "https://storage.googleapis.com/crfm-helm-public/long-context/benchmark_output/releases/v1.0.0/groups.json", "source_benchmark_name": "OpenAI MRCR", "source_model_name": null, "source_model_slug": null, "evidence_scope": "HELM aggregate token table for a 100-instance OpenAI MRCR long-context run", "tokens": { "prompt_tokens": 0.0, "completion_tokens": 7658.16, "input_tokens": null, "output_tokens": null, "reasoning_tokens": null, "answer_tokens": null, "total_tokens": 7658.16 }, "total_cost_usd": null, "reported_items": 100, "reported_samples": null, "relative_tokens_to_source_anchor": 1.0, "relative_cost_to_source_anchor": null, "notes": "Candidate/proxy for BenchPress mrcr_v2. HELM reports a 100-instance OpenAI MRCR slice; BenchPress canonical metadata uses the full 2,400-row OpenAI MRCR v2 dataset." } }, { "id": "mrcr_v2_2needle_128k", "name": "OpenAI MRCR v2 (2 needle, 128k)", "category": "Long Context", "metric": "%", "num_problems": 500, "source_url": "https://huggingface.co/datasets/openai/mrcr", "canonical_setting": { "version": "OpenAI MRCR v2 (2 needle, 128k)", "metric_type": "pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": false, "tools": "none", "judge": "difflib SequenceMatcher string-similarity ratio with required hash prefix", "sampling": "pass@1", "notes": "Official source is the openai/mrcr HuggingFace dataset, with results reported in the OpenAI GPT-4.1 blog. This row is the 2-needle 128k slice: one needle setting, five token-length bins up to 131,072 tokens, and 100 samples per bin, giving 500 scored prompts. The full 2-needle dataset has 800 rows across all eight bins up to 1M." }, "cost": { "source_id": "helm_long_context_v1_0_0", "source_name": "HELM Long Context v1.0.0", "source_url": "https://storage.googleapis.com/crfm-helm-public/long-context/benchmark_output/releases/v1.0.0/groups.json", "source_benchmark_name": "OpenAI MRCR", "source_model_name": null, "source_model_slug": null, "evidence_scope": "reported HELM prompt/completion token totals for the OpenAI MRCR source benchmark scenario", "tokens": { "prompt_tokens": 0.0, "completion_tokens": 7658.16, "input_tokens": null, "output_tokens": null, "reasoning_tokens": null, "answer_tokens": null, "total_tokens": 7658.16 }, "total_cost_usd": null, "reported_items": 100.0, "reported_samples": null, "relative_tokens_to_source_anchor": 1.0, "relative_cost_to_source_anchor": null, "notes": "HELM token evidence only; source row is OpenAI MRCR family-level and not explicitly the BenchPress 2-needle 128k slice. Use as proxy/candidate unless exact slice mapping is accepted." } }, { "id": "mrcr_v2_2needle_1m", "name": "OpenAI MRCR v2 (2 needle, 1M)", "category": "Long Context", "metric": "%", "num_problems": null, "source_url": "https://openai.com/index/gpt-4-1/", "canonical_setting": { "version": "OpenAI MRCR v2 (2 needle, 1M)", "metric_type": "pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": false, "tools": "none", "notes": "Per OpenAI GPT-4.1 blog." } }, { "id": "mrcr_v2_2needle_256k", "name": "OpenAI MRCR v2 (2-needle, 256k)", "description": "OpenAI Multi-hop Retrieval and Comprehension v2, 2-needle, 256k context window", "canonical_setting": { "judge": "rule-based", "notes": "Per OpenAI GPT-5 developer blog https://openai.com/index/introducing-gpt-5-for-developers/" }, "category": "Long Context", "cost": { "source_id": "helm_long_context_v1_0_0", "source_name": "HELM Long Context v1.0.0", "source_url": "https://storage.googleapis.com/crfm-helm-public/long-context/benchmark_output/releases/v1.0.0/groups.json", "source_benchmark_name": "OpenAI MRCR", "source_model_name": null, "source_model_slug": null, "evidence_scope": "reported HELM prompt/completion token totals for the OpenAI MRCR source benchmark scenario", "tokens": { "prompt_tokens": 0.0, "completion_tokens": 7658.16, "input_tokens": null, "output_tokens": null, "reasoning_tokens": null, "answer_tokens": null, "total_tokens": 7658.16 }, "total_cost_usd": null, "reported_items": 100.0, "reported_samples": null, "relative_tokens_to_source_anchor": 1.0, "relative_cost_to_source_anchor": null, "notes": "HELM token evidence only; source row is OpenAI MRCR family-level and not explicitly the BenchPress 2-needle 256k slice. Use as proxy/candidate unless exact slice mapping is accepted." } }, { "id": "mrcr_v2_8needle", "name": "OpenAI MRCR v2 (8-needle)", "category": "Long Context", "metric": "%", "num_problems": 800, "source_url": "https://huggingface.co/datasets/openai/mrcr", "canonical_setting": { "version": "OpenAI MRCR v2 (8-needle)", "metric_type": "pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": false, "tools": "none", "judge": "difflib SequenceMatcher string-similarity ratio with required hash prefix", "sampling": "pass@1", "notes": "Official source is the openai/mrcr HuggingFace dataset. This row is the 8-needle slice: one needle setting, eight token-length bins from 4k through 1M, and 100 samples per bin, giving 800 scored prompts. Some model cards report restricted context slices, but the benchmark variant name here does not restrict to 128k." }, "cost": { "source_id": "helm_long_context_v1_0_0", "source_name": "HELM Long Context v1.0.0", "source_url": "https://storage.googleapis.com/crfm-helm-public/long-context/benchmark_output/releases/v1.0.0/groups.json", "source_benchmark_name": "OpenAI MRCR", "evidence_scope": "reported HELM completion-token total for the OpenAI MRCR long-context scenario, used as a task-family proxy for the BenchPress 8-needle MRCR row", "tokens": { "prompt_tokens": 0.0, "completion_tokens": 7658.16, "input_tokens": null, "output_tokens": null, "reasoning_tokens": null, "answer_tokens": null, "total_tokens": 7658.16 }, "total_cost_usd": null, "reported_items": 100.0, "reported_samples": null, "relative_tokens_to_source_anchor": 1.0, "relative_cost_to_source_anchor": null, "notes": "Proxy evidence: HELM reports an OpenAI MRCR scenario token total, while BenchPress names the OpenAI MRCR v2 8-needle slice. Use as MRCR task-family token evidence only; no dollar cost or hidden reasoning tokens are reported." } }, { "id": "mt_aime_2024", "name": "MT-AIME2024", "category": "Math", "metric": "%", "source_url": "https://huggingface.co/datasets/amphora/MCLM", "num_problems": 1650, "canonical_setting": { "version": "MCLM MT-AIME2024 (multilingual AIME 2024, 55 languages, Son et al. 2025)", "metric_type": "pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": false, "tools": "none", "judge": "rule-based verifier", "sampling": "pass@1", "notes": "Official source is the MCLM dataset and Son et al. (arXiv:2502.17407). MT-AIME2024 translates the full AIME 2024 set into 55 languages; the dataset stores 30 rows with 55 language columns, so the canonical evaluation contains 1,650 language-specific scored prompts." } }, { "id": "mt_bench_101", "name": "MT-Bench-101", "category": "Chat", "metric": "Score (1-10)", "num_problems": null, "source_url": "https://github.com/InternLM/InternLM", "canonical_setting": { "version": "MT-Bench-101 (Score 1-10)", "metric_type": "raw", "range": [ 1, 10 ], "higher_is_better": true, "multimodal_input": false, "tools": "none", "notes": "Per InternLM3 GitHub README. MT-Bench-101 scored 1-10." } }, { "id": "mtvqa", "name": "MTVQA", "category": "Vision VQA", "metric": "%", "num_problems": null, "source_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "canonical_setting": { "version": "MTVQA", "metric_type": "pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": true, "tools": "none", "notes": "Per Doubao Seed 2.0 Pro model card." } }, { "id": "muirbench", "name": "MUIRBench", "category": "Vision VQA", "metric": "%", "num_problems": null, "source_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "canonical_setting": { "version": "MUIRBench", "metric_type": "pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": true, "tools": "none", "notes": "Per Doubao Seed 2.0 Pro model card." } }, { "id": "multi_if", "name": "Multi-IF", "category": "Instruction Following", "metric": "%", "num_problems": 13503, "source_url": "https://huggingface.co/datasets/facebook/Multi-IF", "canonical_setting": { "version": "Multi-IF (8-language, 3-turn conversations)", "metric_type": "pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": false, "tools": "none", "judge": "script-based verifiable-instruction checks", "sampling": "single response per turn", "notes": "Official sources are the facebook/Multi-IF HuggingFace dataset and He et al. (arXiv:2410.15553). The dataset has 4,501 multilingual conversations across 8 languages, and each conversation has three turns; the cost proxy counts the 13,503 model-turn generations that must be evaluated. The reported metric averages instruction-level strict accuracy, conversation-level strict accuracy, instruction-level loose accuracy, and conversation-level loose accuracy across languages and turns." } }, { "id": "multi_swe_bench", "name": "Multi-SWE-bench", "category": "Coding", "metric": "%", "source_url": "https://huggingface.co/datasets/ByteDance-Seed/Multi-SWE-bench", "num_problems": 1632, "canonical_setting": { "version": "Multi-SWE-bench (full 7-language issue-resolving benchmark)", "metric_type": "pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": false, "tools": "repository checkout plus Docker/unit-test execution", "judge": "execution-based patch validation", "sampling": "pass@1", "notes": "Official sources are the ByteDance-Seed/Multi-SWE-bench HuggingFace dataset and Zan et al. (arXiv:2504.02605). The full benchmark covers Java, TypeScript, JavaScript, Go, Rust, C, and C++ with 1,632 human-validated issue-resolving instances curated from 2,456 candidates. Later auxiliary releases such as mini, flash, RL, and Python supplement files are not counted in this canonical full benchmark row." } }, { "id": "multichallenge", "name": "MultiChallenge", "category": "Instruction Following", "metric": "%", "num_problems": 273, "source_url": "https://github.com/ekwinox117/multi-challenge", "canonical_setting": { "version": "MultiChallenge", "metric_type": "pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": false, "tools": "none", "judge": "automated LLM judge with instance-level rubrics", "sampling": "single final response per conversation", "notes": "Official sources are the MultiChallenge paper (arXiv:2501.17399) and the released benchmark_questions.jsonl in the project repository. The benchmark contains 273 maximum-10-turn test conversations: 113 inference-memory, 69 instruction-retention, 41 reliable-version-editing, and 50 self-coherence conversations. Cost counts one model completion per conversation because each row provides conversation history ending in a target question." } }, { "id": "multichallenge_o3mini_grader", "name": "MultiChallenge (o3-mini grader)", "category": "Instruction Following", "metric": "%", "num_problems": 273, "source_url": "https://github.com/ekwinox117/multi-challenge", "canonical_setting": { "version": "Scale MultiChallenge official GitHub benchmark_questions.jsonl / paper Table 1", "metric_type": "pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": false, "tools": "none", "judge": "o3-mini grader / LLM-as-judge with instance-level binary rubrics", "sampling": "attempts=1 unless otherwise reported", "notes": "MultiChallenge has 273 test conversations in the paper and official GitHub data. Each item requires one model response to a multi-turn conversation history, then an LLM grader evaluates the final response against an instance-level rubric. HF currently reports 266 rows, treated as a conflicting mirror/snapshot rather than the canonical paper/repo count." } }, { "id": "multilingual_mmlu", "name": "Multilingual MMLU", "category": "Multilingual", "source_url": "https://huggingface.co/microsoft/Phi-4-mini-instruct", "num_problems": null, "canonical_setting": { "version": "Multilingual MMLU (5-shot)", "metric_type": "pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": false, "notes": "5-shot MMLU across multiple languages." } }, { "id": "multipl_e_avg", "name": "MultiPL-E (average)", "category": "Coding", "metric": "%", "num_problems": 12667, "source_url": "https://huggingface.co/datasets/nuprl/MultiPL-E", "canonical_setting": { "version": "MultiPL-E full public HF dataset, averaged across language/config rows", "metric_type": "pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": false, "tools": "code execution", "sampling": "0-shot/pass@1 as reported by Mistral Medium 3 blog", "notes": "MultiPL-E is a multilingual code-generation benchmark translated from HumanEval and MBPP. The HF dataset-server reports 12,667 total test rows across 47 configs (3,811 HumanEval rows and 8,856 MBPP rows). If the score source used only a HumanEval subset, 3,811 is the narrower count; either interpretation is Tier 3 under the cost proxy." } }, { "id": "nl2repo_bench", "name": "NL2Repo-Bench", "category": "Repository Code", "metric": "%", "num_problems": 104, "source_url": "https://arxiv.org/abs/2512.12730", "canonical_setting": { "version": "NL2Repo-Bench full benchmark", "metric_type": "pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": false, "tools": "agentic code execution", "sampling": "Pass@1 unless otherwise reported", "notes": "NL2Repo-Bench contains 104 repository-generation tasks. Each task gives a natural-language requirements document and empty workspace; generated repositories are evaluated with original upstream pytest suites. Item count is task instances, not upstream test cases." } }, { "id": "nl2repo_pass1", "name": "NL2Repo (Pass@1)", "category": "Repository Code", "metric": "%", "num_problems": null, "source_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "canonical_setting": { "version": "NL2Repo (Pass@1)", "metric_type": "pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": false, "tools": "code execution", "notes": "Per Doubao Seed 2.0 Pro model card." } }, { "id": "ntrex", "name": "NTREX", "category": "Multilingual", "source_url": "https://cohere.com/research/papers/command-a-technical-report.pdf", "num_problems": null, "canonical_setting": { "version": "NTREX (COMET-20)", "metric_type": "pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": false, "notes": "Machine translation COMET-20 score." } }, { "id": "ocrbench", "name": "OCRBench", "category": "Multimodal", "source_url": "https://huggingface.co/moonshotai/Kimi-K2.5", "num_problems": null, "canonical_setting": { "version": "OCRBench", "metric_type": "pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": true, "notes": "OCR benchmark." } }, { "id": "ocrbench_v2", "name": "OCRBench v2", "category": "Document/Chart", "metric": "%", "num_problems": null, "source_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "canonical_setting": { "version": "OCRBench v2", "metric_type": "pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": true, "tools": "none", "notes": "Per Doubao Seed 2.0 Pro model card." } }, { "id": "odvbench", "name": "ODVBench", "category": "Video", "metric": "%", "num_problems": null, "source_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "canonical_setting": { "version": "ODVBench", "metric_type": "pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": true, "tools": "none", "notes": "Per Doubao Seed 2.0 Pro model card." } }, { "id": "officeqa", "name": "OfficeQA", "category": "Office", "metric": "exact-match accuracy (%)", "num_problems": 246, "source_url": "https://www.databricks.com/blog/introducing-officeqa-benchmark-end-to-end-grounded-reasoning", "canonical_setting": { "version": "OfficeQA", "metric_type": "pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": false, "tools": "agentic", "notes": "OfficeQA productivity benchmark (different from OfficeQA Pro). Per OpenAI GPT-5.4 blog." } }, { "id": "officeqa_pro", "name": "OfficeQA Pro", "category": "Office", "metric": "exact-match accuracy (%)", "num_problems": 133, "source_url": "https://arxiv.org/abs/2603.08655", "canonical_setting": { "version": "OfficeQA Pro 133-question release", "metric_type": "pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": true, "tools": "document and rendered-image analysis", "sampling": "pass@1", "judge": "OfficeQA Pro exact-match accuracy", "harness": "OfficeQA Pro official harness", "notes": "Every PDF is rendered as images; no machine-readable PDF text is supplied." }, "cost": { "source_id": "officeqa_pro_paper_table1_agent_cost_per_sample", "source_name": "OfficeQA Pro paper Table 1", "source_url": "https://arxiv.org/abs/2603.08655", "source_data_url": "https://arxiv.org/html/2603.08655v1", "source_benchmark_name": "OfficeQA Pro", "source_model_name": "Claude Opus 4.6, full corpus PDF setting", "source_model_slug": "claude-opus-4.6; corpus=full; format=pdf", "evidence_scope": "paper-reported average agent dollar cost per sample; derived total over 133 questions", "tokens": { "prompt_tokens": null, "completion_tokens": null, "input_tokens": null, "output_tokens": null, "reasoning_tokens": null, "answer_tokens": null, "total_tokens": null }, "total_cost_usd": 605.15, "reported_items": 133, "reported_samples": 1, "relative_tokens_to_source_anchor": null, "relative_cost_to_source_anchor": null, "notes": "OfficeQA Pro Table 1 reports Claude Opus 4.6 full-corpus PDF cost of $4.55 per sample over 133 questions, so total is $4.55*133=$605.15. Agent/framework/model/configuration-specific dollar evidence; raw token counts are not reported." } }, { "id": "ojbench", "name": "OJBench", "category": "Coding", "source_url": "https://arxiv.org/abs/2506.16395", "num_problems": 232, "canonical_setting": { "version": "OJBench Pass@1", "metric_type": "pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": false, "tools": "code execution", "sampling": "Pass@1 for the BenchPress row; official paper also reports Pass@8 in separate settings", "notes": "OJBench comprises 232 NOI/ICPC programming competition problems. The BenchPress row follows score sources that report OJBench (Pass@1), so the source-backed model-generation count is 232 rather than Pass@8 or dual-language variants." }, "metric": "%" }, { "id": "omnidocbench", "name": "OmniDocBench (normalized edit distance, lower is better)", "category": "Vision", "metric": "edit distance (lower=better)", "num_problems": 1651, "source_url": "https://huggingface.co/datasets/opendatalab/OmniDocBench", "canonical_setting": { "version": "OmniDocBench v1.6 full benchmark", "metric_type": "normalized_edit_distance", "range": [ 0, 1 ], "higher_is_better": false, "multimodal_input": true, "tools": "none", "notes": "Official OmniDocBench v1.6 contains 1,651 PDF pages. Count one model output per page for document parsing; HF parquet row count may differ slightly, but official README/page count is canonical. Stored scores use normalized edit distance on [0,1]; Kimi K3 reports 1-NED percentages, which are converted back to NED." } }, { "id": "omnidocbench_1.5", "name": "OmniDocBench 1.5", "category": "Vision", "metric": "normalized edit distance (lower=better)", "num_problems": 1355, "source_url": "https://github.com/opendatalab/OmniDocBench", "canonical_setting": { "version": "OmniDocBench v1.5", "metric_type": "normalized_edit_distance", "higher_is_better": false, "range": [ 0, 1 ], "multimodal_input": true, "tools": "none", "notes": "Average normalized edit distance; lower is better. The 1,355-page count is inferred from the official v1.6 update history." } }, { "id": "omnimath", "name": "OmniMath", "category": "Math", "source_url": "https://arxiv.org/abs/2410.07985", "num_problems": null, "canonical_setting": { "version": "OmniMath", "metric_type": "pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": false, "notes": "Olympiad-level math benchmark." }, "cost": { "source_id": "helm_capabilities_v1_15_0", "source_name": "HELM Capabilities v1.15.0", "source_url": "https://raw.githubusercontent.com/stanford-crfm/helm/main/docs/benchmark.md", "source_benchmark_name": "Omni-MATH", "evidence_scope": "observed model-inference totals for the source evaluation run", "tokens": { "prompt_tokens": 22672.914, "completion_tokens": 253765.647, "total_tokens": 276438.561 }, "total_cost_usd": null, "relative_tokens_to_source_anchor": 3.121, "relative_cost_to_source_anchor": null, "notes": "Inference-side token/cost evidence only; judge, human-review, tool/environment, and infrastructure costs are separate protocol factors." } }, { "id": "osworld", "name": "OSWorld", "category": "Agentic", "metric": "% success", "num_problems": 369, "source_url": "https://os-world.github.io/", "canonical_setting": { "version": "OSWorld (369)", "metric_type": "pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": true, "notes": "tools=agentic (scaffold-defined). Preferred = official harness tools (bash/editor for SWE-bench, browser for OSWorld/BrowseComp, official APIs for tau-bench/MCPAtlas/Toolathlon, terminal for Terminal-Bench). Non-official scaffolds → matches_canonical=false.", "tools": "agentic" } }, { "id": "ovbench", "name": "OVBench", "category": "Video", "metric": "%", "num_problems": null, "source_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "canonical_setting": { "version": "OVBench", "metric_type": "pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": true, "tools": "none", "notes": "Per Doubao Seed 2.0 Pro model card." } }, { "id": "ovobench", "name": "OVO-Bench", "category": "Streaming Video", "metric": "aggregate online-video score (%)", "num_problems": 2814, "source_url": "https://arxiv.org/abs/2501.05510", "canonical_setting": { "version": "OVO-Bench", "metric_type": "pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": true, "tools": "none", "sampling": "pass@1", "judge": "task-specific rule/timing evaluation", "harness": "official", "notes": "2,814 meta-annotations over 644 videos." } }, { "id": "paperbench", "name": "PaperBench", "category": "Coding", "source_url": "https://arxiv.org/abs/2507.20534", "num_problems": null, "canonical_setting": { "version": "PaperBench Code-Dev", "metric_type": "pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": false, "notes": "Code dev from papers." }, "cost": { "source_id": "paperbench_code_dev_o3mini_simplejudge_and_rollout_cost", "source_name": "PaperBench paper SimpleJudge and Code-Dev cost discussion", "source_url": "https://arxiv.org/abs/2504.01848", "source_data_url": "https://arxiv.org/html/2504.01848v3", "source_benchmark_name": "PaperBench Code-Dev", "source_model_name": "o3-mini SimpleJudge", "source_model_slug": "o3-mini-2025-01-31;reasoning_effort=high;simplejudge", "evidence_scope": "paper-reported Code-Dev judge cost and expected rollout cost over 20 papers", "tokens": { "prompt_tokens": null, "completion_tokens": null, "input_tokens": null, "output_tokens": null, "reasoning_tokens": null, "answer_tokens": null, "total_tokens": null }, "total_cost_usd": 4200.0, "reported_items": 20, "reported_samples": 1, "relative_tokens_to_source_anchor": null, "relative_cost_to_source_anchor": null, "notes": "Paper reports PaperBench Code-Dev grading costs are reduced to $10 per paper on average and Code-Dev rollouts would cost $4,000 per eval run. Derived total stored includes $200 judge cost plus $4,000 rollout cost. Code-Dev token counts are not directly reported." } }, { "id": "phibench", "name": "PhiBench (Microsoft Internal)", "category": "General", "source_url": "https://arxiv.org/abs/2412.08905", "num_problems": null, "canonical_setting": { "version": "PhiBench 2.21 (Microsoft internal)", "metric_type": "pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": false, "notes": "Microsoft Phi team internal eval." } }, { "id": "phybench", "name": "Phybench", "category": "Physics", "metric": "%", "num_problems": null, "source_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "canonical_setting": { "version": "Phybench", "metric_type": "pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": false, "tools": "none", "notes": "Per Doubao Seed 2.0 Pro model card." } }, { "id": "phyx_openended", "name": "PhyX (open-ended)", "category": "Vision STEM", "metric": "%", "num_problems": null, "source_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "canonical_setting": { "version": "PhyX (open-ended)", "metric_type": "pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": true, "tools": "none", "notes": "Per Doubao Seed 2.0 Pro model card." } }, { "id": "point_bench", "name": "Point-Bench", "category": "Vision Counting", "metric": "%", "num_problems": null, "source_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "canonical_setting": { "version": "Point-Bench", "metric_type": "pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": true, "tools": "none", "notes": "Per Doubao Seed 2.0 Pro model card." } }, { "id": "popqa", "name": "PopQA", "category": "QA", "source_url": "https://huggingface.co/datasets/akariasai/PopQA", "num_problems": 14267, "canonical_setting": { "version": "PopQA test set", "metric_type": "pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": false, "tools": "none", "notes": "Official PopQA HuggingFace dataset contains 14,267 test rows. Count one factual QA generation per row; do not use rounded 14k marketing count." } }, { "id": "procbench", "name": "ProcBench", "category": "Reasoning", "metric": "%", "num_problems": null, "source_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "canonical_setting": { "version": "ProcBench", "metric_type": "pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": false, "tools": "none", "notes": "Per Doubao Seed 2.0 Pro model card." } }, { "id": "realworldqa", "name": "RealWorldQA", "category": "Vision Perception", "metric": "%", "num_problems": null, "source_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "canonical_setting": { "version": "RealWorldQA", "metric_type": "pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": true, "tools": "none", "notes": "Per Doubao Seed 2.0 Pro model card." } }, { "id": "refspatialbench", "name": "RefSpatialBench", "category": "Vision Spatial", "metric": "%", "num_problems": null, "source_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "canonical_setting": { "version": "RefSpatialBench", "metric_type": "pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": true, "tools": "none", "notes": "Per Doubao Seed 2.0 Pro model card." } }, { "id": "repoqa", "name": "RepoQA", "category": "Coding", "source_url": "https://arxiv.org/abs/2406.06025", "num_problems": 500, "canonical_setting": { "version": "RepoQA SNF, 32K context, threshold 0.8", "metric_type": "pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": false, "tools": "none", "notes": "RepoQA contains 500 code-search tasks from 50 repositories across 5 languages. Count task instances rather than repositories or candidate functions." } }, { "id": "researchrubrics", "name": "ResearchRubrics", "category": "Deep Research", "metric": "weighted rubric compliance score (%)", "num_problems": 101, "source_url": "https://huggingface.co/datasets/ScaleAI/researchrubrics/tree/85de3115053d1453ed612caacf4a405edc1ad756", "canonical_setting": { "version": "ResearchRubrics official 101-task release", "metric_type": "pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": false, "tools": "research tools", "sampling": "one report per task", "judge": "binary rubric satisfaction with positive-weight average", "harness": "official ResearchRubrics evaluation pipeline", "notes": "The pinned processed_data.jsonl contains 101 research tasks. Each report is scored against weighted binary rubrics." } }, { "id": "ruler_128k", "name": "RULER 128K", "category": "Long Context", "metric": "accuracy (%)", "num_problems": 6500, "source_url": "https://github.com/NVIDIA/RULER", "canonical_setting": { "version": "RULER v1 13-task suite at 128K", "metric_type": "pct", "higher_is_better": true, "range": [ 0, 100 ], "multimodal_input": false, "tools": "none", "notes": "13 tasks times 500 generated examples." }, "cost": { "source_id": "helm_long_context_v1_0_0_ruler_proxy", "source_name": "HELM Long Context v1.0.0", "source_url": "https://storage.googleapis.com/crfm-helm-public/long-context/benchmark_output/releases/v1.0.0/groups.json", "source_benchmark_name": "RULER HotPotQA", "evidence_scope": "reported HELM prompt/completion token totals for a RULER long-context source row; proxy for BenchPress RULER 128K", "tokens": { "prompt_tokens": 6068122.679999996, "completion_tokens": 122.49, "input_tokens": null, "output_tokens": null, "reasoning_tokens": null, "answer_tokens": null, "total_tokens": 6068245.169999996 }, "total_cost_usd": null, "reported_items": 100.0, "reported_samples": null, "relative_tokens_to_source_anchor": null, "relative_cost_to_source_anchor": null, "notes": "HELM token evidence only. This maps a RULER subtask row (HotPotQA) to BenchPress ruler_128k as a benchmark-length proxy, not the full RULER suite. The same source also reports RULER SQuAD at 5,650,821.36 total tokens." } }, { "id": "ruler_32k", "name": "RULER 32K", "category": "Long Context", "metric": "accuracy (%)", "num_problems": 6500, "source_url": "https://github.com/NVIDIA/RULER", "canonical_setting": { "version": "RULER v1 13-task suite at 32K", "metric_type": "pct", "higher_is_better": true, "range": [ 0, 100 ], "multimodal_input": false, "tools": "none", "notes": "13 tasks times 500 generated examples." } }, { "id": "safety", "name": "Safety (OLMES suite)", "category": "Safety", "source_url": "https://arxiv.org/abs/2501.00656", "num_problems": null, "canonical_setting": { "version": "OLMES safety suite", "metric_type": "pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": false, "notes": "Allen AI internal safety eval suite." } }, { "id": "scicode", "name": "SciCode", "category": "Coding", "metric": "% correct", "num_problems": 338, "source_url": "https://scicode-bench.github.io/", "canonical_setting": { "version": "SciCode full subproblem benchmark", "metric_type": "pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": false, "tools": "code execution", "notes": "SciCode contains 338 executable scientific-code subproblems. Count subproblems because each requires a code solution evaluated by tests." }, "cost": { "source_id": "artificial_analysis_eval_token_cost_gemini_2_5_pro_anchor", "source_name": "Artificial Analysis per-evaluation token usage and cost with Gemini 2.5 Pro anchor", "source_url": "https://artificialanalysis.ai/evaluations/scicode", "source_benchmark_name": "SciCode", "source_model_name": "Gemini 2.5 Pro", "source_model_slug": "gemini-2-5-pro", "evidence_scope": "observed model-inference totals for the source evaluation run", "tokens": { "prompt_tokens": null, "completion_tokens": null, "input_tokens": 1053511, "output_tokens": 2112372, "reasoning_tokens": 1909473, "answer_tokens": 202899, "total_tokens": 3165883 }, "total_cost_usd": 22.440608750000003, "reported_items": null, "reported_samples": null, "relative_tokens_to_source_anchor": 1.954, "relative_cost_to_source_anchor": 1.423, "notes": "Inference-side token/cost evidence only; source_model_* identifies the reported model row. Dollar cost is run/model specific, not benchmark-intrinsic. primary anchor model: Gemini 2.5 Pro" } }, { "id": "screenspot_pro", "name": "ScreenSpot-Pro", "category": "Multimodal", "source_url": "https://github.com/likaixin2000/ScreenSpot-Pro-GUI-Grounding", "num_problems": 1581, "canonical_setting": { "version": "ScreenSpot-Pro full benchmark", "metric_type": "pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": true, "tools": "none", "notes": "ScreenSpot-Pro contains 1,581 GUI grounding targets: 604 icon targets and 977 text targets. Count one model grounding response per target." } }, { "id": "seal_0", "name": "Seal-0", "category": "Search Agent", "metric": "%", "num_problems": null, "source_url": "https://huggingface.co/moonshotai/Kimi-K2.5", "canonical_setting": { "version": "Seal-0", "metric_type": "pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": false, "tools": "agentic", "notes": "Per Kimi K2.5 model card." } }, { "id": "sfe", "name": "SFE", "category": "Vision STEM", "metric": "%", "num_problems": null, "source_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "canonical_setting": { "version": "SFE", "metric_type": "pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": true, "tools": "none", "notes": "Per Doubao Seed 2.0 Pro model card." } }, { "id": "simplebench", "name": "SimpleBench", "category": "Reasoning", "metric": "% correct", "num_problems": 1000, "source_url": "https://simple-bench.com/", "canonical_setting": { "version": "SimpleBench (1000)", "metric_type": "pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": false, "notes": "tools=none preferred (pure-reasoning eval). If only with-tool scores (python/web/RAG) are available, accept and mark cell matches_canonical=false.", "tools": "none" } }, { "id": "simpleqa", "name": "SimpleQA", "category": "Knowledge", "metric": "% correct", "num_problems": 4326, "source_url": "https://openai.com/index/introducing-simpleqa/", "canonical_setting": { "version": "SimpleQA (OpenAI 4326 questions)", "metric_type": "pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": false, "notes": "tools=none preferred (pure-reasoning eval). If only with-tool scores (python/web/RAG) are available, accept and mark cell matches_canonical=false.", "tools": "none" } }, { "id": "simpleqa_verified", "name": "SimpleQA-Verified", "category": "Knowledge", "metric": "% correct (pass@1)", "num_problems": null, "source_url": "https://huggingface.co/deepseek-ai/DeepSeek-V4-Pro", "canonical_setting": { "version": "SimpleQA-Verified", "metric_type": "pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": false, "notes": "tools=none preferred (pure-reasoning eval). If only with-tool scores (python/web/RAG) are available, accept and mark cell matches_canonical=false.", "tools": "none" } }, { "id": "simplevqa", "name": "SimpleVQA", "category": "Vision", "metric": "accuracy (%)", "num_problems": 2025, "source_url": "https://huggingface.co/datasets/m-a-p/SimpleVQA/tree/037cf89fb6f1212691756b66d1ecde6c2ce89e54", "canonical_setting": { "version": "SimpleVQA official 2,025-item test release", "metric_type": "pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": true, "tools": "none", "sampling": "pass@1", "judge": "LLM-as-judge", "harness": "official SimpleVQA", "notes": "The pinned official simpleVQA_final_modified.json contains 2,025 items. Tool-assisted observations remain score-level non-default settings." } }, { "id": "smt_2025", "name": "SMT 2025", "category": "Math", "metric": "% correct (pass@1)", "num_problems": 53, "source_url": "https://huggingface.co/datasets/MathArena/smt_2025", "canonical_setting": { "version": "SMT 2025", "metric_type": "pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": false, "sampling": "samples=4", "notes": "tools=none preferred (pure-reasoning eval). If only with-tool scores (python/web/RAG) are available, accept and mark cell matches_canonical=false.", "tools": "none" }, "cost": { "source_id": "matharena_smt_2025_outputs_kimi_k2_thinking_cost", "source_name": "MathArena SMT 2025 output logs for Kimi K2 Thinking", "source_url": "https://matharena.ai/competition_tables/smt--smt_2025", "source_data_url": "https://huggingface.co/datasets/MathArena/smt_2025_outputs/resolve/main/data/train-00000-of-00001.parquet", "source_benchmark_name": "SMT 2025", "source_model_name": "Kimi K2 Thinking", "source_model_slug": "moonshot/k2-thinking", "evidence_scope": "official output-log rows with input/output tokens and model-price-derived cost for all 53 problems x 4 samples", "tokens": { "prompt_tokens": null, "completion_tokens": null, "input_tokens": 31809, "output_tokens": 4563498, "reasoning_tokens": null, "answer_tokens": null, "total_tokens": 4595307 }, "total_cost_usd": 11.4278304, "reported_items": 53, "reported_samples": 4, "relative_tokens_to_source_anchor": null, "relative_cost_to_source_anchor": null, "notes": "Rows cover all 53 SMT 2025 problems with 4 samples each for Kimi K2 Thinking. Cost uses MathArena model config pricing: 31,809 input tokens*$0.60/M plus 4,563,498 output tokens*$2.50/M = $11.4278304." } }, { "id": "spreadsheetbench_verified", "name": "SpreadsheetBench Verified", "category": "Coding", "metric": "%", "num_problems": null, "source_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "canonical_setting": { "version": "SpreadsheetBench Verified", "metric_type": "pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": false, "tools": "code execution", "notes": "Per Doubao Seed 2.0 Pro model card." } }, { "id": "superchem", "name": "Superchem (text-only)", "category": "Chemistry", "metric": "%", "num_problems": null, "source_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "canonical_setting": { "version": "Superchem (text-only)", "metric_type": "pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": false, "tools": "none", "notes": "Per Doubao Seed 2.0 Pro model card." } }, { "id": "supergpqa", "name": "SuperGPQA", "category": "Science", "metric": "%", "num_problems": 26529, "source_url": "https://huggingface.co/datasets/m-a-p/SuperGPQA", "canonical_setting": { "version": "SuperGPQA full benchmark", "metric_type": "pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": false, "tools": "none", "notes": "Official SuperGPQA dataset has 26,529 rows/questions spanning science disciplines. Count one answer per row." } }, { "id": "swe_bench_multilingual", "name": "SWE-bench Multilingual", "category": "Coding", "metric": "% resolved (pass@1)", "num_problems": 300, "source_url": "https://huggingface.co/datasets/SWE-bench/SWE-bench_Multilingual/tree/846e647b9f33c0b51b739d005d13d85493c9af09", "canonical_setting": { "version": "SWE-bench Multilingual official 300-instance test set", "metric_type": "pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": false, "tools": "repository shell/editor agent", "sampling": "pass@1", "judge": "language-appropriate isolated test suites", "harness": "source-reported fixed coding scaffold", "notes": "The pinned official dataset contains 300 issues across nine languages. The StepFun label SWE-MTLG is mapped to this released identity; scaffold remains score-level provenance." }, "cost": { "source_id": "swe_bench_multilingual_leaderboard_gemini_3_pro_cost", "source_name": "SWE-bench Multilingual leaderboard reported run cost", "source_url": "https://www.swebench.com/", "source_benchmark_name": "SWE-bench Multilingual", "source_model_name": "Gemini 3 Pro", "source_model_slug": "mini-swe-agent-v2.0.0a0; gemini-3-pro", "evidence_scope": "official embedded leaderboard run-cost row for the 300-instance Multilingual split", "tokens": { "prompt_tokens": null, "completion_tokens": null, "input_tokens": null, "output_tokens": null, "reasoning_tokens": null, "answer_tokens": null, "total_tokens": null }, "total_cost_usd": 306.3256754, "reported_items": 300, "reported_samples": null, "relative_tokens_to_source_anchor": null, "relative_cost_to_source_anchor": null, "notes": "Dollar-only evidence. Embedded row also reports instance_cost=1.0210855846666667, instance_calls=48.85, and 300 per_instance_details with per-instance cost/api_calls; no token totals. Cost is scaffold/model specific." } }, { "id": "swe_bench_multimodal", "name": "SWE-bench Multimodal", "category": "Coding", "metric": "% resolved", "num_problems": null, "source_url": "https://www.swebench.com/", "canonical_setting": { "version": "SWE-bench Multimodal", "metric_type": "pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": true, "notes": "tools=agentic. No single standard public scaffold exists for SWE-bench Multimodal; harness choice is model-side (recorded in cell.reported_setting.harness). Any lab-published harness counts as canonical.", "tools": "agentic" }, "cost": { "source_id": "swe_bench_multimodal_paper_table3_avg_cost_per_task", "source_name": "SWE-bench Multimodal paper Table 3", "source_url": "https://arxiv.org/html/2410.03859", "source_data_url": "https://www.swebench.com/multimodal.html", "source_benchmark_name": "SWE-bench Multimodal", "source_model_name": "RAG + Claude 3.5 Sonnet baseline", "source_model_slug": "rag;claude-3.5-sonnet", "evidence_scope": "paper-reported average per-task inference cost; derived full-run cost over 517 official instances", "tokens": { "prompt_tokens": null, "completion_tokens": null, "input_tokens": null, "output_tokens": null, "reasoning_tokens": null, "answer_tokens": null, "total_tokens": null }, "total_cost_usd": 77.55, "reported_items": 517, "reported_samples": 1, "relative_tokens_to_source_anchor": null, "relative_cost_to_source_anchor": null, "notes": "Paper Table 3 reports average per-task inference costs for SWE-bench Multimodal baselines; the lowest reported source-backed row is RAG + Claude 3.5 Sonnet at $0.15/task. Derived full-run cost is $0.15*517=$77.55. Cost is baseline/harness-specific, not a universal leaderboard cost; token counts are not reported." } }, { "id": "swe_bench_pro", "name": "SWE-bench Pro", "category": "Coding", "metric": "% resolved (pass@1)", "num_problems": 731, "source_url": "https://huggingface.co/datasets/ScaleAI/SWE-bench_Pro/tree/7ab5114912baf22bb098818e604c02fe7ad2c11f", "canonical_setting": { "version": "SWE-bench Pro public 731-instance test set", "metric_type": "pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": false, "tools": "repository shell/editor agent", "sampling": "pass@1", "judge": "isolated repository test suites", "harness": "source-reported fixed coding scaffold", "notes": "The pinned official public release contains 731 instances. Scaffold and effort remain score-level settings." }, "cost": { "source_id": "swe_bench_pro_nilenso_trajectory_report_gpt5_high", "source_name": "SWE-Bench Pro public trajectory cost/token analysis", "source_url": "https://nilenso.github.io/swe-bench-pro-cost-token-time-analysis/report.txt", "source_data_url": "https://nilenso.github.io/swe-bench-pro-cost-token-time-analysis/", "source_benchmark_name": "SWE-Bench Pro", "source_model_name": "GPT-5 + SWE-Agent", "source_model_slug": "gpt-5; agent=swe-agent; reasoning_effort=high", "evidence_scope": "derived totals from public Scale AI SWE-Bench Pro trajectories for the 616 paired submitted GPT-5 tasks", "tokens": { "prompt_tokens": null, "completion_tokens": null, "input_tokens": 1930805533, "output_tokens": 4374442, "reasoning_tokens": null, "answer_tokens": null, "total_tokens": 1935179975 }, "total_cost_usd": 1217.8967125, "reported_items": 616, "reported_samples": null, "relative_tokens_to_source_anchor": null, "relative_cost_to_source_anchor": null, "notes": "Third-party report derived from public Scale AI SWE-Bench Pro leaderboard trajectories. The text report explicitly gives 616 paired submitted tasks, mean GPT-5 input tokens=3,134,425, mean visible output tokens=7,101, and total GPT-5 cost=$1,217.90; exact totals here are summed from the interactive report's embedded per-instance DATA. Values cover the report's 616 paired submitted tasks, not all 731 public instances. Output tokens are visible output counted by tiktoken across response text and tool-call arguments, excluding GPT-5 hidden reasoning tokens. Dollar cost is Scale internal litellm proxy instance_cost and is model/run specific." } }, { "id": "swe_bench_verified", "name": "SWE-bench Verified", "category": "Coding", "metric": "% resolved (pass@1)", "num_problems": 500, "source_url": "https://huggingface.co/datasets/SWE-bench/SWE-bench_Verified/tree/78f471bf655a3137b2e8a75af1501690ec009ec3", "canonical_setting": { "version": "SWE-bench Verified official 500-instance test set", "metric_type": "pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": false, "tools": "repository shell/editor agent", "sampling": "pass@1", "judge": "isolated repository test suites", "harness": "source-reported fixed coding scaffold", "notes": "The pinned official Verified release contains 500 instances. Scaffold, effort, Advisor mode, and internal reproductions remain score-level settings." }, "cost": { "source_id": "swe_bench_verified_leaderboard_mini_swe_agent_claude45_opus_high_cost", "source_name": "SWE-bench Verified leaderboard reported run cost", "source_url": "https://www.swebench.com/", "source_benchmark_name": "SWE-bench Verified", "source_model_name": "mini-SWE-agent + Claude 4.5 Opus (high reasoning)", "source_model_slug": "mini-swe-agent; claude-4-5-opus; reasoning_effort=high", "evidence_scope": "reported leaderboard run cost for one 500-instance SWE-bench Verified row", "tokens": { "prompt_tokens": null, "completion_tokens": null, "input_tokens": null, "output_tokens": null, "reasoning_tokens": null, "answer_tokens": null, "total_tokens": null }, "total_cost_usd": 376.9539985, "reported_items": 500, "reported_samples": null, "relative_tokens_to_source_anchor": null, "relative_cost_to_source_anchor": null, "notes": "Dollar-only evidence from the official SWE-bench leaderboard embedded data. Token counts are not reported. Cost is model/scaffold/run specific; row also reports instance_cost=0.753907997 and instance_calls=32.896." } }, { "id": "swe_evo", "name": "SWE-Evo", "category": "Agentic Coding", "metric": "%", "num_problems": null, "source_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "canonical_setting": { "version": "SWE-Evo", "metric_type": "pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": false, "tools": "code execution", "notes": "Per Doubao Seed 2.0 Pro model card." } }, { "id": "swelancer", "name": "SWE-Lancer IC Diamond", "category": "Coding", "metric": "%", "num_problems": 198, "source_url": "https://github.com/openai/frontier-evals/tree/main/project/swelancer", "canonical_setting": { "version": "SWE-Lancer IC SWE Diamond, current verified offline release", "metric_type": "pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": false, "tools": "agentic", "sampling": "pass@1, one attempt per task", "judge": "end-to-end tests", "notes": "Current official SWE-Lancer release contains 198 verified-offline IC SWE Diamond tasks; original paper reported 237 IC SWE Diamond tasks and the current README says 39 were dropped. Excludes SWE Manager Diamond." } }, { "id": "swelancer_freelance_dollars", "name": "SWE-Lancer IC SWE Diamond Freelance ($)", "category": "Coding", "metric": "dollars", "num_problems": 198, "source_url": "https://github.com/openai/frontier-evals/tree/main/project/swelancer", "canonical_setting": { "version": "SWE-Lancer IC SWE Diamond Freelance ($), current verified offline release", "metric_type": "dollars", "range": [ 0, 200000 ], "higher_is_better": true, "multimodal_input": false, "tools": "agentic", "sampling": "pass@1, one attempt per task", "judge": "end-to-end tests", "notes": "Current official SWE-Lancer release contains 198 verified-offline IC SWE Diamond tasks; original paper reported 237 IC SWE Diamond tasks and the current README says 39 were dropped. Excludes SWE Manager Diamond." } }, { "id": "tau1_bench_avg", "name": "τ-bench (Yao 2024, averaged)", "category": "Tool use", "source_url": "https://cohere.com/research/papers/command-a-technical-report.pdf", "num_problems": null, "canonical_setting": { "version": "τ-bench averaged across retail+airline domains", "metric_type": "pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": false, "notes": "τ-bench (Yao 2024) averaged. Distinct from per-domain cells (tau_bench_retail/airline/telecom)." } }, { "id": "tau2_bench_airline", "name": "τ²-bench Airline", "category": "Agentic", "metric": "% success", "num_problems": 50, "source_url": "https://arxiv.org/abs/2506.07982", "canonical_setting": { "version": "τ²-bench (Sierra AI 2025) — airline domain, base split", "metric_type": "pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": false, "tools": "agentic", "sampling": "pass^1 / one trial per task", "judge": "state-based task success", "notes": "Paper Table 1 and current official split file both give 50 Airline tasks (30 train + 20 test). Dual-control text setting: LLM-controlled agent and simulated user; not comparable to original tau-bench." }, "cost": { "source_id": "tau2_bench_paper_gpt41_agent_user_sim_per_task_cost", "source_name": "tau2-bench paper reported GPT-4.1 agent and user-simulator cost", "source_url": "https://arxiv.org/abs/2506.07982", "source_data_url": "https://arxiv.org/e-print/2506.07982", "source_benchmark_name": "tau2-bench Airline", "source_model_name": "gpt-4.1-2025-04-14 agent + user simulator", "source_model_slug": "gpt-4.1-2025-04-14-agent-user-sim", "evidence_scope": "derived one-trial airline-domain dollar cost from paper-reported average per-task agent and user-simulator API costs", "tokens": { "prompt_tokens": null, "completion_tokens": null, "input_tokens": null, "output_tokens": null, "reasoning_tokens": null, "answer_tokens": null, "total_tokens": null }, "total_cost_usd": 7.25, "reported_items": 50, "reported_samples": 1, "relative_tokens_to_source_anchor": null, "relative_cost_to_source_anchor": null, "notes": "Dollar-only evidence. Paper reports setup-level average cost per task: $0.086 for the GPT-4.1 agent plus $0.059 for the GPT-4.1 user simulator. Airline has 50 tasks, so 50 * $0.145 = $7.25 for one trial. This is not separately measured airline-only token usage." } }, { "id": "tau2_bench_avg", "name": "τ²-Bench (avg of retail/airline/telecom)", "category": "Tool Use", "metric": "macro task success (%)", "num_problems": 279, "source_url": "https://arxiv.org/abs/2506.07982", "canonical_setting": { "version": "Tau2-bench macro average of airline, retail and telecom", "metric_type": "pct", "higher_is_better": true, "range": [ 0, 100 ], "multimodal_input": false, "tools": "agentic", "notes": "Unweighted macro average over the three domain success rates; 50 + 115 + 114 underlying tasks." }, "cost": { "source_id": "artificial_analysis_eval_token_cost_gemini_2_5_pro_anchor", "source_name": "Artificial Analysis per-evaluation token usage and cost with Gemini 2.5 Pro anchor", "source_url": "https://artificialanalysis.ai/evaluations/tau2-bench", "source_benchmark_name": "Tau2-Bench", "source_model_name": "Gemini 2.5 Pro", "source_model_slug": "gemini-2-5-pro", "evidence_scope": "observed model-inference totals for the source evaluation run", "tokens": { "prompt_tokens": null, "completion_tokens": null, "input_tokens": 21458049, "output_tokens": 871618, "reasoning_tokens": 748345, "answer_tokens": 123273, "total_tokens": 22329667 }, "total_cost_usd": 35.53874125, "reported_items": null, "reported_samples": null, "relative_tokens_to_source_anchor": 13.781, "relative_cost_to_source_anchor": 2.253, "notes": "Inference-side token/cost evidence only; source_model_* identifies the reported model row. Dollar cost is run/model specific, not benchmark-intrinsic. primary anchor model: Gemini 2.5 Pro" } }, { "id": "tau2_bench_retail", "name": "τ²-bench Retail", "category": "Agentic", "metric": "% success", "num_problems": 115, "source_url": "https://arxiv.org/abs/2506.07982", "canonical_setting": { "version": "τ²-bench (Sierra AI 2025) — retail domain, paper-defined task set", "metric_type": "pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": false, "tools": "agentic", "sampling": "pass^1 / one trial per task", "judge": "state-based task success", "notes": "Paper Table 1 reports 115 Retail tasks. Current official repo base split has 114 after later task-fix releases; keep paper count for the tau2-bench 2025 row unless the row is redefined to current-release tau3 semantics." }, "cost": { "source_id": "tau2_bench_paper_gpt41_agent_user_sim_per_task_cost", "source_name": "tau2-bench paper reported GPT-4.1 agent and user-simulator cost", "source_url": "https://arxiv.org/abs/2506.07982", "source_data_url": "https://arxiv.org/e-print/2506.07982", "source_benchmark_name": "tau2-bench Retail", "source_model_name": "gpt-4.1-2025-04-14 agent + user simulator", "source_model_slug": "gpt-4.1-2025-04-14-agent-user-sim", "evidence_scope": "derived one-trial retail-domain dollar cost from paper-reported average per-task agent and user-simulator API costs", "tokens": { "prompt_tokens": null, "completion_tokens": null, "input_tokens": null, "output_tokens": null, "reasoning_tokens": null, "answer_tokens": null, "total_tokens": null }, "total_cost_usd": 16.675, "reported_items": 115, "reported_samples": 1, "relative_tokens_to_source_anchor": null, "relative_cost_to_source_anchor": null, "notes": "Dollar-only evidence. Paper reports setup-level average cost per task: $0.086 for the GPT-4.1 agent plus $0.059 for the GPT-4.1 user simulator. Retail has 115 tasks in the paper, so 115 * $0.145 = $16.675 for one trial. This is not separately measured retail-only token usage." } }, { "id": "tau2_bench_telecom", "name": "τ²-bench Telecom", "category": "Agentic", "metric": "% success", "num_problems": 114, "source_url": "https://arxiv.org/abs/2506.07982", "canonical_setting": { "version": "τ²-bench (Sierra AI 2025) — telecom domain, base split", "metric_type": "pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": false, "tools": "agentic", "sampling": "pass^1 / one trial per task", "judge": "state-based task success", "notes": "Paper Table 1 and current official split file give 114 Telecom base tasks; the full generated Telecom pool has 2285 tasks and is excluded." }, "cost": { "source_id": "tau2_bench_paper_gpt41_agent_user_sim_per_task_cost", "source_name": "tau2-bench paper reported GPT-4.1 agent and user-simulator cost", "source_url": "https://arxiv.org/abs/2506.07982", "source_benchmark_name": "tau2-bench", "source_model_name": "gpt-4.1-2025-04-14 agent + user simulator", "source_model_slug": "gpt-4.1-2025-04-14-agent-user-sim", "evidence_scope": "reported average per-task API cost for the GPT-4.1 agent and GPT-4.1 user simulator", "tokens": { "prompt_tokens": null, "completion_tokens": null, "input_tokens": null, "output_tokens": null, "reasoning_tokens": null, "answer_tokens": null, "total_tokens": null }, "total_cost_usd": 0.145, "reported_items": 1, "reported_samples": 1, "relative_tokens_to_source_anchor": null, "relative_cost_to_source_anchor": null, "notes": "Dollar-only per-task evidence: paper reports $0.086 agent plus $0.059 user-simulator average cost per task, and approximately $40 for all domains at one trial per task. This is tau2-bench setup-level evidence rather than a telecom-only measured total." } }, { "id": "tau3_bench", "name": "τ³-Bench", "category": "Tool Use", "metric": "%", "num_problems": 1500, "source_url": "https://z.ai/blog/glm-5.1", "canonical_setting": { "version": "τ³-Bench all-domain text setting reported by Z.ai GLM-5.1 blog", "metric_type": "pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": false, "tools": "agentic", "sampling": "4 trials; count folded into num_problems", "judge": "state-based task success", "notes": "Z.ai footnote says tau3-bench uses all domains with an extra user-simulator prompt, banking terminal_use retrieval, GPT-5.2-low user simulator, and 4 trials. Count = (airline 50 + retail 114 + telecom 114 + banking_knowledge 97) * 4 = 1500 scored simulations." } }, { "id": "tau_bench_airline", "name": "tau-bench Airline", "category": "Agentic", "metric": "% success", "num_problems": 50, "source_url": "https://arxiv.org/abs/2406.12045", "canonical_setting": { "version": "tau-bench airline domain", "metric_type": "pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": false, "tools": "agentic", "sampling": "pass^1 / one trial per task", "judge": "state-based task success", "notes": "Original tau-bench Airline has 50 user-agent interaction tasks. Count task conversations; multi-turn/tool burden is represented by agentic cost factors rather than multiplying by every action step." } }, { "id": "tau_bench_retail", "name": "Tau-Bench Retail", "category": "Agentic", "metric": "% success", "num_problems": 115, "source_url": "https://arxiv.org/abs/2406.12045", "canonical_setting": { "version": "tau-bench retail domain", "metric_type": "pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": false, "tools": "agentic", "sampling": "pass^1 / one trial per task", "judge": "state-based task success", "notes": "Original tau-bench Retail has 115 user-agent interaction tasks. Count task conversations; multi-turn/tool burden is represented by agentic cost factors rather than multiplying by every action step." } }, { "id": "tau_bench_telecom", "name": "Tau-Bench Telecom", "category": "Agentic", "metric": "% success", "num_problems": null, "source_url": "https://arxiv.org/abs/2406.12045", "canonical_setting": { "version": "tau-bench telecom", "metric_type": "pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": false, "notes": "tools=agentic (scaffold-defined). Preferred = official harness tools (bash/editor for SWE-bench, browser for OSWorld/BrowseComp, official APIs for tau-bench/MCPAtlas/Toolathlon, terminal for Terminal-Bench). Non-official scaffolds → matches_canonical=false.", "tools": "agentic" }, "cost": { "source_id": "sierra_tau2_telecom_gpt41_default_cost_usage", "source_name": "Sierra tau2-bench final telecom result logs", "source_url": "https://github.com/sierra-research/tau2-bench", "source_data_url": "https://raw.githubusercontent.com/sierra-research/tau2-bench/main/data/tau2/results/final/gpt-4.1-2025-04-14_telecom_default_gpt-4.1-2025-04-14_4trials.json", "source_benchmark_name": "tau2-bench Telecom default, 4 trials", "source_model_name": "gpt-4.1-2025-04-14 agent + user simulator", "source_model_slug": "gpt-4.1-2025-04-14-agent-user-sim", "evidence_scope": "official final result log with agent plus user-simulator cost and usage totals", "tokens": { "prompt_tokens": 98778208, "completion_tokens": 869183, "input_tokens": null, "output_tokens": null, "reasoning_tokens": null, "answer_tokens": null, "total_tokens": 99647391 }, "total_cost_usd": 89.555448, "reported_items": 114, "reported_samples": 4, "relative_tokens_to_source_anchor": null, "relative_cost_to_source_anchor": null, "notes": "Official final tau2 result log for telecom default 4-trial run reports agent_cost=$50.140428 and user_cost=$39.415020, total $89.555448 over 114 tasks x 4 trials. Usage totals: prompt_tokens=98,778,208 and completion_tokens=869,183." } }, { "id": "tempcompass", "name": "TempCompass", "category": "Video", "metric": "%", "num_problems": null, "source_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "canonical_setting": { "version": "TempCompass", "metric_type": "pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": true, "tools": "none", "notes": "Per Doubao Seed 2.0 Pro model card." } }, { "id": "terminal_bench", "name": "Terminal-Bench 2.0", "category": "Agentic", "metric": "% tasks solved", "num_problems": 445, "source_url": "https://arxiv.org/html/2601.11868v1", "canonical_setting": { "version": "Terminal-Bench 2.0 official 89-task release", "metric_type": "pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": false, "tools": "terminal agent scaffold", "sampling": "at least 5 trials per logical task", "logical_tasks": 89, "required_trials_per_task": 5, "judge": "programmatic end-to-end task tests", "harness": "source-reported terminal scaffold", "notes": "The immutable arXiv v1 paper defines 89 logical Terminal-Bench 2.0 tasks and evaluates every supported model-agent combination at least five times. num_problems therefore records 445 required generations. The paper's Table 2 token-count note refers to a 74-task execution subset and does not redefine the benchmark total. The StepFun mixed row maps Kimi K2.6, GPT-5.5, and Claude Opus 4.7 to this identity per the locked footnote." }, "cost": { "source_id": "terminal_bench_2_table2_gemini3pro_terminus2_tokens", "source_name": "Terminal-Bench 2.0 paper Table 2 token counts", "source_url": "https://arxiv.org/html/2601.11868v1", "source_benchmark_name": "Terminal-Bench 2.0", "source_model_name": "Gemini 3 Pro + Terminus 2", "source_model_slug": "gemini-3-pro; agent=terminus-2", "evidence_scope": "reported token counts for running all 74 Terminal-Bench 2.0 tasks for one agent-model row", "tokens": { "prompt_tokens": null, "completion_tokens": null, "input_tokens": 5100000, "output_tokens": 2200000, "reasoning_tokens": null, "answer_tokens": null, "total_tokens": 7300000 }, "total_cost_usd": null, "reported_items": 74, "reported_samples": null, "relative_tokens_to_source_anchor": null, "relative_cost_to_source_anchor": null, "notes": "Terminal-Bench 2.0 Table 2 reports input/output token counts for all 74 tasks. Values are rounded as printed in the paper table (5.1M input, 2.2M output). Model/agent row specific; not benchmark-intrinsic." } }, { "id": "terminal_bench_1", "name": "Terminal-Bench 1.0", "category": "Agentic", "metric": "% solved", "num_problems": null, "source_url": "https://terminal-bench.com/", "canonical_setting": { "version": "Terminal-Bench 1.0", "metric_type": "pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": false, "notes": "tools=agentic (scaffold-defined). Preferred = official harness tools (bash/editor for SWE-bench, browser for OSWorld/BrowseComp, official APIs for tau-bench/MCPAtlas/Toolathlon, terminal for Terminal-Bench). Non-official scaffolds → matches_canonical=false.", "tools": "agentic" } }, { "id": "terminal_bench_hard", "name": "Terminal-Bench Hard", "category": "Coding", "metric": "%", "num_problems": null, "source_url": "https://z.ai/blog/glm-4.7", "canonical_setting": { "version": "Terminal-Bench Hard", "metric_type": "pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": false, "tools": "agentic", "notes": "Per GLM-4.7 blog." }, "cost": { "source_id": "artificial_analysis_eval_token_cost_gemini_2_5_pro_anchor", "source_name": "Artificial Analysis per-evaluation token usage and cost with Gemini 2.5 Pro anchor", "source_url": "https://artificialanalysis.ai/evaluations/terminalbench-hard", "source_benchmark_name": "Terminal-Bench Hard", "source_model_name": "Gemini 2.5 Pro", "source_model_slug": "gemini-2-5-pro", "evidence_scope": "observed model-inference totals for the source evaluation run", "tokens": { "prompt_tokens": null, "completion_tokens": null, "input_tokens": 12836710, "output_tokens": 1925752, "reasoning_tokens": 1502276, "answer_tokens": 423476, "total_tokens": 14762462 }, "total_cost_usd": 35.3034075, "reported_items": null, "reported_samples": null, "relative_tokens_to_source_anchor": 9.11, "relative_cost_to_source_anchor": 2.238, "notes": "Inference-side token/cost evidence only; source_model_* identifies the reported model row. Dollar cost is run/model specific, not benchmark-intrinsic. primary anchor model: Gemini 2.5 Pro" } }, { "id": "tob_complex_workflows", "name": "ToB-ComplexWorkflows", "category": "Real-world", "metric": "%", "num_problems": null, "source_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "canonical_setting": { "version": "ToB-ComplexWorkflows", "metric_type": "pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": false, "tools": "none", "notes": "Per Doubao Seed 2.0 Pro model card." } }, { "id": "tob_compositional_tasks", "name": "ToB-CompositionalTasks", "category": "Real-world", "metric": "%", "num_problems": null, "source_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "canonical_setting": { "version": "ToB-CompositionalTasks", "metric_type": "pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": false, "tools": "none", "notes": "Per Doubao Seed 2.0 Pro model card." } }, { "id": "tob_information_extraction", "name": "ToB-InformationExtraction", "category": "Real-world", "metric": "%", "num_problems": null, "source_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "canonical_setting": { "version": "ToB-InformationExtraction", "metric_type": "pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": false, "tools": "none", "notes": "Per Doubao Seed 2.0 Pro model card." } }, { "id": "tob_k12_education", "name": "ToB-K12Education", "category": "Real-world", "metric": "%", "num_problems": null, "source_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "canonical_setting": { "version": "ToB-K12Education", "metric_type": "pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": false, "tools": "none", "notes": "Per Doubao Seed 2.0 Pro model card." } }, { "id": "tob_referenceqa", "name": "ToB-ReferenceQ&A", "category": "Real-world", "metric": "%", "num_problems": null, "source_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "canonical_setting": { "version": "ToB-ReferenceQ&A", "metric_type": "pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": false, "tools": "none", "notes": "Per Doubao Seed 2.0 Pro model card." } }, { "id": "tob_text_classification", "name": "ToB-TextClassification", "category": "Real-world", "metric": "%", "num_problems": null, "source_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "canonical_setting": { "version": "ToB-TextClassification", "metric_type": "pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": false, "tools": "none", "notes": "Per Doubao Seed 2.0 Pro model card." } }, { "id": "tomato", "name": "TOMATO", "category": "Video", "metric": "multiple-choice temporal-reasoning accuracy (%)", "num_problems": 1484, "source_url": "https://arxiv.org/abs/2410.23266", "canonical_setting": { "version": "TOMATO", "metric_type": "pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": true, "tools": "none", "sampling": "pass@1", "judge": "exact multiple-choice grader", "harness": "official", "notes": "1,484 questions over 1,417 videos." } }, { "id": "toolathlon", "name": "Toolathlon (Original)", "category": "Agentic", "metric": "% correct (pass@1)", "num_problems": 108, "source_url": "https://toolathlon.xyz/", "canonical_setting": { "version": "Original Toolathlon 108-task release before 2026-06-30", "metric_type": "pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": false, "tools": "agentic tool use", "judge": "dedicated deterministic state evaluators", "notes": "Original 108-task benchmark across 32 applications and 604 tools. Distinct from Toolathlon-Verified introduced on 2026-06-30 with revised tasks/evaluators." }, "cost": { "source_id": "toolathlon_trajectories_gemini_3_pro_preview_run1", "source_name": "Toolathlon trajectories Gemini 3 Pro Preview run 1", "source_url": "https://huggingface.co/datasets/hkust-nlp/Toolathlon-Trajectories", "source_data_url": "https://huggingface.co/datasets/hkust-nlp/Toolathlon-Trajectories/resolve/main/gemini-3-pro-preview_1.jsonl", "source_benchmark_name": "Toolathlon", "source_model_name": "Gemini 3 Pro Preview", "source_model_slug": "gemini-3-pro-preview", "evidence_scope": "official public trajectory JSONL; summed agent_cost over all 108 tasks for one model run", "tokens": { "prompt_tokens": null, "completion_tokens": null, "input_tokens": 68397190, "output_tokens": 978812, "reasoning_tokens": null, "answer_tokens": null, "total_tokens": 69376002 }, "total_cost_usd": 148.5402, "reported_items": 108, "reported_samples": 1, "relative_tokens_to_source_anchor": null, "relative_cost_to_source_anchor": null, "notes": "Run/model-specific agent benchmark cost. JSONL rows expose per-task total_input_tokens, total_output_tokens, total_requests, and total_cost." } }, { "id": "treebench", "name": "TreeBench", "category": "Vision Spatial", "metric": "%", "num_problems": null, "source_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "canonical_setting": { "version": "TreeBench", "metric_type": "pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": true, "tools": "none", "notes": "Per Doubao Seed 2.0 Pro model card." } }, { "id": "truthfulqa", "name": "TruthfulQA", "category": "Factuality", "source_url": "https://github.com/sylinrl/TruthfulQA", "num_problems": 817, "canonical_setting": { "version": "TruthfulQA generation benchmark", "metric_type": "pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": false, "tools": "none", "notes": "TruthfulQA contains 817 questions designed to test imitative falsehoods. Count one text generation per question." }, "cost": { "source_id": "helm_classic_v0_4_0", "source_name": "HELM Classic v0.4.0", "source_url": "https://storage.googleapis.com/crfm-helm-public/classic/benchmark_output/releases/v0.4.0/groups.json", "source_benchmark_name": "TruthfulQA", "evidence_scope": "reported HELM prompt/completion token totals for the source benchmark scenario", "tokens": { "prompt_tokens": 1100944.4691282208, "completion_tokens": 2307.323482433849, "input_tokens": null, "output_tokens": null, "reasoning_tokens": null, "answer_tokens": null, "total_tokens": 1103251.7926106546 }, "total_cost_usd": null, "reported_items": 382.9490861618799, "reported_samples": null, "relative_tokens_to_source_anchor": 0.158, "relative_cost_to_source_anchor": null, "notes": "HELM token evidence only; no dollar cost or hidden reasoning tokens are reported." } }, { "id": "tvbench", "name": "TVBench", "category": "Video", "metric": "%", "num_problems": null, "source_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "canonical_setting": { "version": "TVBench", "metric_type": "pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": true, "tools": "none", "notes": "Per Doubao Seed 2.0 Pro model card." } }, { "id": "usamo_2025", "name": "USAMO 2025", "category": "Math", "metric": "% of 42 points", "num_problems": 6, "source_url": "https://huggingface.co/datasets/MathArena/usamo_2025", "canonical_setting": { "version": "USAMO 2025", "metric_type": "pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": false, "notes": "tools=none preferred (pure-reasoning eval). If only with-tool scores (python/web/RAG) are available, accept and mark cell matches_canonical=false.", "tools": "none" }, "cost": { "source_id": "matharena_usamo_2025_outputs_all_models_cost", "source_name": "MathArena USAMO 2025 output logs", "source_url": "https://matharena.ai/competition_tables/usamo--usamo_2025", "source_data_url": "https://datasets-server.huggingface.co/rows?dataset=MathArena/usamo_2025_outputs&config=default&split=train", "source_benchmark_name": "USAMO 2025", "source_model_name": "All 10 MathArena-reported models", "source_model_slug": "claude-3.7-sonnet-think; deepseek-r1; deepseek-r1-0528; gemini-2.5-pro; grok-3-think; qwq-32b; gemini-2.0-flash-thinking; o1-pro-high; o3-mini-high; o4-mini-high", "evidence_scope": "official output-log rows with per-row input tokens, output tokens, and cost for all 6 problems x 4 samples x 10 models", "tokens": { "prompt_tokens": null, "completion_tokens": null, "input_tokens": 50672, "output_tokens": 3432986, "reasoning_tokens": null, "answer_tokens": null, "total_tokens": 3483658 }, "total_cost_usd": 248.69105409, "reported_items": 6, "reported_samples": 4, "relative_tokens_to_source_anchor": null, "relative_cost_to_source_anchor": null, "notes": "Exact sum over 240 public MathArena per-output rows with input_tokens, output_tokens, and cost fields: 6 problems times 4 answers times 10 models. Preserves reported zero/N/A token or cost rows, e.g. Grok 3 Think and gemini-2.0-flash-thinking cost." } }, { "id": "usamo_2026", "name": "USAMO 2026", "category": "Math", "metric": "% of 42 points", "num_problems": 6, "source_url": "https://matharena.ai/usamo/", "canonical_setting": { "version": "USAMO 2026", "metric_type": "pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": false, "notes": "tools=none preferred (pure-reasoning eval). If only with-tool scores (python/web/RAG) are available, accept and mark cell matches_canonical=false.", "tools": "none" }, "cost": { "source_id": "matharena_usamo_2026_outputs_gpt54_xhigh_cost", "source_name": "MathArena USAMO 2026 output logs for GPT-5.4 (xhigh)", "source_url": "https://matharena.ai/usamo/", "source_data_url": "https://huggingface.co/datasets/MathArena/usamo_2026_outputs/resolve/main/data/train-00000-of-00001.parquet", "source_benchmark_name": "USAMO 2026", "source_model_name": "GPT-5.4 (xhigh)", "source_model_slug": "openai/gpt-54", "evidence_scope": "official output-log rows with per-row input tokens, output tokens, and cost for all 6 problems x 4 samples", "tokens": { "prompt_tokens": null, "completion_tokens": null, "input_tokens": 49614, "output_tokens": 1365072, "reasoning_tokens": null, "answer_tokens": null, "total_tokens": 1414686 }, "total_cost_usd": 20.600115, "reported_items": 6, "reported_samples": 4, "relative_tokens_to_source_anchor": null, "relative_cost_to_source_anchor": null, "notes": "Rows where model_config=openai/gpt-54 cover idx_answer 0,1,2,3 for all 6 USAMO 2026 problems. MathArena states each model is run 4 times on each problem and displayed table cost is average cost of one model run on one problem. This covers model answer generation only, not LLM-jury grading or human verification." } }, { "id": "vending_bench_2", "name": "Vending-Bench 2", "category": "Agentic", "source_url": "https://andonlabs.com/evals/vending-bench-2", "num_problems": 15000, "canonical_setting": { "version": "Vending-Bench 2 (long-horizon planning)", "metric_type": "dollars", "range": [ 0, 100000 ], "higher_is_better": true, "multimodal_input": false, "tools": "agentic browser/email/order-management tools", "sampling": "5 runs; lower-bound messages folded into num_problems", "judge": "year-end bank account balance", "notes": "Official Vending-Bench 2 reports leaderboard scores as the average across 5 full-year simulation runs. The page states that running a model for a full year results in 3,000-6,000 messages total, so this cost count uses the source-backed lower bound of actual model messages: 5 runs times 3,000 messages = 15,000. The true per-model count can be up to 30,000 messages; either way the benchmark is Tier 3." }, "cost": { "source_id": "vending_bench_2_official_cost_chart_gemini_3_pro", "source_name": "Vending-Bench 2 official cost-per-run chart data", "source_url": "https://andonlabs.com/evals/vending-bench-2", "source_data_url": "https://andonlabs.com/_app/immutable/nodes/18.ps2zASBK.js", "source_benchmark_name": "Vending-Bench 2", "source_model_name": "Gemini 3 Pro", "source_model_slug": "gemini-3-pro", "evidence_scope": "official mean per-run input/output token usage and cost for one model row over full-year Vending-Bench 2 runs", "tokens": { "prompt_tokens": null, "completion_tokens": null, "input_tokens": 70579113, "output_tokens": 474840, "reasoning_tokens": null, "answer_tokens": null, "total_tokens": 71053953 }, "total_cost_usd": 146.86, "reported_items": 1, "reported_samples": 5, "relative_tokens_to_source_anchor": null, "relative_cost_to_source_anchor": null, "notes": "Official Vending-Bench 2 page reports score versus mean cost per run and states costs are calculated from provider input/output token pricing without caching. This row stores the embedded chart data for Gemini 3 Pro: n=5 runs, mean_cost=$146.86, avg_input_tokens=70,579,113, avg_output_tokens=474,840. Values are one full-year simulation run averages, model/run specific." } }, { "id": "vibe_eval", "name": "Vibe-Eval", "category": "Multimodal", "source_url": "https://github.com/reka-ai/reka-vibe-eval", "num_problems": 269, "canonical_setting": { "version": "Vibe-Eval v1 overall", "metric_type": "pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": true, "judge": "Reka Core evaluator scores each response on a 1-5 scale", "sampling": "single response per prompt", "notes": "Official paper and HF dataset report 269 visual-understanding prompts, including 100 hard prompts. Count model generations as one response per example_id; evaluator calls are scoring overhead." }, "cost": { "source_id": "vibe_eval_paper_reka_core_free_evaluator_api", "source_name": "Vibe-Eval paper/blog Reka Core evaluator free API access", "source_url": "https://arxiv.org/abs/2405.02287", "source_data_url": "https://github.com/reka-ai/reka-vibe-eval/blob/main/evaluate.py", "source_benchmark_name": "Vibe-Eval", "source_model_name": "Reka Core evaluator", "source_model_slug": "reka-core-20240501", "evidence_scope": "source-backed free evaluator API access for lightweight Vibe-Eval evaluation", "tokens": { "prompt_tokens": null, "completion_tokens": null, "input_tokens": null, "output_tokens": null, "reasoning_tokens": null, "answer_tokens": null, "total_tokens": null }, "total_cost_usd": 0.0, "reported_items": 269, "reported_samples": 1, "relative_tokens_to_source_anchor": null, "relative_cost_to_source_anchor": null, "notes": "Vibe-Eval paper/blog state that Reka Core is the automatic evaluator and API access is free/free-of-charge for lightweight Vibe-Eval evaluation. Paper results run the evaluator three times and average scores, while official evaluate.py makes one Reka Core API call per supplied generation row. No prompt/completion token counts are published; model generation cost is excluded." } }, { "id": "video_mme", "name": "Video-MME", "category": "Multimodal", "source_url": "https://raw.githubusercontent.com/MME-Benchmarks/Video-MME/06c2315b892f88578f81d73205d07cf576f292b9/README.md", "num_problems": 2700, "canonical_setting": { "version": "Video-MME overall, 900 videos / 2,700 QA pairs", "metric_type": "pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": true, "metric": "multiple-choice QA accuracy", "sampling": "one response per question", "notes": "Official immutable repository defines 900 videos and 2,700 human-annotated question-answer pairs.", "tools": "none", "judge": "answer-key multiple-choice accuracy" }, "metric": "% multiple-choice accuracy", "higher_is_better": true }, { "id": "video_mmmu", "name": "Video-MMMU", "category": "Video/Multimodal", "metric": "%", "num_problems": 900, "source_url": "https://videommmu.github.io/", "canonical_setting": { "version": "Video-MMMU overall", "metric_type": "pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": true, "tools": "none", "metric": "accuracy over human-annotated video QA questions", "sampling": "single response per question", "notes": "Official paper/project report 300 expert-level videos and 900 human-annotated questions across Perception, Comprehension, and Adaptation. Count one model generation per question for the overall percent score; do not count the model-report source page as the benchmark definition." } }, { "id": "videoeval_pro", "name": "VideoEval-Pro", "category": "Video", "metric": "%", "num_problems": null, "source_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "canonical_setting": { "version": "VideoEval-Pro", "metric_type": "pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": true, "tools": "none", "notes": "Per Doubao Seed 2.0 Pro model card." } }, { "id": "videoholmes", "name": "VideoHolmes", "category": "Video", "metric": "% multiple-choice accuracy", "num_problems": 1837, "source_url": "https://raw.githubusercontent.com/TencentARC/Video-Holmes/52ef8da286ccad03036a65e7b67c160bc3a24fb9/README.md", "canonical_setting": { "version": "Video-Holmes full evaluation set", "metric_type": "pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": true, "tools": "none", "notes": "Official immutable repository defines 1,837 questions from 270 suspense short films across seven tasks.", "sampling": "one response per question", "judge": "answer-key multiple-choice accuracy" }, "higher_is_better": true }, { "id": "videoreasonbench", "name": "VideoReasonBench", "category": "Video", "metric": "%", "num_problems": null, "source_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "canonical_setting": { "version": "VideoReasonBench", "metric_type": "pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": true, "tools": "none", "notes": "Per Doubao Seed 2.0 Pro model card." }, "cost": { "source_id": "videoreasonbench_paper_gemini25flash_no_thinking_response_tokens", "source_name": "VideoReasonBench paper token-count analysis", "source_url": "https://arxiv.org/abs/2505.23359", "source_data_url": "https://huggingface.co/datasets/lyx97/reasoning_videos", "source_benchmark_name": "VideoReasonBench", "source_model_name": "Gemini-2.5-Flash with thinking_budget=0", "source_model_slug": "gemini-2.5-flash; thinking_budget=0", "evidence_scope": "paper-reported average response/output tokens; derived output-token total over 1,440 questions", "tokens": { "prompt_tokens": null, "completion_tokens": null, "input_tokens": null, "output_tokens": 2678544, "reasoning_tokens": null, "answer_tokens": null, "total_tokens": 2678544 }, "total_cost_usd": null, "reported_items": 1440, "reported_samples": 1, "relative_tokens_to_source_anchor": null, "relative_cost_to_source_anchor": null, "notes": "VideoReasonBench paper reports 1,440 questions and 1,860.1 average response tokens for Gemini-2.5-Flash with no explicit thinking resources. Derived output-token total is 1,860.1*1,440=2,678,544. Output-only evidence; no input/video tokens or dollar cost are reported." } }, { "id": "videosimpleqa", "name": "VideoSimpleQA", "category": "Video", "metric": "source-reported score (%)", "num_problems": 1504, "source_url": "https://huggingface.co/datasets/VideoSimpleQA/VideoSimpleQA", "canonical_setting": { "version": "VideoSimpleQA", "metric_type": "pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": true, "tools": "none", "sampling": "pass@1", "judge": "official configurable LLM grader", "harness": "official", "notes": "1,504 QA pairs over 1,079 videos. The Seed source does not identify whether the displayed score is the official accuracy or F1 view." } }, { "id": "visfactor", "name": "VisFactor", "category": "Vision Perception", "metric": "%", "num_problems": null, "source_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "canonical_setting": { "version": "VisFactor", "metric_type": "pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": true, "tools": "none", "notes": "Per Doubao Seed 2.0 Pro model card." } }, { "id": "vispeak", "name": "ViSpeak", "category": "Video", "metric": "%", "num_problems": null, "source_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "canonical_setting": { "version": "ViSpeak", "metric_type": "pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": true, "tools": "none", "notes": "Per Doubao Seed 2.0 Pro model card." } }, { "id": "visulogic", "name": "VisuLogic", "category": "Vision Puzzles", "metric": "%", "num_problems": null, "source_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "canonical_setting": { "version": "VisuLogic", "metric_type": "pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": true, "tools": "none", "notes": "Per Doubao Seed 2.0 Pro model card." } }, { "id": "vitabench", "name": "VitaBench", "category": "Tool Use", "metric": "%", "num_problems": null, "source_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "canonical_setting": { "version": "VitaBench", "metric_type": "pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": false, "tools": "tool calls", "notes": "Per Doubao Seed 2.0 Pro model card." } }, { "id": "viverbench", "name": "ViVerBench", "category": "Vision VQA", "metric": "%", "num_problems": null, "source_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "canonical_setting": { "version": "ViVerBench", "metric_type": "pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": true, "tools": "none", "notes": "Per Doubao Seed 2.0 Pro model card." } }, { "id": "vlms_are_biased", "name": "VLMsAreBiased", "category": "Vision Perception", "metric": "%", "num_problems": null, "source_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "canonical_setting": { "version": "VLMsAreBiased", "metric_type": "pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": true, "tools": "none", "notes": "Per Doubao Seed 2.0 Pro model card." } }, { "id": "vlms_are_blind", "name": "VLMsAreBlind", "category": "Vision Perception", "metric": "%", "num_problems": null, "source_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "canonical_setting": { "version": "VLMsAreBlind", "metric_type": "pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": true, "tools": "none", "notes": "Per Doubao Seed 2.0 Pro model card." } }, { "id": "vpct", "name": "VPCT", "category": "Vision Puzzles", "metric": "%", "num_problems": null, "source_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "canonical_setting": { "version": "VPCT", "metric_type": "pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": true, "tools": "none", "notes": "Per Doubao Seed 2.0 Pro model card." } }, { "id": "widesearch", "name": "WideSearch (item-F1)", "category": "Search Agent", "metric": "%", "num_problems": 200, "source_url": "https://huggingface.co/datasets/ByteDance-Seed/WideSearch", "canonical_setting": { "version": "WideSearch overall (item-F1)", "metric_type": "pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": false, "tools": "search/browser access required", "judge": "evaluation pipeline mixes exact/URL/numeric matching with LLM-assisted cell judgment", "metric": "item-F1 over required table fields", "sampling": "single final response per task", "notes": "Official paper and dataset report 200 broad information-seeking tasks, split 100 English and 100 Chinese. Count one model generation per task; item-F1 expands the scoring units, not the number of model generations." } }, { "id": "wildbench", "name": "WildBench", "category": "Chat", "metric": "Raw Score", "num_problems": null, "source_url": "https://github.com/InternLM/InternLM", "canonical_setting": { "version": "WildBench (Raw Score)", "metric_type": "raw", "range": [ null, null ], "higher_is_better": true, "multimodal_input": false, "tools": "none", "notes": "Per InternLM3 GitHub README. WildBench raw score." }, "cost": { "source_id": "helm_capabilities_v1_15_0", "source_name": "HELM Capabilities v1.15.0", "source_url": "https://raw.githubusercontent.com/stanford-crfm/helm/main/docs/benchmark.md", "source_benchmark_name": "WildBench", "evidence_scope": "observed model-inference totals for the source evaluation run", "tokens": { "prompt_tokens": 0.0, "completion_tokens": 205215.381, "total_tokens": 205215.381 }, "total_cost_usd": null, "relative_tokens_to_source_anchor": 2.317, "relative_cost_to_source_anchor": null, "notes": "Inference-side token/cost evidence only; judge, human-review, tool/environment, and infrastructure costs are separate protocol factors." } }, { "id": "world_travel_text", "name": "WorldTravel (TEXT)", "category": "Real-world", "metric": "%", "num_problems": null, "source_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "canonical_setting": { "version": "WorldTravel (TEXT)", "metric_type": "pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": false, "tools": "none", "notes": "Per Doubao Seed 2.0 Pro model card." } }, { "id": "world_travel_vlm", "name": "WorldTravel (VLM)", "category": "Real-world", "metric": "%", "num_problems": null, "source_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "canonical_setting": { "version": "WorldTravel (VLM)", "metric_type": "pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": true, "tools": "none", "notes": "Per Doubao Seed 2.0 Pro model card." } }, { "id": "worldvqa", "name": "WorldVQA", "category": "Multimodal Knowledge", "metric": "accuracy (%)", "num_problems": 3000, "source_url": "https://huggingface.co/datasets/moonshotai/WorldVQA/tree/29e1d54b27ffb34cdffb4cdc95d29afcf101f1f7", "canonical_setting": { "version": "WorldVQA", "metric_type": "pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": true, "tools": "none", "sampling": "pass@1", "judge": "official default gpt-oss-120b judge", "harness": "official", "notes": "Released first-eight-category leaderboard excluding People." } }, { "id": "xbench_deepsearch", "name": "xbench-DeepSearch", "category": "Search Agent", "metric": "%", "num_problems": 100, "source_url": "https://huggingface.co/datasets/xbench/DeepSearch", "canonical_setting": { "version": "xbench-DeepSearch original release", "metric_type": "pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": false, "tools": "search/retrieval environment required", "metric": "accuracy", "sampling": "single answer per task", "notes": "Official xbench DeepSearch HF dataset reports 100 encrypted rows/tasks and describes a search/information-retrieval evaluation. Count one model generation per task; do not use the MiniMax M2 model card as the benchmark definition. Later DeepSearch-2510 is a separate variant and also has 100 rows." } }, { "id": "xlrs_macro", "name": "XLRS-Bench (macro)", "category": "Vision STEM", "metric": "%", "num_problems": null, "source_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "canonical_setting": { "version": "XLRS-Bench (macro)", "metric_type": "pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": true, "tools": "none", "notes": "Per Doubao Seed 2.0 Pro model card." } }, { "id": "xpert_bench", "name": "XPertBench", "category": "Economic", "metric": "%", "num_problems": null, "source_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "canonical_setting": { "version": "XPertBench", "metric_type": "pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": false, "tools": "none", "notes": "Per Doubao Seed 2.0 Pro model card." } }, { "id": "zerobench_main", "name": "ZeroBench (main)", "category": "Vision Puzzles", "metric": "%", "num_problems": null, "source_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "canonical_setting": { "version": "ZeroBench (main)", "metric_type": "pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": true, "tools": "none", "notes": "Per Doubao Seed 2.0 Pro model card." }, "cost": { "source_id": "zerobench_v2_table3_o1_main_completion_tokens_cost", "source_name": "ZeroBench paper Table 3 completion tokens and cost", "source_url": "https://arxiv.org/abs/2502.09696", "source_data_url": "https://arxiv.org/html/2502.09696v2", "source_benchmark_name": "ZeroBench main questions", "source_model_name": "o1", "source_model_slug": "o1-2024-12-17", "evidence_scope": "paper-reported average completion tokens and completion-token cost per main question; derived total over 100 questions", "tokens": { "prompt_tokens": null, "completion_tokens": 749600, "input_tokens": null, "output_tokens": null, "reasoning_tokens": null, "answer_tokens": null, "total_tokens": 749600 }, "total_cost_usd": 47.2, "reported_items": 100, "reported_samples": 1, "relative_tokens_to_source_anchor": null, "relative_cost_to_source_anchor": null, "notes": "ZeroBench paper Table 3 reports o1 main-question average completion tokens=7,496 and cost=$0.472 per question. Derived totals: 7,496*100=749,600 completion tokens and $0.472*100=$47.20. Input/image prompt costs are not included." } }, { "id": "zerobench_sub", "name": "ZeroBench (sub)", "category": "Vision Puzzles", "metric": "%", "num_problems": null, "source_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "canonical_setting": { "version": "ZeroBench (sub)", "metric_type": "pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": true, "tools": "none", "notes": "Per Doubao Seed 2.0 Pro model card." }, "cost": { "source_id": "zerobench_v2_table3_o1_sub_completion_tokens_cost", "source_name": "ZeroBench paper Table 3 completion tokens and cost", "source_url": "https://arxiv.org/abs/2502.09696", "source_data_url": "https://arxiv.org/html/2502.09696v2", "source_benchmark_name": "ZeroBench subquestions", "source_model_name": "o1", "source_model_slug": "o1-2024-12-17", "evidence_scope": "paper-reported average completion tokens and completion-token cost per subquestion; derived total over 334 subquestions", "tokens": { "prompt_tokens": null, "completion_tokens": 1299594, "input_tokens": null, "output_tokens": null, "reasoning_tokens": null, "answer_tokens": null, "total_tokens": 1299594 }, "total_cost_usd": 81.83, "reported_items": 334, "reported_samples": 1, "relative_tokens_to_source_anchor": null, "relative_cost_to_source_anchor": null, "notes": "ZeroBench paper Table 3 reports o1 subquestion average completion tokens=3,891 and cost=$0.245 per subquestion. Derived totals: 3,891*334=1,299,594 completion tokens and $0.245*334=$81.83. Input/image prompt costs are not included." } }, { "id": "zerobench_tools", "name": "ZeroBench main (with tools)", "category": "Multimodal Reasoning", "metric": "accuracy (%)", "num_problems": 100, "source_url": "https://arxiv.org/abs/2502.09696", "canonical_setting": { "version": "ZeroBench main (with tools)", "metric_type": "pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": true, "tools": "search/code/web tools", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official", "notes": "Main 100-question set with tool access." } }, { "id": "kimi_code_bench_v2", "name": "Kimi Code Bench v2", "category": "Coding", "metric": "score (%)", "num_problems": null, "source_url": "https://huggingface.co/moonshotai/Kimi-K2.7-Code", "canonical_setting": { "version": "Kimi Code Bench v2", "metric_type": "pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": false, "notes": "Moonshot in-house coding-agent benchmark spanning 10+ languages and production software-engineering tasks. The source does not report the task count." } }, { "id": "program_bench", "name": "ProgramBench", "category": "Coding", "metric": "macro-average behavioral tests passed (%)", "num_problems": 200, "source_url": "https://programbench.com/", "canonical_setting": { "version": "ProgramBench 200-task suite", "metric_type": "pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": false, "tools": "compiled executable and documentation; no source, decompilation, or internet", "sampling": "unknown", "judge": "248,000+ fuzz-generated behavioral tests", "harness": "official ProgramBench sandbox", "notes": "Canonical score is the macro-average behavioral-tests-passed rate. Full task resolution is a distinct future metric and is not mixed into this id." } }, { "id": "mls_bench_lite", "name": "MLS-Bench-Lite", "category": "Coding", "metric": "score (0-100)", "num_problems": 30, "source_url": "https://mls-bench.com/", "canonical_setting": { "version": "Official MLS-Bench-Lite 30-task subset", "metric_type": "pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": false, "notes": "Official 30-task subset of MLS-Bench. Agents receive five hours to develop and submit scalable ML methods." } }, { "id": "kimi_claw_24_7", "name": "Kimi Claw 24/7 Bench", "category": "Agentic", "metric": "% average pass rate", "num_problems": 17, "source_url": "https://huggingface.co/moonshotai/Kimi-K2.7-Code", "canonical_setting": { "version": "Kimi Claw 24/7 Bench", "metric_type": "pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": false, "notes": "17 persistent multi-day professional scenarios covering 610 evaluation points in the OpenClaw harness. Final score is the average pass rate over evaluation points and three runs." } }, { "id": "mcpmark_verified", "name": "MCPMark-Verified", "category": "Agentic", "metric": "% success", "num_problems": null, "source_url": "https://huggingface.co/moonshotai/Kimi-K2.7-Code", "canonical_setting": { "version": "MCPMark-Verified human-verified edition", "metric_type": "pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": false, "notes": "Human-verified edition across Notion, GitHub, Filesystem, Postgres, and Playwright. Official configuration uses a 100-step tool-call budget, 32K max tokens per step, and averages three runs. Task count was not reported." } }, { "id": "aa_briefcase_elo", "name": "AA-Briefcase (Elo)", "category": "Agentic", "metric": "Elo rating", "num_problems": 91, "source_url": "https://artificialanalysis.ai/evaluations/aa-briefcase", "canonical_setting": { "version": "AA-Briefcase live 91-task series", "metric_type": "elo_rating", "range": null, "higher_is_better": true, "multimodal_input": true, "notes": "Composite Artificial Analysis rating over rubric correctness, analytical quality, and presentation quality; live snapshot scores can drift." } }, { "id": "agents_last_exam", "name": "Agents' Last Exam", "category": "Agentic", "metric": "% tasks passed", "num_problems": null, "source_url": "https://agents-last-exam.org/docs/ale/index.html", "canonical_setting": { "version": "Living public benchmark; no fixed scored version", "metric_type": "pass_rate_pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": true, "notes": "Professional computer-use tasks with hidden-reference grading. The public corpus is growing, so the scored item count is not stable." } }, { "id": "automation_bench", "name": "AutomationBench", "category": "Agentic", "metric": "% tasks passed", "num_problems": 600, "source_url": "https://github.com/zapier/AutomationBench", "canonical_setting": { "version": "Public 600-task scored set", "metric_type": "strict_pass_rate_pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": false, "notes": "A task passes only when every final-state assertion passes. Distinct from AutomationBench-AA." } }, { "id": "corpfin_v2", "name": "CorpFin v2", "category": "Finance", "metric": "% accuracy", "num_problems": 858, "source_url": "https://www.vals.ai/benchmarks/corp_fin_v2", "canonical_setting": { "version": "CorpFin v2 held-out test set", "metric_type": "accuracy_pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": false, "notes": "858 questions from 43 credit agreements, evaluated under documented context variants." } }, { "id": "deep_swe_v1_1", "name": "DeepSWE v1.1", "category": "Agentic Coding", "metric": "% resolved (pass@1)", "num_problems": 113, "source_url": "https://github.com/datacurve-ai/deep-swe", "canonical_setting": { "version": "DeepSWE v1.1", "metric_type": "pass_at_1_pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": false, "tools": "agentic repository shell/editor", "sampling": "pass@1; repeat count is score-level provenance", "judge": "program-based functional and regression verifiers", "harness": "Pier newer than 0.3.0 with a separate pristine verifier environment", "notes": "Official 113-task v1.1 corpus across five languages. The official leaderboard uses Pier and mini-swe-agent; Meta's chart uses selected agent products and is not leaderboard-harness-identical." } }, { "id": "finance_agent_v2", "name": "Finance Agent v2", "category": "Finance", "metric": "% weighted partial credit", "num_problems": 927, "source_url": "https://www.vals.ai/benchmarks/fabv2", "canonical_setting": { "version": "Finance Agent v2", "metric_type": "weighted_partial_credit_pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": false, "notes": "Finance Agent v2 has 927 questions: 27 public, 450 private validation, 450 held-out test. Published scores use the hidden 450-task test and dealbreaker-gated severity-weighted partial credit.", "judge": "three-model jury: GPT-5.4, Gemini 3.1 Pro, Claude Sonnet 4.6", "sampling": "3 runs/model on 450 held-out scored tasks", "tools": "Vals six-tool finance-agent harness" } }, { "id": "frontier_swe", "name": "FrontierSWE", "category": "Agentic Coding", "metric": "dominance (%)", "num_problems": 17, "source_url": "https://www.frontierswe.com/", "canonical_setting": { "version": "FrontierSWE initial 17-task release; 20-hour budget", "metric_type": "dominance_pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": false, "tools": "agentic code execution", "sampling": "mean@5", "judge": "continuous partial-credit task scoring and dominance", "harness": "public FrontierSWE harness", "notes": "Dominance is win probability against a random opponent over continuous task scores; Grok 4.5 uses Grok CLI." } }, { "id": "harvey_lab_aa", "name": "Harvey LAB-AA", "category": "Agentic", "metric": "% rubric criteria passed", "num_problems": 120, "source_url": "https://artificialanalysis.ai/evaluations/harvey-lab-aa", "canonical_setting": { "version": "Artificial Analysis 120-task LAB-AA implementation", "metric_type": "criterion_pass_rate_pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": true, "notes": "Criterion pass rate over 120 private legal tasks across 24 practice areas." } }, { "id": "job_bench", "name": "JobBench", "category": "Agentic", "metric": "% weighted rubric score", "num_problems": 65, "source_url": "https://github.com/Job-Bench/job-bench-eval", "canonical_setting": { "version": "Main 65-task leaderboard split", "metric_type": "weighted_rubric_score_pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": true, "notes": "Professional-work deliverables scored by weighted rubrics; excludes the easy smoke-test split." } }, { "id": "legal_research_bench", "name": "Legal Research Bench", "category": "Agentic", "metric": "% all-pass accuracy", "num_problems": null, "source_url": "https://www.vals.ai/benchmarks/legal_research", "canonical_setting": { "version": "Current Vals held-out suite", "metric_type": "all_pass_accuracy_pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": false, "notes": "A question passes only when every required rubric item passes; exact hidden count is not public." } }, { "id": "osworld_2_0", "name": "OSWorld 2.0", "category": "Agentic", "metric": "% weighted checkpoint score", "num_problems": 108, "source_url": "https://osworld-v2.xlang.ai/", "canonical_setting": { "version": "OSWorld 2.0", "metric_type": "weighted_checkpoint_score_pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": true, "notes": "Separate 108-workflow benchmark with weighted checkpoint partial scoring at the standard 500-step budget." } }, { "id": "osworld_verified", "name": "OSWorld-Verified", "category": "Agentic", "metric": "% task success", "num_problems": 369, "source_url": "https://xlang.ai/blog/osworld-verified", "canonical_setting": { "version": "OSWorld-Verified revision announced 2025-07-28", "metric_type": "task_success_rate_pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": true, "notes": "Repaired OSWorld 1.x lineage, distinct from OSWorld 2.0. Some runs exclude eight Google Drive tasks; score-level notes must disclose exclusions when known." } }, { "id": "perception_bench", "name": "PerceptionBench", "category": "Vision Perception", "metric": "% accuracy", "num_problems": 3000, "source_url": "https://github.com/MoonshotAI/PerceptionBench", "canonical_setting": { "version": "Initial 3,000-question release", "metric_type": "accuracy_pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": true, "notes": "Open-ended visual perception questions across ten atomic capabilities; binary judge verdict per response." } }, { "id": "posttrain_bench", "name": "PostTrainBench", "category": "Agentic Research", "metric": "weighted average objective score (%)", "num_problems": 28, "source_url": "https://github.com/aisa-group/PostTrainBench", "canonical_setting": { "version": "PostTrainBench", "metric_type": "pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": false, "tools": "post-training pipeline modification", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "one H100 for 10 hours", "notes": "Seven objectives across four base models; count actual objective runs as 28." } }, { "id": "saas_bench", "name": "SaaS-Bench", "category": "Agentic", "metric": "% checkpoint score", "num_problems": 106, "source_url": "https://github.com/UniPat-AI/SaaS-Bench", "canonical_setting": { "version": "Initial 106-task release", "metric_type": "checkpoint_score_pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": true, "notes": "Workflow checkpoint score across 23 self-hosted SaaS applications; distinct from strict resolved-task rate." } }, { "id": "spreadsheetbench_2", "name": "SpreadsheetBench 2", "category": "Office", "metric": "% modification accuracy", "num_problems": 321, "source_url": "https://github.com/RUCKBReasoning/SpreadsheetBench-2", "canonical_setting": { "version": "SpreadsheetBench 2", "metric_type": "modification_accuracy_pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": true, "notes": "Spreadsheet modification benchmark covering debugging, financial models, templates, and visualization." } }, { "id": "swe_marathon_h20_2026_07_09", "name": "SWE-Marathon H20 Snapshot (2026-07-09)", "category": "Agentic Coding", "metric": "% resolved (pass@1)", "num_problems": 20, "source_url": "https://github.com/abundant-ai/swe-marathon", "canonical_setting": { "version": "H20-calibrated pre-final-v1.1 branch dated 2026-07-09", "metric_type": "pass_at_1_pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": false, "notes": "Source-specific Moonshot H20 calibration before final v1.1; kept separate from canonical SWE-Marathon releases." } }, { "id": "tau3_banking", "name": "τ³-Banking", "category": "Tool Use", "metric": "% passed (pass@1)", "num_problems": 97, "source_url": "https://github.com/sierra-research/tau2-bench", "canonical_setting": { "version": "τ³-bench banking_knowledge", "metric_type": "pass_at_1_pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": false, "notes": "Knowledge-retrieval and transactional banking customer-service scenarios. K3 source does not pin the corrected release." } }, { "id": "terminal_bench_2_1", "name": "Terminal-Bench 2.1", "category": "Agentic Coding", "metric": "% resolved (pass@1)", "num_problems": 89, "source_url": "https://www.tbench.ai/news/terminal-bench-2-1", "canonical_setting": { "version": "Terminal-Bench 2.1", "metric_type": "pass_at_1_pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": false, "tools": "terminal agent in the official container task", "sampling": "pass@1; leaderboard submission requires at least five trials per task", "judge": "task executable verifier", "harness": "Harbor with submitted agent/model/sandbox provenance", "notes": "All 89 tasks. The official release page says 28 tasks changed from 2.0; the pinned official repository README says 26. The task count agrees and the changed-task-count conflict remains explicit." } }, { "id": "toolathlon_verified", "name": "Toolathlon-Verified", "category": "Tool Use", "metric": "mean pass@1 (%)", "num_problems": 108, "source_url": "https://toolathlon.xyz/docs/blog/toolathlon-verified", "canonical_setting": { "version": "Toolathlon-Verified released 2026-06-30", "metric_type": "mean_pass_at_1_pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": true, "notes": "Official repaired release, separate from original Toolathlon; mean pass@1 across three runs." } }, { "id": "coding_experience", "name": "Coding Experience", "category": "Coding", "metric": "score (%)", "num_problems": null, "source_url": "https://github.com/MoonshotAI/Kimi-K3/blob/main/k3_tech_report.pdf", "canonical_setting": { "version": "Kimi internal Coding Experience; version unspecified", "metric_type": "pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": false, "notes": "Practical coding-agent experience in real development workflows; provider-aligned Claude Code, Kimi Code, or Codex harness." } }, { "id": "clawbench_2_0", "name": "24/7 ClawBench 2.0", "category": "Agentic", "metric": "score (%)", "num_problems": null, "source_url": "https://github.com/MoonshotAI/Kimi-K3/blob/main/k3_tech_report.pdf", "canonical_setting": { "version": "24/7 ClawBench 2.0", "metric_type": "pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": false, "notes": "Always-on, multi-day assistant tasks with concurrent events and interruptions; OpenClaw harness." } }, { "id": "mira_bench", "name": "MIRA Bench", "category": "Agentic", "metric": "score (%)", "num_problems": null, "source_url": "https://github.com/MoonshotAI/Kimi-K3/blob/main/k3_tech_report.pdf", "canonical_setting": { "version": "Kimi internal MIRA Bench; version unspecified", "metric_type": "pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": false, "notes": "Multi-agent enterprise collaboration, delegation, and routing; MIRA harness." } }, { "id": "kaet", "name": "KAET", "category": "Agentic", "metric": "score (%)", "num_problems": null, "source_url": "https://github.com/MoonshotAI/Kimi-K3/blob/main/k3_tech_report.pdf", "canonical_setting": { "version": "Kimi Autonomous Execution Tasks; version unspecified", "metric_type": "pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": false, "notes": "Long-horizon autonomous execution simulating user requests and enterprise operations; Kimi Code harness." } }, { "id": "clif_bench", "name": "CLIF Bench", "category": "Agentic", "metric": "score (%)", "num_problems": null, "source_url": "https://github.com/MoonshotAI/Kimi-K3/blob/main/k3_tech_report.pdf", "canonical_setting": { "version": "Context Learning and Instruction Following Bench; version unspecified", "metric_type": "pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": false, "notes": "In-context learning with instructions interleaving multiple complex skills; Kimi Code harness." } }, { "id": "agentic_vision_bench", "name": "Agentic Vision Bench", "category": "Vision Agent", "metric": "score (%)", "num_problems": null, "source_url": "https://github.com/MoonshotAI/Kimi-K3/blob/main/k3_tech_report.pdf", "canonical_setting": { "version": "Kimi internal Agentic Vision Bench; version unspecified", "metric_type": "pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": true, "notes": "Tests whether agents notice and use key visual facts during execution; Kimi Code harness." } }, { "id": "swarm_bench", "name": "SwarmBench", "category": "Agentic", "metric": "score (%)", "num_problems": null, "source_url": "https://github.com/MoonshotAI/Kimi-K3/blob/main/k3_tech_report.pdf", "canonical_setting": { "version": "Kimi internal SwarmBench; version unspecified", "metric_type": "pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": false, "notes": "Agent-swarm orchestration through coordinated decomposition and parallel execution; Kimi Agent harness." } }, { "id": "online_experience", "name": "Online Experience", "category": "Agentic", "metric": "score (%)", "num_problems": null, "source_url": "https://github.com/MoonshotAI/Kimi-K3/blob/main/k3_tech_report.pdf", "canonical_setting": { "version": "Kimi internal Online Experience; version unspecified", "metric_type": "pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": false, "notes": "Real online-agent usage and commonly requested deliverable file types; Kimi Agent harness." } }, { "id": "finance_bench", "name": "Finance Bench", "category": "Finance", "metric": "score (%)", "num_problems": null, "source_url": "https://github.com/MoonshotAI/Kimi-K3/blob/main/k3_tech_report.pdf", "canonical_setting": { "version": "Kimi internal Finance Bench; version unspecified", "metric_type": "pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": false, "notes": "Kimi internal realistic financial work from source materials to reviewable deliverables; distinct from public FinanceBench." } }, { "id": "kwv_bench", "name": "KWVBench", "category": "Vision Agent", "metric": "score (%)", "num_problems": null, "source_url": "https://github.com/MoonshotAI/Kimi-K3/blob/main/k3_tech_report.pdf", "canonical_setting": { "version": "Kimi internal Knowledge Work Vision Bench; version unspecified", "metric_type": "pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": true, "notes": "Atomic visual capabilities distilled from real knowledge-work scenarios." } }, { "id": "deck_bench", "name": "DECKBench", "category": "Office", "metric": "score (%)", "num_problems": null, "source_url": "https://github.com/MoonshotAI/Kimi-K3/blob/main/k3_tech_report.pdf", "canonical_setting": { "version": "Kimi internal DECKBench; version unspecified", "metric_type": "pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": false, "notes": "Presentation-deck generation from real-usage task descriptions." } }, { "id": "agent_behavior_bench", "name": "Agent Behavior Bench", "category": "Agentic", "metric": "score (%)", "num_problems": null, "source_url": "https://github.com/MoonshotAI/Kimi-K3/blob/main/k3_tech_report.pdf", "canonical_setting": { "version": "Kimi internal Agent Behavior Bench; version unspecified", "metric_type": "pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": false, "notes": "Process quality, tool use, efficiency, and discipline alongside task completion; Kimi Work harness." } }, { "id": "faithfulness", "name": "Faithfulness", "category": "Factuality", "metric": "1 - hallucination rate (%)", "num_problems": null, "source_url": "https://github.com/MoonshotAI/Kimi-K3/blob/main/k3_tech_report.pdf", "canonical_setting": { "version": "Kimi internal Faithfulness; version unspecified", "metric_type": "pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": false, "notes": "Responses are fact-checked; the reported higher-is-better metric is one minus hallucination rate." } }, { "id": "chat_all_in_one_bench", "name": "Chat All-in-One Bench", "category": "Chat", "metric": "score (%)", "num_problems": null, "source_url": "https://github.com/MoonshotAI/Kimi-K3/blob/main/k3_tech_report.pdf", "canonical_setting": { "version": "Kimi internal Chat All-in-One Bench; version unspecified", "metric_type": "pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": false, "notes": "Conversational experience across every stage of product usage; Kimi Work harness." } }, { "id": "aa_intelligence_index_v4_1", "name": "Artificial Analysis Intelligence Index v4.1", "category": "Composite", "metric": "composite score", "num_problems": null, "source_url": "https://artificialanalysis.ai/methodology/intelligence-benchmarking", "canonical_setting": { "version": "Artificial Analysis Intelligence Index v4.1", "metric_type": "weighted_composite_score", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": false, "notes": "Version-pinned English text-only composite. Stored scores are the K3 report's 2026-07-23 live leaderboard snapshot." } }, { "id": "vals_index", "name": "Vals Index", "category": "Composite", "metric": "weighted accuracy (%)", "num_problems": null, "source_url": "https://www.vals.ai/benchmarks/vals_index", "canonical_setting": { "version": "Live Vals Index snapshot", "metric_type": "weighted_accuracy_pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": false, "notes": "GDP-weighted professional benchmark composite; methodology can change over time." } }, { "id": "webdev_arena_elo", "name": "WebDevArena Elo", "category": "Human Preference", "metric": "Arena score", "num_problems": null, "source_url": "https://arena.ai/leaderboard/code/webdev", "canonical_setting": { "version": "Live leaderboard snapshot", "metric_type": "arena_score", "range": null, "higher_is_better": true, "multimodal_input": false, "notes": "Blind pairwise preference over generated web applications; scores drift as votes accumulate." } }, { "id": "text_arena_elo", "name": "Arena Text Elo", "category": "Human Preference", "metric": "Elo rating", "num_problems": null, "source_url": "https://arena.ai/leaderboard/text", "canonical_setting": { "version": "Live Arena Text leaderboard snapshot", "metric_type": "elo", "higher_is_better": true, "range": null, "multimodal_input": false, "tools": "none", "judge": "blind human pairwise preference votes", "notes": "Dynamic Elo snapshot; score observations must record snapshot date because ratings drift." } }, { "id": "agent_arena", "name": "AgentArena", "category": "Agentic", "metric": "net improvement", "num_problems": null, "source_url": "https://arena.ai/blog/agent-arena-methodology", "canonical_setting": { "version": "Live leaderboard snapshot", "metric_type": "mean_treatment_effect", "range": null, "higher_is_better": true, "multimodal_input": false, "notes": "Causal treatment-effect estimate over real Agent Mode sessions; no fixed task count or invariant baseline." } }, { "id": "moonshot_exploit_development_suite", "name": "Moonshot Exploit Development Suite", "category": "Cyber", "metric": "% tasks solved", "num_problems": 36, "source_url": "https://github.com/MoonshotAI/Kimi-K3/blob/main/k3_tech_report.pdf", "canonical_setting": { "version": "Kimi K3 report 36-task in-house exploit suite", "metric_type": "task_solve_rate_pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": false, "notes": "16 user-space CVE exploitation tasks and 20 historical Linux-kernel CVE tasks in QEMU; human-verified solvable." } }, { "id": "exploitbench", "name": "ExploitBench", "category": "Cyber", "metric": "% capability success", "num_problems": 41, "source_url": "https://arxiv.org/abs/2605.14153", "canonical_setting": { "version": "ExploitBench V8 41-task corpus (2026)", "metric_type": "capability_ladder_success_pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": false, "notes": "Public 41-vulnerability exploitation benchmark with deterministically verified capability ladders." } }, { "id": "dsbench_fullstack", "name": "DSBench-FullStack", "category": "Agentic Coding", "metric": "score (%)", "num_problems": null, "source_url": "https://huggingface.co/deepseek-ai/DeepSeek-V4-Flash-0731", "canonical_setting": { "version": "DeepSeek internal DSBench-FullStack; version unspecified", "metric_type": "pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": false, "notes": "DeepSeek internal full-stack development test set. The official model card does not report its task count." } }, { "id": "dsbench_hard", "name": "DSBench-Hard", "category": "Agentic Coding", "metric": "score (%)", "num_problems": null, "source_url": "https://huggingface.co/deepseek-ai/DeepSeek-V4-Flash-0731", "canonical_setting": { "version": "DeepSeek internal DSBench-Hard; version unspecified", "metric_type": "pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": false, "notes": "DeepSeek internal test set of difficult coding-agent problems. The official model card does not report its task count." } }, { "id": "hle_tools_text", "name": "HLE Text (w/ tools)", "category": "Reasoning & Knowledge", "metric": "accuracy (%)", "num_problems": 2158, "source_url": "https://z.ai/blog/glm-5.2", "canonical_setting": { "version": "HLE finalized text-only subset (2,158 questions) with tools", "metric_type": "pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": false, "tools": "search, code execution, and web browsing", "judge": "answer-key scoring for closed-ended answers", "sampling": "pass@1", "notes": "GLM-5.2 explicitly marks unstarred HLE-with-tools values as text-only and starred values as the full text+image set. Text-only and full-set scores are distinct benchmark identities." } }, { "id": "swe_marathon_v1_0", "name": "SWE-Marathon v1.0", "category": "Agentic Coding", "metric": "% resolved (pass@1)", "num_problems": 20, "source_url": "https://github.com/abundant-ai/swe-marathon/releases/tag/v1.0", "canonical_setting": { "version": "SWE-Marathon v1.0 released 2026-06-10", "metric_type": "pass_at_1_pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": false, "tools": "agentic code execution", "sampling": "trials=5; reported as pass@1", "notes": "Official v1.0 evaluates 20 ultra-long-horizon tasks with five trials per agent-model-task pair. It predates and is distinct from the July v1.1/H20-calibrated snapshot already stored in BP." } }, { "id": "hy_backend_2_0", "name": "Hy-Backend 2.0 (Internal)", "category": "Agentic Coding", "source_url": "https://huggingface.co/tencent/Hy3", "num_problems": null, "metric": "reported score (%)", "canonical_setting": { "version": "Hy-Backend 2.0 (Internal)", "metric_type": "pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": false, "sampling": "pass@1", "tools": "Claude Code; GPT-5.5 uses CodeX", "notes": "Tencent internal backend benchmark; public item count and metric definition are not published." } }, { "id": "hy_swe_max", "name": "Hy-SWE Max (Internal)", "category": "Agentic Coding", "source_url": "https://huggingface.co/tencent/Hy3", "num_problems": null, "metric": "reported score (%)", "canonical_setting": { "version": "Hy-SWE Max (Internal)", "metric_type": "pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": false, "sampling": "pass@1", "tools": "Claude Code; GPT-5.5 uses CodeX", "notes": "Tencent internal software-engineering benchmark; public item count and metric definition are not published." } }, { "id": "hy_company_bench", "name": "Hy-CompanyBench (Internal)", "category": "Agentic Coding", "source_url": "https://huggingface.co/tencent/Hy3", "num_problems": null, "metric": "reported score (%)", "canonical_setting": { "version": "Hy-CompanyBench (Internal)", "metric_type": "pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": false, "sampling": "pass@1", "tools": "Claude Code; GPT-5.5 uses CodeX", "notes": "Tencent internal company-task benchmark; public item count and metric definition are not published." } }, { "id": "wildclaw_bench_35_text", "name": "WildClawBench (35, text-only)", "category": "Agentic", "source_url": "https://huggingface.co/tencent/Hy3", "num_problems": 35, "metric": "reported score (%)", "canonical_setting": { "version": "WildClawBench (35, text-only)", "metric_type": "pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": false, "sampling": "pass@1", "tools": "OpenClaw harness", "notes": "Official Hy3 footnote defines the text-only 35-query subset." } }, { "id": "skills_bench_text_79", "name": "SkillsBench (79, text-only)", "category": "Agentic", "source_url": "https://huggingface.co/tencent/Hy3", "num_problems": 79, "metric": "reported score (%)", "canonical_setting": { "version": "SkillsBench (79, text-only)", "metric_type": "pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": false, "sampling": "average over 3 runs", "tools": "Claude Code", "notes": "Self-contained 79-task subset; multimodal tasks excluded." } }, { "id": "e_bench_internal", "name": "e-bench (Internal)", "category": "Agentic", "source_url": "https://huggingface.co/tencent/Hy3", "num_problems": null, "metric": "reported score (%)", "canonical_setting": { "version": "e-bench (Internal)", "metric_type": "pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": false, "sampling": "pass@1", "tools": "benchmark-specific agent tools", "notes": "Tencent internal working-agent benchmark; public item count and protocol are not published." } }, { "id": "hy_finmodel_bench", "name": "Hy-FinModelBench (Internal)", "category": "Agentic", "source_url": "https://huggingface.co/tencent/Hy3", "num_problems": null, "metric": "reported score (%)", "canonical_setting": { "version": "Hy-FinModelBench (Internal)", "metric_type": "pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": false, "sampling": "pass@1", "tools": "benchmark-specific agent tools", "notes": "Tencent internal financial-modeling benchmark; public item count and protocol are not published." } }, { "id": "prod_bench_internal", "name": "ProdBench (Internal, pass^3)", "category": "Agentic", "source_url": "https://huggingface.co/tencent/Hy3", "num_problems": null, "metric": "reported score (%)", "canonical_setting": { "version": "ProdBench (Internal, pass^3)", "metric_type": "pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": false, "sampling": "pass^3", "tools": "OpenClaw harness", "notes": "Tencent internal productivity benchmark evaluated with OpenClaw; public item count is not published." } }, { "id": "hy_skillsworld", "name": "Hy-SkillsWorld (Internal)", "category": "Agentic", "source_url": "https://huggingface.co/tencent/Hy3", "num_problems": null, "metric": "reported score (%)", "canonical_setting": { "version": "Hy-SkillsWorld (Internal)", "metric_type": "pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": false, "sampling": "pass@1", "tools": "benchmark-specific agent tools", "notes": "Tencent internal skills benchmark; public item count and protocol are not published." } }, { "id": "hy_euler_pro", "name": "Hy-Euler Pro (Internal, tools)", "category": "Reasoning & Knowledge", "source_url": "https://huggingface.co/tencent/Hy3", "num_problems": null, "metric": "reported score (%)", "canonical_setting": { "version": "Hy-Euler Pro (Internal, tools)", "metric_type": "pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": false, "sampling": "pass@1", "tools": "benchmark-specific tools", "notes": "Tencent internal STEM-agent benchmark; public item count and metric definition are not published." } }, { "id": "horizon_math_pass12", "name": "HorizonMath (pass@12)", "category": "Reasoning & Knowledge", "source_url": "https://github.com/ewang26/HorizonMath", "num_problems": 113, "metric": "reported score (%)", "canonical_setting": { "version": "HorizonMath (pass@12)", "metric_type": "pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": false, "sampling": "pass@12", "tools": "none", "notes": "Official HorizonMath repository defines 113 automatically verified research problems across eight domains." } }, { "id": "hy_math_internal", "name": "Hy-Math (Internal)", "category": "Reasoning & Knowledge", "source_url": "https://huggingface.co/tencent/Hy3", "num_problems": null, "metric": "reported score (%)", "canonical_setting": { "version": "Hy-Math (Internal)", "metric_type": "pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": false, "sampling": "pass@1", "tools": "none", "notes": "Tencent internal mathematics benchmark; public item count and protocol are not published." } }, { "id": "cmt_benchmark", "name": "CMT-Benchmark", "category": "Reasoning & Knowledge", "source_url": "https://github.com/JamesRoggeveen/cmt_benchmark_data", "num_problems": 50, "metric": "reported score (%)", "canonical_setting": { "version": "CMT-Benchmark", "metric_type": "pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": false, "sampling": "pass@1", "tools": "none", "notes": "Official CMT-Benchmark paper/repository defines 50 expert-authored condensed-matter problems." } }, { "id": "cl_bench_life", "name": "CL-Bench Life", "category": "Long Context", "source_url": "https://huggingface.co/datasets/tencent/CL-bench-Life", "num_problems": 405, "metric": "reported score (%)", "canonical_setting": { "version": "CL-Bench Life", "metric_type": "pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": false, "sampling": "pass@1", "tools": "none", "notes": "Official Tencent dataset defines 405 context-task pairs and 5,348 evaluation rubrics." } }, { "id": "forte_avg3", "name": "FORTE (Avg@3)", "category": "Agents", "metric": "Avg@3 (%)", "num_problems": 180, "source_url": "https://github.com/AGI-Eval-Official/FORTE", "canonical_setting": { "version": "FORTE June 2026", "metric_type": "pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": true, "tools": "office-computing environment", "sampling": "trials=3", "judge": "LLM-as-judge over expert rubrics; all-or-nothing per run", "harness": "OpenClaw in Docker", "notes": "Official README declares 180 full tasks across 15 professions. The official leaderboard JSON values appear arithmetically compatible with a 183-task denominator, an unresolved publisher inconsistency. LongCat used 45-minute task timeouts; public demo schemas use 40 minutes." } }, { "id": "rwsearch", "name": "RWSearch", "category": "Agents", "metric": "publisher Score (0-100)", "num_problems": 200, "source_url": "https://github.com/AGI-Eval-Official/RW-Search", "canonical_setting": { "version": "RWSearch 2026", "metric_type": "pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": false, "tools": "Search and Browse", "context_management": "none", "notes": "Official repository defines 200 Chinese real-world search questions with unique objective answers. The publisher labels the metric only as Score and does not disclose the full matcher or aggregation." } }, { "id": "advancedif", "name": "AdvancedIF", "category": "Instruction Following", "metric": "overall pass rate (%)", "num_problems": 1645, "source_url": "https://arxiv.org/abs/2511.10507", "canonical_setting": { "version": "AdvancedIF public test split", "metric_type": "pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": false, "tools": "none", "judge": "LLM-as-judge; public evaluator default o3-mini-2025-01-31", "sampling": "pass@1", "notes": "1,645 expert-rubric prompts across system steerability, carried context, and complex instruction following. Canonical score is the percentage of samples where all rubrics pass." } }, { "id": "xlrs_bench_micro", "name": "XLRS-Bench (micro)", "category": "Vision STEM", "metric": "micro-average (%)", "num_problems": 45942, "source_url": "https://arxiv.org/abs/2503.23771", "canonical_setting": { "version": "XLRS-Bench full", "metric_type": "pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": true, "tools": "none", "sampling": "pass@1", "notes": "Full ultra-high-resolution remote-sensing benchmark with 45,942 annotations across 16 tasks. The official full-benchmark Avg. is a micro average; XLRS-Bench-lite uses a macro average." } }, { "id": "microvqa", "name": "MicroVQA", "category": "Vision STEM", "metric": "mean accuracy (%)", "num_problems": 1042, "source_url": "https://arxiv.org/abs/2503.13399", "canonical_setting": { "version": "MicroVQA v0.0.1 test split", "metric_type": "pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": true, "tools": "none", "judge": "rule-based multiple-choice accuracy", "sampling": "pass@1", "notes": "1,042 expert-curated microscopy multiple-choice questions covering perception, hypothesis generation, and experiment proposal." } }, { "id": "sgi_bench", "name": "SGI-Bench", "category": "Scientific Agents", "metric": "SGI-Score", "num_problems": null, "source_url": "https://arxiv.org/abs/2512.16969", "canonical_setting": { "version": "SGI-Bench full gated suite", "metric_type": "pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": true, "tools": "web search, PDF parser, Python interpreter, file reader", "judge": "task-specific official metrics aggregated into SGI-Score", "harness": "official SGI-Bench agentic evaluation", "notes": "Scientist-aligned full inquiry-cycle suite spanning 10 disciplines and more than 1,000 gated expert-curated samples. The publisher does not state one exact total item count." } }, { "id": "researchclawbench", "name": "ResearchClawBench", "category": "Scientific Agents", "metric": "weighted rubric score (%)", "num_problems": 40, "source_url": "https://arxiv.org/abs/2606.07591", "canonical_setting": { "version": "ResearchClawBench core", "metric_type": "pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": true, "tools": "scientific research environment", "judge": "weighted task-specific rubrics", "harness": "ResearchHarness", "notes": "40 real-science research tasks across 10 domains." } }, { "id": "gdpval_normalized_elo", "name": "GDPVal (normalized Elo)", "category": "Economic", "metric": "normalized Elo (0-100)", "num_problems": 220, "source_url": "https://huggingface.co/datasets/openai/gdpval", "canonical_setting": { "version": "GDPVal public 220-task set; normalized Elo on 0-100 scale", "metric_type": "index", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": false, "tools": "office workflow, web search, sandboxed code", "sampling": "pass@1", "judge": "Gemini 3.1 Pro rubric judge", "notes": "Distinct from raw GDPVal Artificial Analysis Elo. NVIDIA reports normalized=(Elo-500)/2000 on a 0-100 display scale." } }, { "id": "profbench", "name": "ProfBench (Search)", "category": "Search Agent", "metric": "search score", "num_problems": 40, "source_url": "https://github.com/NVlabs/ProfBench", "canonical_setting": { "version": "ProfBench full 40-task release", "metric_type": "pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": false, "tools": "web search and browsing", "sampling": "16-run average", "notes": "Professional-domain rubric tasks with search and browsing.", "judge": "rubric-based criterion grading" } }, { "id": "pinchbench", "name": "PinchBench", "category": "Agentic", "metric": "% passed", "num_problems": 53, "source_url": "https://github.com/pinchbench/skill", "canonical_setting": { "version": "PinchBench public 53-task release", "metric_type": "pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": false, "tools": "OpenClaw coding environment", "sampling": "pass@1", "notes": "Open public subset evaluated in the OpenClaw environment." } }, { "id": "tau3_airline", "name": "tau3-bench Airline", "category": "Tool Use", "metric": "% passed (8-trial average)", "num_problems": 400, "source_url": "https://github.com/sierra-research/tau2-bench", "canonical_setting": { "version": "tau3 airline domain", "metric_type": "pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": false, "tools": "domain customer-service tools", "sampling": "50 tasks x 8 trials", "notes": "50 tasks x 8 trials; extra simulator prompt and GPT-5.2-low simulator.", "judge": "state-based task success" } }, { "id": "tau3_retail", "name": "tau3-bench Retail", "category": "Tool Use", "metric": "% passed (8-trial average)", "num_problems": 912, "source_url": "https://github.com/sierra-research/tau2-bench", "canonical_setting": { "version": "tau3 retail domain", "metric_type": "pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": false, "tools": "domain customer-service tools", "sampling": "114 tasks x 8 trials", "notes": "114 tasks x 8 trials; extra simulator prompt and GPT-5.2-low simulator.", "judge": "state-based task success" } }, { "id": "tau3_telecom", "name": "tau3-bench Telecom", "category": "Tool Use", "metric": "% passed (8-trial average)", "num_problems": 912, "source_url": "https://github.com/sierra-research/tau2-bench", "canonical_setting": { "version": "tau3 telecom domain", "metric_type": "pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": false, "tools": "domain customer-service tools", "sampling": "114 tasks x 8 trials", "notes": "114 tasks x 8 trials; extra simulator prompt and GPT-5.2-low simulator.", "judge": "state-based task success" } }, { "id": "ioi_2025", "name": "IOI 2025", "category": "Coding", "metric": "contest points (0-600)", "num_problems": 6, "source_url": "https://ioi2025.bo/", "canonical_setting": { "version": "International Olympiad in Informatics 2025", "metric_type": "score", "range": [ 0, 600 ], "higher_is_better": true, "multimodal_input": false, "tools": "code execution", "sampling": "pass@1", "notes": "Six official contest problems; score is points, not percent." } }, { "id": "aa_omniscience_accuracy", "name": "AA Omniscience Accuracy", "category": "Factuality", "metric": "% correct", "num_problems": 60000, "source_url": "https://artificialanalysis.ai/articles/aa-omniscience-knowledge-hallucination-benchmark", "canonical_setting": { "version": "AA-Omniscience 6,000-question benchmark", "metric_type": "pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": false, "tools": "none", "sampling": "10-run average", "notes": "Count is 6,000 questions x 10 repeats. Accuracy=c/(c+p+i+a).", "judge": "official answer and abstention classification" } }, { "id": "aa_omniscience_non_hallucination", "name": "AA Omniscience Non-Hallucination", "category": "Hallucination", "metric": "% non-hallucination", "num_problems": 60000, "source_url": "https://artificialanalysis.ai/articles/aa-omniscience-knowledge-hallucination-benchmark", "canonical_setting": { "version": "AA-Omniscience 6,000-question benchmark", "metric_type": "pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": false, "tools": "none", "sampling": "10-run average", "notes": "Count is 6,000 questions x 10 repeats. Non-hallucination=100-i/(p+i+a).", "judge": "official answer and abstention classification" } }, { "id": "ruler_1m", "name": "RULER 1M", "category": "Long Context", "metric": "%", "num_problems": null, "source_url": "https://github.com/NVIDIA/RULER", "canonical_setting": { "version": "RULER at 1M tokens", "metric_type": "pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": false, "tools": "none", "sampling": "pass@1", "notes": "Official RULER aggregate at the exact 1M context length." } }, { "id": "wmt24pp", "name": "WMT24++", "category": "Multilingual", "metric": "XCOMET-XXL", "num_problems": 54890, "source_url": "https://huggingface.co/datasets/google/wmt24pp", "canonical_setting": { "version": "WMT24++ English-to-55-language evaluation", "metric_type": "pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": false, "tools": "none", "sampling": "54,890 translations", "notes": "998 English paragraphs x 55 target languages = 54,890 translations.", "judge": "XCOMET-XXL" } }, { "id": "agieval_en", "name": "AGIEval English", "category": "Reasoning & Knowledge", "metric": "% exact match", "num_problems": null, "source_url": "https://github.com/microsoft/AGIEval", "canonical_setting": { "version": "AGIEval English aggregate", "metric_type": "pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": false, "tools": "none", "sampling": "pass@1", "notes": "English tasks evaluated with benchmark-specific 3-shot or 5-shot CoT." } }, { "id": "math_test", "name": "MATH Test", "category": "Math", "metric": "% exact match", "num_problems": 5000, "source_url": "https://github.com/hendrycks/math", "canonical_setting": { "version": "MATH test split (5,000 problems)", "metric_type": "pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": false, "tools": "none", "sampling": "pass@1", "notes": "Minerva 4-shot exact-match setting; distinct from full 12,500-example MATH." } }, { "id": "mbpp_sanitized", "name": "MBPP Sanitized", "category": "Coding", "metric": "sampled pass@1 %", "num_problems": 13664, "source_url": "https://github.com/evalplus/evalplus", "canonical_setting": { "version": "EvalPlus MBPP sanitized 427-problem set", "metric_type": "pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": false, "tools": "none", "sampling": "427 problems x 32 samples", "notes": "Count is 427 problems x 32 samples = 13,664 model generations." } }, { "id": "openbookqa", "name": "OpenBookQA", "category": "Knowledge", "metric": "% normalized accuracy", "num_problems": 500, "source_url": "https://github.com/allenai/OpenBookQA", "canonical_setting": { "version": "OpenBookQA test split", "metric_type": "pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": false, "tools": "none", "sampling": "pass@1", "notes": "Official 500-question test split, zero-shot normalized accuracy." } }, { "id": "piqa", "name": "PIQA", "category": "Reasoning", "metric": "% normalized accuracy", "num_problems": 1838, "source_url": "https://yonatanbisk.com/piqa/", "canonical_setting": { "version": "PIQA validation split", "metric_type": "pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": false, "tools": "none", "sampling": "pass@1", "notes": "Official 1,838-question validation split, zero-shot normalized accuracy." } }, { "id": "winogrande", "name": "WinoGrande", "category": "Reasoning", "metric": "% accuracy", "num_problems": 1267, "source_url": "https://winogrande.allenai.org/", "canonical_setting": { "version": "WinoGrande validation split", "metric_type": "pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": false, "tools": "none", "sampling": "pass@1", "notes": "Official 1,267-question validation split, 5-shot accuracy." } }, { "id": "race", "name": "RACE", "category": "Reasoning & Knowledge", "metric": "% accuracy", "num_problems": 4934, "source_url": "https://www.cs.cmu.edu/~glai1/data/race/", "canonical_setting": { "version": "RACE test split", "metric_type": "pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": false, "tools": "none", "sampling": "pass@1", "notes": "Official 4,934-question test split, zero-shot accuracy." } }, { "id": "ruler_64k", "name": "RULER 64K", "category": "Long Context", "metric": "%", "num_problems": null, "source_url": "https://github.com/NVIDIA/RULER", "canonical_setting": { "version": "RULER at 64K tokens", "metric_type": "pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": false, "tools": "none", "sampling": "pass@1", "notes": "Official RULER aggregate at the exact 64K context length." } }, { "id": "ruler_256k", "name": "RULER 256K", "category": "Long Context", "metric": "%", "num_problems": null, "source_url": "https://github.com/NVIDIA/RULER", "canonical_setting": { "version": "RULER at 256K tokens", "metric_type": "pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": false, "tools": "none", "sampling": "pass@1", "notes": "Official RULER aggregate at the exact 256K context length." } }, { "id": "ruler_512k", "name": "RULER 512K", "category": "Long Context", "metric": "%", "num_problems": null, "source_url": "https://github.com/NVIDIA/RULER", "canonical_setting": { "version": "RULER at 512K tokens", "metric_type": "pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": false, "tools": "none", "sampling": "pass@1", "notes": "Official RULER aggregate at the exact 512K context length." } }, { "id": "imo_proofbench_advanced", "name": "IMO-ProofBench Advanced", "category": "Math", "metric": "% of 210 points", "num_problems": 30, "source_url": "https://github.com/google-deepmind/superhuman/tree/main/imobench", "canonical_setting": { "version": "IMO-ProofBench Advanced 30-problem split", "metric_type": "pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": false, "tools": "none", "sampling": "pass@1", "notes": "Thirty proof problems worth seven points each; 210 points total." } }, { "id": "putnam_2025", "name": "Putnam 2025", "category": "Math", "metric": "% of 120 points", "num_problems": 12, "source_url": "https://maa.org/math-competitions/william-lowell-putnam-mathematical-competition", "canonical_setting": { "version": "William Lowell Putnam Mathematical Competition 2025", "metric_type": "pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": false, "tools": "none", "sampling": "pass@1", "notes": "Twelve proof problems worth ten points each; 120 points total." } }, { "id": "internal_research_debugging", "name": "Internal Research Debugging Evaluation", "category": "AI Self-Improvement", "metric": "% tasks passed", "num_problems": null, "source_url": "https://openai.com/index/gpt-5-6/", "canonical_setting": { "higher_is_better": true, "metric_type": "pct", "multimodal_input": false, "notes": "Named internal evaluation; task count not disclosed. The official source does not disclose a stable public task count.", "range": [ 0, 100 ], "sampling": "pass@1", "tools": "benchmark-specified", "version": "Internal Research Debugging Evaluation" } }, { "id": "kernelgen_1p", "name": "KernelGen 1P", "category": "AI Self-Improvement", "metric": "% reward", "num_problems": null, "source_url": "https://openai.com/index/gpt-5-6/", "canonical_setting": { "higher_is_better": true, "metric_type": "pct", "multimodal_input": false, "notes": "The official source does not disclose a stable public task count.", "range": [ 0, 100 ], "sampling": "pass@1", "tools": "benchmark-specified", "version": "KernelGen 1P" } }, { "id": "nanogpt_self_improvement", "name": "NanoGPT", "category": "AI Self-Improvement", "metric": "% normalized reward", "num_problems": null, "source_url": "https://openai.com/index/gpt-5-6/", "canonical_setting": { "higher_is_better": true, "metric_type": "pct", "multimodal_input": false, "notes": "The official source does not disclose a stable public task count.", "range": [ 0, 100 ], "sampling": "pass@1", "tools": "benchmark-specified", "version": "NanoGPT" } }, { "id": "posttrain_bench_lite", "name": "PostTrainBench Lite", "category": "AI Self-Improvement", "metric": "% score", "num_problems": null, "source_url": "https://openai.com/index/gpt-5-6/", "canonical_setting": { "higher_is_better": true, "metric_type": "pct", "multimodal_input": false, "notes": "Distinct Lite variant from the existing PostTrainBench row. The official source does not disclose a stable public task count.", "range": [ 0, 100 ], "sampling": "pass@1", "tools": "benchmark-specified", "version": "PostTrainBench Lite" } }, { "id": "rsi_index", "name": "RSI Index", "category": "AI Self-Improvement", "metric": "index score", "num_problems": null, "source_url": "https://openai.com/index/gpt-5-6/", "canonical_setting": { "higher_is_better": true, "metric_type": "index", "multimodal_input": false, "notes": "The official source does not disclose a stable public task count.", "range": [ 0, 100 ], "sampling": "pass@1", "tools": "benchmark-specified", "version": "RSI Index" } }, { "id": "aa_coding_agent_index_v1_1", "name": "Artificial Analysis Coding Agent Index v1.1", "category": "Agentic Coding", "metric": "index score", "num_problems": null, "source_url": "https://artificialanalysis.ai/methodology/coding-agent", "canonical_setting": { "higher_is_better": true, "metric_type": "index", "multimodal_input": false, "notes": "The official source does not disclose a stable public task count.", "range": [ 0, 100 ], "sampling": "pass@1", "tools": "benchmark-specified", "version": "Artificial Analysis Coding Agent Index v1.1" } }, { "id": "aav_capsid_packaging_prediction", "name": "AAV Capsid Packaging Prediction", "category": "Biology", "metric": "Spearman correlation", "num_problems": null, "source_url": "https://deploymentsafety.openai.com/gpt-5-6/aav-capsid-packaging-prediction", "canonical_setting": { "higher_is_better": true, "metric_type": "correlation", "multimodal_input": false, "notes": "The official source does not disclose a stable public task count.", "range": [ -1, 1 ], "sampling": "pass@1", "tools": "none", "version": "AAV Capsid Packaging Prediction" } }, { "id": "dna_tf_binding_design", "name": "DNA sequence design for transcription factor binding", "category": "Biology", "metric": "pass@1 (%)", "num_problems": null, "source_url": "https://deploymentsafety.openai.com/gpt-5-6/dna-sequence-design-for-transcription-factor-binding", "canonical_setting": { "higher_is_better": true, "metric_type": "pct", "multimodal_input": false, "notes": "The official source does not disclose a stable public task count.", "range": [ 0, 100 ], "sampling": "pass@1", "tools": "none", "version": "DNA sequence design for transcription factor binding" } }, { "id": "hard_negative_protein_binding", "name": "Hard-negative protein binding prediction", "category": "Biology", "metric": "pass@4 (%)", "num_problems": null, "source_url": "https://deploymentsafety.openai.com/gpt-5-6/hard-negative-protein-binding-prediction", "canonical_setting": { "higher_is_better": true, "metric_type": "pct", "multimodal_input": false, "notes": "The official source does not disclose a stable public task count.", "range": [ 0, 100 ], "sampling": "pass@4", "tools": "none", "version": "Hard-negative protein binding prediction" } }, { "id": "benchcad", "name": "BenchCAD", "category": "Computer Use", "metric": "% tasks passed", "num_problems": null, "source_url": "https://openai.com/index/gpt-5-6/", "canonical_setting": { "higher_is_better": true, "metric_type": "pct", "multimodal_input": true, "notes": "The official source does not disclose a stable public task count.", "range": [ 0, 100 ], "sampling": "pass@1", "tools": "none", "version": "BenchCAD" } }, { "id": "exploitgym", "name": "ExploitGym", "category": "Cyber", "metric": "% tasks passed", "num_problems": null, "source_url": "https://openai.com/index/gpt-5-6/", "canonical_setting": { "higher_is_better": true, "metric_type": "pct", "multimodal_input": false, "notes": "The official source does not disclose a stable public task count.", "range": [ 0, 100 ], "sampling": "pass@1", "tools": "benchmark-specified", "version": "ExploitGym" } }, { "id": "frontiercyber_easy", "name": "FrontierCyber Easy", "category": "Cyber", "metric": "success rate (%)", "num_problems": null, "source_url": "https://deploymentsafety.openai.com/gpt-5-6", "canonical_setting": { "higher_is_better": true, "metric_type": "pct", "multimodal_input": false, "notes": "Irregular agent harness; exact per-model denominators are stored on scores. The official source does not disclose a stable public task count.", "range": [ 0, 100 ], "sampling": "pass@1", "tools": "benchmark-specified", "version": "FrontierCyber Easy" } }, { "id": "frontiercyber_elite", "name": "FrontierCyber Elite", "category": "Cyber", "metric": "success rate (%)", "num_problems": null, "source_url": "https://deploymentsafety.openai.com/gpt-5-6", "canonical_setting": { "higher_is_better": true, "metric_type": "pct", "multimodal_input": false, "notes": "Irregular agent harness; exact per-model denominators are stored on scores. The official source does not disclose a stable public task count.", "range": [ 0, 100 ], "sampling": "pass@1", "tools": "benchmark-specified", "version": "FrontierCyber Elite" } }, { "id": "frontiercyber_hard", "name": "FrontierCyber Hard", "category": "Cyber", "metric": "success rate (%)", "num_problems": null, "source_url": "https://deploymentsafety.openai.com/gpt-5-6", "canonical_setting": { "higher_is_better": true, "metric_type": "pct", "multimodal_input": false, "notes": "Irregular agent harness; exact per-model denominators are stored on scores. The official source does not disclose a stable public task count.", "range": [ 0, 100 ], "sampling": "pass@1", "tools": "benchmark-specified", "version": "FrontierCyber Hard" } }, { "id": "frontiercyber_medium", "name": "FrontierCyber Medium", "category": "Cyber", "metric": "success rate (%)", "num_problems": null, "source_url": "https://deploymentsafety.openai.com/gpt-5-6", "canonical_setting": { "higher_is_better": true, "metric_type": "pct", "multimodal_input": false, "notes": "Irregular agent harness; exact per-model denominators are stored on scores. The official source does not disclose a stable public task count.", "range": [ 0, 100 ], "sampling": "pass@1", "tools": "benchmark-specified", "version": "FrontierCyber Medium" } }, { "id": "sec_bench_pro", "name": "SEC-Bench Pro", "category": "Cyber", "metric": "% tasks passed", "num_problems": null, "source_url": "https://openai.com/index/gpt-5-6/", "canonical_setting": { "higher_is_better": true, "metric_type": "pct", "multimodal_input": false, "notes": "The official source does not disclose a stable public task count.", "range": [ 0, 100 ], "sampling": "pass@1", "tools": "none", "version": "SEC-Bench Pro" } }, { "id": "first_person_fairness_harm_overall", "name": "First-Person Fairness Evaluation", "category": "Fairness", "metric": "harm_overall (%)", "num_problems": null, "source_url": "https://deploymentsafety.openai.com/gpt-5-6/first-person-fairness-evaluation", "canonical_setting": { "higher_is_better": false, "metric_type": "pct", "multimodal_input": false, "notes": "Expected male-versus-female biased-answer difference based on evaluation performance divided by 10; lower is better. The official source does not disclose a stable public task count.", "range": [ 0, 100 ], "sampling": "pass@1", "tools": "none", "version": "First-Person Fairness Evaluation" } }, { "id": "big_finance_bench", "name": "Big Finance Bench", "category": "Finance", "metric": "% tasks passed", "num_problems": null, "source_url": "https://openai.com/index/gpt-5-6/", "canonical_setting": { "higher_is_better": true, "metric_type": "pct", "multimodal_input": false, "notes": "Distinct from the existing Finance Bench row. The official source does not disclose a stable public task count.", "range": [ 0, 100 ], "sampling": "pass@1", "tools": "none", "version": "Big Finance Bench" } }, { "id": "healthbench_length_adjusted", "name": "HealthBench (length-adjusted)", "category": "Health", "metric": "length-adjusted score", "num_problems": 5000, "source_url": "https://huggingface.co/datasets/openai/healthbench", "canonical_setting": { "higher_is_better": true, "metric_type": "pct", "multimodal_input": false, "notes": "", "range": [ 0, 100 ], "sampling": "pass@1", "tools": "none", "version": "HealthBench (length-adjusted)" } }, { "id": "healthbench_consensus_length_adjusted", "name": "HealthBench Consensus (length-adjusted)", "category": "Health", "metric": "length-adjusted score", "num_problems": 3671, "source_url": "https://huggingface.co/datasets/openai/healthbench", "canonical_setting": { "higher_is_better": true, "metric_type": "pct", "multimodal_input": false, "notes": "", "range": [ 0, 100 ], "sampling": "pass@1", "tools": "none", "version": "HealthBench Consensus (length-adjusted)" } }, { "id": "healthbench_hard_length_adjusted", "name": "HealthBench Hard (length-adjusted)", "category": "Health", "metric": "length-adjusted score", "num_problems": 1000, "source_url": "https://huggingface.co/datasets/openai/healthbench", "canonical_setting": { "higher_is_better": true, "metric_type": "pct", "multimodal_input": false, "notes": "", "range": [ 0, 100 ], "sampling": "pass@1", "tools": "none", "version": "HealthBench Hard (length-adjusted)" } }, { "id": "healthbench_professional", "name": "HealthBench Professional", "category": "Health", "metric": "unadjusted score", "num_problems": null, "source_url": "https://deploymentsafety.openai.com/gpt-5-6/healthbench", "canonical_setting": { "higher_is_better": true, "metric_type": "pct", "multimodal_input": false, "notes": "The official source does not disclose a stable public task count.", "range": [ 0, 100 ], "sampling": "pass@1", "tools": "none", "version": "HealthBench Professional" } }, { "id": "healthbench_professional_length_adjusted", "name": "HealthBench Professional (length-adjusted)", "category": "Health", "metric": "length-adjusted score", "num_problems": null, "source_url": "https://deploymentsafety.openai.com/gpt-5-6/healthbench", "canonical_setting": { "higher_is_better": true, "metric_type": "pct", "multimodal_input": false, "notes": "The official source does not disclose a stable public task count.", "range": [ 0, 100 ], "sampling": "pass@1", "tools": "none", "version": "HealthBench Professional (length-adjusted)" } }, { "id": "management_consulting_tasks_internal", "name": "Management Consulting Tasks (Internal)", "category": "Knowledge", "metric": "% tasks passed", "num_problems": null, "source_url": "https://openai.com/index/gpt-5-6/", "canonical_setting": { "higher_is_better": true, "metric_type": "pct", "multimodal_input": false, "notes": "Named internal evaluation; task count not disclosed. The official source does not disclose a stable public task count.", "range": [ 0, 100 ], "sampling": "pass@1", "tools": "none", "version": "Management Consulting Tasks (Internal)" } }, { "id": "graphwalks_bfs_1m", "name": "GraphWalks BFS 1M", "category": "Long Context", "metric": "% F1", "num_problems": null, "source_url": "https://huggingface.co/datasets/openai/graphwalks/resolve/f338bb265735a56a79f4b0f5def722c9c3268ead/README.md", "canonical_setting": { "higher_is_better": true, "metric_type": "pct", "multimodal_input": false, "notes": "Exact-length BFS point under the immutable official GraphWalks definition after the 2026-02-27 parents ground-truth and BFS prompt fixes. The deterministic judge computes set-overlap F1. Task counts remain observation-specific; no benchmark-wide count is invented.", "range": [ 0, 100 ], "sampling": "one response per graph prompt", "tools": "none", "version": "GraphWalks post-2026-02-27 ground-truth/prompt fix; BFS exact 1M point", "judge": "deterministic set-overlap F1 against answer node list" } }, { "id": "graphwalks_bfs_256k", "name": "GraphWalks BFS 256K", "category": "Long Context", "metric": "% F1", "num_problems": null, "source_url": "https://huggingface.co/datasets/openai/graphwalks/resolve/f338bb265735a56a79f4b0f5def722c9c3268ead/README.md", "canonical_setting": { "higher_is_better": true, "metric_type": "pct", "multimodal_input": false, "notes": "Exact-length BFS point under the immutable official GraphWalks definition after the 2026-02-27 parents ground-truth and BFS prompt fixes. The deterministic judge computes set-overlap F1. Task counts remain observation-specific; no benchmark-wide count is invented.", "range": [ 0, 100 ], "sampling": "one response per graph prompt", "tools": "none", "version": "GraphWalks post-2026-02-27 ground-truth/prompt fix; BFS exact 256K point", "judge": "deterministic set-overlap F1 against answer node list" } }, { "id": "mrcr_v2_8needle_256k_512k", "name": "OpenAI MRCR v2 (8-needle, 256K-512K)", "category": "Long Context", "metric": "% correct", "num_problems": 200, "source_url": "https://huggingface.co/datasets/openai/mrcr", "canonical_setting": { "higher_is_better": true, "metric_type": "pct", "multimodal_input": false, "notes": "", "range": [ 0, 100 ], "sampling": "pass@1", "tools": "none", "version": "OpenAI MRCR v2 (8-needle, 256K-512K)" } }, { "id": "frontiermath_tier_1_3_v2", "name": "FrontierMath Tier 1-3 v2", "category": "Math", "metric": "% correct", "num_problems": null, "source_url": "https://epoch.ai/benchmarks/frontiermath", "canonical_setting": { "higher_is_better": true, "metric_type": "pct", "multimodal_input": false, "notes": "Version 2 is distinct from the existing FrontierMath row. The official source does not disclose a stable public task count.", "range": [ 0, 100 ], "sampling": "pass@1", "tools": "none", "version": "FrontierMath Tier 1-3 v2" } }, { "id": "frontiermath_tier_4_v2", "name": "FrontierMath Tier 4 v2", "category": "Math", "metric": "% correct", "num_problems": null, "source_url": "https://epoch.ai/benchmarks/frontiermath", "canonical_setting": { "higher_is_better": true, "metric_type": "pct", "multimodal_input": false, "notes": "Version 2 is distinct from the existing FrontierMath Tier 4 row. The official source does not disclose a stable public task count.", "range": [ 0, 100 ], "sampling": "pass@1", "tools": "none", "version": "FrontierMath Tier 4 v2" } }, { "id": "gdp_pdf", "name": "GDP.pdf", "category": "Multimodal", "metric": "strict task pass rate (%)", "num_problems": 100, "source_url": "https://raw.githubusercontent.com/surge-ai/gdp-pdf/7a72a514a6ab19c90babb00adc817e4ae86b9c1b/README.md", "canonical_setting": { "higher_is_better": true, "metric_type": "all_pass_pct", "multimodal_input": true, "notes": "Headline all_pass/mean: a task passes only when every rubric criterion is satisfied. The repository and HF size endpoint independently lock 100 tasks.", "range": [ 0, 100 ], "sampling": "pass@1; published leaderboard uses five epochs", "tools": "none (per-benchmark override)", "version": "GDP.pdf 100-task held-out test set", "judge": "Gemini 3.5 Flash rubric judge", "harness": "Surge GDP.pdf Inspect AI harness", "dataset_split": "test" } }, { "id": "arc_agi_3", "name": "ARC-AGI-3", "category": "Reasoning", "metric": "% tasks passed", "num_problems": null, "source_url": "https://arcprize.org/arc-agi/3/", "canonical_setting": { "higher_is_better": true, "metric_type": "pct", "multimodal_input": false, "notes": "The official source does not disclose a stable public task count.", "range": [ 0, 100 ], "sampling": "pass@1", "tools": "none", "version": "ARC-AGI-3" } }, { "id": "genebench_pro", "name": "GeneBench Pro", "category": "Science", "metric": "% tasks passed", "num_problems": null, "source_url": "https://openai.com/index/gpt-5-6/", "canonical_setting": { "higher_is_better": true, "metric_type": "pct", "multimodal_input": false, "notes": "Distinct professional variant from the existing GeneBench row. The official source does not disclose a stable public task count.", "range": [ 0, 100 ], "sampling": "pass@1", "tools": "none", "version": "GeneBench Pro" } }, { "id": "lifescibench", "name": "LifeSciBench", "category": "Science", "metric": "% tasks passed", "num_problems": null, "source_url": "https://openai.com/index/gpt-5-6/", "canonical_setting": { "higher_is_better": true, "metric_type": "pct", "multimodal_input": false, "notes": "The official source does not disclose a stable public task count.", "range": [ 0, 100 ], "sampling": "pass@1", "tools": "none", "version": "LifeSciBench" } }, { "id": "medchembench_internal", "name": "MedChemBench (Internal)", "category": "Science", "metric": "% tasks passed", "num_problems": null, "source_url": "https://openai.com/index/gpt-5-6/", "canonical_setting": { "higher_is_better": true, "metric_type": "pct", "multimodal_input": false, "notes": "Named internal evaluation; task count not disclosed. The official source does not disclose a stable public task count.", "range": [ 0, 100 ], "sampling": "pass@1", "tools": "none", "version": "MedChemBench (Internal)" } }, { "id": "long_form_virology_task_1_sequence_design", "name": "Long-form virology: Task 1 / Sequence design", "category": "Biology", "metric": "score (0-1)", "num_problems": null, "source_url": "https://www.anthropic.com/claude-sonnet-5-system-card", "canonical_setting": { "version": "Long-form virology: Task 1 / Sequence design", "metric_type": "score", "range": [ 0, 1 ], "higher_is_better": true, "multimodal_input": false, "tools": "none", "notes": "Anthropic internal automated CB-1 evaluation; exact task count is not disclosed." } }, { "id": "long_form_virology_task_1_protocol_design", "name": "Long-form virology: Task 1 / Protocol design", "category": "Biology", "metric": "score (0-1)", "num_problems": null, "source_url": "https://www.anthropic.com/claude-sonnet-5-system-card", "canonical_setting": { "version": "Long-form virology: Task 1 / Protocol design", "metric_type": "score", "range": [ 0, 1 ], "higher_is_better": true, "multimodal_input": false, "tools": "none", "notes": "Anthropic internal automated CB-1 evaluation; exact task count is not disclosed." } }, { "id": "long_form_virology_task_1_end_to_end", "name": "Long-form virology: Task 1 / End-to-end", "category": "Biology", "metric": "score (0-1)", "num_problems": null, "source_url": "https://www.anthropic.com/claude-sonnet-5-system-card", "canonical_setting": { "version": "Long-form virology: Task 1 / End-to-end", "metric_type": "score", "range": [ 0, 1 ], "higher_is_better": true, "multimodal_input": false, "tools": "none", "notes": "Anthropic internal automated CB-1 evaluation; exact task count is not disclosed." } }, { "id": "long_form_virology_task_2_sequence_design", "name": "Long-form virology: Task 2 / Sequence design", "category": "Biology", "metric": "score (0-1)", "num_problems": null, "source_url": "https://www.anthropic.com/claude-sonnet-5-system-card", "canonical_setting": { "version": "Long-form virology: Task 2 / Sequence design", "metric_type": "score", "range": [ 0, 1 ], "higher_is_better": true, "multimodal_input": false, "tools": "none", "notes": "Anthropic internal automated CB-1 evaluation; exact task count is not disclosed." } }, { "id": "long_form_virology_task_2_protocol_design", "name": "Long-form virology: Task 2 / Protocol design", "category": "Biology", "metric": "score (0-1)", "num_problems": null, "source_url": "https://www.anthropic.com/claude-sonnet-5-system-card", "canonical_setting": { "version": "Long-form virology: Task 2 / Protocol design", "metric_type": "score", "range": [ 0, 1 ], "higher_is_better": true, "multimodal_input": false, "tools": "none", "notes": "Anthropic internal automated CB-1 evaluation; exact task count is not disclosed." } }, { "id": "long_form_virology_task_2_end_to_end", "name": "Long-form virology: Task 2 / End-to-end", "category": "Biology", "metric": "score (0-1)", "num_problems": null, "source_url": "https://www.anthropic.com/claude-sonnet-5-system-card", "canonical_setting": { "version": "Long-form virology: Task 2 / End-to-end", "metric_type": "score", "range": [ 0, 1 ], "higher_is_better": true, "multimodal_input": false, "tools": "none", "notes": "Anthropic internal automated CB-1 evaluation; exact task count is not disclosed." } }, { "id": "ai_rd_kernel_best_speedup", "name": "AI R&D Kernel Best Speedup", "category": "Agentic Science", "metric": "speedup (x)", "num_problems": null, "source_url": "https://www.anthropic.com/claude-sonnet-5-system-card", "canonical_setting": { "version": "AI R&D Kernel Best Speedup", "metric_type": "ratio", "range": [ 0, null ], "higher_is_better": true, "multimodal_input": false, "tools": "none", "notes": "Anthropic internal AI R&D rule-out evaluation; inherited protocol is cited to the Mythos 5 System Card." } }, { "id": "ai_rd_time_series_forecasting_mse", "name": "AI R&D Time-Series Forecasting", "category": "Agentic Science", "metric": "MSE", "num_problems": null, "source_url": "https://www.anthropic.com/claude-sonnet-5-system-card", "canonical_setting": { "version": "AI R&D Time-Series Forecasting", "metric_type": "error", "range": [ 0, null ], "higher_is_better": false, "multimodal_input": false, "tools": "none", "notes": "Anthropic internal AI R&D rule-out evaluation; inherited protocol is cited to the Mythos 5 System Card." } }, { "id": "ai_rd_llm_training_speedup", "name": "AI R&D LLM Training Speedup", "category": "Agentic Science", "metric": "speedup (x)", "num_problems": null, "source_url": "https://www.anthropic.com/claude-sonnet-5-system-card", "canonical_setting": { "version": "AI R&D LLM Training Speedup", "metric_type": "ratio", "range": [ 0, null ], "higher_is_better": true, "multimodal_input": false, "tools": "none", "notes": "Anthropic internal AI R&D rule-out evaluation; inherited protocol is cited to the Mythos 5 System Card." } }, { "id": "ai_rd_quadruped_rl", "name": "AI R&D Quadruped RL", "category": "Agentic Science", "metric": "score", "num_problems": null, "source_url": "https://www.anthropic.com/claude-sonnet-5-system-card", "canonical_setting": { "version": "AI R&D Quadruped RL", "metric_type": "score", "range": [ null, null ], "higher_is_better": true, "multimodal_input": false, "tools": "none", "notes": "Anthropic internal AI R&D rule-out evaluation; inherited protocol is cited to the Mythos 5 System Card." } }, { "id": "ai_rd_novel_compiler", "name": "AI R&D Novel Compiler", "category": "Agentic Science", "metric": "% complex tests passed", "num_problems": null, "source_url": "https://www.anthropic.com/claude-sonnet-5-system-card", "canonical_setting": { "version": "AI R&D Novel Compiler", "metric_type": "pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": false, "tools": "none", "notes": "Anthropic internal AI R&D rule-out evaluation; inherited protocol is cited to the Mythos 5 System Card." } }, { "id": "oss_fuzz_control_flow_hijack_count", "name": "OSS-Fuzz Exploit Primitive: Control-Flow Hijack", "category": "Cyber", "metric": "targets reaching grade 1.0", "num_problems": 830, "source_url": "https://www.anthropic.com/claude-sonnet-5-system-card", "canonical_setting": { "version": "OSS-Fuzz Exploit Primitive: Control-Flow Hijack", "metric_type": "count", "range": [ 0, 830 ], "higher_is_better": true, "multimodal_input": false, "tools": "benchmark-specified", "notes": "Internal evaluation over about 830 OSS-Fuzz entry points from 228 projects." } }, { "id": "oss_fuzz_any_progress", "name": "OSS-Fuzz Exploit Primitive: Any Progress", "category": "Cyber", "metric": "% targets with grade > 0", "num_problems": 830, "source_url": "https://www.anthropic.com/claude-sonnet-5-system-card", "canonical_setting": { "version": "OSS-Fuzz Exploit Primitive: Any Progress", "metric_type": "pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": false, "tools": "benchmark-specified", "notes": "Internal evaluation over about 830 OSS-Fuzz entry points from 228 projects." } }, { "id": "firefox_147_exploit_development_working_exploit", "name": "Firefox 147 Exploit Development: Working Exploit", "category": "Cyber", "metric": "% trials with working exploit", "num_problems": 250, "source_url": "https://www.anthropic.com/claude-sonnet-5-system-card", "canonical_setting": { "version": "Firefox 147 Exploit Development: Working Exploit", "metric_type": "pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": false, "tools": "benchmark-specified", "notes": "50 crash categories x 5 trials; security mitigations disabled." } }, { "id": "firefox_147_exploit_development_any_success", "name": "Firefox 147 Exploit Development: Any Success", "category": "Cyber", "metric": "% trials with grade >= 0.5", "num_problems": 250, "source_url": "https://www.anthropic.com/claude-sonnet-5-system-card", "canonical_setting": { "version": "Firefox 147 Exploit Development: Any Success", "metric_type": "pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": false, "tools": "benchmark-specified", "notes": "50 crash categories x 5 trials; security mitigations disabled." } }, { "id": "frontiercode_main_v1", "name": "FrontierCode Main v1", "category": "Agentic Coding", "metric": "score (%)", "num_problems": 100, "source_url": "https://cognition.com/blog/frontier-code", "canonical_setting": { "version": "FrontierCode Main v1", "metric_type": "pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": false, "tools": "benchmark-specified", "notes": "The plotted values match FrontierCode Main (100 hardest tasks), not Extended or Diamond." } }, { "id": "automation_bench_private_heldout", "name": "AutomationBench Private Held-Out", "category": "Agentic", "metric": "% tasks passed", "num_problems": null, "source_url": "https://zapier.com/benchmarks", "canonical_setting": { "version": "AutomationBench Private Held-Out", "metric_type": "pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": false, "tools": "benchmark-specified", "notes": "Distinct private held-out leaderboard set; do not merge into the current public 600-task row." } }, { "id": "legal_agent_benchmark_public", "name": "Legal Agent Benchmark: Full Public Set", "category": "Legal", "metric": "all-pass rate (%)", "num_problems": 1235, "source_url": "https://www.harvey.ai/blog/introducing-harveys-legal-agent-benchmark", "canonical_setting": { "version": "Legal Agent Benchmark: Full Public Set", "metric_type": "pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": false, "tools": "benchmark-specified", "notes": "1,235 tested tasks after 16 pre-test exclusions from 1,251." } }, { "id": "legal_agent_benchmark_harvey_held_out", "name": "Legal Agent Benchmark: Harvey Held-Out", "category": "Legal", "metric": "all-pass rate (%)", "num_problems": null, "source_url": "https://www.anthropic.com/claude-sonnet-5-system-card", "canonical_setting": { "version": "Legal Agent Benchmark: Harvey Held-Out", "metric_type": "pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": false, "tools": "benchmark-specified", "notes": "Separate proprietary held-out set; task count is not disclosed." } }, { "id": "cursorbench_3_1", "name": "CursorBench 3.1", "category": "Agentic Coding", "metric": "score (%)", "num_problems": null, "source_url": "https://cursor.com/cursorbench", "canonical_setting": { "version": "CursorBench 3.1", "metric_type": "pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": false, "tools": "benchmark-specified", "notes": "The June 30 System Card predates CursorBench 3.2 (July 8); preserve it as CursorBench 3.1." } }, { "id": "arxivmath_2026_04_05", "name": "ArxivMath April-May 2026", "category": "Math", "metric": "accuracy (%)", "num_problems": 81, "source_url": "https://www.anthropic.com/claude-sonnet-5-system-card", "canonical_setting": { "version": "ArxivMath April-May 2026", "metric_type": "pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": false, "tools": "none", "notes": "Exact monthly releases: April 2026 (41) plus May 2026 (40)." } }, { "id": "gdp_pdf_mean_criteria_pass_rate", "name": "GDP.pdf Mean Criteria Pass Rate", "category": "Multimodal", "metric": "mean criteria pass rate (%)", "num_problems": 100, "source_url": "https://surgehq.ai/benchmarks/gdp-pdf", "canonical_setting": { "version": "GDP.pdf Mean Criteria Pass Rate", "metric_type": "pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": false, "tools": "none", "notes": "Distinct from the current gdp_pdf '% tasks passed' row." } }, { "id": "chartmuseum", "name": "ChartMuseum", "category": "Multimodal", "metric": "accuracy (%)", "num_problems": 1162, "source_url": "https://github.com/Liyan06/ChartMuseum", "canonical_setting": { "version": "ChartMuseum", "metric_type": "pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": true, "tools": "none", "notes": "Full 1,162-question benchmark." } }, { "id": "real_world_finance_v2_elo", "name": "Real-World Finance v2", "category": "Finance", "metric": "Elo rating", "num_problems": 294, "source_url": "https://www.anthropic.com/claude-sonnet-5-system-card", "canonical_setting": { "version": "Real-World Finance v2", "metric_type": "elo", "range": [ null, null ], "higher_is_better": true, "multimodal_input": false, "tools": "benchmark-specified", "notes": "Anthropic internal 294-task quantitative-finance evaluation." } }, { "id": "global_mmlu", "name": "Global MMLU", "category": "Multilingual", "metric": "average accuracy (%)", "num_problems": 589764, "source_url": "https://huggingface.co/datasets/CohereLabs/Global-MMLU", "canonical_setting": { "version": "Global MMLU", "metric_type": "pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": false, "tools": "none", "notes": "Full test split across 42 languages." } }, { "id": "milu", "name": "MILU", "category": "Multilingual", "metric": "average accuracy (%)", "num_problems": 79617, "source_url": "https://huggingface.co/datasets/ai4bharat/MILU", "canonical_setting": { "version": "MILU", "metric_type": "pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": false, "tools": "none", "notes": "Full 79,617-question test benchmark across 11 languages." } }, { "id": "include_base_44", "name": "INCLUDE-base-44", "category": "Multilingual", "metric": "average accuracy (%)", "num_problems": 22637, "source_url": "https://huggingface.co/datasets/CohereLabs/include-base-44", "canonical_setting": { "version": "INCLUDE-base-44", "metric_type": "pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": false, "tools": "none", "notes": "Full 44-language base benchmark." } }, { "id": "biomysterybench_human_solvable", "name": "BioMysteryBench / Human solvable", "category": "Biology", "metric": "score (0-1)", "num_problems": null, "source_url": "https://www.anthropic.com/research/Evaluating-Claude-For-Bioinformatics-With-BioMysteryBench", "canonical_setting": { "version": "BioMysteryBench / Human solvable", "metric_type": "score", "range": [ 0, 1 ], "higher_is_better": true, "multimodal_input": false, "tools": "none", "notes": "Official BioMysteryBench subset; subset count is not disclosed in the System Card." } }, { "id": "biomysterybench_human_difficult", "name": "BioMysteryBench / Human difficult", "category": "Biology", "metric": "score (0-1)", "num_problems": null, "source_url": "https://www.anthropic.com/research/Evaluating-Claude-For-Bioinformatics-With-BioMysteryBench", "canonical_setting": { "version": "BioMysteryBench / Human difficult", "metric_type": "score", "range": [ 0, 1 ], "higher_is_better": true, "multimodal_input": false, "tools": "none", "notes": "Official BioMysteryBench subset; subset count is not disclosed in the System Card." } }, { "id": "spatialbench_verified", "name": "SpatialBench Verified", "category": "Biology", "metric": "score (0-1)", "num_problems": 115, "source_url": "https://www.anthropic.com/claude-sonnet-5-system-card", "canonical_setting": { "version": "SpatialBench Verified", "metric_type": "score", "range": [ 0, 1 ], "higher_is_better": true, "multimodal_input": false, "tools": "none", "notes": "LatchBio externally validated spatial-transcriptomics benchmark." } }, { "id": "singlecellbench", "name": "SingleCellBench", "category": "Biology", "metric": "score (0-1)", "num_problems": 195, "source_url": "https://www.anthropic.com/claude-sonnet-5-system-card", "canonical_setting": { "version": "SingleCellBench", "metric_type": "score", "range": [ 0, 1 ], "higher_is_better": true, "multimodal_input": false, "tools": "none", "notes": "LatchBio single-cell RNA sequencing benchmark." } }, { "id": "structural_biology_open_ended_internal", "name": "Structural biology open-ended", "category": "Biology", "metric": "score (0-1)", "num_problems": null, "source_url": "https://www.anthropic.com/claude-sonnet-5-system-card", "canonical_setting": { "version": "Structural biology open-ended", "metric_type": "score", "range": [ 0, 1 ], "higher_is_better": true, "multimodal_input": false, "tools": "none", "notes": "Anthropic internal open-ended structural biology evaluation." } }, { "id": "proteingym_hard", "name": "ProteinGym Hard", "category": "Biology", "metric": "score (0-1)", "num_problems": null, "source_url": "https://proteingym.org/", "canonical_setting": { "version": "ProteinGym Hard", "metric_type": "score", "range": [ 0, 1 ], "higher_is_better": true, "multimodal_input": false, "tools": "none", "notes": "Exact hard split count is not disclosed in the System Card." } }, { "id": "organic_chemistry_internal", "name": "Organic chemistry", "category": "Science", "metric": "score (0-1)", "num_problems": null, "source_url": "https://www.anthropic.com/claude-sonnet-5-system-card", "canonical_setting": { "version": "Organic chemistry", "metric_type": "score", "range": [ 0, 1 ], "higher_is_better": true, "multimodal_input": false, "tools": "none", "notes": "Anthropic internal organic chemistry evaluation." } }, { "id": "protocol_troubleshooting_internal", "name": "Protocol troubleshooting", "category": "Biology", "metric": "score (0-1)", "num_problems": null, "source_url": "https://www.anthropic.com/claude-sonnet-5-system-card", "canonical_setting": { "version": "Protocol troubleshooting", "metric_type": "score", "range": [ 0, 1 ], "higher_is_better": true, "multimodal_input": false, "tools": "none", "notes": "Anthropic internal molecular-biology protocol troubleshooting evaluation." } }, { "id": "ai_rd_llm_training_hard_speedup", "name": "AI R&D LLM Training Hard Speedup", "category": "Agentic Science", "metric": "speedup (x)", "num_problems": null, "source_url": "https://www.anthropic.com/claude-opus-5-system-card", "canonical_setting": { "version": "AI R&D LLM Training hard variant", "metric_type": "ratio", "range": [ 0, null ], "higher_is_better": true, "multimodal_input": false, "tools": "benchmark-specified", "notes": "Separate hard variant from the existing easy LLM-training speedup row." } }, { "id": "arxivmath_2026_06_no_tools", "name": "ArxivMath June 2026 (No Tools)", "category": "Math", "metric": "accuracy (%)", "num_problems": 49, "source_url": "https://matharena.ai/", "canonical_setting": { "version": "ArxivMath June 2026", "metric_type": "pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": false, "tools": "none", "notes": "Distinct June 2026 release; four runs per problem for Anthropic internal values." } }, { "id": "arxivmath_2026_06_with_tools", "name": "ArxivMath June 2026 (With Tools)", "category": "Math", "metric": "accuracy (%)", "num_problems": 49, "source_url": "https://matharena.ai/", "canonical_setting": { "version": "ArxivMath June 2026", "metric_type": "pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": false, "tools": "benchmark-specified", "notes": "Distinct June 2026 release; four runs per problem." } }, { "id": "biomysterybench_human_difficult_revised_2026_07", "name": "BioMysteryBench Human Difficult (Revised July 2026)", "category": "Biology", "metric": "score (0-1)", "num_problems": null, "source_url": "https://www.anthropic.com/claude-opus-5-system-card", "canonical_setting": { "version": "BioMysteryBench revised July 2026", "metric_type": "score", "range": [ 0, 1 ], "higher_is_better": true, "multimodal_input": false, "tools": "benchmark-specified", "notes": "Distinct revised subset after removal of 6 Human Difficult problems." } }, { "id": "biomysterybench_human_solvable_revised_2026_07", "name": "BioMysteryBench Human Solvable (Revised July 2026)", "category": "Biology", "metric": "score (0-1)", "num_problems": null, "source_url": "https://www.anthropic.com/claude-opus-5-system-card", "canonical_setting": { "version": "BioMysteryBench revised July 2026", "metric_type": "score", "range": [ 0, 1 ], "higher_is_better": true, "multimodal_input": false, "tools": "benchmark-specified", "notes": "Distinct revised subset after removal of 3 Human Solvable problems." } }, { "id": "chartography_no_tools", "name": "Chartography (No Tools)", "category": "Multimodal", "metric": "score (%)", "num_problems": 100, "source_url": "https://www.surgehq.ai/blog/chartography", "canonical_setting": { "version": "Chartography", "metric_type": "pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": true, "tools": "none", "notes": "100 specialized chart types; five runs." } }, { "id": "chartography_with_tools", "name": "Chartography (With Tools)", "category": "Multimodal", "metric": "score (%)", "num_problems": 100, "source_url": "https://www.surgehq.ai/blog/chartography", "canonical_setting": { "version": "Chartography", "metric_type": "pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": true, "tools": "benchmark-specified", "notes": "100 specialized chart types; five runs." } }, { "id": "cursorbench_3_2", "name": "CursorBench 3.2", "category": "Agentic Coding", "metric": "score (%)", "num_problems": null, "source_url": "https://cursor.com/cursorbench", "canonical_setting": { "version": "CursorBench 3.2", "metric_type": "pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": false, "tools": "Cursor coding agent", "notes": "Ambiguous multi-file tasks from real Cursor sessions; task count is not disclosed." } }, { "id": "draco", "name": "DRACO", "category": "Agentic Data Analysis", "metric": "normalized score (%)", "num_problems": 100, "source_url": "https://www.anthropic.com/claude-opus-5-system-card", "canonical_setting": { "version": "DRACO", "metric_type": "pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": false, "tools": "web search, web fetch, programmatic calls, code execution", "notes": "100 curated tasks; four grading categories; five independent grading runs." } }, { "id": "frontierbench_v0_1", "name": "FrontierBench v0.1", "category": "Agentic", "metric": "% tasks completed", "num_problems": 74, "source_url": "https://github.com/harbor-framework/frontier-bench", "canonical_setting": { "version": "FrontierBench v0.1", "metric_type": "pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": false, "tools": "terminal agent", "notes": "74 professional terminal tasks. Harness differences are retained in reported_setting." } }, { "id": "frontiercode_extended_v1_1", "name": "FrontierCode v1.1 Extended", "category": "Agentic Coding", "metric": "score (%)", "num_problems": 150, "source_url": "https://cognition.com/blog/frontier-code-1.1", "canonical_setting": { "version": "FrontierCode v1.1 Extended", "metric_type": "pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": false, "tools": "coding agent", "notes": "Private full 150-task set; five runs per available effort." } }, { "id": "frontiercode_main_v1_1", "name": "FrontierCode v1.1 Main", "category": "Agentic Coding", "metric": "score (%)", "num_problems": 100, "source_url": "https://cognition.com/blog/frontier-code-1.1", "canonical_setting": { "version": "FrontierCode v1.1 Main", "metric_type": "pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": false, "tools": "coding agent", "notes": "Private 100-task Main subset; five runs per available effort." } }, { "id": "gmmlu", "name": "GMMLU", "category": "Multilingual", "metric": "average accuracy (%)", "num_problems": null, "source_url": "https://www.anthropic.com/claude-opus-5-system-card", "canonical_setting": { "version": "GMMLU", "metric_type": "pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": false, "tools": "none", "notes": "42-language evaluation; the System Card does not disclose a stable task count." } }, { "id": "organic_chemistry_v2_internal", "name": "Organic Chemistry V2", "category": "Science", "metric": "score (0-1)", "num_problems": null, "source_url": "https://www.anthropic.com/claude-opus-5-system-card", "canonical_setting": { "version": "Anthropic internal Organic Chemistry V2", "metric_type": "score", "range": [ 0, 1 ], "higher_is_better": true, "multimodal_input": false, "tools": "benchmark-specified", "notes": "Versioned separately from the earlier internal Organic Chemistry evaluation." } }, { "id": "protein_design_internal", "name": "Protein Design", "category": "Biology", "metric": "score (0-1)", "num_problems": null, "source_url": "https://www.anthropic.com/claude-opus-5-system-card", "canonical_setting": { "version": "Anthropic internal Protein Design", "metric_type": "score", "range": [ 0, 1 ], "higher_is_better": true, "multimodal_input": false, "tools": "none", "notes": "Combined constraint-satisfaction, folding-confidence, and novelty score." } }, { "id": "protocol_understanding_internal", "name": "Protocol Understanding (Benchling)", "category": "Biology", "metric": "score (0-1)", "num_problems": null, "source_url": "https://www.anthropic.com/claude-opus-5-system-card", "canonical_setting": { "version": "Protocol Understanding (Benchling)", "metric_type": "score", "range": [ 0, 1 ], "higher_is_better": true, "multimodal_input": false, "tools": "bash, file editor, web search", "notes": "Distinct from Protocol Troubleshooting." } }, { "id": "riemannbench_no_tools", "name": "RiemannBench (No Tools)", "category": "Math", "metric": "score (%)", "num_problems": 25, "source_url": "https://arxiv.org/abs/2604.06802", "canonical_setting": { "version": "RiemannBench corrected references/grading", "metric_type": "pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": false, "tools": "none", "notes": "Private 25-problem research-mathematics benchmark; mean over four attempts." } }, { "id": "riemannbench_with_tools", "name": "RiemannBench (With Tools)", "category": "Math", "metric": "score (%)", "num_problems": 25, "source_url": "https://arxiv.org/abs/2604.06802", "canonical_setting": { "version": "RiemannBench corrected references/grading", "metric_type": "pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": false, "tools": "benchmark-specified", "notes": "Private 25-problem research-mathematics benchmark; mean over four attempts." } }, { "id": "cursorbench_3_0", "name": "CursorBench 3.0", "category": "Agentic Coding", "metric": "score (%)", "num_problems": null, "source_url": "https://cursor.com/resources/Composer2.pdf", "canonical_setting": { "version": "CursorBench 3.0", "metric_type": "pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": false, "tools": "Cursor coding agent", "notes": "The Composer 2 report calls this CursorBench-3. Its 2026-03-25 evaluation predates CursorBench 3.1 (2026-05-19) and follows the 3.0 launch (2026-03-11), so the exact version identity is CursorBench 3.0. Task count is undisclosed." } }, { "id": "mrcr_v2_8needle_512k_1m", "name": "MRCR v2 8-Needle 512K-1M", "category": "Long Context", "metric": "mean SequenceMatcher ratio (%)", "num_problems": 100, "source_url": "https://huggingface.co/datasets/openai/mrcr", "canonical_setting": { "version": "OpenAI MRCR; 8 needles; (524,288, 1,048,576] token bin", "metric_type": "pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": false, "tools": "none", "notes": "Official dataset has 100 samples per context-length bin. Scoring uses the mean difflib SequenceMatcher ratio with the required hash prefix. Distinct from the all-bin 800-task aggregate." } }, { "id": "mcpatlas_full_1000", "name": "MCP-Atlas Full 1,000", "category": "Agentic", "metric": "pass rate at >=0.75 claim coverage (%)", "num_problems": 1000, "source_url": "https://arxiv.org/abs/2602.00933v3", "canonical_setting": { "version": "MCP-Atlas full 1,000-task evaluation", "metric_type": "pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": false, "tools": "controlled target and distractor tools across 36 real MCP servers and 220 tools", "sampling": "pass@1 over all 1,000 tasks", "judge": "claim-level 1/0.5/0 scoring; Gemini 3.1 Pro Preview primary judge with GPT-5.4 and Claude Opus 4.6 sensitivity judges", "harness": "Scale AI MCP-Atlas agent harness and containerized scoring pipeline", "notes": "Full set is 500 public plus 500 held-out private tasks. A task passes when mean claim coverage is at least 0.75." } }, { "id": "osworld_2_0_binary", "name": "OSWorld 2.0 Binary", "category": "Agentic", "metric": "binary success (%)", "num_problems": 108, "source_url": "https://osworld-v2.xlang.ai/", "canonical_setting": { "version": "OSWorld 2.0 binary scoring", "metric_type": "pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": false, "tools": "computer-use environment", "notes": "Binary success metric printed separately from OSWorld 2.0 partial-credit score over the 108-task release." } }, { "id": "webarena_verified_full", "name": "WebArena-Verified Full", "category": "Agentic", "metric": "task success (%)", "num_problems": 812, "source_url": "https://github.com/ServiceNow/webarena-verified", "canonical_setting": { "version": "WebArena-Verified full 812-task set", "metric_type": "pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": false, "tools": "browser", "notes": "Full verified set, distinct from the 258-task Hard subset and from original WebArena." } }, { "id": "gdpval_aa_v2_elo", "name": "GDPval-AA v2 Elo", "category": "Professional", "metric": "Elo", "num_problems": 220, "source_url": "https://artificialanalysis.ai/methodology/intelligence-benchmarking#gdpval-aa", "canonical_setting": { "version": "GDPval-AA v2", "metric_type": "elo", "range": null, "higher_is_better": true, "multimodal_input": false, "tools": "Web Fetch, Web Search, View Image, Code Exec, Finish, Abandon Task", "sampling": "one agentic submission per task", "judge": "blind pairwise panel of three frontier LLM judges; Bradley-Terry Elo", "harness": "Artificial Analysis Stirrup in a fresh E2B sandbox; 250-turn limit", "notes": "All 220 public OpenAI GDPval gold tasks across 44 occupations. The v2 Elo scale is anchored to human-expert deliverables at 1000." } }, { "id": "meta_internal_coding_bench", "name": "Meta Internal Coding Bench", "category": "Agentic Coding", "metric": "% resolved (pass@1)", "num_problems": 440, "source_url": "https://research.meta.ai/static/muse-spark-1-2-methodology", "canonical_setting": { "version": "Meta Internal Coding Bench, August 2026", "metric_type": "pass_at_1_pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": false, "tools": "internal agentic coding environment", "sampling": "two attempts per task; average task-level success rate", "judge": "compile and unit-test verifier in dedicated grading containers", "harness": "Meta internal agentic harness; internet disabled", "notes": "Private 440-task benchmark derived from real internal pull requests." } }, { "id": "vibecodebench_v1_1_test", "name": "Vibe Code Bench v1.1 Test", "category": "Agentic Coding", "metric": "mean per-application accuracy (%)", "num_problems": 50, "source_url": "https://www.vals.ai/benchmarks/vibe-code", "canonical_setting": { "version": "Vibe Code Bench v1.1 held-out test set", "metric_type": "pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": false, "tools": "agentic coding environment", "notes": "Official benchmark has 50 validation and 50 held-out test tasks; leaderboard scores use the 50-task held-out test set. A workflow passes when at least 90% of its substeps succeed." } }, { "id": "swe_atlas_codebase_qna", "name": "SWE-Atlas-QnA", "category": "Agentic Coding", "metric": "task resolve rate (%)", "num_problems": 124, "source_url": "https://huggingface.co/datasets/SWE-Atlas/SWE-Atlas-QnA", "canonical_setting": { "version": "SWE-Atlas-QnA public 124-task split", "metric_type": "pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": false, "tools": "codebase reading and exploration", "sampling": "unknown", "judge": "repository-grounded task-resolution evaluator", "harness": "SWE-Atlas-QnA official dataset and score-level agent harness", "notes": "Repository question answering, not patch generation; 124 tasks across 11 repositories. Agent harness remains score-level provenance." } }, { "id": "wmdp_bio_accuracy", "name": "WMDP-Bio", "category": "Safety Capability", "metric": "accuracy (%)", "num_problems": 1273, "source_url": "https://wmdp.ai", "canonical_setting": { "version": "WMDP-Bio current public release", "metric_type": "pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": false, "tools": "none; no safeguards", "sampling": "pass@1", "judge": "multiple-choice exact accuracy", "harness": "WMDP official evaluation", "notes": "Public Bio split has 1,273 questions; xAI capability rows are evaluated without safeguards." } }, { "id": "wmdp_chem_accuracy", "name": "WMDP-Chem", "category": "Safety Capability", "metric": "accuracy (%)", "num_problems": 408, "source_url": "https://wmdp.ai", "canonical_setting": { "version": "WMDP-Chem current public release", "metric_type": "pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": false, "tools": "none; no safeguards", "sampling": "pass@1", "judge": "multiple-choice exact accuracy", "harness": "WMDP official evaluation", "notes": "Public Chemistry split has 408 questions; xAI capability rows are evaluated without safeguards." } }, { "id": "lab_bench_protocolqa_accuracy", "name": "LAB-Bench ProtocolQA", "category": "Safety Capability", "metric": "accuracy (%)", "num_problems": 108, "source_url": "https://github.com/Future-House/LAB-Bench", "canonical_setting": { "version": "LAB-Bench ProtocolQA official 108-question MCQ split", "metric_type": "pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": false, "tools": "none", "sampling": "pass@1", "judge": "official multiple-choice exact-match accuracy", "harness": "LAB-Bench official ProtocolQA harness", "notes": "The xAI cards report a distinct open-ended ProtocolQA adaptation. Those scores remain noncanonical score-level variants rather than redefining this benchmark id." } }, { "id": "ipho_2025_theory", "name": "IPhO 2025 Theory", "category": "Science", "metric": "normalized theory marks (%)", "num_problems": 3, "source_url": "https://www.ipho2025.fr/official-questions-ipho-france-2025", "canonical_setting": { "version": "IPhO 2025 full theory examination: 3 used problems", "metric_type": "pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": true, "tools": "none", "judge": "official partial-credit marking scheme", "sampling": "pass@1", "notes": "The organizer publishes three used theory problems. Google's scores average eight runs and use Gemini as judge, so those observations are retained as noncanonical settings." } }, { "id": "icho_2025_theory", "name": "IChO 2025 Theory", "category": "Science", "metric": "normalized theory marks (%)", "num_problems": 9, "source_url": "https://www.icho-official.org/results/results.php?id=57&year=2025", "canonical_setting": { "version": "IChO 2025 full theory examination: 9 problems", "metric_type": "pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": true, "tools": "none", "judge": "official partial-credit marking scheme", "sampling": "pass@1", "notes": "Official theory marks are normalized to 100. Google's scores average eight runs and use Gemini as judge, so those observations are retained as noncanonical settings." } }, { "id": "blueprint_bench_2", "name": "Blueprint-Bench 2", "category": "Spatial Reasoning", "metric": "normalized connectivity score (%)", "num_problems": 50, "source_url": "https://andonlabs.com/evals/blueprint-bench-2", "canonical_setting": { "version": "Blueprint-Bench 2: 50-apartment sequential evaluation", "metric_type": "pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": true, "tools": "persistent cross-apartment notepad", "judge": "D4-invariant deterministic connectivity composite", "sampling": "leaderboard protocol", "notes": "Composite weights: Jaccard 50%, degree 20%, density 10%, room count 10%, door count 5%, orientation 5%. Full v2 dataset/evaluator is not public, so reported cells are stored with matches_canonical=false." } }, { "id": "gdm_mrcr_v2_8needle_upto_128k", "name": "GDM MRCR v2 8-Needle up to 128K", "category": "Long Context", "metric": "mean strict MRCR score (%)", "num_problems": 484, "source_url": "https://github.com/google-deepmind/eval_hub/tree/master/eval_hub/mrcr_v2", "canonical_setting": { "version": "mrcr_v2p1 8-needle upto_128K cumulative CSV", "metric_type": "pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": false, "tools": "none", "judge": "official hash check + SequenceMatcher strict scorer", "sampling": "pass@1", "notes": "Google DeepMind MRCR v2, not OpenAI MRCR v2. Count 484 data rows in the official fixed CSV SHA 706a254439905ed6286d1184a03a018667e0ea0f49ad3312aedb8c27054ab0bd." } }, { "id": "gdm_mrcr_v2_8needle_1m", "name": "GDM MRCR v2 8-Needle at 1M", "category": "Long Context", "metric": "mean strict MRCR score (%)", "num_problems": null, "source_url": "https://github.com/google-deepmind/eval_hub/tree/master/eval_hub/mrcr_v2", "canonical_setting": { "version": "mrcr_v2p1 8-needle (524288,1048576] pointwise CSV", "metric_type": "pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": false, "tools": "none", "judge": "official hash check + SequenceMatcher strict scorer", "sampling": "pass@1", "notes": "Google DeepMind MRCR v2, not OpenAI MRCR v2. Official 1.5GB fixed object and scorer are public; the README does not publish the data-row count, so num_problems remains null." } }, { "id": "mle_bench_partial30_avg_position_k2", "name": "MLE-Bench Partial 30 Average Position (k=2)", "category": "Agentic Coding", "metric": "Average Position Score (%)", "num_problems": 60, "source_url": "https://github.com/openai/mle-bench", "canonical_setting": { "version": "official systemcard Partial 30 split; k=2 runs", "metric_type": "pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": false, "tools": "interactive Bash, internet, isolated H100 sandbox", "judge": "Kaggle private-leaderboard rank transformed by (N-r+1)/N", "sampling": "2 independent runs per competition", "notes": "30 fixed competitions from experiments/splits/systemcard.txt and two independent runs each: 60 model episodes. Failed/missing submission scores zero; metric is not Any-Medal accuracy." } }, { "id": "infographicvqa", "name": "InfographicVQA", "category": "Vision", "metric": "ANLS (%)", "num_problems": 3288, "source_url": "https://arxiv.org/abs/2104.12756", "canonical_setting": { "version": "InfographicVQA test split", "metric_type": "anls", "higher_is_better": true, "range": [ 0, 100 ], "multimodal_input": true, "tools": "none", "notes": "Official test split; ANLS uses normalized Levenshtein similarity." } }, { "id": "covost2_xx_en_7lang_macro", "name": "CoVoST2 XX-to-English 7-Language Macro", "category": "Audio", "metric": "macro CorpusBLEU", "num_problems": 62325, "source_url": "https://huggingface.co/datasets/facebook/covost2", "canonical_setting": { "version": "CoVoST2 ja/de/fr/es/it/ru/zh-CN to English macro", "metric_type": "bleu", "higher_is_better": true, "range": [ 0, 100 ], "multimodal_input": true, "tools": "none", "notes": "Simple macro over seven direction-level CorpusBLEU scores; Gemma uses a transcribe-then-translate prompt." } }, { "id": "loft_text_retrieval_128k", "name": "LOFT Text Retrieval at 128K", "category": "Long Context", "metric": "Recall@k (%)", "num_problems": null, "source_url": "https://github.com/google-deepmind/loft", "canonical_setting": { "version": "LOFT text-retrieval aggregate at 128K", "metric_type": "pct", "higher_is_better": true, "range": [ 0, 100 ], "multimodal_input": false, "tools": "none", "notes": "Gemma reports an aggregate across LOFT text-retrieval datasets; exact included datasets and weighting are not published." } }, { "id": "graphwalks_lt128k_combined", "name": "GraphWalks Combined below 128K", "category": "Long Context", "metric": "F1 (%)", "num_problems": 650, "source_url": "https://huggingface.co/datasets/openai/graphwalks", "canonical_setting": { "version": "GraphWalks BFS plus parent-node prompts below 128K", "metric_type": "f1", "higher_is_better": true, "range": [ 0, 100 ], "multimodal_input": false, "tools": "none", "notes": "Combined score over 300 BFS and 350 parent-node rows; the report does not publish the aggregation weighting." } }, { "id": "covost2_ja_en", "name": "CoVoST2 ja to English", "category": "Audio", "metric": "CorpusBLEU", "num_problems": 684, "source_url": "https://huggingface.co/datasets/facebook/covost2", "canonical_setting": { "version": "CoVoST2 ja to English test direction", "metric_type": "bleu", "higher_is_better": true, "range": [ 0, 100 ], "multimodal_input": true, "tools": "none", "notes": "Speech-to-English translation; Gemma uses a transcribe-then-translate prompt." } }, { "id": "covost2_de_en", "name": "CoVoST2 de to English", "category": "Audio", "metric": "CorpusBLEU", "num_problems": 13511, "source_url": "https://huggingface.co/datasets/facebook/covost2", "canonical_setting": { "version": "CoVoST2 de to English test direction", "metric_type": "bleu", "higher_is_better": true, "range": [ 0, 100 ], "multimodal_input": true, "tools": "none", "notes": "Speech-to-English translation; Gemma uses a transcribe-then-translate prompt." } }, { "id": "covost2_fr_en", "name": "CoVoST2 fr to English", "category": "Audio", "metric": "CorpusBLEU", "num_problems": 14760, "source_url": "https://huggingface.co/datasets/facebook/covost2", "canonical_setting": { "version": "CoVoST2 fr to English test direction", "metric_type": "bleu", "higher_is_better": true, "range": [ 0, 100 ], "multimodal_input": true, "tools": "none", "notes": "Speech-to-English translation; Gemma uses a transcribe-then-translate prompt." } }, { "id": "covost2_es_en", "name": "CoVoST2 es to English", "category": "Audio", "metric": "CorpusBLEU", "num_problems": 13221, "source_url": "https://huggingface.co/datasets/facebook/covost2", "canonical_setting": { "version": "CoVoST2 es to English test direction", "metric_type": "bleu", "higher_is_better": true, "range": [ 0, 100 ], "multimodal_input": true, "tools": "none", "notes": "Speech-to-English translation; Gemma uses a transcribe-then-translate prompt." } }, { "id": "covost2_it_en", "name": "CoVoST2 it to English", "category": "Audio", "metric": "CorpusBLEU", "num_problems": 8951, "source_url": "https://huggingface.co/datasets/facebook/covost2", "canonical_setting": { "version": "CoVoST2 it to English test direction", "metric_type": "bleu", "higher_is_better": true, "range": [ 0, 100 ], "multimodal_input": true, "tools": "none", "notes": "Speech-to-English translation; Gemma uses a transcribe-then-translate prompt." } }, { "id": "covost2_ru_en", "name": "CoVoST2 ru to English", "category": "Audio", "metric": "CorpusBLEU", "num_problems": 6300, "source_url": "https://huggingface.co/datasets/facebook/covost2", "canonical_setting": { "version": "CoVoST2 ru to English test direction", "metric_type": "bleu", "higher_is_better": true, "range": [ 0, 100 ], "multimodal_input": true, "tools": "none", "notes": "Speech-to-English translation; Gemma uses a transcribe-then-translate prompt." } }, { "id": "covost2_zh_cn_en", "name": "CoVoST2 zh-CN to English", "category": "Audio", "metric": "CorpusBLEU", "num_problems": 4898, "source_url": "https://huggingface.co/datasets/facebook/covost2", "canonical_setting": { "version": "CoVoST2 zh-CN to English test direction", "metric_type": "bleu", "higher_is_better": true, "range": [ 0, 100 ], "multimodal_input": true, "tools": "none", "notes": "Speech-to-English translation; Gemma uses a transcribe-then-translate prompt." } }, { "id": "fleurs_asr_en_us", "name": "FLEURS ASR en", "category": "Audio", "metric": "WER", "num_problems": 647, "source_url": "https://huggingface.co/datasets/google/fleurs", "canonical_setting": { "version": "FLEURS en test split", "metric_type": "wer", "higher_is_better": false, "range": null, "multimodal_input": true, "tools": "none", "notes": "Automatic speech recognition transcription; lower is better." } }, { "id": "fleurs_asr_ko_kr_cer", "name": "FLEURS ASR ko", "category": "Audio", "metric": "CER", "num_problems": 382, "source_url": "https://huggingface.co/datasets/google/fleurs", "canonical_setting": { "version": "FLEURS ko test split", "metric_type": "cer", "higher_is_better": false, "range": null, "multimodal_input": true, "tools": "none", "notes": "Automatic speech recognition transcription; lower is better." } }, { "id": "fleurs_asr_ja_jp_cer", "name": "FLEURS ASR ja", "category": "Audio", "metric": "CER", "num_problems": 650, "source_url": "https://huggingface.co/datasets/google/fleurs", "canonical_setting": { "version": "FLEURS ja test split", "metric_type": "cer", "higher_is_better": false, "range": null, "multimodal_input": true, "tools": "none", "notes": "Automatic speech recognition transcription; lower is better." } }, { "id": "fleurs_asr_de_de", "name": "FLEURS ASR de", "category": "Audio", "metric": "WER", "num_problems": 862, "source_url": "https://huggingface.co/datasets/google/fleurs", "canonical_setting": { "version": "FLEURS de test split", "metric_type": "wer", "higher_is_better": false, "range": null, "multimodal_input": true, "tools": "none", "notes": "Automatic speech recognition transcription; lower is better." } }, { "id": "fleurs_asr_fr_fr", "name": "FLEURS ASR fr", "category": "Audio", "metric": "WER", "num_problems": 676, "source_url": "https://huggingface.co/datasets/google/fleurs", "canonical_setting": { "version": "FLEURS fr test split", "metric_type": "wer", "higher_is_better": false, "range": null, "multimodal_input": true, "tools": "none", "notes": "Automatic speech recognition transcription; lower is better." } }, { "id": "fleurs_asr_hi_in", "name": "FLEURS ASR hi", "category": "Audio", "metric": "WER", "num_problems": 418, "source_url": "https://huggingface.co/datasets/google/fleurs", "canonical_setting": { "version": "FLEURS hi test split", "metric_type": "wer", "higher_is_better": false, "range": null, "multimodal_input": true, "tools": "none", "notes": "Automatic speech recognition transcription; lower is better." } }, { "id": "fleurs_asr_es_419", "name": "FLEURS ASR es", "category": "Audio", "metric": "WER", "num_problems": 908, "source_url": "https://huggingface.co/datasets/google/fleurs", "canonical_setting": { "version": "FLEURS es test split", "metric_type": "wer", "higher_is_better": false, "range": null, "multimodal_input": true, "tools": "none", "notes": "Automatic speech recognition transcription; lower is better." } }, { "id": "fleurs_asr_it_it", "name": "FLEURS ASR it", "category": "Audio", "metric": "WER", "num_problems": 865, "source_url": "https://huggingface.co/datasets/google/fleurs", "canonical_setting": { "version": "FLEURS it test split", "metric_type": "wer", "higher_is_better": false, "range": null, "multimodal_input": true, "tools": "none", "notes": "Automatic speech recognition transcription; lower is better." } }, { "id": "fleurs_asr_pt_br", "name": "FLEURS ASR pt-br", "category": "Audio", "metric": "WER", "num_problems": 919, "source_url": "https://huggingface.co/datasets/google/fleurs", "canonical_setting": { "version": "FLEURS pt-br test split", "metric_type": "wer", "higher_is_better": false, "range": null, "multimodal_input": true, "tools": "none", "notes": "Automatic speech recognition transcription; lower is better." } }, { "id": "fleurs_asr_ru_ru", "name": "FLEURS ASR ru", "category": "Audio", "metric": "WER", "num_problems": 775, "source_url": "https://huggingface.co/datasets/google/fleurs", "canonical_setting": { "version": "FLEURS ru test split", "metric_type": "wer", "higher_is_better": false, "range": null, "multimodal_input": true, "tools": "none", "notes": "Automatic speech recognition transcription; lower is better." } }, { "id": "fleurs_asr_ar_eg", "name": "FLEURS ASR ar", "category": "Audio", "metric": "WER", "num_problems": 428, "source_url": "https://huggingface.co/datasets/google/fleurs", "canonical_setting": { "version": "FLEURS ar test split", "metric_type": "wer", "higher_is_better": false, "range": null, "multimodal_input": true, "tools": "none", "notes": "Automatic speech recognition transcription; lower is better." } }, { "id": "fleurs_asr_zh_cn_cer", "name": "FLEURS ASR zh", "category": "Audio", "metric": "CER", "num_problems": 945, "source_url": "https://huggingface.co/datasets/google/fleurs", "canonical_setting": { "version": "FLEURS zh test split", "metric_type": "cer", "higher_is_better": false, "range": null, "multimodal_input": true, "tools": "none", "notes": "Automatic speech recognition transcription; lower is better." } }, { "id": "mtob_eng_kgv_half_book", "name": "MTOB English to Kalamang: Half Book", "category": "Long Context", "metric": "chrF", "num_problems": 100, "source_url": "https://github.com/lukemelas/mtob", "canonical_setting": { "version": "MTOB eng-to-kgv with approximately 128K half-book context", "metric_type": "chrf", "higher_is_better": true, "range": null, "multimodal_input": false, "tools": "none", "notes": "One hundred held-out translation sentence pairs." } }, { "id": "mtob_eng_kgv_full_book", "name": "MTOB English to Kalamang: Full Book", "category": "Long Context", "metric": "chrF", "num_problems": 100, "source_url": "https://github.com/lukemelas/mtob", "canonical_setting": { "version": "MTOB eng-to-kgv with approximately 256K full-book context", "metric_type": "chrf", "higher_is_better": true, "range": null, "multimodal_input": false, "tools": "none", "notes": "One hundred held-out translation sentence pairs." } }, { "id": "mtob_kgv_eng_half_book", "name": "MTOB Kalamang to English: Half Book", "category": "Long Context", "metric": "chrF", "num_problems": 100, "source_url": "https://github.com/lukemelas/mtob", "canonical_setting": { "version": "MTOB kgv-to-eng with approximately 128K half-book context", "metric_type": "chrf", "higher_is_better": true, "range": null, "multimodal_input": false, "tools": "none", "notes": "One hundred held-out translation sentence pairs." } }, { "id": "mtob_kgv_eng_full_book", "name": "MTOB Kalamang to English: Full Book", "category": "Long Context", "metric": "chrF", "num_problems": 100, "source_url": "https://github.com/lukemelas/mtob", "canonical_setting": { "version": "MTOB kgv-to-eng with approximately 256K full-book context", "metric_type": "chrf", "higher_is_better": true, "range": null, "multimodal_input": false, "tools": "none", "notes": "One hundred held-out translation sentence pairs." } }, { "id": "codeelo_rating", "name": "CodeElo", "category": "Coding", "metric": "Elo rating", "num_problems": 408, "source_url": "https://github.com/QwenLM/CodeElo", "canonical_setting": { "version": "CodeElo 408-problem Codeforces evaluation", "metric_type": "elo", "range": null, "higher_is_better": true, "multimodal_input": false, "tools": "none", "judge": "Codeforces online judge", "sampling": "samples=8 (3264 submissions)", "notes": "Eight samples per problem produce 3,264 judged submissions. The Google report does not disclose its replicate count, so its observations do not match canonical sampling." } }, { "id": "lbpp_v2_multilingual", "name": "LBPP v2 Multilingual", "category": "Coding", "metric": "functional pass@1 (%)", "num_problems": 943, "source_url": "https://huggingface.co/datasets/CohereLabs/lbpp", "canonical_setting": { "version": "LBPP v2 multilingual six-language test set after mandatory Python canary removal", "metric_type": "pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": false, "tools": "none", "judge": "sandboxed unit tests", "sampling": "pass@1", "notes": "The public multilingual configuration has 944 rows across Python, C++, Go, Java, JavaScript, and Rust; lbpp/python/042 is the mandatory canary and is excluded, leaving 943 scored tasks." } }, { "id": "natural2code", "name": "Natural2Code (Google internal)", "category": "Coding", "metric": "% correct", "num_problems": null, "source_url": "https://arxiv.org/abs/2608.00146", "canonical_setting": { "version": "Google internal Natural2Code evaluation", "metric_type": "pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": false, "tools": "none", "judge": "Google internal evaluator (undisclosed)", "sampling": "pass@1", "notes": "Internal/proprietary benchmark. Public task definition, count, and judge are unavailable; source observations remain valid provider-reported scores." } }, { "id": "pubmedqa_pqal_decision_accuracy", "name": "PubMedQA PQA-L Decision Accuracy", "category": "Knowledge/Medical", "metric": "decision accuracy (%)", "num_problems": 500, "source_url": "https://github.com/pubmedqa/pubmedqa", "canonical_setting": { "version": "PubMedQA PQA-L official 500-item test set: yes/no/maybe decision", "metric_type": "pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": false, "tools": "none", "judge": "exact yes/no/maybe decision accuracy", "sampling": "pass@1", "notes": "The official DiffusionGemma adapter maps the 500 PQA-L test IDs to generated categorical and long-form outputs." } }, { "id": "pubmedqa_pqal_long_answer_bleu", "name": "PubMedQA PQA-L Long-Answer BLEU", "category": "Knowledge/Medical", "metric": "corpus BLEU", "num_problems": 500, "source_url": "https://github.com/pubmedqa/pubmedqa", "canonical_setting": { "version": "PubMedQA PQA-L official 500-item test set: generated long answers", "metric_type": "bleu", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": false, "tools": "none", "judge": "corpus BLEU against PQA-L long answers", "sampling": "pass@1", "notes": "Corpus BLEU over the same 500 generated responses used for the decision metric by the official DiffusionGemma adapter." } }, { "id": "deep_swe_v1_0", "name": "DeepSWE v1.0", "category": "Agentic Coding", "metric": "% resolved (pass@1)", "num_problems": 113, "source_url": "https://deepswe.datacurve.ai/", "canonical_setting": { "version": "DeepSWE v1.0 113-task release", "metric_type": "pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": false, "tools": "agentic repository shell/editor", "sampling": "4 trials; pass@1 per attempt", "judge": "behavioral and correctness verifiers", "harness": "provider harnesses run by Artificial Analysis", "notes": "Contamination-resistant repository issues; provider harness is score-level provenance. The xAI card does not disclose the trial count, so its scores are noncanonical variants." } }, { "id": "apex_swe", "name": "APEX-SWE", "category": "Agentic Coding", "metric": "pass@1 (%)", "num_problems": 200, "source_url": "https://www.mercor.com/apex/apex-swe-leaderboard/", "canonical_setting": { "version": "APEX-SWE 200-task release", "metric_type": "pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": false, "tools": "agentic software-engineering environment", "sampling": "pass@1", "judge": "Mercor integration and observability task verifier", "harness": "Mercor APEX-SWE harness", "notes": "Integration and observability software-engineering tasks." } }, { "id": "swe_marathon_v1_1", "name": "SWE-Marathon v1.1", "category": "Agentic Coding", "metric": "resolution rate (pass@1, %)", "num_problems": 20, "source_url": "https://www.swe-marathon.org/", "canonical_setting": { "version": "SWE-Marathon v1.1 full 20-task set", "metric_type": "pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": false, "tools": "agentic code execution", "sampling": "pass@1", "judge": "multi-layer resolution verification", "harness": "SWE-Marathon v1.1 harness", "notes": "Ultra-long-horizon tasks with reward-hacking-resistant verification." } }, { "id": "falseclaimbench_accuracy", "name": "FalseClaimBench", "category": "Agent Reliability", "metric": "fully true claim accuracy (%)", "num_problems": null, "source_url": "https://media.x.ai/v1/website/4p5-5184fdf9.pdf", "canonical_setting": { "version": "SpaceXAI internal FalseClaimBench", "metric_type": "pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": false, "tools": "workspace editing and command tools", "sampling": "unknown", "judge": "claim-to-final-workspace-state verifier", "harness": "SpaceXAI internal harness", "notes": "Internal task count is undisclosed; checks whether claimed work was actually performed." } }, { "id": "eebench_v1_core_reward", "name": "EEBench v1 Core", "category": "Engineering", "metric": "reward (%)", "num_problems": 13, "source_url": "https://eebench.org/", "canonical_setting": { "version": "EEBench v1 core corpus", "metric_type": "pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": false, "tools": "electrical-engineering design tools", "sampling": "3 trials", "judge": "physical correctness and functionality reward", "harness": "EEBench public leaderboard harness", "notes": "Public v1 core corpus contains 13 tasks; the leaderboard averages three trials." } }, { "id": "parametric_cad_bench_combined", "name": "Parametric CAD Bench", "category": "Engineering", "metric": "combined reward (%)", "num_problems": 100, "source_url": "https://www.gnucleus.ai/cad-bench", "canonical_setting": { "version": "Parametric CAD Bench combined score", "metric_type": "pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": false, "tools": "parametric CAD agent tools", "sampling": "unknown", "judge": "combined executability and geometric reward", "harness": "gNucleus CAD-Bench harness", "notes": "Official benchmark has 100 parametric CAD tasks." } }, { "id": "cadgenbench_generation_cad_score", "name": "CadGenBench Generation", "category": "Engineering", "metric": "generation CAD score (%)", "num_problems": 49, "source_url": "https://github.com/huggingface/cadgenbench", "canonical_setting": { "version": "CadGenBench 49-task generation split", "metric_type": "pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": false, "tools": "CAD generation agent tools", "sampling": "unknown", "judge": "generation CAD score", "harness": "CadGenBench generation harness", "notes": "Generation split contains 49 public tasks." } }, { "id": "spacexai_mts_accuracy", "name": "SpaceXAI MTS Eval", "category": "AI R&D", "metric": "correct (%)", "num_problems": 29, "source_url": "https://media.x.ai/v1/website/4p5-5184fdf9.pdf", "canonical_setting": { "version": "SpaceXAI internal MTS 29-task suite", "metric_type": "pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": false, "tools": "Grok Build coding and research tools", "sampling": "unknown", "judge": "task correctness under fixed rollout budget", "harness": "Grok Build", "notes": "The card explicitly identifies 29 internal model-development tasks; refused tasks are excluded where stated." } }, { "id": "single_turn_hallucination_rate", "name": "Single-Turn Hallucination", "category": "Factuality", "metric": "hallucination rate (%)", "num_problems": null, "source_url": "https://media.x.ai/v1/website/4p5-5184fdf9.pdf", "canonical_setting": { "version": "SpaceXAI internal single-turn factuality suite", "metric_type": "rate_pct", "range": [ 0, 100 ], "higher_is_better": false, "multimodal_input": false, "tools": "none", "sampling": "unknown", "judge": "separate factual-support grader", "harness": "SpaceXAI internal harness", "notes": "Internal task count is undisclosed; lower is better." } }, { "id": "hackerbench_v0_2_harmful_compliance", "name": "HackerBench v0.2 Harmful Compliance", "category": "Safety", "metric": "harmful/dual-use compliance rate (%)", "num_problems": null, "source_url": "https://media.x.ai/v1/website/4p5-5184fdf9.pdf", "canonical_setting": { "version": "HackerBench v0.2 harmful and dual-use split", "metric_type": "rate_pct", "range": [ 0, 100 ], "higher_is_better": false, "multimodal_input": false, "tools": "constrained cyber task tools", "sampling": "unknown", "judge": "harmful or dual-use compliance grader", "harness": "HackerBench v0.2 with standard safeguards", "notes": "Internal task count is undisclosed; lower is better." } }, { "id": "hackerbench_v0_2_benign_refusal", "name": "HackerBench v0.2 Benign Refusal", "category": "Safety", "metric": "benign refusal rate (%)", "num_problems": null, "source_url": "https://media.x.ai/v1/website/4p5-5184fdf9.pdf", "canonical_setting": { "version": "HackerBench v0.2 benign split", "metric_type": "rate_pct", "range": [ 0, 100 ], "higher_is_better": false, "multimodal_input": false, "tools": "constrained cyber task tools", "sampling": "unknown", "judge": "benign refusal grader", "harness": "HackerBench v0.2 with standard safeguards", "notes": "Internal task count is undisclosed; lower is better." } }, { "id": "vct_accuracy", "name": "Virology Capabilities Test", "category": "Safety Capability", "metric": "accuracy (%)", "num_problems": 322, "source_url": "https://arxiv.org/abs/2504.16137", "canonical_setting": { "version": "VCT 322-item release", "metric_type": "pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": true, "tools": "multimodal input; no safeguards", "sampling": "unknown", "judge": "exact VCT accuracy", "harness": "VCT capability harness", "notes": "Dual-use virology troubleshooting capability benchmark." } }, { "id": "biosecurity_vct_accuracy", "name": "Biosecurity VCT", "category": "Safety Capability", "metric": "overall accuracy (%)", "num_problems": null, "source_url": "https://media.x.ai/v1/website/card-4p6-4cd2dc57.pdf", "canonical_setting": { "version": "SpaceXAI internal Biosecurity VCT", "metric_type": "pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": false, "tools": "none; no safeguards", "sampling": "unknown", "judge": "overall should-refuse and should-comply accuracy", "harness": "SpaceXAI internal harness", "notes": "Internal task count is undisclosed." } }, { "id": "biousebench_severity5_refusal_rate", "name": "BioUseBench Severity-5", "category": "Safety", "metric": "severity-5 refusal rate (%)", "num_problems": null, "source_url": "https://media.x.ai/v1/website/card-4p6-4cd2dc57.pdf", "canonical_setting": { "version": "SpaceXAI internal BioUseBench severity-5 split", "metric_type": "pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": false, "tools": "standard safeguards", "sampling": "unknown", "judge": "severity-5 refusal grader", "harness": "SpaceXAI internal harness", "notes": "Internal high-elo dual-use biology split; task count is undisclosed." } }, { "id": "wmdp_cyber_accuracy", "name": "WMDP-Cyber", "category": "Safety Capability", "metric": "accuracy (%)", "num_problems": 1987, "source_url": "https://wmdp.ai", "canonical_setting": { "version": "WMDP-Cyber public release", "metric_type": "pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": false, "tools": "none; no safeguards", "sampling": "pass@1", "judge": "multiple-choice exact accuracy", "harness": "WMDP official evaluation", "notes": "Public WMDP-Cyber split contains 1,987 questions." } }, { "id": "xai_standard_jailbreak_compliance", "name": "SpaceXAI Standard Jailbreaks", "category": "Safety", "metric": "compliance rate (%)", "num_problems": null, "source_url": "https://media.x.ai/v1/website/4p5-5184fdf9.pdf", "canonical_setting": { "version": "SpaceXAI standard jailbreak suite", "metric_type": "rate_pct", "range": [ 0, 100 ], "higher_is_better": false, "multimodal_input": false, "tools": "standard safeguards", "sampling": "unknown", "judge": "should-refuse compliance grader", "harness": "SpaceXAI internal harness", "notes": "Internal task count is undisclosed; lower is better." } }, { "id": "strongreject_compliance", "name": "StrongREJECT Compliance", "category": "Safety", "metric": "compliance rate (%)", "num_problems": 313, "source_url": "https://arxiv.org/abs/2402.10260", "canonical_setting": { "version": "StrongREJECT 313-prompt suite", "metric_type": "rate_pct", "range": [ 0, 100 ], "higher_is_better": false, "multimodal_input": false, "tools": "standard safeguards", "sampling": "pass@1", "judge": "StrongREJECT compliance grader", "harness": "StrongREJECT evaluation", "notes": "Public StrongREJECT suite contains 313 prompts; lower is better." } }, { "id": "xai_long_horizon_jailbreak_compliance", "name": "SpaceXAI Long-Horizon Jailbreaks", "category": "Safety", "metric": "compliance rate (%)", "num_problems": null, "source_url": "https://media.x.ai/v1/website/card-4p6-4cd2dc57.pdf", "canonical_setting": { "version": "SpaceXAI long-horizon jailbreak suite", "metric_type": "rate_pct", "range": [ 0, 100 ], "higher_is_better": false, "multimodal_input": false, "tools": "standard safeguards", "sampling": "unknown", "judge": "long-horizon should-refuse compliance grader", "harness": "SpaceXAI internal harness", "notes": "Includes multiturn, Crescendo, and malicious-use long-horizon attacks; count undisclosed." } }, { "id": "xai_general_refusal_compliance", "name": "SpaceXAI General Refusal Compliance", "category": "Safety", "metric": "compliance rate (%)", "num_problems": null, "source_url": "https://media.x.ai/v1/website/4p5-5184fdf9.pdf", "canonical_setting": { "version": "SpaceXAI multilingual general-refusal suite", "metric_type": "rate_pct", "range": [ 0, 100 ], "higher_is_better": false, "multimodal_input": false, "tools": "standard safeguards", "sampling": "unknown", "judge": "multilingual should-refuse compliance grader", "harness": "SpaceXAI internal harness", "notes": "Internal multilingual task count is undisclosed; lower is better." } }, { "id": "xai_child_safety_compliance", "name": "SpaceXAI Child Safety Compliance", "category": "Safety", "metric": "compliance rate (%)", "num_problems": null, "source_url": "https://media.x.ai/v1/website/4p5-5184fdf9.pdf", "canonical_setting": { "version": "SpaceXAI CSAM and child-safety suite", "metric_type": "rate_pct", "range": [ 0, 100 ], "higher_is_better": false, "multimodal_input": false, "tools": "standard safeguards", "sampling": "unknown", "judge": "CSAM and child-safety compliance grader", "harness": "SpaceXAI internal harness", "notes": "Internal multi-turn suite; lower is better." } }, { "id": "xai_bio_refusal_accuracy", "name": "SpaceXAI Bio Refusal", "category": "Safety", "metric": "refusal accuracy (%)", "num_problems": null, "source_url": "https://media.x.ai/v1/website/4p5-5184fdf9.pdf", "canonical_setting": { "version": "SpaceXAI Autointent-Bio dangerous-query split", "metric_type": "pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": false, "tools": "full safeguards", "sampling": "unknown", "judge": "dangerous-query refusal accuracy", "harness": "SpaceXAI internal harness", "notes": "Internal task count is undisclosed." } }, { "id": "xai_chem_refusal_accuracy", "name": "SpaceXAI Chem Refusal", "category": "Safety", "metric": "refusal accuracy (%)", "num_problems": null, "source_url": "https://media.x.ai/v1/website/4p5-5184fdf9.pdf", "canonical_setting": { "version": "SpaceXAI Autointent-Chem dangerous-query split", "metric_type": "pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": false, "tools": "full safeguards", "sampling": "unknown", "judge": "dangerous-query refusal accuracy", "harness": "SpaceXAI internal harness", "notes": "Internal task count is undisclosed." } }, { "id": "xai_rn_refusal_accuracy", "name": "SpaceXAI Radiological/Nuclear Refusal", "category": "Safety", "metric": "refusal accuracy (%)", "num_problems": null, "source_url": "https://labs.scale.com/leaderboard/fortress", "canonical_setting": { "version": "FORTRESS-RN dangerous-query split", "metric_type": "pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": false, "tools": "full safeguards", "sampling": "unknown", "judge": "dangerous-query refusal accuracy", "harness": "FORTRESS-RN evaluation", "notes": "The xAI card reports the R/N refusal row; exact evaluated count is undisclosed." } }, { "id": "xai_self_harm_compliance", "name": "SpaceXAI Self-Harm Compliance", "category": "Safety", "metric": "compliance rate (%)", "num_problems": null, "source_url": "https://media.x.ai/v1/website/4p5-5184fdf9.pdf", "canonical_setting": { "version": "SpaceXAI self-harm suite", "metric_type": "rate_pct", "range": [ 0, 100 ], "higher_is_better": false, "multimodal_input": false, "tools": "standard safeguards", "sampling": "unknown", "judge": "self-harm assistance or failed-redirection compliance grader", "harness": "SpaceXAI internal harness", "notes": "Internal task count is undisclosed; lower is better." } }, { "id": "xai_epistemic_bias_rate", "name": "SpaceXAI Epistemic Bias", "category": "Behavior", "metric": "bias rate (%)", "num_problems": null, "source_url": "https://media.x.ai/v1/website/4p5-5184fdf9.pdf", "canonical_setting": { "version": "SpaceXAI epistemic-bias suite", "metric_type": "rate_pct", "range": [ 0, 100 ], "higher_is_better": false, "multimodal_input": false, "tools": "none", "sampling": "unknown", "judge": "opposing-framing bias grader", "harness": "SpaceXAI internal harness", "notes": "Internal task count is undisclosed; lower is better." } }, { "id": "xai_mask_rectified_dishonesty", "name": "MASK-Rectified Dishonesty", "category": "Behavior", "metric": "dishonesty rate (%)", "num_problems": null, "source_url": "https://www.mask-benchmark.ai/", "canonical_setting": { "version": "SpaceXAI MASK-Rectified evaluation", "metric_type": "rate_pct", "range": [ 0, 100 ], "higher_is_better": false, "multimodal_input": false, "tools": "none", "sampling": "unknown", "judge": "MASK-Rectified dishonesty grader", "harness": "SpaceXAI rectified harness", "notes": "Rectifies model-aware role-playing cases; exact evaluated count is undisclosed." } }, { "id": "xai_sycophancy_rate", "name": "SpaceXAI Sycophancy", "category": "Behavior", "metric": "sycophancy rate (%)", "num_problems": null, "source_url": "https://media.x.ai/v1/website/4p5-5184fdf9.pdf", "canonical_setting": { "version": "SpaceXAI internal sycophancy suite", "metric_type": "rate_pct", "range": [ 0, 100 ], "higher_is_better": false, "multimodal_input": false, "tools": "none", "sampling": "unknown", "judge": "accuracy-drop sycophancy grader", "harness": "SpaceXAI internal harness", "notes": "Internal task count is undisclosed; lower is better." } }, { "id": "terminal_bench_3_0", "name": "Terminal-Bench 3.0", "category": "Agentic Coding", "metric": "task success rate (%)", "num_problems": 74, "source_url": "https://www.tbench.ai/", "canonical_setting": { "version": "Terminal-Bench 3.0 / FrontierBench 74-task release", "metric_type": "pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": false, "tools": "container terminal", "sampling": "3 trials", "judge": "verified task-success evaluator", "harness": "Terminal-Bench 3.0 fixed harness", "notes": "Expanded successor to Terminal-Bench 2.1; public release contains 74 tasks and uses three trials by default. The xAI card does not disclose trial count, so its scores are noncanonical variants." } }, { "id": "xai_inferenceeval_accuracy", "name": "SpaceXAI InferenceEval", "category": "AI R&D", "metric": "accuracy (%)", "num_problems": null, "source_url": "https://media.x.ai/v1/website/card-4p6-4cd2dc57.pdf", "canonical_setting": { "version": "SpaceXAI internal InferenceEval", "metric_type": "pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": false, "tools": "Grok Build coding tools and GPU tests", "sampling": "unknown", "judge": "hidden GPU unit score gated by integration probe", "harness": "SpaceXAI internal harness", "notes": "Internal task count is undisclosed." } }, { "id": "xai_kernelbenchinternal_accuracy_efficiency", "name": "SpaceXAI KernelBenchInternal", "category": "AI R&D", "metric": "accuracy / efficiency score (%)", "num_problems": null, "source_url": "https://media.x.ai/v1/website/card-4p6-4cd2dc57.pdf", "canonical_setting": { "version": "SpaceXAI internal KernelBench-inspired suite", "metric_type": "pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": false, "tools": "sandboxed GPU coding tools", "sampling": "unknown", "judge": "correctness and measured speedup composite", "harness": "SpaceXAI internal harness", "notes": "Internal task count is undisclosed; the score combines correctness and speedup." } }, { "id": "cve_bench_reward", "name": "CVE-Bench", "category": "Cybersecurity", "metric": "reward (%)", "num_problems": 40, "source_url": "https://arxiv.org/abs/2503.17332", "canonical_setting": { "version": "CVE-Bench 40-environment release", "metric_type": "pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": false, "tools": "unrestricted sandboxed web-application tools", "sampling": "unknown", "judge": "CVE-Bench exploit reward", "harness": "CVE-Bench sandbox", "notes": "Public benchmark contains 40 CVE environments." } }, { "id": "xai_securecodereview_reward", "name": "SpaceXAI SecureCodeReview", "category": "Cybersecurity", "metric": "reward (%)", "num_problems": null, "source_url": "https://media.x.ai/v1/website/card-4p6-4cd2dc57.pdf", "canonical_setting": { "version": "SpaceXAI internal SecureCodeReview", "metric_type": "pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": false, "tools": "fixed secure-code-review tools", "sampling": "unknown", "judge": "security-fix reward with regression checks", "harness": "SpaceXAI internal harness", "notes": "Internal task count is undisclosed." } }, { "id": "harvey_lab_vals_final_score", "name": "Harvey Legal Agent Benchmark (Vals)", "category": "Professional", "metric": "final score (%)", "num_problems": null, "source_url": "https://www.vals.ai/benchmarks/hlab", "canonical_setting": { "version": "Vals implementation of Harvey LAB", "metric_type": "pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": false, "tools": "Vals legal-agent file tools", "sampling": "unknown", "judge": "expert all-pass final score", "harness": "Vals Harvey LAB harness", "notes": "Long-horizon legal work over client-matter files; count is undisclosed." } }, { "id": "workspace_bench_openclaw_100", "name": "Workspace Bench (100-task OpenClaw total)", "category": "Agentic Office", "metric": "total score (%)", "num_problems": 100, "source_url": "https://arxiv.org/abs/2605.03596", "canonical_setting": { "version": "Workspace Bench (100-task OpenClaw total)", "metric_type": "pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": false, "tools": "OpenClaw workspace tools", "sampling": "pass@1", "judge": "rubric-based evaluator", "harness": "100-task OpenClaw setting", "notes": "Official Seed2.1 Table 1 explicitly reports a 100-task OpenClaw setting. This source-defined subset is distinct from the full public Workspace-Bench release." } }, { "id": "workspace_bench_openclaw_100_pass30", "name": "Workspace Bench (100-task OpenClaw Pass@30)", "category": "Agentic Office", "metric": "% tasks with rubric score >= 30", "num_problems": 100, "source_url": "https://arxiv.org/abs/2605.03596", "canonical_setting": { "version": "Workspace Bench (100-task OpenClaw Pass@30)", "metric_type": "pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": false, "tools": "OpenClaw workspace tools", "sampling": "pass@1", "judge": "rubric-based evaluator", "harness": "100-task OpenClaw setting", "notes": "Pass@threshold is the source's rubric-score threshold metric, not repeated sampling." } }, { "id": "workspace_bench_openclaw_100_pass50", "name": "Workspace Bench (100-task OpenClaw Pass@50)", "category": "Agentic Office", "metric": "% tasks with rubric score >= 50", "num_problems": 100, "source_url": "https://arxiv.org/abs/2605.03596", "canonical_setting": { "version": "Workspace Bench (100-task OpenClaw Pass@50)", "metric_type": "pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": false, "tools": "OpenClaw workspace tools", "sampling": "pass@1", "judge": "rubric-based evaluator", "harness": "100-task OpenClaw setting", "notes": "Pass@threshold is the source's rubric-score threshold metric, not repeated sampling." } }, { "id": "workspace_bench_openclaw_100_pass70", "name": "Workspace Bench (100-task OpenClaw Pass@70)", "category": "Agentic Office", "metric": "% tasks with rubric score >= 70", "num_problems": 100, "source_url": "https://arxiv.org/abs/2605.03596", "canonical_setting": { "version": "Workspace Bench (100-task OpenClaw Pass@70)", "metric_type": "pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": false, "tools": "OpenClaw workspace tools", "sampling": "pass@1", "judge": "rubric-based evaluator", "harness": "100-task OpenClaw setting", "notes": "Pass@threshold is the source's rubric-score threshold metric, not repeated sampling." } }, { "id": "workspace_bench_openclaw_100_pass90", "name": "Workspace Bench (100-task OpenClaw Pass@90)", "category": "Agentic Office", "metric": "% tasks with rubric score >= 90", "num_problems": 100, "source_url": "https://arxiv.org/abs/2605.03596", "canonical_setting": { "version": "Workspace Bench (100-task OpenClaw Pass@90)", "metric_type": "pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": false, "tools": "OpenClaw workspace tools", "sampling": "pass@1", "judge": "rubric-based evaluator", "harness": "100-task OpenClaw setting", "notes": "Pass@threshold is the source's rubric-score threshold metric, not repeated sampling." } }, { "id": "workspace_bench_openclaw_100_pass100", "name": "Workspace Bench (100-task OpenClaw Pass@100)", "category": "Agentic Office", "metric": "% tasks with rubric score >= 100", "num_problems": 100, "source_url": "https://arxiv.org/abs/2605.03596", "canonical_setting": { "version": "Workspace Bench (100-task OpenClaw Pass@100)", "metric_type": "pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": false, "tools": "OpenClaw workspace tools", "sampling": "pass@1", "judge": "rubric-based evaluator", "harness": "100-task OpenClaw setting", "notes": "Pass@threshold is the source's rubric-score threshold metric, not repeated sampling." } }, { "id": "presentbench", "name": "PresentBench", "category": "Agentic Office", "metric": "rubric score (%)", "num_problems": 238, "source_url": "https://arxiv.org/abs/2603.07244", "canonical_setting": { "version": "PresentBench", "metric_type": "pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": true, "tools": "slide-generation environment", "sampling": "pass@1", "judge": "fine-grained instance-specific rubrics", "harness": "official", "notes": "Official PresentBench paper; exact scored task count pending metadata audit." } }, { "id": "agent_startup_bench", "name": "Agent Startup Bench", "category": "Agentic Professional", "metric": "expert-reviewed score (%)", "num_problems": null, "source_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2.1/Seed2_1_Model_Card.pdf", "canonical_setting": { "version": "Agent Startup Bench", "metric_type": "pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": false, "tools": "research and professional-deliverable tools", "sampling": "pass@1", "judge": "expert review", "harness": "official", "notes": "Seed-developed benchmark based on research and interviews with real AI-native startups; no public fixed count disclosed." } }, { "id": "agents_last_exam_average_score", "name": "Agents' Last Exam (average overall score)", "category": "Agentic", "metric": "average overall score", "num_problems": null, "source_url": "https://agents-last-exam.org/docs/ale/index.html", "canonical_setting": { "version": "Agents' Last Exam (average overall score)", "metric_type": "score", "range": null, "higher_is_better": true, "multimodal_input": true, "tools": "computer-use environment", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official", "notes": "Distinct from the existing full-pass-rate metric." } }, { "id": "one_million_bench", "name": "OneMillion Bench", "category": "Agentic Professional", "metric": "reported score (%)", "num_problems": 400, "source_url": "https://arxiv.org/abs/2603.07980", "canonical_setting": { "version": "OneMillion Bench", "metric_type": "pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": false, "tools": "agentic professional-task environment", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official", "notes": "Official OneMillion-Bench paper; exact scored task count pending metadata audit." } }, { "id": "gdpval_seed_reported_score", "name": "GDPVal (Seed source-reported score)", "category": "Agentic Professional", "metric": "source-reported score (0-100)", "num_problems": 220, "source_url": "https://huggingface.co/datasets/openai/gdpval", "canonical_setting": { "version": "GDPVal (Seed source-reported score)", "metric_type": "score", "range": null, "higher_is_better": true, "multimodal_input": true, "tools": "document/spreadsheet deliverable workflow", "sampling": "pass@1", "judge": "source-specific evaluator; normalization undisclosed", "harness": "official", "notes": "The Seed source reports 0-100 values without identifying AA Elo, OpenAI wins-or-ties, GDPVal-Diamond, or another published aggregation. A distinct metric prevents conflation." } }, { "id": "xdailybench", "name": "xDailyBench", "category": "Agentic Daily Life", "metric": "rubric score (%)", "num_problems": null, "source_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2.1/Seed2_1_Model_Card.pdf", "canonical_setting": { "version": "xDailyBench", "metric_type": "pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": false, "tools": "benchmark-specified", "sampling": "pass@1", "judge": "multi-dimensional rubric evaluator", "harness": "official", "notes": "Seed-developed benchmark covering more than 30 vertical scenarios; no fixed public item count disclosed." } }, { "id": "doubao_multi_turn_bench", "name": "Doubao Multi-Turn Bench", "category": "Conversation", "metric": "rubric score (%)", "num_problems": null, "source_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2.1/Seed2_1_Model_Card.pdf", "canonical_setting": { "version": "Doubao Multi-Turn Bench", "metric_type": "pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": false, "tools": "none", "sampling": "pass@1", "judge": "rubric-based evaluator", "harness": "official", "notes": "Seed-developed benchmark filtered from real Doubao conversations; no fixed public item count disclosed." } }, { "id": "seedclawbench", "name": "SeedClawBench", "category": "Agentic", "metric": "Agent-as-Judge score (%)", "num_problems": 100, "source_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2.1/Seed2_1_Model_Card.pdf", "canonical_setting": { "version": "SeedClawBench", "metric_type": "pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": false, "tools": "OpenClaw-style tools and skills", "sampling": "pass@1", "judge": "Agent-as-Judge with read-only evidence tools", "harness": "official", "notes": "100 tasks constructed from more than 5,000 Ark tasks and more than 600 crowdsourced tasks; 89.69% human-machine rubric agreement on the pilot." } }, { "id": "claw_eval_multimodal_pass3", "name": "Claw-Eval Multimodal (Pass^3)", "category": "Multimodal Agentic", "metric": "all-three-pass rate (%)", "num_problems": 303, "source_url": "https://raw.githubusercontent.com/claw-eval/claw-eval/9ac81fc3f18e9711bc0e0e3a96f0883887ae88b4/README.md", "canonical_setting": { "version": "Claw-Eval v1.1 multimodal Pass^3", "metric_type": "pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": true, "tools": "official multimodal Claw-Eval agent environment", "sampling": "Pass^3: three successful trajectories required", "judge": "full-trajectory benchmark grading", "harness": "official Claw-Eval v1.1", "notes": "Distinct 101-task multimodal subset with exactly 303 generated trajectories." } }, { "id": "wildclaw_bench_60_openclaw", "name": "WildClawBench (60-task full OpenClaw suite)", "category": "Multimodal Agentic", "metric": "weighted overall score (%)", "num_problems": 60, "source_url": "https://github.com/InternLM/WildClawBench", "canonical_setting": { "version": "WildClawBench (60-task full OpenClaw suite)", "metric_type": "pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": true, "tools": "multimodal agent harness", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official", "notes": "Full suite contains 35 text-oriented and 25 multimodal tasks; distinct from the existing 35-task text-only subset." } }, { "id": "image2floorplan_avg_score", "name": "Image2FloorPlan (average graph-similarity score)", "category": "Multimodal Spatial", "metric": "average normalized graph-similarity score (%)", "num_problems": null, "source_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2.1/Seed2_1_Model_Card.pdf", "canonical_setting": { "version": "Image2FloorPlan (average graph-similarity score)", "metric_type": "pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": true, "tools": "high-resolution image perception", "sampling": "pass@1", "judge": "weighted room-connectivity graph similarity", "harness": "official", "notes": "Seed-developed in-house benchmark; fixed item count undisclosed." } }, { "id": "mobileworld", "name": "MobileWorld", "category": "Computer Use", "metric": "task success rate (%)", "num_problems": 201, "source_url": "https://arxiv.org/abs/2512.19432", "canonical_setting": { "version": "MobileWorld", "metric_type": "pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": true, "tools": "mobile GUI actions", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official", "notes": "Seed excludes the MCP test split due to deployment issues." } }, { "id": "creativework", "name": "CreativeWork", "category": "Computer Use", "metric": "task success rate (%)", "num_problems": null, "source_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2.1/Seed2_1_Model_Card.pdf", "canonical_setting": { "version": "CreativeWork", "metric_type": "pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": true, "tools": "GUI + MCP in Notion, Canva, and Figma", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official", "notes": "Seed-developed benchmark; fixed item count undisclosed." } }, { "id": "gameworld", "name": "GameWorld", "category": "Computer Use", "metric": "task success rate (%)", "num_problems": 170, "source_url": "https://arxiv.org/abs/2604.07429", "canonical_setting": { "version": "GameWorld", "metric_type": "pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": true, "tools": "browser-game GUI actions", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official", "notes": "Exact scored task count pending metadata audit." } }, { "id": "program_bench_fully_resolved", "name": "ProgramBench (fully resolved)", "category": "Coding", "metric": "fully resolved tasks (%)", "num_problems": 200, "source_url": "https://programbench.com/", "canonical_setting": { "version": "ProgramBench (fully resolved)", "metric_type": "pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": false, "tools": "compiled executable and documentation", "sampling": "pass@1", "judge": "248,000+ behavioral tests", "harness": "official ProgramBench sandbox", "notes": "Distinct from the macro-average behavioral-tests-passed metric." } }, { "id": "program_bench_almost_resolved_95", "name": "ProgramBench (almost resolved >=95%)", "category": "Coding", "metric": "tasks with at least 95% pass rate (%)", "num_problems": 200, "source_url": "https://programbench.com/", "canonical_setting": { "version": "ProgramBench (almost resolved >=95%)", "metric_type": "pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": false, "tools": "compiled executable and documentation", "sampling": "pass@1", "judge": "248,000+ behavioral tests", "harness": "official ProgramBench sandbox", "notes": "Distinct from fully resolved and macro-average metrics." } }, { "id": "zerobench_sub_tools", "name": "ZeroBench subquestions (with tools)", "category": "Multimodal Reasoning", "metric": "accuracy (%)", "num_problems": 334, "source_url": "https://arxiv.org/abs/2502.09696", "canonical_setting": { "version": "ZeroBench subquestions (with tools)", "metric_type": "pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": true, "tools": "source tool-augmented setting", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official", "notes": "Distinct from ZeroBench main with tools and the no-tool subquestion row." } }, { "id": "swe_atlas", "name": "SWE-Atlas", "category": "Agentic Coding", "metric": "task resolve rate (%)", "num_problems": 500, "source_url": "https://arxiv.org/abs/2605.08366", "canonical_setting": { "version": "SWE-Atlas", "metric_type": "pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": false, "tools": "repository shell/editor", "sampling": "pass@1", "judge": "test-based task resolution", "harness": "SWE-agent", "notes": "Patch-generation benchmark; distinct from SWE-Atlas-QnA." } }, { "id": "webbench", "name": "WebBench", "category": "Agentic Coding", "metric": "live task completion rate (%)", "num_problems": 2454, "source_url": "https://github.com/Halluminate/WebBench", "canonical_setting": { "version": "WebBench", "metric_type": "pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": false, "tools": "browser workflows on 452 live sites", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official", "notes": "Pin the filtered 2,454-task release, not the historical 5,750-task count." } }, { "id": "seedkernelbench_avg_speedup", "name": "SeedKernelBench (average speedup)", "category": "Agentic Coding", "metric": "average speedup ratio", "num_problems": null, "source_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2.1/Seed2_1_Model_Card.pdf", "canonical_setting": { "version": "SeedKernelBench (average speedup)", "metric_type": "score", "range": null, "higher_is_better": true, "multimodal_input": false, "tools": "GPU kernel implementation and benchmarking", "sampling": "pass@1", "judge": "measured speedup", "harness": "official", "notes": "Higher is better; source reports ratios such as 9.21x." } }, { "id": "trae_repo_env", "name": "Trae Repo Env", "category": "Agentic Coding", "metric": "reported score (%)", "num_problems": null, "source_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2.1/Seed2_1_Model_Card.pdf", "canonical_setting": { "version": "Trae Repo Env", "metric_type": "pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": false, "tools": "Trae coding-agent environment", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official", "notes": "Seed/Trae internal benchmark; fixed item count undisclosed." } }, { "id": "trae_artifacts", "name": "Trae Artifacts", "category": "Agentic Coding", "metric": "reported score (%)", "num_problems": null, "source_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2.1/Seed2_1_Model_Card.pdf", "canonical_setting": { "version": "Trae Artifacts", "metric_type": "pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": false, "tools": "Trae coding-agent environment", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official", "notes": "Seed/Trae internal benchmark; fixed item count undisclosed." } }, { "id": "trae_error_fix_python", "name": "Trae Error Fix Python", "category": "Agentic Coding", "metric": "reported score (%)", "num_problems": null, "source_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2.1/Seed2_1_Model_Card.pdf", "canonical_setting": { "version": "Trae Error Fix Python", "metric_type": "pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": false, "tools": "Trae coding-agent environment", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official", "notes": "Seed/Trae internal benchmark; fixed item count undisclosed." } }, { "id": "trae_error_fix_js", "name": "Trae Error Fix JavaScript", "category": "Agentic Coding", "metric": "reported score (%)", "num_problems": null, "source_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2.1/Seed2_1_Model_Card.pdf", "canonical_setting": { "version": "Trae Error Fix JavaScript", "metric_type": "pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": false, "tools": "Trae coding-agent environment", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official", "notes": "Seed/Trae internal benchmark; fixed item count undisclosed." } }, { "id": "trae_error_fix_java", "name": "Trae Error Fix Java", "category": "Agentic Coding", "metric": "reported score (%)", "num_problems": null, "source_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2.1/Seed2_1_Model_Card.pdf", "canonical_setting": { "version": "Trae Error Fix Java", "metric_type": "pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": false, "tools": "Trae coding-agent environment", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official", "notes": "Seed/Trae internal benchmark; fixed item count undisclosed." } }, { "id": "trae_error_fix_go", "name": "Trae Error Fix Go", "category": "Agentic Coding", "metric": "reported score (%)", "num_problems": null, "source_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2.1/Seed2_1_Model_Card.pdf", "canonical_setting": { "version": "Trae Error Fix Go", "metric_type": "pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": false, "tools": "Trae coding-agent environment", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official", "notes": "Seed/Trae internal benchmark; fixed item count undisclosed." } }, { "id": "trae_code_gen_python", "name": "Trae Code Gen Python", "category": "Agentic Coding", "metric": "reported score (%)", "num_problems": null, "source_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2.1/Seed2_1_Model_Card.pdf", "canonical_setting": { "version": "Trae Code Gen Python", "metric_type": "pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": false, "tools": "Trae coding-agent environment", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official", "notes": "Seed/Trae internal benchmark; fixed item count undisclosed." } }, { "id": "trae_code_gen_js", "name": "Trae Code Gen JavaScript", "category": "Agentic Coding", "metric": "reported score (%)", "num_problems": null, "source_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2.1/Seed2_1_Model_Card.pdf", "canonical_setting": { "version": "Trae Code Gen JavaScript", "metric_type": "pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": false, "tools": "Trae coding-agent environment", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official", "notes": "Seed/Trae internal benchmark; fixed item count undisclosed." } }, { "id": "frontiercs_overall_v1", "name": "FrontierCS v1 (overall)", "category": "Agentic Research", "metric": "mean continuous score (%)", "num_problems": 256, "source_url": "https://arxiv.org/abs/2512.15699", "canonical_setting": { "version": "FrontierCS v1 (overall)", "metric_type": "pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": false, "tools": "research and code-execution environment", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official", "notes": "Source-date v1 snapshot: 188 algorithmic + 68 research tasks. Overall is another metric view over the same 256 generations; do not add all three rows for cost aggregation." } }, { "id": "frontiercs_algorithmic_v1", "name": "FrontierCS v1 (algorithmic)", "category": "Agentic Research", "metric": "mean continuous score (%)", "num_problems": 188, "source_url": "https://arxiv.org/abs/2512.15699", "canonical_setting": { "version": "FrontierCS v1 (algorithmic)", "metric_type": "pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": false, "tools": "research and code-execution environment", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official", "notes": "Source-date algorithmic track; excludes later FrontierCS 2.0 additions." } }, { "id": "frontiercs_research_v1", "name": "FrontierCS v1 (research)", "category": "Agentic Research", "metric": "mean continuous score (%)", "num_problems": 68, "source_url": "https://arxiv.org/abs/2512.15699", "canonical_setting": { "version": "FrontierCS v1 (research)", "metric_type": "pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": false, "tools": "research and code-execution environment", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official", "notes": "Source-date research/systems track with task-specific partial credit." } }, { "id": "mathverse_vision_only", "name": "MathVerse (Vision-Only)", "category": "Multimodal Math", "metric": "accuracy (%)", "num_problems": null, "source_url": "https://github.com/ZrrSkywalker/MathVerse", "canonical_setting": { "version": "MathVerse (Vision-Only)", "metric_type": "pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": true, "tools": "none", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official", "notes": "Vision-only MathVerse setting." } }, { "id": "measurebench", "name": "MeasureBench", "category": "Multimodal Perception", "metric": "average real-and-synthetic score (%)", "num_problems": 2442, "source_url": "https://arxiv.org/abs/2510.26865", "canonical_setting": { "version": "MeasureBench", "metric_type": "pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": true, "tools": "none", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official", "notes": "Source reports the average over real and synthetic subsets." } }, { "id": "worldbench", "name": "WorldBench", "category": "Multimodal Knowledge", "metric": "accuracy (%)", "num_problems": 2000, "source_url": "https://arxiv.org/abs/2606.06538", "canonical_setting": { "version": "WorldBench", "metric_type": "pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": true, "tools": "none", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official", "notes": "WorldVQA paper benchmark family; exact WorldBench split count pending audit." } }, { "id": "embspatial_bench", "name": "EmbSpatialBench", "category": "Multimodal Spatial", "metric": "accuracy (%)", "num_problems": 3640, "source_url": "https://arxiv.org/abs/2406.05756", "canonical_setting": { "version": "EmbSpatialBench", "metric_type": "pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": true, "tools": "none", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official", "notes": "Official paper/source resolution pending metadata audit." } }, { "id": "kina", "name": "KINA", "category": "Knowledge", "metric": "multiple-choice accuracy (%)", "num_problems": 899, "source_url": "https://arxiv.org/abs/2606.05104", "canonical_setting": { "version": "KINA", "metric_type": "pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": false, "tools": "none", "sampling": "pass@1", "judge": "letter extraction", "harness": "official KINA/lighteval runner", "notes": "" } }, { "id": "msqa", "name": "Multicultural SimpleQA", "category": "Multilingual Knowledge", "metric": "accuracy (%)", "num_problems": 1086, "source_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2.1/Seed2_1_Model_Card.pdf", "canonical_setting": { "version": "Multicultural SimpleQA", "metric_type": "pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": false, "tools": "none", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official", "notes": "Seed-developed internal benchmark across 11 major languages." } }, { "id": "live_mathematician_bench_2026_06", "name": "LiveMathematicianBench (Seed2.1 June 2026 snapshot)", "category": "Reasoning", "metric": "reported score (%)", "num_problems": null, "source_url": "https://arxiv.org/abs/2604.01754", "canonical_setting": { "version": "LiveMathematicianBench (Seed2.1 June 2026 snapshot)", "metric_type": "pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": false, "tools": "none", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official", "notes": "Live benchmark; the exact Seed model-card snapshot count is not published. Do not silently substitute the later 608-item live release." } }, { "id": "fs_researcher", "name": "FS-Researcher", "category": "Agentic Research", "metric": "reported score (%)", "num_problems": null, "source_url": "https://arxiv.org/abs/2602.01566", "canonical_setting": { "version": "FS-Researcher", "metric_type": "pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": false, "tools": "file-system-based research agent environment", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official", "notes": "Table 11 cites FS-Researcher, which is distinct from FrontierScience-Research in Tables 6-7." } }, { "id": "code_arena_frontend_elo", "name": "Code Arena: Frontend Elo", "category": "Coding", "metric": "Arena Elo", "num_problems": null, "source_url": "https://arena.ai/leaderboard/code", "canonical_setting": { "version": "Code Arena: Frontend Elo", "metric_type": "score", "range": null, "higher_is_better": true, "multimodal_input": false, "tools": "none", "sampling": "dynamic leaderboard aggregation", "judge": "human preference", "harness": "Arena Code leaderboard", "notes": "Historical 107,962-vote snapshot reproduced in the Seed release blog." } }, { "id": "seed21_claudecode_glm51_win_rate", "name": "Seed2.1 ClaudeCode vs GLM-5.1 (win rate)", "category": "Human Preference / Internal", "metric": "Seed model win rate (%)", "num_problems": 178, "source_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2.1/Seed2_1_Model_Card.pdf", "canonical_setting": { "version": "Seed2.1 ClaudeCode vs GLM-5.1 (win rate)", "metric_type": "pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": false, "tools": "real-repository coding environment", "sampling": "pass@1", "judge": "anonymous developer preference or multidimensional human rating", "harness": "Seed ClaudeCode/Trae crowdsourced evaluation", "notes": "Official Seed2.1 model-card internal evaluation." } }, { "id": "seed21_claudecode_glm51_net_win_rate", "name": "Seed2.1 ClaudeCode vs GLM-5.1 (net win rate)", "category": "Human Preference / Internal", "metric": "net win rate (percentage points)", "num_problems": 178, "source_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2.1/Seed2_1_Model_Card.pdf", "canonical_setting": { "version": "Seed2.1 ClaudeCode vs GLM-5.1 (net win rate)", "metric_type": "score", "range": null, "higher_is_better": true, "multimodal_input": false, "tools": "real-repository coding environment", "sampling": "pass@1", "judge": "anonymous developer preference or multidimensional human rating", "harness": "Seed ClaudeCode/Trae crowdsourced evaluation", "notes": "Official Seed2.1 model-card internal evaluation." } }, { "id": "seed21_claudecode_opus46_win_rate", "name": "Seed2.1 ClaudeCode vs Claude Opus 4.6 (win rate)", "category": "Human Preference / Internal", "metric": "Seed model win rate (%)", "num_problems": 230, "source_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2.1/Seed2_1_Model_Card.pdf", "canonical_setting": { "version": "Seed2.1 ClaudeCode vs Claude Opus 4.6 (win rate)", "metric_type": "pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": false, "tools": "real-repository coding environment", "sampling": "pass@1", "judge": "anonymous developer preference or multidimensional human rating", "harness": "Seed ClaudeCode/Trae crowdsourced evaluation", "notes": "Official Seed2.1 model-card internal evaluation." } }, { "id": "seed21_claudecode_opus46_tie_rate", "name": "Seed2.1 ClaudeCode vs Claude Opus 4.6 (tie rate)", "category": "Human Preference / Internal", "metric": "tie rate (%)", "num_problems": 230, "source_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2.1/Seed2_1_Model_Card.pdf", "canonical_setting": { "version": "Seed2.1 ClaudeCode vs Claude Opus 4.6 (tie rate)", "metric_type": "pct", "range": [ 0, 100 ], "higher_is_better": false, "multimodal_input": false, "tools": "real-repository coding environment", "sampling": "pass@1", "judge": "anonymous developer preference or multidimensional human rating", "harness": "Seed ClaudeCode/Trae crowdsourced evaluation", "notes": "Official Seed2.1 model-card internal evaluation." } }, { "id": "seed21_claudecode_opus46_loss_rate", "name": "Seed2.1 ClaudeCode vs Claude Opus 4.6 (loss rate)", "category": "Human Preference / Internal", "metric": "Seed model loss rate (%)", "num_problems": 230, "source_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2.1/Seed2_1_Model_Card.pdf", "canonical_setting": { "version": "Seed2.1 ClaudeCode vs Claude Opus 4.6 (loss rate)", "metric_type": "pct", "range": [ 0, 100 ], "higher_is_better": false, "multimodal_input": false, "tools": "real-repository coding environment", "sampling": "pass@1", "judge": "anonymous developer preference or multidimensional human rating", "harness": "Seed ClaudeCode/Trae crowdsourced evaluation", "notes": "Official Seed2.1 model-card internal evaluation." } }, { "id": "seed21_claudecode_opus46_net_win_rate", "name": "Seed2.1 ClaudeCode vs Claude Opus 4.6 (net win rate)", "category": "Human Preference / Internal", "metric": "net win rate (percentage points)", "num_problems": 230, "source_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2.1/Seed2_1_Model_Card.pdf", "canonical_setting": { "version": "Seed2.1 ClaudeCode vs Claude Opus 4.6 (net win rate)", "metric_type": "score", "range": null, "higher_is_better": true, "multimodal_input": false, "tools": "real-repository coding environment", "sampling": "pass@1", "judge": "anonymous developer preference or multidimensional human rating", "harness": "Seed ClaudeCode/Trae crowdsourced evaluation", "notes": "Official Seed2.1 model-card internal evaluation." } }, { "id": "seed21_trae_opus47_preference_win_rate", "name": "Seed2.1 Trae vs Claude Opus 4.7 (preference win rate)", "category": "Human Preference / Internal", "metric": "Seed model preference win rate (%)", "num_problems": 167, "source_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2.1/Seed2_1_Model_Card.pdf", "canonical_setting": { "version": "Seed2.1 Trae vs Claude Opus 4.7 (preference win rate)", "metric_type": "pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": false, "tools": "real-repository coding environment", "sampling": "pass@1", "judge": "anonymous developer preference or multidimensional human rating", "harness": "Seed ClaudeCode/Trae crowdsourced evaluation", "notes": "Official Seed2.1 model-card internal evaluation." } }, { "id": "seed21_trae_mean_rating", "name": "Seed2.1 Trae mean rating", "category": "Human Preference / Internal", "metric": "mean rating across six dimensions", "num_problems": 167, "source_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2.1/Seed2_1_Model_Card.pdf", "canonical_setting": { "version": "Seed2.1 Trae mean rating", "metric_type": "score", "range": null, "higher_is_better": true, "multimodal_input": false, "tools": "real-repository coding environment", "sampling": "pass@1", "judge": "anonymous developer preference or multidimensional human rating", "harness": "Seed ClaudeCode/Trae crowdsourced evaluation", "notes": "Official Seed2.1 model-card internal evaluation." } }, { "id": "seed21_trae_fully_correct_rate", "name": "Seed2.1 Trae fully correct artifact rate", "category": "Human Preference / Internal", "metric": "fully correct ready-to-use rate (%)", "num_problems": 167, "source_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2.1/Seed2_1_Model_Card.pdf", "canonical_setting": { "version": "Seed2.1 Trae fully correct artifact rate", "metric_type": "pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": false, "tools": "real-repository coding environment", "sampling": "pass@1", "judge": "anonymous developer preference or multidimensional human rating", "harness": "Seed ClaudeCode/Trae crowdsourced evaluation", "notes": "Official Seed2.1 model-card internal evaluation." } }, { "id": "seed21_trae_acceptable_delivery_rate", "name": "Seed2.1 Trae acceptable delivery rate", "category": "Human Preference / Internal", "metric": "acceptable delivery rate (%)", "num_problems": 167, "source_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2.1/Seed2_1_Model_Card.pdf", "canonical_setting": { "version": "Seed2.1 Trae acceptable delivery rate", "metric_type": "pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": false, "tools": "real-repository coding environment", "sampling": "pass@1", "judge": "anonymous developer preference or multidimensional human rating", "harness": "Seed ClaudeCode/Trae crowdsourced evaluation", "notes": "Official Seed2.1 model-card internal evaluation." } }, { "id": "seed21_trae_severely_broken_rate", "name": "Seed2.1 Trae severely broken delivery rate", "category": "Human Preference / Internal", "metric": "severely broken delivery rate (%)", "num_problems": 167, "source_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2.1/Seed2_1_Model_Card.pdf", "canonical_setting": { "version": "Seed2.1 Trae severely broken delivery rate", "metric_type": "pct", "range": [ 0, 100 ], "higher_is_better": false, "multimodal_input": false, "tools": "real-repository coding environment", "sampling": "pass@1", "judge": "anonymous developer preference or multidimensional human rating", "harness": "Seed ClaudeCode/Trae crowdsourced evaluation", "notes": "Official Seed2.1 model-card internal evaluation." } }, { "id": "seed21_trae_delivery_completeness", "name": "Seed2.1 Trae delivery completeness", "category": "Human Preference / Internal", "metric": "delivery completeness score", "num_problems": 167, "source_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2.1/Seed2_1_Model_Card.pdf", "canonical_setting": { "version": "Seed2.1 Trae delivery completeness", "metric_type": "score", "range": null, "higher_is_better": true, "multimodal_input": false, "tools": "real-repository coding environment", "sampling": "pass@1", "judge": "anonymous developer preference or multidimensional human rating", "harness": "Seed ClaudeCode/Trae crowdsourced evaluation", "notes": "Official Seed2.1 model-card internal evaluation." } }, { "id": "seed21_trae_fully_usable_rate", "name": "Seed2.1 Trae fully usable rate", "category": "Human Preference / Internal", "metric": "fully usable rate (%)", "num_problems": 167, "source_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2.1/Seed2_1_Model_Card.pdf", "canonical_setting": { "version": "Seed2.1 Trae fully usable rate", "metric_type": "pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": false, "tools": "real-repository coding environment", "sampling": "pass@1", "judge": "anonymous developer preference or multidimensional human rating", "harness": "Seed ClaudeCode/Trae crowdsourced evaluation", "notes": "Official Seed2.1 model-card internal evaluation." } }, { "id": "seed21_trae_unusable_rate", "name": "Seed2.1 Trae unusable rate", "category": "Human Preference / Internal", "metric": "unusable rate (%)", "num_problems": 167, "source_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2.1/Seed2_1_Model_Card.pdf", "canonical_setting": { "version": "Seed2.1 Trae unusable rate", "metric_type": "pct", "range": [ 0, 100 ], "higher_is_better": false, "multimodal_input": false, "tools": "real-repository coding environment", "sampling": "pass@1", "judge": "anonymous developer preference or multidimensional human rating", "harness": "Seed ClaudeCode/Trae crowdsourced evaluation", "notes": "Official Seed2.1 model-card internal evaluation." } }, { "id": "seed21_trae_instruction_following", "name": "Seed2.1 Trae instruction-following score", "category": "Human Preference / Internal", "metric": "instruction-following score", "num_problems": 167, "source_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2.1/Seed2_1_Model_Card.pdf", "canonical_setting": { "version": "Seed2.1 Trae instruction-following score", "metric_type": "score", "range": null, "higher_is_better": true, "multimodal_input": false, "tools": "real-repository coding environment", "sampling": "pass@1", "judge": "anonymous developer preference or multidimensional human rating", "harness": "Seed ClaudeCode/Trae crowdsourced evaluation", "notes": "Official Seed2.1 model-card internal evaluation." } }, { "id": "seed21_trae_boundary_adherence", "name": "Seed2.1 Trae boundary-adherence score", "category": "Human Preference / Internal", "metric": "boundary-adherence score", "num_problems": 167, "source_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2.1/Seed2_1_Model_Card.pdf", "canonical_setting": { "version": "Seed2.1 Trae boundary-adherence score", "metric_type": "score", "range": null, "higher_is_better": true, "multimodal_input": false, "tools": "real-repository coding environment", "sampling": "pass@1", "judge": "anonymous developer preference or multidimensional human rating", "harness": "Seed ClaudeCode/Trae crowdsourced evaluation", "notes": "Official Seed2.1 model-card internal evaluation." } }, { "id": "seed21_dirtyfilter_validation_recall", "name": "Seed2.1 DirtyFilter validation recall", "category": "Data Cleaning Agent", "metric": "validation recall (%)", "num_problems": 8037, "source_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2.1/Seed2_1_Model_Card.pdf", "canonical_setting": { "version": "Seed2.1 DirtyFilter validation recall", "metric_type": "pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": false, "tools": "rule-only autonomous evaluator/optimizer loop", "sampling": "pass@1", "judge": "validation-set labels", "harness": "300M-token autonomous rule-iteration budget", "notes": "Figure 19; all models use the same harness and validation set." } }, { "id": "seed21_dirtyfilter_precision", "name": "Seed2.1 DirtyFilter precision", "category": "Data Cleaning Agent", "metric": "precision (%)", "num_problems": 8037, "source_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2.1/Seed2_1_Model_Card.pdf", "canonical_setting": { "version": "Seed2.1 DirtyFilter precision", "metric_type": "pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": false, "tools": "rule-only autonomous evaluator/optimizer loop", "sampling": "pass@1", "judge": "validation-set labels", "harness": "300M-token autonomous rule-iteration budget", "notes": "Figure 19; all models use the same harness and validation set." } }, { "id": "atlas_mrcr_8needle_auc_1m", "name": "ATLAS MRCR 8-Needle AUC through 1M", "category": "Long Context", "metric": "normalized length AUC of exact match (%)", "num_problems": 792, "source_url": "https://arxiv.org/html/2605.28079", "canonical_setting": { "version": "ATLAS full eight-slice MRCR 8-needle through 1M", "metric_type": "pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": false, "tools": "none", "sampling": "one model response per instance", "judge": "deterministic benchmark-native evaluator", "harness": "ATLAS full eight-slice evaluation through 1M", "notes": "Eight slices from 8K through 1M with 106/96/98/100/100/100/100/92 instances. Distinct from a pointwise or unweighted MRCR aggregate." } }, { "id": "atlas_oolong_synth_auc_1m", "name": "ATLAS OOLong-Synth AUC through 1M", "category": "Long Context", "metric": "normalized length AUC of answer-level score (%)", "num_problems": 800, "source_url": "https://arxiv.org/html/2605.28079", "canonical_setting": { "version": "ATLAS full eight-slice OOLong-Synth through 1M", "metric_type": "pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": false, "tools": "none", "sampling": "one model response per instance", "judge": "deterministic benchmark-native evaluator", "harness": "ATLAS full eight-slice evaluation through 1M", "notes": "100 instances at each of 8K, 16K, 32K, 64K, 128K, 256K, 512K and 1M; benchmark-native categorical/date/numeric/frequency answer scoring." } }, { "id": "atlas_graphwalks_extend_auc_1m", "name": "ATLAS GraphWalks Extend AUC through 1M", "category": "Long Context", "metric": "normalized length AUC of node-set F1 (%)", "num_problems": 800, "source_url": "https://arxiv.org/html/2605.28079", "canonical_setting": { "version": "ATLAS full eight-slice GraphWalks Extend through 1M", "metric_type": "pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": false, "tools": "none", "sampling": "one model response per instance", "judge": "deterministic benchmark-native evaluator", "harness": "ATLAS full eight-slice evaluation through 1M", "notes": "Official 128K and 1M data plus six ATLAS-generated slices using the official BFS/parent-node generation procedure." } }, { "id": "atlas_loft_text_retrieval_extend_auc_1m", "name": "ATLAS LOFT Text Retrieval Extend AUC through 1M", "category": "Long Context", "metric": "normalized length AUC of MRecall@K (%)", "num_problems": 800, "source_url": "https://arxiv.org/html/2605.28079", "canonical_setting": { "version": "ATLAS full eight-slice LOFT Text Retrieval Extend through 1M", "metric_type": "pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": false, "tools": "none", "sampling": "one model response per instance", "judge": "deterministic benchmark-native evaluator", "harness": "ATLAS full eight-slice evaluation through 1M", "notes": "LOFT retrieval plus HELMET-RAG evidence grounding. Remaining length slices are obtained by downsampling the 1M retrieval subset." } }, { "id": "atlas_helmet_icl_extend_auc_1m", "name": "ATLAS HELMET-ICL Extend AUC through 1M", "category": "Long Context", "metric": "normalized length AUC of classification accuracy (%)", "num_problems": 800, "source_url": "https://arxiv.org/html/2605.28079", "canonical_setting": { "version": "ATLAS full eight-slice HELMET-ICL Extend through 1M", "metric_type": "pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": false, "tools": "none", "sampling": "one model response per instance", "judge": "deterministic benchmark-native evaluator", "harness": "ATLAS full eight-slice evaluation through 1M", "notes": "Official HELMET construction scripts generate all ATLAS slices." } }, { "id": "atlas_longcodeqa_auc_1m", "name": "ATLAS LongCodeQA AUC through 1M", "category": "Long Context", "metric": "normalized length AUC of exact option accuracy (%)", "num_problems": null, "source_url": "https://arxiv.org/html/2605.28079", "canonical_setting": { "version": "ATLAS LongCodeQA six-slice evaluation from 32K through 1M", "metric_type": "pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": false, "tools": "none", "sampling": "one model response per instance", "judge": "deterministic benchmark-native evaluator", "harness": "ATLAS six-slice evaluation from 32K through 1M", "notes": "Table 15 reports the LongCodeQA subtask. ATLAS describes it as exact matching of the predicted option letter. Exact LongCodeQA counts are not separated from LongSWE in the published LongCodeBench composite counts." } }, { "id": "atlas_amembench_acu_auc_1m", "name": "ATLAS AMemBench-ACU AUC through 1M", "category": "Long Context", "metric": "normalized length AUC of QPEM (%)", "num_problems": 800, "source_url": "https://arxiv.org/html/2605.28079", "canonical_setting": { "version": "ATLAS full eight-slice AMemBench-ACU through 1M", "metric_type": "pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": false, "tools": "none", "sampling": "one model response per instance", "judge": "deterministic benchmark-native evaluator", "harness": "ATLAS full eight-slice evaluation through 1M", "notes": "100 static transcript instances per slice; quasi-prefix exact match after normalization." } }, { "id": "biology_instructions", "name": "Biology-Instructions", "category": "Science", "metric": "source-reported aggregate score (%)", "num_problems": null, "source_url": "https://huggingface.co/datasets/SciReason/bio_instruction/tree/c536cf1a0727baba3c42c79972b8a64bb08c5488", "canonical_setting": { "version": "SciReason/bio_instruction at c536cf1a0727baba3c42c79972b8a64bb08c5488", "metric_type": "pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": false, "tools": "none", "sampling": "pass@1/source-defined", "judge": "task-specific metrics", "notes": "Multi-omics task bundle across DNA, RNA, protein, and mixed tasks. Registry includes MCC, PCC, Spearman, Fmax, accuracy, AUC, R², and mixed scores. One aggregate formula and one non-overlapping item count are not published in the score source; keep num_problems null." } }, { "id": "mol_instructions", "name": "Mol-Instructions", "category": "Science", "metric": "source-reported aggregate score (%)", "num_problems": null, "source_url": "https://huggingface.co/datasets/SciReason/Mol-Instructions-test/tree/a581cc374ec90be8d808acf37d788f1f65d0395b", "canonical_setting": { "version": "SciReason/Mol-Instructions-test at a581cc374ec90be8d808acf37d788f1f65d0395b", "metric_type": "pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": false, "tools": "none", "sampling": "pass@1/source-defined", "judge": "task-specific exact/semantic metrics", "notes": "Heterogeneous molecular and protein instruction suite. Source table does not state a single aggregate formula or non-overlapping total; keep num_problems null." } }, { "id": "moleculariq", "name": "MolecularIQ", "category": "Science", "metric": "aggregate accuracy (%)", "num_problems": 5111, "source_url": "https://huggingface.co/datasets/ml-jku/moleculariq-v0.0/tree/aa3d7c6c2a67c20977f3fefa5169e65829166450", "canonical_setting": { "version": "moleculariq-v0.0 test split", "metric_type": "pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": false, "tools": "none", "sampling": "pass@1", "judge": "task-specific rule-based scoring", "notes": "Official dataset card reports a 5,111-example test split. Additional task-specific splits overlap/derive from the benchmark and are not added to the canonical count." } }, { "id": "scireasoner", "name": "SciReasoner", "category": "Science", "metric": "source-reported aggregate score (%)", "num_problems": null, "source_url": "https://arxiv.org/abs/2509.21320v3", "canonical_setting": { "version": "SciReasoner evaluation suite in arXiv:2509.21320v3", "metric_type": "pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": false, "tools": "none", "sampling": "source-defined", "judge": "task-specific scientific reasoning metrics", "notes": "Broad suite spanning chemistry, biology, materials, sequence, prediction, classification, generation, and extraction. Examined official sources do not publish one aggregate item count/formula; keep num_problems null." } }, { "id": "hle_multimodal", "name": "HLE Multimodal", "category": "Multimodal", "metric": "% correct", "num_problems": 342, "source_url": "https://huggingface.co/datasets/cais/hle/tree/5a81a4c7271a2a2a312b9a690f0c2fde837e4c29", "canonical_setting": { "version": "Multimodal-only subset of finalized 2,500-question HLE", "metric_type": "pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": true, "tools": "none", "sampling": "pass@1", "judge": "official HLE answer evaluation", "notes": "Count is 2,500 finalized HLE minus the campaign-established 2,158 text-only rows = 342. Must not reuse hle_vl, which is a different search-enabled setting." } }, { "id": "lmarena_search_elo", "name": "LMArena Search Arena Elo", "category": "Search Agent", "metric": "Arena score", "num_problems": null, "source_url": "https://arena.ai/leaderboard/search", "canonical_setting": { "version": "LMArena Search live Arena score", "metric_type": "elo", "range": null, "higher_is_better": true, "multimodal_input": false, "tools": "search/grounding environment varies by model", "judge": "human pairwise preference votes", "style_control": "off", "sampling": "live pairwise battles", "notes": "Dated live-leaderboard score, distinct from text-only chatbot_arena_elo. No fixed static item set." } }, { "id": "aa_lcr_mistral_custom_gpt_4_1_mini_middle_out", "name": "AA Long Context Reasoning (Mistral custom)", "category": "Long Context", "metric": "% correct", "num_problems": 100, "source_url": "https://huggingface.co/mistralai/Mistral-Small-4-119B-2603/blob/a11f36bebf709121056b1dbcc943d1c6afbe494d/README.md", "canonical_setting": { "version": "Mistral Small 4 custom implementation of AA-LCR", "metric_type": "pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": false, "tools": "none", "sampling": "pass@1; repeat count not disclosed by Mistral", "judge": "gpt-4.1-mini-2025-04-14 equality judge", "harness": "Mistral custom AA-LCR implementation", "prompt_style": "AA-LCR 100 hard open-answer questions over approximately 100k-token inputs; exact Mistral prompt undisclosed", "temperature": "undisclosed", "context_handling": "middle-out for models with shorter context lengths", "notes": "Distinct from campaign aa_lcr: Mistral changes the judge and adds middle-out context handling. The underlying AA-LCR set has 100 questions; Mistral does not disclose repeats or total generations." } }, { "id": "aa_omniscience_index", "name": "AA Omniscience Index", "category": "Factuality", "metric": "index (-100 to 100)", "num_problems": 60000, "source_url": "https://artificialanalysis.ai/evaluations/omniscience", "canonical_setting": { "version": "AA-Omniscience 6,000 questions x 10 runs", "metric_type": "index", "range": [ -100, 100 ], "higher_is_better": true, "multimodal_input": false, "notes": "Rewards correct answers and penalizes hallucinations." } }, { "id": "aa_omniscience_public_index", "name": "AA Omniscience Public Index", "category": "Factuality", "metric": "index (-100 to 100)", "num_problems": null, "source_url": "https://www.liquid.ai/blog/lfm2-5-2-6b", "canonical_setting": { "version": "AA-Omniscience-Public", "metric_type": "index", "range": [ -100, 100 ], "higher_is_better": true, "multimodal_input": false, "notes": "Public subset/index reported by Liquid AI; item count is not stated." } }, { "id": "aa_omniscience_public_accuracy", "name": "AA Omniscience Public Accuracy", "category": "Factuality", "metric": "% correct", "num_problems": null, "source_url": "https://huggingface.co/LiquidAI/LFM2.5-2.6B", "canonical_setting": { "version": "AA-Omniscience-Public", "metric_type": "pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": false, "notes": "Public-subset accuracy reported by Liquid AI." } }, { "id": "aa_omniscience_public_non_hallucination", "name": "AA Omniscience Public Non-Hallucination", "category": "Hallucination", "metric": "% non-hallucination", "num_problems": null, "source_url": "https://huggingface.co/LiquidAI/LFM2.5-2.6B", "canonical_setting": { "version": "AA-Omniscience-Public", "metric_type": "pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": false, "notes": "Public-subset non-hallucination rate reported by Liquid AI." } }, { "id": "ifstruct_v1", "name": "IFStruct v1.0", "category": "Instruction Following", "metric": "% binary structural compliance", "num_problems": 2000, "source_url": "https://huggingface.co/datasets/LiquidAI/ifstruct-v1.0", "canonical_setting": { "version": "IFStruct v1.0 frozen public test set", "metric_type": "pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": false, "notes": "No constrained decoding; JSON/YAML structure and schema only." } }, { "id": "tool_sandbox", "name": "ToolSandbox", "category": "Tool Use", "metric": "% aggregate milestone similarity", "num_problems": null, "source_url": "https://github.com/apple/ToolSandbox", "canonical_setting": { "version": "ToolSandbox arXiv:2408.04682v2", "metric_type": "pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": false, "notes": "Stateful conversational tool use with milestone/minefield scoring." } }, { "id": "claw_eval_en_average", "name": "Claw-Eval English Average", "category": "Agentic", "metric": "Pass^3 average (%)", "num_problems": 179, "source_url": "https://huggingface.co/datasets/claw-eval/Claw-Eval", "canonical_setting": { "version": "Claw-Eval public English tasks", "metric_type": "pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": false, "notes": "101 general, 72 multimodal, and 6 multi-turn English tasks; N=3." } }, { "id": "browsecomp_plus_openclaw", "name": "BrowseComp-Plus (OpenClaw)", "category": "Agentic Search", "metric": "% correct", "num_problems": 830, "source_url": "https://huggingface.co/datasets/Tevatron/browsecomp-plus", "canonical_setting": { "version": "BrowseComp-Plus fixed-corpus test set", "metric_type": "pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": false, "notes": "OpenClaw agent over the fixed approximately 100K-document corpus." } }, { "id": "air_bench_2024", "name": "AIR-Bench 2024", "category": "Safety", "metric": "safe-engagement score (%)", "num_problems": null, "source_url": "https://arxiv.org/abs/2407.17436", "canonical_setting": { "version": "AIR-Bench 2024 aggregate", "metric_type": "pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": false, "notes": "Category-specific LLM judges score policy-grounded safe engagement across regulatory and policy-derived harms." } }, { "id": "cyberseceval4_instruct", "name": "CyberSecEval 4 Instruct", "category": "Safety", "metric": "secure-code score (%)", "num_problems": null, "source_url": "https://github.com/meta-llama/PurpleLlama", "canonical_setting": { "version": "CyberSecEval 4 insecure-code-generation Instruct", "metric_type": "pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": false, "notes": "Static-analysis evaluation of responses to coding requests designed to elicit known insecure patterns." } }, { "id": "cyberseceval4_autocomplete", "name": "CyberSecEval 4 Autocomplete", "category": "Safety", "metric": "secure-code score (%)", "num_problems": null, "source_url": "https://github.com/meta-llama/PurpleLlama", "canonical_setting": { "version": "CyberSecEval 4 insecure-code-generation Autocomplete", "metric_type": "pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": false, "notes": "Static-analysis evaluation of code completions where context leads up to a known insecure pattern." } }, { "id": "longfact_claim_precision", "name": "LongFact Claim Precision", "category": "Factuality", "metric": "claim-level precision (%)", "num_problems": null, "source_url": "https://arxiv.org/abs/2403.18802", "canonical_setting": { "version": "LongFact prompts with simplified claim extraction", "metric_type": "pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": false, "notes": "Uses the LongFact prompt set and the simplified claim extraction plus LLM-judge protocol described by Microsoft." } }, { "id": "corpusqa_gpt54_judge", "name": "CorpusQA with GPT-5.4 Judge", "category": "Long Context", "metric": "AI-judge score (%)", "num_problems": 1316, "source_url": "https://arxiv.org/abs/2601.14952", "canonical_setting": { "version": "CorpusQA with Microsoft GPT-5.4-high judge", "metric_type": "pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": false, "notes": "1,316 multi-document free-form QA instances. Microsoft replaces the default DeepSeek-V3 judge with GPT-5.4 high." } }, { "id": "amo_bench", "name": "AMO Bench", "category": "Math", "metric": "accuracy (%)", "num_problems": null, "source_url": "https://microsoft.ai/pdf/MAI-Code-1-Flash-Model-Card.PDF", "canonical_setting": { "version": "MAI-Code-1-Flash model-card release evaluation", "metric_type": "pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": false, "notes": "Olympiad-math benchmark; public task count is not stated." } }, { "id": "advancedif_rubric_level", "name": "AdvancedIF Rubric-Level Score", "category": "Instruction Following", "metric": "rubric-level score (%)", "num_problems": 1645, "source_url": "https://arxiv.org/abs/2511.10507", "canonical_setting": { "version": "AdvancedIF public test split; rubric-level aggregation", "metric_type": "pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": false, "notes": "Microsoft explicitly reports rubric-level scores, distinct from the all-rubrics-pass benchmark primary." } }, { "id": "advancedif_average", "name": "AdvancedIF Average", "category": "Instruction Following", "metric": "source-reported average score (%)", "num_problems": 1645, "source_url": "https://arxiv.org/abs/2511.10507", "canonical_setting": { "version": "AdvancedIF public test split; source-reported average", "metric_type": "pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": false, "notes": "MAI-Code-1-Flash chart labels this metric as an average, not the all-rubrics-pass primary." } }, { "id": "ifbench_single_multiturn_average", "name": "IFBench Single/Multi-Turn Average", "category": "Instruction Following", "metric": "single/multi-turn average score (%)", "num_problems": 1687, "source_url": "https://arxiv.org/abs/2507.02833", "canonical_setting": { "version": "IFBench single-turn and multi-turn average", "metric_type": "pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": false, "notes": "300 single-turn plus 1,387 multi-turn examples. MAI-Code-1-Flash reports the average of the two aggregate evaluation scores." } }, { "id": "truthfulqa_mc", "name": "TruthfulQA Multiple Choice", "category": "Factuality", "metric": "multiple-choice accuracy (%)", "num_problems": 817, "source_url": "https://github.com/sylinrl/TruthfulQA", "canonical_setting": { "version": "TruthfulQA recommended multiple-choice setting", "metric_type": "pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": false, "notes": "Distinct from the generation benchmark already registered as truthfulqa." } }, { "id": "longbench_v2_256k", "name": "LongBench-V2 256K Subset", "category": "Long Context", "metric": "multiple-choice accuracy (%)", "num_problems": 408, "source_url": "https://huggingface.co/datasets/THUDM/LongBench-v2", "canonical_setting": { "version": "LongBench-V2 questions fitting within 256K input context", "metric_type": "pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": false, "notes": "Microsoft follows the official setup but limits input context to 256K, leaving 408 unique questions." } }, { "id": "graphwalks_bfs_bucket_0_4k", "name": "GraphWalks BFS bucket 0–4K", "category": "Long Context", "metric": "% F1", "num_problems": null, "source_url": "https://huggingface.co/datasets/openai/graphwalks/resolve/f338bb265735a56a79f4b0f5def722c9c3268ead/README.md", "canonical_setting": { "version": "GraphWalks post-2026-02-27 ground-truth/prompt fix; BFS exact context bucket 0–4K", "metric_type": "pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": false, "tools": "none", "sampling": "one response per graph prompt", "judge": "deterministic set-overlap F1 against answer node list", "notes": "Exact bucket identity from the Xiaomi embedded JavaScript data and the immutable official GraphWalks README after the 2026-02-27 parents-ground-truth and BFS-prompt fixes. The source reports cell-specific n values, which are preserved in each observation setting rather than inventing a benchmark-wide count." } }, { "id": "graphwalks_bfs_bucket_4k_8k", "name": "GraphWalks BFS bucket 4K–8K", "category": "Long Context", "metric": "% F1", "num_problems": null, "source_url": "https://huggingface.co/datasets/openai/graphwalks/resolve/f338bb265735a56a79f4b0f5def722c9c3268ead/README.md", "canonical_setting": { "version": "GraphWalks post-2026-02-27 ground-truth/prompt fix; BFS exact context bucket 4K–8K", "metric_type": "pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": false, "tools": "none", "sampling": "one response per graph prompt", "judge": "deterministic set-overlap F1 against answer node list", "notes": "Exact bucket identity from the Xiaomi embedded JavaScript data and the immutable official GraphWalks README after the 2026-02-27 parents-ground-truth and BFS-prompt fixes. The source reports cell-specific n values, which are preserved in each observation setting rather than inventing a benchmark-wide count." } }, { "id": "graphwalks_bfs_bucket_8k_16k", "name": "GraphWalks BFS bucket 8K–16K", "category": "Long Context", "metric": "% F1", "num_problems": null, "source_url": "https://huggingface.co/datasets/openai/graphwalks/resolve/f338bb265735a56a79f4b0f5def722c9c3268ead/README.md", "canonical_setting": { "version": "GraphWalks post-2026-02-27 ground-truth/prompt fix; BFS exact context bucket 8K–16K", "metric_type": "pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": false, "tools": "none", "sampling": "one response per graph prompt", "judge": "deterministic set-overlap F1 against answer node list", "notes": "Exact bucket identity from the Xiaomi embedded JavaScript data and the immutable official GraphWalks README after the 2026-02-27 parents-ground-truth and BFS-prompt fixes. The source reports cell-specific n values, which are preserved in each observation setting rather than inventing a benchmark-wide count." } }, { "id": "graphwalks_bfs_bucket_16k_32k", "name": "GraphWalks BFS bucket 16K–32K", "category": "Long Context", "metric": "% F1", "num_problems": null, "source_url": "https://huggingface.co/datasets/openai/graphwalks/resolve/f338bb265735a56a79f4b0f5def722c9c3268ead/README.md", "canonical_setting": { "version": "GraphWalks post-2026-02-27 ground-truth/prompt fix; BFS exact context bucket 16K–32K", "metric_type": "pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": false, "tools": "none", "sampling": "one response per graph prompt", "judge": "deterministic set-overlap F1 against answer node list", "notes": "Exact bucket identity from the Xiaomi embedded JavaScript data and the immutable official GraphWalks README after the 2026-02-27 parents-ground-truth and BFS-prompt fixes. The source reports cell-specific n values, which are preserved in each observation setting rather than inventing a benchmark-wide count." } }, { "id": "graphwalks_bfs_bucket_32k_64k", "name": "GraphWalks BFS bucket 32K–64K", "category": "Long Context", "metric": "% F1", "num_problems": null, "source_url": "https://huggingface.co/datasets/openai/graphwalks/resolve/f338bb265735a56a79f4b0f5def722c9c3268ead/README.md", "canonical_setting": { "version": "GraphWalks post-2026-02-27 ground-truth/prompt fix; BFS exact context bucket 32K–64K", "metric_type": "pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": false, "tools": "none", "sampling": "one response per graph prompt", "judge": "deterministic set-overlap F1 against answer node list", "notes": "Exact bucket identity from the Xiaomi embedded JavaScript data and the immutable official GraphWalks README after the 2026-02-27 parents-ground-truth and BFS-prompt fixes. The source reports cell-specific n values, which are preserved in each observation setting rather than inventing a benchmark-wide count." } }, { "id": "graphwalks_bfs_bucket_64k_128k", "name": "GraphWalks BFS bucket 64K–128K", "category": "Long Context", "metric": "% F1", "num_problems": null, "source_url": "https://huggingface.co/datasets/openai/graphwalks/resolve/f338bb265735a56a79f4b0f5def722c9c3268ead/README.md", "canonical_setting": { "version": "GraphWalks post-2026-02-27 ground-truth/prompt fix; BFS exact context bucket 64K–128K", "metric_type": "pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": false, "tools": "none", "sampling": "one response per graph prompt", "judge": "deterministic set-overlap F1 against answer node list", "notes": "Exact bucket identity from the Xiaomi embedded JavaScript data and the immutable official GraphWalks README after the 2026-02-27 parents-ground-truth and BFS-prompt fixes. The source reports cell-specific n values, which are preserved in each observation setting rather than inventing a benchmark-wide count." } }, { "id": "graphwalks_bfs_bucket_128k_256k", "name": "GraphWalks BFS bucket 128K–256K", "category": "Long Context", "metric": "% F1", "num_problems": null, "source_url": "https://huggingface.co/datasets/openai/graphwalks/resolve/f338bb265735a56a79f4b0f5def722c9c3268ead/README.md", "canonical_setting": { "version": "GraphWalks post-2026-02-27 ground-truth/prompt fix; BFS exact context bucket 128K–256K", "metric_type": "pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": false, "tools": "none", "sampling": "one response per graph prompt", "judge": "deterministic set-overlap F1 against answer node list", "notes": "Exact bucket identity from the Xiaomi embedded JavaScript data and the immutable official GraphWalks README after the 2026-02-27 parents-ground-truth and BFS-prompt fixes. The source reports cell-specific n values, which are preserved in each observation setting rather than inventing a benchmark-wide count." } }, { "id": "graphwalks_bfs_bucket_256k_512k", "name": "GraphWalks BFS bucket 256K–512K", "category": "Long Context", "metric": "% F1", "num_problems": null, "source_url": "https://huggingface.co/datasets/openai/graphwalks/resolve/f338bb265735a56a79f4b0f5def722c9c3268ead/README.md", "canonical_setting": { "version": "GraphWalks post-2026-02-27 ground-truth/prompt fix; BFS exact context bucket 256K–512K", "metric_type": "pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": false, "tools": "none", "sampling": "one response per graph prompt", "judge": "deterministic set-overlap F1 against answer node list", "notes": "Exact bucket identity from the Xiaomi embedded JavaScript data and the immutable official GraphWalks README after the 2026-02-27 parents-ground-truth and BFS-prompt fixes. The source reports cell-specific n values, which are preserved in each observation setting rather than inventing a benchmark-wide count." } }, { "id": "graphwalks_parents_bucket_0_4k", "name": "GraphWalks Parents bucket 0–4K", "category": "Long Context", "metric": "% F1", "num_problems": null, "source_url": "https://huggingface.co/datasets/openai/graphwalks/resolve/f338bb265735a56a79f4b0f5def722c9c3268ead/README.md", "canonical_setting": { "version": "GraphWalks post-2026-02-27 ground-truth/prompt fix; Parents exact context bucket 0–4K", "metric_type": "pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": false, "tools": "none", "sampling": "one response per graph prompt", "judge": "deterministic set-overlap F1 against answer node list", "notes": "Exact bucket identity from the Xiaomi embedded JavaScript data and the immutable official GraphWalks README after the 2026-02-27 parents-ground-truth and BFS-prompt fixes. The source reports cell-specific n values, which are preserved in each observation setting rather than inventing a benchmark-wide count." } }, { "id": "graphwalks_parents_bucket_4k_8k", "name": "GraphWalks Parents bucket 4K–8K", "category": "Long Context", "metric": "% F1", "num_problems": null, "source_url": "https://huggingface.co/datasets/openai/graphwalks/resolve/f338bb265735a56a79f4b0f5def722c9c3268ead/README.md", "canonical_setting": { "version": "GraphWalks post-2026-02-27 ground-truth/prompt fix; Parents exact context bucket 4K–8K", "metric_type": "pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": false, "tools": "none", "sampling": "one response per graph prompt", "judge": "deterministic set-overlap F1 against answer node list", "notes": "Exact bucket identity from the Xiaomi embedded JavaScript data and the immutable official GraphWalks README after the 2026-02-27 parents-ground-truth and BFS-prompt fixes. The source reports cell-specific n values, which are preserved in each observation setting rather than inventing a benchmark-wide count." } }, { "id": "graphwalks_parents_bucket_8k_16k", "name": "GraphWalks Parents bucket 8K–16K", "category": "Long Context", "metric": "% F1", "num_problems": null, "source_url": "https://huggingface.co/datasets/openai/graphwalks/resolve/f338bb265735a56a79f4b0f5def722c9c3268ead/README.md", "canonical_setting": { "version": "GraphWalks post-2026-02-27 ground-truth/prompt fix; Parents exact context bucket 8K–16K", "metric_type": "pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": false, "tools": "none", "sampling": "one response per graph prompt", "judge": "deterministic set-overlap F1 against answer node list", "notes": "Exact bucket identity from the Xiaomi embedded JavaScript data and the immutable official GraphWalks README after the 2026-02-27 parents-ground-truth and BFS-prompt fixes. The source reports cell-specific n values, which are preserved in each observation setting rather than inventing a benchmark-wide count." } }, { "id": "graphwalks_parents_bucket_16k_32k", "name": "GraphWalks Parents bucket 16K–32K", "category": "Long Context", "metric": "% F1", "num_problems": null, "source_url": "https://huggingface.co/datasets/openai/graphwalks/resolve/f338bb265735a56a79f4b0f5def722c9c3268ead/README.md", "canonical_setting": { "version": "GraphWalks post-2026-02-27 ground-truth/prompt fix; Parents exact context bucket 16K–32K", "metric_type": "pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": false, "tools": "none", "sampling": "one response per graph prompt", "judge": "deterministic set-overlap F1 against answer node list", "notes": "Exact bucket identity from the Xiaomi embedded JavaScript data and the immutable official GraphWalks README after the 2026-02-27 parents-ground-truth and BFS-prompt fixes. The source reports cell-specific n values, which are preserved in each observation setting rather than inventing a benchmark-wide count." } }, { "id": "graphwalks_parents_bucket_32k_64k", "name": "GraphWalks Parents bucket 32K–64K", "category": "Long Context", "metric": "% F1", "num_problems": null, "source_url": "https://huggingface.co/datasets/openai/graphwalks/resolve/f338bb265735a56a79f4b0f5def722c9c3268ead/README.md", "canonical_setting": { "version": "GraphWalks post-2026-02-27 ground-truth/prompt fix; Parents exact context bucket 32K–64K", "metric_type": "pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": false, "tools": "none", "sampling": "one response per graph prompt", "judge": "deterministic set-overlap F1 against answer node list", "notes": "Exact bucket identity from the Xiaomi embedded JavaScript data and the immutable official GraphWalks README after the 2026-02-27 parents-ground-truth and BFS-prompt fixes. The source reports cell-specific n values, which are preserved in each observation setting rather than inventing a benchmark-wide count." } }, { "id": "graphwalks_parents_bucket_64k_128k", "name": "GraphWalks Parents bucket 64K–128K", "category": "Long Context", "metric": "% F1", "num_problems": null, "source_url": "https://huggingface.co/datasets/openai/graphwalks/resolve/f338bb265735a56a79f4b0f5def722c9c3268ead/README.md", "canonical_setting": { "version": "GraphWalks post-2026-02-27 ground-truth/prompt fix; Parents exact context bucket 64K–128K", "metric_type": "pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": false, "tools": "none", "sampling": "one response per graph prompt", "judge": "deterministic set-overlap F1 against answer node list", "notes": "Exact bucket identity from the Xiaomi embedded JavaScript data and the immutable official GraphWalks README after the 2026-02-27 parents-ground-truth and BFS-prompt fixes. The source reports cell-specific n values, which are preserved in each observation setting rather than inventing a benchmark-wide count." } }, { "id": "graphwalks_parents_bucket_128k_256k", "name": "GraphWalks Parents bucket 128K–256K", "category": "Long Context", "metric": "% F1", "num_problems": null, "source_url": "https://huggingface.co/datasets/openai/graphwalks/resolve/f338bb265735a56a79f4b0f5def722c9c3268ead/README.md", "canonical_setting": { "version": "GraphWalks post-2026-02-27 ground-truth/prompt fix; Parents exact context bucket 128K–256K", "metric_type": "pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": false, "tools": "none", "sampling": "one response per graph prompt", "judge": "deterministic set-overlap F1 against answer node list", "notes": "Exact bucket identity from the Xiaomi embedded JavaScript data and the immutable official GraphWalks README after the 2026-02-27 parents-ground-truth and BFS-prompt fixes. The source reports cell-specific n values, which are preserved in each observation setting rather than inventing a benchmark-wide count." } }, { "id": "graphwalks_parents_bucket_256k_512k", "name": "GraphWalks Parents bucket 256K–512K", "category": "Long Context", "metric": "% F1", "num_problems": null, "source_url": "https://huggingface.co/datasets/openai/graphwalks/resolve/f338bb265735a56a79f4b0f5def722c9c3268ead/README.md", "canonical_setting": { "version": "GraphWalks post-2026-02-27 ground-truth/prompt fix; Parents exact context bucket 256K–512K", "metric_type": "pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": false, "tools": "none", "sampling": "one response per graph prompt", "judge": "deterministic set-overlap F1 against answer node list", "notes": "Exact bucket identity from the Xiaomi embedded JavaScript data and the immutable official GraphWalks README after the 2026-02-27 parents-ground-truth and BFS-prompt fixes. The source reports cell-specific n values, which are preserved in each observation setting rather than inventing a benchmark-wide count." } }, { "id": "graphwalks_bfs_32k", "name": "GraphWalks BFS at 32K", "category": "Long Context", "metric": "% F1", "num_problems": null, "source_url": "https://huggingface.co/datasets/openai/graphwalks/resolve/f338bb265735a56a79f4b0f5def722c9c3268ead/README.md", "canonical_setting": { "version": "GraphWalks post-2026-02-27 ground-truth/prompt fix; BFS exact 32K point", "metric_type": "pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": false, "tools": "none", "sampling": "one response per graph prompt", "judge": "deterministic set-overlap F1 against answer node list", "notes": "Exact-length point from the locked Xiaomi line chart. The benchmark definition is the immutable official GraphWalks README after the 2026-02-27 fixes. The chart prints F1 on [0,1]; campaign scores are the same values expressed as percentages." } }, { "id": "graphwalks_bfs_64k", "name": "GraphWalks BFS at 64K", "category": "Long Context", "metric": "% F1", "num_problems": null, "source_url": "https://huggingface.co/datasets/openai/graphwalks/resolve/f338bb265735a56a79f4b0f5def722c9c3268ead/README.md", "canonical_setting": { "version": "GraphWalks post-2026-02-27 ground-truth/prompt fix; BFS exact 64K point", "metric_type": "pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": false, "tools": "none", "sampling": "one response per graph prompt", "judge": "deterministic set-overlap F1 against answer node list", "notes": "Exact-length point from the locked Xiaomi line chart. The benchmark definition is the immutable official GraphWalks README after the 2026-02-27 fixes. The chart prints F1 on [0,1]; campaign scores are the same values expressed as percentages." } }, { "id": "graphwalks_bfs_128k", "name": "GraphWalks BFS at 128K", "category": "Long Context", "metric": "% F1", "num_problems": null, "source_url": "https://huggingface.co/datasets/openai/graphwalks/resolve/f338bb265735a56a79f4b0f5def722c9c3268ead/README.md", "canonical_setting": { "version": "GraphWalks post-2026-02-27 ground-truth/prompt fix; BFS exact 128K point", "metric_type": "pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": false, "tools": "none", "sampling": "one response per graph prompt", "judge": "deterministic set-overlap F1 against answer node list", "notes": "Exact-length point from the locked Xiaomi line chart. The benchmark definition is the immutable official GraphWalks README after the 2026-02-27 fixes. The chart prints F1 on [0,1]; campaign scores are the same values expressed as percentages." } }, { "id": "graphwalks_bfs_512k", "name": "GraphWalks BFS at 512K", "category": "Long Context", "metric": "% F1", "num_problems": null, "source_url": "https://huggingface.co/datasets/openai/graphwalks/resolve/f338bb265735a56a79f4b0f5def722c9c3268ead/README.md", "canonical_setting": { "version": "GraphWalks post-2026-02-27 ground-truth/prompt fix; BFS exact 512K point", "metric_type": "pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": false, "tools": "none", "sampling": "one response per graph prompt", "judge": "deterministic set-overlap F1 against answer node list", "notes": "Exact-length point from the locked Xiaomi line chart. The benchmark definition is the immutable official GraphWalks README after the 2026-02-27 fixes. The chart prints F1 on [0,1]; campaign scores are the same values expressed as percentages." } }, { "id": "graphwalks_parents_32k", "name": "GraphWalks Parents at 32K", "category": "Long Context", "metric": "% F1", "num_problems": null, "source_url": "https://huggingface.co/datasets/openai/graphwalks/resolve/f338bb265735a56a79f4b0f5def722c9c3268ead/README.md", "canonical_setting": { "version": "GraphWalks post-2026-02-27 ground-truth/prompt fix; Parents exact 32K point", "metric_type": "pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": false, "tools": "none", "sampling": "one response per graph prompt", "judge": "deterministic set-overlap F1 against answer node list", "notes": "Exact-length point from the locked Xiaomi line chart. The benchmark definition is the immutable official GraphWalks README after the 2026-02-27 fixes. The chart prints F1 on [0,1]; campaign scores are the same values expressed as percentages." } }, { "id": "graphwalks_parents_64k", "name": "GraphWalks Parents at 64K", "category": "Long Context", "metric": "% F1", "num_problems": null, "source_url": "https://huggingface.co/datasets/openai/graphwalks/resolve/f338bb265735a56a79f4b0f5def722c9c3268ead/README.md", "canonical_setting": { "version": "GraphWalks post-2026-02-27 ground-truth/prompt fix; Parents exact 64K point", "metric_type": "pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": false, "tools": "none", "sampling": "one response per graph prompt", "judge": "deterministic set-overlap F1 against answer node list", "notes": "Exact-length point from the locked Xiaomi line chart. The benchmark definition is the immutable official GraphWalks README after the 2026-02-27 fixes. The chart prints F1 on [0,1]; campaign scores are the same values expressed as percentages." } }, { "id": "graphwalks_parents_128k", "name": "GraphWalks Parents at 128K", "category": "Long Context", "metric": "% F1", "num_problems": null, "source_url": "https://huggingface.co/datasets/openai/graphwalks/resolve/f338bb265735a56a79f4b0f5def722c9c3268ead/README.md", "canonical_setting": { "version": "GraphWalks post-2026-02-27 ground-truth/prompt fix; Parents exact 128K point", "metric_type": "pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": false, "tools": "none", "sampling": "one response per graph prompt", "judge": "deterministic set-overlap F1 against answer node list", "notes": "Exact-length point from the locked Xiaomi line chart. The benchmark definition is the immutable official GraphWalks README after the 2026-02-27 fixes. The chart prints F1 on [0,1]; campaign scores are the same values expressed as percentages." } }, { "id": "graphwalks_parents_256k", "name": "GraphWalks Parents at 256K", "category": "Long Context", "metric": "% F1", "num_problems": null, "source_url": "https://huggingface.co/datasets/openai/graphwalks/resolve/f338bb265735a56a79f4b0f5def722c9c3268ead/README.md", "canonical_setting": { "version": "GraphWalks post-2026-02-27 ground-truth/prompt fix; Parents exact 256K point", "metric_type": "pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": false, "tools": "none", "sampling": "one response per graph prompt", "judge": "deterministic set-overlap F1 against answer node list", "notes": "Exact-length point from the locked Xiaomi line chart. The benchmark definition is the immutable official GraphWalks README after the 2026-02-27 fixes. The chart prints F1 on [0,1]; campaign scores are the same values expressed as percentages." } }, { "id": "graphwalks_parents_512k", "name": "GraphWalks Parents at 512K", "category": "Long Context", "metric": "% F1", "num_problems": null, "source_url": "https://huggingface.co/datasets/openai/graphwalks/resolve/f338bb265735a56a79f4b0f5def722c9c3268ead/README.md", "canonical_setting": { "version": "GraphWalks post-2026-02-27 ground-truth/prompt fix; Parents exact 512K point", "metric_type": "pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": false, "tools": "none", "sampling": "one response per graph prompt", "judge": "deterministic set-overlap F1 against answer node list", "notes": "Exact-length point from the locked Xiaomi line chart. The benchmark definition is the immutable official GraphWalks README after the 2026-02-27 fixes. The chart prints F1 on [0,1]; campaign scores are the same values expressed as percentages." } }, { "id": "graphwalks_parents_1m", "name": "GraphWalks Parents at 1M", "category": "Long Context", "metric": "% F1", "num_problems": null, "source_url": "https://huggingface.co/datasets/openai/graphwalks/resolve/f338bb265735a56a79f4b0f5def722c9c3268ead/README.md", "canonical_setting": { "version": "GraphWalks post-2026-02-27 ground-truth/prompt fix; Parents exact 1M point", "metric_type": "pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": false, "tools": "none", "sampling": "one response per graph prompt", "judge": "deterministic set-overlap F1 against answer node list", "notes": "Exact-length point from the locked Xiaomi line chart. The benchmark definition is the immutable official GraphWalks README after the 2026-02-27 fixes. The chart prints F1 on [0,1]; campaign scores are the same values expressed as percentages." } }, { "id": "livecodebench_v6_2408_2505", "name": "LiveCodeBench v6 (2024-08 to 2025-05)", "category": "Coding", "metric": "pass@1 / mean pass@1 (%)", "num_problems": 454, "source_url": "https://raw.githubusercontent.com/LiveCodeBench/LiveCodeBench/28fef95ea8c9f7a547c8329f2cd3d32b92c1fa24/README.md", "canonical_setting": { "version": "LiveCodeBench v6, 454 problems, 2024-08 through 2025-05", "metric_type": "pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": false, "notes": "The pinned official LiveCodeBench repository defines release versions and date-window filtering. The InclusionAI report is score evidence for this exact 454-problem 2024-08 to 2025-05 window, distinct from the 1,055-problem release_v6." } }, { "id": "claw_eval_general_pass3", "name": "Claw-Eval General (Pass^3)", "category": "Agentic", "metric": "all-three-pass rate (%)", "num_problems": 483, "source_url": "https://raw.githubusercontent.com/claw-eval/claw-eval/9ac81fc3f18e9711bc0e0e3a96f0883887ae88b4/README.md", "canonical_setting": { "version": "Claw-Eval v1.1 general split, 161 tasks, three trials", "metric_type": "pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": false, "notes": "Exactly 161 general tasks x 3 trajectories. Distinct from the 199-task non-multimodal general+multi-turn aggregate and from the 101-task multimodal split." } }, { "id": "pinchbench_123", "name": "PinchBench 123-task manifest at commit 27afe091b6ae04ec4b6aa9f5459bd280da0fd61d (2026-04-24)", "category": "Agentic", "metric": "mean task score (%)", "num_problems": 123, "source_url": "https://raw.githubusercontent.com/pinchbench/skill/27afe091b6ae04ec4b6aa9f5459bd280da0fd61d/manifest.yaml", "canonical_setting": { "version": "PinchBench 123-task manifest at commit 27afe091b6ae04ec4b6aa9f5459bd280da0fd61d (2026-04-24)", "metric_type": "pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": false, "notes": "The immutable manifest has 123 unique task IDs. Each task is scored on [0,1], including partial rubric credit; a run score is the arithmetic mean across tasks. Best Score and average across runs are distinct aggregations. No release tag is inferred. This is distinct from the campaign's 53-task identity." } }, { "id": "vstar", "name": "V* Bench", "category": "Vision", "metric": "multiple-choice accuracy (%)", "num_problems": 191, "source_url": "https://huggingface.co/datasets/craigwu/vstar_bench/tree/d9ae62c903da0c98336e85c5ee89cd863b04b4da", "canonical_setting": { "version": "V* Bench official 191-question test_questions release", "metric_type": "pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": true, "tools": "none; score-level settings may add a Python visual-search tool", "notes": "The pinned official test_questions.jsonl has 191 items. StepFun's reported V* cells use a Python visual tool and are therefore non-default score settings, not a distinct benchmark." } }, { "id": "hr_bench_4k", "name": "HR-Bench 4K", "category": "Vision", "metric": "accuracy (%)", "num_problems": 800, "source_url": "https://huggingface.co/datasets/DreamMr/HR-Bench/tree/83b9013d6293b85dc507e87199ca52517536939c", "canonical_setting": { "version": "HR-Bench 4K official option-rotation release", "metric_type": "pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": true, "tools": "none; score-level settings may add a Python visual tool", "notes": "The benchmark has 200 logical questions: 100 FSP and 100 FCP. The official 4K split contains 800 evaluated rows after four cyclic answer-option rotations, so num_problems records 800 physical model generations." } }, { "id": "hr_bench_8k", "name": "HR-Bench 8K", "category": "Vision", "metric": "accuracy (%)", "num_problems": 800, "source_url": "https://huggingface.co/datasets/DreamMr/HR-Bench/tree/83b9013d6293b85dc507e87199ca52517536939c", "canonical_setting": { "version": "HR-Bench 8K official option-rotation release", "metric_type": "pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": true, "tools": "none; score-level settings may add a Python visual tool", "notes": "The benchmark has 200 logical questions: 100 FSP and 100 FCP. The official 8K split contains 800 evaluated rows after four cyclic answer-option rotations, so num_problems records 800 physical model generations." } }, { "id": "android_daily", "name": "AndroidDaily", "category": "GUI Agent", "metric": "pass@1 task success rate (%)", "num_problems": 350, "source_url": "https://arxiv.org/html/2605.27761v1", "canonical_setting": { "version": "AndroidDaily v1: 350 tasks across 94 closed-source applications", "metric_type": "pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": true, "tools": "physical Android phone-use action environment", "notes": "The official paper defines 350 tasks over 94 applications and reports pass@1 success without multi-seed variance. Each rollout may run for up to 40 minutes. StepFun's launch chart uses its Step-specific phone-use stack; that harness remains score-level provenance." } }, { "id": "osworld_2_0_2026_06_24_partial", "name": "OSWorld 2.0 (2026-06-24 partial score)", "category": "Agentic", "metric": "weighted checkpoint partial score (%)", "num_problems": 108, "source_url": "https://raw.githubusercontent.com/xlang-ai/OSWorld-V2/8b6b59660b59832a42a345db8f86fa9f98c37573/benchmark_releases/osworld-v2-2026.06.24.json", "canonical_setting": { "version": "osworld-v2-2026.06.24, before 2026-08-08 patch", "metric_type": "weighted_checkpoint_score_pct", "range": [ 0, 100 ], "higher_is_better": true, "notes": "Distinct official 108-task release manifest from 2026-06-24, before the 2026-08-08 patch. The metric is the weighted checkpoint partial score." } }, { "id": "labbench2", "name": "LABBench2", "category": "Science", "metric": "accuracy (%)", "num_problems": 1912, "source_url": "https://raw.githubusercontent.com/EdisonScientific/labbench2/c028ecdcf144b55ffcd92b68be45081df5628c20/README.md", "canonical_setting": { "version": "LABBench2 full 1,912-task release", "metric_type": "accuracy_pct", "range": [ 0, 100 ], "higher_is_better": true, "multimodal_input": true, "tools": "benchmark/model-specific", "sampling": "pass@1", "judge": "task-specific exact/programmatic evaluator", "harness": "official LABBench2 evaluation harness", "dataset_split": "full 1,912-task suite", "notes": "Paper Table 1 totals 1,912 evaluation units across literature, databases, protocols, sequence, cloning, figure, and table tasks." } } ], "scores": [ { "model_id": "claude-haiku-4.5", "benchmark_id": "tau2_bench_airline", "score": 63.6, "reference_url": "https://www.anthropic.com/news/claude-haiku-4-5", "reported_setting": { "mode": "thinking", "effort": "128K thinking budget", "sampling": "pass@1 (avg 10 runs)", "judge": "official harness", "harness": "τ²-bench official", "prompt_style": "official + prompt addendum to Airline Agent Policy and User prompt (per Anthropic blog)", "temperature": "default", "tools": "agentic (τ²-bench official tools + tool use)" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Anthropic Haiku 4.5 blog (cross-model comparison table), airline domain, 128K thinking, avg 10 runs. Prompt addendum added to Agent Policy and User prompt for airline domain → matches_canonical=false." }, { "model_id": "claude-haiku-4.5", "benchmark_id": "tau2_bench_retail", "score": 83.2, "reference_url": "https://www.anthropic.com/news/claude-haiku-4-5", "reported_setting": { "mode": "thinking", "effort": "128K thinking budget", "sampling": "pass@1 (avg 10 runs)", "judge": "official harness", "harness": "τ²-bench official", "prompt_style": "default", "temperature": "default", "tools": "agentic (τ²-bench official tools + tool use)" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "Anthropic Haiku 4.5 blog (cross-model comparison table), retail domain, 128K thinking, avg 10 runs." }, { "model_id": "claude-haiku-4.5", "benchmark_id": "tau2_bench_telecom", "score": 83.0, "reference_url": "https://www.anthropic.com/news/claude-haiku-4-5", "reported_setting": { "mode": "thinking", "effort": "128K thinking budget", "sampling": "pass@1 (avg 10 runs)", "judge": "official harness", "harness": "τ²-bench official", "prompt_style": "official + prompt addendum to Telecom Agent Policy and User prompt (per Anthropic blog)", "temperature": "default", "tools": "agentic (τ²-bench official tools + tool use)" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Anthropic Haiku 4.5 blog (cross-model comparison table), telecom domain, 128K thinking, avg 10 runs. Prompt addendum added to Agent Policy and User prompt for telecom domain → matches_canonical=false.", "candidates": [ { "score": 54.7, "reference_url": "https://microsoft.ai/pdf/MAI-Code-1-Flash-Model-Card.PDF", "source_type": "official_model_card", "reported_setting": { "mode": "adaptive reasoning", "effort": "adaptive solution length control", "tools": "Tau2 telecom tools", "sampling": "single/multi-turn average where applicable", "judge": "benchmark-specific", "harness": "Microsoft MAI-Code-1-Flash comparative evaluation", "prompt_style": "source does not state", "temperature": "source does not state", "context": "256K", "notes": "Official instruction-following or agentic-tool-use score; benchmark=tau2_bench_telecom." }, "notes": "Displayed exactly as 54.7. Official Microsoft-reported result." } ] }, { "model_id": "claude-sonnet-4", "benchmark_id": "tau2_bench_airline", "score": 63.0, "reference_url": "https://www.anthropic.com/news/claude-haiku-4-5", "reported_setting": { "mode": "thinking", "effort": "128K thinking budget", "sampling": "pass@1 (avg 10 runs)", "judge": "official harness", "harness": "τ²-bench official", "prompt_style": "official + prompt addendum to Airline Agent Policy and User prompt (per Anthropic blog)", "temperature": "default", "tools": "agentic (τ²-bench official tools + tool use)" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Anthropic Haiku 4.5 blog (cross-model comparison table), airline domain, 128K thinking, avg 10 runs. Prompt addendum added to Agent Policy and User prompt for airline domain → matches_canonical=false." }, { "model_id": "claude-sonnet-4", "benchmark_id": "tau2_bench_retail", "score": 83.8, "reference_url": "https://www.anthropic.com/news/claude-haiku-4-5", "reported_setting": { "mode": "thinking", "effort": "128K thinking budget", "sampling": "pass@1 (avg 10 runs)", "judge": "official harness", "harness": "τ²-bench official", "prompt_style": "default", "temperature": "default", "tools": "agentic (τ²-bench official tools + tool use)" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "Anthropic Haiku 4.5 blog (cross-model comparison table), retail domain, 128K thinking, avg 10 runs." }, { "model_id": "claude-sonnet-4", "benchmark_id": "tau2_bench_telecom", "score": 49.6, "reference_url": "https://www.anthropic.com/news/claude-haiku-4-5", "reported_setting": { "mode": "thinking", "effort": "128K thinking budget", "sampling": "pass@1 (avg 10 runs)", "judge": "official harness", "harness": "τ²-bench official", "prompt_style": "official + prompt addendum to Telecom Agent Policy and User prompt (per Anthropic blog)", "temperature": "default", "tools": "agentic (τ²-bench official tools + tool use)" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Anthropic Haiku 4.5 blog (cross-model comparison table), telecom domain, 128K thinking, avg 10 runs. Prompt addendum added to Agent Policy and User prompt for telecom domain → matches_canonical=false.", "candidates": [ { "score": 65, "reference_url": "https://huggingface.co/MiniMaxAI/MiniMax-M2", "source_type": "model_card", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "agentic", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "notes": "MiniMax M2 model card: claude-sonnet-4=65." } ] }, { "model_id": "claude-sonnet-4.5", "benchmark_id": "tau2_bench_airline", "score": 70.0, "reference_url": "https://www.anthropic.com/news/claude-haiku-4-5", "reported_setting": { "mode": "thinking", "effort": "128K thinking budget", "sampling": "pass@1 (avg 10 runs)", "judge": "official harness", "harness": "τ²-bench official", "prompt_style": "official + prompt addendum to Airline Agent Policy and User prompt (per Anthropic blog)", "temperature": "default", "tools": "agentic (τ²-bench official tools + tool use)" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Anthropic Haiku 4.5 blog (cross-model comparison table), airline domain, 128K thinking, avg 10 runs. Prompt addendum added to Agent Policy and User prompt for airline domain → matches_canonical=false." }, { "model_id": "claude-sonnet-4.5", "benchmark_id": "tau2_bench_retail", "score": 86.2, "reference_url": "https://www.anthropic.com/news/claude-haiku-4-5", "reported_setting": { "mode": "thinking", "effort": "128K thinking budget", "sampling": "pass@1 (avg 10 runs)", "judge": "official harness", "harness": "τ²-bench official", "prompt_style": "default", "temperature": "default", "tools": "agentic (τ²-bench official tools + tool use)" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "Anthropic Haiku 4.5 blog (cross-model comparison table), retail domain, 128K thinking, avg 10 runs." }, { "model_id": "claude-sonnet-4.5", "benchmark_id": "tau2_bench_telecom", "score": 98.0, "reference_url": "https://www.anthropic.com/news/claude-haiku-4-5", "reported_setting": { "mode": "thinking", "effort": "128K thinking budget", "sampling": "pass@1 (avg 10 runs)", "judge": "official harness", "harness": "τ²-bench official", "prompt_style": "official + prompt addendum to Telecom Agent Policy and User prompt (per Anthropic blog)", "temperature": "default", "tools": "agentic (τ²-bench official tools + tool use)" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Anthropic Haiku 4.5 blog (cross-model comparison table), telecom domain, 128K thinking, avg 10 runs. Prompt addendum added to Agent Policy and User prompt for telecom domain → matches_canonical=false.", "candidates": [ { "score": 78, "reference_url": "https://huggingface.co/MiniMaxAI/MiniMax-M2", "source_type": "model_card", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "agentic", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "notes": "MiniMax M2 model card: claude-sonnet-4.5=78." } ] }, { "model_id": "gpt-5", "benchmark_id": "tau2_bench_airline", "score": 62.6, "reference_url": "https://www.anthropic.com/news/claude-haiku-4-5", "reported_setting": { "mode": "thinking", "effort": "128K thinking budget", "sampling": "pass@1 (avg 10 runs)", "judge": "official harness", "harness": "τ²-bench official", "prompt_style": "official + prompt addendum to Airline Agent Policy and User prompt (per Anthropic blog)", "temperature": "default", "tools": "agentic (τ²-bench official tools + tool use)" }, "matches_canonical": false, "source_type": "third_party", "audit_status": "verified", "notes": "Anthropic Haiku 4.5 blog (cross-model comparison table), airline domain, 128K thinking, avg 10 runs. Prompt addendum added to Agent Policy and User prompt for airline domain → matches_canonical=false. Reported by Anthropic, not OpenAI primary — source_type=tech_report (third-party measurement on competitor model)." }, { "model_id": "gpt-5", "benchmark_id": "tau2_bench_retail", "score": 81.1, "reference_url": "https://www.anthropic.com/news/claude-haiku-4-5", "reported_setting": { "mode": "thinking", "effort": "128K thinking budget", "sampling": "pass@1 (avg 10 runs)", "judge": "official harness", "harness": "τ²-bench official", "prompt_style": "default", "temperature": "default", "tools": "agentic (τ²-bench official tools + tool use)" }, "matches_canonical": false, "source_type": "third_party", "audit_status": "verified", "notes": "Anthropic Haiku 4.5 blog (cross-model comparison table), retail domain, 128K thinking, avg 10 runs. Reported by Anthropic, not OpenAI primary — source_type=tech_report (third-party measurement on competitor model)." }, { "model_id": "gpt-5", "benchmark_id": "tau2_bench_telecom", "score": 96.7, "reference_url": "https://www.anthropic.com/news/claude-haiku-4-5", "reported_setting": { "mode": "thinking", "effort": "128K thinking budget", "sampling": "pass@1 (avg 10 runs)", "judge": "official harness", "harness": "τ²-bench official", "prompt_style": "official + prompt addendum to Telecom Agent Policy and User prompt (per Anthropic blog)", "temperature": "default", "tools": "agentic (τ²-bench official tools + tool use)" }, "matches_canonical": false, "source_type": "third_party", "audit_status": "verified", "notes": "Anthropic Haiku 4.5 blog (cross-model comparison table), telecom domain, 128K thinking, avg 10 runs. Prompt addendum added to Agent Policy and User prompt for telecom domain → matches_canonical=false. Reported by Anthropic, not OpenAI primary — source_type=tech_report (third-party measurement on competitor model).", "candidates": [ { "score": 85, "reference_url": "https://huggingface.co/MiniMaxAI/MiniMax-M2", "source_type": "model_card", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "agentic", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "notes": "MiniMax M2 model card: gpt-5=85." } ] }, { "model_id": "o3-mini-high", "benchmark_id": "aime_2024", "score": 87.3, "reference_url": "https://openai.com/index/gpt-4-1/", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-4.1 blog: o3-mini-high=87.3.", "candidates": [ { "score": 88.0, "reference_url": "https://arxiv.org/abs/2504.21318", "source_type": "tech_report", "reported_setting": { "mode": "unknown", "effort": "unknown", "tools": "unknown", "sampling": "unknown", "judge": "rule-based", "harness": "Microsoft Phi internal eval", "prompt_style": "unknown", "temperature": "unknown" }, "notes": "Phi-4-reasoning paper Table 1 (o3-mini-high column): aime_2024=88.0 (alt measurement, mc=false)" } ] }, { "model_id": "o3-mini-high", "benchmark_id": "gpqa_diamond", "score": 77.2, "reference_url": "https://openai.com/index/gpt-4-1/", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-4.1 blog: o3-mini-high=77.2.", "candidates": [ { "score": 79.7, "reference_url": "https://arxiv.org/abs/2504.13914", "source_type": "tech_report", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "notes": "Seed-Thinking-v1.5 paper Table 2: o3-mini-high=79.7." }, { "score": 77.7, "reference_url": "https://arxiv.org/abs/2504.21318", "source_type": "tech_report", "reported_setting": { "mode": "unknown", "effort": "unknown", "tools": "unknown", "sampling": "unknown", "judge": "rule-based", "harness": "Microsoft Phi internal eval", "prompt_style": "unknown", "temperature": "unknown" }, "notes": "Phi-4-reasoning paper Table 1 (o3-mini-high column): gpqa_diamond=77.7 (alt measurement, mc=false)" }, { "score": 79.7, "reference_url": "https://www.anthropic.com/news/claude-3-7-sonnet", "source_type": "official_blog", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "notes": "Anthropic 3.7 Sonnet blog (third-party eval): o3-mini-high GPQA Diamond=79.7%. BP primary=77.2 from gpt-4-1 blog." } ] }, { "model_id": "o3-mini-high", "benchmark_id": "codeforces_rating", "score": 2130, "reference_url": "https://arxiv.org/abs/2504.21318", "reported_setting": { "mode": "unknown", "effort": "unknown", "tools": "unknown", "sampling": "unknown", "judge": "rule-based", "harness": "Microsoft Phi internal eval", "prompt_style": "unknown", "temperature": "unknown" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Phi-4-reasoning paper Table 1 (o3-mini-high column): codeforces_rating=2130 (re-sourced from https://arxiv.org/abs/2504.21318, matches BP value)" }, { "model_id": "o3-mini-high", "benchmark_id": "mmlu", "score": 86.9, "reference_url": "https://openai.com/index/gpt-4-1/", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-4.1 blog: o3-mini-high=86.9." }, { "model_id": "o3-mini-high", "benchmark_id": "math_500", "score": 97.9, "reference_url": "https://github.com/openai/simple-evals", "audit_status": "verified", "reported_setting": "MATH-500, 0-shot (footnote: MATH-500 not full MATH for newer models)", "source_type": "tech_report", "matches_canonical": true }, { "model_id": "o3-mini-high", "benchmark_id": "humaneval", "score": 97.6, "reference_url": "https://github.com/openai/simple-evals", "audit_status": "verified", "reported_setting": "HumanEval, 0-shot", "source_type": "tech_report", "matches_canonical": true }, { "model_id": "o3-mini-high", "benchmark_id": "simpleqa", "score": 13.8, "reference_url": "https://arxiv.org/abs/2504.13914", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Seed-Thinking-v1.5 paper Table 2: o3-mini-high=13.8." }, { "model_id": "o3-mini-high", "benchmark_id": "hle", "score": 13.0, "reference_url": "https://scale.com/leaderboard/humanitys_last_exam", "audit_status": "dropped", "notes": " [Dropped: Model not found in current Scale.com HLE leaderboard (50 entries, 2026-04-29 scrape). Ghost cell per R5g-a-ghost-cell.]" }, { "model_id": "o3-mini-high", "benchmark_id": "frontiermath", "score": 12.4, "reference_url": "https://epoch.ai/benchmarks/frontiermath", "audit_status": "verified", "source_type": "leaderboard", "matches_canonical": true, "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none" }, "notes": "epoch.ai FrontierMath-2025-02-28-Private: o3-mini-2025-01-31_high = 12.4% (mean_score=0.124). Canonical effort=high matches." }, { "model_id": "o3-mini-high", "benchmark_id": "aime_2025", "score": 86.5, "reference_url": "https://arxiv.org/abs/2504.13914", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Seed-Thinking-v1.5 paper Table 2: o3-mini-high=86.5.", "candidates": [ { "score": 82.5, "reference_url": "https://arxiv.org/abs/2504.21318", "source_type": "tech_report", "reported_setting": { "mode": "unknown", "effort": "unknown", "tools": "unknown", "sampling": "unknown", "judge": "rule-based", "harness": "Microsoft Phi internal eval", "prompt_style": "unknown", "temperature": "unknown" }, "notes": "Phi-4-reasoning paper Table 1 (o3-mini-high column): aime_2025=82.5 (alt measurement, mc=false)" } ] }, { "model_id": "o3-mini-high", "benchmark_id": "chatbot_arena_elo", "score": 1363, "reference_url": "https://lmarena.ai/leaderboard/text", "audit_status": "verified", "source_type": "leaderboard", "matches_canonical": true, "notes": "lmarena.ai text arena (fetched 2026-04-29). Arena model name: o3-mini-high (rank 144). ELO updates continuously; score reflects latest available." }, { "model_id": "o3-mini-high", "benchmark_id": "arena_hard", "score": 81.9, "reference_url": "https://arxiv.org/abs/2504.21318", "reported_setting": { "mode": "unknown", "effort": "unknown", "tools": "unknown", "sampling": "unknown", "judge": "rule-based", "harness": "Microsoft Phi internal eval", "prompt_style": "unknown", "temperature": "unknown" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Phi-4-reasoning paper Table 2 (o3-mini-high column): arena_hard=81.9. [R5d: prior unverified value 43.0 from https://github.com/lmarena/arena-hard-auto deleted.]" }, { "model_id": "o3-mini-high", "benchmark_id": "hmmt_feb_2025", "score": 67.5, "reference_url": "https://arxiv.org/abs/2504.21318", "reported_setting": { "mode": "unknown", "effort": "unknown", "tools": "unknown", "sampling": "unknown", "judge": "rule-based", "harness": "Microsoft Phi internal eval", "prompt_style": "unknown", "temperature": "unknown" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Phi-4-reasoning paper Table 1 (o3-mini-high column): hmmt_2025=67.5. [R5d: prior unverified value 50.0 from https://matharena.ai/?comp=hmmt--hmmt_feb_2025 deleted.]" }, { "model_id": "o3-mini-high", "benchmark_id": "ifeval", "score": 93.9, "reference_url": "https://openai.com/index/gpt-4-1/", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-4.1 blog: o3-mini-high=93.9.", "candidates": [ { "score": 93.7, "reference_url": "https://arxiv.org/abs/2504.13914", "source_type": "tech_report", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "notes": "Seed-Thinking-v1.5 paper Table 2: o3-mini-high=93.7." }, { "score": 91.5, "reference_url": "https://arxiv.org/abs/2504.21318", "source_type": "tech_report", "reported_setting": { "mode": "unknown", "effort": "unknown", "tools": "unknown", "sampling": "unknown", "judge": "rule-based", "harness": "Microsoft Phi internal eval", "prompt_style": "unknown", "temperature": "unknown" }, "notes": "Phi-4-reasoning paper Table 2 (o3-mini-high column): ifeval=91.5 (alt measurement, mc=false)" } ] }, { "model_id": "o3-mini-high", "benchmark_id": "livecodebench", "score": 68.8, "reference_url": "https://arxiv.org/abs/2504.21318", "reported_setting": { "mode": "unknown", "effort": "unknown", "tools": "unknown", "sampling": "unknown", "judge": "rule-based", "harness": "Microsoft Phi internal eval", "prompt_style": "unknown", "temperature": "unknown" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Phi-4-reasoning paper Table 1 (o3-mini-high column): livecodebench=68.8. [R5d: prior unverified value 74.1 from https://openai.com/index/openai-o3-mini/ deleted.]" }, { "model_id": "o3-mini-high", "benchmark_id": "mmlu_pro", "score": 82.4, "reference_url": "https://arxiv.org/abs/2504.13914", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Seed-Thinking-v1.5 paper Table 2: o3-mini-high=82.4.", "candidates": [ { "score": 79.4, "reference_url": "https://arxiv.org/abs/2504.21318", "source_type": "tech_report", "reported_setting": { "mode": "unknown", "effort": "unknown", "tools": "unknown", "sampling": "unknown", "judge": "rule-based", "harness": "Microsoft Phi internal eval", "prompt_style": "unknown", "temperature": "unknown" }, "notes": "Phi-4-reasoning paper Table 2 (o3-mini-high column): mmlu_pro=79.4 (alt measurement, mc=false)" } ] }, { "model_id": "o3-mini-high", "benchmark_id": "swe_bench_verified", "score": 61.0, "reference_url": "https://openai.com/index/introducing-gpt-4-5/", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "agentic (bash + edit)", "sampling": "pass@1", "judge": "rule-based", "harness": "official (GPT-4.5 blog)", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-4.5 blog: o3-mini-high=61.0. (GPT-4.1 blog 477-subset value 49.3 dropped — was different N).", "candidates": [ { "score": 49.3, "reference_url": "https://arxiv.org/abs/2504.13914", "source_type": "tech_report", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "agentic (bash + edit)", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "notes": "Seed-Thinking-v1.5 paper Table 2: o3-mini-high=49.3." } ] }, { "model_id": "o3-mini-high", "benchmark_id": "arc_agi_1", "score": 25.8, "reference_url": "https://arxiv.org/abs/2504.13914", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Seed-Thinking-v1.5 paper Table 2: o3-mini-high=25.8." }, { "model_id": "o3-mini-high", "benchmark_id": "arc_agi_2", "score": 3, "reference_url": "https://arcprize.org/leaderboard", "audit_status": "verified", "source_type": "leaderboard", "matches_canonical": true, "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1 (single-pass)", "judge": "rule-based", "harness": "official (arcprize.org)", "system_type": "CoT" }, "notes": "arcprize.org leaderboard audit: o3-mini (High) on leaderboard" }, { "model_id": "o3-mini-high", "benchmark_id": "usamo_2025", "score": 2.08, "reference_url": "https://matharena.ai/?comp=usamo--usamo_2025", "audit_status": "verified", "matches_canonical": true, "reported_setting": "reasoning_effort:high", "source_type": "leaderboard" }, { "model_id": "gpt-4.5", "benchmark_id": "simpleqa", "score": 62.5, "reference_url": "https://www.helicone.ai/blog/gpt-4.5-benchmarks", "audit_status": "dropped", "rule_id": "R5h-third-party-blog", "notes": "DROPPED (R5h): random third-party blog/aggregator, no primary trail" }, { "model_id": "gpt-4.5", "benchmark_id": "mmlu", "score": 90.8, "reference_url": "https://openai.com/index/gpt-4-1/", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-4.1 blog: gpt-4.5=90.8." }, { "model_id": "gpt-4.5", "benchmark_id": "gpqa_diamond", "score": 69.5, "reference_url": "https://openai.com/index/gpt-4-1/", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-4.1 blog: gpt-4.5=69.5." }, { "model_id": "gpt-4.5", "benchmark_id": "swe_bench_verified", "score": 38.0, "reference_url": "https://openai.com/index/introducing-gpt-4-5/", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "agentic (bash + edit)", "sampling": "pass@1", "judge": "rule-based", "harness": "official (GPT-4.5 blog)", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-4.5 blog: gpt-4.5=38.0. (GPT-4.1 blog 477-subset value 38.0 dropped — was different N).", "candidates": [] }, { "model_id": "gpt-4.5", "benchmark_id": "aime_2024", "score": 36.7, "reference_url": "https://openai.com/index/gpt-4-1/", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-4.1 blog: gpt-4.5=36.7." }, { "model_id": "gpt-4.5", "benchmark_id": "arena_hard", "score": 51.4, "reference_url": "https://github.com/lmarena/arena-hard-auto", "audit_status": "verified", "reported_setting": "Arena-Hard-v2.0 Creative Writing, Ensemble GPT-4.1+Gemini-2.5-Pro judge", "source_type": "leaderboard", "matches_canonical": true }, { "model_id": "gpt-4.5", "benchmark_id": "arc_agi_1", "score": 10.3, "reference_url": "https://arcprize.org/leaderboard", "audit_status": "verified", "source_type": "leaderboard", "matches_canonical": true, "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1 (single-pass)", "judge": "rule-based", "harness": "official (arcprize.org)", "system_type": "Base LLM" }, "notes": "arcprize.org leaderboard audit: GPT-4.5 Base LLM on leaderboard" }, { "model_id": "gpt-4.5", "benchmark_id": "arc_agi_2", "score": 0.8, "reference_url": "https://arcprize.org/leaderboard", "audit_status": "verified", "source_type": "leaderboard", "matches_canonical": true, "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1 (single-pass)", "judge": "rule-based", "harness": "official (arcprize.org)", "system_type": "Base LLM" }, "notes": "arcprize.org leaderboard audit: GPT-4.5 Base LLM on leaderboard" }, { "model_id": "gpt-4.5", "benchmark_id": "humaneval", "score": 86.6, "reference_url": "https://www.helicone.ai/blog/gpt-4.5-benchmarks", "audit_status": "dropped", "rule_id": "R5h-third-party-blog", "notes": "DROPPED (R5h): random third-party blog/aggregator, no primary trail" }, { "model_id": "gpt-4.5", "benchmark_id": "ifeval", "score": 88.2, "reference_url": "https://openai.com/index/gpt-4-1/", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-4.1 blog: gpt-4.5=88.2." }, { "model_id": "gpt-4.5", "benchmark_id": "mmlu_pro", "score": 74.3, "reference_url": "https://www.helicone.ai/blog/gpt-4.5-benchmarks", "audit_status": "dropped", "rule_id": "R5h-third-party-blog", "notes": "DROPPED (R5h): random third-party blog/aggregator, no primary trail" }, { "model_id": "gpt-4.1", "benchmark_id": "swe_bench_verified", "score": 54.6, "reference_url": "https://openai.com/index/introducing-gpt-5-for-developers/", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "agentic (bash + edit)", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "dropped", "notes": "OpenAI GPT-5 dev blog: gpt-4.1=54.6 (high reasoning effort). DROPPED: source uses 477/500 subset (omit 23 problems), not full SWE-bench Verified." }, { "model_id": "gpt-4.1", "benchmark_id": "mmlu", "score": 90.2, "reference_url": "https://openai.com/index/gpt-4-1/", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-4.1 blog: gpt-4.1=90.2." }, { "model_id": "gpt-4.1", "benchmark_id": "gpqa_diamond", "score": 66.3, "reference_url": "https://openai.com/index/introducing-gpt-5-for-developers/", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-5 dev blog: gpt-4.1=66.3 (high reasoning effort)." }, { "model_id": "gpt-4.1", "benchmark_id": "ifeval", "score": 87.4, "reference_url": "https://openai.com/index/gpt-4-1/", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-4.1 blog: gpt-4.1=87.4." }, { "model_id": "gpt-4.1", "benchmark_id": "mmmu", "score": 74.8, "reference_url": "https://openai.com/index/introducing-gpt-5-for-developers/", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-5 dev blog: gpt-4.1=74.8 (high reasoning effort)." }, { "model_id": "gpt-4.1", "benchmark_id": "arena_hard", "score": 61.5, "reference_url": "https://github.com/lmarena/arena-hard-auto", "audit_status": "verified", "reported_setting": "Arena-Hard-v2.0 Creative Writing, Ensemble GPT-4.1+Gemini-2.5-Pro judge", "source_type": "leaderboard", "matches_canonical": true }, { "model_id": "gpt-4.1", "benchmark_id": "aime_2024", "score": 48.1, "reference_url": "https://openai.com/index/gpt-4-1/", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-4.1 blog: gpt-4.1=48.1." }, { "model_id": "gpt-4.1", "benchmark_id": "aime_2025", "score": 46.4, "reference_url": "https://openai.com/index/introducing-gpt-5-for-developers/", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-5 dev blog: gpt-4.1=46.4 (high reasoning effort)." }, { "model_id": "gpt-4.1", "benchmark_id": "hle", "score": 5.4, "reference_url": "https://openai.com/index/introducing-gpt-5-for-developers/", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-5 dev blog: gpt-4.1=5.4 (high reasoning effort)." }, { "model_id": "gpt-4.1", "benchmark_id": "hmmt_feb_2025", "score": 28.9, "reference_url": "https://openai.com/index/introducing-gpt-5-for-developers/", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-5 dev blog: gpt-4.1=28.9 (high reasoning effort)." }, { "model_id": "gpt-4.1", "benchmark_id": "humaneval", "score": 92.0, "reference_url": "https://openai.com/index/gpt-4-1/", "audit_status": "dropped", "notes": " [R5g: ghost cell — openai.com/gpt-4-1 blog does NOT contain HumanEval data; score 92.0 origin unknown. Dropped per R5g.]" }, { "model_id": "gpt-4.1", "benchmark_id": "livecodebench", "score": 44.7, "reference_url": "https://arxiv.org/abs/2507.20534", "audit_status": "verified", "source_type": "tech_report", "matches_canonical": true, "reported_setting": { "mode": "non-thinking", "effort": "default", "tools": "none (benchmark-specified)", "sampling": "pass@1", "judge": "rule-based", "harness": "Kimi team evaluation (official API)", "prompt_style": "default", "temperature": "unified (Kimi team setting)", "context": "8192 max output tokens" }, "notes": "Via third-party self-test by Moonshot AI (Kimi K2 tech report, Table 3, p.16). LiveCodeBench v6 (Aug 2024–May 2025), non-thinking, 8192 max output tokens." }, { "model_id": "gpt-4.1", "benchmark_id": "math_500", "score": 92.4, "reference_url": "https://arxiv.org/abs/2507.20534", "audit_status": "verified", "source_type": "tech_report", "matches_canonical": true, "reported_setting": { "mode": "non-thinking", "effort": "default", "tools": "none (benchmark-specified)", "sampling": "pass@1", "judge": "rule-based", "harness": "Kimi team evaluation (official API)", "prompt_style": "default", "temperature": "unified (Kimi team setting)", "context": "8192 max output tokens" }, "notes": "Via third-party self-test by Moonshot AI (Kimi K2 tech report, Table 3, p.16). MATH-500 Acc, non-thinking, 8192 max output tokens." }, { "model_id": "gpt-4.1", "benchmark_id": "mmlu_pro", "score": 81.8, "reference_url": "https://arxiv.org/abs/2507.20534", "audit_status": "verified", "source_type": "tech_report", "matches_canonical": true, "reported_setting": { "mode": "non-thinking", "effort": "default", "tools": "none (benchmark-specified)", "sampling": "pass@1 (EM)", "judge": "rule-based", "harness": "Kimi team evaluation (official API)", "prompt_style": "default", "temperature": "unified (Kimi team setting)", "context": "8192 max output tokens" }, "notes": "Via third-party self-test by Moonshot AI (Kimi K2 tech report, Table 3, p.16). MMLU-Pro EM, non-thinking, 8192 max output tokens." }, { "model_id": "gpt-4.1", "benchmark_id": "simpleqa", "score": 42.3, "reference_url": "https://arxiv.org/abs/2507.20534", "audit_status": "verified", "source_type": "tech_report", "matches_canonical": true, "reported_setting": { "mode": "non-thinking", "effort": "default", "tools": "none (benchmark-specified)", "sampling": "pass@1", "judge": "rule-based", "harness": "Kimi team evaluation (official API)", "prompt_style": "default", "temperature": "unified (Kimi team setting)", "context": "8192 max output tokens" }, "notes": "Via third-party self-test by Moonshot AI (Kimi K2 tech report, Table 3, p.16). SimpleQA Correct, non-thinking, 8192 max output tokens." }, { "model_id": "gpt-4.1", "benchmark_id": "arc_agi_1", "score": 5.5, "reference_url": "https://arcprize.org/leaderboard", "audit_status": "verified", "source_type": "leaderboard", "matches_canonical": true, "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1 (single-pass)", "judge": "rule-based", "harness": "official (arcprize.org)", "system_type": "Base LLM" }, "notes": "arcprize.org leaderboard audit: GPT-4.1 Base LLM on leaderboard" }, { "model_id": "gpt-4.1", "benchmark_id": "arc_agi_2", "score": 0.4, "reference_url": "https://arcprize.org/leaderboard", "audit_status": "verified", "source_type": "leaderboard", "matches_canonical": true, "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1 (single-pass)", "judge": "rule-based", "harness": "official (arcprize.org)", "system_type": "Base LLM" }, "notes": "arcprize.org leaderboard audit: GPT-4.1 Base LLM on leaderboard" }, { "model_id": "gpt-4.1", "benchmark_id": "codeforces_rating", "score": 1807, "reference_url": "https://openai.com/index/gpt-4-1/", "audit_status": "dropped", "notes": " [R5g: ghost cell — openai.com/gpt-4-1 blog does NOT contain Codeforces data; score 1807 origin unknown. Dropped per R5g.]" }, { "model_id": "gpt-4.1", "benchmark_id": "mrcr_v2", "score": 80, "reference_url": "https://awesomeagents.ai/leaderboards/long-context-benchmarks-leaderboard/", "audit_status": "dropped", "rule_id": "R5h-third-party-blog", "notes": "DROPPED (R5h): random third-party blog/aggregator, no primary trail" }, { "model_id": "gpt-4.1", "benchmark_id": "terminal_bench_1", "score": 30.3, "reference_url": "https://www.tbench.ai/leaderboard/terminal-bench/1.0", "audit_status": "verified", "source_type": "leaderboard", "rule_ids": [ "R5d" ], "notes": "Confirmed 30.3% on tbench.ai Terminal-Bench 1.0 leaderboard (Terminus 1 scaffold, 2025-05-15). model=gpt-4.1" }, { "model_id": "gpt-4.1-mini", "benchmark_id": "ifeval", "score": 84.1, "reference_url": "https://openai.com/index/gpt-4-1/", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-4.1 blog: gpt-4.1-mini=84.1." }, { "model_id": "gpt-4.1-mini", "benchmark_id": "mmmu", "score": 72.7, "reference_url": "https://openai.com/index/introducing-gpt-5-for-developers/", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-5 dev blog: gpt-4.1-mini=72.7 (high reasoning effort)." }, { "model_id": "gpt-4.1-mini", "benchmark_id": "arena_hard", "score": 28.2, "reference_url": "https://github.com/lmarena/arena-hard-auto", "audit_status": "verified", "reported_setting": "Arena-Hard-v2.0 Creative Writing, Ensemble GPT-4.1+Gemini-2.5-Pro judge", "source_type": "leaderboard", "matches_canonical": true }, { "model_id": "gpt-4.1-mini", "benchmark_id": "gpqa_diamond", "score": 65.0, "reference_url": "https://openai.com/index/introducing-gpt-5-for-developers/", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-5 dev blog: gpt-4.1-mini=65.0 (high reasoning effort)." }, { "model_id": "gpt-4.1-mini", "benchmark_id": "humaneval", "score": 90.0, "reference_url": "https://www.helicone.ai/blog/gpt-4.1-full-developer-guide", "audit_status": "dropped", "rule_id": "R5h-third-party-blog", "notes": "DROPPED (R5h): random third-party blog/aggregator, no primary trail" }, { "model_id": "gpt-4.1-mini", "benchmark_id": "mmlu", "score": 87.5, "reference_url": "https://openai.com/index/gpt-4-1/", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-4.1 blog: gpt-4.1-mini=87.5." }, { "model_id": "gpt-4.1-mini", "benchmark_id": "simpleqa", "score": 20.0, "reference_url": "https://www.helicone.ai/blog/gpt-4.1-full-developer-guide", "audit_status": "dropped", "rule_id": "R5h-third-party-blog", "notes": "DROPPED (R5h): random third-party blog/aggregator, no primary trail" }, { "model_id": "gpt-4.1-mini", "benchmark_id": "arc_agi_1", "score": 3.5, "reference_url": "https://arcprize.org/leaderboard", "audit_status": "verified", "source_type": "leaderboard", "matches_canonical": true, "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1 (single-pass)", "judge": "rule-based", "harness": "official (arcprize.org)", "system_type": "Base LLM" }, "notes": "arcprize.org leaderboard audit: GPT-4.1-Mini Base LLM on leaderboard" }, { "model_id": "gpt-4.1-mini", "benchmark_id": "arc_agi_2", "score": 0, "reference_url": "https://arcprize.org/leaderboard", "audit_status": "verified", "source_type": "leaderboard", "matches_canonical": true, "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1 (single-pass)", "judge": "rule-based", "harness": "official (arcprize.org)", "system_type": "Base LLM" }, "notes": "arcprize.org leaderboard audit: GPT-4.1-Mini Base LLM on leaderboard" }, { "model_id": "gpt-4.1-mini", "benchmark_id": "bigcodebench", "score": 48.9, "reference_url": "https://bigcode-bench.github.io/", "audit_status": "verified", "reported_setting": "BigCodeBench Complete Instruct pass@1", "source_type": "leaderboard", "matches_canonical": true }, { "model_id": "gpt-4.1-nano", "benchmark_id": "mmlu", "score": 80.1, "reference_url": "https://openai.com/index/gpt-4-1/", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-4.1 blog: gpt-4.1-nano=80.1." }, { "model_id": "gpt-4.1-nano", "benchmark_id": "gpqa_diamond", "score": 50.3, "reference_url": "https://openai.com/index/introducing-gpt-5-for-developers/", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-5 dev blog: gpt-4.1-nano=50.3 (high reasoning effort)." }, { "model_id": "gpt-4.1-nano", "benchmark_id": "arena_hard", "score": 10.7, "reference_url": "https://github.com/lmarena/arena-hard-auto", "audit_status": "verified", "reported_setting": "Arena-Hard-v2.0 Creative Writing, Ensemble GPT-4.1+Gemini-2.5-Pro judge", "source_type": "leaderboard", "matches_canonical": true }, { "model_id": "gpt-4.1-nano", "benchmark_id": "humaneval", "score": 78.0, "reference_url": "https://www.datacamp.com/blog/gpt-4-1", "audit_status": "dropped", "notes": " [Dropped: DataCamp is a tutorial/learning platform aggregator, not primary source. Rule R5h-tutorial-blog.]" }, { "model_id": "gpt-4.1-nano", "benchmark_id": "ifeval", "score": 74.5, "reference_url": "https://openai.com/index/gpt-4-1/", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-4.1 blog: gpt-4.1-nano=74.5." }, { "model_id": "gpt-4.1-nano", "benchmark_id": "arc_agi_1", "score": 0, "reference_url": "https://arcprize.org/leaderboard", "audit_status": "verified", "source_type": "leaderboard", "matches_canonical": true, "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1 (single-pass)", "judge": "rule-based", "harness": "official (arcprize.org)", "system_type": "Base LLM" }, "notes": "arcprize.org leaderboard audit: GPT-4.1-Nano Base LLM on leaderboard" }, { "model_id": "gpt-4.1-nano", "benchmark_id": "arc_agi_2", "score": 0, "reference_url": "https://arcprize.org/leaderboard", "audit_status": "verified", "source_type": "leaderboard", "matches_canonical": true, "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1 (single-pass)", "judge": "rule-based", "harness": "official (arcprize.org)", "system_type": "Base LLM" }, "notes": "arcprize.org leaderboard audit: GPT-4.1-Nano Base LLM on leaderboard" }, { "model_id": "o3-high", "benchmark_id": "aime_2024", "score": 96.7, "reference_url": "https://openai.com/index/introducing-o3-and-o4-mini/", "audit_status": "needs_review", "audit_note": "openai.com/index/introducing-o3-and-o4-mini/ returns 403 (Cloudflare blocked). Value 96.7 widely cited from official Dec 2024 o3 announcement; primary URL inaccessible to bot.", "audited_by": "lychee-audit-bot", "audited_at": "2026-07-26T00:00:00Z", "candidates": [ { "score": 91.6, "reference_url": "https://llm-stats.com/benchmarks/aime-2024", "source_type": "third_party_aggregator", "reported_setting": { "effort": "default (o3 base, not high)" }, "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=o3, slug=o3-2025-04-16, provider=OpenAI" } ] }, { "model_id": "o3-high", "benchmark_id": "aime_2025", "score": 88.9, "reference_url": "https://openai.com/index/introducing-gpt-5-for-developers/", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-5 dev blog: o3-high=88.9 (high reasoning effort).", "candidates": [ { "score": 86.4, "reference_url": "https://llm-stats.com/benchmarks/aime-2025", "source_type": "third_party_aggregator", "reported_setting": { "effort": "default (o3 base, not high)" }, "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=o3, slug=o3-2025-04-16, provider=OpenAI" } ] }, { "model_id": "o3-high", "benchmark_id": "gpqa_diamond", "score": 83.3, "reference_url": "https://openai.com/index/introducing-gpt-5-for-developers/", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-5 dev blog: o3-high=83.3 (high reasoning effort).", "candidates": [ { "score": 83.4, "reference_url": "https://github.com/openai/simple-evals", "source_type": "official_blog", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "notes": "openai/simple-evals: o3-high GPQA=83.4 [fn8: includes answer regex tweak for GPQA]. Differs from verified 83.3 (GPT-5 dev blog)." }, { "score": 83.3, "reference_url": "https://llm-stats.com/benchmarks/gpqa", "source_type": "third_party_aggregator", "reported_setting": { "effort": "default (o3 base, not high)" }, "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=o3, slug=o3-2025-04-16, provider=OpenAI" } ] }, { "model_id": "o3-high", "benchmark_id": "swe_bench_verified", "score": 69.1, "reference_url": "https://openai.com/index/introducing-gpt-5-for-developers/", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "agentic (bash + edit)", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "dropped", "notes": "OpenAI GPT-5 dev blog: o3-high=69.1 (high reasoning effort). DROPPED: source uses 477/500 subset (omit 23 problems), not full SWE-bench Verified.", "candidates": [ { "score": 69.1, "reference_url": "https://llm-stats.com/benchmarks/swe-bench-verified", "source_type": "third_party_aggregator", "reported_setting": { "effort": "default (o3 base, not high)" }, "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=o3, slug=o3-2025-04-16, provider=OpenAI" } ] }, { "model_id": "o3-high", "benchmark_id": "frontiermath", "score": 15.8, "reference_url": "https://openai.com/index/introducing-gpt-5-for-developers/", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "python tool", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "with python tool (per OpenAI dev blog)", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-5 dev blog: o3-high=15.8 (high reasoning effort).", "candidates": [ { "score": 15.8, "reference_url": "https://llm-stats.com/benchmarks/frontiermath", "source_type": "third_party_aggregator", "reported_setting": { "effort": "default (o3 base, not high)" }, "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=o3, slug=o3-2025-04-16, provider=OpenAI" } ] }, { "model_id": "o3-high", "benchmark_id": "livecodebench", "score": 75.8, "reference_url": "https://www.alphaxiv.org/benchmarks/uc-berkeley/livecodebench", "audit_status": "dropped", "rule_id": "R5h-third-party-blog", "notes": "DROPPED (R5h): random third-party blog/aggregator, no primary trail" }, { "model_id": "o3-high", "benchmark_id": "math_500", "score": 98.1, "reference_url": "https://github.com/openai/simple-evals", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "openai/simple-evals README (no tools, [^10]): o3-high=98.1 [fn6: MATH=MATH-500 for post-o1 models]. Prior ref (artificialanalysis.ai) reported 99.2 — overwritten." }, { "model_id": "o3-high", "benchmark_id": "codeforces_rating", "score": 2706, "reference_url": "https://www.datacamp.com/blog/o4-mini", "audit_status": "dropped", "notes": " [Dropped: DataCamp is a tutorial/learning platform aggregator, not primary source. Rule R5h-tutorial-blog.]" }, { "model_id": "o3-high", "benchmark_id": "arc_agi_2", "score": 6.5, "reference_url": "https://arcprize.org/leaderboard", "audit_status": "verified", "source_type": "leaderboard", "matches_canonical": true, "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1 (single-pass)", "judge": "rule-based", "harness": "official (arcprize.org)", "system_type": "CoT" }, "notes": "arcprize.org leaderboard audit: o3 (High) on leaderboard" }, { "model_id": "o3-high", "benchmark_id": "arena_hard", "score": 85.9, "reference_url": "https://llm-stats.com/benchmarks/arena-hard", "audit_status": "dropped", "rule_id": "R5h-third-party-blog", "notes": "DROPPED (R5h): random third-party blog/aggregator, no primary trail" }, { "model_id": "o3-high", "benchmark_id": "mmlu", "score": 93.3, "reference_url": "https://github.com/openai/simple-evals", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "openai/simple-evals README (no tools, [^10]): o3-high=93.3." }, { "model_id": "o3-high", "benchmark_id": "humaneval", "score": 88.4, "reference_url": "https://github.com/openai/simple-evals", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "openai/simple-evals README (no tools, [^10]): o3-high=88.4." }, { "model_id": "o3-high", "benchmark_id": "simpleqa", "score": 48.6, "reference_url": "https://github.com/openai/simple-evals", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "openai/simple-evals README (no tools, [^10]): o3-high=48.6." }, { "model_id": "o3-high", "benchmark_id": "hle", "score": 20.2, "reference_url": "https://openai.com/index/introducing-gpt-5-for-developers/", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-5 dev blog: o3-high=20.2 (high reasoning effort).", "candidates": [ { "score": 14.7, "reference_url": "https://llm-stats.com/benchmarks/humanity's-last-exam", "source_type": "third_party_aggregator", "reported_setting": { "effort": "default (o3 base, not high)" }, "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=o3, slug=o3-2025-04-16, provider=OpenAI" } ] }, { "model_id": "o3-high", "benchmark_id": "browsecomp", "score": 49.7, "reference_url": "https://www.helicone.ai/blog/o3-and-o4-mini-for-developers", "audit_status": "dropped", "rule_id": "R5h-third-party-blog", "notes": "DROPPED (R5h): random third-party blog/aggregator, no primary trail", "candidates": [ { "score": 49.7, "reference_url": "https://llm-stats.com/benchmarks/browsecomp", "source_type": "third_party_aggregator", "reported_setting": { "effort": "default (o3 base, not high)" }, "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=o3, slug=o3-2025-04-16, provider=OpenAI" } ] }, { "model_id": "o3-high", "benchmark_id": "mmmu", "score": 82.9, "reference_url": "https://openai.com/index/introducing-gpt-5-for-developers/", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-5 dev blog: o3-high=82.9 (high reasoning effort).", "candidates": [ { "score": 82.9, "reference_url": "https://llm-stats.com/benchmarks/mmmu", "source_type": "third_party_aggregator", "reported_setting": { "effort": "default (o3 base, not high)" }, "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=o3, slug=o3-2025-04-16, provider=OpenAI" } ] }, { "model_id": "o3-high", "benchmark_id": "arc_agi_1", "score": 60.8, "reference_url": "https://arcprize.org/leaderboard", "audit_status": "verified", "source_type": "leaderboard", "matches_canonical": true, "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1 (single-pass)", "judge": "rule-based", "harness": "official (arcprize.org)", "system_type": "CoT" }, "notes": "arcprize.org leaderboard audit: o3 (High) on leaderboard", "candidates": [ { "score": 88.0, "reference_url": "https://llm-stats.com/benchmarks/arc-agi", "source_type": "third_party_aggregator", "reported_setting": { "effort": "default (o3 base, not high)" }, "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=o3, slug=o3-2025-04-16, provider=OpenAI" } ] }, { "model_id": "o3-high", "benchmark_id": "hmmt_feb_2025", "score": 81.7, "reference_url": "https://openai.com/index/introducing-gpt-5-for-developers/", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-5 dev blog: o3-high=81.7 (high reasoning effort)." }, { "model_id": "o3-high", "benchmark_id": "ifeval", "score": 92.1, "reference_url": "https://huggingface.co/Qwen/Qwen3-235B-A22B-Thinking-2507", "audit_status": "dropped", "rule_id": "R5g-ghost", "audit_note": "Card reports OpenAI O3 IFEval=92.1 at medium effort (no asterisk). JSON model o3-high is high effort. Effort mismatch.", "audited_by": "lychee-audit-bot", "audited_at": "2026-04-29T13:17:14Z" }, { "model_id": "o3-high", "benchmark_id": "imo_2025", "score": 16.67, "reference_url": "https://matharena.ai/?comp=imo--imo_2025", "audit_status": "verified", "matches_canonical": true, "reported_setting": "reasoning_effort:high", "source_type": "leaderboard" }, { "model_id": "o3-high", "benchmark_id": "mmlu_pro", "score": 85.9, "reference_url": "https://huggingface.co/Qwen/Qwen3-235B-A22B-Thinking-2507", "audit_status": "dropped", "rule_id": "R5g-ghost", "audit_note": "Card reports OpenAI O3 MMLU-Pro=85.9 at medium effort. JSON model o3-high is high effort. Effort mismatch.", "audited_by": "lychee-audit-bot", "audited_at": "2026-04-29T13:17:14Z" }, { "model_id": "o3-high", "benchmark_id": "mmmu_pro", "score": 76.4, "reference_url": "https://openai.com/index/introducing-gpt-5-for-developers/", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "avg across standard and vision sets", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-5 dev blog: o3-high=76.4 (high reasoning effort).", "candidates": [ { "score": 76.4, "reference_url": "https://llm-stats.com/benchmarks/mmmu-pro", "source_type": "third_party_aggregator", "reported_setting": { "effort": "default (o3 base, not high)" }, "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=o3, slug=o3-2025-04-16, provider=OpenAI" } ] }, { "model_id": "o3-high", "benchmark_id": "usamo_2025", "score": 21.9, "reference_url": "https://files.sri.inf.ethz.ch/matharena/usamo_report.pdf", "audit_status": "dropped", "rule_id": "R5g-ghost", "audit_note": "Model not in USAMO report. Paper evaluates GEMINI-2.5-PRO/R1/GROK3/FLASH-THINKING/CLAUDE3.7/QWQ/O1-PRO/O3-MINI only.", "audited_by": "lychee-audit-bot", "audited_at": "2026-04-29T13:11:23Z" }, { "model_id": "o3-high", "benchmark_id": "aa_lcr", "score": 69, "reference_url": "https://artificialanalysis.ai/evaluations/artificial-analysis-long-context-reasoning", "audit_status": "verified", "source_type": "third_party", "notes": "ArtificialAnalysis aggregator; used as fallback when no primary source available (R5d-aggregator-fallback). AA LCR eval page: o3 (default=high effort) lcr=0.693 → 69.3%, displayed as 69. AA slug: o3.", "reported_setting": { "mode": "thinking", "effort": "high (AA default)", "sampling": "pass@1", "harness": "AA standard evaluation", "notes": "AA evaluates o3 at high effort by default per aa/evaluations/artificial-analysis-long-context-reasoning" } }, { "model_id": "o3-high", "benchmark_id": "brumo_2025", "score": 95.83, "reference_url": "https://matharena.ai/?comp=brumo--brumo_2025", "audit_status": "verified", "matches_canonical": true, "reported_setting": "reasoning_effort:high", "source_type": "leaderboard" }, { "model_id": "o3-high", "benchmark_id": "cmimc_2025", "score": 79.38, "reference_url": "https://matharena.ai/?comp=cmimc--cmimc_2025", "audit_status": "verified", "matches_canonical": true, "reported_setting": "reasoning_effort:high", "source_type": "leaderboard" }, { "model_id": "o3-high", "benchmark_id": "critpt", "score": 1.4, "reference_url": "https://github.com/CritPt-Benchmark/CritPt", "audit_status": "verified", "reported_setting": "CritPt leaderboard, ε=1.5, % tasks solved", "source_type": "leaderboard", "matches_canonical": true }, { "model_id": "o3-high", "benchmark_id": "ifbench", "score": 69.3, "reference_url": "https://github.com/allenai/IFBench", "audit_status": "verified", "reported_setting": "IFBench score, 0-shot", "source_type": "leaderboard", "matches_canonical": true, "notes": "Source lists \"OpenAI o3\" without explicit effort level" }, { "model_id": "o3-high", "benchmark_id": "smt_2025", "score": 87.74, "reference_url": "https://matharena.ai/?comp=smt--smt_2025", "audit_status": "verified", "matches_canonical": true, "reported_setting": "reasoning_effort:high", "source_type": "leaderboard" }, { "model_id": "o3-high", "benchmark_id": "terminal_bench_1", "score": 30.2, "reference_url": "https://www.tbench.ai/leaderboard/terminal-bench/1.0", "audit_status": "verified", "source_type": "leaderboard", "rule_ids": [ "R5d" ], "notes": "Confirmed 30.2% on tbench.ai Terminal-Bench 1.0 leaderboard (Terminus 1 scaffold, 2025-05-15). model listed as o3 (high effort)" }, { "model_id": "o4-mini-high", "benchmark_id": "aime_2024", "score": 93.4, "reference_url": "https://www.datacamp.com/blog/o4-mini", "audit_status": "dropped", "notes": " [Dropped: DataCamp is a tutorial/learning platform aggregator, not primary source. Rule R5h-tutorial-blog.]", "candidates": [ { "score": 93.4, "reference_url": "https://llm-stats.com/benchmarks/aime-2024", "source_type": "third_party_aggregator", "reported_setting": { "effort": "default (o4-mini base, not high)" }, "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=o4-mini, slug=o4-mini, provider=OpenAI" } ] }, { "model_id": "o4-mini-high", "benchmark_id": "aime_2025", "score": 92.7, "reference_url": "https://openai.com/index/introducing-gpt-5-for-developers/", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-5 dev blog: o4-mini-high=92.7 (high reasoning effort).", "candidates": [ { "score": 92.7, "reference_url": "https://llm-stats.com/benchmarks/aime-2025", "source_type": "third_party_aggregator", "reported_setting": { "effort": "default (o4-mini base, not high)" }, "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=o4-mini, slug=o4-mini, provider=OpenAI" } ] }, { "model_id": "o4-mini-high", "benchmark_id": "gpqa_diamond", "score": 81.4, "reference_url": "https://openai.com/index/introducing-gpt-5-for-developers/", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-5 dev blog: o4-mini-high=81.4 (high reasoning effort).", "candidates": [ { "score": 81.3, "reference_url": "https://github.com/openai/simple-evals", "source_type": "official_blog", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "notes": "openai/simple-evals: o4-mini-high GPQA=81.3 [fn8: includes answer regex tweak for GPQA]. Differs from verified 81.4 (GPT-5 dev blog)." }, { "score": 81.4, "reference_url": "https://llm-stats.com/benchmarks/gpqa", "source_type": "third_party_aggregator", "reported_setting": { "effort": "default (o4-mini base, not high)" }, "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=o4-mini, slug=o4-mini, provider=OpenAI" } ] }, { "model_id": "o4-mini-high", "benchmark_id": "swe_bench_verified", "score": 68.1, "reference_url": "https://openai.com/index/introducing-gpt-5-for-developers/", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "agentic (bash + edit)", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "dropped", "notes": "OpenAI GPT-5 dev blog: o4-mini-high=68.1 (high reasoning effort). DROPPED: source uses 477/500 subset (omit 23 problems), not full SWE-bench Verified.", "candidates": [ { "score": 68.1, "reference_url": "https://llm-stats.com/benchmarks/swe-bench-verified", "source_type": "third_party_aggregator", "reported_setting": { "effort": "default (o4-mini base, not high)" }, "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=o4-mini, slug=o4-mini, provider=OpenAI" } ] }, { "model_id": "o4-mini-high", "benchmark_id": "frontiermath", "score": 15.4, "reference_url": "https://openai.com/index/introducing-gpt-5-for-developers/", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "python tool", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "with python tool (per OpenAI dev blog)", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-5 dev blog: o4-mini-high=15.4 (high reasoning effort)." }, { "model_id": "o4-mini-high", "benchmark_id": "livecodebench", "score": 80.2, "reference_url": "https://www.alphaxiv.org/benchmarks/uc-berkeley/livecodebench", "audit_status": "dropped", "rule_id": "R5h-third-party-blog", "notes": "DROPPED (R5h): random third-party blog/aggregator, no primary trail" }, { "model_id": "o4-mini-high", "benchmark_id": "humaneval", "score": 99.3, "reference_url": "https://github.com/openai/simple-evals", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "openai/simple-evals README (no tools, [^10]): o4-mini-high=99.3. Prior ref (rdworldonline) reported 98.5 — overwritten." }, { "model_id": "o4-mini-high", "benchmark_id": "codeforces_rating", "score": 2719, "reference_url": "https://www.datacamp.com/blog/o4-mini", "audit_status": "dropped", "notes": " [Dropped: DataCamp is a tutorial/learning platform aggregator, not primary source. Rule R5h-tutorial-blog.]" }, { "model_id": "o4-mini-high", "benchmark_id": "arena_hard", "score": 79.1, "reference_url": "https://llm-stats.com/benchmarks/arena-hard", "audit_status": "dropped", "rule_id": "R5h-third-party-blog", "notes": "DROPPED (R5h): random third-party blog/aggregator, no primary trail" }, { "model_id": "o4-mini-high", "benchmark_id": "mmlu", "score": 90.3, "reference_url": "https://github.com/openai/simple-evals", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "openai/simple-evals README (no tools, [^10]): o4-mini-high=90.3." }, { "model_id": "o4-mini-high", "benchmark_id": "math_500", "score": 98.2, "reference_url": "https://github.com/openai/simple-evals", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "openai/simple-evals README (no tools, [^10]): o4-mini-high=98.2 [fn6: MATH=MATH-500]." }, { "model_id": "o4-mini-high", "benchmark_id": "simpleqa", "score": 19.3, "reference_url": "https://github.com/openai/simple-evals", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "openai/simple-evals README (no tools, [^10]): o4-mini-high=19.3." }, { "model_id": "o4-mini-high", "benchmark_id": "hle", "score": 14.7, "reference_url": "https://openai.com/index/introducing-gpt-5-for-developers/", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-5 dev blog: o4-mini-high=14.7 (high reasoning effort).", "candidates": [ { "score": 14.7, "reference_url": "https://llm-stats.com/benchmarks/humanity's-last-exam", "source_type": "third_party_aggregator", "reported_setting": { "effort": "default (o4-mini base, not high)" }, "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=o4-mini, slug=o4-mini, provider=OpenAI" } ] }, { "model_id": "o4-mini-high", "benchmark_id": "arc_agi_2", "score": 6.1, "reference_url": "https://arcprize.org/leaderboard", "audit_status": "verified", "source_type": "leaderboard", "matches_canonical": true, "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1 (single-pass)", "judge": "rule-based", "harness": "official (arcprize.org)", "system_type": "CoT" }, "notes": "arcprize.org leaderboard audit: o4-mini (High) on leaderboard" }, { "model_id": "o4-mini-high", "benchmark_id": "mmmu", "score": 81.6, "reference_url": "https://openai.com/index/introducing-gpt-5-for-developers/", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-5 dev blog: o4-mini-high=81.6 (high reasoning effort).", "candidates": [ { "score": 81.6, "reference_url": "https://llm-stats.com/benchmarks/mmmu", "source_type": "third_party_aggregator", "reported_setting": { "effort": "default (o4-mini base, not high)" }, "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=o4-mini, slug=o4-mini, provider=OpenAI" } ] }, { "model_id": "o4-mini-high", "benchmark_id": "hmmt_feb_2025", "score": 85.0, "reference_url": "https://openai.com/index/introducing-gpt-5-for-developers/", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-5 dev blog: o4-mini-high=85.0 (high reasoning effort)." }, { "model_id": "o4-mini-high", "benchmark_id": "ifeval", "score": 92.4, "reference_url": "https://huggingface.co/Qwen/Qwen3-235B-A22B-Thinking-2507", "audit_status": "verified", "source_type": "model_card", "matches_canonical": true, "audit_note": "Qwen3-235B card: OpenAI O4-mini* IFEval=92.4 (asterisk=high effort, matches o4-mini-high canonical).", "audited_by": "lychee-audit-bot", "audited_at": "2026-04-29T13:17:14Z" }, { "model_id": "o4-mini-high", "benchmark_id": "mmlu_pro", "score": 81.9, "reference_url": "https://huggingface.co/Qwen/Qwen3-235B-A22B-Thinking-2507", "audit_status": "dropped", "rule_id": "R5g-ghost", "audit_note": "Card reports O4-mini MMLU-Pro=81.9 without asterisk (medium effort). JSON model o4-mini-high is high effort. Effort mismatch.", "audited_by": "lychee-audit-bot", "audited_at": "2026-04-29T13:17:14Z" }, { "model_id": "o4-mini-high", "benchmark_id": "usamo_2025", "score": 19.3, "reference_url": "https://files.sri.inf.ethz.ch/matharena/usamo_report.pdf", "audit_status": "dropped", "rule_id": "R5g-ghost", "audit_note": "Model not in USAMO report. Paper evaluates GEMINI-2.5-PRO/R1/GROK3/FLASH-THINKING/CLAUDE3.7/QWQ/O1-PRO/O3-MINI only.", "audited_by": "lychee-audit-bot", "audited_at": "2026-04-29T13:11:23Z" }, { "model_id": "o4-mini-high", "benchmark_id": "arc_agi_1", "score": 58.7, "reference_url": "https://arcprize.org/leaderboard", "audit_status": "verified", "source_type": "leaderboard", "matches_canonical": true, "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1 (single-pass)", "judge": "rule-based", "harness": "official (arcprize.org)", "system_type": "CoT" }, "notes": "arcprize.org leaderboard audit: o4-mini (High) on leaderboard" }, { "model_id": "o4-mini-high", "benchmark_id": "brumo_2025", "score": 86.67, "reference_url": "https://matharena.ai/?comp=brumo--brumo_2025", "audit_status": "verified", "matches_canonical": true, "reported_setting": "reasoning_effort:high", "source_type": "leaderboard" }, { "model_id": "o4-mini-high", "benchmark_id": "cmimc_2025", "score": 84.38, "reference_url": "https://matharena.ai/?comp=cmimc--cmimc_2025", "audit_status": "verified", "matches_canonical": true, "reported_setting": "reasoning_effort:high", "source_type": "leaderboard" }, { "model_id": "o4-mini-high", "benchmark_id": "critpt", "score": 0.6, "reference_url": "https://github.com/CritPt-Benchmark/CritPt", "audit_status": "verified", "reported_setting": "CritPt leaderboard, ε=1.5, % tasks solved", "source_type": "leaderboard", "matches_canonical": true }, { "model_id": "o4-mini-high", "benchmark_id": "imo_2025", "score": 14.29, "reference_url": "https://matharena.ai/?comp=imo--imo_2025", "audit_status": "verified", "matches_canonical": true, "reported_setting": "reasoning_effort:high", "source_type": "leaderboard" }, { "model_id": "o4-mini-high", "benchmark_id": "smt_2025", "score": 88.68, "reference_url": "https://matharena.ai/?comp=smt--smt_2025", "audit_status": "verified", "matches_canonical": true, "reported_setting": "reasoning_effort:high", "source_type": "leaderboard" }, { "model_id": "o4-mini-high", "benchmark_id": "terminal_bench_1", "score": 18.5, "reference_url": "https://www.tbench.ai/leaderboard/terminal-bench/1.0", "audit_status": "verified", "source_type": "leaderboard", "rule_ids": [ "R5d" ], "notes": "Confirmed 18.5% on tbench.ai Terminal-Bench 1.0 leaderboard (Terminus 1 scaffold, 2025-05-15). model listed as o4-mini (high effort)" }, { "model_id": "gpt-5", "benchmark_id": "aime_2025", "score": 94.6, "reference_url": "https://openai.com/index/gpt-5-1-for-developers/", "candidates": [ { "score": 94, "reference_url": "https://huggingface.co/MiniMaxAI/MiniMax-M2", "source_type": "model_card", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "notes": "MiniMax M2 model card: gpt-5=94." }, { "score": 94.6, "reference_url": "https://llm-stats.com/benchmarks/aime-2025", "source_type": "third_party_aggregator", "reported_setting": { "effort": "high" }, "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=GPT-5 High, slug=gpt-5-high-2025-08-07, provider=OpenAI" }, { "score": 88.9, "reference_url": "https://llm-stats.com/benchmarks/aime-2025", "source_type": "third_party_aggregator", "reported_setting": { "effort": "medium" }, "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=GPT-5 Medium, slug=gpt-5-medium-2025-08-07, provider=OpenAI" } ], "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official with apply_patch", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-5.1 dev blog (cross-model column): GPT-5 = 94.6." }, { "model_id": "gpt-5", "benchmark_id": "gpqa_diamond", "score": 85.7, "reference_url": "https://openai.com/index/gpt-5-1-for-developers/", "candidates": [ { "score": 88.4, "reference_url": "https://openai.com/index/introducing-gpt-5/", "source_type": "third_party", "reported_setting": {}, "notes": " (demoted by 5.1 dev blog)" }, { "score": 85, "reference_url": "https://huggingface.co/MiniMaxAI/MiniMax-M2", "source_type": "model_card", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "notes": "MiniMax M2 model card: gpt-5=85." }, { "score": 88.1, "reference_url": "https://llm-stats.com/benchmarks/gpqa", "source_type": "third_party_aggregator", "reported_setting": { "effort": "medium" }, "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=GPT-5 Medium, slug=gpt-5-medium-2025-08-07, provider=OpenAI" }, { "score": 87.3, "reference_url": "https://llm-stats.com/benchmarks/gpqa", "source_type": "third_party_aggregator", "reported_setting": { "effort": "high" }, "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=GPT-5 High, slug=gpt-5-high-2025-08-07, provider=OpenAI" } ], "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official with apply_patch", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-5.1 dev blog (cross-model column): GPT-5 = 85.7." }, { "model_id": "gpt-5", "benchmark_id": "swe_bench_verified", "score": 74.9, "reference_url": "https://huggingface.co/deepseek-ai/DeepSeek-V3.2/resolve/main/assets/paper.pdf", "candidates": [ { "score": 72.8, "reference_url": "https://openai.com/index/gpt-5-1-for-developers/", "source_type": "official_blog", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "agentic (apply_patch JSON)", "sampling": "pass@1", "judge": "rule-based", "harness": "JSON-based apply_patch", "prompt_style": "default", "temperature": "default" }, "notes": "GPT-5.1 dev blog cross-model column reported 72.8 (demoted by GPT-5 own dev blog)." }, { "score": 74.9, "reference_url": "https://openai.com/index/introducing-gpt-5/", "source_type": "third_party", "reported_setting": {}, "notes": " (demoted by 5.1 dev blog)" }, { "score": 72.8, "reference_url": "https://openai.com/index/gpt-5-1-for-developers/", "source_type": "official_blog", "notes": "gpt-5.1 blog cross-model col: GPT-5 SWE-bench Verified=72.8% (vs dev blog 74.9%)" } ], "reported_setting": { "mode": "thinking", "effort": "high", "tools": "agentic (bash + edit)", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "DeepSeek V3.2 tech report Table 2: gpt-5=74.9." }, { "model_id": "gpt-5", "benchmark_id": "mmmu", "score": 84.2, "reference_url": "https://openai.com/index/gpt-5-1-for-developers/", "candidates": [], "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official with apply_patch", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-5.1 dev blog (cross-model column): GPT-5 = 84.2." }, { "model_id": "gpt-5", "benchmark_id": "math_500", "score": 99.4, "reference_url": "https://artificialanalysis.ai/evaluations/math-500", "audit_status": "verified", "source_type": "third_party", "notes": "ArtificialAnalysis aggregator; used as fallback when no primary source available (R5d-aggregator-fallback). AA MATH-500 eval page: gpt-5 math_500=0.994 → 99.4%. AA slug: gpt-5 (deprecated, reasoning variant).", "reported_setting": { "mode": "thinking", "effort": "high (AA default)", "sampling": "pass@1", "harness": "AA standard evaluation", "notes": "Per AA evaluations/math-500: gpt-5 math_500=99.4%" } }, { "model_id": "gpt-5", "benchmark_id": "tau_bench_telecom", "score": 97.0, "reference_url": "https://www.digitalocean.com/resources/articles/gpt-5-overview", "audit_status": "dropped", "rule_id": "R5h-third-party-blog", "notes": "DROPPED (R5h): random third-party blog/aggregator, no primary trail" }, { "model_id": "gpt-5", "benchmark_id": "hle", "score": 24.8, "reference_url": "https://openai.com/index/introducing-gpt-5-for-developers/", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-5 dev blog: gpt-5=24.8 (high reasoning effort).", "candidates": [ { "score": 26.3, "reference_url": "https://huggingface.co/deepseek-ai/DeepSeek-V3.2/resolve/main/assets/paper.pdf", "source_type": "tech_report", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "notes": "DeepSeek V3.2 tech report Table 2: gpt-5=26.3." }, { "score": 26.3, "reference_url": "https://z.ai/blog/glm-4.7", "source_type": "official_blog", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "notes": "GLM-4.7 blog: gpt-5=26.3." }, { "score": 26.3, "reference_url": "https://huggingface.co/moonshotai/Kimi-K2-Thinking", "source_type": "model_card", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "1.0" }, "notes": "Kimi K2-Thinking model card: gpt-5=26.3." } ] }, { "model_id": "gpt-5", "benchmark_id": "browsecomp", "score": 54.9, "reference_url": "https://huggingface.co/deepseek-ai/DeepSeek-V3.2/resolve/main/assets/paper.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "agentic", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "DeepSeek V3.2 tech report Table 2: gpt-5=54.9." }, { "model_id": "gpt-5", "benchmark_id": "mmlu_pro", "score": 87.5, "reference_url": "https://huggingface.co/deepseek-ai/DeepSeek-V3.2/resolve/main/assets/paper.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "DeepSeek V3.2 tech report Table 2: gpt-5=87.5.", "candidates": [ { "score": 87.1, "reference_url": "https://huggingface.co/moonshotai/Kimi-K2-Thinking", "source_type": "model_card", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "1.0" }, "notes": "Kimi K2-Thinking model card: gpt-5=87.1." }, { "score": 87, "reference_url": "https://huggingface.co/MiniMaxAI/MiniMax-M2", "source_type": "model_card", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "notes": "MiniMax M2 model card: gpt-5=87." } ] }, { "model_id": "gpt-5", "benchmark_id": "aa_lcr", "score": 76, "reference_url": "https://huggingface.co/MiniMaxAI/MiniMax-M2", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "MiniMax M2 model card: gpt-5=76." }, { "model_id": "gpt-5", "benchmark_id": "simpleqa", "score": 55.0, "reference_url": "https://cdn.openai.com/gpt-5-system-card.pdf", "audit_status": "verified", "source_type": "official_paper", "matches_canonical": true, "audited_by": "lychee-audit-bot", "audited_at": "2026-04-29T13:09:38Z", "audit_note": "Table 8 in system card: gpt-5-thinking SimpleQA(noweb) accuracy=0.55 (=55%)." }, { "model_id": "gpt-5", "benchmark_id": "frontiermath", "score": 26.3, "reference_url": "https://openai.com/index/introducing-gpt-5-for-developers/", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "python tool", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "with python tool (per OpenAI dev blog)", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-5 dev blog: gpt-5=26.3 (high reasoning effort)." }, { "model_id": "gpt-5", "benchmark_id": "livecodebench", "score": 84.5, "reference_url": "https://huggingface.co/deepseek-ai/DeepSeek-V3.2/resolve/main/assets/paper.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "DeepSeek V3.2 tech report Table 2: gpt-5=84.5.", "candidates": [ { "score": 85, "reference_url": "https://huggingface.co/MiniMaxAI/MiniMax-M2", "source_type": "model_card", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "notes": "MiniMax M2 model card: gpt-5=85." } ] }, { "model_id": "gpt-5", "benchmark_id": "codeforces_rating", "score": 2537, "reference_url": "https://huggingface.co/deepseek-ai/DeepSeek-V3.2/resolve/main/assets/paper.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "DeepSeek V3.2 tech report Table 2: gpt-5=2537." }, { "model_id": "gpt-5", "benchmark_id": "swe_bench_pro", "score": 41.78, "reference_url": "https://scale.com/leaderboard/swe_bench_pro_public", "audit_status": "verified", "source_type": "leaderboard", "matches_canonical": false, "notes": " [Verified against Scale.com SWE-Bench Pro Public leaderboard (2026-04-29): gpt-5-2025-08-07 (High) = 41.78. Scale uses standardized evaluation scaffolding.]" }, { "model_id": "gpt-5", "benchmark_id": "arc_agi_1", "score": 65.7, "reference_url": "https://arcprize.org/leaderboard", "audit_status": "verified", "source_type": "leaderboard", "matches_canonical": true, "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1 (single-pass)", "judge": "rule-based", "harness": "official (arcprize.org)", "system_type": "CoT" }, "notes": "arcprize.org leaderboard audit: GPT-5 (High) on leaderboard" }, { "model_id": "gpt-5", "benchmark_id": "arc_agi_2", "score": 9.9, "reference_url": "https://arcprize.org/leaderboard", "audit_status": "verified", "source_type": "leaderboard", "matches_canonical": true, "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1 (single-pass)", "judge": "rule-based", "harness": "official (arcprize.org)", "system_type": "CoT" }, "notes": "arcprize.org leaderboard audit: GPT-5 (High) on leaderboard" }, { "model_id": "gpt-5", "benchmark_id": "chatbot_arena_elo", "score": 1460, "reference_url": "https://lmarena.ai/", "audit_status": "needs_review", "audit_note": "lmarena.ai is a dynamic leaderboard; ELO values change over time. Cannot verify static value 1460 from snapshot-dependent source via automated audit.", "audited_by": "lychee-audit-bot", "audited_at": "2026-07-26T00:00:00Z" }, { "model_id": "gpt-5", "benchmark_id": "humaneval", "score": 94.0, "reference_url": "https://cdn.openai.com/gpt-5-system-card.pdf", "audit_status": "dropped", "rule_id": "R5g-ghost", "audit_note": "Not in GPT-5 system card PDF. Card covers safety/hallucination evals; no humaneval table.", "audited_by": "lychee-audit-bot", "audited_at": "2026-04-29T13:09:38Z" }, { "model_id": "gpt-5", "benchmark_id": "ifeval", "score": 90.0, "reference_url": "https://cdn.openai.com/gpt-5-system-card.pdf", "audit_status": "dropped", "rule_id": "R5g-ghost", "audit_note": "Not in GPT-5 system card PDF. Card covers safety/hallucination evals; no ifeval table.", "audited_by": "lychee-audit-bot", "audited_at": "2026-04-29T13:09:38Z" }, { "model_id": "gpt-5", "benchmark_id": "mmlu", "score": 91.0, "reference_url": "https://cdn.openai.com/gpt-5-system-card.pdf", "audit_status": "dropped", "rule_id": "R5g-ghost", "audit_note": "Not in GPT-5 system card PDF. Card covers safety/hallucination evals; no mmlu table.", "audited_by": "lychee-audit-bot", "audited_at": "2026-04-29T13:09:38Z" }, { "model_id": "gpt-5", "benchmark_id": "osworld", "score": 40.0, "reference_url": "https://www.digitalocean.com/resources/articles/gpt-5-overview", "audit_status": "dropped", "rule_id": "R5h-third-party-blog", "notes": "DROPPED (R5h): random third-party blog/aggregator, no primary trail" }, { "model_id": "gpt-5", "benchmark_id": "terminal_bench", "score": 35.2, "reference_url": "https://huggingface.co/deepseek-ai/DeepSeek-V3.2/resolve/main/assets/paper.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "agentic", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "DeepSeek V3.2 tech report Table 2: gpt-5=35.2.", "candidates": [ { "score": 43.8, "reference_url": "https://huggingface.co/MiniMaxAI/MiniMax-M2", "source_type": "model_card", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "agentic", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "notes": "MiniMax M2 model card: gpt-5=43.8." } ] }, { "model_id": "gpt-5", "benchmark_id": "aa_intelligence_index", "score": 69, "reference_url": "https://huggingface.co/MiniMaxAI/MiniMax-M2", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "MiniMax M2 model card: gpt-5=69." }, { "model_id": "gpt-5", "benchmark_id": "aime_2024", "score": 94.6, "reference_url": "https://openai.com/index/introducing-gpt-5/", "audit_status": "dropped", "audit_notes": "R5g ghost cell: introducing-gpt-5/ blog reports only AIME 2025 (94.6%) for gpt-5, not AIME 2024. The 94.6% value duplicates gpt-5/aime_2025 which is already verified separately. Cell benchmark_id mismatch: drop." }, { "model_id": "gpt-5", "benchmark_id": "brumo_2025", "score": 91.67, "reference_url": "https://matharena.ai/?comp=brumo--brumo_2025", "audit_status": "verified", "matches_canonical": true, "reported_setting": "reasoning_effort:high", "source_type": "leaderboard" }, { "model_id": "gpt-5", "benchmark_id": "cmimc_2025", "score": 90, "reference_url": "https://matharena.ai/?comp=cmimc--cmimc_2025", "audit_status": "verified", "matches_canonical": true, "reported_setting": "reasoning_effort:high", "source_type": "leaderboard" }, { "model_id": "gpt-5", "benchmark_id": "critpt", "score": 5.7, "reference_url": "https://github.com/CritPt-Benchmark/CritPt", "audit_status": "verified", "reported_setting": "CritPt leaderboard, ε=1.5, % tasks solved", "source_type": "leaderboard", "matches_canonical": true }, { "model_id": "gpt-5", "benchmark_id": "hmmt_nov_2025", "score": 89.2, "reference_url": "https://huggingface.co/deepseek-ai/DeepSeek-V3.2/resolve/main/assets/paper.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "DeepSeek V3.2 tech report Table 2: gpt-5=89.2." }, { "model_id": "gpt-5", "benchmark_id": "imo_2025", "score": 38.1, "reference_url": "https://matharena.ai/?comp=imo--imo_2025", "audit_status": "verified", "matches_canonical": true, "reported_setting": "reasoning_effort:high", "source_type": "leaderboard" }, { "model_id": "gpt-5", "benchmark_id": "livebench", "score": 71.3, "reference_url": "https://livebench.ai/", "audit_status": "verified", "source_type": "leaderboard", "matches_canonical": false, "reported_setting": { "mode": "thinking", "effort": "n/a", "tools": "n/a" }, "notes": "livebench.ai table_2026_01_08 (latest as of 2026-04-28): gpt-5-pro-2025-10-06 avg=71.3. Matched as closest GPT-5 entry; canonical effort=high not confirmed for this variant (no \"-high\" suffix). See review. BP had 70.5." }, { "model_id": "gpt-5", "benchmark_id": "matharena_apex_2025", "score": 1.04, "reference_url": "https://matharena.ai/?comp=apex--apex_2025", "audit_status": "verified", "matches_canonical": true, "reported_setting": "reasoning_effort:high", "source_type": "leaderboard" }, { "model_id": "gpt-5", "benchmark_id": "simplebench", "score": 61.6, "reference_url": "https://lmcouncil.ai/benchmarks", "audit_status": "dropped", "rule_id": "R5h-third-party-blog", "notes": "DROPPED (R5h): random third-party blog/aggregator, no primary trail | DROPPED (R5g-c stale, never verified, source not audited in any pass)" }, { "model_id": "gpt-5", "benchmark_id": "smt_2025", "score": 91.98, "reference_url": "https://matharena.ai/?comp=smt--smt_2025", "audit_status": "verified", "matches_canonical": true, "reported_setting": "reasoning_effort:high", "source_type": "leaderboard" }, { "model_id": "gpt-5", "benchmark_id": "terminal_bench_1", "score": 41.3, "reference_url": "https://www.tbench.ai/leaderboard/terminal-bench/1.0", "audit_status": "verified", "source_type": "leaderboard", "rule_ids": [ "R5d" ], "notes": "Confirmed 41.3% on tbench.ai Terminal-Bench 1.0 leaderboard (Terminus 2 scaffold, 2025-08-11)." }, { "model_id": "gpt-oss-120b", "benchmark_id": "aime_2025", "score": 97.9, "reference_url": "https://arxiv.org/abs/2508.10925", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "with tools", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "harmony", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "gpt-oss model card Table 3 (high effort): AIME 2025 with-tools 97.9", "candidates": [ { "score": 92.5, "reference_url": "https://llm-stats.com/benchmarks/aime-2025", "source_type": "third_party_aggregator", "reported_setting": { "effort": "high" }, "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=GPT OSS 120B High, slug=gpt-oss-120b-high, provider=OpenAI" }, { "score": 89, "reference_url": "https://huggingface.co/mistralai/Mistral-Small-4-119B-2603/resolve/a11f36bebf709121056b1dbcc943d1c6afbe494d/images/livecode.png?download=true", "source_type": "official_model_card", "reported_setting": { "mode": "thinking", "effort": "chart reasoning mode; exact effort level undisclosed", "tools": "none", "sampling": "pass@1; repeat count undisclosed", "judge": "benchmark-specified", "harness": "Mistral official model-card evaluation", "prompt_style": "default", "temperature": "default", "context": "default (128K)" }, "notes": "Displayed exactly: '89'. Research observation: small-livecode.png:1:2:score:reasoning." }, { "score": 45, "reference_url": "https://huggingface.co/mistralai/Mistral-Small-4-119B-2603/resolve/a11f36bebf709121056b1dbcc943d1c6afbe494d/images/livecode.png?download=true", "source_type": "official_model_card", "reported_setting": { "mode": "non-thinking", "effort": "none", "tools": "none", "sampling": "pass@1; repeat count undisclosed", "judge": "benchmark-specified", "harness": "Mistral official model-card evaluation", "prompt_style": "default", "temperature": "default", "context": "default (128K)" }, "notes": "Displayed exactly: '45'. Research observation: small-livecode.png:1:2:score:instruct." } ] }, { "model_id": "gpt-oss-120b", "benchmark_id": "gpqa_diamond", "score": 80.9, "reference_url": "https://arxiv.org/abs/2508.10925", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "with tools", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "harmony", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "gpt-oss model card Table 3 (high effort): GPQA Diamond with-tools 80.9", "candidates": [ { "score": 80.9, "reference_url": "https://llm-stats.com/benchmarks/gpqa", "source_type": "third_party_aggregator", "reported_setting": { "effort": "high" }, "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=GPT OSS 120B High, slug=gpt-oss-120b-high, provider=OpenAI" } ] }, { "model_id": "gpt-oss-120b", "benchmark_id": "mmlu_pro", "score": 90.0, "reference_url": "https://www.clarifai.com/blog/openai-gpt-oss-benchmarks", "audit_status": "dropped", "notes": " | DROPPED: BP value 90.0 from clarifai blog matches gpt-oss-120b MMLU=90.0 in Table 3 — clarifai mislabeled MMLU as MMLU-Pro. Table 3 does not report MMLU-Pro for gpt-oss. Ghost cell per R5g(b).", "candidates": [ { "score": 80.7, "reference_url": "https://llm-stats.com/benchmarks/mmlu-pro", "source_type": "third_party_aggregator", "reported_setting": { "effort": "high" }, "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=GPT OSS 120B High, slug=gpt-oss-120b-high, provider=OpenAI" } ] }, { "model_id": "gpt-oss-120b", "benchmark_id": "codeforces_rating", "score": 2622, "reference_url": "https://arxiv.org/abs/2508.10925", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "with tools", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "harmony", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "gpt-oss model card Table 3 (high effort): Codeforces with-tools high 2622 (was 2620). [R5d: prior unverified value 2620 from https://smythos.com/developers/ai-models/openai-gpt-oss-120b-and-20b/ deleted.]" }, { "model_id": "gpt-oss-120b", "benchmark_id": "tau_bench_retail", "score": 67.8, "reference_url": "https://arxiv.org/abs/2508.10925", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "with tools", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "harmony", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "gpt-oss model card Table 3 (high effort): Tau-Bench Retail with-tools 67.8" }, { "model_id": "gpt-oss-120b", "benchmark_id": "mmlu", "score": 90.0, "reference_url": "https://arxiv.org/abs/2508.10925", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "with tools", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "harmony", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "gpt-oss model card Table 3 (high effort): MMLU 90.0" }, { "model_id": "gpt-oss-120b", "benchmark_id": "swe_bench_verified", "score": 62.4, "reference_url": "https://arxiv.org/abs/2508.10925", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "with tools", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "harmony", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "gpt-oss model card Table 3 (high effort): SWE-Bench Verified 62.4" }, { "model_id": "gpt-oss-120b", "benchmark_id": "aime_2024", "score": 96.6, "reference_url": "https://arxiv.org/abs/2508.10925", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "with tools", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "harmony", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "gpt-oss model card Table 3 (high effort): AIME 2024 with-tools 96.6" }, { "model_id": "gpt-oss-120b", "benchmark_id": "humaneval", "score": 92.0, "reference_url": "https://arxiv.org/abs/2508.10925", "audit_status": "dropped", "rule_id": "R5g-ghost", "audit_note": "arxiv 2508.10925 (gpt-oss model card) has no humaneval evaluation. Paper covers MMLU/SWE-Bench/Tau-Bench/HLE/HealthBench/MMMLU.", "audited_by": "lychee-audit-bot", "audited_at": "2026-04-29T13:12:50Z" }, { "model_id": "gpt-oss-120b", "benchmark_id": "ifeval", "score": 88.0, "reference_url": "https://arxiv.org/abs/2508.10925", "audit_status": "dropped", "rule_id": "R5g-ghost", "audit_note": "arxiv 2508.10925 (gpt-oss model card) has no ifeval evaluation. Paper covers MMLU/SWE-Bench/Tau-Bench/HLE/HealthBench/MMMLU.", "audited_by": "lychee-audit-bot", "audited_at": "2026-04-29T13:12:50Z" }, { "model_id": "gpt-oss-120b", "benchmark_id": "livecodebench", "score": 75.0, "reference_url": "https://arxiv.org/abs/2508.10925", "audit_status": "dropped", "rule_id": "R5g-ghost", "audit_note": "arxiv 2508.10925 (gpt-oss model card) has no livecodebench evaluation. Paper covers MMLU/SWE-Bench/Tau-Bench/HLE/HealthBench/MMMLU.", "audited_by": "lychee-audit-bot", "audited_at": "2026-04-29T13:12:50Z" }, { "model_id": "gpt-oss-120b", "benchmark_id": "math_500", "score": 98.0, "reference_url": "https://www.clarifai.com/blog/openai-gpt-oss-benchmarks", "audit_status": "dropped", "rule_id": "R5h-third-party-blog", "notes": "DROPPED (R5h): random third-party blog/aggregator, no primary trail" }, { "model_id": "gpt-oss-120b", "benchmark_id": "terminal_bench", "score": 18.7, "reference_url": "https://www.tbench.ai/leaderboard/terminal-bench/2.0", "audit_status": "verified", "source_type": "leaderboard", "rule_ids": [ "R5d" ], "notes": "Confirmed 18.7% on tbench.ai Terminal-Bench 2.0 leaderboard (Terminus 2 scaffold, 2025-11-01)." }, { "model_id": "gpt-oss-20b", "benchmark_id": "aime_2025", "score": 98.7, "reference_url": "https://arxiv.org/abs/2508.10925", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "with tools", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "harmony", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "gpt-oss model card Table 3 (high effort): AIME 2025 with-tools 98.7", "candidates": [ { "score": 98.7, "reference_url": "https://llm-stats.com/benchmarks/aime-2025", "source_type": "third_party_aggregator", "reported_setting": { "effort": "high" }, "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=GPT OSS 20B High, slug=gpt-oss-20b-high, provider=OpenAI" }, { "score": 68.53, "reference_url": "https://www.liquid.ai/blog/lfm2-5-8b-a1b", "source_type": "official_blog", "reported_setting": { "mode": "source/model default", "effort": "source does not state", "tools": "none", "sampling": "source does not state trial count", "judge": "benchmark-specified", "harness": "Liquid AI official comparative evaluation", "prompt_style": "source does not state", "temperature": "source does not state evaluation parameters", "context": "source does not state evaluation context" }, "notes": "Displayed exactly as 68.53. Official Liquid AI reported result." } ] }, { "model_id": "gpt-oss-20b", "benchmark_id": "gpqa_diamond", "score": 71.5, "reference_url": "https://arxiv.org/abs/2508.10925", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "harmony", "temperature": "default" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "gpt-oss model card Table 3 (high effort): GPQA Diamond no-tools 71.5", "candidates": [ { "score": 74.2, "reference_url": "https://llm-stats.com/benchmarks/gpqa", "source_type": "third_party_aggregator", "reported_setting": { "effort": "high" }, "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=GPT OSS 20B High, slug=gpt-oss-20b-high, provider=OpenAI" } ] }, { "model_id": "gpt-oss-20b", "benchmark_id": "mmlu_pro", "score": 85.3, "reference_url": "https://www.clarifai.com/blog/openai-gpt-oss-benchmarks", "audit_status": "dropped", "notes": " | DROPPED: BP value 85.3 from clarifai blog matches gpt-oss-20b MMLU=85.3 in Table 3 — clarifai mislabeled MMLU as MMLU-Pro. Ghost cell per R5g(b)." }, { "model_id": "gpt-oss-20b", "benchmark_id": "mmlu", "score": 85.3, "reference_url": "https://arxiv.org/abs/2508.10925", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "with tools", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "harmony", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "gpt-oss model card Table 3 (high effort): MMLU 85.3" }, { "model_id": "gpt-oss-20b", "benchmark_id": "ifeval", "score": 86.73, "reference_url": "https://www.liquid.ai/blog/lfm2-5-8b-a1b", "reported_setting": { "mode": "source/model default", "effort": "source does not state", "tools": "none", "sampling": "source does not state trial count", "judge": "benchmark-specified", "harness": "Liquid AI official comparative evaluation", "prompt_style": "source does not state", "temperature": "source does not state evaluation parameters", "context": "source does not state evaluation context" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Displayed exactly as 86.73. Official Liquid AI reported result." }, { "model_id": "gpt-oss-20b", "benchmark_id": "livecodebench", "score": 70.0, "reference_url": "https://arxiv.org/abs/2508.10925", "audit_status": "dropped", "rule_id": "R5g-ghost", "audit_note": "arxiv 2508.10925 (gpt-oss model card) has no livecodebench evaluation. Paper covers MMLU/SWE-Bench/Tau-Bench/HLE/HealthBench/MMMLU.", "audited_by": "lychee-audit-bot", "audited_at": "2026-04-29T13:12:50Z" }, { "model_id": "gpt-oss-20b", "benchmark_id": "math_500", "score": 92.4, "reference_url": "https://www.liquid.ai/blog/lfm2-5-8b-a1b", "reported_setting": { "mode": "source/model default", "effort": "source does not state", "tools": "none", "sampling": "source does not state trial count", "judge": "benchmark-specified", "harness": "Liquid AI official comparative evaluation", "prompt_style": "source does not state", "temperature": "source does not state evaluation parameters", "context": "source does not state evaluation context" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Displayed exactly as 92.40. Official Liquid AI reported result." }, { "model_id": "gpt-oss-20b", "benchmark_id": "terminal_bench", "score": 3.1, "reference_url": "https://www.tbench.ai/leaderboard/terminal-bench/2.0", "audit_status": "verified", "source_type": "leaderboard", "rule_ids": [ "R5d" ], "notes": "Confirmed 3.1% on tbench.ai Terminal-Bench 2.0 leaderboard (Terminus 2 scaffold, 2025-11-01)." }, { "model_id": "gpt-oss-20b", "benchmark_id": "aime_2024", "score": 96.0, "reference_url": "https://arxiv.org/abs/2508.10925", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "with tools", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "harmony", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "gpt-oss model card Table 3 (high effort): AIME 2024 with-tools 96.0 (was 98.7 from llm-stats — wrong, that was AIME 2025 score). [R5d: prior unverified value 98.7 from https://llm-stats.com/models/compare/gpt-oss-120b-vs-gpt-oss-20b deleted.]" }, { "model_id": "gpt-oss-20b", "benchmark_id": "codeforces_rating", "score": 2516, "reference_url": "https://arxiv.org/abs/2508.10925", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "with tools", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "harmony", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "gpt-oss model card Table 3 (high effort): Codeforces with-tools high 2516 (was 1985 — likely no-tools or low effort). [R5d: prior unverified value 1985 from https://arxiv.org/html/2508.10925v1 deleted.]" }, { "model_id": "gpt-oss-20b", "benchmark_id": "humaneval", "score": 85, "reference_url": "https://arxiv.org/html/2508.10925v1", "audit_status": "dropped", "rule_id": "R5g-ghost", "audit_note": "arxiv 2508.10925 (gpt-oss model card) has no humaneval evaluation. Paper covers MMLU/SWE-Bench/Tau-Bench/HLE/HealthBench/MMMLU.", "audited_by": "lychee-audit-bot", "audited_at": "2026-04-29T13:12:50Z" }, { "model_id": "gpt-oss-20b", "benchmark_id": "swe_bench_verified", "score": 60.7, "reference_url": "https://arxiv.org/abs/2508.10925", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "with tools", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "harmony", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "gpt-oss model card Table 3 (high effort): SWE-Bench Verified with-tools high 60.7 (was 52 from clarifai — likely medium effort). [R5d: prior unverified value 52 from https://www.clarifai.com/blog/openai-gpt-oss-benchmarks-how-it-compares-to-glm-4.5-qwen3-deepseek-and-kimi-k2 deleted.]" }, { "model_id": "gpt-5.1", "benchmark_id": "aime_2025", "score": 94.0, "reference_url": "https://openai.com/index/introducing-gpt-5-2/", "candidates": [ { "score": 99.6, "reference_url": "https://llm-stats.com/benchmarks/aime-2025", "source_type": "third_party_aggregator", "reported_setting": { "effort": "high" }, "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=GPT-5.1 High, slug=gpt-5.1-high-2025-11-12, provider=OpenAI" }, { "score": 98.4, "reference_url": "https://llm-stats.com/benchmarks/aime-2025", "source_type": "third_party_aggregator", "reported_setting": { "effort": "medium" }, "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=GPT-5.1 Medium, slug=gpt-5.1-medium-2025-11-12, provider=OpenAI" }, { "score": 94.0, "reference_url": "https://llm-stats.com/benchmarks/aime-2025", "source_type": "third_party_aggregator", "reported_setting": { "effort": "instant" }, "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=GPT-5.1 Instant, slug=gpt-5.1-instant-2025-11-12, provider=OpenAI" }, { "score": 94.0, "reference_url": "https://llm-stats.com/benchmarks/aime-2025", "source_type": "third_party_aggregator", "reported_setting": { "effort": "thinking" }, "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=GPT-5.1 Thinking, slug=gpt-5.1-thinking-2025-11-12, provider=OpenAI" } ], "reported_setting": { "mode": "thinking", "effort": "xhigh", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-5.2 blog (cross-model column): GPT-5.1 = 94.0." }, { "model_id": "gpt-5.1", "benchmark_id": "gpqa_diamond", "score": 88.1, "reference_url": "https://openai.com/index/introducing-gpt-5-2/", "candidates": [ { "score": 88.1, "reference_url": "https://llm-stats.com/benchmarks/gpqa", "source_type": "third_party_aggregator", "reported_setting": { "effort": "high" }, "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=GPT-5.1 High, slug=gpt-5.1-high-2025-11-12, provider=OpenAI" }, { "score": 88.1, "reference_url": "https://llm-stats.com/benchmarks/gpqa", "source_type": "third_party_aggregator", "reported_setting": { "effort": "instant" }, "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=GPT-5.1 Instant, slug=gpt-5.1-instant-2025-11-12, provider=OpenAI" }, { "score": 88.1, "reference_url": "https://llm-stats.com/benchmarks/gpqa", "source_type": "third_party_aggregator", "reported_setting": { "effort": "thinking" }, "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=GPT-5.1 Thinking, slug=gpt-5.1-thinking-2025-11-12, provider=OpenAI" } ], "reported_setting": { "mode": "thinking", "effort": "xhigh", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-5.2 blog (cross-model column): GPT-5.1 = 88.1." }, { "model_id": "gpt-5.1", "benchmark_id": "frontiermath", "score": 31.0, "reference_url": "https://openai.com/index/introducing-gpt-5-2/", "candidates": [ { "score": 26.7, "reference_url": "https://openai.com/index/gpt-5-1-for-developers/", "source_type": "official_blog", "notes": "gpt-5.1 blog says FrontierMath w/ Python = 26.7%. Primary=31.0 from gpt-5.2 blog (Tier 1-3 w/ Python). Possible scope difference." }, { "score": 26.7, "reference_url": "https://llm-stats.com/benchmarks/frontiermath", "source_type": "third_party_aggregator", "reported_setting": { "effort": "instant" }, "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=GPT-5.1 Instant, slug=gpt-5.1-instant-2025-11-12, provider=OpenAI" }, { "score": 26.7, "reference_url": "https://llm-stats.com/benchmarks/frontiermath", "source_type": "third_party_aggregator", "reported_setting": { "effort": "thinking" }, "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=GPT-5.1 Thinking, slug=gpt-5.1-thinking-2025-11-12, provider=OpenAI" } ], "reported_setting": { "mode": "thinking", "effort": "xhigh", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-5.2 blog (cross-model column): GPT-5.1 = 31.0." }, { "model_id": "gpt-5.1", "benchmark_id": "arc_agi_2", "score": 17.6, "reference_url": "https://openai.com/index/introducing-gpt-5-2/", "candidates": [], "reported_setting": { "mode": "thinking", "effort": "xhigh", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-5.2 blog (cross-model column): GPT-5.1 = 17.6." }, { "model_id": "gpt-5.1", "benchmark_id": "swe_bench_verified", "score": 76.3, "reference_url": "https://openai.com/index/introducing-gpt-5-2/", "candidates": [ { "score": 76.3, "reference_url": "https://llm-stats.com/benchmarks/swe-bench-verified", "source_type": "third_party_aggregator", "reported_setting": { "effort": "instant" }, "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=GPT-5.1 Instant, slug=gpt-5.1-instant-2025-11-12, provider=OpenAI" }, { "score": 76.3, "reference_url": "https://llm-stats.com/benchmarks/swe-bench-verified", "source_type": "third_party_aggregator", "reported_setting": { "effort": "thinking" }, "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=GPT-5.1 Thinking, slug=gpt-5.1-thinking-2025-11-12, provider=OpenAI" } ], "reported_setting": { "mode": "thinking", "effort": "xhigh", "tools": "agentic (bash + edit)", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-5.2 blog (cross-model column): GPT-5.1 = 76.3." }, { "model_id": "gpt-5.1", "benchmark_id": "aa_lcr", "score": 75.0, "reference_url": "https://artificialanalysis.ai/evaluations/artificial-analysis-long-context-reasoning", "audit_status": "verified", "source_type": "third_party", "notes": "ArtificialAnalysis aggregator; used as fallback when no primary source available (R5d-aggregator-fallback). AA LCR eval page: gpt-5-1 lcr=0.750 → 75.0%. AA slug: gpt-5-1.", "reported_setting": { "mode": "thinking", "effort": "default", "sampling": "pass@1", "harness": "AA standard evaluation", "notes": "Per AA evaluations/artificial-analysis-long-context-reasoning: gpt-5-1 lcr=75.0%" } }, { "model_id": "gpt-5.1", "benchmark_id": "chatbot_arena_elo", "score": 1439, "reference_url": "https://lmarena.ai/leaderboard/text", "audit_status": "verified", "source_type": "leaderboard", "matches_canonical": true, "notes": "lmarena.ai text arena (fetched 2026-04-29). Arena model name: gpt-5.1 (rank 52). ELO updates continuously; score reflects latest available." }, { "model_id": "gpt-5.1", "benchmark_id": "mmmu_pro", "score": 85.4, "reference_url": "https://www.vellum.ai/blog/gpt-5-2-benchmarks", "audit_status": "dropped", "rule_id": "R5h-third-party-blog", "notes": "DROPPED (R5h): random third-party blog/aggregator, no primary trail" }, { "model_id": "gpt-5.1", "benchmark_id": "browsecomp", "score": 50.8, "reference_url": "https://openai.com/index/introducing-gpt-5-2/", "candidates": [ { "score": 60.0, "reference_url": "https://www.vellum.ai/blog/gpt-5-2-benchmarks", "source_type": "third_party", "reported_setting": {}, "notes": "previous primary (demoted: superseded by GPT-5.2 own blog)" } ], "reported_setting": { "mode": "thinking", "effort": "xhigh", "tools": "agentic", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-5.2 blog (cross-model column): GPT-5.1 = 50.8." }, { "model_id": "gpt-5.1", "benchmark_id": "hle", "score": 25.7, "reference_url": "https://openai.com/index/introducing-gpt-5-2/", "candidates": [ { "score": 32.0, "reference_url": "https://www.vellum.ai/blog/flagship-model-report", "source_type": "third_party", "reported_setting": {}, "notes": "previous primary (demoted: superseded by GPT-5.2 own blog)" } ], "reported_setting": { "mode": "thinking", "effort": "xhigh", "tools": "none (no-tools column)", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-5.2 blog (cross-model column): GPT-5.1 = 25.7." }, { "model_id": "gpt-5.1", "benchmark_id": "livecodebench", "score": 82.0, "reference_url": "https://www.vellum.ai/blog/flagship-model-report", "audit_status": "dropped", "rule_id": "R5h-third-party-blog", "notes": "DROPPED (R5h): random third-party blog/aggregator, no primary trail" }, { "model_id": "gpt-5.1", "benchmark_id": "math_500", "score": 99.0, "reference_url": "https://artificialanalysis.ai/evaluations/math-500", "audit_status": "needs_review", "notes": "Reference URL is AA math-500 eval page, but gpt-5-1 (gpt-5.1) is NOT present on that page; AA shows math_500=null for this model. Possible ghost score (R5g-a). Cannot verify 99.0% from AA source." }, { "model_id": "gpt-5.1", "benchmark_id": "matharena_apex_2025", "score": 1.04, "reference_url": "https://matharena.ai/?comp=apex--apex_2025", "audit_status": "verified", "matches_canonical": true, "reported_setting": "reasoning_effort:high", "source_type": "leaderboard" }, { "model_id": "gpt-5.1", "benchmark_id": "mmlu", "score": 90.0, "reference_url": "https://www.vellum.ai/blog/gpt-5-2-benchmarks", "audit_status": "dropped", "rule_id": "R5h-third-party-blog", "notes": "DROPPED (R5h): random third-party blog/aggregator, no primary trail" }, { "model_id": "gpt-5.1", "benchmark_id": "mmmu", "score": 85.4, "reference_url": "https://openai.com/index/gpt-5-1-for-developers/", "candidates": [ { "score": 84.2, "reference_url": "https://automatio.ai/models/gpt-5-3-codex", "source_type": "third_party", "reported_setting": {}, "notes": " (demoted by 5.1 dev blog)" }, { "score": 85.4, "reference_url": "https://llm-stats.com/benchmarks/mmmu", "source_type": "third_party_aggregator", "reported_setting": { "effort": "thinking" }, "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=GPT-5.1 Thinking, slug=gpt-5.1-thinking-2025-11-12, provider=OpenAI" }, { "score": 85.4, "reference_url": "https://llm-stats.com/benchmarks/mmmu", "source_type": "third_party_aggregator", "reported_setting": { "effort": "instant" }, "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=GPT-5.1 Instant, slug=gpt-5.1-instant-2025-11-12, provider=OpenAI" } ], "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official with apply_patch", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-5.1 dev blog: 85.4 (high reasoning effort)." }, { "model_id": "gpt-5.1", "benchmark_id": "simpleqa", "score": 55.0, "reference_url": "https://www.vellum.ai/blog/gpt-5-2-benchmarks", "audit_status": "dropped", "rule_id": "R5h-third-party-blog", "notes": "DROPPED (R5h): random third-party blog/aggregator, no primary trail" }, { "model_id": "gpt-5.1", "benchmark_id": "swe_bench_pro", "score": 50.8, "reference_url": "https://openai.com/index/introducing-gpt-5-2/", "candidates": [ { "score": 45.0, "reference_url": "https://www.vellum.ai/blog/flagship-model-report", "source_type": "third_party", "reported_setting": {}, "notes": "previous primary (demoted: superseded by GPT-5.2 own blog)" } ], "reported_setting": { "mode": "thinking", "effort": "xhigh", "tools": "agentic (bash + edit)", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-5.2 blog (cross-model column): GPT-5.1 = 50.8." }, { "model_id": "gpt-5.1", "benchmark_id": "terminal_bench", "score": 47.6, "reference_url": "https://z.ai/blog/glm-4.7", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "agentic", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "GLM-4.7 blog: gpt-5.1=47.6." }, { "model_id": "gpt-5.1", "benchmark_id": "arc_agi_1", "score": 72.8, "reference_url": "https://openai.com/index/introducing-gpt-5-2/", "candidates": [], "reported_setting": { "mode": "thinking", "effort": "xhigh", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-5.2 blog (cross-model column): GPT-5.1 = 72.8." }, { "model_id": "gpt-5.1", "benchmark_id": "brumo_2025", "score": 93.33, "reference_url": "https://matharena.ai/?comp=brumo--brumo_2025", "audit_status": "verified", "matches_canonical": true, "reported_setting": "reasoning_effort:high", "source_type": "leaderboard" }, { "model_id": "gpt-5.1", "benchmark_id": "cmimc_2025", "score": 91.88, "reference_url": "https://matharena.ai/?comp=cmimc--cmimc_2025", "audit_status": "verified", "matches_canonical": true, "reported_setting": "reasoning_effort:high", "source_type": "leaderboard" }, { "model_id": "gpt-5.1", "benchmark_id": "hmmt_nov_2025", "score": 91.67, "reference_url": "https://matharena.ai/?comp=hmmt--hmmt_nov_2025", "audit_status": "verified", "matches_canonical": true, "reported_setting": "reasoning_effort:high", "source_type": "leaderboard" }, { "model_id": "gpt-5.1", "benchmark_id": "mmlu_pro", "score": 87.0, "reference_url": "https://z.ai/blog/glm-4.7", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "GLM-4.7 blog: gpt-5.1=87.0." }, { "model_id": "gpt-5.1", "benchmark_id": "smt_2025", "score": 91.04, "reference_url": "https://matharena.ai/?comp=smt--smt_2025", "audit_status": "verified", "matches_canonical": true, "reported_setting": "reasoning_effort:high", "source_type": "leaderboard" }, { "model_id": "gpt-5.2", "benchmark_id": "aime_2025", "score": 100.0, "reference_url": "https://openai.com/index/introducing-gpt-5-2/", "candidates": [ { "score": 100.0, "reference_url": "https://deepmind.google/models/gemini/flash/", "source_type": "official_blog", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "notes": "DeepMind /models/gemini/flash/ Performance table: AIME 2025 no tools 100%" }, { "score": 98.0, "reference_url": "https://huggingface.co/MiniMaxAI/MiniMax-M2.5", "source_type": "model_card", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "notes": "MiniMax M2.5 model card: gpt-5.2=98.0." }, { "score": 99.0, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "source_type": "model_card", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "notes": "Doubao Seed 2.0 model card Table 3: gpt-5.2=99.0." }, { "score": 100.0, "reference_url": "https://llm-stats.com/benchmarks/aime-2025", "source_type": "third_party_aggregator", "reported_setting": { "effort": "pro" }, "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=GPT-5.2 Pro, slug=gpt-5.2-pro-2025-12-11, provider=OpenAI" }, { "score": 100.0, "reference_url": "https://microsoft.ai/models/mai-thinking-1/", "source_type": "official_model_card", "reported_setting": { "mode": "thinking", "effort": "xhigh", "tools": "none (per-bench override)", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default (with prompt adjustment for τ²-bench)", "temperature": "default", "context": "default", "notes": "Per OpenAI gpt-5.2 blog: reasoning effort=xhigh, research environment." }, "notes": "Displayed exactly as 100.0. Official Microsoft-reported result. Comparator value is quoted from that provider's official model card/release; use the campaign canonical setting." } ], "reported_setting": { "mode": "thinking", "effort": "xhigh", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-5.2 blog: 100.0 (xhigh reasoning effort)." }, { "model_id": "gpt-5.2", "benchmark_id": "gpqa_diamond", "score": 92.4, "reference_url": "https://openai.com/index/introducing-gpt-5-2/", "candidates": [ { "score": 93.7, "reference_url": "https://www.anthropic.com/news/claude-sonnet-4-6", "source_type": "third_party", "reported_setting": { "note": "see primary" }, "notes": "Reported by Anthropic Sonnet 4.6 blog (third-party)." }, { "score": 92.4, "reference_url": "https://deepmind.google/models/gemini/pro/", "source_type": "official_blog", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "notes": "DeepMind /models/gemini/pro/ cross-model Performance table: GPQA Diamond no tools 92.4% (Thinking xhigh)" }, { "score": 92.4, "reference_url": "https://deepmind.google/models/gemini/flash/", "source_type": "official_blog", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "notes": "DeepMind /models/gemini/flash/ Performance table: GPQA Diamond 92.4%" }, { "score": 90.0, "reference_url": "https://huggingface.co/MiniMaxAI/MiniMax-M2.5", "source_type": "model_card", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "notes": "MiniMax M2.5 model card: gpt-5.2=90.0." }, { "score": 93.2, "reference_url": "https://llm-stats.com/benchmarks/gpqa", "source_type": "third_party_aggregator", "reported_setting": { "effort": "pro" }, "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=GPT-5.2 Pro, slug=gpt-5.2-pro-2025-12-11, provider=OpenAI" }, { "score": 92.4, "reference_url": "https://microsoft.ai/models/mai-thinking-1/", "source_type": "official_model_card", "reported_setting": { "mode": "thinking", "effort": "xhigh", "tools": "none (per-bench override)", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default (with prompt adjustment for τ²-bench)", "temperature": "default", "context": "default", "notes": "Per OpenAI gpt-5.2 blog: reasoning effort=xhigh, research environment." }, "notes": "Displayed exactly as 92.4. Official Microsoft-reported result. Comparator value is quoted from that provider's official model card/release; use the campaign canonical setting." } ], "reported_setting": { "mode": "thinking", "effort": "xhigh", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-5.2 blog: 92.4 (xhigh reasoning effort)." }, { "model_id": "gpt-5.2", "benchmark_id": "frontiermath", "score": 40.3, "reference_url": "https://openai.com/index/introducing-gpt-5-2/", "reported_setting": { "mode": "thinking", "effort": "xhigh", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-5.2 blog: 40.3 (xhigh reasoning effort).", "candidates": [ { "score": 40.7, "reference_url": "https://openai.com/index/introducing-gpt-5-4/", "source_type": "official_blog", "reported_setting": { "mode": "thinking", "effort": "xhigh", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "notes": "OpenAI GPT-5.4 blog table: 40.7. (demoted: superseded by GPT-5.2 own blog)" } ] }, { "model_id": "gpt-5.2", "benchmark_id": "swe_bench_verified", "score": 80.0, "reference_url": "https://openai.com/index/introducing-gpt-5-2/", "candidates": [ { "score": 77.9, "reference_url": "https://www.anthropic.com/news/claude-sonnet-4-6", "source_type": "third_party", "reported_setting": { "note": "see primary" }, "notes": "Reported by Anthropic Sonnet 4.6 blog (third-party, Codex-Max)." }, { "score": 80.0, "reference_url": "https://deepmind.google/models/gemini/pro/", "source_type": "official_blog", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "notes": "DeepMind /models/gemini/pro/ cross-model Performance table: SWE-Bench Verified single attempt 80.0%" }, { "score": 80.0, "reference_url": "https://deepmind.google/models/gemini/flash/", "source_type": "official_blog", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "notes": "DeepMind /models/gemini/flash/ Performance table: SWE-bench Verified 80.0%" }, { "score": 80.0, "reference_url": "https://microsoft.ai/models/mai-thinking-1/", "source_type": "official_model_card", "reported_setting": { "mode": "thinking", "effort": "xhigh", "tools": "none (per-bench override)", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default (with prompt adjustment for τ²-bench)", "temperature": "default", "context": "default", "notes": "Per OpenAI gpt-5.2 blog: reasoning effort=xhigh, research environment." }, "notes": "Displayed exactly as 80.0. Official Microsoft-reported result. Comparator value is quoted from that provider's official model card/release; use the campaign canonical setting." } ], "reported_setting": { "mode": "thinking", "effort": "xhigh", "tools": "agentic (bash + edit)", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-5.2 blog: 80.0 (xhigh reasoning effort)." }, { "model_id": "gpt-5.2", "benchmark_id": "swe_bench_pro", "score": 55.6, "reference_url": "https://openai.com/index/introducing-gpt-5-2/", "reported_setting": { "mode": "thinking", "effort": "xhigh", "tools": "agentic (bash + edit)", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-5.2 blog: 55.6 (xhigh reasoning effort).", "candidates": [ { "score": 55.6, "reference_url": "https://openai.com/index/introducing-gpt-5-3-codex/", "source_type": "official_blog", "reported_setting": { "mode": "thinking", "effort": "xhigh", "tools": "agentic (bash + edit)", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "notes": "OpenAI GPT-5.3-Codex blog: 55.6 (xhigh reasoning effort)." }, { "score": 55.6, "reference_url": "https://deepmind.google/models/gemini/pro/", "source_type": "official_blog", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "notes": "DeepMind /models/gemini/pro/ cross-model Performance table: SWE-Bench Pro single attempt 55.6%" }, { "score": 55.6, "reference_url": "https://microsoft.ai/models/mai-thinking-1/", "source_type": "official_model_card", "reported_setting": { "mode": "thinking", "effort": "xhigh", "tools": "none (per-bench override)", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default (with prompt adjustment for τ²-bench)", "temperature": "default", "context": "default", "notes": "Per OpenAI gpt-5.2 blog: reasoning effort=xhigh, research environment." }, "notes": "Displayed exactly as 55.6. Official Microsoft-reported result. Comparator value is quoted from that provider's official model card/release; use the campaign canonical setting." } ] }, { "model_id": "gpt-5.2", "benchmark_id": "arc_agi_2", "score": 52.9, "reference_url": "https://openai.com/index/introducing-gpt-5-2/", "reported_setting": { "mode": "thinking", "effort": "xhigh", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-5.2 blog: 52.9 (xhigh reasoning effort).", "candidates": [ { "score": 52.9, "reference_url": "https://deepmind.google/models/gemini/pro/", "source_type": "official_blog", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "notes": "DeepMind /models/gemini/pro/ cross-model Performance table: ARC-AGI-2 ARC Prize Verified 52.9% (Thinking xhigh)" }, { "score": 52.9, "reference_url": "https://deepmind.google/models/gemini/flash/", "source_type": "official_blog", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "notes": "DeepMind /models/gemini/flash/ Performance table: ARC-AGI-2 ARC Prize Verified 52.9%" }, { "score": 57.5, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "source_type": "model_card", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "notes": "Doubao Seed 2.0 model card Table 3: gpt-5.2=57.5." } ] }, { "model_id": "gpt-5.2", "benchmark_id": "browsecomp", "score": 65.8, "reference_url": "https://openai.com/index/introducing-gpt-5-2/", "reported_setting": { "mode": "thinking", "effort": "xhigh", "tools": "agentic", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-5.2 blog: 65.8 (xhigh reasoning effort).", "candidates": [ { "score": 65.8, "reference_url": "https://deepmind.google/models/gemini/pro/", "source_type": "official_blog", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "web+code", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "notes": "DeepMind /models/gemini/pro/ cross-model Performance table: BrowseComp Search+Python+Browse 65.8%" }, { "score": 77.9, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "source_type": "model_card", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "varies", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "notes": "Doubao Seed 2.0 model card: gpt-5.2=77.9." }, { "score": 77.9, "reference_url": "https://llm-stats.com/benchmarks/browsecomp", "source_type": "third_party_aggregator", "reported_setting": { "effort": "pro" }, "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=GPT-5.2 Pro, slug=gpt-5.2-pro-2025-12-11, provider=OpenAI" } ] }, { "model_id": "gpt-5.2", "benchmark_id": "critpt", "score": 11.6, "reference_url": "https://artificialanalysis.ai/evaluations/critpt", "audit_status": "verified", "source_type": "third_party", "notes": "ArtificialAnalysis aggregator; used as fallback when no primary source available (R5d-aggregator-fallback). AA CritPT eval page: gpt-5-2 critpt=0.116 → 11.6%. AA slug: gpt-5-2.", "reported_setting": { "mode": "thinking", "effort": "xhigh (AA default for gpt-5-2)", "sampling": "pass@1", "harness": "AA standard evaluation", "notes": "Per AA evaluations/critpt: gpt-5-2 critpt=11.6%" } }, { "model_id": "gpt-5.2", "benchmark_id": "scicode", "score": 52.0, "reference_url": "https://deepmind.google/models/gemini/pro/", "candidates": [ { "score": 52.1, "reference_url": "https://huggingface.co/moonshotai/Kimi-K2.5", "source_type": "model_card", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "1.0" }, "notes": "Kimi K2.5 model card: gpt-5.2=52.1." }, { "score": 49.7, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "source_type": "model_card", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "varies", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "notes": "Doubao Seed 2.0 model card: gpt-5.2=49.7." } ], "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "source_type": "official_blog", "audit_status": "verified", "matches_canonical": true, "notes": "DeepMind /models/gemini/pro/ Performance table (PROMOTED from candidate; prior unverified value 54.6 from https://artificialanalysis.ai/evaluations/scicode deleted per audit rule). DeepMind /models/gemini/pro/ cross-model Performance table: SciCode 52%" }, { "model_id": "gpt-5.2", "benchmark_id": "aa_lcr", "score": 72.3, "reference_url": "https://huggingface.co/moonshotai/Kimi-K2.5", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "1.0" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Kimi K2.5 model card: gpt-5.2=72.3.", "candidates": [ { "score": 72.7, "reference_url": "https://huggingface.co/Qwen/Qwen3.5-397B-A17B", "source_type": "model_card", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "notes": "Qwen3.5-397B-A17B HF model card cross-model table (Qwen self-test of 3rd-party model, mc=false): alternate measurement 72.7 (3rd-party Qwen self-test)." }, { "score": 73.0, "reference_url": "https://huggingface.co/MiniMaxAI/MiniMax-M2.5", "source_type": "model_card", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "notes": "MiniMax M2.5 model card: gpt-5.2=73.0." } ] }, { "model_id": "gpt-5.2", "benchmark_id": "tau_bench_telecom", "score": 98.7, "reference_url": "https://openai.com/index/introducing-gpt-5-2/", "audit_status": "dropped", "audit_notes": "R5g ghost cell: introducing-gpt-5-2/ blog reports τ²-bench Telecom (98.7%), not original τ-bench. τ²-bench data already in tau2_bench_telecom (verified). Drop." }, { "model_id": "gpt-5.2", "benchmark_id": "mmmu_pro", "score": 79.5, "reference_url": "https://openai.com/index/introducing-gpt-5-2/", "candidates": [ { "score": 84.3, "reference_url": "https://www.anthropic.com/news/claude-sonnet-4-6", "source_type": "third_party", "reported_setting": { "note": "see primary" }, "notes": "Reported by Anthropic Sonnet 4.6 blog (third-party)." }, { "score": 79.5, "reference_url": "https://deepmind.google/models/gemini/pro/", "source_type": "official_blog", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "notes": "DeepMind /models/gemini/pro/ cross-model Performance table: MMMU-Pro no tools 79.5%" }, { "score": 79.5, "reference_url": "https://deepmind.google/models/gemini/flash/", "source_type": "official_blog", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "notes": "DeepMind /models/gemini/flash/ Performance table: MMMU-Pro 79.5%" } ], "reported_setting": { "mode": "thinking", "effort": "xhigh", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-5.2 blog: 79.5 (xhigh reasoning effort)." }, { "model_id": "gpt-5.2", "benchmark_id": "simpleqa", "score": 58.0, "reference_url": "https://llm-stats.com/models/gpt-5.2-2025-12-11", "audit_status": "dropped", "rule_id": "R5h-third-party-blog", "notes": "DROPPED (R5h): random third-party blog/aggregator, no primary trail" }, { "model_id": "gpt-5.2", "benchmark_id": "ifeval", "score": 94.8, "reference_url": "https://huggingface.co/Qwen/Qwen3.5-397B-A17B", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Qwen3.5-397B-A17B HF model card cross-model table (Qwen self-test of 3rd-party model, mc=false): ifeval 94.8 (3rd-party Qwen self-test). [R5d: prior unverified value 95.0 from https://llm-stats.com/models/gpt-5.2-2025-12-11 deleted.]" }, { "model_id": "gpt-5.2", "benchmark_id": "livecodebench", "score": 87.7, "reference_url": "https://huggingface.co/Qwen/Qwen3.5-397B-A17B", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Qwen3.5-397B-A17B HF model card cross-model table (Qwen self-test of 3rd-party model, mc=false): livecodebench 87.7 (3rd-party Qwen self-test). [R5d: prior unverified value 80.0 from https://llm-stats.com/models/gpt-5.2-2025-12-11 deleted.]" }, { "model_id": "gpt-5.2", "benchmark_id": "humaneval", "score": 95.0, "reference_url": "https://llm-stats.com/models/gpt-5.2-2025-12-11", "audit_status": "dropped", "rule_id": "R5h-third-party-blog", "notes": "DROPPED (R5h): random third-party blog/aggregator, no primary trail" }, { "model_id": "gpt-5.2", "benchmark_id": "mmlu", "score": 88.0, "reference_url": "https://llm-stats.com/models/gpt-5.2-2025-12-11", "audit_status": "dropped", "rule_id": "R5h-third-party-blog", "notes": "DROPPED (R5h): random third-party blog/aggregator, no primary trail" }, { "model_id": "gpt-5.2", "benchmark_id": "math_500", "score": 99.4, "reference_url": "https://artificialanalysis.ai/evaluations/math-500", "audit_status": "needs_review", "notes": "Reference URL is AA math-500 eval page, but gpt-5-2 (gpt-5.2) is NOT present on that page; AA shows math_500=null for this model. Possible ghost score (R5g-a). Cannot verify 99.4% from AA source." }, { "model_id": "gpt-5.2", "benchmark_id": "mrcr_v2", "score": 83.8, "reference_url": "https://deepmind.google/models/gemini/pro/", "candidates": [ { "score": 81.9, "reference_url": "https://deepmind.google/models/gemini/flash/", "source_type": "official_blog", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default", "context": "128k" }, "notes": "DeepMind /models/gemini/flash/ Performance table: MRCR v2 8-needle 128k 81.9%" } ], "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default", "context": "128k" }, "source_type": "official_blog", "audit_status": "verified", "matches_canonical": true, "notes": "DeepMind /models/gemini/pro/ Performance table (PROMOTED from candidate; prior unverified value 70.0 from https://www.datacamp.com/blog/gpt-5-2 deleted per audit rule). DeepMind /models/gemini/pro/ cross-model Performance table: MRCR v2 8-needle 128k average 83.8%" }, { "model_id": "gpt-5.2", "benchmark_id": "osworld", "score": 47.3, "reference_url": "https://openai.com/index/introducing-gpt-5-4/", "reported_setting": { "mode": "thinking", "effort": "xhigh", "tools": "agentic", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-5.4 blog table: 47.3.", "candidates": [ { "score": 37.9, "reference_url": "https://openai.com/index/introducing-gpt-5-3-codex/", "source_type": "official_blog", "reported_setting": { "mode": "thinking", "effort": "xhigh", "tools": "agentic", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "notes": "OpenAI GPT-5.3-Codex blog: 37.9 (xhigh reasoning effort)." } ] }, { "model_id": "gpt-5.2", "benchmark_id": "hle", "score": 34.5, "reference_url": "https://openai.com/index/introducing-gpt-5-2/", "reported_setting": { "mode": "thinking", "effort": "xhigh", "tools": "none (no-tools column)", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-5.2 blog: 34.5 (xhigh reasoning effort).", "candidates": [ { "score": 34.5, "reference_url": "https://deepmind.google/models/gemini/pro/", "source_type": "official_blog", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "notes": "DeepMind /models/gemini/pro/ cross-model Performance table: HLE no tools 34.5% (Thinking xhigh)" }, { "score": 45.5, "reference_url": "https://deepmind.google/models/gemini/pro/", "source_type": "official_blog", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "web+code", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "notes": "DeepMind /models/gemini/pro/ cross-model Performance table: HLE Search+Code 45.5% (Thinking xhigh)" }, { "score": 34.5, "reference_url": "https://deepmind.google/models/gemini/flash/", "source_type": "official_blog", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "notes": "DeepMind /models/gemini/flash/ Performance table: HLE no tools 34.5% (GPT-5.2 Extra high)" }, { "score": 45.5, "reference_url": "https://deepmind.google/models/gemini/flash/", "source_type": "official_blog", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "web+code", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "notes": "DeepMind /models/gemini/flash/ Performance table: HLE Search+Code 45.5%" }, { "score": 35.4, "reference_url": "https://z.ai/blog/glm-5", "source_type": "official_blog", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "notes": "GLM-5 blog: gpt-5.2=35.4." }, { "score": 35.5, "reference_url": "https://huggingface.co/Qwen/Qwen3.5-397B-A17B", "source_type": "model_card", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "notes": "Qwen3.5-397B-A17B HF model card cross-model table (Qwen self-test of 3rd-party model, mc=false): alternate measurement 35.5 (3rd-party Qwen self-test)." }, { "score": 31.4, "reference_url": "https://huggingface.co/MiniMaxAI/MiniMax-M2.5", "source_type": "model_card", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "notes": "MiniMax M2.5 model card: gpt-5.2=31.4." }, { "score": 29.9, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "source_type": "model_card", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "notes": "Doubao Seed 2.0 model card Table 3: gpt-5.2=29.9." }, { "score": 36.6, "reference_url": "https://llm-stats.com/benchmarks/humanity's-last-exam", "source_type": "third_party_aggregator", "reported_setting": { "effort": "pro" }, "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=GPT-5.2 Pro, slug=gpt-5.2-pro-2025-12-11, provider=OpenAI" } ] }, { "model_id": "gpt-5.2", "benchmark_id": "mmmu", "score": 86.7, "reference_url": "https://huggingface.co/Qwen/Qwen3.5-397B-A17B", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Qwen3.5-397B-A17B HF model card cross-model table (Qwen self-test of 3rd-party model, mc=false): mmmu 86.7 (matches prior unverified value, re-sourced to HF Qwen3.5 card).", "candidates": [ { "score": 83.7, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "source_type": "model_card", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "notes": "Doubao Seed 2.0 model card: gpt-5.2=83.7." } ] }, { "model_id": "gpt-5.2", "benchmark_id": "chatbot_arena_elo", "score": 1438, "reference_url": "https://lmarena.ai/leaderboard/text", "audit_status": "verified", "source_type": "leaderboard", "matches_canonical": true, "notes": "lmarena.ai text arena (fetched 2026-04-29). Arena model name: gpt-5.2 (rank 54). ELO updates continuously; score reflects latest available.", "candidates": [ { "score": 1476, "reference_url": "https://ernie.baidu.com/blog/posts/ernie-5.1-0508-release/creative-capability.png", "source_type": "official_blog", "reported_setting": { "leaderboard": "Arena AI Text", "metric": "Arena score", "style_control": "on", "mode": "chat/non-thinking", "snapshot_date": "2026-04-30" }, "notes": "April 30, 2026 Text Arena snapshot; style control on. Research observation obs-059. Displayed rank 12. Identical asset is published in four page placements. The same PNG bytes are published under two filenames and four English/Chinese page placements." } ] }, { "model_id": "gpt-5.2", "benchmark_id": "aa_intelligence_index", "score": 70.0, "reference_url": "https://x.com/ArtificialAnlys/status/1943166841150644622", "audit_status": "needs_review", "audit_note": "Source is a Twitter/X post (x.com/ArtificialAnlys/status/1943166841150644622); requires browser to verify tweet content. Value 70.0 not verified via automated fetch.", "audited_by": "lychee-audit-bot", "audited_at": "2026-07-26T00:00:00Z" }, { "model_id": "gpt-5.2", "benchmark_id": "arc_agi_1", "score": 86.2, "reference_url": "https://openai.com/index/introducing-gpt-5-2/", "reported_setting": { "mode": "thinking", "effort": "xhigh", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-5.2 blog: 86.2 (xhigh reasoning effort).", "candidates": [ { "score": 89.9, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "source_type": "model_card", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "notes": "Doubao Seed 2.0 model card Table 3: gpt-5.2=89.9." }, { "score": 90.5, "reference_url": "https://llm-stats.com/benchmarks/arc-agi", "source_type": "third_party_aggregator", "reported_setting": { "effort": "pro" }, "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=GPT-5.2 Pro, slug=gpt-5.2-pro-2025-12-11, provider=OpenAI" } ] }, { "model_id": "gpt-5.2", "benchmark_id": "codeforces_rating", "score": 3148, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card Table 3: gpt-5.2=3148." }, { "model_id": "gpt-5.2", "benchmark_id": "mmlu_pro", "score": 86.7, "reference_url": "https://huggingface.co/moonshotai/Kimi-K2.5", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "1.0" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Kimi K2.5 model card: gpt-5.2=86.7.", "candidates": [ { "score": 87.4, "reference_url": "https://huggingface.co/Qwen/Qwen3.5-397B-A17B", "source_type": "model_card", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "notes": "Qwen3.5-397B-A17B HF model card cross-model table (Qwen self-test of 3rd-party model, mc=false): alternate measurement 87.4 (3rd-party Qwen self-test)." }, { "score": 85.9, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "source_type": "model_card", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "notes": "Doubao Seed 2.0 model card Table 3: gpt-5.2=85.9." } ] }, { "model_id": "gpt-5.2", "benchmark_id": "tau_bench_retail", "score": 88.0, "reference_url": "https://openai.com/index/introducing-gpt-5-2/", "audit_status": "dropped", "audit_notes": "R5g ghost cell: introducing-gpt-5-2/ blog reports τ²-bench Retail (82.0%), not original τ-bench. τ²-bench data in tau2_bench_retail (verified). Cell value 88.0 also mismatches blog value 82.0. Drop." }, { "model_id": "gpt-5.2", "benchmark_id": "terminal_bench", "score": 62.2, "reference_url": "https://openai.com/index/introducing-gpt-5-4/", "candidates": [ { "score": 53.5, "reference_url": "https://www.anthropic.com/news/claude-sonnet-4-6", "source_type": "third_party", "reported_setting": { "note": "see primary" }, "notes": "Reported by Anthropic Sonnet 4.6 blog (third-party)." }, { "score": 62.2, "reference_url": "https://openai.com/index/introducing-gpt-5-3-codex/", "source_type": "official_blog", "reported_setting": { "mode": "thinking", "effort": "xhigh", "tools": "agentic", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "notes": "OpenAI GPT-5.3-Codex blog: 62.2 (xhigh reasoning effort)." }, { "score": 54.0, "reference_url": "https://deepmind.google/models/gemini/pro/", "source_type": "official_blog", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "Terminus-2", "prompt_style": "default", "temperature": "default" }, "notes": "DeepMind /models/gemini/pro/ cross-model Performance table: Terminal-Bench 2.0 Terminus-2 54.0%" }, { "score": 62.2, "reference_url": "https://deepmind.google/models/gemini/pro/", "source_type": "official_blog", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "Codex", "prompt_style": "default", "temperature": "default" }, "notes": "DeepMind /models/gemini/pro/ cross-model Performance table: Terminal-Bench 2.0 Codex (best self-reported) 62.2%" }, { "score": 54.0, "reference_url": "https://z.ai/blog/glm-5", "source_type": "official_blog", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "agentic", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "notes": "GLM-5 blog: gpt-5.2=54.0." }, { "score": 54.0, "reference_url": "https://huggingface.co/moonshotai/Kimi-K2.5", "source_type": "model_card", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "agentic", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "1.0" }, "notes": "Kimi K2.5 model card: gpt-5.2=54.0." }, { "score": 54.0, "reference_url": "https://huggingface.co/Qwen/Qwen3.5-397B-A17B", "source_type": "model_card", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "notes": "Qwen3.5-397B-A17B HF model card cross-model table (Qwen self-test of 3rd-party model, mc=false): alternate measurement 54.0 (3rd-party Qwen self-test)." }, { "score": 62.4, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "source_type": "model_card", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "varies", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "notes": "(merged from terminal_bench_2) Doubao Seed 2.0 model card: gpt-5.2=62.4." }, { "score": 60.5, "reference_url": "https://cursor.com/resources/Composer2.pdf", "source_type": "tech_report", "reported_setting": { "effort": "high", "tools": "Cursor coding agent", "sampling": "pass@1", "harness": "Cursor harness" }, "notes": "Composer 2 technical report Table 1 / Terminal-Bench 2.0 / GPT-5.2 / value 1: For third-party public benchmarks, slash-separated values are Cursor/official-harness then self-reported. SWE prompts prepend 'please solve this github issue'; Terminal prompts add solution-format instructions." } ], "reported_setting": { "mode": "thinking", "effort": "xhigh", "tools": "agentic", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-5.4 blog table: 62.2." }, { "model_id": "gpt-5.2", "benchmark_id": "aime_2024", "score": 100, "reference_url": "https://www.vellum.ai/blog/gpt-5-2-benchmarks", "audit_status": "dropped", "rule_id": "R5h-third-party-blog", "notes": "DROPPED (R5h): random third-party blog/aggregator, no primary trail" }, { "model_id": "gpt-5.2", "benchmark_id": "brumo_2025", "score": 98.33, "reference_url": "https://matharena.ai/?comp=brumo--brumo_2025", "audit_status": "verified", "matches_canonical": false, "reported_setting": "reasoning_effort:high", "source_type": "leaderboard" }, { "model_id": "gpt-5.2", "benchmark_id": "cmimc_2025", "score": 91.25, "reference_url": "https://matharena.ai/?comp=cmimc--cmimc_2025", "audit_status": "verified", "matches_canonical": false, "reported_setting": "reasoning_effort:high", "source_type": "leaderboard" }, { "model_id": "gpt-5.2", "benchmark_id": "hmmt_nov_2025", "score": 97.1, "reference_url": "https://z.ai/blog/glm-5", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "GLM-5 blog: gpt-5.2=97.1.", "candidates": [ { "score": 100.0, "reference_url": "https://huggingface.co/Qwen/Qwen3.5-397B-A17B", "source_type": "model_card", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "notes": "Qwen3.5-397B-A17B HF model card cross-model table (Qwen self-test of 3rd-party model, mc=false): alternate measurement 100.0 (3rd-party Qwen self-test)." }, { "score": 100.0, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "source_type": "model_card", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "notes": "Doubao Seed 2.0 model card Table 3: gpt-5.2=100.0." } ] }, { "model_id": "gpt-5.2", "benchmark_id": "matharena_apex_2025", "score": 13.54, "reference_url": "https://matharena.ai/?comp=apex--apex_2025", "audit_status": "verified", "matches_canonical": false, "reported_setting": "reasoning_effort:high", "source_type": "leaderboard", "candidates": [ { "score": 18.2, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "source_type": "model_card", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "notes": "(merged from apex) Doubao Seed 2.0 model card Table 3: gpt-5.2=18.2." } ] }, { "model_id": "gpt-5.2", "benchmark_id": "smt_2025", "score": 91.98, "reference_url": "https://matharena.ai/?comp=smt--smt_2025", "audit_status": "verified", "matches_canonical": false, "reported_setting": "reasoning_effort:high", "source_type": "leaderboard" }, { "model_id": "gpt-5.3-codex", "benchmark_id": "swe_bench_pro", "score": 56.8, "reference_url": "https://openai.com/index/introducing-gpt-5-3-codex/", "reported_setting": { "mode": "thinking", "effort": "xhigh", "tools": "agentic (bash + edit)", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-5.3-Codex blog: 56.8 (xhigh reasoning effort).", "candidates": [ { "score": 56.8, "reference_url": "https://deepmind.google/models/gemini/pro/", "source_type": "official_blog", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "notes": "DeepMind /models/gemini/pro/ cross-model Performance table: SWE-Bench Pro single attempt 56.8%" } ] }, { "model_id": "gpt-5.3-codex", "benchmark_id": "osworld", "score": 64.7, "reference_url": "https://openai.com/index/introducing-gpt-5-3-codex/", "reported_setting": { "mode": "thinking", "effort": "xhigh", "tools": "agentic", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-5.3-Codex blog: 64.7 (xhigh reasoning effort).", "candidates": [ { "score": 74.0, "reference_url": "https://openai.com/index/introducing-gpt-5-4/", "source_type": "official_blog", "reported_setting": { "mode": "thinking", "effort": "xhigh", "tools": "agentic", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "notes": "OpenAI GPT-5.4 blog: 74.0. (was primary, demoted when 5.3-Codex own blog promoted to primary)" } ] }, { "model_id": "gpt-5.3-codex", "benchmark_id": "swe_bench_verified", "score": 56.8, "reference_url": "https://openai.com/index/introducing-gpt-5-3-codex/", "audit_status": "dropped", "audit_notes": "R5g ghost cell: introducing-gpt-5-3-codex/ blog has no SWE-bench Verified data (only SWE-Bench Pro=56.8, which happens to match this value by coincidence). Drop." }, { "model_id": "gpt-5.3-codex", "benchmark_id": "humaneval", "score": 93.0, "reference_url": "https://openai.com/index/introducing-gpt-5-3-codex/", "audit_status": "dropped", "audit_notes": "R5g ghost cell: introducing-gpt-5-3-codex/ blog contains no HumanEval data. Drop." }, { "model_id": "gpt-5.3-codex", "benchmark_id": "aime_2025", "score": 94.0, "reference_url": "https://automatio.ai/models/gpt-5-3-codex", "audit_status": "dropped", "rule_id": "R5h-third-party-blog", "notes": "DROPPED (R5h): random third-party blog/aggregator, no primary trail" }, { "model_id": "gpt-5.3-codex", "benchmark_id": "gpqa_diamond", "score": 92.6, "reference_url": "https://openai.com/index/introducing-gpt-5-4/", "reported_setting": { "mode": "thinking", "effort": "xhigh", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-5.4 blog table: 92.6." }, { "model_id": "gpt-5.3-codex", "benchmark_id": "gsm8k", "score": 99.0, "reference_url": "https://automatio.ai/models/gpt-5-3-codex", "audit_status": "dropped", "rule_id": "R5h-third-party-blog", "notes": "DROPPED (R5h): random third-party blog/aggregator, no primary trail" }, { "model_id": "gpt-5.3-codex", "benchmark_id": "livecodebench", "score": 85.0, "reference_url": "https://openai.com/index/introducing-gpt-5-3-codex/", "audit_status": "dropped", "audit_notes": "R5g ghost cell: introducing-gpt-5-3-codex/ blog contains no LiveCodeBench data. Drop." }, { "model_id": "gpt-5.3-codex", "benchmark_id": "math_500", "score": 96.0, "reference_url": "https://automatio.ai/models/gpt-5-3-codex", "audit_status": "dropped", "rule_id": "R5h-third-party-blog", "notes": "DROPPED (R5h): random third-party blog/aggregator, no primary trail" }, { "model_id": "gpt-5.3-codex", "benchmark_id": "mmlu", "score": 94.0, "reference_url": "https://automatio.ai/models/gpt-5-3-codex", "audit_status": "dropped", "rule_id": "R5h-third-party-blog", "notes": "DROPPED (R5h): random third-party blog/aggregator, no primary trail" }, { "model_id": "gpt-5.3-codex", "benchmark_id": "mmmu", "score": 84.0, "reference_url": "https://automatio.ai/models/gpt-5-3-codex", "audit_status": "dropped", "rule_id": "R5h-third-party-blog", "notes": "DROPPED (R5h): random third-party blog/aggregator, no primary trail" }, { "model_id": "gpt-5.3-codex", "benchmark_id": "terminal_bench", "score": 77.3, "reference_url": "https://openai.com/index/introducing-gpt-5-3-codex/", "reported_setting": { "mode": "thinking", "effort": "xhigh", "tools": "agentic", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-5.3-Codex blog: 77.3 (xhigh reasoning effort).", "candidates": [ { "score": 64.7, "reference_url": "https://deepmind.google/models/gemini/pro/", "source_type": "official_blog", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "Terminus-2", "prompt_style": "default", "temperature": "default" }, "notes": "DeepMind /models/gemini/pro/ cross-model Performance table: Terminal-Bench 2.0 Terminus-2 64.7%" }, { "score": 77.3, "reference_url": "https://deepmind.google/models/gemini/pro/", "source_type": "official_blog", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "Codex", "prompt_style": "default", "temperature": "default" }, "notes": "DeepMind /models/gemini/pro/ cross-model Performance table: Terminal-Bench 2.0 Codex (best self-reported) 77.3%" }, { "score": 64.8, "reference_url": "https://cursor.com/resources/Composer2.pdf", "source_type": "tech_report", "reported_setting": { "effort": "high", "tools": "Cursor coding agent", "sampling": "pass@1", "harness": "Cursor harness; 3 safety-filter refusals scored 0" }, "notes": "Composer 2 technical report Table 1 / Terminal-Bench 2.0 / GPT-5.3 Codex / value 1: For third-party public benchmarks, slash-separated values are Cursor/official-harness then self-reported. SWE prompts prepend 'please solve this github issue'; Terminal prompts add solution-format instructions." } ] }, { "model_id": "gpt-5.3-codex", "benchmark_id": "ifeval", "score": 92, "reference_url": "https://automatio.ai/models/gpt-5-3-codex", "audit_status": "dropped", "rule_id": "R5h-third-party-blog", "notes": "DROPPED (R5h): random third-party blog/aggregator, no primary trail" }, { "model_id": "gpt-5.4", "benchmark_id": "swe_bench_pro", "score": 57.7, "reference_url": "https://openai.com/index/introducing-gpt-5-5/", "reported_setting": { "mode": "thinking", "effort": "xhigh", "tools": "agentic (bash + edit)", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-5.5 blog table: 57.7.", "candidates": [ { "score": 57.7, "reference_url": "https://microsoft.ai/wp-content/uploads/2026/06/main_20260602_2.pdf", "source_type": "official_paper", "reported_setting": { "mode": "thinking", "effort": "xhigh", "tools": "none (per-bench override)", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default", "context": "default", "notes": "Per OpenAI gpt-5.4 blog: reasoning effort=xhigh, research environment (may differ from production ChatGPT)." }, "notes": "Displayed exactly as 57.7. Official Microsoft-reported result. Comparator value is quoted from that provider's official model card/release; use the campaign canonical setting." } ] }, { "model_id": "gpt-5.4", "benchmark_id": "terminal_bench", "score": 75.1, "reference_url": "https://openai.com/index/introducing-gpt-5-5/", "reported_setting": { "mode": "thinking", "effort": "xhigh", "tools": "agentic", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-5.5 blog table: 75.1.", "candidates": [ { "score": 65.4, "reference_url": "https://huggingface.co/moonshotai/Kimi-K2.6", "source_type": "model_card", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "agentic", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "1.0" }, "notes": "Kimi K2.6 model card: gpt-5.4=65.4." }, { "score": 66.5, "reference_url": "https://cursor.com/resources/Composer2.pdf", "source_type": "tech_report", "reported_setting": { "effort": "high", "tools": "Cursor coding agent", "sampling": "pass@1", "harness": "Cursor harness; 5 safety-filter refusals scored 0" }, "notes": "Composer 2 technical report Table 1 / Terminal-Bench 2.0 / GPT-5.4 / value 1: For third-party public benchmarks, slash-separated values are Cursor/official-harness then self-reported. SWE prompts prepend 'please solve this github issue'; Terminal prompts add solution-format instructions." }, { "score": 75.1, "reference_url": "https://microsoft.ai/wp-content/uploads/2026/06/main_20260602_2.pdf", "source_type": "official_paper", "reported_setting": { "mode": "thinking", "effort": "xhigh", "tools": "none (per-bench override)", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default", "context": "default", "notes": "Per OpenAI gpt-5.4 blog: reasoning effort=xhigh, research environment (may differ from production ChatGPT)." }, "notes": "Displayed exactly as 75.1. Official Microsoft-reported result. Comparator value is quoted from that provider's official model card/release; use the campaign canonical setting." } ] }, { "model_id": "gpt-5.4", "benchmark_id": "mmmu_pro", "score": 81.2, "reference_url": "https://openai.com/index/introducing-gpt-5-5/", "reported_setting": { "mode": "thinking", "effort": "xhigh", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-5.5 blog table: 81.2." }, { "model_id": "gpt-5.4", "benchmark_id": "browsecomp", "score": 82.7, "reference_url": "https://openai.com/index/introducing-gpt-5-5/", "reported_setting": { "mode": "thinking", "effort": "xhigh", "tools": "agentic", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-5.5 blog table: 82.7." }, { "model_id": "gpt-5.4", "benchmark_id": "frontiermath", "score": 47.6, "reference_url": "https://openai.com/index/introducing-gpt-5-5/", "reported_setting": { "mode": "thinking", "effort": "xhigh", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-5.5 blog table: 47.6." }, { "model_id": "gpt-5.4", "benchmark_id": "gpqa_diamond", "score": 92.8, "reference_url": "https://openai.com/index/introducing-gpt-5-5/", "reported_setting": { "mode": "thinking", "effort": "xhigh", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-5.5 blog table: 92.8.", "candidates": [ { "score": 93.0, "reference_url": "https://huggingface.co/deepseek-ai/DeepSeek-V4-Pro", "source_type": "third_party", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "notes": "DeepSeek V4-Pro model card (third-party): gpt-5.4 (per their own canonical config) = 93.0." }, { "score": 92.0, "reference_url": "https://z.ai/blog/glm-5.1", "source_type": "official_blog", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "notes": "GLM-5.1 blog: gpt-5.4=92.0." }, { "score": 92.8, "reference_url": "https://microsoft.ai/wp-content/uploads/2026/06/main_20260602_2.pdf", "source_type": "official_paper", "reported_setting": { "mode": "thinking", "effort": "xhigh", "tools": "none (per-bench override)", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default", "context": "default", "notes": "Per OpenAI gpt-5.4 blog: reasoning effort=xhigh, research environment (may differ from production ChatGPT)." }, "notes": "Displayed exactly as 92.8. Official Microsoft-reported result. Comparator value is quoted from that provider's official model card/release; use the campaign canonical setting." } ] }, { "model_id": "gpt-5.4", "benchmark_id": "hle", "score": 39.8, "reference_url": "https://openai.com/index/introducing-gpt-5-5/", "reported_setting": { "mode": "thinking", "effort": "xhigh", "tools": "none (no-tools column)", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-5.5 blog table: 39.8." }, { "model_id": "gpt-5.4", "benchmark_id": "arc_agi_1", "score": 93.7, "reference_url": "https://openai.com/index/introducing-gpt-5-5/", "reported_setting": { "mode": "thinking", "effort": "xhigh", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-5.5 blog table: 93.7." }, { "model_id": "gpt-5.4", "benchmark_id": "arc_agi_2", "score": 73.3, "reference_url": "https://openai.com/index/introducing-gpt-5-5/", "reported_setting": { "mode": "thinking", "effort": "xhigh", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-5.5 blog table: 73.3." }, { "model_id": "gpt-5.5", "benchmark_id": "swe_bench_pro", "score": 58.6, "reference_url": "https://openai.com/index/introducing-gpt-5-5/", "reported_setting": { "mode": "thinking", "effort": "xhigh", "tools": "agentic (bash + edit)", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-5.5 blog table: 58.6.", "candidates": [ { "score": 59.4, "reference_url": "https://openai.com/index/gpt-5-6/", "source_type": "official_blog", "reported_setting": { "mode": "thinking", "effort": "source does not state", "tools": "none", "sampling": "source does not state", "judge": "benchmark-specified", "harness": "official release evaluation", "prompt_style": "source does not state", "temperature": "source does not state", "context": "source does not state" }, "notes": "OpenAI GPT-5.6 release table (table 1): GPT‑5.5; SWE-Bench Pro=59.4." }, { "score": 58.6, "reference_url": "https://x.ai/news/grok-4-5", "source_type": "official_blog", "reported_setting": { "mode": "thinking", "effort": "xhigh", "tools": "agentic repository shell/editor", "sampling": "pass@1", "judge": "verification reward and repository tests", "harness": "fixed SWE-bench Pro agent scaffold", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown" }, "notes": "release45 SWE-Bench Pro Resolve rate (%); source effort=xhigh; benchmark protocol matches canonical=True. Unknown source fields remain unknown." }, { "score": 58.6, "reference_url": "https://seed.bytedance.com/en/blog/seed2-1-officially-released-advancing-ai-productivity", "source_type": "official_blog", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "Trae or benchmark-specific coding-agent environment", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official", "prompt_style": "official default", "temperature": "unknown" }, "notes": "Release blog EN: Coding chart (https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/user-upload/4xfa4mqq4wlbt.png)." } ] }, { "model_id": "gpt-5.5", "benchmark_id": "terminal_bench", "score": 82.7, "reference_url": "https://openai.com/index/introducing-gpt-5-5/", "reported_setting": { "mode": "thinking", "effort": "xhigh", "tools": "agentic", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-5.5 blog table: 82.7." }, { "model_id": "gpt-5.5", "benchmark_id": "mmmu_pro", "score": 81.2, "reference_url": "https://openai.com/index/introducing-gpt-5-5/", "reported_setting": { "mode": "thinking", "effort": "xhigh", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-5.5 blog table: 81.2.", "candidates": [ { "score": 83.2, "reference_url": "https://github.com/MoonshotAI/Kimi-K3/blob/main/k3_tech_report.pdf", "source_type": "tech_report", "reported_setting": { "mode": "thinking", "effort": "xhigh", "tools": "python", "sampling": "avg-of-3", "judge": "benchmark-specified", "harness": "Python tool; avg 3 runs", "prompt_style": "default", "temperature": "default", "context": "default" }, "notes": "Kimi K3 technical report: MMMU-Pro (with Python) = 83.2. Source setting: Python tool; avg 3 runs. Origin: Moonshot evaluation." }, { "score": 81.68, "reference_url": "https://huggingface.co/internlm/Intern-S2-Preview-397B", "source_type": "model_card", "reported_setting": { "mode": "source does not state", "effort": "source does not state", "tools": "none", "sampling": "source does not state", "judge": "benchmark-specific; exact version not stated", "harness": "VLMEvalKit or AgentCompass; exact adapter not stated", "prompt_style": "source does not state", "temperature": "source does not state", "context": "source does not state", "notes": "Shanghai AI Lab evaluated all model columns using OpenCompass, VLMEvalKit, and AgentCompass. Per-model mode/effort/sampling are not reported; only Intern-S2 maximum evaluation lengths are disclosed." }, "notes": "Intern-S2-Preview-397B official model card: MMMU Pro = 81.68. This is a Shanghai AI Lab cross-model evaluation; the source does not disclose this model's exact mode, effort, or sampling." }, { "score": 81.2, "reference_url": "https://seed.bytedance.com/en/blog/seed2-1-officially-released-advancing-ai-productivity", "source_type": "official_blog", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "none", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official", "prompt_style": "official default", "temperature": "unknown" }, "notes": "Release blog EN: Vision capability chart (https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/user-upload/4xfa4mqq4yer3.png)." } ] }, { "model_id": "gpt-5.5", "benchmark_id": "browsecomp", "score": 84.4, "reference_url": "https://openai.com/index/introducing-gpt-5-5/", "reported_setting": { "mode": "thinking", "effort": "xhigh", "tools": "agentic", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-5.5 blog table: 84.4.", "candidates": [ { "score": 90.1, "reference_url": "https://llm-stats.com/benchmarks/browsecomp", "source_type": "third_party_aggregator", "reported_setting": { "effort": "pro" }, "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=GPT-5.5 Pro, slug=gpt-5.5-pro, provider=OpenAI" }, { "score": 34.68, "reference_url": "https://openai.com/index/gpt-5-6/", "source_type": "official_blog", "reported_setting": { "effort": "none", "eval_variant": null, "code_mode": "false" }, "notes": "OpenAI GPT-5.6 release figure (BrowseComp): GPT-5.5; BrowseComp=34.68. Figure 15: Agentic browsing tasks.. Cost, latency, and output-token coordinates are retained as review metadata, not score-matrix metrics." }, { "score": 57.82, "reference_url": "https://openai.com/index/gpt-5-6/", "source_type": "official_blog", "reported_setting": { "effort": "low", "eval_variant": null, "code_mode": "false" }, "notes": "OpenAI GPT-5.6 release figure (BrowseComp): GPT-5.5; BrowseComp=57.82. Figure 15: Agentic browsing tasks.. Cost, latency, and output-token coordinates are retained as review metadata, not score-matrix metrics." }, { "score": 78.12, "reference_url": "https://openai.com/index/gpt-5-6/", "source_type": "official_blog", "reported_setting": { "effort": "medium", "eval_variant": null, "code_mode": "false" }, "notes": "OpenAI GPT-5.6 release figure (BrowseComp): GPT-5.5; BrowseComp=78.12. Figure 15: Agentic browsing tasks.. Cost, latency, and output-token coordinates are retained as review metadata, not score-matrix metrics." }, { "score": 83.89, "reference_url": "https://openai.com/index/gpt-5-6/", "source_type": "official_blog", "reported_setting": { "effort": "high", "eval_variant": null, "code_mode": "false" }, "notes": "OpenAI GPT-5.6 release figure (BrowseComp): GPT-5.5; BrowseComp=83.89. Figure 15: Agentic browsing tasks.. Cost, latency, and output-token coordinates are retained as review metadata, not score-matrix metrics." }, { "score": 84.4, "reference_url": "https://seed.bytedance.com/en/blog/seed2-1-officially-released-advancing-ai-productivity", "source_type": "official_blog", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "search", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official", "prompt_style": "official default", "temperature": "unknown" }, "notes": "Release blog EN: Language/search chart (https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/user-upload/4xfa4mqq510uc.png)." } ] }, { "model_id": "gpt-5.5", "benchmark_id": "frontiermath", "score": 51.7, "reference_url": "https://openai.com/index/introducing-gpt-5-5/", "reported_setting": { "mode": "thinking", "effort": "xhigh", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-5.5 blog table: 51.7.", "candidates": [ { "score": 39.6, "reference_url": "https://llm-stats.com/benchmarks/frontiermath", "source_type": "third_party_aggregator", "reported_setting": { "effort": "pro" }, "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=GPT-5.5 Pro, slug=gpt-5.5-pro, provider=OpenAI" } ] }, { "model_id": "gpt-5.5", "benchmark_id": "gpqa_diamond", "score": 93.6, "reference_url": "https://openai.com/index/introducing-gpt-5-5/", "reported_setting": { "mode": "thinking", "effort": "xhigh", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-5.5 blog table: 93.6.", "candidates": [ { "score": 93.5, "reference_url": "https://github.com/MoonshotAI/Kimi-K3/blob/main/k3_tech_report.pdf", "source_type": "tech_report", "reported_setting": { "mode": "thinking", "effort": "xhigh", "tools": "none", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "max/xhigh; no tools", "prompt_style": "default", "temperature": "default", "context": "default" }, "notes": "Kimi K3 technical report: GPQA Diamond = 93.5. Source setting: max/xhigh; no tools. Origin: Moonshot evaluation." } ] }, { "model_id": "gpt-5.5", "benchmark_id": "hle", "score": 41.4, "reference_url": "https://openai.com/index/introducing-gpt-5-5/", "reported_setting": { "mode": "thinking", "effort": "xhigh", "tools": "none (no-tools column)", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-5.5 blog table: 41.4.", "candidates": [ { "score": 57.2, "reference_url": "https://llm-stats.com/benchmarks/humanity's-last-exam", "source_type": "third_party_aggregator", "reported_setting": { "effort": "pro" }, "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=GPT-5.5 Pro, slug=gpt-5.5-pro, provider=OpenAI" }, { "score": 43.1, "reference_url": "https://www.anthropic.com/claude-sonnet-5-system-card", "source_type": "model_card", "reported_setting": { "effort": "source does not state", "tools": "none", "sampling": "pass@1", "harness": "hle", "deployment": "gpt-5.5-pro" }, "notes": "Anthropic Claude Sonnet 5 System Card Figure 8.9.1.A, page 122: GPT-5.5 Pro; Humanity’s Last Exam [without tools]=43.1%. Source setting: effort=source does not state; tools=none; sampling=pass@1; harness=hle; deployment=gpt-5.5-pro." }, { "score": 44.8, "reference_url": "https://ai.meta.com/static-resource/muse-spark-1.1-evaluation-report", "source_type": "tech_report", "reported_setting": { "mode": "thinking", "effort": "xhigh", "tools": "none", "sampling": "pass@1 unless source row states otherwise", "harness": "Meta MSL general-capability evaluation", "temperature": "1.0 for Muse; provider/source setting for comparisons", "context": "benchmark-specific", "configuration": "default" }, "notes": "Figure 44 · HLE no tools; metric=headline_metric. Exact printed value in the general-capability summary." } ] }, { "model_id": "gpt-5.5", "benchmark_id": "arc_agi_1", "score": 95.0, "reference_url": "https://openai.com/index/introducing-gpt-5-5/", "reported_setting": { "mode": "thinking", "effort": "xhigh", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-5.5 blog table: 95.0." }, { "model_id": "gpt-5.5", "benchmark_id": "arc_agi_2", "score": 85.0, "reference_url": "https://openai.com/index/introducing-gpt-5-5/", "reported_setting": { "mode": "thinking", "effort": "xhigh", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-5.5 blog table: 85.0.", "candidates": [ { "score": 84.6, "reference_url": "https://deepmind.google/models/model-cards/gemini-3-5-flash/", "source_type": "official_model_card", "reported_setting": { "mode": "thinking/reasoning", "effort": "xhigh/best available", "tools": "none", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "ARC Prize verified leaderboard", "temperature": "default", "snapshot": "May 2026" }, "notes": "Google Gemini 3.5 Flash May 2026 table: ARC-AGI-2; ARC Prize Verified; semi-private. Settings and provenance are preserved per cell." }, { "score": 85.0, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2.1/Seed2_1_Model_Card.pdf", "source_type": "tech_report", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "none", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official", "prompt_style": "official default", "temperature": "unknown" }, "notes": "Model card Table 11 (p53)." } ] }, { "model_id": "claude-3.7-sonnet", "benchmark_id": "gpqa_diamond", "score": 78.2, "reference_url": "https://www.anthropic.com/news/claude-4", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1 (single-pass)", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "top_p=0.95" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "Anthropic Claude 4 blog cross-model table: Sonnet 3.7 = 78.2% (with extended thinking).", "candidates": [ { "score": 76.8, "reference_url": "https://arxiv.org/abs/2504.21318", "source_type": "tech_report", "reported_setting": { "mode": "unknown", "effort": "unknown", "tools": "unknown", "sampling": "unknown", "judge": "rule-based", "harness": "Microsoft Phi internal eval", "prompt_style": "unknown", "temperature": "unknown" }, "notes": "Phi-4-reasoning paper Table 1 (claude-3.7-sonnet column): gpqa_diamond=76.8 (alt measurement, mc=false)" } ] }, { "model_id": "claude-3.7-sonnet", "benchmark_id": "aime_2024", "score": 23.3, "reference_url": "https://www.anthropic.com/news/claude-3-7-sonnet", "reported_setting": { "mode": "non-thinking", "effort": null, "tools": "none", "sampling": "pass@1", "judge": "official harness", "harness": "official", "prompt_style": "default", "temperature": 0.0 }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "Anthropic blog, no extended thinking. 80.0% (extended w/ parallel test-time compute) excluded.", "candidates": [ { "score": 55.3, "reference_url": "https://arxiv.org/abs/2504.21318", "source_type": "tech_report", "reported_setting": { "mode": "unknown", "effort": "unknown", "tools": "unknown", "sampling": "unknown", "judge": "rule-based", "harness": "Microsoft Phi internal eval", "prompt_style": "unknown", "temperature": "unknown" }, "notes": "Phi-4-reasoning paper Table 1 (claude-3.7-sonnet column): aime_2024=55.3 (alt measurement, mc=false)" }, { "score": 61.3, "reference_url": "https://www.anthropic.com/news/claude-3-7-sonnet", "source_type": "official_blog", "reported_setting": { "mode": "thinking", "effort": "64K token budget", "tools": "none", "sampling": "pass@1 (avg 16 trials)", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "notes": "Anthropic 3.7 Sonnet blog extended thinking (64K) column: AIME-2024=61.3% (pass@1); 80.0% parallel TTC" } ] }, { "model_id": "claude-3.7-sonnet", "benchmark_id": "math_500", "score": 82.2, "reference_url": "https://www.anthropic.com/news/claude-3-7-sonnet", "reported_setting": { "mode": "non-thinking", "effort": null, "tools": "none", "sampling": "pass@1", "judge": "official harness", "harness": "official", "prompt_style": "default", "temperature": 0.0 }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "Anthropic blog, no extended thinking. 96.2% (extended) excluded.", "candidates": [ { "score": 96.2, "reference_url": "https://www.anthropic.com/news/claude-3-7-sonnet", "source_type": "official_blog", "reported_setting": { "mode": "thinking", "effort": "64K token budget", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "notes": "Anthropic 3.7 Sonnet blog extended thinking (64K) column: MATH-500=96.2%" } ] }, { "model_id": "claude-3.7-sonnet", "benchmark_id": "swe_bench_verified", "score": 62.3, "reference_url": "https://www.anthropic.com/news/claude-4", "reported_setting": { "mode": "non-thinking", "effort": "default", "tools": "agentic (bash + edit)", "sampling": "pass@1 (avg 10 trials)", "judge": "rule-based", "harness": "simple bash+editor scaffold", "prompt_style": "default", "temperature": "top_p=0.95" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "Anthropic Claude 4 blog cross-model table: Sonnet 3.7 = 62.3% / 70.3% w/ parallel." }, { "model_id": "claude-3.7-sonnet", "benchmark_id": "ifeval", "score": 90.8, "reference_url": "https://www.anthropic.com/news/claude-3-7-sonnet", "reported_setting": { "mode": "non-thinking", "effort": null, "tools": "none", "sampling": "pass@1", "judge": "official harness", "harness": "official", "prompt_style": "default", "temperature": 0.0 }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "Anthropic blog, no extended thinking. 93.2% (extended) excluded.", "candidates": [ { "score": 93.2, "reference_url": "https://www.anthropic.com/news/claude-3-7-sonnet", "source_type": "official_blog", "reported_setting": { "mode": "thinking", "effort": "64K token budget", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "notes": "Anthropic 3.7 Sonnet blog extended thinking (64K) column: IFEval=93.2%" } ] }, { "model_id": "claude-3.7-sonnet", "benchmark_id": "humaneval", "score": 94.0, "reference_url": "https://www.datacamp.com/blog/claude-3-7-sonnet", "audit_status": "dropped", "notes": " [Dropped: DataCamp is a tutorial/learning platform aggregator, not primary source. Rule R5h-tutorial-blog.]" }, { "model_id": "claude-3.7-sonnet", "benchmark_id": "livecodebench", "score": 52.6, "reference_url": "https://arxiv.org/abs/2504.21318", "reported_setting": { "mode": "unknown", "effort": "unknown", "tools": "unknown", "sampling": "unknown", "judge": "rule-based", "harness": "Microsoft Phi internal eval", "prompt_style": "unknown", "temperature": "unknown" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Phi-4-reasoning paper Table 1 (claude-3.7-sonnet column): livecodebench=52.6. [R5d: prior unverified value 65.0 from https://www.datacamp.com/blog/claude-3-7-sonnet deleted.]" }, { "model_id": "claude-3.7-sonnet", "benchmark_id": "tau_bench_retail", "score": 81.2, "reference_url": "https://www.anthropic.com/news/claude-4", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "agentic (TAU-bench official tools)", "sampling": "pass@1 (single-pass)", "judge": "rule-based", "harness": "τ-bench official + max steps 100", "prompt_style": "prompt addendum to Agent Policy (Anthropic)", "temperature": "top_p=0.95" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Anthropic Claude 4 blog cross-model table." }, { "model_id": "claude-3.7-sonnet", "benchmark_id": "tau_bench_airline", "score": 58.4, "reference_url": "https://www.anthropic.com/news/claude-4", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "agentic (TAU-bench official tools)", "sampling": "pass@1 (single-pass)", "judge": "rule-based", "harness": "τ-bench official + max steps 100", "prompt_style": "prompt addendum to Agent Policy (Anthropic)", "temperature": "top_p=0.95" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Anthropic Claude 4 blog cross-model table." }, { "model_id": "claude-3.7-sonnet", "benchmark_id": "tau_bench_telecom", "score": 49.0, "reference_url": "https://artificialanalysis.ai/models/claude-3-7-sonnet", "audit_status": "needs_review", "notes": "Reference URL is AA model page (claude-3-7-sonnet). AA only publishes aggregate tau2 score (no domain breakdown); AA shows claude-3-7-sonnet tau2=0.5 aggregate but no telecom-domain score. Cannot verify 49.0 from AA source." }, { "model_id": "claude-3.7-sonnet", "benchmark_id": "aime_2025", "score": 54.8, "reference_url": "https://www.anthropic.com/news/claude-4", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1 (single-pass)", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "top_p=0.95" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "Anthropic Claude 4 blog cross-model table.", "candidates": [ { "score": 53.0, "reference_url": "https://arxiv.org/abs/2504.21318", "source_type": "tech_report", "reported_setting": { "mode": "unknown", "effort": "unknown", "tools": "unknown", "sampling": "unknown", "judge": "rule-based", "harness": "Microsoft Phi internal eval", "prompt_style": "unknown", "temperature": "unknown" }, "notes": "Phi-4-reasoning paper Table 1 (claude-3.7-sonnet column): aime_2025=53.0 (alt measurement, mc=false)" } ] }, { "model_id": "claude-3.7-sonnet", "benchmark_id": "frontiermath", "score": 3.1, "reference_url": "https://epoch.ai/benchmarks/frontiermath", "audit_status": "verified", "source_type": "leaderboard", "matches_canonical": true, "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none" }, "notes": "epoch.ai FrontierMath-2025-02-28-Private: claude-3-7-sonnet-20250219 (no thinking budget) = 3.1%. Canonical notes state no extended thinking for FrontierMath. BP had 5.0 (unverified, sourced from unknown)." }, { "model_id": "claude-3.7-sonnet", "benchmark_id": "arc_agi_2", "score": 0.7, "reference_url": "https://arcprize.org/leaderboard", "audit_status": "verified", "source_type": "leaderboard", "matches_canonical": true, "reported_setting": { "mode": "thinking", "effort": "16K", "tools": "none", "sampling": "pass@1 (single-pass)", "judge": "rule-based", "harness": "official (arcprize.org)", "system_type": "CoT" }, "notes": "arcprize.org leaderboard audit: Claude 3.7 (16K) on leaderboard" }, { "model_id": "claude-3.7-sonnet", "benchmark_id": "mmmu", "score": 75.0, "reference_url": "https://www.anthropic.com/news/claude-4", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1 (single-pass)", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "top_p=0.95" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "Anthropic Claude 4 blog cross-model table (MMMU validation)." }, { "model_id": "claude-3.7-sonnet", "benchmark_id": "osworld", "score": 28.0, "reference_url": "https://llm-stats.com/benchmarks/osworld", "audit_status": "dropped", "rule_id": "R5h-third-party-blog", "notes": "DROPPED (R5h): random third-party blog/aggregator, no primary trail" }, { "model_id": "claude-3.7-sonnet", "benchmark_id": "chatbot_arena_elo", "score": 1372, "reference_url": "https://lmarena.ai/leaderboard/text", "audit_status": "verified", "source_type": "leaderboard", "matches_canonical": true, "notes": "lmarena.ai text arena (fetched 2026-04-29). Arena model name: claude-3-7-sonnet-20250219 (rank 138). ELO updates continuously; score reflects latest available." }, { "model_id": "claude-3.7-sonnet", "benchmark_id": "arena_hard", "score": 63.9, "reference_url": "https://github.com/lmarena/arena-hard-auto", "audit_status": "verified", "reported_setting": "Arena-Hard-v2.0 Creative Writing, Ensemble GPT-4.1+Gemini-2.5-Pro judge", "source_type": "leaderboard", "matches_canonical": true }, { "model_id": "claude-3.7-sonnet", "benchmark_id": "mmlu", "score": 89.0, "reference_url": "https://www.anthropic.com/news/claude-3-7-sonnet", "audit_status": "dropped", "notes": "DROPPED: Blog shows MMMLU (not MMLU), no MMLU row in source; 89.0 not found anywhere in blog table" }, { "model_id": "claude-3.7-sonnet", "benchmark_id": "simpleqa", "score": 26.0, "reference_url": "https://artificialanalysis.ai/models/claude-3-7-sonnet", "audit_status": "needs_review", "notes": "Reference URL is AA model page (claude-3-7-sonnet). AA does not publish SimpleQA scores; field is not present in AA model data for any model. Cannot verify from AA source. Original source needs identification." }, { "model_id": "claude-3.7-sonnet", "benchmark_id": "arc_agi_1", "score": 28.6, "reference_url": "https://arcprize.org/leaderboard", "audit_status": "verified", "source_type": "leaderboard", "matches_canonical": true, "reported_setting": { "mode": "thinking", "effort": "16K", "tools": "none", "sampling": "pass@1 (single-pass)", "judge": "rule-based", "harness": "official (arcprize.org)", "system_type": "CoT" }, "notes": "arcprize.org leaderboard audit: Claude 3.7 (16K) on leaderboard" }, { "model_id": "claude-3.7-sonnet", "benchmark_id": "brumo_2025", "score": 65.83, "reference_url": "https://matharena.ai/?comp=brumo--brumo_2025", "audit_status": "verified", "matches_canonical": true, "reported_setting": "extended_thinking", "source_type": "leaderboard" }, { "model_id": "claude-3.7-sonnet", "benchmark_id": "codeforces_rating", "score": 1640, "reference_url": "https://automatio.ai/models/claude-3-7-sonnet", "audit_status": "dropped", "rule_id": "R5h-third-party-blog", "notes": "DROPPED (R5h): random third-party blog/aggregator, no primary trail" }, { "model_id": "claude-3.7-sonnet", "benchmark_id": "mmlu_pro", "score": 78, "reference_url": "https://automatio.ai/models/claude-3-7-sonnet", "audit_status": "dropped", "rule_id": "R5h-third-party-blog", "notes": "DROPPED (R5h): random third-party blog/aggregator, no primary trail" }, { "model_id": "claude-3.7-sonnet", "benchmark_id": "mmmlu", "score": 85.9, "reference_url": "https://www.anthropic.com/news/claude-4", "reported_setting": { "mode": "non-thinking", "effort": "default", "tools": "none", "sampling": "pass@1 (single-pass)", "judge": "rule-based", "harness": "official", "prompt_style": "avg 14 non-English languages", "temperature": "top_p=0.95" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "Anthropic Claude 4 blog cross-model table.", "candidates": [ { "score": 86.1, "reference_url": "https://www.anthropic.com/news/claude-3-7-sonnet", "source_type": "official_blog", "reported_setting": { "mode": "thinking", "effort": "64K token budget", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "notes": "Anthropic 3.7 Sonnet blog extended thinking (64K) col: MMMLU=86.1%; no-thinking=83.2%" } ] }, { "model_id": "claude-3.7-sonnet", "benchmark_id": "smt_2025", "score": 56.6, "reference_url": "https://matharena.ai/?comp=smt--smt_2025", "audit_status": "verified", "matches_canonical": true, "reported_setting": "extended_thinking", "source_type": "leaderboard" }, { "model_id": "claude-3.7-sonnet", "benchmark_id": "swe_bench_pro", "score": 25.6, "reference_url": "https://www.anthropic.com/news/claude-3-7-sonnet", "audit_status": "dropped", "notes": "DROPPED: SWE-bench Pro not in blog table; only SWE-bench Verified is shown. Value 25.6 not in source." }, { "model_id": "claude-3.7-sonnet", "benchmark_id": "terminal_bench_1", "score": 35.2, "reference_url": "https://www.tbench.ai/leaderboard/terminal-bench/1.0", "audit_status": "verified", "source_type": "leaderboard", "rule_ids": [ "R5d" ], "notes": "Confirmed 35.2% on tbench.ai Terminal-Bench 1.0 leaderboard (Claude Code (Anthropic) scaffold, 2025-05-16)." }, { "model_id": "claude-3.7-sonnet", "benchmark_id": "usamo_2025", "score": 3.65, "reference_url": "https://matharena.ai/?comp=usamo--usamo_2025", "audit_status": "verified", "matches_canonical": true, "reported_setting": "extended_thinking", "source_type": "leaderboard" }, { "model_id": "claude-sonnet-4", "benchmark_id": "gpqa_diamond", "score": 75.4, "reference_url": "https://www.anthropic.com/news/claude-4", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1 (single-pass)", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "top_p=0.95" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "Anthropic Claude 4 blog table: 75.4% single-pass / 83.8% w/ parallel.", "candidates": [ { "score": 78, "reference_url": "https://huggingface.co/MiniMaxAI/MiniMax-M2", "source_type": "model_card", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "notes": "MiniMax M2 model card: claude-sonnet-4=78." } ] }, { "model_id": "claude-sonnet-4", "benchmark_id": "swe_bench_verified", "score": 72.7, "reference_url": "https://www.anthropic.com/news/claude-4", "reported_setting": { "mode": "non-thinking", "effort": "default", "tools": "agentic (bash + edit)", "sampling": "pass@1 (avg 10 trials)", "judge": "rule-based", "harness": "simple bash+editor scaffold", "prompt_style": "default", "temperature": "top_p=0.95" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "Anthropic Claude 4 blog table: 72.7% single-pass / 80.2% w/ parallel." }, { "model_id": "claude-sonnet-4", "benchmark_id": "mmlu", "score": 86.5, "reference_url": "https://www.datacamp.com/blog/claude-4", "audit_status": "dropped", "notes": " [Dropped: DataCamp is a tutorial/learning platform aggregator, not primary source. Rule R5h-tutorial-blog.]" }, { "model_id": "claude-sonnet-4", "benchmark_id": "tau_bench_retail", "score": 80.5, "reference_url": "https://www.anthropic.com/news/claude-4", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "agentic (TAU-bench official tools)", "sampling": "pass@1 (single-pass)", "judge": "rule-based", "harness": "τ-bench official + max steps 100", "prompt_style": "prompt addendum to Agent Policy (Anthropic)", "temperature": "top_p=0.95" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Anthropic Claude 4 blog table. Prompt addendum + max steps 100." }, { "model_id": "claude-sonnet-4", "benchmark_id": "aime_2025", "score": 70.5, "reference_url": "https://www.anthropic.com/news/claude-4", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1 (single-pass)", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "top_p=0.95" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "Anthropic Claude 4 blog table: 70.5% single-pass / 85.0% w/ parallel. nucleus sampling top_p=0.95.", "candidates": [ { "score": 74, "reference_url": "https://huggingface.co/MiniMaxAI/MiniMax-M2", "source_type": "model_card", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "notes": "MiniMax M2 model card: claude-sonnet-4=74." } ] }, { "model_id": "claude-sonnet-4", "benchmark_id": "mmmu", "score": 74.4, "reference_url": "https://www.anthropic.com/news/claude-4", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1 (single-pass)", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "top_p=0.95" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "Anthropic Claude 4 blog table (MMMU validation)." }, { "model_id": "claude-sonnet-4", "benchmark_id": "swe_bench_pro", "score": 42.7, "reference_url": "https://scale.com/leaderboard/swe_bench_pro_public", "audit_status": "verified", "source_type": "leaderboard", "matches_canonical": false, "notes": " [Verified against Scale.com SWE-Bench Pro Public leaderboard (2026-04-29): claude-4-Sonnet = 42.7. Scale uses standardized evaluation scaffolding.]" }, { "model_id": "claude-sonnet-4", "benchmark_id": "aime_2024", "score": 43.4, "reference_url": "https://arxiv.org/abs/2507.20534", "audit_status": "verified", "source_type": "tech_report", "matches_canonical": false, "reported_setting": { "mode": "non-thinking", "effort": "default", "tools": "none (benchmark-specified)", "sampling": "avg@64", "judge": "rule-based", "harness": "Kimi team evaluation (official API)", "prompt_style": "default", "temperature": "unified (Kimi team setting)", "context": "8192 max output tokens" }, "notes": "Via third-party self-test by Moonshot AI (Kimi K2 tech report, Table 3, p.16). AIME2024 Avg@64, non-thinking mode (Kimi team eval), 8192 max output tokens." }, { "model_id": "claude-sonnet-4", "benchmark_id": "arena_hard", "score": 51.6, "reference_url": "https://arxiv.org/abs/2507.20534", "audit_status": "verified", "source_type": "tech_report", "matches_canonical": false, "reported_setting": { "mode": "non-thinking", "effort": "default", "tools": "none (benchmark-specified)", "sampling": "pass@1 (winrate)", "judge": "rule-based", "harness": "Kimi team evaluation (official API)", "prompt_style": "default", "temperature": "unified (Kimi team setting)", "context": "8192 max output tokens" }, "notes": "Via third-party self-test by Moonshot AI (Kimi K2 tech report, Table 3, p.16). Arena Hard v2.0 Hard Prompt winrate, non-thinking, 8192 max output tokens." }, { "model_id": "claude-sonnet-4", "benchmark_id": "hle", "score": 5.8, "reference_url": "https://arxiv.org/abs/2507.20534", "audit_status": "verified", "source_type": "tech_report", "matches_canonical": false, "reported_setting": { "mode": "non-thinking", "effort": "default", "tools": "none (benchmark-specified)", "sampling": "pass@1", "judge": "rule-based", "harness": "Kimi team evaluation (official API)", "prompt_style": "default", "temperature": "unified (Kimi team setting)", "context": "8192 max output tokens" }, "notes": "Via third-party self-test by Moonshot AI (Kimi K2 tech report, Table 3, p.16). HLE Text-Only Acc, non-thinking, 8192 max output tokens." }, { "model_id": "claude-sonnet-4", "benchmark_id": "hmmt_feb_2025", "score": 15.9, "reference_url": "https://arxiv.org/abs/2507.20534", "audit_status": "verified", "source_type": "tech_report", "matches_canonical": false, "reported_setting": { "mode": "non-thinking", "effort": "default", "tools": "none (benchmark-specified)", "sampling": "avg@32", "judge": "rule-based", "harness": "Kimi team evaluation (official API)", "prompt_style": "default", "temperature": "unified (Kimi team setting)", "context": "8192 max output tokens" }, "notes": "Via third-party self-test by Moonshot AI (Kimi K2 tech report, Table 3, p.16). HMMT 2025 Avg@32, non-thinking, 8192 max output tokens." }, { "model_id": "claude-sonnet-4", "benchmark_id": "humaneval", "score": 88.0, "reference_url": "https://www.datacamp.com/blog/claude-4", "audit_status": "dropped", "notes": " [Dropped: DataCamp is a tutorial/learning platform aggregator, not primary source. Rule R5h-tutorial-blog.]" }, { "model_id": "claude-sonnet-4", "benchmark_id": "ifeval", "score": 87.6, "reference_url": "https://arxiv.org/abs/2507.20534", "audit_status": "verified", "source_type": "tech_report", "matches_canonical": false, "reported_setting": { "mode": "non-thinking", "effort": "default", "tools": "none (benchmark-specified)", "sampling": "pass@1", "judge": "rule-based", "harness": "Kimi team evaluation (official API)", "prompt_style": "default", "temperature": "unified (Kimi team setting)", "context": "8192 max output tokens" }, "notes": "Via third-party self-test by Moonshot AI (Kimi K2 tech report, Table 3, p.16). IFEval Prompt Strict, non-thinking, 8192 max output tokens." }, { "model_id": "claude-sonnet-4", "benchmark_id": "livecodebench", "score": 66, "reference_url": "https://huggingface.co/MiniMaxAI/MiniMax-M2", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "MiniMax M2 model card: claude-sonnet-4=66." }, { "model_id": "claude-sonnet-4", "benchmark_id": "math_500", "score": 94.0, "reference_url": "https://arxiv.org/abs/2507.20534", "audit_status": "verified", "source_type": "tech_report", "matches_canonical": false, "reported_setting": { "mode": "non-thinking", "effort": "default", "tools": "none (benchmark-specified)", "sampling": "pass@1", "judge": "rule-based", "harness": "Kimi team evaluation (official API)", "prompt_style": "default", "temperature": "unified (Kimi team setting)", "context": "8192 max output tokens" }, "notes": "Via third-party self-test by Moonshot AI (Kimi K2 tech report, Table 3, p.16). MATH-500 Acc, non-thinking mode (Kimi team eval), 8192 max output tokens." }, { "model_id": "claude-sonnet-4", "benchmark_id": "mmlu_pro", "score": 84, "reference_url": "https://huggingface.co/MiniMaxAI/MiniMax-M2", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "MiniMax M2 model card: claude-sonnet-4=84." }, { "model_id": "claude-sonnet-4", "benchmark_id": "simpleqa", "score": 15.9, "reference_url": "https://arxiv.org/abs/2507.20534", "audit_status": "verified", "source_type": "tech_report", "matches_canonical": false, "reported_setting": { "mode": "non-thinking", "effort": "default", "tools": "none (benchmark-specified)", "sampling": "pass@1", "judge": "rule-based", "harness": "Kimi team evaluation (official API)", "prompt_style": "default", "temperature": "unified (Kimi team setting)", "context": "8192 max output tokens" }, "notes": "Via third-party self-test by Moonshot AI (Kimi K2 tech report, Table 3, p.16). SimpleQA Correct, non-thinking, 8192 max output tokens." }, { "model_id": "claude-sonnet-4", "benchmark_id": "arc_agi_1", "score": 40, "reference_url": "https://arcprize.org/leaderboard", "audit_status": "verified", "source_type": "leaderboard", "matches_canonical": true, "reported_setting": { "mode": "thinking", "effort": "16K thinking", "tools": "none", "sampling": "pass@1 (single-pass)", "judge": "rule-based", "harness": "official (arcprize.org)", "system_type": "CoT" }, "notes": "arcprize.org leaderboard audit: Claude Sonnet 4 (Thinking 16K) on leaderboard" }, { "model_id": "claude-sonnet-4", "benchmark_id": "arc_agi_2", "score": 5.9, "reference_url": "https://arcprize.org/leaderboard", "audit_status": "verified", "source_type": "leaderboard", "matches_canonical": true, "reported_setting": { "mode": "thinking", "effort": "16K thinking", "tools": "none", "sampling": "pass@1 (single-pass)", "judge": "rule-based", "harness": "official (arcprize.org)", "system_type": "CoT" }, "notes": "arcprize.org leaderboard audit: Claude Sonnet 4 (Thinking 16K) on leaderboard" }, { "model_id": "claude-sonnet-4", "benchmark_id": "ifbench", "score": 55, "reference_url": "https://huggingface.co/MiniMaxAI/MiniMax-M2", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "MiniMax M2 model card: claude-sonnet-4=55." }, { "model_id": "claude-sonnet-4", "benchmark_id": "osworld", "score": 42, "reference_url": "https://www.datacamp.com/blog/claude-4", "audit_status": "dropped", "notes": " [Dropped: DataCamp is a tutorial/learning platform aggregator, not primary source. Rule R5h-tutorial-blog.]" }, { "model_id": "claude-sonnet-4", "benchmark_id": "terminal_bench", "score": 35.5, "reference_url": "https://www.anthropic.com/news/claude-4", "reported_setting": { "mode": "non-thinking", "effort": "default", "tools": "agentic", "sampling": "pass@1 (single-pass)", "judge": "rule-based", "harness": "Claude Code agent framework", "prompt_style": "default", "temperature": "top_p=0.95" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "Anthropic Claude 4 blog table: 35.5% single-pass (Claude Code) / 41.3% w/ parallel. Same agent as non-Claude=33.5%.", "candidates": [ { "score": 36.4, "reference_url": "https://huggingface.co/MiniMaxAI/MiniMax-M2", "source_type": "model_card", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "agentic", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "notes": "MiniMax M2 model card: claude-sonnet-4=36.4." } ] }, { "model_id": "claude-sonnet-4", "benchmark_id": "terminal_bench_1", "score": 36.4, "reference_url": "https://www.tbench.ai/leaderboard/terminal-bench/1.0", "audit_status": "verified", "source_type": "leaderboard", "rule_ids": [ "R5d" ], "notes": "Confirmed 36.4% on tbench.ai Terminal-Bench 1.0 leaderboard (Terminus 2 scaffold, 2025-08-05)." }, { "model_id": "claude-opus-4", "benchmark_id": "gpqa_diamond", "score": 79.6, "reference_url": "https://www.anthropic.com/news/claude-4", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1 (single-pass)", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "top_p=0.95" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "Anthropic Claude 4 blog table: 79.6% single-pass / 83.3% w/ parallel test-time compute." }, { "model_id": "claude-opus-4", "benchmark_id": "swe_bench_verified", "score": 72.5, "reference_url": "https://www.anthropic.com/news/claude-4", "reported_setting": { "mode": "non-thinking", "effort": "default", "tools": "agentic (bash + edit)", "sampling": "pass@1 (avg 10 trials)", "judge": "rule-based", "harness": "simple bash+editor scaffold", "prompt_style": "default", "temperature": "top_p=0.95" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "Anthropic Claude 4 blog table: 72.5% single-pass / 79.4% w/ parallel test-time compute. We record single-pass." }, { "model_id": "claude-opus-4", "benchmark_id": "mmlu", "score": 88.8, "reference_url": "https://www.datacamp.com/blog/claude-4", "audit_status": "dropped", "notes": " [Dropped: DataCamp is a tutorial/learning platform aggregator, not primary source. Rule R5h-tutorial-blog.]" }, { "model_id": "claude-opus-4", "benchmark_id": "aime_2025", "score": 75.5, "reference_url": "https://www.anthropic.com/news/claude-4", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1 (single-pass)", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "top_p=0.95" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "Anthropic Claude 4 blog table: 75.5% single-pass / 90.0% w/ parallel test-time compute. nucleus sampling top_p=0.95." }, { "model_id": "claude-opus-4", "benchmark_id": "tau_bench_retail", "score": 81.4, "reference_url": "https://www.anthropic.com/news/claude-4", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "agentic (TAU-bench official tools)", "sampling": "pass@1 (single-pass)", "judge": "rule-based", "harness": "τ-bench official + max steps 100", "prompt_style": "prompt addendum to Agent Policy (Anthropic)", "temperature": "top_p=0.95" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Anthropic Claude 4 blog table. With extended thinking + tool use; prompt addendum to Retail Agent Policy; max steps 100." }, { "model_id": "claude-opus-4", "benchmark_id": "aime_2024", "score": 75.5, "reference_url": "https://www.datacamp.com/blog/claude-4", "audit_status": "dropped", "notes": " [Dropped: DataCamp is a tutorial/learning platform aggregator, not primary source. Rule R5h-tutorial-blog.]" }, { "model_id": "claude-opus-4", "benchmark_id": "mmmu", "score": 76.5, "reference_url": "https://www.anthropic.com/news/claude-4", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1 (single-pass)", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "top_p=0.95" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "Anthropic Claude 4 blog table (MMMU validation). With extended thinking up to 64K." }, { "model_id": "claude-opus-4", "benchmark_id": "arena_hard", "score": 59.7, "reference_url": "https://arxiv.org/abs/2507.20534", "audit_status": "verified", "source_type": "tech_report", "matches_canonical": false, "reported_setting": { "mode": "non-thinking", "effort": "default", "tools": "none (benchmark-specified)", "sampling": "pass@1 (winrate)", "judge": "rule-based", "harness": "Kimi team evaluation (official API)", "prompt_style": "default", "temperature": "unified (Kimi team setting)", "context": "8192 max output tokens" }, "notes": "Via third-party self-test by Moonshot AI (Kimi K2 tech report, Table 3, p.16). Arena Hard v2.0 Hard Prompt winrate, non-thinking, 8192 max output tokens." }, { "model_id": "claude-opus-4", "benchmark_id": "frontiermath", "score": 4.5, "reference_url": "https://epoch.ai/benchmarks/frontiermath", "audit_status": "verified", "source_type": "leaderboard", "matches_canonical": true, "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none" }, "notes": "epoch.ai FrontierMath-2025-02-28-Private: claude-opus-4-20250514 (Claude Opus 4) = 4.5% (default thinking, no explicit budget). Canonical effort=default matches. BP had 10.0 (unverified)." }, { "model_id": "claude-opus-4", "benchmark_id": "hle", "score": 7.1, "reference_url": "https://arxiv.org/abs/2507.20534", "audit_status": "verified", "source_type": "tech_report", "matches_canonical": false, "reported_setting": { "mode": "non-thinking", "effort": "default", "tools": "none (benchmark-specified)", "sampling": "pass@1", "judge": "rule-based", "harness": "Kimi team evaluation (official API)", "prompt_style": "default", "temperature": "unified (Kimi team setting)", "context": "8192 max output tokens" }, "notes": "Via third-party self-test by Moonshot AI (Kimi K2 tech report, Table 3, p.16). HLE Text-Only Acc, non-thinking, 8192 max output tokens." }, { "model_id": "claude-opus-4", "benchmark_id": "hmmt_feb_2025", "score": 15.9, "reference_url": "https://arxiv.org/abs/2507.20534", "audit_status": "verified", "source_type": "tech_report", "matches_canonical": false, "reported_setting": { "mode": "non-thinking", "effort": "default", "tools": "none (benchmark-specified)", "sampling": "avg@32", "judge": "rule-based", "harness": "Kimi team evaluation (official API)", "prompt_style": "default", "temperature": "unified (Kimi team setting)", "context": "8192 max output tokens" }, "notes": "Via third-party self-test by Moonshot AI (Kimi K2 tech report, Table 3, p.16). HMMT 2025 Avg@32, non-thinking, 8192 max output tokens." }, { "model_id": "claude-opus-4", "benchmark_id": "ifeval", "score": 89.7, "reference_url": "https://huggingface.co/Qwen/Qwen3-235B-A22B-Thinking-2507", "audit_status": "verified", "source_type": "model_card", "matches_canonical": true, "audit_note": "Qwen3-235B card: Claude4 Opus Thinking IFEval=89.7. Matches claude-opus-4 (thinking mode).", "audited_by": "lychee-audit-bot", "audited_at": "2026-04-29T13:17:14Z" }, { "model_id": "claude-opus-4", "benchmark_id": "livecodebench", "score": 47.4, "reference_url": "https://arxiv.org/abs/2507.20534", "audit_status": "verified", "source_type": "tech_report", "matches_canonical": false, "reported_setting": { "mode": "non-thinking", "effort": "default", "tools": "none (benchmark-specified)", "sampling": "pass@1", "judge": "rule-based", "harness": "Kimi team evaluation (official API)", "prompt_style": "default", "temperature": "unified (Kimi team setting)", "context": "8192 max output tokens" }, "notes": "Via third-party self-test by Moonshot AI (Kimi K2 tech report, Table 3, p.16). LiveCodeBench v6 (Aug 2024–May 2025) Pass@1, non-thinking, 8192 max output tokens." }, { "model_id": "claude-opus-4", "benchmark_id": "math_500", "score": 94.4, "reference_url": "https://arxiv.org/abs/2507.20534", "audit_status": "verified", "source_type": "tech_report", "matches_canonical": false, "reported_setting": { "mode": "non-thinking", "effort": "default", "tools": "none (benchmark-specified)", "sampling": "pass@1", "judge": "rule-based", "harness": "Kimi team evaluation (official API)", "prompt_style": "default", "temperature": "unified (Kimi team setting)", "context": "8192 max output tokens" }, "notes": "Via third-party self-test by Moonshot AI (Kimi K2 tech report, Table 3, p.16). MATH-500 Acc, non-thinking mode (Kimi team eval), 8192 max output tokens." }, { "model_id": "claude-opus-4", "benchmark_id": "mmlu_pro", "score": 86.6, "reference_url": "https://arxiv.org/abs/2507.20534", "audit_status": "verified", "source_type": "tech_report", "matches_canonical": false, "reported_setting": { "mode": "non-thinking", "effort": "default", "tools": "none (benchmark-specified)", "sampling": "pass@1 (EM)", "judge": "rule-based", "harness": "Kimi team evaluation (official API)", "prompt_style": "default", "temperature": "unified (Kimi team setting)", "context": "8192 max output tokens" }, "notes": "Via third-party self-test by Moonshot AI (Kimi K2 tech report, Table 3, p.16). MMLU-Pro EM, non-thinking, 8192 max output tokens." }, { "model_id": "claude-opus-4", "benchmark_id": "simpleqa", "score": 22.8, "reference_url": "https://arxiv.org/abs/2507.20534", "audit_status": "verified", "source_type": "tech_report", "matches_canonical": false, "reported_setting": { "mode": "non-thinking", "effort": "default", "tools": "none (benchmark-specified)", "sampling": "pass@1", "judge": "rule-based", "harness": "Kimi team evaluation (official API)", "prompt_style": "default", "temperature": "unified (Kimi team setting)", "context": "8192 max output tokens" }, "notes": "Via third-party self-test by Moonshot AI (Kimi K2 tech report, Table 3, p.16). SimpleQA Correct, non-thinking, 8192 max output tokens." }, { "model_id": "claude-opus-4", "benchmark_id": "arc_agi_1", "score": 35.7, "reference_url": "https://arcprize.org/leaderboard", "audit_status": "verified", "source_type": "leaderboard", "matches_canonical": true, "reported_setting": { "mode": "thinking", "effort": "16K thinking", "tools": "none", "sampling": "pass@1 (single-pass)", "judge": "rule-based", "harness": "official (arcprize.org)", "system_type": "CoT" }, "notes": "arcprize.org leaderboard audit: Claude Opus 4 (Thinking 16K) on leaderboard" }, { "model_id": "claude-opus-4", "benchmark_id": "arc_agi_2", "score": 8.6, "reference_url": "https://arcprize.org/leaderboard", "audit_status": "verified", "source_type": "leaderboard", "matches_canonical": true, "reported_setting": { "mode": "thinking", "effort": "16K thinking", "tools": "none", "sampling": "pass@1 (single-pass)", "judge": "rule-based", "harness": "official (arcprize.org)", "system_type": "CoT" }, "notes": "arcprize.org leaderboard audit: Claude Opus 4 (Thinking 16K) on leaderboard" }, { "model_id": "claude-opus-4", "benchmark_id": "codeforces_rating", "score": 1886, "reference_url": "https://www.datacamp.com/blog/claude-4", "audit_status": "dropped", "notes": " [Dropped: DataCamp is a tutorial/learning platform aggregator, not primary source. Rule R5h-tutorial-blog.]" }, { "model_id": "claude-opus-4", "benchmark_id": "critpt", "score": 0.3, "reference_url": "https://github.com/CritPt-Benchmark/CritPt", "audit_status": "verified", "reported_setting": "CritPt leaderboard, ε=1.5, % tasks solved", "source_type": "leaderboard", "matches_canonical": true }, { "model_id": "claude-opus-4", "benchmark_id": "osworld", "score": 38.2, "reference_url": "https://www.datacamp.com/blog/claude-4", "audit_status": "dropped", "notes": " [Dropped: DataCamp is a tutorial/learning platform aggregator, not primary source. Rule R5h-tutorial-blog.]" }, { "model_id": "claude-opus-4", "benchmark_id": "swe_bench_pro", "score": 35.8, "reference_url": "https://www.anthropic.com/news/claude-opus-4-1", "audit_status": "dropped", "rule_id": "R5g-ghost", "audit_note": "anthropic.com/news/claude-opus-4-1 mentions SWE-bench Verified=74.5% only; swe_bench_pro metric (35.8) not in source.", "audited_by": "lychee-audit-bot", "audited_at": "2026-07-26T00:00:00Z" }, { "model_id": "claude-opus-4", "benchmark_id": "terminal_bench", "score": 43.2, "reference_url": "https://www.anthropic.com/news/claude-4", "reported_setting": { "mode": "non-thinking", "effort": "default", "tools": "agentic", "sampling": "pass@1 (single-pass)", "judge": "rule-based", "harness": "Claude Code agent framework", "prompt_style": "default", "temperature": "top_p=0.95" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "Anthropic Claude 4 blog table: 43.2% single-pass (Claude Code agent) / 50.0% w/ parallel. Same agent as non-Claude=39.2%." }, { "model_id": "claude-opus-4", "benchmark_id": "terminal_bench_1", "score": 39, "reference_url": "https://www.tbench.ai/leaderboard/terminal-bench/1.0", "audit_status": "verified", "source_type": "leaderboard", "rule_ids": [ "R5d" ], "notes": "Confirmed 39.0% on tbench.ai Terminal-Bench 1.0 leaderboard (Terminus 2 scaffold, 2025-08-05). score=39.0" }, { "model_id": "claude-opus-4.1", "benchmark_id": "swe_bench_verified", "score": 74.5, "reference_url": "https://www.anthropic.com/news/claude-opus-4-1", "reported_setting": { "mode": "non-thinking", "effort": "default", "tools": "agentic (bash + edit)", "sampling": "pass@1 (avg 10 trials)", "judge": "rule-based", "harness": "simple bash+editor scaffold", "prompt_style": "default", "temperature": "top_p=0.95" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "Anthropic Opus 4.1 blog table: 74.5%, single-pass." }, { "model_id": "claude-opus-4.1", "benchmark_id": "gpqa_diamond", "score": 80.9, "reference_url": "https://www.anthropic.com/news/claude-opus-4-1", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1 (single-pass)", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "top_p=0.95" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "Anthropic Opus 4.1 blog table: 80.9% (with extended thinking)." }, { "model_id": "claude-opus-4.1", "benchmark_id": "mmlu", "score": 88.8, "reference_url": "https://www.anthropic.com/news/claude-opus-4-1", "audit_status": "needs_review", "rule_ids": [ "R5g_c" ], "notes": "Anthropic blog benchmark table is a CDN PNG image; cannot extract text value. R5g(c): value not found in source text." }, { "model_id": "claude-opus-4.1", "benchmark_id": "mmlu_pro", "score": 87.92, "reference_url": "https://artificialanalysis.ai/evaluations/mmlu-pro", "audit_status": "needs_review", "rule_ids": [ "R5g_c" ], "notes": "Reference URL is AA mmlu-pro eval page. AA shows claude-4-1-opus-thinking mmlu_pro=0.88→88.0%, but BP has 87.92. Close but not exact match; AA may have rounded. Only thinking variant available on AA. Current AA value: 88.0%." }, { "model_id": "claude-opus-4.1", "benchmark_id": "swe_bench_pro", "score": 22.7, "reference_url": "https://scale.com/leaderboard/swe_bench_pro_public", "audit_status": "verified", "rule_ids": [ "R4" ], "notes": "Confirmed 22.7% from scale.com analysis text: \"Claude Opus 4.1 decreases from 22.7% to 17.8% resolution\"; historical public score." }, { "model_id": "claude-opus-4.1", "benchmark_id": "aime_2025", "score": 78.0, "reference_url": "https://www.anthropic.com/news/claude-opus-4-1", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1 (single-pass)", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "top_p=0.95" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "Anthropic Opus 4.1 blog table: 78.0%, with extended thinking, top_p=0.95." }, { "model_id": "claude-opus-4.1", "benchmark_id": "frontiermath", "score": 5.9, "reference_url": "https://epoch.ai/benchmarks/frontiermath", "audit_status": "verified", "rule_ids": [ "R5g_c" ], "notes": "epoch.ai FrontierMath-2025-02-28-Private: claude-opus-4-1-20250805 (no explicit thinking budget) = 5.9%. Canonical effort=default matches. BP had 15.0.", "source_type": "leaderboard", "matches_canonical": true, "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none" } }, { "model_id": "claude-opus-4.1", "benchmark_id": "osworld", "score": 44.4, "reference_url": "https://assets.anthropic.com/m/64823ba7485345a7/Claude-Opus-4-5-System-Card.pdf", "audit_status": "verified", "notes": "FLAGGED cleared: OSWorld=44.4% confirmed in cross-model Table 2.3.A (p.19) of Opus 4.5 system card.", "rule_ids": [ "R3" ] }, { "model_id": "claude-opus-4.1", "benchmark_id": "terminal_bench", "score": 43.3, "reference_url": "https://www.anthropic.com/news/claude-opus-4-1", "reported_setting": { "mode": "non-thinking", "effort": "default", "tools": "agentic", "sampling": "pass@1 (avg 5 trials)", "judge": "rule-based", "harness": "Terminus 1", "prompt_style": "default", "temperature": "top_p=0.95" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "Anthropic Opus 4.1 blog table: 43.3%, Terminus 1, avg 5 trials. Differs from Opus 4 (Claude Code framework=43.2).", "candidates": [ { "score": 38.0, "reference_url": "https://www.tbench.ai/leaderboard/terminal-bench/2.0", "source_type": "leaderboard", "notes": "tbench 2.0 Terminus 2 result; BP primary (43.3%) from Anthropic blog differs." } ] }, { "model_id": "claude-opus-4.1", "benchmark_id": "hle", "score": 35, "reference_url": "https://www.anthropic.com/news/claude-opus-4-1", "audit_status": "flagged", "notes": "Anthropic blog benchmark table is CDN PNG; cannot confirm or deny flag. Keeping flagged." }, { "model_id": "claude-opus-4.1", "benchmark_id": "humaneval", "score": 93, "reference_url": "https://www.anthropic.com/news/claude-opus-4-1", "audit_status": "flagged", "notes": "Anthropic blog benchmark table is CDN PNG; cannot confirm or deny flag. Keeping flagged." }, { "model_id": "claude-opus-4.1", "benchmark_id": "livebench", "score": 61.4, "reference_url": "https://livebench.ai/", "audit_status": "verified", "rule_ids": [ "R5g_c" ], "notes": "livebench.ai table_2026_01_08 (latest as of 2026-04-28): claude-4-1-opus-20250805-thinking-32k avg=61.4. Canonical thinking mode matches. BP had 54.5.", "source_type": "leaderboard", "matches_canonical": true, "reported_setting": { "mode": "thinking", "effort": "default", "context": "32k thinking", "tools": "n/a" } }, { "model_id": "claude-opus-4.1", "benchmark_id": "livecodebench", "score": 63.2, "reference_url": "https://www.anthropic.com/news/claude-opus-4-1", "audit_status": "flagged", "notes": "Anthropic blog benchmark table is CDN PNG; cannot confirm or deny flag. Keeping flagged." }, { "model_id": "claude-opus-4.1", "benchmark_id": "simpleqa", "score": 43.5, "reference_url": "https://www.anthropic.com/news/claude-opus-4-1", "audit_status": "flagged", "notes": "Anthropic blog benchmark table is CDN PNG; cannot confirm or deny flag. Keeping flagged." }, { "model_id": "claude-opus-4.1", "benchmark_id": "terminal_bench_1", "score": 43.8, "reference_url": "https://www.tbench.ai/leaderboard/terminal-bench/1.0", "audit_status": "verified", "rule_ids": [ "R4" ], "notes": "Confirmed 43.8% on tbench.ai Terminal-Bench 1.0 leaderboard (Terminus 2 scaffold, 2025-09-30)." }, { "model_id": "claude-sonnet-4.5", "benchmark_id": "swe_bench_verified", "score": 77.2, "reference_url": "https://www.anthropic.com/news/claude-opus-4-5", "reported_setting": { "mode": "non-thinking", "effort": "high", "tools": "agentic (bash + edit)", "sampling": "pass@1 (avg 5 trials)", "judge": "rule-based", "harness": "simple bash+editor scaffold", "prompt_style": "interleaved scratchpads", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "Anthropic Opus 4.5 blog table: Sonnet 4.5 = 77.2% (no thinking).", "candidates": [ { "score": 77.2, "reference_url": "https://deepmind.google/models/gemini/flash/", "source_type": "official_blog", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "notes": "DeepMind /models/gemini/flash/ Performance table: SWE-bench Verified 77.2%" }, { "score": 77.2, "reference_url": "https://huggingface.co/mistralai/Mistral-Medium-3.5-128B/resolve/22b2b868a15677cfa6061277ed2f653d1349a9ab/images/image4.png?download=true", "source_type": "official_model_card", "reported_setting": { "mode": "thinking", "effort": "high (maximum)", "tools": "agentic coding scaffold and SWE-bench verifier", "sampling": "source does not state trial count", "judge": "benchmark-specified", "harness": "Mistral self-reported SWE-bench Verified evaluation", "prompt_style": "interleaved scratchpads", "temperature": "default (top_p, temperature)", "context": "200K (default)" }, "notes": "Displayed exactly: '77.2'. Research observation: medium-image4.png:1:2:score:reported. Chart footnotes: *All model evaluations were run with maximum reasoning settings. | **Sonnet 4.6 encountered higher rates of reasoning truncation due to external API restrictions than other models, which affected its performance. | *** Self-reported | **** Self-reported, Mistral is using context management and a discard-all strategy at 100k tokens. | τ³ scores as reported by Sierra for Claude Sonnet 4.5 and Qwen3.5. Others with user simulator: gpt-5.2 with reasoning_effort: low. 4 trials. Banking domain evaluated with terminal- or embedding-based agentic search retrieval, only highest score is reported." } ] }, { "model_id": "claude-sonnet-4.5", "benchmark_id": "gpqa_diamond", "score": 83.4, "reference_url": "https://www.anthropic.com/news/claude-opus-4-5", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1 (avg 5 trials)", "judge": "rule-based", "harness": "official", "prompt_style": "interleaved scratchpads", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "Anthropic Opus 4.5 blog table: Sonnet 4.5 = 83.4%.", "candidates": [ { "score": 83.4, "reference_url": "https://deepmind.google/models/gemini/flash/", "source_type": "official_blog", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "notes": "DeepMind /models/gemini/flash/ Performance table: GPQA Diamond 83.4%" }, { "score": 83.0, "reference_url": "https://huggingface.co/MiniMaxAI/MiniMax-M2.5", "source_type": "model_card", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "notes": "MiniMax M2.5 model card: claude-sonnet-4.5=83.0." }, { "score": 83, "reference_url": "https://huggingface.co/MiniMaxAI/MiniMax-M2", "source_type": "model_card", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "notes": "MiniMax M2 model card: claude-sonnet-4.5=83." }, { "score": 84.3, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "source_type": "model_card", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "notes": "Doubao Seed 2.0 model card Table 3: claude-sonnet-4.5=84.3." } ] }, { "model_id": "claude-sonnet-4.5", "benchmark_id": "tau_bench_retail", "score": 86.2, "reference_url": "https://assets.anthropic.com/m/64823ba7485345a7/Claude-Opus-4-5-System-Card.pdf", "audit_status": "verified", "rule_ids": [ "R3" ], "notes": "Confirmed τ²-Bench Retail=86.2% in cross-model Table 2.3.A (p.19) of Opus 4.5 system card." }, { "model_id": "claude-sonnet-4.5", "benchmark_id": "tau_bench_telecom", "score": 98.0, "reference_url": "https://assets.anthropic.com/m/64823ba7485345a7/Claude-Opus-4-5-System-Card.pdf", "audit_status": "verified", "rule_ids": [ "R3" ], "notes": "Confirmed τ²-Bench Telecom=98.0% in cross-model Table 2.3.A (p.19) of Opus 4.5 system card." }, { "model_id": "claude-sonnet-4.5", "benchmark_id": "osworld", "score": 61.4, "reference_url": "https://www.anthropic.com/news/claude-opus-4-5", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "agentic (browser/computer use)", "sampling": "pass@1 (avg 5 trials)", "judge": "rule-based", "harness": "OSWorld official", "prompt_style": "interleaved scratchpads", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "Anthropic Opus 4.5 blog table: Sonnet 4.5 = 61.4%." }, { "model_id": "claude-sonnet-4.5", "benchmark_id": "livecodebench", "score": 64.0, "reference_url": "https://huggingface.co/deepseek-ai/DeepSeek-V3.2/resolve/main/assets/paper.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "DeepSeek V3.2 tech report Table 2: claude-sonnet-4.5=64.0.", "candidates": [ { "score": 71, "reference_url": "https://huggingface.co/MiniMaxAI/MiniMax-M2", "source_type": "model_card", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "notes": "MiniMax M2 model card: claude-sonnet-4.5=71." } ] }, { "model_id": "claude-sonnet-4.5", "benchmark_id": "aime_2025", "score": 87.0, "reference_url": "https://deepmind.google/models/gemini/flash/", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "DeepMind /models/gemini/flash/ Performance table: AIME 2025 no tools 87.0% [Promoted to verified, prior unverified value deleted per R5d.]", "candidates": [ { "score": 100.0, "reference_url": "https://deepmind.google/models/gemini/flash/", "source_type": "official_blog", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "code", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "notes": "DeepMind /models/gemini/flash/ Performance table: AIME 2025 with code 100%" }, { "score": 88.0, "reference_url": "https://huggingface.co/MiniMaxAI/MiniMax-M2.5", "source_type": "model_card", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "notes": "MiniMax M2.5 model card: claude-sonnet-4.5=88.0." }, { "score": 88, "reference_url": "https://huggingface.co/MiniMaxAI/MiniMax-M2", "source_type": "model_card", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "notes": "MiniMax M2 model card: claude-sonnet-4.5=88." }, { "score": 86.7, "reference_url": "https://huggingface.co/mistralai/Mistral-Medium-3.5-128B/resolve/22b2b868a15677cfa6061277ed2f653d1349a9ab/images/image1.png?download=true", "source_type": "official_model_card", "reported_setting": { "mode": "thinking", "effort": "high (maximum)", "tools": "none", "sampling": "samples=16 (reported as avg@16)", "judge": "benchmark-specified", "harness": "Mistral official maximum-reasoning comparison", "prompt_style": "interleaved scratchpads", "temperature": "default (top_p, temperature)", "context": "200K (default)" }, "notes": "Displayed exactly: '86.7'. Research observation: medium-image1.png:1:2:score:reported. Chart footnotes: *All model evaluations were run with maximum reasoning settings. | **Sonnet 4.6 encountered higher rates of reasoning truncation due to external API restrictions than other models, which affected its performance." } ] }, { "model_id": "claude-sonnet-4.5", "benchmark_id": "humaneval", "score": 85.0, "reference_url": "https://www.getpassionfruit.com/blog/gpt-5-1-vs-claude-4-5-sonnet-vs-gemini-3-pro-vs-deepseek-v3-2-the-definitive-2025-ai-model-comparison", "audit_status": "needs_review", "rule_ids": [ "R5h" ], "notes": "Source is third-party blog (getpassionfruit.com); R5h priority. Not verifiable from official Anthropic source." }, { "model_id": "claude-sonnet-4.5", "benchmark_id": "mmmu", "score": 77.8, "reference_url": "https://www.anthropic.com/news/claude-opus-4-5", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1 (avg 5 trials)", "judge": "rule-based", "harness": "official", "prompt_style": "interleaved scratchpads", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "Anthropic Opus 4.5 blog table: Sonnet 4.5 = 77.8%." }, { "model_id": "claude-sonnet-4.5", "benchmark_id": "mmmu_pro", "score": 63.4, "reference_url": "https://www.anthropic.com/news/claude-sonnet-4-6", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "none", "sampling": "pass@1 (avg 5 trials)", "judge": "rule-based", "harness": "official", "prompt_style": "updated grading: separate Sonnet 4 grader; no 'think step-by-step' prefix", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "Anthropic Sonnet 4.6 blog table: claude-sonnet-4.5 = 63.4.", "candidates": [ { "score": 68.0, "reference_url": "https://deepmind.google/models/gemini/flash/", "source_type": "official_blog", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "notes": "DeepMind /models/gemini/flash/ Performance table: MMMU-Pro 68.0%" } ] }, { "model_id": "claude-sonnet-4.5", "benchmark_id": "swe_bench_pro", "score": 48.4, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "varies", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card: claude-sonnet-4.5=48.4." }, { "model_id": "claude-sonnet-4.5", "benchmark_id": "mrcr_v2", "score": 47.1, "reference_url": "https://deepmind.google/models/gemini/flash/", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default", "context": "128k" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "DeepMind /models/gemini/flash/ Performance table: MRCR v2 8-needle 128k 47.1% [Promoted to verified, prior unverified value deleted per R5d.]" }, { "model_id": "claude-sonnet-4.5", "benchmark_id": "mmlu", "score": 86.5, "reference_url": "https://www.leanware.co/insights/claude-sonnet-4-5-overview", "audit_status": "needs_review", "rule_ids": [ "R5h" ], "notes": "Source is third-party blog (leanware.co); R5h priority. Official source has image-only table." }, { "model_id": "claude-sonnet-4.5", "benchmark_id": "codeforces_rating", "score": 1480, "reference_url": "https://huggingface.co/deepseek-ai/DeepSeek-V3.2/resolve/main/assets/paper.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "DeepSeek V3.2 tech report Table 2: claude-sonnet-4.5=1480.", "candidates": [ { "score": 1485, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "source_type": "model_card", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "notes": "Doubao Seed 2.0 model card Table 3: claude-sonnet-4.5=1485." } ] }, { "model_id": "claude-sonnet-4.5", "benchmark_id": "hle", "score": 17.7, "reference_url": "https://www.anthropic.com/news/claude-sonnet-4-6", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "none (no-tools column)", "sampling": "pass@1 (avg 5 trials)", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "Anthropic Sonnet 4.6 blog table: claude-sonnet-4.5 = 17.7.", "candidates": [ { "score": 13.7, "reference_url": "https://deepmind.google/models/gemini/flash/", "source_type": "official_blog", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "notes": "DeepMind /models/gemini/flash/ Performance table: HLE no tools 13.7% (Claude Sonnet 4.5 Thinking)" }, { "score": 13.7, "reference_url": "https://huggingface.co/deepseek-ai/DeepSeek-V3.2/resolve/main/assets/paper.pdf", "source_type": "tech_report", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "notes": "DeepSeek V3.2 tech report Table 2: claude-sonnet-4.5=13.7." }, { "score": 13.7, "reference_url": "https://z.ai/blog/glm-4.7", "source_type": "official_blog", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "notes": "GLM-4.7 blog: claude-sonnet-4.5=13.7." }, { "score": 19.8, "reference_url": "https://huggingface.co/moonshotai/Kimi-K2-Thinking", "source_type": "model_card", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "1.0" }, "notes": "Kimi K2-Thinking model card: claude-sonnet-4.5=19.8." }, { "score": 17.3, "reference_url": "https://huggingface.co/MiniMaxAI/MiniMax-M2.5", "source_type": "model_card", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "notes": "MiniMax M2.5 model card: claude-sonnet-4.5=17.3." }, { "score": 14.5, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "source_type": "model_card", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "notes": "Doubao Seed 2.0 model card Table 3: claude-sonnet-4.5=14.5." } ] }, { "model_id": "claude-sonnet-4.5", "benchmark_id": "mmlu_pro", "score": 88.2, "reference_url": "https://huggingface.co/deepseek-ai/DeepSeek-V3.2/resolve/main/assets/paper.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "DeepSeek V3.2 tech report Table 2: claude-sonnet-4.5=88.2.", "candidates": [ { "score": 87.5, "reference_url": "https://huggingface.co/moonshotai/Kimi-K2-Thinking", "source_type": "model_card", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "1.0" }, "notes": "Kimi K2-Thinking model card: claude-sonnet-4.5=87.5." }, { "score": 88, "reference_url": "https://huggingface.co/MiniMaxAI/MiniMax-M2", "source_type": "model_card", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "notes": "MiniMax M2 model card: claude-sonnet-4.5=88." }, { "score": 88.0, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "source_type": "model_card", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "notes": "Doubao Seed 2.0 model card Table 3: claude-sonnet-4.5=88.0." } ] }, { "model_id": "claude-sonnet-4.5", "benchmark_id": "terminal_bench", "score": 50.0, "reference_url": "https://www.anthropic.com/news/claude-opus-4-5", "reported_setting": { "mode": "thinking", "effort": "128K thinking", "tools": "agentic", "sampling": "pass@1 (avg 5 trials)", "judge": "rule-based", "harness": "Terminus-2", "prompt_style": "interleaved scratchpads", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "Anthropic Opus 4.5 blog table: Sonnet 4.5 = 50.0%, Terminus-2 harness.", "candidates": [ { "score": 51.0, "reference_url": "https://www.anthropic.com/news/claude-sonnet-4-6", "source_type": "third_party", "reported_setting": { "note": "see primary" }, "notes": "Sonnet 4.6 blog reports 51.0 (vs primary verified=50.0 from Opus 4.5 blog)." }, { "score": 42.8, "reference_url": "https://deepmind.google/models/gemini/flash/", "source_type": "official_blog", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "Terminus-2", "prompt_style": "default", "temperature": "default" }, "notes": "DeepMind /models/gemini/flash/ Performance table: Terminal-Bench 2.0 Terminus-2 42.8%" }, { "score": 42.8, "reference_url": "https://huggingface.co/deepseek-ai/DeepSeek-V3.2/resolve/main/assets/paper.pdf", "source_type": "tech_report", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "agentic", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "notes": "DeepSeek V3.2 tech report Table 2: claude-sonnet-4.5=42.8." }, { "score": 42.8, "reference_url": "https://z.ai/blog/glm-4.7", "source_type": "official_blog", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "agentic", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "notes": "GLM-4.7 blog: claude-sonnet-4.5=42.8." }, { "score": 45.2, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "source_type": "model_card", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "varies", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "notes": "(merged from terminal_bench_2) Doubao Seed 2.0 model card: claude-sonnet-4.5=45.2." } ] }, { "model_id": "claude-sonnet-4.5", "benchmark_id": "aime_2024", "score": 88, "reference_url": "https://www.leanware.co/insights/claude-sonnet-4-5-overview", "audit_status": "needs_review", "rule_ids": [ "R5h" ], "notes": "Source is third-party blog (leanware.co); R5h priority." }, { "model_id": "claude-sonnet-4.5", "benchmark_id": "arc_agi_1", "score": 70.9, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card Table 3: claude-sonnet-4.5=70.9." }, { "model_id": "claude-sonnet-4.5", "benchmark_id": "arc_agi_2", "score": 13.6, "reference_url": "https://www.anthropic.com/news/claude-opus-4-5", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1 (avg 5 trials)", "judge": "rule-based", "harness": "official", "prompt_style": "interleaved scratchpads", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "Anthropic Opus 4.5 blog table: Sonnet 4.5 = 13.6%.", "candidates": [ { "score": 13.6, "reference_url": "https://deepmind.google/models/gemini/flash/", "source_type": "official_blog", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "notes": "DeepMind /models/gemini/flash/ Performance table: ARC-AGI-2 ARC Prize Verified 13.6%" } ] }, { "model_id": "claude-sonnet-4.5", "benchmark_id": "brumo_2025", "score": 90.83, "reference_url": "https://matharena.ai/?comp=brumo--brumo_2025", "audit_status": "verified", "rule_ids": [ "R4" ], "notes": "Confirmed brumo_2025=90.83% on matharena.ai model page." }, { "model_id": "claude-sonnet-4.5", "benchmark_id": "cmimc_2025", "score": 66.88, "reference_url": "https://matharena.ai/?comp=cmimc--cmimc_2025", "audit_status": "verified", "rule_ids": [ "R4" ], "notes": "Confirmed cmimc_2025=66.88% on matharena.ai model page." }, { "model_id": "claude-sonnet-4.5", "benchmark_id": "ifeval", "score": 88.5, "reference_url": "https://www.leanware.co/insights/claude-sonnet-4-5-overview", "audit_status": "needs_review", "rule_ids": [ "R5h" ], "notes": "Source is third-party blog (leanware.co); R5h priority." }, { "model_id": "claude-sonnet-4.5", "benchmark_id": "livebench", "score": 67.9, "reference_url": "https://livebench.ai/", "audit_status": "verified", "rule_ids": [ "R5g_c" ], "notes": "livebench.ai table_2026_01_08 (latest as of 2026-04-28): claude-sonnet-4-5-20250929-thinking-64k avg=67.9. Canonical thinking+high effort matches. BP had 53.7.", "source_type": "leaderboard", "matches_canonical": true, "reported_setting": { "mode": "thinking", "effort": "high", "context": "64k thinking", "tools": "n/a" } }, { "model_id": "claude-sonnet-4.5", "benchmark_id": "math_500", "score": 95.8, "reference_url": "https://www.leanware.co/insights/claude-sonnet-4-5-overview", "audit_status": "needs_review", "rule_ids": [ "R5h" ], "notes": "Source is third-party blog (leanware.co); R5h priority." }, { "model_id": "claude-sonnet-4.5", "benchmark_id": "matharena_apex_2025", "score": 1.56, "reference_url": "https://matharena.ai/?comp=apex--apex_2025", "audit_status": "verified", "rule_ids": [ "R4" ], "notes": "Confirmed matharena_apex_2025=1.56% on matharena.ai model page.", "candidates": [ { "score": 1.0, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "source_type": "model_card", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "notes": "(merged from apex) Doubao Seed 2.0 model card Table 3: claude-sonnet-4.5=1.0." } ] }, { "model_id": "claude-sonnet-4.5", "benchmark_id": "simpleqa", "score": 47, "reference_url": "https://www.vellum.ai/blog/claude-opus-4-5-benchmarks", "audit_status": "dropped", "rule_ids": [ "R5h" ], "notes": "Source is third-party blog (vellum.ai); R5h priority. DROPPED (R5h): random third-party blog/aggregator, no primary trail", "rule_id": "R5h-third-party-blog" }, { "model_id": "claude-sonnet-4.5", "benchmark_id": "smt_2025", "score": 83.96, "reference_url": "https://matharena.ai/?comp=smt--smt_2025", "audit_status": "verified", "rule_ids": [ "R4" ], "notes": "Confirmed smt_2025=83.96% on matharena.ai model page." }, { "model_id": "claude-sonnet-4.5", "benchmark_id": "terminal_bench_1", "score": 51, "reference_url": "https://www.tbench.ai/leaderboard/terminal-bench/1.0", "audit_status": "verified", "rule_ids": [ "R4" ], "notes": "Confirmed 51.0% on tbench.ai Terminal-Bench 1.0 leaderboard (Terminus 2 scaffold, 2025-10-31)." }, { "model_id": "claude-haiku-4.5", "benchmark_id": "mmmlu", "score": 83.0, "reference_url": "https://www.anthropic.com/news/claude-haiku-4-5", "reported_setting": { "mode": "thinking", "effort": "128K thinking budget", "sampling": "pass@1", "judge": "official harness", "harness": "official", "prompt_style": "default", "temperature": "default", "tools": "none" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "Anthropic blog, avg 10 runs over 14 non-English languages, 128K thinking budget." }, { "model_id": "claude-haiku-4.5", "benchmark_id": "mmmu", "score": 73.2, "reference_url": "https://www.anthropic.com/news/claude-haiku-4-5", "reported_setting": { "mode": "thinking", "effort": "128K thinking budget", "sampling": "pass@1", "judge": "official harness", "harness": "official", "prompt_style": "default", "temperature": "default", "tools": "none" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "Anthropic blog (MMMU validation), 128K thinking, avg 10 runs." }, { "model_id": "claude-haiku-4.5", "benchmark_id": "swe_bench_verified", "score": 73.3, "reference_url": "https://www.anthropic.com/news/claude-haiku-4-5", "reported_setting": { "mode": "thinking", "effort": "128K thinking budget", "sampling": "pass@1", "judge": "official harness", "harness": "official", "prompt_style": "default", "temperature": "default", "tools": "agentic (bash + file editing)" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Anthropic blog, full 500 instances, avg 50 trials, 128K thinking, default sampling. NOTE: Anthropic added prompt addendum (\"use tools 100+ times, implement own tests first\") — deviates from default prompt → matches_canonical=false.", "candidates": [ { "score": 73.3, "reference_url": "https://huggingface.co/internlm/Intern-S2-Preview/resolve/3b53ea8d2f8ed46abefa507e3f6f56ee0c662c4a/figs/performance.png", "source_type": "official_model_card", "reported_setting": { "before_slash": "officially reported in model technical report", "after_slash": "AgentCompass unified framework with Mini-SWE-Agent", "footnote": "[2]", "official_report_details": { "harness": "Anthropic official SWE-bench Verified evaluation", "instances": 500, "trials_average": 50, "thinking_budget": "128K", "prompt_addendum": "use tools 100+ times; implement own tests first", "campaign_matches_canonical": false } }, "notes": "Displayed exactly: 73.30 / 68.8[2]; Alternatives: [{\"footnote\": \"[2]\", \"role\": \"official_report\", \"value\": \"73.30\"}, {\"footnote\": \"[2]\", \"harness\": \"AgentCompass with Mini-SWE-Agent\", \"role\": \"unified_re_evaluation\", \"value\": \"68.8\"}]" }, { "score": 68.8, "reference_url": "https://huggingface.co/internlm/Intern-S2-Preview/resolve/3b53ea8d2f8ed46abefa507e3f6f56ee0c662c4a/figs/performance.png", "source_type": "official_model_card", "reported_setting": { "role": "unified_re_evaluation", "harness": "AgentCompass with Mini-SWE-Agent", "footnote": "[2]" }, "notes": "Alternative reported in compound source cell; displayed cell: 73.30 / 68.8[2]" }, { "score": 66.6, "reference_url": "https://microsoft.ai/pdf/MAI-Code-1-Flash-Model-Card.PDF", "source_type": "official_model_card", "reported_setting": { "mode": "adaptive reasoning", "effort": "adaptive solution length control", "tools": "GitHub Copilot / VS Code production tools", "sampling": "pass@1; benchmark-specific where stated", "judge": "benchmark-specified", "harness": "GitHub Copilot VS Code production harness", "prompt_style": "official Microsoft evaluation prompts", "temperature": "source does not state globally", "context": "256K", "notes": "Coding-focused 137B-total / 5B-active model. Core coding scores use the same production harness for both models." }, "notes": "Displayed exactly as 66.6. Official Microsoft-reported result." } ] }, { "model_id": "claude-haiku-4.5", "benchmark_id": "gpqa_diamond", "score": 73.0, "reference_url": "https://www.anthropic.com/news/claude-haiku-4-5", "reported_setting": { "mode": "thinking", "effort": "128K thinking budget", "sampling": "pass@1", "judge": "official harness", "harness": "official", "prompt_style": "default", "temperature": "default", "tools": "none" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "Anthropic blog, 128K thinking, avg over 10 runs. Prior BP value 80.9 not found on blog (corrected).", "candidates": [ { "score": 73.2, "reference_url": "https://microsoft.ai/pdf/MAI-Code-1-Flash-Model-Card.PDF", "source_type": "official_model_card", "reported_setting": { "mode": "adaptive reasoning", "effort": "adaptive solution length control", "tools": "none", "sampling": "source does not state trial count", "judge": "benchmark-specific", "harness": "Microsoft MAI-Code-1-Flash comparative evaluation", "prompt_style": "source does not state", "temperature": "source does not state", "context": "256K model context; per-task usage reported separately", "notes": "Accuracy column from the official model-card/release table; benchmark=gpqa_diamond." }, "notes": "Displayed exactly as 73.2. Official Microsoft-reported result." } ] }, { "model_id": "claude-haiku-4.5", "benchmark_id": "aime_2025", "score": 96.3, "reference_url": "https://www.anthropic.com/news/claude-haiku-4-5", "reported_setting": { "mode": "thinking", "effort": "128K thinking budget", "sampling": "pass@1", "judge": "official harness", "harness": "official", "prompt_style": "default", "temperature": "default", "tools": "python interpreter" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Anthropic blog, with python tool, avg 10 runs of pass@1 over 16 trials, 128K thinking. Blog also reports 80.7% no-tools (canonical-matching). Updated source from airank.dev to Anthropic primary.", "candidates": [ { "score": 83, "reference_url": "https://huggingface.co/mistralai/Mistral-Small-4-119B-2603/resolve/a11f36bebf709121056b1dbcc943d1c6afbe494d/images/livecode.png?download=true", "source_type": "official_model_card", "reported_setting": { "mode": "thinking", "effort": "128K thinking budget", "tools": "none", "sampling": "pass@1; repeat count undisclosed", "judge": "benchmark-specified", "harness": "Mistral official model-card evaluation", "prompt_style": "default", "temperature": "default (Anthropic default)", "context": "source does not state" }, "notes": "Displayed exactly: '83'. Research observation: small-livecode.png:1:3:score:reasoning." }, { "score": 34, "reference_url": "https://huggingface.co/mistralai/Mistral-Small-4-119B-2603/resolve/a11f36bebf709121056b1dbcc943d1c6afbe494d/images/livecode.png?download=true", "source_type": "official_model_card", "reported_setting": { "mode": "non-thinking", "effort": "none", "tools": "none", "sampling": "pass@1; repeat count undisclosed", "judge": "benchmark-specified", "harness": "Mistral official model-card evaluation", "prompt_style": "default", "temperature": "default (Anthropic default)", "context": "source does not state" }, "notes": "Displayed exactly: '34'. Research observation: small-livecode.png:1:3:score:instruct." } ] }, { "model_id": "claude-haiku-4.5", "benchmark_id": "osworld", "score": 50.7, "reference_url": "https://www.anthropic.com/news/claude-haiku-4-5", "reported_setting": { "mode": "thinking", "effort": "128K thinking budget", "sampling": "pass@1", "judge": "official harness", "harness": "official", "prompt_style": "default", "temperature": "default", "tools": "agentic (OSWorld-Verified, 100 max steps)" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "Anthropic blog, official OSWorld-Verified framework, avg 4 runs, 128K total + 2K per-step thinking budget." }, { "model_id": "claude-haiku-4.5", "benchmark_id": "swe_bench_pro", "score": 39.45, "reference_url": "https://scale.com/leaderboard/swe_bench_pro_public", "audit_status": "verified", "rule_ids": [ "R4" ], "notes": "Confirmed 39.45% on scale.com SWE-bench Pro Public leaderboard (current live entry).", "candidates": [ { "score": 39.5, "reference_url": "https://deepmind.google/models/model-cards/gemini-3-5-flash-lite/", "source_type": "official_model_card", "reported_setting": { "mode": "thinking/reasoning", "effort": "extended thinking/best available", "tools": "code execution", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "Google internal Antigravity for Gemini; provider for others", "temperature": "default", "snapshot": "July 2026" }, "notes": "Google Gemini 3.5 Flash-Lite July 2026 matrix: SWE-Bench Pro (Public). Exact source effort, tools, sampling and harness are preserved." }, { "score": 35.2, "reference_url": "https://microsoft.ai/pdf/MAI-Code-1-Flash-Model-Card.PDF", "source_type": "official_model_card", "reported_setting": { "mode": "adaptive reasoning", "effort": "adaptive solution length control", "tools": "GitHub Copilot / VS Code production tools", "sampling": "pass@1; benchmark-specific where stated", "judge": "benchmark-specified", "harness": "GitHub Copilot VS Code production harness", "prompt_style": "official Microsoft evaluation prompts", "temperature": "source does not state globally", "context": "256K", "notes": "Coding-focused 137B-total / 5B-active model. Core coding scores use the same production harness for both models." }, "notes": "Displayed exactly as 35.2. Official Microsoft-reported result." } ] }, { "model_id": "claude-haiku-4.5", "benchmark_id": "terminal_bench", "score": 41.0, "reference_url": "https://www.anthropic.com/news/claude-haiku-4-5", "reported_setting": { "mode": "thinking", "effort": "128K thinking budget", "sampling": "pass@1", "judge": "official harness", "harness": "official", "prompt_style": "default", "temperature": "default", "tools": "agentic (Terminus 2 default scaffold)" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "Anthropic blog, Terminus 2 default agent, avg 11 runs (6 no-thinking 40.21%, 5 with 32K thinking 41.75%). Replaces prior tbench.ai leaderboard value 28.3.", "candidates": [ { "score": 41.6, "reference_url": "https://microsoft.ai/pdf/MAI-Code-1-Flash-Model-Card.PDF", "source_type": "official_model_card", "reported_setting": { "mode": "adaptive reasoning", "effort": "adaptive solution length control", "tools": "GitHub Copilot / VS Code production tools", "sampling": "pass@1; benchmark-specific where stated", "judge": "benchmark-specified", "harness": "GitHub Copilot VS Code production harness", "prompt_style": "official Microsoft evaluation prompts", "temperature": "source does not state globally", "context": "256K", "notes": "Coding-focused 137B-total / 5B-active model. Core coding scores use the same production harness for both models." }, "notes": "Displayed exactly as 41.6. Official Microsoft-reported result." } ] }, { "model_id": "claude-haiku-4.5", "benchmark_id": "arc_agi_1", "score": 47.7, "reference_url": "https://arcprize.org/leaderboard", "audit_status": "verified", "rule_ids": [ "R5g_c" ], "notes": "arcprize.org leaderboard audit: Claude Haiku 4.5 (Thinking 32K) on leaderboard; canonical is 128K", "source_type": "leaderboard", "matches_canonical": false, "reported_setting": { "mode": "thinking", "effort": "32K thinking", "tools": "none", "sampling": "pass@1 (single-pass)", "judge": "rule-based", "harness": "official (arcprize.org)", "system_type": "CoT" } }, { "model_id": "claude-haiku-4.5", "benchmark_id": "arc_agi_2", "score": 4, "reference_url": "https://arcprize.org/leaderboard", "audit_status": "verified", "rule_ids": [ "R5g_c" ], "notes": "arcprize.org leaderboard audit: Claude Haiku 4.5 (Thinking 32K) on leaderboard; canonical is 128K", "source_type": "leaderboard", "matches_canonical": false, "reported_setting": { "mode": "thinking", "effort": "32K thinking", "tools": "none", "sampling": "pass@1 (single-pass)", "judge": "rule-based", "harness": "official (arcprize.org)", "system_type": "CoT" } }, { "model_id": "claude-haiku-4.5", "benchmark_id": "humaneval", "score": 91.5, "reference_url": "https://www.anthropic.com/claude/haiku", "audit_status": "needs_review", "rule_ids": [ "R5g_c" ], "notes": "Anthropic claude/haiku model page benchmark table is CDN PNG image; cannot extract text value. R5g(c)." }, { "model_id": "claude-haiku-4.5", "benchmark_id": "ifeval", "score": 85, "reference_url": "https://www.anthropic.com/claude/haiku", "audit_status": "needs_review", "rule_ids": [ "R5g_c" ], "notes": "Anthropic claude/haiku model page benchmark table is CDN PNG image; cannot extract text value. R5g(c)." }, { "model_id": "claude-haiku-4.5", "benchmark_id": "livebench", "score": 61.0, "reference_url": "https://livebench.ai/", "audit_status": "verified", "rule_ids": [ "R5g_c" ], "notes": "livebench.ai table_2026_01_08 (latest as of 2026-04-28): claude-haiku-4-5-20251001-thinking-64k avg=61.0. Canonical requires 128K thinking budget but livebench only has 64k variant; used 64k as best available. BP had 45.3.", "source_type": "leaderboard", "matches_canonical": false, "reported_setting": { "mode": "thinking", "effort": "n/a", "context": "64k thinking", "tools": "n/a" } }, { "model_id": "claude-haiku-4.5", "benchmark_id": "livecodebench", "score": 52, "reference_url": "https://artificialanalysis.ai/models/claude-4-5-haiku", "audit_status": "needs_review", "rule_ids": [ "R5g_c" ], "notes": "AA model page (claude-4-5-haiku): non-reasoning variant livecodebench=0.511→51.1%; reasoning variant=0.615→61.5%. BP has 52, which matches neither. BP canonical is thinking mode but neither AA variant shows 52. Current AA (non-reasoning): 51.1%, (reasoning): 61.5%." }, { "model_id": "claude-haiku-4.5", "benchmark_id": "math_500", "score": 90.2, "reference_url": "https://www.anthropic.com/claude/haiku", "audit_status": "needs_review", "rule_ids": [ "R5g_c" ], "notes": "Anthropic claude/haiku model page benchmark table is CDN PNG image; cannot extract text value. R5g(c)." }, { "model_id": "claude-haiku-4.5", "benchmark_id": "mmlu_pro", "score": 82.16, "reference_url": "https://huggingface.co/internlm/Intern-S2-Preview/resolve/3b53ea8d2f8ed46abefa507e3f6f56ee0c662c4a/figs/performance.png", "reported_setting": { "table_framework_claim": "OpenCompass and VLMEvalKit used to evaluate all models", "exact_per_cell_settings": "not stated" }, "matches_canonical": false, "source_type": "official_model_card", "audit_status": "verified", "notes": "Displayed exactly: 82.16" }, { "model_id": "claude-opus-4.5", "benchmark_id": "swe_bench_verified", "score": 80.9, "reference_url": "https://www.anthropic.com/news/claude-opus-4-5", "reported_setting": { "mode": "non-thinking", "effort": "high", "tools": "agentic (bash + edit)", "sampling": "pass@1 (avg 5 trials)", "judge": "rule-based", "harness": "simple bash+editor scaffold", "prompt_style": "interleaved scratchpads", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "Anthropic Opus 4.5 blog table: 80.9% (no thinking). avg 5 trials." }, { "model_id": "claude-opus-4.5", "benchmark_id": "gpqa_diamond", "score": 87.0, "reference_url": "https://www.anthropic.com/news/claude-opus-4-5", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1 (avg 5 trials)", "judge": "rule-based", "harness": "official", "prompt_style": "interleaved scratchpads", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "Anthropic Opus 4.5 blog table: 87.0% (with extended thinking 64K).", "candidates": [ { "score": 86.9, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "source_type": "model_card", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "notes": "Doubao Seed 2.0 model card Table 3: claude-opus-4.5=86.9." } ] }, { "model_id": "claude-opus-4.5", "benchmark_id": "mmlu", "score": 90.8, "reference_url": "https://assets.anthropic.com/m/64823ba7485345a7/Claude-Opus-4-5-System-Card.pdf", "audit_status": "verified", "rule_ids": [ "R3" ], "notes": "MMMLU=90.77 in Table 2.3.A (p.19) of Opus 4.5 system card; rounds to 90.8 as reported." }, { "model_id": "claude-opus-4.5", "benchmark_id": "mmlu_pro", "score": 89.3, "reference_url": "https://huggingface.co/moonshotai/Kimi-K2.5", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "1.0" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Kimi K2.5 model card: claude-opus-4.5=89.3.", "candidates": [ { "score": 89.5, "reference_url": "https://huggingface.co/Qwen/Qwen3.5-397B-A17B", "source_type": "model_card", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "notes": "Qwen3.5-397B-A17B HF model card cross-model table (Qwen self-test of 3rd-party model, mc=false): alternate measurement 89.5 (3rd-party Qwen self-test)." } ] }, { "model_id": "claude-opus-4.5", "benchmark_id": "osworld", "score": 66.3, "reference_url": "https://www.anthropic.com/news/claude-opus-4-5", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "agentic (browser/computer use)", "sampling": "pass@1 (avg 5 trials)", "judge": "rule-based", "harness": "OSWorld official", "prompt_style": "interleaved scratchpads", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "Anthropic Opus 4.5 blog table: 66.3% (computer use)." }, { "model_id": "claude-opus-4.5", "benchmark_id": "arc_agi_2", "score": 37.6, "reference_url": "https://www.anthropic.com/news/claude-opus-4-5", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1 (avg 5 trials)", "judge": "rule-based", "harness": "official", "prompt_style": "interleaved scratchpads", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "Anthropic Opus 4.5 blog table: 37.6% (Verified, novel problem solving).", "candidates": [ { "score": 29.1, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "source_type": "model_card", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "notes": "Doubao Seed 2.0 model card Table 3: claude-opus-4.5=29.1." } ] }, { "model_id": "claude-opus-4.5", "benchmark_id": "browsecomp", "score": 67.8, "reference_url": "https://www.anthropic.com/news/claude-sonnet-4-6", "reported_setting": { "mode": "non-thinking", "effort": "max", "tools": "agentic (web search + fetch + programmatic tool calling, context compaction at 50k up to 10M)", "sampling": "pass@1 (avg 5 trials)", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "Anthropic Sonnet 4.6 blog table: claude-opus-4.5 = 67.8.", "candidates": [ { "score": 37.0, "reference_url": "https://z.ai/blog/glm-5", "source_type": "official_blog", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "agentic", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "notes": "GLM-5 blog: claude-opus-4.5=37.0." }, { "score": 37.0, "reference_url": "https://huggingface.co/moonshotai/Kimi-K2.5", "source_type": "model_card", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "agentic", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "1.0" }, "notes": "Kimi K2.5 model card: claude-opus-4.5=37.0." } ] }, { "model_id": "claude-opus-4.5", "benchmark_id": "frontiermath", "score": 20.7, "reference_url": "https://epoch.ai/benchmarks/frontiermath", "audit_status": "verified", "rule_ids": [ "R5g_c" ], "notes": "epoch.ai FrontierMath-2025-02-28-Private: claude-opus-4-5-20251101_32K (32k thinking) = 20.7%. Canonical requires 64K thinking budget but epoch only has 16k (20.3%) and 32k (20.7%); used 32k as closest. BP had 21.0 (likely rounded from 20.7).", "source_type": "leaderboard", "matches_canonical": false, "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "context": "32k thinking" } }, { "model_id": "claude-opus-4.5", "benchmark_id": "hle", "score": 30.8, "reference_url": "https://www.anthropic.com/news/claude-sonnet-4-6", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "none (no-tools column)", "sampling": "pass@1 (avg 5 trials)", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "Anthropic Sonnet 4.6 blog table: claude-opus-4.5 = 30.8.", "candidates": [ { "score": 28.4, "reference_url": "https://z.ai/blog/glm-5", "source_type": "official_blog", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "notes": "GLM-5 blog: claude-opus-4.5=28.4." }, { "score": 28.4, "reference_url": "https://huggingface.co/MiniMaxAI/MiniMax-M2.5", "source_type": "model_card", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "notes": "MiniMax M2.5 model card: claude-opus-4.5=28.4." }, { "score": 23.7, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "source_type": "model_card", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "notes": "Doubao Seed 2.0 model card Table 3: claude-opus-4.5=23.7." } ] }, { "model_id": "claude-opus-4.5", "benchmark_id": "tau_bench_telecom", "score": 98.2, "reference_url": "https://assets.anthropic.com/m/64823ba7485345a7/Claude-Opus-4-5-System-Card.pdf", "audit_status": "verified", "rule_ids": [ "R3" ], "notes": "Confirmed τ²-Bench Telecom=98.2% in Table 2.3.A (p.19) of Opus 4.5 system card." }, { "model_id": "claude-opus-4.5", "benchmark_id": "tau_bench_retail", "score": 88.9, "reference_url": "https://assets.anthropic.com/m/64823ba7485345a7/Claude-Opus-4-5-System-Card.pdf", "audit_status": "verified", "rule_ids": [ "R3" ], "notes": "Confirmed τ²-Bench Retail=88.9% in Table 2.3.A (p.19) of Opus 4.5 system card." }, { "model_id": "claude-opus-4.5", "benchmark_id": "swe_bench_pro", "score": 55.4, "reference_url": "https://huggingface.co/moonshotai/Kimi-K2.5", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "agentic (bash + edit)", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "1.0" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Kimi K2.5 model card: claude-opus-4.5=55.4." }, { "model_id": "claude-opus-4.5", "benchmark_id": "aime_2025", "score": 92.8, "reference_url": "https://huggingface.co/moonshotai/Kimi-K2.5", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "1.0" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Kimi K2.5 model card: claude-opus-4.5=92.8.", "candidates": [ { "score": 91.0, "reference_url": "https://huggingface.co/MiniMaxAI/MiniMax-M2.5", "source_type": "model_card", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "notes": "MiniMax M2.5 model card: claude-opus-4.5=91.0." }, { "score": 91.3, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "source_type": "model_card", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "notes": "Doubao Seed 2.0 model card Table 3: claude-opus-4.5=91.3." } ] }, { "model_id": "claude-opus-4.5", "benchmark_id": "simpleqa", "score": 72.0, "reference_url": "https://artificialanalysis.ai/articles/claude-opus-4-5-benchmarks-and-analysis", "audit_status": "needs_review", "rule_ids": [ "R5g_c" ], "notes": "Reference URL is AA article page (claude-opus-4-5-benchmarks-and-analysis). AA does not publish SimpleQA scores; field is not present in AA model data for any model. Cannot verify from AA source. Original source needs identification." }, { "model_id": "claude-opus-4.5", "benchmark_id": "chatbot_arena_elo", "score": 1469, "reference_url": "https://lmarena.ai/leaderboard/text", "audit_status": "verified", "notes": "lmarena.ai text arena (fetched 2026-04-29). Arena model name: claude-opus-4-5-20251101 (rank 19). ELO updates continuously; score reflects latest available.", "source_type": "leaderboard", "matches_canonical": true }, { "model_id": "claude-opus-4.5", "benchmark_id": "mmmu", "score": 80.7, "reference_url": "https://www.anthropic.com/news/claude-opus-4-5", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1 (avg 5 trials)", "judge": "rule-based", "harness": "official", "prompt_style": "interleaved scratchpads", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "Anthropic Opus 4.5 blog table: 80.7% (validation, with extended thinking).", "candidates": [ { "score": 81.6, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "source_type": "model_card", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "notes": "Doubao Seed 2.0 model card: claude-opus-4.5=81.6." } ] }, { "model_id": "claude-opus-4.5", "benchmark_id": "math_500", "score": 85.0, "reference_url": "https://www.vellum.ai/blog/claude-opus-4-5-benchmarks", "audit_status": "dropped", "rule_ids": [ "R5h" ], "notes": "Source is third-party blog (vellum.ai); R5h priority. DROPPED (R5h): random third-party blog/aggregator, no primary trail", "rule_id": "R5h-third-party-blog" }, { "model_id": "claude-opus-4.5", "benchmark_id": "ifeval", "score": 90.9, "reference_url": "https://huggingface.co/Qwen/Qwen3.5-397B-A17B", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Qwen3.5-397B-A17B HF model card cross-model table (Qwen self-test of 3rd-party model, mc=false): ifeval 90.9 (3rd-party Qwen self-test). [R5d: prior unverified value 90.0 from https://www.vellum.ai/blog/claude-opus-4-5-benchmarks deleted.]" }, { "model_id": "claude-opus-4.5", "benchmark_id": "livecodebench", "score": 84.8, "reference_url": "https://huggingface.co/Qwen/Qwen3.5-397B-A17B", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Qwen3.5-397B-A17B HF model card cross-model table (Qwen self-test of 3rd-party model, mc=false): livecodebench 84.8 (3rd-party Qwen self-test). [R5d: prior unverified value 68.0 from https://www.anthropic.com/news/claude-opus-4-5 deleted.]" }, { "model_id": "claude-opus-4.5", "benchmark_id": "video_mmmu", "score": 84.4, "reference_url": "https://huggingface.co/moonshotai/Kimi-K2.5", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "1.0" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Kimi K2.5 model card: claude-opus-4.5=84.4." }, { "model_id": "claude-opus-4.5", "benchmark_id": "aa_lcr", "score": 71.3, "reference_url": "https://huggingface.co/moonshotai/Kimi-K2.5", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "1.0" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Kimi K2.5 model card: claude-opus-4.5=71.3.", "candidates": [ { "score": 74.0, "reference_url": "https://huggingface.co/Qwen/Qwen3.5-397B-A17B", "source_type": "model_card", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "notes": "Qwen3.5-397B-A17B HF model card cross-model table (Qwen self-test of 3rd-party model, mc=false): alternate measurement 74.0 (3rd-party Qwen self-test)." }, { "score": 74.0, "reference_url": "https://huggingface.co/MiniMaxAI/MiniMax-M2.5", "source_type": "model_card", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "notes": "MiniMax M2.5 model card: claude-opus-4.5=74.0." } ] }, { "model_id": "claude-opus-4.5", "benchmark_id": "terminal_bench", "score": 59.3, "reference_url": "https://www.anthropic.com/news/claude-opus-4-5", "reported_setting": { "mode": "thinking", "effort": "128K thinking", "tools": "agentic", "sampling": "pass@1 (avg 5 trials)", "judge": "rule-based", "harness": "Terminus-2", "prompt_style": "interleaved scratchpads", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "Anthropic Opus 4.5 blog table: 59.3%, Terminus-2 harness, 128K thinking, avg 5 trials.", "candidates": [ { "score": 59.8, "reference_url": "https://www.anthropic.com/news/claude-sonnet-4-6", "source_type": "third_party", "reported_setting": { "note": "see primary" }, "notes": "Sonnet 4.6 blog reports 59.8 (vs primary verified=59.3 from Opus 4.5 blog)." }, { "score": 60.2, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "source_type": "model_card", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "varies", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "notes": "(merged from terminal_bench_2) Doubao Seed 2.0 model card: claude-opus-4.5=60.2." }, { "score": 57.8, "reference_url": "https://www.tbench.ai/leaderboard/terminal-bench/2.0", "source_type": "leaderboard", "notes": "tbench 2.0 Terminus 2 result; BP primary (59.3%) from Anthropic blog slightly differs." }, { "score": 52.1, "reference_url": "https://cursor.com/resources/Composer2.pdf", "source_type": "tech_report", "reported_setting": { "effort": "high", "tools": "Cursor coding agent", "sampling": "pass@1", "harness": "Claude Code leaderboard" }, "notes": "Composer 2 technical report Table 1 / Terminal-Bench 2.0 / Opus 4.5 High / value 1: For third-party public benchmarks, slash-separated values are Cursor/official-harness then self-reported. SWE prompts prepend 'please solve this github issue'; Terminal prompts add solution-format instructions." } ] }, { "model_id": "claude-opus-4.5", "benchmark_id": "aime_2024", "score": 90, "reference_url": "https://artificialanalysis.ai/articles/claude-opus-4-5-benchmarks-and-analysis", "audit_status": "needs_review", "rule_ids": [ "R5g_c" ], "notes": "Reference URL is AA article page. Current AA data: claude-4-opus-thinking (reasoning) aime=0.757→75.7%; claude-4-opus (non-thinking) aime=0.563→56.3%. BP has 90 which matches neither. 90% may be Anthropic lab-claimed value mistakenly attributed to AA. Current AA (thinking): 75.7%." }, { "model_id": "claude-opus-4.5", "benchmark_id": "arc_agi_1", "score": 84.0, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card Table 3: claude-opus-4.5=84.0." }, { "model_id": "claude-opus-4.5", "benchmark_id": "codeforces_rating", "score": 1701, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card Table 3: claude-opus-4.5=1701." }, { "model_id": "claude-opus-4.5", "benchmark_id": "humaneval", "score": 95.1, "reference_url": "https://artificialanalysis.ai/articles/claude-opus-4-5-benchmarks-and-analysis", "audit_status": "needs_review", "rule_ids": [ "R5g_c" ], "notes": "Reference URL is AA article page. Current AA data: claude-4-opus (non-thinking) humaneval=0.970→97.0%, claude-4-opus-thinking=N/A. BP has 95.1% which does not match AA current value (97.0%). Possible stale/different evaluation config." }, { "model_id": "claude-opus-4.5", "benchmark_id": "livebench", "score": 76.0, "reference_url": "https://livebench.ai/", "audit_status": "verified", "rule_ids": [ "R5g_c" ], "notes": "livebench.ai table_2026_01_08 (latest as of 2026-04-28): claude-opus-4-5-20251101-thinking-64k-high-effort avg=76.0. EXACT MATCH with BP value 76. Canonical thinking+high+64K matches.", "source_type": "leaderboard", "matches_canonical": true, "reported_setting": { "mode": "thinking", "effort": "high", "context": "64k thinking", "tools": "n/a" } }, { "model_id": "claude-opus-4.5", "benchmark_id": "simplebench", "score": 62, "reference_url": "https://lmcouncil.ai/benchmarks", "audit_status": "dropped", "rule_ids": [ "R5g_c" ], "notes": "lmcouncil.ai scores are JS-rendered; could not extract value. R5g(c). DROPPED (R5h): random third-party blog/aggregator, no primary trail | DROPPED (R5g-c stale, never verified, source not audited in any pass)", "rule_id": "R5h-third-party-blog" }, { "model_id": "claude-opus-4.6", "benchmark_id": "gpqa_diamond", "score": 91.3, "reference_url": "https://www-cdn.anthropic.com/8b8380204f74670be75e81c820ca8dda846ab289.pdf", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "none", "sampling": "pass@1 (avg 5 trials)", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Mythos sys card Table 6.3.A: Opus 4.6 = 91.3%. Verified per Anthropic standard config (thinking max effort, default sampling, avg 5 trials).", "candidates": [ { "score": 91.3, "reference_url": "https://deepmind.google/models/gemini/pro/", "source_type": "official_blog", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "notes": "DeepMind /models/gemini/pro/ cross-model Performance table: GPQA Diamond no tools 91.3% (Thinking Max)" }, { "score": 90.0, "reference_url": "https://huggingface.co/MiniMaxAI/MiniMax-M2.5", "source_type": "model_card", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "notes": "MiniMax M2.5 model card: claude-opus-4.6=90.0." }, { "score": 91.3, "reference_url": "https://microsoft.ai/wp-content/uploads/2026/06/main_20260602_2.pdf", "source_type": "official_paper", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "none (per-bench override)", "sampling": "pass@1 (avg 5 trials)", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default", "context": "up to 1M", "notes": "Per Mythos System Card Table 6.3.A: standard config = adaptive thinking max effort, default sampling, avg 5 trials, context up to 1M." }, "notes": "Displayed exactly as 91.3. Official Microsoft-reported result. Comparator value is quoted from that provider's official model card/release; use the campaign canonical setting." } ] }, { "model_id": "claude-opus-4.6", "benchmark_id": "swe_bench_verified", "score": 80.8, "reference_url": "https://www-cdn.anthropic.com/8b8380204f74670be75e81c820ca8dda846ab289.pdf", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "agentic", "sampling": "pass@1 (avg 5 trials)", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Mythos sys card Table 6.3.A: Opus 4.6 = 80.8%. Verified per Anthropic standard config (thinking max effort, default sampling, avg 5 trials).", "candidates": [ { "score": 80.8, "reference_url": "https://deepmind.google/models/gemini/pro/", "source_type": "official_blog", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "notes": "DeepMind /models/gemini/pro/ cross-model Performance table: SWE-Bench Verified single attempt 80.8%" }, { "score": 80.8, "reference_url": "https://microsoft.ai/wp-content/uploads/2026/06/main_20260602_2.pdf", "source_type": "official_paper", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "none (per-bench override)", "sampling": "pass@1 (avg 5 trials)", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default", "context": "up to 1M", "notes": "Per Mythos System Card Table 6.3.A: standard config = adaptive thinking max effort, default sampling, avg 5 trials, context up to 1M." }, "notes": "Displayed exactly as 80.8. Official Microsoft-reported result. Comparator value is quoted from that provider's official model card/release; use the campaign canonical setting." }, { "score": 80.8, "reference_url": "https://www-cdn.anthropic.com/6a5fa276ac68b9aeb0c8b6af5fa36326e0e166dd/Claude%20Opus%204.6%20System%20Card.pdf", "source_type": "official_blog", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "source-reported agentic tools", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official Anthropic evaluation", "prompt_style": "provider official evaluation prompt", "temperature": "source/provider setting", "context": "source/provider setting", "input_modalities": "text", "trials": "source-reported" }, "notes": "Exact provider-official score and reported setting." }, { "score": 78.7, "reference_url": "https://static.stepfun.com/blog/step-3.7-flash/", "source_type": "official_blog", "reported_setting": { "quantization_config": "common release evaluation; artifact precision not established", "configuration": "internal reproduce", "mode": "source does not state", "effort": "source does not state", "tools_search": "agentic coding tools", "shots": "source does not state", "trials_samples": "source does not state", "aggregation_pass_k": "% resolved", "context": "source does not state", "output_cap": "source does not state", "temperature": "source does not state", "top_p": "source does not state", "judge": "benchmark-specified; exact StepFun judge not stated", "harness_agent": "Claude Opus 4.6 internal reproduce harness", "dataset_version_split": "SWE-bench Verified 500", "multimodal_input": false }, "notes": "Advisor is an inference configuration of Step 3.7 Flash, not a separate model. Cost is excluded separately." } ] }, { "model_id": "claude-opus-4.6", "benchmark_id": "aime_2025", "score": 95.6, "reference_url": "https://huggingface.co/MiniMaxAI/MiniMax-M2.5", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "MiniMax M2.5 model card: claude-opus-4.6=95.6.", "candidates": [ { "score": 99.8, "reference_url": "https://microsoft.ai/wp-content/uploads/2026/06/main_20260602_2.pdf", "source_type": "official_paper", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "none (per-bench override)", "sampling": "pass@1 (avg 5 trials)", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default", "context": "up to 1M", "notes": "Per Mythos System Card Table 6.3.A: standard config = adaptive thinking max effort, default sampling, avg 5 trials, context up to 1M." }, "notes": "Displayed exactly as 99.8. Official Microsoft-reported result. Comparator value is quoted from that provider's official model card/release; use the campaign canonical setting." } ] }, { "model_id": "claude-opus-4.6", "benchmark_id": "mmlu_pro", "score": 89.1, "reference_url": "https://huggingface.co/deepseek-ai/DeepSeek-V4-Pro", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "DeepSeek V4-Pro model card (third-party): claude-opus-4.6 (per their own canonical config) = 89.1.", "candidates": [ { "score": 89.5, "reference_url": "https://ernie.baidu.com/blog/posts/ernie-5.1-0508-release/benchmark.png", "source_type": "official_blog", "reported_setting": { "metric": "accuracy" }, "notes": "Accuracy row matches MMLU-Pro; source does not disclose tools. Research observation obs-027." } ] }, { "model_id": "claude-opus-4.6", "benchmark_id": "osworld", "score": 72.7, "reference_url": "https://www-cdn.anthropic.com/8b8380204f74670be75e81c820ca8dda846ab289.pdf", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "agentic", "sampling": "pass@1 (avg 5 trials)", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Mythos sys card Table 6.3.A: Opus 4.6 = 72.7%. Verified per Anthropic standard config (thinking max effort, default sampling, avg 5 trials)." }, { "model_id": "claude-opus-4.6", "benchmark_id": "arc_agi_2", "score": 68.8, "reference_url": "https://www.anthropic.com/news/claude-sonnet-4-6", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "none", "sampling": "pass@1 (avg 5 trials)", "judge": "rule-based", "harness": "official", "prompt_style": "120k thinking budget", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "Anthropic Sonnet 4.6 blog table: claude-opus-4.6 = 68.8.", "candidates": [ { "score": 68.8, "reference_url": "https://deepmind.google/models/gemini/pro/", "source_type": "official_blog", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "notes": "DeepMind /models/gemini/pro/ cross-model Performance table: ARC-AGI-2 ARC Prize Verified 68.8% (Thinking Max)" } ] }, { "model_id": "claude-opus-4.6", "benchmark_id": "browsecomp", "score": 83.7, "reference_url": "https://www.anthropic.com/news/claude-opus-4-7", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "agentic", "sampling": "pass@1 (avg 5 trials)", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "Anthropic Opus 4.7 blog table: claude-opus-4.6 = 83.7%.", "candidates": [ { "score": 84.0, "reference_url": "https://deepmind.google/models/gemini/pro/", "source_type": "official_blog", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "web+code", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "notes": "DeepMind /models/gemini/pro/ cross-model Performance table: BrowseComp Search+Python+Browse 84.0%" } ] }, { "model_id": "claude-opus-4.6", "benchmark_id": "hle", "score": 40.0, "reference_url": "https://www-cdn.anthropic.com/8b8380204f74670be75e81c820ca8dda846ab289.pdf", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "none", "sampling": "pass@1 (avg 5 trials)", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Mythos sys card Table 6.3.A: Opus 4.6 = 40.0%. Verified per Anthropic standard config (thinking max effort, default sampling, avg 5 trials).", "candidates": [ { "score": 40.0, "reference_url": "https://deepmind.google/models/gemini/pro/", "source_type": "official_blog", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "notes": "DeepMind /models/gemini/pro/ cross-model Performance table: HLE no tools 40.0% (Thinking Max)" }, { "score": 53.1, "reference_url": "https://deepmind.google/models/gemini/pro/", "source_type": "official_blog", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "web+code", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "notes": "DeepMind /models/gemini/pro/ cross-model Performance table: HLE Search+Code 53.1% (Thinking Max)" }, { "score": 36.7, "reference_url": "https://z.ai/blog/glm-5.1", "source_type": "official_blog", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "notes": "GLM-5.1 blog: claude-opus-4.6=36.7." }, { "score": 30.7, "reference_url": "https://huggingface.co/MiniMaxAI/MiniMax-M2.5", "source_type": "model_card", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "notes": "MiniMax M2.5 model card: claude-opus-4.6=30.7." } ] }, { "model_id": "claude-opus-4.6", "benchmark_id": "frontiermath", "score": 40.7, "reference_url": "https://epoch.ai/benchmarks/frontiermath", "audit_status": "verified", "rule_ids": [ "R5g_c" ], "notes": "epoch.ai FrontierMath-2025-02-28-Private: claude-opus-4-6_max = 40.7% (adaptive thinking, max effort). Canonical effort=max matches. BP had 40.0 (matched 32k thinking variant, not max).", "source_type": "leaderboard", "matches_canonical": true, "reported_setting": { "mode": "thinking", "effort": "max", "tools": "none" } }, { "model_id": "claude-opus-4.6", "benchmark_id": "mrcr_v2", "score": 84.0, "reference_url": "https://deepmind.google/models/gemini/pro/", "candidates": [], "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default", "context": "128k" }, "source_type": "official_blog", "audit_status": "verified", "matches_canonical": true, "notes": "DeepMind /models/gemini/pro/ Performance table (PROMOTED from candidate; prior unverified value 93.0 from https://www.vellum.ai/blog/claude-opus-4-6-benchmarks deleted per audit rule). DeepMind /models/gemini/pro/ cross-model Performance table: MRCR v2 8-needle 128k average 84.0%" }, { "model_id": "claude-opus-4.6", "benchmark_id": "simpleqa", "score": 72.0, "reference_url": "https://www.vellum.ai/blog/claude-opus-4-6-benchmarks", "audit_status": "dropped", "rule_ids": [ "R5h" ], "notes": "Source is third-party blog (vellum.ai); R5h priority. DROPPED (R5h): random third-party blog/aggregator, no primary trail", "rule_id": "R5h-third-party-blog" }, { "model_id": "claude-opus-4.6", "benchmark_id": "ifeval", "score": 92.2, "reference_url": "https://longcat.chat/blog/longcat-2.0/", "reported_setting": { "mode": "thinking", "effort": "source does not state", "tools": "none", "sampling": "source does not state", "judge": "benchmark-specified", "harness": "Meituan unified in-house harness", "prompt_style": "default", "temperature": "0.7; top_k=-1; top_p=0.95", "context": "source does not state", "notes": "Measured in-house by Meituan unless otherwise noted." }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "LongCat-2.0 official tech blog: IFEval = 92.2. Measured in-house by Meituan under the reported unified harness." }, { "model_id": "claude-opus-4.6", "benchmark_id": "humaneval", "score": 95.0, "reference_url": "https://www.vellum.ai/blog/claude-opus-4-6-benchmarks", "audit_status": "dropped", "rule_ids": [ "R5h" ], "notes": "Source is third-party blog (vellum.ai); R5h priority. DROPPED (R5h): random third-party blog/aggregator, no primary trail", "rule_id": "R5h-third-party-blog" }, { "model_id": "claude-opus-4.6", "benchmark_id": "livecodebench", "score": 88.8, "reference_url": "https://huggingface.co/deepseek-ai/DeepSeek-V4-Pro", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "DeepSeek V4-Pro model card (third-party): claude-opus-4.6 (per their own canonical config) = 88.8." }, { "model_id": "claude-opus-4.6", "benchmark_id": "tau_bench_retail", "score": 91.9, "reference_url": "https://www.vellum.ai/blog/claude-opus-4-6-benchmarks", "audit_status": "dropped", "rule_ids": [ "R5h" ], "notes": "Source is third-party blog (vellum.ai); R5h priority. DROPPED (R5h): random third-party blog/aggregator, no primary trail", "rule_id": "R5h-third-party-blog" }, { "model_id": "claude-opus-4.6", "benchmark_id": "critpt", "score": 12.6, "reference_url": "https://artificialanalysis.ai/evaluations/critpt", "audit_status": "verified", "rule_ids": [ "R5g_c" ], "notes": "ArtificialAnalysis aggregator; used as fallback when no primary source available (R5d-aggregator-fallback). AA CritPT eval page: claude-opus-4-6-adaptive critpt=0.126 → 12.6%. AA adaptive = max-effort adaptive reasoning variant.", "source_type": "third_party", "reported_setting": { "mode": "thinking", "effort": "max (adaptive reasoning)", "sampling": "pass@1", "harness": "AA standard evaluation", "notes": "AA slug: claude-opus-4-6-adaptive; critpt=12.6%" } }, { "model_id": "claude-opus-4.6", "benchmark_id": "gdpval_aa_elo", "score": 1606, "reference_url": "https://www.anthropic.com/news/claude-sonnet-4-6", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "none", "sampling": "pass@1 (avg 5 trials)", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "Anthropic Sonnet 4.6 blog table: claude-opus-4.6 = 1606.", "candidates": [ { "score": 1619, "reference_url": "https://huggingface.co/deepseek-ai/DeepSeek-V4-Pro", "source_type": "third_party", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "notes": "DeepSeek V4-Pro model card (third-party): claude-opus-4.6 (per their own canonical config) = 1619." }, { "score": 1606, "reference_url": "https://www-cdn.anthropic.com/6a5fa276ac68b9aeb0c8b6af5fa36326e0e166dd/Claude%20Opus%204.6%20System%20Card.pdf", "source_type": "official_blog", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "none", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official Anthropic evaluation", "prompt_style": "provider official evaluation prompt", "temperature": "source/provider setting", "context": "source/provider setting", "input_modalities": "text", "trials": "source-reported" }, "notes": "Exact provider-official score and reported setting." } ] }, { "model_id": "claude-opus-4.6", "benchmark_id": "aa_intelligence_index", "score": 53, "reference_url": "https://artificialanalysis.ai/evaluations/artificial-analysis-intelligence-index", "audit_status": "verified", "rule_ids": [ "R5g_d" ], "notes": "ArtificialAnalysis aggregator; used as fallback when no primary source available (R5d-aggregator-fallback). AA Intelligence Index: claude-opus-4-6-adaptive = 52.95 (displayed as 53). AA adaptive = max-effort adaptive reasoning variant.", "source_type": "third_party", "reported_setting": { "mode": "thinking", "effort": "max (adaptive reasoning)", "harness": "AA Intelligence Index composite", "notes": "AA slug: claude-opus-4-6-adaptive; II=52.95 rounds to 53" } }, { "model_id": "claude-opus-4.6", "benchmark_id": "mmmu", "score": 77.0, "reference_url": "https://www.vellum.ai/blog/claude-opus-4-6-benchmarks", "audit_status": "dropped", "rule_ids": [ "R5h" ], "notes": "Source is third-party blog (vellum.ai); R5h priority. DROPPED (R5h): random third-party blog/aggregator, no primary trail", "rule_id": "R5h-third-party-blog" }, { "model_id": "claude-opus-4.6", "benchmark_id": "math_500", "score": 93.0, "reference_url": "https://www.vellum.ai/blog/claude-opus-4-6-benchmarks", "audit_status": "dropped", "rule_ids": [ "R5h" ], "notes": "Source is third-party blog (vellum.ai); R5h priority. DROPPED (R5h): random third-party blog/aggregator, no primary trail", "rule_id": "R5h-third-party-blog" }, { "model_id": "claude-opus-4.6", "benchmark_id": "mmlu", "score": 90.8, "reference_url": "https://www.vellum.ai/blog/claude-opus-4-6-benchmarks", "audit_status": "dropped", "rule_ids": [ "R5h" ], "notes": "Source is third-party blog (vellum.ai); R5h priority. DROPPED (R5h): random third-party blog/aggregator, no primary trail", "rule_id": "R5h-third-party-blog" }, { "model_id": "claude-opus-4.6", "benchmark_id": "chatbot_arena_elo", "score": 1496, "reference_url": "https://lmarena.ai/leaderboard/text", "audit_status": "verified", "notes": "lmarena.ai text arena (fetched 2026-04-29). Arena model name: claude-opus-4-6 (rank 3). ELO updates continuously; score reflects latest available.", "source_type": "leaderboard", "matches_canonical": true, "candidates": [ { "score": 1502, "reference_url": "https://ernie.baidu.com/blog/posts/ernie-5.1-0508-release/creative-capability.png", "source_type": "official_blog", "reported_setting": { "leaderboard": "Arena AI Text", "metric": "Arena score", "style_control": "on", "mode": "thinking", "snapshot_date": "2026-04-30" }, "notes": "April 30, 2026 Text Arena snapshot; style control on. Research observation obs-049. Displayed rank 2. Identical asset is published in four page placements. The same PNG bytes are published under two filenames and four English/Chinese page placements." }, { "score": 1497, "reference_url": "https://ernie.baidu.com/blog/posts/ernie-5.1-0508-release/creative-capability.png", "source_type": "official_blog", "reported_setting": { "leaderboard": "Arena AI Text", "metric": "Arena score", "style_control": "on", "mode": "non-thinking/unspecified", "snapshot_date": "2026-04-30" }, "notes": "April 30, 2026 Text Arena snapshot; style control on. Research observation obs-050. Displayed rank 3. Identical asset is published in four page placements. The same PNG bytes are published under two filenames and four English/Chinese page placements." } ] }, { "model_id": "claude-opus-4.6", "benchmark_id": "tau_bench_telecom", "score": 98.2, "reference_url": "https://www.anthropic.com/news/claude-opus-4-6", "audit_status": "needs_review", "rule_ids": [ "R5g_c" ], "notes": "Anthropic blog benchmark table is a CDN PNG image; cannot extract text value. R5g(c). Note: Opus 4.5 system card shows tau_bench_telecom=98.2% for Opus 4.5; Opus 4.6 value is unverified." }, { "model_id": "claude-opus-4.6", "benchmark_id": "swe_bench_pro", "score": 53.4, "reference_url": "https://www-cdn.anthropic.com/8b8380204f74670be75e81c820ca8dda846ab289.pdf", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "agentic", "sampling": "pass@1 (avg 5 trials)", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Mythos sys card Table 6.3.A: Opus 4.6 = 53.4%. Verified per Anthropic standard config (thinking max effort, default sampling, avg 5 trials).", "candidates": [ { "score": 57.3, "reference_url": "https://huggingface.co/deepseek-ai/DeepSeek-V4-Pro", "source_type": "third_party", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "agentic (bash + edit)", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "notes": "DeepSeek V4-Pro model card (third-party): claude-opus-4.6 (per their own canonical config) = 57.3." }, { "score": 57.3, "reference_url": "https://z.ai/blog/glm-5.1", "source_type": "official_blog", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "agentic (bash + edit)", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "notes": "GLM-5.1 blog: claude-opus-4.6=57.3." }, { "score": 53.4, "reference_url": "https://microsoft.ai/wp-content/uploads/2026/06/main_20260602_2.pdf", "source_type": "official_paper", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "none (per-bench override)", "sampling": "pass@1 (avg 5 trials)", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default", "context": "up to 1M", "notes": "Per Mythos System Card Table 6.3.A: standard config = adaptive thinking max effort, default sampling, avg 5 trials, context up to 1M." }, "notes": "Displayed exactly as 53.4. Official Microsoft-reported result. Comparator value is quoted from that provider's official model card/release; use the campaign canonical setting." } ] }, { "model_id": "claude-opus-4.6", "benchmark_id": "arc_agi_1", "score": 93.0, "reference_url": "https://arcprize.org/leaderboard", "audit_status": "verified", "rule_ids": [ "R5g_c" ], "notes": "arcprize.org leaderboard audit: Claude Opus 4.6 (120K, Max) on leaderboard", "source_type": "leaderboard", "matches_canonical": true, "reported_setting": { "mode": "thinking", "effort": "max (120K)", "tools": "none", "sampling": "pass@1 (single-pass)", "judge": "rule-based", "harness": "official (arcprize.org)", "system_type": "CoT" } }, { "model_id": "claude-opus-4.6", "benchmark_id": "codeforces_rating", "score": 2650, "reference_url": "https://www.vellum.ai/blog/claude-opus-4-6-benchmarks", "audit_status": "dropped", "rule_ids": [ "R5h" ], "notes": "Source is third-party blog (vellum.ai); R5h priority. DROPPED (R5h): random third-party blog/aggregator, no primary trail", "rule_id": "R5h-third-party-blog" }, { "model_id": "claude-opus-4.6", "benchmark_id": "mmmu_pro", "score": 73.9, "reference_url": "https://www.anthropic.com/news/claude-sonnet-4-6", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "none", "sampling": "pass@1 (avg 5 trials)", "judge": "rule-based", "harness": "official", "prompt_style": "updated grading: separate Sonnet 4 grader; no 'think step-by-step' prefix", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "Anthropic Sonnet 4.6 blog table: claude-opus-4.6 = 73.9.", "candidates": [ { "score": 73.9, "reference_url": "https://deepmind.google/models/gemini/pro/", "source_type": "official_blog", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "notes": "DeepMind /models/gemini/pro/ cross-model Performance table: MMMU-Pro no tools 73.9%" }, { "score": 73.9, "reference_url": "https://www-cdn.anthropic.com/6a5fa276ac68b9aeb0c8b6af5fa36326e0e166dd/Claude%20Opus%204.6%20System%20Card.pdf", "source_type": "official_model_card", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "none", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official Anthropic evaluation", "prompt_style": "provider official evaluation prompt", "temperature": "source/provider setting", "context": "source/provider setting", "input_modalities": "image and text", "trials": "source-reported" }, "notes": "Exact provider-official score and reported setting." } ] }, { "model_id": "claude-opus-4.6", "benchmark_id": "terminal_bench", "score": 65.4, "reference_url": "https://www-cdn.anthropic.com/8b8380204f74670be75e81c820ca8dda846ab289.pdf", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "agentic", "sampling": "pass@1 (avg 5 trials)", "judge": "rule-based", "harness": "Harbor scaffold + Terminus-2", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Mythos sys card Table 6.3.A: Opus 4.6 = 65.4%. Verified per Anthropic standard config (thinking max effort, default sampling, avg 5 trials).", "candidates": [ { "score": 65.4, "reference_url": "https://deepmind.google/models/gemini/pro/", "source_type": "official_blog", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "Terminus-2", "prompt_style": "default", "temperature": "default" }, "notes": "DeepMind /models/gemini/pro/ cross-model Performance table: Terminal-Bench 2.0 Terminus-2 65.4%" }, { "score": 58.0, "reference_url": "https://cursor.com/resources/Composer2.pdf", "source_type": "tech_report", "reported_setting": { "effort": "high", "tools": "Cursor coding agent", "sampling": "pass@1", "harness": "Claude Code leaderboard" }, "notes": "Composer 2 technical report Table 1 / Terminal-Bench 2.0 / Opus 4.6 High / value 1: For third-party public benchmarks, slash-separated values are Cursor/official-harness then self-reported. SWE prompts prepend 'please solve this github issue'; Terminal prompts add solution-format instructions." }, { "score": 65.4, "reference_url": "https://microsoft.ai/wp-content/uploads/2026/06/main_20260602_2.pdf", "source_type": "official_paper", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "none (per-bench override)", "sampling": "pass@1 (avg 5 trials)", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default", "context": "up to 1M", "notes": "Per Mythos System Card Table 6.3.A: standard config = adaptive thinking max effort, default sampling, avg 5 trials, context up to 1M." }, "notes": "Displayed exactly as 65.4. Official Microsoft-reported result. Comparator value is quoted from that provider's official model card/release; use the campaign canonical setting." }, { "score": 65.4, "reference_url": "https://www-cdn.anthropic.com/6a5fa276ac68b9aeb0c8b6af5fa36326e0e166dd/Claude%20Opus%204.6%20System%20Card.pdf", "source_type": "official_blog", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "source-reported agentic tools", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "Terminus-2", "prompt_style": "provider official evaluation prompt", "temperature": "source/provider setting", "context": "source/provider setting", "input_modalities": "text", "trials": "source-reported" }, "notes": "Exact provider-official score and reported setting." } ] }, { "model_id": "claude-opus-4.6", "benchmark_id": "simplebench", "score": 67.6, "reference_url": "https://lmcouncil.ai/benchmarks", "audit_status": "dropped", "rule_ids": [ "R5g_c" ], "notes": "lmcouncil.ai scores are JS-rendered; could not extract value. R5g(c). DROPPED (R5h): random third-party blog/aggregator, no primary trail | DROPPED (R5g-c stale, never verified, source not audited in any pass)", "rule_id": "R5h-third-party-blog" }, { "model_id": "claude-sonnet-4.6", "benchmark_id": "swe_bench_verified", "score": 79.6, "reference_url": "https://www.anthropic.com/news/claude-sonnet-4-6", "reported_setting": { "mode": "non-thinking", "effort": "max", "tools": "agentic (bash + edit)", "sampling": "pass@1 (avg 10 trials)", "judge": "rule-based", "harness": "simple bash+editor scaffold", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "Anthropic Sonnet 4.6 blog table: claude-sonnet-4.6 = 79.6.", "candidates": [ { "score": 79.6, "reference_url": "https://deepmind.google/models/gemini/pro/", "source_type": "official_blog", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "notes": "DeepMind /models/gemini/pro/ cross-model Performance table: SWE-Bench Verified single attempt 79.6%" }, { "score": 79.6, "reference_url": "https://huggingface.co/mistralai/Mistral-Medium-3.5-128B/resolve/22b2b868a15677cfa6061277ed2f653d1349a9ab/images/image4.png?download=true", "source_type": "official_model_card", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "agentic coding scaffold and SWE-bench verifier", "sampling": "source does not state trial count", "judge": "benchmark-specified", "harness": "Mistral self-reported SWE-bench Verified evaluation", "prompt_style": "default", "temperature": "default", "context": "default" }, "notes": "Displayed exactly: '79.6'. Research observation: medium-image4.png:1:3:score:reported. Chart footnotes: *All model evaluations were run with maximum reasoning settings. | **Sonnet 4.6 encountered higher rates of reasoning truncation due to external API restrictions than other models, which affected its performance. | *** Self-reported | **** Self-reported, Mistral is using context management and a discard-all strategy at 100k tokens. | τ³ scores as reported by Sierra for Claude Sonnet 4.5 and Qwen3.5. Others with user simulator: gpt-5.2 with reasoning_effort: low. 4 trials. Banking domain evaluated with terminal- or embedding-based agentic search retrieval, only highest score is reported." }, { "score": 79.6, "reference_url": "https://microsoft.ai/wp-content/uploads/2026/06/main_20260602_2.pdf", "source_type": "official_paper", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "none (per-bench override)", "sampling": "pass@1 (avg 5-15 trials)", "judge": "rule-based", "harness": "Terminus-2 (terminal-bench, thinking off); official otherwise", "prompt_style": "default", "temperature": "default", "context": "default", "notes": "Per Anthropic Sonnet 4.6 announcement: max thinking effort default; Terminal-Bench=thinking off, Terminus-2; SWE-bench=avg 10 trials; BrowseComp/HLE-with-tools have specific tool configs (web search/fetch + 50k context compaction)." }, "notes": "Displayed exactly as 79.6. Official Microsoft-reported result. Comparator value is quoted from that provider's official model card/release; use the campaign canonical setting." } ] }, { "model_id": "claude-sonnet-4.6", "benchmark_id": "osworld", "score": 72.5, "reference_url": "https://www.anthropic.com/news/claude-sonnet-4-6", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "agentic", "sampling": "pass@1 (avg 5 trials)", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "Anthropic Sonnet 4.6 blog table: claude-sonnet-4.6 = 72.5." }, { "model_id": "claude-sonnet-4.6", "benchmark_id": "arc_agi_2", "score": 58.3, "reference_url": "https://www.anthropic.com/news/claude-sonnet-4-6", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "none", "sampling": "pass@1 (avg 5 trials)", "judge": "rule-based", "harness": "official", "prompt_style": "120k thinking budget", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "Anthropic Sonnet 4.6 blog table: claude-sonnet-4.6 = 58.3.", "candidates": [ { "score": 58.3, "reference_url": "https://deepmind.google/models/gemini/pro/", "source_type": "official_blog", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "notes": "DeepMind /models/gemini/pro/ cross-model Performance table: ARC-AGI-2 ARC Prize Verified 58.3% (Thinking Max)" } ] }, { "model_id": "claude-sonnet-4.6", "benchmark_id": "gdpval_aa_elo", "score": 1633, "reference_url": "https://www.anthropic.com/news/claude-sonnet-4-6", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "none", "sampling": "pass@1 (avg 5 trials)", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "Anthropic Sonnet 4.6 blog table: claude-sonnet-4.6 = 1633.", "candidates": [ { "score": 1395.0, "reference_url": "https://www.anthropic.com/claude-sonnet-5-system-card", "source_type": "model_card", "reported_setting": { "effort": "max", "tools": "none", "sampling": "avg 5 trials unless source states otherwise", "harness": "table8" }, "notes": "Anthropic Claude Sonnet 5 System Card Table 8.1.A, page 115: Claude Sonnet 4.6; GDPval-AA v2 [summary]=1395. Source setting: effort=max; tools=none; sampling=avg 5 trials unless source states otherwise; harness=table8." } ] }, { "model_id": "claude-sonnet-4.6", "benchmark_id": "aa_intelligence_index", "score": 51, "reference_url": "https://artificialanalysis.ai/evaluations/artificial-analysis-intelligence-index", "audit_status": "needs_review", "rule_ids": [ "R5g_d" ], "notes": "AA Intelligence Index: claude-sonnet-4-6-adaptive = 51.72 (rounds to 52), but BP has 51. 1-point discrepancy vs current AA data. Possible stale or truncation-vs-round difference. Current AA value: 51.72." }, { "model_id": "claude-sonnet-4.6", "benchmark_id": "aime_2025", "score": 95.6, "reference_url": "https://anthropic.com/claude-sonnet-4-6-system-card", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "none", "sampling": "pass@1 (avg 10 trials)", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Claude Sonnet 4.6 System Card section 2.10: AIME 2025 95.6% without tools, avg over 10 trials, max effort. (Card notes potential contamination concern, see Section 2.2 of Opus 4.5 System Card.) [R5d: prior unverified value 97.0 from anthropic blog deleted.]", "candidates": [ { "score": 86.9, "reference_url": "https://huggingface.co/mistralai/Mistral-Medium-3.5-128B/resolve/22b2b868a15677cfa6061277ed2f653d1349a9ab/images/image1.png?download=true", "source_type": "official_model_card", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "none", "sampling": "samples=16 (reported as avg@16)", "judge": "benchmark-specified", "harness": "Mistral official maximum-reasoning comparison", "prompt_style": "default", "temperature": "default", "context": "default" }, "notes": "Displayed exactly: '86.9'. Research observation: medium-image1.png:1:3:score:reported. Chart footnotes: *All model evaluations were run with maximum reasoning settings. | **Sonnet 4.6 encountered higher rates of reasoning truncation due to external API restrictions than other models, which affected its performance." }, { "score": 95.6, "reference_url": "https://microsoft.ai/wp-content/uploads/2026/06/main_20260602_2.pdf", "source_type": "official_paper", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "none (per-bench override)", "sampling": "pass@1 (avg 5-15 trials)", "judge": "rule-based", "harness": "Terminus-2 (terminal-bench, thinking off); official otherwise", "prompt_style": "default", "temperature": "default", "context": "default", "notes": "Per Anthropic Sonnet 4.6 announcement: max thinking effort default; Terminal-Bench=thinking off, Terminus-2; SWE-bench=avg 10 trials; BrowseComp/HLE-with-tools have specific tool configs (web search/fetch + 50k context compaction)." }, "notes": "Displayed exactly as 95.6. Official Microsoft-reported result. Comparator value is quoted from that provider's official model card/release; use the campaign canonical setting." } ] }, { "model_id": "claude-sonnet-4.6", "benchmark_id": "browsecomp", "score": 74.7, "reference_url": "https://www.anthropic.com/news/claude-sonnet-4-6", "reported_setting": { "mode": "non-thinking", "effort": "max", "tools": "agentic (web search + fetch + programmatic tool calling, context compaction at 50k up to 10M)", "sampling": "pass@1 (avg 5 trials)", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "Anthropic Sonnet 4.6 blog table: claude-sonnet-4.6 = 74.7.", "candidates": [ { "score": 74.7, "reference_url": "https://deepmind.google/models/gemini/pro/", "source_type": "official_blog", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "web+code", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "notes": "DeepMind /models/gemini/pro/ cross-model Performance table: BrowseComp Search+Python+Browse 74.7%" }, { "score": 76.2, "reference_url": "https://www.anthropic.com/claude-sonnet-5-system-card", "source_type": "model_card", "reported_setting": { "effort": "max", "tools": "none", "sampling": "avg 5 trials unless source states otherwise", "harness": "table8" }, "notes": "Anthropic Claude Sonnet 5 System Card Table 8.1.A, page 115: Claude Sonnet 4.6; BrowseComp / single agent [summary]=76.2. Source setting: effort=max; tools=none; sampling=avg 5 trials unless source states otherwise; harness=table8." }, { "score": 70.5, "reference_url": "https://www.anthropic.com/claude-sonnet-5-system-card", "source_type": "model_card", "reported_setting": { "effort": "max", "tools": "web search, fetch, programmatic tools, code execution", "sampling": "pass@1", "harness": "browse", "token_cap": "1M" }, "notes": "Anthropic Claude Sonnet 5 System Card Figure 8.9.2.A, page 124: Claude Sonnet 4.6; BrowseComp [1M]=70.5%. Source setting: effort=max; tools=web search, fetch, programmatic tools, code execution; sampling=pass@1; harness=browse; token_cap=1M." }, { "score": 74.6, "reference_url": "https://www.anthropic.com/claude-sonnet-5-system-card", "source_type": "model_card", "reported_setting": { "effort": "max", "tools": "web search, fetch, programmatic tools, code execution", "sampling": "pass@1", "harness": "browse", "token_cap": "3M" }, "notes": "Anthropic Claude Sonnet 5 System Card Figure 8.9.2.A, page 124: Claude Sonnet 4.6; BrowseComp [3M]=74.6%. Source setting: effort=max; tools=web search, fetch, programmatic tools, code execution; sampling=pass@1; harness=browse; token_cap=3M." }, { "score": 76.2, "reference_url": "https://www.anthropic.com/claude-sonnet-5-system-card", "source_type": "model_card", "reported_setting": { "effort": "max", "tools": "web search, fetch, programmatic tools, code execution", "sampling": "pass@1", "harness": "browse", "token_cap": "10M" }, "notes": "Anthropic Claude Sonnet 5 System Card Figure 8.9.2.A, page 124: Claude Sonnet 4.6; BrowseComp [10M]=76.2%. Source setting: effort=max; tools=web search, fetch, programmatic tools, code execution; sampling=pass@1; harness=browse; token_cap=10M." }, { "score": 74.7, "reference_url": "https://huggingface.co/mistralai/Mistral-Medium-3.5-128B/resolve/22b2b868a15677cfa6061277ed2f653d1349a9ab/images/image4.png?download=true", "source_type": "official_model_card", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "web search/browser environment", "sampling": "source does not state trial count", "judge": "benchmark-specified", "harness": "Mistral official BrowseComp evaluation", "prompt_style": "default", "temperature": "default", "context": "default" }, "notes": "Displayed exactly: '74.7'. Research observation: medium-image4.png:6:3:score:reported. Chart footnotes: *All model evaluations were run with maximum reasoning settings. | **Sonnet 4.6 encountered higher rates of reasoning truncation due to external API restrictions than other models, which affected its performance. | *** Self-reported | **** Self-reported, Mistral is using context management and a discard-all strategy at 100k tokens. | τ³ scores as reported by Sierra for Claude Sonnet 4.5 and Qwen3.5. Others with user simulator: gpt-5.2 with reasoning_effort: low. 4 trials. Banking domain evaluated with terminal- or embedding-based agentic search retrieval, only highest score is reported." } ] }, { "model_id": "claude-sonnet-4.6", "benchmark_id": "frontiermath", "score": 32.4, "reference_url": "https://epoch.ai/benchmarks/frontiermath", "audit_status": "verified", "rule_ids": [ "R5g_c" ], "notes": "epoch.ai FrontierMath-2025-02-28-Private: claude-sonnet-4-6_16K = 32.4% (16k thinking). Canonical requests max effort but epoch.ai only has 16k thinking variant for Sonnet 4.6; no max entry exists. BP had 35.0.", "source_type": "leaderboard", "matches_canonical": false, "reported_setting": { "mode": "thinking", "effort": "n/a", "tools": "none", "context": "16k thinking" } }, { "model_id": "claude-sonnet-4.6", "benchmark_id": "gpqa_diamond", "score": 89.9, "reference_url": "https://www.anthropic.com/news/claude-sonnet-4-6", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "none", "sampling": "pass@1 (avg 5 trials)", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "Anthropic Sonnet 4.6 blog table: claude-sonnet-4.6 = 89.9.", "candidates": [ { "score": 89.9, "reference_url": "https://deepmind.google/models/gemini/pro/", "source_type": "official_blog", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "notes": "DeepMind /models/gemini/pro/ cross-model Performance table: GPQA Diamond no tools 89.9% (Thinking Max)" }, { "score": 89.9, "reference_url": "https://microsoft.ai/wp-content/uploads/2026/06/main_20260602_2.pdf", "source_type": "official_paper", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "none (per-bench override)", "sampling": "pass@1 (avg 5-15 trials)", "judge": "rule-based", "harness": "Terminus-2 (terminal-bench, thinking off); official otherwise", "prompt_style": "default", "temperature": "default", "context": "default", "notes": "Per Anthropic Sonnet 4.6 announcement: max thinking effort default; Terminal-Bench=thinking off, Terminus-2; SWE-bench=avg 10 trials; BrowseComp/HLE-with-tools have specific tool configs (web search/fetch + 50k context compaction)." }, "notes": "Displayed exactly as 89.9. Official Microsoft-reported result. Comparator value is quoted from that provider's official model card/release; use the campaign canonical setting." } ] }, { "model_id": "claude-sonnet-4.6", "benchmark_id": "hle", "score": 33.2, "reference_url": "https://www.anthropic.com/news/claude-sonnet-4-6", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "none (no-tools column)", "sampling": "pass@1 (avg 5 trials)", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "Anthropic Sonnet 4.6 blog table: claude-sonnet-4.6 = 33.2.", "candidates": [ { "score": 33.2, "reference_url": "https://deepmind.google/models/gemini/pro/", "source_type": "official_blog", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "notes": "DeepMind /models/gemini/pro/ cross-model Performance table: HLE no tools 33.2% (Thinking Max)" }, { "score": 49.0, "reference_url": "https://deepmind.google/models/gemini/pro/", "source_type": "official_blog", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "web+code", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "notes": "DeepMind /models/gemini/pro/ cross-model Performance table: HLE Search+Code 49.0% (Thinking Max)" }, { "score": 34.6, "reference_url": "https://www.anthropic.com/claude-sonnet-5-system-card", "source_type": "model_card", "reported_setting": { "effort": "max", "tools": "none", "sampling": "avg 5 trials unless source states otherwise", "harness": "table8" }, "notes": "Anthropic Claude Sonnet 5 System Card Table 8.1.A, page 115: Claude Sonnet 4.6; Humanity’s Last Exam / no tools [summary]=34.6. Source setting: effort=max; tools=none; sampling=avg 5 trials unless source states otherwise; harness=table8." } ] }, { "model_id": "claude-sonnet-4.6", "benchmark_id": "humaneval", "score": 93.0, "reference_url": "https://www.anthropic.com/news/claude-sonnet-4-6", "audit_status": "needs_review", "rule_ids": [ "R5g_c" ], "notes": "Anthropic blog benchmark table is a CDN PNG image; cannot extract text value. Web search shows conflicting values (90.8% vs JSON 93.0%). R5g(c)." }, { "model_id": "claude-sonnet-4.6", "benchmark_id": "ifeval", "score": 92.0, "reference_url": "https://www.anthropic.com/news/claude-sonnet-4-6", "audit_status": "needs_review", "rule_ids": [ "R5g_c" ], "notes": "Anthropic blog benchmark table is a CDN PNG image; cannot extract text value. Web search shows 89.9% vs JSON 92.0%. R5g(c)." }, { "model_id": "claude-sonnet-4.6", "benchmark_id": "livecodebench", "score": 74.0, "reference_url": "https://www.anthropic.com/news/claude-sonnet-4-6", "audit_status": "needs_review", "rule_ids": [ "R5g_c" ], "notes": "Anthropic blog benchmark table is a CDN PNG image; cannot extract text value. Web search shows 72.4% vs JSON 74.0%. R5g(c)." }, { "model_id": "claude-sonnet-4.6", "benchmark_id": "mmlu", "score": 90.0, "reference_url": "https://www.anthropic.com/news/claude-sonnet-4-6", "audit_status": "needs_review", "rule_ids": [ "R5g_c" ], "notes": "Anthropic blog benchmark table is a CDN PNG image; cannot extract text value. Web search shows 89.7% vs JSON 90.0%. R5g(c)." }, { "model_id": "claude-sonnet-4.6", "benchmark_id": "mmlu_pro", "score": 87.0, "reference_url": "https://microsoft.ai/wp-content/uploads/2026/06/main_20260602_2.pdf", "reported_setting": { "mode": "thinking", "effort": "maximum reasoning effort", "tools": "none; benchmark-provided tools where required", "sampling": "source does not state trial count", "judge": "benchmark-specific Microsoft implementation", "harness": "Microsoft independent evaluation suite", "prompt_style": "benchmark-specific", "temperature": "source does not state", "context": "maximum sequence length", "notes": "Sonnet 4.6 result generated by Microsoft's own evaluation." }, "matches_canonical": false, "source_type": "official_paper", "audit_status": "verified", "notes": "Displayed exactly as 87. Official Microsoft-reported result." }, { "model_id": "claude-sonnet-4.6", "benchmark_id": "mmmu", "score": 74.2, "reference_url": "https://automatio.ai/models/claude-sonnet-4-6", "audit_status": "dropped", "rule_ids": [ "R5h" ], "notes": "Source is third-party blog (automatio.ai); R5h priority. DROPPED (R5h): random third-party blog/aggregator, no primary trail", "rule_id": "R5h-third-party-blog" }, { "model_id": "claude-sonnet-4.6", "benchmark_id": "simpleqa", "score": 68.0, "reference_url": "https://www.anthropic.com/news/claude-sonnet-4-6", "audit_status": "needs_review", "rule_ids": [ "R5g_c" ], "notes": "Anthropic blog benchmark table is a CDN PNG image; cannot extract text value. R5g(c)." }, { "model_id": "claude-sonnet-4.6", "benchmark_id": "tau_bench_retail", "score": 89.0, "reference_url": "https://www.anthropic.com/news/claude-sonnet-4-6", "audit_status": "dropped", "notes": " | DROPPED: BP cell value 89.0 from anthropic.com/news/claude-sonnet-4-6 is mislabeled — Anthropic Sonnet 4.6 reports tau2-bench (v2) not tau-bench (v1). Use tau2_bench_retail (already verified at 91.7) instead." }, { "model_id": "claude-sonnet-4.6", "benchmark_id": "tau_bench_telecom", "score": 97.0, "reference_url": "https://www.anthropic.com/news/claude-sonnet-4-6", "audit_status": "dropped", "notes": " | DROPPED: BP cell value 97.0 from anthropic.com/news/claude-sonnet-4-6 is mislabeled — Anthropic Sonnet 4.6 reports tau2-bench (v2) not tau-bench (v1). Use tau2_bench_telecom (already verified at 97.9) instead." }, { "model_id": "claude-sonnet-4.6", "benchmark_id": "arc_agi_1", "score": 86.5, "reference_url": "https://anthropic.com/claude-sonnet-4-6-system-card", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default", "notes": "120k thinking tokens, ARC Prize Foundation private set" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Claude Sonnet 4.6 System Card section 2.7: ARC-AGI-1 86.5% reported by ARC Prize Foundation with 120k thinking tokens and high effort on private set. matches_canonical=false (high effort vs canonical max). [R5d: prior unverified value 86.0 from arcprize.org leaderboard deleted.]" }, { "model_id": "claude-sonnet-4.6", "benchmark_id": "codeforces_rating", "score": 2010, "reference_url": "https://www.nxcode.io/resources/news/claude-sonnet-4-6-complete-guide-benchmarks-pricing-2026", "audit_status": "needs_review", "rule_ids": [ "R5h" ], "notes": "Source is third-party blog (nxcode.io); R5h priority." }, { "model_id": "claude-sonnet-4.6", "benchmark_id": "math_500", "score": 96.5, "reference_url": "https://www.nxcode.io/resources/news/claude-sonnet-4-6-complete-guide-benchmarks-pricing-2026", "audit_status": "needs_review", "rule_ids": [ "R5h" ], "notes": "Source is third-party blog (nxcode.io); R5h priority." }, { "model_id": "claude-sonnet-4.6", "benchmark_id": "mmmu_pro", "score": 74.5, "reference_url": "https://www.anthropic.com/news/claude-sonnet-4-6", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "none", "sampling": "pass@1 (avg 5 trials)", "judge": "rule-based", "harness": "official", "prompt_style": "updated grading: separate Sonnet 4 grader; no 'think step-by-step' prefix", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "Anthropic Sonnet 4.6 blog table: claude-sonnet-4.6 = 74.5.", "candidates": [ { "score": 74.5, "reference_url": "https://deepmind.google/models/gemini/pro/", "source_type": "official_blog", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "notes": "DeepMind /models/gemini/pro/ cross-model Performance table: MMMU-Pro no tools 74.5%" } ] }, { "model_id": "claude-sonnet-4.6", "benchmark_id": "mrcr_v2", "score": 84.9, "reference_url": "https://deepmind.google/models/gemini/pro/", "candidates": [], "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default", "context": "128k" }, "source_type": "official_blog", "audit_status": "verified", "matches_canonical": true, "notes": "DeepMind /models/gemini/pro/ Performance table (PROMOTED from candidate; prior unverified value 82 from https://awesomeagents.ai/leaderboards/long-context-benchmarks-leaderboard/ deleted per audit rule). DeepMind /models/gemini/pro/ cross-model Performance table: MRCR v2 8-needle 128k average 84.9%" }, { "model_id": "claude-sonnet-4.6", "benchmark_id": "swe_bench_pro", "score": 58.1, "reference_url": "https://www.anthropic.com/claude-sonnet-5-system-card", "reported_setting": { "effort": "max", "tools": "none", "sampling": "avg 5 trials unless source states otherwise", "harness": "table8" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Anthropic Claude Sonnet 5 System Card Table 8.1.A, page 115: Claude Sonnet 4.6; SWE-bench Pro [summary]=58.1. Source setting: effort=max; tools=none; sampling=avg 5 trials unless source states otherwise; harness=table8." }, { "model_id": "claude-sonnet-4.6", "benchmark_id": "terminal_bench", "score": 59.1, "reference_url": "https://www.anthropic.com/news/claude-sonnet-4-6", "reported_setting": { "mode": "non-thinking", "effort": "max", "tools": "agentic", "sampling": "pass@1 (avg 5-15 trials)", "judge": "rule-based", "harness": "Terminus-2 (1× guaranteed/3× ceiling)", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "Anthropic Sonnet 4.6 blog table: claude-sonnet-4.6 = 59.1.", "candidates": [ { "score": 59.1, "reference_url": "https://deepmind.google/models/gemini/pro/", "source_type": "official_blog", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "Terminus-2", "prompt_style": "default", "temperature": "default" }, "notes": "DeepMind /models/gemini/pro/ cross-model Performance table: Terminal-Bench 2.0 Terminus-2 59.1%" }, { "score": 59.1, "reference_url": "https://microsoft.ai/wp-content/uploads/2026/06/main_20260602_2.pdf", "source_type": "official_paper", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "none (per-bench override)", "sampling": "pass@1 (avg 5-15 trials)", "judge": "rule-based", "harness": "Terminus-2 (terminal-bench, thinking off); official otherwise", "prompt_style": "default", "temperature": "default", "context": "default", "notes": "Per Anthropic Sonnet 4.6 announcement: max thinking effort default; Terminal-Bench=thinking off, Terminus-2; SWE-bench=avg 10 trials; BrowseComp/HLE-with-tools have specific tool configs (web search/fetch + 50k context compaction)." }, "notes": "Displayed exactly as 59.1. Official Microsoft-reported result. Comparator value is quoted from that provider's official model card/release; use the campaign canonical setting." } ] }, { "model_id": "claude-mythos", "benchmark_id": "swe_bench_verified", "score": 93.9, "reference_url": "https://www-cdn.anthropic.com/8b8380204f74670be75e81c820ca8dda846ab289.pdf", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "agentic (bash/edit + thinking)", "sampling": "pass@1 (avg 5 trials)", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Mythos sys card §6.4: 5 trials, standard config." }, { "model_id": "claude-mythos", "benchmark_id": "swe_bench_pro", "score": 77.8, "reference_url": "https://www-cdn.anthropic.com/8b8380204f74670be75e81c820ca8dda846ab289.pdf", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "agentic", "sampling": "pass@1 (avg 5 trials)", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Mythos sys card §6.4: 5 trials." }, { "model_id": "claude-mythos", "benchmark_id": "terminal_bench", "score": 82.0, "reference_url": "https://www-cdn.anthropic.com/8b8380204f74670be75e81c820ca8dda846ab289.pdf", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "agentic", "sampling": "pass@1 (avg 5 trials)", "judge": "rule-based", "harness": "Harbor scaffold + Terminus-2", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Mythos sys card §6.5." }, { "model_id": "claude-mythos", "benchmark_id": "gpqa_diamond", "score": 94.5, "reference_url": "https://www-cdn.anthropic.com/8b8380204f74670be75e81c820ca8dda846ab289.pdf", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "none", "sampling": "pass@1 (avg 5 trials)", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Mythos sys card: 94.55%.", "candidates": [ { "score": 94.6, "reference_url": "https://openai.com/index/gpt-5-6/", "source_type": "official_blog", "reported_setting": { "mode": "thinking", "effort": "source does not state", "tools": "none", "sampling": "source does not state", "judge": "benchmark-specified", "harness": "official release evaluation", "prompt_style": "source does not state", "temperature": "source does not state", "context": "source does not state" }, "notes": "OpenAI GPT-5.6 release table (table 7): Claude Mythos Preview; GPQA Diamond=94.6." } ] }, { "model_id": "claude-mythos", "benchmark_id": "hle", "score": 56.8, "reference_url": "https://www-cdn.anthropic.com/8b8380204f74670be75e81c820ca8dda846ab289.pdf", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "none", "sampling": "pass@1 (avg 5 trials)", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Mythos sys card Table 6.3.A: no-tools column. With-tools=64.7 not used." }, { "model_id": "claude-mythos", "benchmark_id": "osworld", "score": 79.6, "reference_url": "https://www-cdn.anthropic.com/8b8380204f74670be75e81c820ca8dda846ab289.pdf", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "agentic", "sampling": "pass@1 (avg 5 trials)", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Mythos sys card Table 6.3.A." }, { "model_id": "claude-mythos", "benchmark_id": "usamo_2025", "score": 97.6, "reference_url": "https://www-cdn.anthropic.com/8b8380204f74670be75e81c820ca8dda846ab289.pdf", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "none", "sampling": "pass@1 (avg 10 trials)", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Mythos sys card §6.8: max effort, no tools, avg 10 trials per problem. MathArena grading methodology (3 LLM judges incl. Gemini 3.1 Pro neutral rewriter)." }, { "model_id": "claude-opus-4.7", "benchmark_id": "swe_bench_pro", "score": 64.3, "reference_url": "https://www.anthropic.com/news/claude-opus-4-7", "reported_setting": { "mode": "non-thinking", "effort": "default", "tools": "agentic (bash + edit)", "sampling": "pass@1 (avg 5 trials)", "judge": "rule-based", "harness": "simple bash+editor scaffold", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "Anthropic Opus 4.7 blog table: claude-opus-4.7 = 64.3%.", "candidates": [ { "score": 64.3, "reference_url": "https://openai.com/index/introducing-gpt-5-5/", "source_type": "third_party", "reported_setting": { "note": "see primary" }, "notes": "OpenAI GPT-5.5 blog reports 64.3 (matches primary)." }, { "score": 64.3, "reference_url": "https://x.ai/news/grok-4-5", "source_type": "official_blog", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "agentic repository shell/editor", "sampling": "pass@1", "judge": "verification reward and repository tests", "harness": "fixed SWE-bench Pro agent scaffold", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown" }, "notes": "release45 SWE-Bench Pro Resolve rate (%); source effort=max; benchmark protocol matches canonical=True. Unknown source fields remain unknown." }, { "score": 64.3, "reference_url": "https://seed.bytedance.com/en/blog/seed2-1-officially-released-advancing-ai-productivity", "source_type": "official_blog", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "Trae or benchmark-specific coding-agent environment", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official", "prompt_style": "official default", "temperature": "unknown" }, "notes": "Release blog EN: Coding chart (https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/user-upload/4xfa4mqq4wlbt.png)." } ] }, { "model_id": "claude-opus-4.7", "benchmark_id": "swe_bench_verified", "score": 87.6, "reference_url": "https://www.anthropic.com/news/claude-opus-4-7", "reported_setting": { "mode": "non-thinking", "effort": "default", "tools": "agentic (bash + edit)", "sampling": "pass@1 (avg 5 trials)", "judge": "rule-based", "harness": "simple bash+editor scaffold", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "Anthropic Opus 4.7 blog table: claude-opus-4.7 = 87.6%." }, { "model_id": "claude-opus-4.7", "benchmark_id": "terminal_bench", "score": 69.4, "reference_url": "https://www.anthropic.com/news/claude-opus-4-7", "reported_setting": { "mode": "non-thinking", "effort": "default", "tools": "agentic", "sampling": "pass@1 (avg 5 trials)", "judge": "rule-based", "harness": "Terminus-2 (1× guaranteed/3× ceiling, avg 5)", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "Anthropic Opus 4.7 blog table: claude-opus-4.7 = 69.4%.", "candidates": [ { "score": 69.4, "reference_url": "https://openai.com/index/introducing-gpt-5-5/", "source_type": "third_party", "reported_setting": { "note": "see primary" }, "notes": "OpenAI GPT-5.5 blog reports 69.4 (matches primary)." } ] }, { "model_id": "claude-opus-4.7", "benchmark_id": "hle", "score": 46.9, "reference_url": "https://www.anthropic.com/news/claude-opus-4-7", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1 (avg 5 trials)", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "Anthropic Opus 4.7 blog table: claude-opus-4.7 = 46.9%.", "candidates": [ { "score": 46.9, "reference_url": "https://openai.com/index/introducing-gpt-5-5/", "source_type": "third_party", "reported_setting": { "note": "see primary" }, "notes": "OpenAI GPT-5.5 blog reports 46.9 (matches primary)." } ] }, { "model_id": "claude-opus-4.7", "benchmark_id": "browsecomp", "score": 79.3, "reference_url": "https://www.anthropic.com/news/claude-opus-4-7", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "agentic", "sampling": "pass@1 (avg 5 trials)", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "Anthropic Opus 4.7 blog table: claude-opus-4.7 = 79.3%.", "candidates": [ { "score": 79.3, "reference_url": "https://openai.com/index/introducing-gpt-5-5/", "source_type": "third_party", "reported_setting": { "note": "see primary" }, "notes": "OpenAI GPT-5.5 blog reports 79.3 (matches primary)." }, { "score": 79.3, "reference_url": "https://seed.bytedance.com/en/blog/seed2-1-officially-released-advancing-ai-productivity", "source_type": "official_blog", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "search", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official", "prompt_style": "official default", "temperature": "unknown" }, "notes": "Release blog EN: Language/search chart (https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/user-upload/4xfa4mqq510uc.png)." } ] }, { "model_id": "claude-opus-4.7", "benchmark_id": "gpqa_diamond", "score": 94.2, "reference_url": "https://www.anthropic.com/news/claude-opus-4-7", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1 (avg 5 trials)", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "Anthropic Opus 4.7 blog table: claude-opus-4.7 = 94.2%.", "candidates": [ { "score": 94.2, "reference_url": "https://openai.com/index/introducing-gpt-5-5/", "source_type": "third_party", "reported_setting": { "note": "see primary" }, "notes": "OpenAI GPT-5.5 blog reports 94.2 (matches primary)." } ] }, { "model_id": "claude-opus-4.7", "benchmark_id": "arc_agi_1", "score": 92.0, "reference_url": "https://arcprize.org/leaderboard", "audit_status": "verified", "source_type": "leaderboard", "matches_canonical": false, "reported_setting": { "mode": "thinking", "effort": "max", "tools": "none", "sampling": "pass@1 (single-pass)", "judge": "rule-based", "harness": "official (arcprize.org)", "system_type": "CoT" }, "notes": "arcprize.org leaderboard audit: Claude 4.7 (Max) on leaderboard; canonical is default effort" }, { "model_id": "claude-opus-4.7", "benchmark_id": "arc_agi_2", "score": 75.8, "reference_url": "https://arcprize.org/leaderboard", "audit_status": "verified", "source_type": "leaderboard", "matches_canonical": false, "reported_setting": { "mode": "thinking", "effort": "max", "tools": "none", "sampling": "pass@1 (single-pass)", "judge": "rule-based", "harness": "official (arcprize.org)", "system_type": "CoT" }, "notes": "arcprize.org leaderboard audit: Claude 4.7 (Max) on leaderboard; canonical is default effort", "candidates": [ { "score": 75.83, "reference_url": "https://arcprize.org/leaderboard", "source_type": "leaderboard", "reported_setting": { "effort": "max" }, "notes": "Figure 8.14 narrative / ARC-AGI-2 / Claude Opus 4.7: Verified semi-private ARC Prize score." }, { "score": 75.8, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2.1/Seed2_1_Model_Card.pdf", "source_type": "tech_report", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "none", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official", "prompt_style": "official default", "temperature": "unknown" }, "notes": "Model card Table 11 (p53)." } ] }, { "model_id": "claude-opus-4.7", "benchmark_id": "frontiermath", "score": 43.8, "reference_url": "https://openai.com/index/introducing-gpt-5-5/", "audit_status": "verified", "matches_canonical": false, "notes": "GPT-5.5 blog table: FrontierMath Tier 1–3=43.8%; third-party self-test by OpenAI", "source_type": "third_party" }, { "model_id": "gemini-2.0-flash", "benchmark_id": "mmlu_pro", "score": 77.6, "reference_url": "https://storage.googleapis.com/deepmind-media/Model-Cards/Gemini-2-0-Flash-Model-Card.pdf", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Gemini 2.0 Flash Model Card (April 15 2025): MMLU-Pro 77.6% [Promoted to verified per R5d, prior unverified value 76.4 from https://www.helicone.ai/blog/gemini-2.0-flash deleted.]", "candidates": [] }, { "model_id": "gemini-2.0-flash", "benchmark_id": "mmmu", "score": 71.7, "reference_url": "https://storage.googleapis.com/deepmind-media/Model-Cards/Gemini-2-0-Flash-Model-Card.pdf", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Gemini 2.0 Flash Model Card (April 15 2025): MMMU 71.7% [Promoted to verified per R5d, prior unverified value 70.7 from https://blog.google/technology/google-deepmind/gemini-model-updates-february-2025/ deleted.]", "candidates": [ { "score": 75.4, "reference_url": "https://llm-stats.com/benchmarks/mmmu", "source_type": "third_party_aggregator", "reported_setting": { "effort": "thinking" }, "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=Gemini 2.0 Flash Thinking, slug=gemini-2.0-flash-thinking, provider=Google" } ] }, { "model_id": "gemini-2.0-flash", "benchmark_id": "gpqa_diamond", "score": 60.1, "reference_url": "https://storage.googleapis.com/deepmind-media/Model-Cards/Gemini-2-0-Flash-Model-Card.pdf", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Gemini 2.0 Flash Model Card (April 15 2025): GPQA Diamond 60.1% [Promoted to verified per R5d, prior unverified value 74.2 from https://www.marktechpost.com/2025/01/21/google-ai-releases-gemini-2-0-flash-thinking-model/ deleted.]", "candidates": [ { "score": 74.2, "reference_url": "https://llm-stats.com/benchmarks/gpqa", "source_type": "third_party_aggregator", "reported_setting": { "effort": "thinking" }, "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=Gemini 2.0 Flash Thinking, slug=gemini-2.0-flash-thinking, provider=Google" } ] }, { "model_id": "gemini-2.0-flash", "benchmark_id": "aime_2024", "score": 73.3, "reference_url": "https://www.marktechpost.com/2025/01/21/google-ai-releases-gemini-2-0-flash-thinking-model/", "audit_status": "dropped", "rule_id": "R5h-third-party-blog", "notes": "DROPPED (R5h): random third-party blog/aggregator, no primary trail", "candidates": [ { "score": 73.3, "reference_url": "https://llm-stats.com/benchmarks/aime-2024", "source_type": "third_party_aggregator", "reported_setting": { "effort": "thinking" }, "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=Gemini 2.0 Flash Thinking, slug=gemini-2.0-flash-thinking, provider=Google" } ] }, { "model_id": "gemini-2.0-flash", "benchmark_id": "arc_agi_2", "score": 1.3, "reference_url": "https://arcprize.org/leaderboard", "audit_status": "verified", "source_type": "leaderboard", "matches_canonical": true, "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1 (single-pass)", "judge": "rule-based", "harness": "official (arcprize.org)", "system_type": "Base LLM" }, "notes": "arcprize.org leaderboard audit: Gemini 2.0 Flash Base LLM on leaderboard" }, { "model_id": "gemini-2.0-flash", "benchmark_id": "arena_hard", "score": 50.0, "reference_url": "https://github.com/lmarena/arena-hard-auto", "audit_status": "verified", "reported_setting": "Arena-Hard-v2.0 Creative Writing, Ensemble GPT-4.1+Gemini-2.5-Pro judge", "source_type": "leaderboard", "matches_canonical": true }, { "model_id": "gemini-2.0-flash", "benchmark_id": "ifeval", "score": 80.0, "reference_url": "https://www.helicone.ai/blog/gemini-2.0-flash", "audit_status": "dropped", "rule_id": "R5h-third-party-blog", "notes": "DROPPED (R5h): random third-party blog/aggregator, no primary trail" }, { "model_id": "gemini-2.0-flash", "benchmark_id": "mmlu", "score": 76.4, "reference_url": "https://www.marktechpost.com/2025/01/21/google-ai-releases-gemini-2-0-flash-thinking-model-gemini-2-0-flash-thinking-exp-01-21-scoring-73-3-on-aime-math-and-74-2-on-gpqa-diamond-science-benchmarks/", "audit_status": "dropped", "notes": " | DROPPED: BP cell value 76.4 from marktechpost.com matches mmlu_pro=76.4 from same source — likely source mislabeled MMLU-Pro as MMLU. Standard MMLU 5-shot 14k not reported in Gemini 2.0 Flash model card. Use mmlu_pro cell instead." }, { "model_id": "gemini-2.0-flash", "benchmark_id": "simpleqa", "score": 29.9, "reference_url": "https://storage.googleapis.com/deepmind-media/Model-Cards/Gemini-2-0-Flash-Model-Card.pdf", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Gemini 2.0 Flash Model Card (April 15 2025): SimpleQA (no search) 29.9% [Promoted to verified per R5d, prior unverified value 27.0 from https://www.helicone.ai/blog/gemini-2.0-flash deleted.]", "candidates": [] }, { "model_id": "gemini-2.0-flash", "benchmark_id": "bigcodebench", "score": 45.9, "reference_url": "https://bigcode-bench.github.io/", "audit_status": "verified", "reported_setting": "BigCodeBench Complete Instruct pass@1", "source_type": "leaderboard", "matches_canonical": true, "notes": "Source entry: Gemini-2.0-Flash-Exp (2025-02-05)" }, { "model_id": "gemini-2.0-flash", "benchmark_id": "humaneval", "score": 82.6, "reference_url": "https://artificialanalysis.ai/models/gemini-2-0-flash", "audit_status": "needs_review", "notes": "Reference URL is AA model page (gemini-2-0-flash). Current AA model page shows humaneval=0.904→90.4%, but BP has 82.6%. Major mismatch (7.8pp). Likely stale data; AA may have updated score. Current AA value: 90.4%." }, { "model_id": "gemini-2.0-flash", "benchmark_id": "livecodebench", "score": 34.5, "reference_url": "https://storage.googleapis.com/deepmind-media/Model-Cards/Gemini-2-0-Flash-Model-Card.pdf", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Gemini 2.0 Flash Model Card (April 15 2025): LiveCodeBench v5 (UI 10/01/2024-02/01/2025) 34.5% [Promoted to verified per R5d, prior unverified value 45.2 from https://artificialanalysis.ai/models/gemini-2-0-flash deleted.]", "candidates": [] }, { "model_id": "gemini-2.0-flash", "benchmark_id": "math_500", "score": 83.9, "reference_url": "https://artificialanalysis.ai/models/gemini-2-0-flash", "audit_status": "needs_review", "notes": "Reference URL is AA model page (gemini-2-0-flash). Current AA model page shows math_500=0.93→93.0%, but BP has 83.9%. Major mismatch (9.1pp). Likely stale data; AA may have updated score. Current AA value: 93.0%." }, { "model_id": "gemini-2.0-flash", "benchmark_id": "swe_bench_verified", "score": 42, "reference_url": "https://artificialanalysis.ai/models/gemini-2-0-flash", "audit_status": "needs_review", "notes": "Reference URL is AA model page (gemini-2-0-flash). AA does not publish swe_bench_verified scores; field is not present in AA model data. Cannot verify 42 from AA source. Original source may be SWE-bench.com." }, { "model_id": "gemini-2.5-pro", "benchmark_id": "aime_2025", "score": 88.0, "reference_url": "https://deepmind.google/models/gemini/flash/", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "DeepMind /models/gemini/flash/ Performance table: AIME 2025 no tools 88.0% [Promoted to verified, prior unverified value deleted per R5d.]", "candidates": [ { "score": 86.7, "reference_url": "https://arxiv.org/abs/2504.13914", "source_type": "tech_report", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "notes": "Seed-Thinking-v1.5 paper Table 2: gemini-2.5-pro=86.7." }, { "score": 86.7, "reference_url": "https://arxiv.org/abs/2504.21318", "source_type": "tech_report", "reported_setting": { "mode": "unknown", "effort": "unknown", "tools": "unknown", "sampling": "unknown", "judge": "rule-based", "harness": "Microsoft Phi internal eval", "prompt_style": "unknown", "temperature": "unknown" }, "notes": "Phi-4-reasoning paper Table 1 (gemini-2.5-pro column): aime_2025=86.7 (alt measurement, mc=false)" } ] }, { "model_id": "gemini-2.5-pro", "benchmark_id": "aime_2024", "score": 92.0, "reference_url": "https://arxiv.org/abs/2504.13914", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Seed-Thinking-v1.5 paper Table 2: gemini-2.5-pro=92.0." }, { "model_id": "gemini-2.5-pro", "benchmark_id": "gpqa_diamond", "score": 86.4, "reference_url": "https://deepmind.google/models/gemini/flash/", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "DeepMind /models/gemini/flash/ Performance table: GPQA Diamond 86.4% [Promoted to verified, prior unverified value deleted per R5d.]", "candidates": [ { "score": 84, "reference_url": "https://huggingface.co/MiniMaxAI/MiniMax-M2", "source_type": "model_card", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "notes": "MiniMax M2 model card: gemini-2.5-pro=84." }, { "score": 84.0, "reference_url": "https://arxiv.org/abs/2504.13914", "source_type": "tech_report", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "notes": "Seed-Thinking-v1.5 paper Table 2: gemini-2.5-pro=84.0." }, { "score": 84.0, "reference_url": "https://arxiv.org/abs/2504.21318", "source_type": "tech_report", "reported_setting": { "mode": "unknown", "effort": "unknown", "tools": "unknown", "sampling": "unknown", "judge": "rule-based", "harness": "Microsoft Phi internal eval", "prompt_style": "unknown", "temperature": "unknown" }, "notes": "Phi-4-reasoning paper Table 1 (gemini-2.5-pro column): gpqa_diamond=84.0 (alt measurement, mc=false)" } ] }, { "model_id": "gemini-2.5-pro", "benchmark_id": "hle", "score": 21.6, "reference_url": "https://deepmind.google/models/gemini/flash/", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "DeepMind /models/gemini/flash/ Performance table: HLE no tools 21.6% (Gemini 2.5 Pro Thinking) [Promoted to verified, prior unverified value deleted per R5d.]" }, { "model_id": "gemini-2.5-pro", "benchmark_id": "swe_bench_verified", "score": 59.6, "reference_url": "https://deepmind.google/models/gemini/flash/", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "DeepMind /models/gemini/flash/ Performance table: SWE-bench Verified 59.6% [Promoted to verified, prior unverified value deleted per R5d.]", "candidates": [ { "score": 67.2, "reference_url": "https://storage.googleapis.com/deepmind-media/Model-Cards/Gemini-2-5-Pro-Model-Card.pdf", "source_type": "model_card", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "multiple attempts (parallel test-time compute)", "judge": "rule-based", "harness": "AI Studio API", "prompt_style": "default", "temperature": "default" }, "notes": "Gemini 2.5 Pro Model Card: multiple attempts 67.2% (parallel test-time compute, drawing multiple trajectories + re-scoring with model judgement). Different setting from canonical single-attempt primary 59.6%." }, { "score": 63.8, "reference_url": "https://huggingface.co/MiniMaxAI/MiniMax-M2", "source_type": "model_card", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "agentic (bash + edit)", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "notes": "MiniMax M2 model card: gemini-2.5-pro=63.8." }, { "score": 63.8, "reference_url": "https://arxiv.org/abs/2504.13914", "source_type": "tech_report", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "agentic (bash + edit)", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "notes": "Seed-Thinking-v1.5 paper Table 2: gemini-2.5-pro=63.8." } ] }, { "model_id": "gemini-2.5-pro", "benchmark_id": "math_500", "score": 97.3, "reference_url": "https://www.helicone.ai/blog/gemini-2.5-full-developer-guide", "audit_status": "dropped", "rule_id": "R5h-third-party-blog", "notes": "DROPPED (R5h): random third-party blog/aggregator, no primary trail" }, { "model_id": "gemini-2.5-pro", "benchmark_id": "livecodebench", "score": 69.0, "reference_url": "https://storage.googleapis.com/deepmind-media/Model-Cards/Gemini-2-5-Pro-Model-Card.pdf", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1 (single attempt)", "judge": "rule-based", "harness": "AI Studio API", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Gemini 2.5 Pro Model Card (June 2025 GA): LiveCodeBench (UI 1/1/2025-5/1/2025) single attempt 69.0% [Promoted to verified per R5d, prior unverified value 70.4 from https://www.helicone.ai/blog/gemini-2.5-full-developer-guide deleted.]", "candidates": [ { "score": 80, "reference_url": "https://huggingface.co/MiniMaxAI/MiniMax-M2", "source_type": "model_card", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "notes": "MiniMax M2 model card: gemini-2.5-pro=80." }, { "score": 69.1, "reference_url": "https://arxiv.org/abs/2504.21318", "source_type": "tech_report", "reported_setting": { "mode": "unknown", "effort": "unknown", "tools": "unknown", "sampling": "unknown", "judge": "rule-based", "harness": "Microsoft Phi internal eval", "prompt_style": "unknown", "temperature": "unknown" }, "notes": "Phi-4-reasoning paper Table 1 (gemini-2.5-pro column): livecodebench=69.1 (alt measurement, mc=false)" } ] }, { "model_id": "gemini-2.5-pro", "benchmark_id": "simpleqa", "score": 54.0, "reference_url": "https://storage.googleapis.com/deepmind-media/Model-Cards/Gemini-2-5-Pro-Model-Card.pdf", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1 (single attempt)", "judge": "rule-based", "harness": "AI Studio API", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Gemini 2.5 Pro Model Card (June 2025 GA): SimpleQA single attempt 54.0% [Promoted to verified per R5d, prior unverified value 52.9 from https://www.helicone.ai/blog/gemini-2.5-full-developer-guide deleted.]", "candidates": [ { "score": 52.9, "reference_url": "https://arxiv.org/abs/2504.13914", "source_type": "tech_report", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "notes": "Seed-Thinking-v1.5 paper Table 2: gemini-2.5-pro=52.9." } ] }, { "model_id": "gemini-2.5-pro", "benchmark_id": "mmmu", "score": 82.0, "reference_url": "https://storage.googleapis.com/deepmind-media/Model-Cards/Gemini-2-5-Pro-Model-Card.pdf", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1 (single attempt)", "judge": "rule-based", "harness": "AI Studio API", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Gemini 2.5 Pro Model Card (June 2025 GA): MMMU single attempt pass@1 82.0% (GA; was 81.7 from Experimental 03-25 column of same card, mislabeled) [Promoted to verified per R5d, prior unverified value 81.7 from https://modelcards.withgoogle.com/assets/documents/gemini-2.5-pro.pdf deleted.]", "candidates": [] }, { "model_id": "gemini-2.5-pro", "benchmark_id": "mmlu", "score": 89.8, "reference_url": "https://modelcards.withgoogle.com/assets/documents/gemini-2.5-pro.pdf", "audit_status": "dropped", "notes": " | DROPPED: BP cell value 89.8 from modelcards.withgoogle.com/.../gemini-2.5-pro.pdf is actually Global MMLU Lite from the Experimental 03-25 column, NOT standard MMLU 5-shot 14k. Source mislabel + wrong-version. Standard MMLU 5-shot is not reported in this card." }, { "model_id": "gemini-2.5-pro", "benchmark_id": "arena_hard", "score": 90.8, "reference_url": "https://github.com/lmarena/arena-hard-auto", "audit_status": "verified", "reported_setting": "Arena-Hard-v2.0 Creative Writing, Ensemble GPT-4.1+Gemini-2.5-Pro judge", "source_type": "leaderboard", "matches_canonical": true }, { "model_id": "gemini-2.5-pro", "benchmark_id": "chatbot_arena_elo", "score": 1448, "reference_url": "https://lmarena.ai/leaderboard/text", "audit_status": "verified", "source_type": "leaderboard", "matches_canonical": true, "notes": "lmarena.ai text arena (fetched 2026-04-29). Arena model name: gemini-2.5-pro (rank 42). ELO updates continuously; score reflects latest available." }, { "model_id": "gemini-2.5-pro", "benchmark_id": "aa_intelligence_index", "score": 60, "reference_url": "https://huggingface.co/MiniMaxAI/MiniMax-M2", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "MiniMax M2 model card: gemini-2.5-pro=60." }, { "model_id": "gemini-2.5-pro", "benchmark_id": "arc_agi_2", "score": 4.9, "reference_url": "https://deepmind.google/models/gemini/flash/", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "DeepMind /models/gemini/flash/ Performance table: ARC-AGI-2 ARC Prize Verified 4.9% [Promoted to verified, prior unverified value deleted per R5d.]" }, { "model_id": "gemini-2.5-pro", "benchmark_id": "codeforces_rating", "score": 2001, "reference_url": "https://huggingface.co/Qwen/Qwen3-235B-A22B-Thinking-2507", "audit_status": "verified", "source_type": "model_card", "matches_canonical": true, "audit_note": "Qwen3-235B card: Gemini-2.5 Pro CFEval=2001.", "audited_by": "lychee-audit-bot", "audited_at": "2026-04-29T13:17:14Z" }, { "model_id": "gemini-2.5-pro", "benchmark_id": "frontiermath", "score": 14.1, "reference_url": "https://epoch.ai/benchmarks/frontiermath", "audit_status": "verified", "source_type": "leaderboard", "matches_canonical": true, "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none" }, "notes": "epoch.ai FrontierMath-2025-02-28-Private: gemini-2.5-pro (Gemini 2.5 Pro Jun 2025) = 14.1%. Canonical model-id gemini-2.5-pro matches. BP had 5.0 (stale/incorrect)." }, { "model_id": "gemini-2.5-pro", "benchmark_id": "hmmt_feb_2025", "score": 82.5, "reference_url": "https://arxiv.org/abs/2504.21318", "reported_setting": { "mode": "unknown", "effort": "unknown", "tools": "unknown", "sampling": "unknown", "judge": "rule-based", "harness": "Microsoft Phi internal eval", "prompt_style": "unknown", "temperature": "unknown" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Phi-4-reasoning paper Table 1 (gemini-2.5-pro column): hmmt_2025=82.5 (re-sourced from https://arxiv.org/abs/2504.21318, matches BP value)" }, { "model_id": "gemini-2.5-pro", "benchmark_id": "ifeval", "score": 91.5, "reference_url": "https://arxiv.org/abs/2504.13914", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Seed-Thinking-v1.5 paper Table 2: gemini-2.5-pro=91.5." }, { "model_id": "gemini-2.5-pro", "benchmark_id": "imo_2025", "score": 31.55, "reference_url": "https://matharena.ai/?comp=imo--imo_2025", "audit_status": "verified", "matches_canonical": true, "reported_setting": "thinking:default", "source_type": "leaderboard" }, { "model_id": "gemini-2.5-pro", "benchmark_id": "matharena_apex_2025", "score": 0.5, "reference_url": "https://www.vellum.ai/blog/google-gemini-3-benchmarks", "audit_status": "dropped", "rule_id": "R5h-third-party-blog", "notes": "DROPPED (R5h): random third-party blog/aggregator, no primary trail" }, { "model_id": "gemini-2.5-pro", "benchmark_id": "mmlu_pro", "score": 86, "reference_url": "https://huggingface.co/MiniMaxAI/MiniMax-M2", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "MiniMax M2 model card: gemini-2.5-pro=86.", "candidates": [ { "score": 86.3, "reference_url": "https://arxiv.org/abs/2504.13914", "source_type": "tech_report", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "notes": "Seed-Thinking-v1.5 paper Table 2: gemini-2.5-pro=86.3." } ] }, { "model_id": "gemini-2.5-pro", "benchmark_id": "usamo_2025", "score": 24.0, "reference_url": "https://files.sri.inf.ethz.ch/matharena/usamo_report.pdf", "audit_status": "verified", "source_type": "academic_paper", "matches_canonical": true, "audited_by": "lychee-audit-bot", "audited_at": "2026-04-29T13:11:23Z", "audit_note": "Table 1: GEMINI-2.5-PRO Total=10.1/42=24.0%. JSON=24.0 matches." }, { "model_id": "gemini-2.5-pro", "benchmark_id": "terminal_bench", "score": 32.6, "reference_url": "https://deepmind.google/models/gemini/flash/", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "Terminus-2", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "DeepMind /models/gemini/flash/ Performance table: Terminal-Bench 2.0 Terminus-2 32.6% [Promoted to verified, prior unverified value deleted per R5d.]", "candidates": [ { "score": 25.3, "reference_url": "https://huggingface.co/MiniMaxAI/MiniMax-M2", "source_type": "model_card", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "agentic", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "notes": "MiniMax M2 model card: gemini-2.5-pro=25.3." } ] }, { "model_id": "gemini-2.5-pro", "benchmark_id": "arc_agi_1", "score": 27.6, "reference_url": "https://arxiv.org/abs/2504.13914", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Seed-Thinking-v1.5 paper Table 2: gemini-2.5-pro=27.6." }, { "model_id": "gemini-2.5-pro", "benchmark_id": "brumo_2025", "score": 90, "reference_url": "https://matharena.ai/?comp=brumo--brumo_2025", "audit_status": "verified", "matches_canonical": true, "reported_setting": "thinking:default", "source_type": "leaderboard" }, { "model_id": "gemini-2.5-pro", "benchmark_id": "cmimc_2025", "score": 58.13, "reference_url": "https://matharena.ai/?comp=cmimc--cmimc_2025", "audit_status": "verified", "matches_canonical": true, "reported_setting": "thinking:default", "source_type": "leaderboard" }, { "model_id": "gemini-2.5-pro", "benchmark_id": "critpt", "score": 2, "reference_url": "https://github.com/CritPt-Benchmark/CritPt", "audit_status": "verified", "reported_setting": "CritPt leaderboard, ε=1.5, % tasks solved", "source_type": "leaderboard", "matches_canonical": true }, { "model_id": "gemini-2.5-pro", "benchmark_id": "hmmt_nov_2025", "score": 80.0, "reference_url": "https://matharena.ai/?comp=hmmt--hmmt_nov_2025", "audit_status": "verified", "matches_canonical": true, "reported_setting": "thinking:default", "source_type": "leaderboard" }, { "model_id": "gemini-2.5-pro", "benchmark_id": "ifbench", "score": 49, "reference_url": "https://huggingface.co/MiniMaxAI/MiniMax-M2", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "MiniMax M2 model card: gemini-2.5-pro=49." }, { "model_id": "gemini-2.5-pro", "benchmark_id": "livebench", "score": 57.5, "reference_url": "https://livebench.ai/", "audit_status": "verified", "source_type": "leaderboard", "matches_canonical": true, "reported_setting": { "mode": "thinking", "effort": "default", "tools": "n/a" }, "notes": "livebench.ai table_2026_01_08 (latest as of 2026-04-28): gemini-2.5-pro-06-05-highthinking avg=57.5. Canonical thinking+default matches. BP had 58.3 (rounding difference)." }, { "model_id": "gemini-2.5-pro", "benchmark_id": "mrcr_v2", "score": 58.0, "reference_url": "https://deepmind.google/models/gemini/flash/", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default", "context": "128k" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "DeepMind /models/gemini/flash/ Performance table: MRCR v2 8-needle 128k 58.0% [Promoted to verified, prior unverified value deleted per R5d.]", "candidates": [ { "score": 16.4, "reference_url": "https://deepmind.google/models/gemini/flash/", "source_type": "official_blog", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default", "context": "1m" }, "notes": "DeepMind /models/gemini/flash/ Performance table: MRCR v2 1M 16.4%" } ] }, { "model_id": "gemini-2.5-pro", "benchmark_id": "simplebench", "score": 62.4, "reference_url": "https://lmcouncil.ai/benchmarks", "audit_status": "dropped", "rule_id": "R5h-third-party-blog", "notes": "DROPPED (R5h): random third-party blog/aggregator, no primary trail | DROPPED (R5g-c stale, never verified, source not audited in any pass)" }, { "model_id": "gemini-2.5-pro", "benchmark_id": "smt_2025", "score": 84.91, "reference_url": "https://matharena.ai/?comp=smt--smt_2025", "audit_status": "verified", "matches_canonical": true, "reported_setting": "thinking:default", "source_type": "leaderboard" }, { "model_id": "gemini-2.5-pro", "benchmark_id": "terminal_bench_1", "score": 25.3, "reference_url": "https://www.tbench.ai/leaderboard/terminal-bench/1.0", "audit_status": "verified", "source_type": "leaderboard", "rule_ids": [ "R5d" ], "notes": "Confirmed 25.3% on tbench.ai Terminal-Bench 1.0 leaderboard (Terminus 1 scaffold, 2025-05-15)." }, { "model_id": "gemini-2.5-flash", "benchmark_id": "aime_2025", "score": 72.0, "reference_url": "https://deepmind.google/models/gemini/flash/", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "DeepMind /models/gemini/flash/ Performance table: AIME 2025 no tools 72.0% [Promoted to verified, prior unverified value deleted per R5d.]", "candidates": [ { "score": 75.7, "reference_url": "https://deepmind.google/models/gemini/flash/", "source_type": "official_blog", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "code", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "notes": "DeepMind /models/gemini/flash/ Performance table: AIME 2025 with code 75.7%" } ] }, { "model_id": "gemini-2.5-flash", "benchmark_id": "arena_hard", "score": 83.9, "reference_url": "https://github.com/lmarena/arena-hard-auto", "audit_status": "verified", "reported_setting": "Arena-Hard-v2.0 Creative Writing, Ensemble GPT-4.1+Gemini-2.5-Pro judge", "source_type": "leaderboard", "matches_canonical": true }, { "model_id": "gemini-2.5-flash", "benchmark_id": "aime_2024", "score": 88.0, "reference_url": "https://modelcards.withgoogle.com/assets/documents/gemini-2.5-flash.pdf", "audit_status": "dropped", "rule_id": "R5g-ghost", "audit_note": "Source URL (modelcards.withgoogle.com/assets/documents/gemini-2.5-flash.pdf) returns 302 redirect to /models/model-cards; PDF is permanently dead/inaccessible.", "audited_by": "lychee-audit-bot", "audited_at": "2026-07-26T00:00:00Z" }, { "model_id": "gemini-2.5-flash", "benchmark_id": "livecodebench", "score": 63.9, "reference_url": "https://storage.googleapis.com/deepmind-media/Model-Cards/Gemini-2-5-Flash-Model-Card.pdf", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "AI Studio API", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Gemini 2.5 Flash Model Card (Dec 2025) GA Thinking column: LiveCodeBench v5 single attempt 63.9" }, { "model_id": "gemini-2.5-flash", "benchmark_id": "codeforces_rating", "score": 1995, "reference_url": "https://huggingface.co/Qwen/Qwen3-30B-A3B-Thinking-2507", "audit_status": "verified", "source_type": "model_card", "matches_canonical": true, "audit_note": "Qwen3-30B-Thinking card: Gemini2.5-Flash-Thinking CFEval=1995.", "audited_by": "lychee-audit-bot", "audited_at": "2026-04-29T13:17:14Z" }, { "model_id": "gemini-2.5-flash", "benchmark_id": "gpqa_diamond", "score": 82.8, "reference_url": "https://deepmind.google/models/gemini/flash/", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "DeepMind /models/gemini/flash/ Performance table: GPQA Diamond 82.8% [Promoted to verified, prior unverified value deleted per R5d.]" }, { "model_id": "gemini-2.5-flash", "benchmark_id": "hmmt_feb_2025", "score": 64.2, "reference_url": "https://huggingface.co/Qwen/Qwen3-30B-A3B-Thinking-2507", "audit_status": "verified", "source_type": "model_card", "matches_canonical": true, "audit_note": "Qwen3-30B-Thinking card: Gemini2.5-Flash-Thinking HMMT25=64.2.", "audited_by": "lychee-audit-bot", "audited_at": "2026-04-29T13:17:14Z" }, { "model_id": "gemini-2.5-flash", "benchmark_id": "ifeval", "score": 84.3, "reference_url": "https://huggingface.co/Qwen/Qwen3-30B-A3B-Instruct-2507", "audit_status": "verified", "source_type": "model_card", "matches_canonical": true, "audit_note": "Qwen3-30B-Instruct card: Gemini-2.5-Flash Non-Thinking IFEval=84.3.", "audited_by": "lychee-audit-bot", "audited_at": "2026-04-29T13:17:14Z" }, { "model_id": "gemini-2.5-flash", "benchmark_id": "mmlu", "score": 85.0, "reference_url": "https://deepmind.google/technologies/gemini/flash/", "audit_status": "dropped", "rule_id": "R5g-ghost", "audit_note": "deepmind.google/technologies/gemini/flash/ now shows Gemini 3 Flash; benchmark table shows AIME 2025 and MMMLU — no standard MMLU=85.0 for Gemini 2.5 Flash.", "audited_by": "lychee-audit-bot", "audited_at": "2026-07-26T00:00:00Z" }, { "model_id": "gemini-2.5-flash", "benchmark_id": "mmlu_pro", "score": 81.1, "reference_url": "https://huggingface.co/Qwen/Qwen3-30B-A3B-Instruct-2507", "audit_status": "verified", "source_type": "model_card", "matches_canonical": true, "audit_note": "Qwen3-30B-Instruct card: Gemini-2.5-Flash Non-Thinking MMLU-Pro=81.1.", "audited_by": "lychee-audit-bot", "audited_at": "2026-04-29T13:17:14Z" }, { "model_id": "gemini-2.5-flash", "benchmark_id": "terminal_bench", "score": 16.9, "reference_url": "https://deepmind.google/models/gemini/flash/", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "Terminus-2", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "DeepMind /models/gemini/flash/ Performance table: Terminal-Bench 2.0 Terminus-2 16.9% [Promoted to verified, prior unverified value deleted per R5d.]" }, { "model_id": "gemini-2.5-flash", "benchmark_id": "arc_agi_1", "score": 33.3, "reference_url": "https://arcprize.org/leaderboard", "audit_status": "verified", "source_type": "leaderboard", "matches_canonical": true, "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1 (single-pass)", "judge": "rule-based", "harness": "official (arcprize.org)", "system_type": "CoT" }, "notes": "arcprize.org leaderboard audit: Gemini 2.5 Flash (Preview) on leaderboard" }, { "model_id": "gemini-2.5-flash", "benchmark_id": "arc_agi_2", "score": 2.5, "reference_url": "https://deepmind.google/models/gemini/flash/", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "DeepMind /models/gemini/flash/ Performance table: ARC-AGI-2 ARC Prize Verified 2.5% [Promoted to verified, prior unverified value deleted per R5d.]" }, { "model_id": "gemini-2.5-flash", "benchmark_id": "brumo_2025", "score": 83.33, "reference_url": "https://matharena.ai/?comp=brumo--brumo_2025", "audit_status": "verified", "matches_canonical": true, "reported_setting": "thinking:default", "source_type": "leaderboard" }, { "model_id": "gemini-2.5-flash", "benchmark_id": "cmimc_2025", "score": 51.88, "reference_url": "https://matharena.ai/?comp=cmimc--cmimc_2025", "audit_status": "verified", "matches_canonical": true, "reported_setting": "thinking:default", "source_type": "leaderboard" }, { "model_id": "gemini-2.5-flash", "benchmark_id": "critpt", "score": 1.1, "reference_url": "https://github.com/CritPt-Benchmark/CritPt", "audit_status": "verified", "reported_setting": "CritPt leaderboard, ε=1.5, % tasks solved", "source_type": "leaderboard", "matches_canonical": true }, { "model_id": "gemini-2.5-flash", "benchmark_id": "hle", "score": 11.0, "reference_url": "https://deepmind.google/models/gemini/flash/", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "DeepMind /models/gemini/flash/ Performance table: HLE no tools 11.0% (Gemini 2.5 Flash Thinking) [Promoted to verified, prior unverified value deleted per R5d.]" }, { "model_id": "gemini-2.5-flash", "benchmark_id": "humaneval", "score": 90.2, "reference_url": "https://llm-stats.com/models/gemini-2.5-flash", "audit_status": "dropped", "rule_id": "R5h-third-party-blog", "notes": "DROPPED (R5h): random third-party blog/aggregator, no primary trail" }, { "model_id": "gemini-2.5-flash", "benchmark_id": "livebench", "score": 46.9, "reference_url": "https://livebench.ai/", "audit_status": "verified", "source_type": "leaderboard", "matches_canonical": false, "reported_setting": { "mode": "thinking", "effort": "default", "tools": "n/a" }, "notes": "livebench.ai table_2026_01_08 (latest as of 2026-04-28): gemini-2.5-flash-06-05-highthinking avg=46.9. Canonical requires Dec 2025 GA version; livebench has Jun 2025 preview (46.9) and Sep 2025 preview (52.3). Used Jun 2025 as closest to model release. BP had 47.7. See review." }, { "model_id": "gemini-2.5-flash", "benchmark_id": "math_500", "score": 95.2, "reference_url": "https://llm-stats.com/models/gemini-2.5-flash", "audit_status": "dropped", "rule_id": "R5h-third-party-blog", "notes": "DROPPED (R5h): random third-party blog/aggregator, no primary trail" }, { "model_id": "gemini-2.5-flash", "benchmark_id": "mmmu", "score": 79.7, "reference_url": "https://storage.googleapis.com/deepmind-media/Model-Cards/Gemini-2-5-Flash-Model-Card.pdf", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "AI Studio API", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Gemini 2.5 Flash Model Card (Dec 2025) GA Thinking column: MMMU single attempt 79.7. [R5d: prior unverified value 73.5 from https://llm-stats.com/models/gemini-2.5-flash deleted.]" }, { "model_id": "gemini-2.5-flash", "benchmark_id": "simpleqa", "score": 26.9, "reference_url": "https://storage.googleapis.com/deepmind-media/Model-Cards/Gemini-2-5-Flash-Model-Card.pdf", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "AI Studio API", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Gemini 2.5 Flash Model Card (Dec 2025) GA Thinking column: SimpleQA 26.9. [R5d: prior unverified value 28.1 from https://llm-stats.com/models/gemini-2.5-flash deleted.]" }, { "model_id": "gemini-2.5-flash", "benchmark_id": "smt_2025", "score": 75.47, "reference_url": "https://matharena.ai/?comp=smt--smt_2025", "audit_status": "verified", "matches_canonical": true, "reported_setting": "thinking:default", "source_type": "leaderboard" }, { "model_id": "gemini-2.5-flash", "benchmark_id": "swe_bench_verified", "score": 60.4, "reference_url": "https://deepmind.google/models/gemini/flash/", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "DeepMind /models/gemini/flash/ Performance table: SWE-bench Verified 60.4% [Promoted to verified, prior unverified value deleted per R5d.]" }, { "model_id": "gemini-2.5-flash", "benchmark_id": "terminal_bench_1", "score": 16.8, "reference_url": "https://www.tbench.ai/leaderboard/terminal-bench/1.0", "audit_status": "verified", "source_type": "leaderboard", "rule_ids": [ "R5d" ], "notes": "Confirmed 16.8% on tbench.ai Terminal-Bench 1.0 leaderboard (Terminus 1 scaffold, 2025-05-17)." }, { "model_id": "gemma-3-27b", "benchmark_id": "mmlu_pro", "score": 67.5, "reference_url": "https://arxiv.org/abs/2503.19786", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Gemma 3 tech report Table 6 IT 27B: MMLU-Pro 67.5", "candidates": [ { "score": 67.6, "reference_url": "https://arxiv.org/abs/2607.02770", "source_type": "tech_report", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "provider-reported", "judge": "benchmark-specified", "harness": "official Google evaluation", "prompt_style": "official", "temperature": "not disclosed", "context": "default" }, "notes": "Gemma 4 Technical Report Table 5; Gemma 4 thinking unless explicitly stated, Gemma 3 27B non-thinking." } ] }, { "model_id": "gemma-3-27b", "benchmark_id": "gpqa_diamond", "score": 42.4, "reference_url": "https://arxiv.org/abs/2503.19786", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Gemma 3 tech report Table 6 IT 27B: GPQA Diamond 42.4" }, { "model_id": "gemma-3-27b", "benchmark_id": "livecodebench", "score": 29.7, "reference_url": "https://arxiv.org/abs/2503.19786", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Gemma 3 tech report Table 6 IT 27B: LiveCodeBench 29.7" }, { "model_id": "gemma-3-27b", "benchmark_id": "humaneval", "score": 87.8, "reference_url": "https://www.emergentmind.com/topics/gemma-3-27b-it", "audit_status": "dropped", "rule_id": "R5h-third-party-blog", "notes": "DROPPED (R5h): random third-party blog/aggregator, no primary trail" }, { "model_id": "gemma-3-27b", "benchmark_id": "chatbot_arena_elo", "score": 1338, "reference_url": "https://arxiv.org/abs/2503.19786", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Gemma 3 tech report Table 6 IT 27B: Chatbot Arena Elo 1338 (Table 5)" }, { "model_id": "gemma-3-27b", "benchmark_id": "mmlu", "score": 76.9, "reference_url": "https://arxiv.org/abs/2503.19786", "audit_status": "dropped", "notes": " | DROPPED: BP value 76.9 with cell URL arxiv:2503.19786 — Table 6 IT does not include MMLU; pre-trained Table 9 shows 75.2. Value 76.9 not present in paper. Ghost cell per R5g(a)." }, { "model_id": "gemma-3-27b", "benchmark_id": "arena_hard", "score": 69.9, "reference_url": "https://github.com/lmarena/arena-hard-auto", "audit_status": "verified", "reported_setting": "Arena-Hard-v2.0 Creative Writing, Ensemble GPT-4.1+Gemini-2.5-Pro judge", "source_type": "leaderboard", "matches_canonical": true }, { "model_id": "gemma-3-27b", "benchmark_id": "gsm8k", "score": 92.3, "reference_url": "https://arxiv.org/abs/2503.19786", "audit_status": "dropped", "notes": " | DROPPED: BP value 92.3 with cell URL arxiv:2503.19786 — Gemma 3 paper Table 6 IT does not report GSM8K; pre-trained 27B GSM8K = 74.6. Value 92.3 not present. Ghost cell per R5g(a)." }, { "model_id": "gemma-3-27b", "benchmark_id": "aime_2024", "score": 22, "reference_url": "https://llm-stats.com/benchmarks", "audit_status": "dropped", "notes": " | DROPPED: BP cell URL llm-stats with value 22; Gemma 3 paper does not report AIME 2024. Third-party ghost; canonical paper has no entry to verify against." }, { "model_id": "gemma-3-27b", "benchmark_id": "ifeval", "score": 90.4, "reference_url": "https://arxiv.org/abs/2607.02770", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "provider-reported", "judge": "benchmark-specified", "harness": "official Google evaluation", "prompt_style": "official", "temperature": "not disclosed", "context": "default" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Gemma 4 Technical Report Table 5; Gemma 4 thinking unless explicitly stated, Gemma 3 27B non-thinking." }, { "model_id": "gemma-3-27b", "benchmark_id": "math_500", "score": 78, "reference_url": "https://llm-stats.com/benchmarks", "audit_status": "dropped", "notes": " | DROPPED: BP cell URL llm-stats with value 78; Gemma 3 paper Table 6 reports MATH (full set, score 89.0) not MATH-500 subset. Wrong-benchmark ghost." }, { "model_id": "gemma-3-27b", "benchmark_id": "simpleqa", "score": 10.0, "reference_url": "https://arxiv.org/abs/2503.19786", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Gemma 3 tech report Table 6 IT 27B: SimpleQA 10.0 (was 22 from llm-stats). [R5d: prior unverified value 22 from https://llm-stats.com/benchmarks deleted.]" }, { "model_id": "gemma-3-27b", "benchmark_id": "swe_bench_verified", "score": 32, "reference_url": "https://llm-stats.com/benchmarks", "audit_status": "dropped", "notes": " | DROPPED: BP cell URL llm-stats with value 32; Gemma 3 paper does not report SWE-bench Verified. Third-party ghost." }, { "model_id": "gemini-3-pro", "benchmark_id": "gpqa_diamond", "score": 91.9, "reference_url": "https://deepmind.google/models/gemini/pro/", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "DeepMind /models/gemini/pro/ Performance table: GPQA Diamond no tools 91.9%", "candidates": [ { "score": 91.9, "reference_url": "https://www.vellum.ai/blog/google-gemini-3-benchmarks", "source_type": "third_party", "reported_setting": {}, "notes": "Demoted from primary on https://deepmind.google/models/gemini/pro/ audit. Original notes: " }, { "score": 91.9, "reference_url": "https://deepmind.google/models/gemini/flash/", "source_type": "official_blog", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "notes": "DeepMind /models/gemini/flash/ Performance table: GPQA Diamond 91.9%" }, { "score": 91.0, "reference_url": "https://huggingface.co/MiniMaxAI/MiniMax-M2.5", "source_type": "model_card", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "notes": "MiniMax M2.5 model card: gemini-3-pro=91.0." } ] }, { "model_id": "gemini-3-pro", "benchmark_id": "aime_2025", "score": 95.0, "reference_url": "https://deepmind.google/models/gemini/flash/", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "DeepMind /models/gemini/flash/ Performance table: AIME 2025 no tools 95.0% [Promoted to verified, prior unverified value deleted per R5d.]", "candidates": [ { "score": 100.0, "reference_url": "https://deepmind.google/models/gemini/flash/", "source_type": "official_blog", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "code", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "notes": "DeepMind /models/gemini/flash/ Performance table: AIME 2025 with code 100%" }, { "score": 96.0, "reference_url": "https://huggingface.co/MiniMaxAI/MiniMax-M2.5", "source_type": "model_card", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "notes": "MiniMax M2.5 model card: gemini-3-pro=96.0." } ] }, { "model_id": "gemini-3-pro", "benchmark_id": "swe_bench_verified", "score": 76.2, "reference_url": "https://deepmind.google/models/gemini/pro/", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "DeepMind /models/gemini/pro/ Performance table: SWE-Bench Verified single attempt 76.2%", "candidates": [ { "score": 76.2, "reference_url": "https://www.vellum.ai/blog/google-gemini-3-benchmarks", "source_type": "third_party", "reported_setting": {}, "notes": "Demoted from primary on https://deepmind.google/models/gemini/pro/ audit. Original notes: " }, { "score": 76.2, "reference_url": "https://deepmind.google/models/gemini/flash/", "source_type": "official_blog", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "notes": "DeepMind /models/gemini/flash/ Performance table: SWE-bench Verified 76.2%" } ] }, { "model_id": "gemini-3-pro", "benchmark_id": "hle", "score": 37.5, "reference_url": "https://deepmind.google/models/gemini/pro/", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "DeepMind /models/gemini/pro/ Performance table: HLE no tools 37.5%", "candidates": [ { "score": 44.4, "reference_url": "https://www.anthropic.com/news/claude-sonnet-4-6", "source_type": "third_party", "reported_setting": { "note": "see primary" }, "notes": "Reported by Anthropic Sonnet 4.6 blog (third-party, no tools)." }, { "score": 37.5, "reference_url": "https://www.vellum.ai/blog/google-gemini-3-benchmarks", "source_type": "third_party", "reported_setting": {}, "notes": "Demoted from primary on https://deepmind.google/models/gemini/pro/ audit. Original notes: " }, { "score": 45.8, "reference_url": "https://deepmind.google/models/gemini/pro/", "source_type": "official_blog", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "web+code", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "notes": "DeepMind /models/gemini/pro/ table: HLE Search (blocklist)+Code 45.8% (3 Pro). Different from no-tools primary 37.5%." }, { "score": 37.5, "reference_url": "https://deepmind.google/models/gemini/flash/", "source_type": "official_blog", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "notes": "DeepMind /models/gemini/flash/ Performance table: HLE no tools 37.5% (Gemini 3 Pro Thinking)" }, { "score": 45.8, "reference_url": "https://deepmind.google/models/gemini/flash/", "source_type": "official_blog", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "web+code", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "notes": "DeepMind /models/gemini/flash/ Performance table: HLE Search+Code 45.8%" }, { "score": 37.7, "reference_url": "https://huggingface.co/deepseek-ai/DeepSeek-V3.2/resolve/main/assets/paper.pdf", "source_type": "tech_report", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "notes": "DeepSeek V3.2 tech report Table 2: gemini-3-pro=37.7." }, { "score": 37.2, "reference_url": "https://z.ai/blog/glm-5", "source_type": "official_blog", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "notes": "GLM-5 blog: gemini-3-pro=37.2." }, { "score": 37.2, "reference_url": "https://huggingface.co/MiniMaxAI/MiniMax-M2.5", "source_type": "model_card", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "notes": "MiniMax M2.5 model card: gemini-3-pro=37.2." }, { "score": 33.3, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "source_type": "model_card", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "notes": "Doubao Seed 2.0 model card Table 3: gemini-3-pro=33.3." } ] }, { "model_id": "gemini-3-pro", "benchmark_id": "arc_agi_2", "score": 31.1, "reference_url": "https://deepmind.google/models/gemini/pro/", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "DeepMind /models/gemini/pro/ Performance table: ARC-AGI-2 ARC Prize Verified 31.1%", "candidates": [ { "score": 31.1, "reference_url": "https://www.vellum.ai/blog/google-gemini-3-benchmarks", "source_type": "third_party", "reported_setting": {}, "notes": "Demoted from primary on https://deepmind.google/models/gemini/pro/ audit. Original notes: " }, { "score": 31.1, "reference_url": "https://deepmind.google/models/gemini/flash/", "source_type": "official_blog", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "notes": "DeepMind /models/gemini/flash/ Performance table: ARC-AGI-2 ARC Prize Verified 31.1%" } ] }, { "model_id": "gemini-3-pro", "benchmark_id": "mmmu_pro", "score": 81.0, "reference_url": "https://deepmind.google/models/gemini/pro/", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "DeepMind /models/gemini/pro/ Performance table: MMMU-Pro no tools 81.0%", "candidates": [ { "score": 81.0, "reference_url": "https://www.vellum.ai/blog/google-gemini-3-benchmarks", "source_type": "third_party", "reported_setting": {}, "notes": "Demoted from primary on https://deepmind.google/models/gemini/pro/ audit. Original notes: " }, { "score": 81.0, "reference_url": "https://deepmind.google/models/gemini/flash/", "source_type": "official_blog", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "notes": "DeepMind /models/gemini/flash/ Performance table: MMMU-Pro 81.0%" }, { "score": 81.0, "reference_url": "https://deepmind.google/models/gemini/pro/", "source_type": "official_model_card", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official Google Gemini evaluation", "prompt_style": "provider official evaluation prompt", "temperature": "source/provider setting", "context": "source/provider setting", "input_modalities": "image and text", "trials": "source-reported" }, "notes": "Exact provider-official score and reported setting." } ] }, { "model_id": "gemini-3-pro", "benchmark_id": "mmlu_pro", "score": 90.1, "reference_url": "https://huggingface.co/deepseek-ai/DeepSeek-V3.2/resolve/main/assets/paper.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "DeepSeek V3.2 tech report Table 2: gemini-3-pro=90.1.", "candidates": [ { "score": 89.8, "reference_url": "https://huggingface.co/Qwen/Qwen3.5-397B-A17B", "source_type": "model_card", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "notes": "Qwen3.5-397B-A17B HF model card cross-model table (Qwen self-test of 3rd-party model, mc=false): alternate measurement 89.8 (3rd-party Qwen self-test)." } ] }, { "model_id": "gemini-3-pro", "benchmark_id": "mrcr_v2", "score": 77.0, "reference_url": "https://deepmind.google/models/gemini/pro/", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default", "context": "128k" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "DeepMind /models/gemini/pro/ Performance table: MRCR v2 8-needle 128k average 77.0%", "candidates": [ { "score": 77.0, "reference_url": "https://www.vellum.ai/blog/google-gemini-3-benchmarks", "source_type": "third_party", "reported_setting": {}, "notes": "Demoted from primary on https://deepmind.google/models/gemini/pro/ audit. Original notes: " }, { "score": 26.3, "reference_url": "https://deepmind.google/models/gemini/pro/", "source_type": "official_blog", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default", "context": "1m" }, "notes": "DeepMind /models/gemini/pro/ table: MRCR v2 1M pointwise 26.3% (3 Pro)." }, { "score": 26.3, "reference_url": "https://deepmind.google/models/gemini/flash/", "source_type": "official_blog", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default", "context": "1m" }, "notes": "DeepMind /models/gemini/flash/ Performance table: MRCR v2 1M pointwise 26.3%" } ] }, { "model_id": "gemini-3-pro", "benchmark_id": "simpleqa", "score": 72.1, "reference_url": "https://officechai.com/ai/gemini-3-1-pro-benchmarks/", "audit_status": "dropped", "rule_id": "R5h-third-party-blog", "notes": "DROPPED (R5h): random third-party blog/aggregator, no primary trail" }, { "model_id": "gemini-3-pro", "benchmark_id": "chatbot_arena_elo", "score": 1486, "reference_url": "https://ernie.baidu.com/blog/posts/ernie-5.1-0508-release/creative-capability.png", "reported_setting": { "leaderboard": "Arena AI Text", "metric": "Arena score", "style_control": "on", "snapshot_date": "2026-04-30" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "April 30, 2026 Text Arena snapshot; style control on. Research observation obs-055. Displayed rank 8. Identical asset is published in four page placements. The same PNG bytes are published under two filenames and four English/Chinese page placements." }, { "model_id": "gemini-3-pro", "benchmark_id": "livecodebench", "score": 90.7, "reference_url": "https://huggingface.co/deepseek-ai/DeepSeek-V3.2/resolve/main/assets/paper.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "DeepSeek V3.2 tech report Table 2: gemini-3-pro=90.7." }, { "model_id": "gemini-3-pro", "benchmark_id": "browsecomp", "score": 59.2, "reference_url": "https://deepmind.google/models/gemini/pro/", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "web+code", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "DeepMind /models/gemini/pro/ Performance table (PROMOTED over prior third-party): BrowseComp Search+Python+Browse 59.2% (was 85.9 from vellum.ai which mislabeled 3.1 Pro number; demoted to candidate)", "candidates": [ { "score": 85.9, "reference_url": "https://www.vellum.ai/blog/google-gemini-3-benchmarks", "source_type": "third_party", "reported_setting": {}, "notes": "Demoted from primary on https://deepmind.google/models/gemini/pro/ audit. Original notes: " }, { "score": 37.8, "reference_url": "https://z.ai/blog/glm-5", "source_type": "official_blog", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "agentic", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "notes": "GLM-5 blog: gemini-3-pro=37.8." }, { "score": 37.8, "reference_url": "https://huggingface.co/moonshotai/Kimi-K2.5", "source_type": "model_card", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "agentic", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "1.0" }, "notes": "Kimi K2.5 model card: gemini-3-pro=37.8." } ] }, { "model_id": "gemini-3-pro", "benchmark_id": "tau_bench_telecom", "score": 99.3, "reference_url": "https://www.vellum.ai/blog/google-gemini-3-benchmarks", "audit_status": "dropped", "notes": " | DROPPED: BP cell value 99.3 from vellum.ai is actually τ²-bench-telecom for 3.1 Pro (not 3 Pro, not τ-bench). Source mislabel + wrong-model. Use tau2_bench_telecom for 3-pro." }, { "model_id": "gemini-3-pro", "benchmark_id": "mmlu", "score": 91.8, "reference_url": "https://www.vellum.ai/blog/google-gemini-3-benchmarks", "audit_status": "dropped", "notes": " | DROPPED: BP cell value 91.8 from vellum.ai is actually MMMLU not MMLU. Source mislabel. Use mmmlu cell instead." }, { "model_id": "gemini-3-pro", "benchmark_id": "humaneval", "score": 93.0, "reference_url": "https://www.vellum.ai/blog/google-gemini-3-benchmarks", "audit_status": "dropped", "rule_id": "R5h-third-party-blog", "notes": "DROPPED (R5h): random third-party blog/aggregator, no primary trail" }, { "model_id": "gemini-3-pro", "benchmark_id": "frontiermath", "score": 38.0, "reference_url": "https://x.com/EpochAIResearch/status/1991945942174761050", "audit_status": "needs_review", "audit_note": "Source is a Twitter/X post (x.com/EpochAIResearch/status/1991945942174761050); requires browser to verify tweet content. Value 38.0 not verified via automated fetch.", "audited_by": "lychee-audit-bot", "audited_at": "2026-07-26T00:00:00Z" }, { "model_id": "gemini-3-pro", "benchmark_id": "codeforces_rating", "score": 2708, "reference_url": "https://huggingface.co/deepseek-ai/DeepSeek-V3.2/resolve/main/assets/paper.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "DeepSeek V3.2 tech report Table 2: gemini-3-pro=2708.", "candidates": [ { "score": 2726, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "source_type": "model_card", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "notes": "Doubao Seed 2.0 model card Table 3: gemini-3-pro=2726." } ] }, { "model_id": "gemini-3-pro", "benchmark_id": "swe_bench_pro", "score": 43.3, "reference_url": "https://deepmind.google/models/gemini/pro/", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "DeepMind /models/gemini/pro/ Performance table: SWE-Bench Pro single attempt 43.3%", "candidates": [ { "score": 43.3, "reference_url": "https://scale.com/leaderboard/swe_bench_pro_public", "source_type": "third_party", "reported_setting": {}, "notes": "Demoted from primary on https://deepmind.google/models/gemini/pro/ audit. Original notes: " }, { "score": 49.7, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "source_type": "model_card", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "varies", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "notes": "Doubao Seed 2.0 model card: gemini-3-pro=49.7." } ] }, { "model_id": "gemini-3-pro", "benchmark_id": "mmmu", "score": 87.2, "reference_url": "https://huggingface.co/Qwen/Qwen3.5-397B-A17B", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Qwen3.5-397B-A17B HF model card cross-model table (Qwen self-test of 3rd-party model, mc=false): mmmu 87.2 (3rd-party Qwen self-test). [R5d: prior unverified value 87.51 from https://www.vellum.ai/blog/google-gemini-3-benchmarks deleted.]", "candidates": [ { "score": 87.0, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "source_type": "model_card", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "notes": "Doubao Seed 2.0 model card: gemini-3-pro=87.0." } ] }, { "model_id": "gemini-3-pro", "benchmark_id": "math_500", "score": 97.3, "reference_url": "https://artificialanalysis.ai/evaluations/math-500", "audit_status": "needs_review", "notes": "Reference URL is AA math-500 eval page, but gemini-3-1-pro-preview is NOT present on that page; AA shows math_500=null for this model. Possible ghost score (R5g-a). Cannot verify 97.3% from AA source." }, { "model_id": "gemini-3-pro", "benchmark_id": "ifeval", "score": 93.5, "reference_url": "https://huggingface.co/Qwen/Qwen3.5-397B-A17B", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Qwen3.5-397B-A17B HF model card cross-model table (Qwen self-test of 3rd-party model, mc=false): ifeval 93.5 (3rd-party Qwen self-test). [R5d: prior unverified value 90.0 from https://www.vellum.ai/blog/google-gemini-3-benchmarks deleted.]" }, { "model_id": "gemini-3-pro", "benchmark_id": "matharena_apex_2025", "score": 24.5, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "audit_status": "verified", "rule_id": "R5h-third-party-blog", "notes": "Doubao Seed 2.0 model card Table 3: gemini-3-pro=24.5.", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "candidates": [] }, { "model_id": "gemini-3-pro", "benchmark_id": "osworld", "score": 55.0, "reference_url": "https://www.vellum.ai/blog/google-gemini-3-benchmarks", "audit_status": "dropped", "rule_id": "R5h-third-party-blog", "notes": "DROPPED (R5h): random third-party blog/aggregator, no primary trail" }, { "model_id": "gemini-3-pro", "benchmark_id": "scicode", "score": 56.0, "reference_url": "https://deepmind.google/models/gemini/pro/", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "DeepMind /models/gemini/pro/ Performance table: SciCode 56% (rounded; existing 56.1 within rounding)", "candidates": [ { "score": 56.1, "reference_url": "https://artificialanalysis.ai/evaluations/scicode", "source_type": "third_party", "reported_setting": {}, "notes": "Demoted from primary on https://deepmind.google/models/gemini/pro/ audit. Original notes: " }, { "score": 56.1, "reference_url": "https://huggingface.co/moonshotai/Kimi-K2.5", "source_type": "model_card", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "1.0" }, "notes": "Kimi K2.5 model card: gemini-3-pro=56.1." }, { "score": 57.7, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "source_type": "model_card", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "varies", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "notes": "Doubao Seed 2.0 model card: gemini-3-pro=57.7." } ] }, { "model_id": "gemini-3-pro", "benchmark_id": "terminal_bench", "score": 56.9, "reference_url": "https://deepmind.google/models/gemini/pro/", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "Terminus-2", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "DeepMind /models/gemini/pro/ Performance table: Terminal-Bench 2.0 Terminus-2 56.9%", "candidates": [ { "score": 57.2, "reference_url": "https://www.anthropic.com/news/claude-sonnet-4-6", "source_type": "third_party", "reported_setting": { "note": "see primary" }, "notes": "Reported by Anthropic Sonnet 4.6 blog (third-party)." }, { "score": 56.9, "reference_url": "https://www.tbench.ai/leaderboard/terminal-bench/2.0", "source_type": "third_party", "reported_setting": {}, "notes": "Demoted from primary on https://deepmind.google/models/gemini/pro/ audit. Original notes: " }, { "score": 54.2, "reference_url": "https://deepmind.google/models/gemini/flash/", "source_type": "official_blog", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "Terminus-2", "prompt_style": "default", "temperature": "default" }, "notes": "DeepMind /models/gemini/flash/ Performance table: Terminal-Bench 2.0 Terminus-2 54.2% (Note: differs from /pro/ page 56.9 — Flash page shows 54.2)" }, { "score": 54.2, "reference_url": "https://huggingface.co/deepseek-ai/DeepSeek-V3.2/resolve/main/assets/paper.pdf", "source_type": "tech_report", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "agentic", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "notes": "DeepSeek V3.2 tech report Table 2: gemini-3-pro=54.2." }, { "score": 54.2, "reference_url": "https://z.ai/blog/glm-5", "source_type": "official_blog", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "agentic", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "notes": "GLM-5 blog: gemini-3-pro=54.2." }, { "score": 54.2, "reference_url": "https://z.ai/blog/glm-4.7", "source_type": "official_blog", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "agentic", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "notes": "GLM-4.7 blog: gemini-3-pro=54.2." }, { "score": 54.2, "reference_url": "https://huggingface.co/moonshotai/Kimi-K2.5", "source_type": "model_card", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "agentic", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "1.0" }, "notes": "Kimi K2.5 model card: gemini-3-pro=54.2." }, { "score": 54.2, "reference_url": "https://huggingface.co/Qwen/Qwen3.5-397B-A17B", "source_type": "model_card", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "notes": "Qwen3.5-397B-A17B HF model card cross-model table (Qwen self-test of 3rd-party model, mc=false): alternate measurement 54.2 (3rd-party Qwen self-test)." } ] }, { "model_id": "gemini-3-pro", "benchmark_id": "aime_2024", "score": 97, "reference_url": "https://www.vellum.ai/blog/google-gemini-3-benchmarks", "audit_status": "dropped", "rule_id": "R5h-third-party-blog", "notes": "DROPPED (R5h): random third-party blog/aggregator, no primary trail" }, { "model_id": "gemini-3-pro", "benchmark_id": "arc_agi_1", "score": 85.0, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card Table 3: gemini-3-pro=85.0." }, { "model_id": "gemini-3-pro", "benchmark_id": "brumo_2025", "score": 98.33, "reference_url": "https://matharena.ai/?comp=brumo--brumo_2025", "audit_status": "verified", "matches_canonical": true, "reported_setting": "thinking:high", "source_type": "leaderboard" }, { "model_id": "gemini-3-pro", "benchmark_id": "cmimc_2025", "score": 90, "reference_url": "https://matharena.ai/?comp=cmimc--cmimc_2025", "audit_status": "verified", "matches_canonical": true, "reported_setting": "thinking:high", "source_type": "leaderboard" }, { "model_id": "gemini-3-pro", "benchmark_id": "critpt", "score": 9.1, "reference_url": "https://github.com/CritPt-Benchmark/CritPt", "audit_status": "verified", "reported_setting": "CritPt leaderboard, ε=1.5, % tasks solved", "source_type": "leaderboard", "matches_canonical": true }, { "model_id": "gemini-3-pro", "benchmark_id": "hmmt_nov_2025", "score": 93.3, "reference_url": "https://huggingface.co/deepseek-ai/DeepSeek-V3.2/resolve/main/assets/paper.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "DeepSeek V3.2 tech report Table 2: gemini-3-pro=93.3.", "candidates": [ { "score": 93.0, "reference_url": "https://z.ai/blog/glm-5", "source_type": "official_blog", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "notes": "GLM-5 blog: gemini-3-pro=93.0." } ] }, { "model_id": "gemini-3-pro", "benchmark_id": "livebench", "score": 73.5, "reference_url": "https://livebench.ai/", "audit_status": "verified", "source_type": "leaderboard", "matches_canonical": true, "reported_setting": { "mode": "thinking", "effort": "high", "tools": "n/a" }, "notes": "livebench.ai table_2026_01_08 (latest as of 2026-04-28): gemini-3-pro-preview-11-2025-high avg=73.5. Canonical thinking+high matches. BP had 73.4 (0.1 rounding diff)." }, { "model_id": "gemini-3-pro", "benchmark_id": "smt_2025", "score": 93.4, "reference_url": "https://matharena.ai/?comp=smt--smt_2025", "audit_status": "verified", "matches_canonical": true, "reported_setting": "thinking:high", "source_type": "leaderboard" }, { "model_id": "gemini-3-pro", "benchmark_id": "tau_bench_retail", "score": 88.5, "reference_url": "https://artificialanalysis.ai/articles/gemini-3-pro-everything-you-need-to-know", "audit_status": "needs_review", "notes": "Reference URL is AA article page (gemini-3-pro-everything-you-need-to-know). AA only publishes aggregate tau2 scores; no retail-domain breakdown available in AA data. gemini-3-1-pro-preview tau2=0.956 aggregate. Cannot verify 88.5 from AA source." }, { "model_id": "gemini-3-flash", "benchmark_id": "gpqa_diamond", "score": 90.4, "reference_url": "https://deepmind.google/models/gemini/flash/", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "DeepMind /models/gemini/flash/ Performance table: GPQA Diamond no tools 90.4%", "candidates": [ { "score": 90.4, "reference_url": "https://www.vellum.ai/blog/google-gemini-3-benchmarks", "source_type": "third_party", "reported_setting": {}, "notes": "Demoted from primary on https://deepmind.google/models/gemini/flash/ audit. Original notes: " } ] }, { "model_id": "gemini-3-flash", "benchmark_id": "mmlu", "score": 91.8, "reference_url": "https://www.vellum.ai/blog/google-gemini-3-benchmarks", "audit_status": "dropped", "notes": " | DROPPED: BP cell 91.8 from vellum.ai is actually MMMLU (multilingual) not MMLU. Source mislabel. Use mmmlu cell instead." }, { "model_id": "gemini-3-flash", "benchmark_id": "mmlu_pro", "score": 88.59, "reference_url": "https://www.vellum.ai/blog/google-gemini-3-benchmarks", "audit_status": "dropped", "rule_id": "R5h-third-party-blog", "notes": "DROPPED (R5h): random third-party blog/aggregator, no primary trail" }, { "model_id": "gemini-3-flash", "benchmark_id": "swe_bench_verified", "score": 78.0, "reference_url": "https://deepmind.google/models/gemini/flash/", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "DeepMind /models/gemini/flash/ Performance table: SWE-bench Verified single attempt 78.0%", "candidates": [ { "score": 78.0, "reference_url": "https://www.vellum.ai/blog/google-gemini-3-benchmarks", "source_type": "third_party", "reported_setting": {}, "notes": "Demoted from primary on https://deepmind.google/models/gemini/flash/ audit. Original notes: " } ] }, { "model_id": "gemini-3-flash", "benchmark_id": "hle", "score": 33.7, "reference_url": "https://deepmind.google/models/gemini/flash/", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "DeepMind /models/gemini/flash/ Performance table: HLE no tools 33.7%", "candidates": [ { "score": 33.7, "reference_url": "https://www.vellum.ai/blog/google-gemini-3-benchmarks", "source_type": "third_party", "reported_setting": {}, "notes": "Demoted from primary on https://deepmind.google/models/gemini/flash/ audit. Original notes: " }, { "score": 43.5, "reference_url": "https://deepmind.google/models/gemini/flash/", "source_type": "official_blog", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "web+code", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "notes": "DeepMind /models/gemini/flash/ table: HLE Search+Code 43.5%" } ] }, { "model_id": "gemini-3-flash", "benchmark_id": "mmmu_pro", "score": 81.2, "reference_url": "https://deepmind.google/models/gemini/flash/", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "DeepMind /models/gemini/flash/ Performance table: MMMU-Pro 81.2%", "candidates": [ { "score": 81.2, "reference_url": "https://www.vellum.ai/blog/google-gemini-3-benchmarks", "source_type": "third_party", "reported_setting": {}, "notes": "Demoted from primary on https://deepmind.google/models/gemini/flash/ audit. Original notes: " } ] }, { "model_id": "gemini-3-flash", "benchmark_id": "aime_2025", "score": 95.2, "reference_url": "https://deepmind.google/models/gemini/flash/", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "DeepMind /models/gemini/flash/ Performance table: AIME 2025 no tools 95.2%", "candidates": [ { "score": 99.7, "reference_url": "https://deepmind.google/models/gemini/flash/", "source_type": "official_blog", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "code", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "notes": "DeepMind /models/gemini/flash/ table: AIME 2025 with code execution 99.7%" } ] }, { "model_id": "gemini-3-flash", "benchmark_id": "mmmu", "score": 87.63, "reference_url": "https://www.vellum.ai/blog/google-gemini-3-benchmarks", "audit_status": "dropped", "rule_id": "R5h-third-party-blog", "notes": "DROPPED (R5h): random third-party blog/aggregator, no primary trail" }, { "model_id": "gemini-3-flash", "benchmark_id": "chatbot_arena_elo", "score": 1473, "reference_url": "https://lmarena.ai/leaderboard/text", "audit_status": "verified", "source_type": "leaderboard", "matches_canonical": true, "notes": "lmarena.ai text arena (fetched 2026-04-29). Arena model name: gemini-3-flash (rank 15). ELO updates continuously; score reflects latest available." }, { "model_id": "gemini-3-flash", "benchmark_id": "browsecomp", "score": 75.0, "reference_url": "https://www.vellum.ai/blog/google-gemini-3-benchmarks", "audit_status": "dropped", "rule_id": "R5h-third-party-blog", "notes": "DROPPED (R5h): random third-party blog/aggregator, no primary trail" }, { "model_id": "gemini-3-flash", "benchmark_id": "frontiermath", "score": 35.6, "reference_url": "https://epoch.ai/benchmarks/frontiermath", "audit_status": "verified", "source_type": "leaderboard", "matches_canonical": true, "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none" }, "notes": "epoch.ai FrontierMath-2025-02-28-Private: gemini-3-flash-preview = 35.6% (Display: Gemini 3 Flash). BP had 30.0." }, { "model_id": "gemini-3-flash", "benchmark_id": "livecodebench", "score": 90.8, "reference_url": "https://artificialanalysis.ai/evaluations/livecodebench", "audit_status": "verified", "source_type": "third_party", "notes": "ArtificialAnalysis aggregator; used as fallback when no primary source available (R5d-aggregator-fallback). AA LiveCodeBench eval page: gemini-3-flash-reasoning livecodebench=0.908 → 90.8%. AA slug: gemini-3-flash-reasoning (reasoning/thinking variant).", "reported_setting": { "mode": "thinking", "effort": "default", "sampling": "pass@1", "harness": "AA standard evaluation", "notes": "AA slug: gemini-3-flash-reasoning; livecodebench=90.8%" } }, { "model_id": "gemini-3-flash", "benchmark_id": "osworld", "score": 50.0, "reference_url": "https://www.vellum.ai/blog/google-gemini-3-benchmarks", "audit_status": "dropped", "rule_id": "R5h-third-party-blog", "notes": "DROPPED (R5h): random third-party blog/aggregator, no primary trail" }, { "model_id": "gemini-3-flash", "benchmark_id": "simpleqa", "score": 60.0, "reference_url": "https://www.vellum.ai/blog/google-gemini-3-benchmarks", "audit_status": "dropped", "rule_id": "R5h-third-party-blog", "notes": "DROPPED (R5h): random third-party blog/aggregator, no primary trail" }, { "model_id": "gemini-3-flash", "benchmark_id": "swe_bench_pro", "score": 34.63, "reference_url": "https://scale.com/leaderboard/swe_bench_pro_public", "audit_status": "verified", "source_type": "leaderboard", "matches_canonical": false, "notes": " [Verified against Scale.com SWE-Bench Pro Public leaderboard (2026-04-29): gemini-3-flash = 34.63. Scale uses standardized evaluation scaffolding.]", "candidates": [ { "score": 49.6, "reference_url": "https://deepmind.google/models/model-cards/gemini-3-5-flash/", "source_type": "official_model_card", "reported_setting": { "mode": "thinking/reasoning", "effort": "source does not state", "tools": "code execution", "sampling": "Gemini avg 5 runs, single attempt per run", "judge": "benchmark-specified", "harness": "internal Antigravity for Gemini; provider/public for others", "temperature": "default", "snapshot": "May 2026" }, "notes": "Google Gemini 3.5 Flash May 2026 table: SWE-Bench Pro (Public); single attempt. Settings and provenance are preserved per cell." } ] }, { "model_id": "gemini-3-flash", "benchmark_id": "terminal_bench", "score": 47.6, "reference_url": "https://deepmind.google/models/gemini/flash/", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "Terminus-2", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "DeepMind /models/gemini/flash/ (PROMOTED over prior third-party): Terminal-Bench 2.0 Terminus-2 47.6% (was 51.7 from tbench leaderboard, demoted to candidate)", "candidates": [ { "score": 51.7, "reference_url": "https://www.tbench.ai/leaderboard/terminal-bench/2.0", "source_type": "third_party", "reported_setting": {}, "notes": "Demoted from primary on https://deepmind.google/models/gemini/flash/ audit. Original notes: " } ] }, { "model_id": "gemini-3-flash", "benchmark_id": "aime_2024", "score": 93, "reference_url": "https://medium.com/@leucopsis/gemini-3-flash-preliminary-review-34e7420e3be7", "audit_status": "dropped", "rule_id": "R5h-third-party-blog", "notes": "DROPPED (R5h): random third-party blog/aggregator, no primary trail" }, { "model_id": "gemini-3-flash", "benchmark_id": "arc_agi_1", "score": 84.7, "reference_url": "https://arcprize.org/leaderboard", "audit_status": "verified", "source_type": "leaderboard", "matches_canonical": false, "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1 (single-pass)", "judge": "rule-based", "harness": "official (arcprize.org)", "system_type": "CoT" }, "notes": "arcprize.org leaderboard audit: Gemini 3 Flash Preview (High) on leaderboard; canonical is default" }, { "model_id": "gemini-3-flash", "benchmark_id": "arc_agi_2", "score": 33.6, "reference_url": "https://deepmind.google/models/gemini/flash/", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "DeepMind /models/gemini/flash/ Performance table: ARC-AGI-2 ARC Prize Verified 33.6%", "candidates": [ { "score": 33.6, "reference_url": "https://arcprize.org/arc-agi/2/", "source_type": "third_party", "reported_setting": {}, "notes": "Demoted from primary on https://deepmind.google/models/gemini/flash/ audit. Original notes: " } ] }, { "model_id": "gemini-3-flash", "benchmark_id": "brumo_2025", "score": 100, "reference_url": "https://matharena.ai/?comp=brumo--brumo_2025", "audit_status": "verified", "matches_canonical": true, "reported_setting": "thinking:default", "source_type": "leaderboard" }, { "model_id": "gemini-3-flash", "benchmark_id": "cmimc_2025", "score": 90.62, "reference_url": "https://matharena.ai/?comp=cmimc--cmimc_2025", "audit_status": "verified", "matches_canonical": true, "reported_setting": "thinking:default", "source_type": "leaderboard" }, { "model_id": "gemini-3-flash", "benchmark_id": "codeforces_rating", "score": 2100, "reference_url": "https://medium.com/@leucopsis/gemini-3-flash-preliminary-review-34e7420e3be7", "audit_status": "dropped", "rule_id": "R5h-third-party-blog", "notes": "DROPPED (R5h): random third-party blog/aggregator, no primary trail" }, { "model_id": "gemini-3-flash", "benchmark_id": "hmmt_nov_2025", "score": 93.33, "reference_url": "https://matharena.ai/?comp=hmmt--hmmt_nov_2025", "audit_status": "verified", "matches_canonical": true, "reported_setting": "thinking:default", "source_type": "leaderboard" }, { "model_id": "gemini-3-flash", "benchmark_id": "ifeval", "score": 88.2, "reference_url": "https://automatio.ai/models/gemini-3-flash", "audit_status": "dropped", "rule_id": "R5h-third-party-blog", "notes": "DROPPED (R5h): random third-party blog/aggregator, no primary trail" }, { "model_id": "gemini-3-flash", "benchmark_id": "livebench", "score": 73.0, "reference_url": "https://livebench.ai/", "audit_status": "verified", "source_type": "leaderboard", "matches_canonical": true, "reported_setting": { "mode": "thinking", "effort": "high", "tools": "n/a" }, "notes": "livebench.ai table_2026_01_08 (latest as of 2026-04-28): gemini-3-flash-preview-high avg=73.0. Canonical thinking mode matches (high = default thinking for Gemini 3 Flash per canonical notes). BP had 72.4." }, { "model_id": "gemini-3-flash", "benchmark_id": "matharena_apex_2025", "score": 15.62, "reference_url": "https://matharena.ai/?comp=apex--apex_2025", "audit_status": "verified", "matches_canonical": true, "reported_setting": "thinking:default", "source_type": "leaderboard" }, { "model_id": "gemini-3-flash", "benchmark_id": "smt_2025", "score": 92.92, "reference_url": "https://matharena.ai/?comp=smt--smt_2025", "audit_status": "verified", "matches_canonical": true, "reported_setting": "thinking:default", "source_type": "leaderboard" }, { "model_id": "gemini-3-flash", "benchmark_id": "tau_bench_retail", "score": 82, "reference_url": "https://artificialanalysis.ai/articles/gemini-3-flash-everything-you-need-to-know", "audit_status": "needs_review", "notes": "Reference URL is AA article page (gemini-3-flash-everything-you-need-to-know). AA only publishes aggregate tau2 scores; no retail-domain breakdown available in AA data. gemini-3-flash tau2=N/A in AA (reasoning variant has other tau data). Cannot verify 82 from AA source." }, { "model_id": "gemini-3.1-pro", "benchmark_id": "arc_agi_2", "score": 77.1, "reference_url": "https://deepmind.google/models/gemini/pro/", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "DeepMind /models/gemini/pro/ Performance table: ARC-AGI-2 ARC Prize Verified 77.1%", "candidates": [ { "score": 77.1, "reference_url": "https://openai.com/index/introducing-gpt-5-5/", "source_type": "third_party", "reported_setting": { "note": "see primary" }, "notes": "OpenAI GPT-5.5 blog reports 77.1." }, { "score": 77.1, "reference_url": "https://techcrunch.com/2026/02/19/googles-new-gemini-pro-model-has-record-benchmark-scores-again/", "source_type": "third_party", "reported_setting": {}, "notes": "Demoted from primary on https://deepmind.google/models/gemini/pro/ audit. Original notes: " }, { "score": 77.1, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2.1/Seed2_1_Model_Card.pdf", "source_type": "tech_report", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "none", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official", "prompt_style": "official default", "temperature": "unknown" }, "notes": "Model card Table 11 (p53)." } ] }, { "model_id": "gemini-3.1-pro", "benchmark_id": "gpqa_diamond", "score": 94.3, "reference_url": "https://deepmind.google/models/gemini/pro/", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "DeepMind /models/gemini/pro/ Performance table: GPQA Diamond no tools 94.3%", "candidates": [] }, { "model_id": "gemini-3.1-pro", "benchmark_id": "hle", "score": 44.4, "reference_url": "https://deepmind.google/models/gemini/pro/", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "DeepMind /models/gemini/pro/ Performance table: HLE no tools 44.4%", "candidates": [ { "score": 45.4, "reference_url": "https://ai.meta.com/static-resource/muse-spark-1.1-evaluation-report", "source_type": "tech_report", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1 unless source row states otherwise", "harness": "Meta MSL general-capability evaluation", "temperature": "1.0 for Muse; provider/source setting for comparisons", "context": "benchmark-specific", "configuration": "default" }, "notes": "Figure 44 · HLE no tools; metric=headline_metric. Exact printed value in the general-capability summary." }, { "score": 44.4, "reference_url": "https://deepmind.google/models/gemini/pro/", "source_type": "official_blog", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official Google Gemini evaluation", "prompt_style": "provider official evaluation prompt", "temperature": "source/provider setting", "context": "source/provider setting", "input_modalities": "text", "trials": "source-reported" }, "notes": "Semantic alternative printed inside the same physical HLE cell; it does not add a physical position. Exact provider-official score and reported setting." } ] }, { "model_id": "gemini-3.1-pro", "benchmark_id": "swe_bench_verified", "score": 80.6, "reference_url": "https://deepmind.google/models/gemini/pro/", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "DeepMind /models/gemini/pro/ Performance table: SWE-Bench Verified single attempt 80.6%", "candidates": [ { "score": 80.6, "reference_url": "https://www.digitalapplied.com/blog/google-gemini-3-1-pro-benchmarks-pricing-guide", "source_type": "third_party", "reported_setting": {}, "notes": "Demoted from primary on https://deepmind.google/models/gemini/pro/ audit. Original notes: " } ] }, { "model_id": "gemini-3.1-pro", "benchmark_id": "scicode", "score": 59.0, "reference_url": "https://deepmind.google/models/gemini/pro/", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "DeepMind /models/gemini/pro/ Performance table: SciCode 59% (rounded; existing 58.9 within rounding)", "candidates": [ { "score": 58.9, "reference_url": "https://artificialanalysis.ai/evaluations/scicode", "source_type": "third_party", "reported_setting": {}, "notes": "Demoted from primary on https://deepmind.google/models/gemini/pro/ audit. Original notes: " }, { "score": 58.9, "reference_url": "https://huggingface.co/moonshotai/Kimi-K2.6", "source_type": "model_card", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "1.0" }, "notes": "Kimi K2.6 model card: gemini-3.1-pro=58.9." }, { "score": 54.44, "reference_url": "https://huggingface.co/internlm/Intern-S2-Preview-397B", "source_type": "model_card", "reported_setting": { "mode": "source does not state", "effort": "source does not state", "tools": "code execution", "sampling": "source does not state", "judge": "benchmark-specific; exact version not stated", "harness": "AgentCompass / benchmark evaluator", "prompt_style": "source does not state", "temperature": "source does not state", "context": "source does not state", "notes": "Shanghai AI Lab evaluated all model columns using OpenCompass, VLMEvalKit, and AgentCompass. Per-model mode/effort/sampling are not reported; only Intern-S2 maximum evaluation lengths are disclosed." }, "notes": "Intern-S2-Preview-397B official model card: SciCode = 54.44. This is a Shanghai AI Lab cross-model evaluation; the source does not disclose this model's exact mode, effort, or sampling." }, { "score": 62.3, "reference_url": "https://seed.bytedance.com/en/blog/seed2-1-officially-released-advancing-ai-productivity", "source_type": "official_blog", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "none", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official", "prompt_style": "official default", "temperature": "unknown" }, "notes": "Release blog EN: Language/search chart (https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/user-upload/4xfa4mqq510uc.png)." } ] }, { "model_id": "gemini-3.1-pro", "benchmark_id": "aime_2025", "score": 100.0, "reference_url": "https://automatio.ai/models/gemini-3-1-pro", "audit_status": "dropped", "rule_id": "R5h-third-party-blog", "notes": "DROPPED (R5h): random third-party blog/aggregator, no primary trail" }, { "model_id": "gemini-3.1-pro", "benchmark_id": "mmlu_pro", "score": 91.0, "reference_url": "https://huggingface.co/deepseek-ai/DeepSeek-V4-Pro", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "DeepSeek V4-Pro model card (third-party): gemini-3.1-pro (per their own canonical config) = 91.0.", "candidates": [ { "score": 87.7, "reference_url": "https://ernie.baidu.com/blog/posts/ernie-5.1-0508-release/benchmark.png", "source_type": "official_blog", "reported_setting": { "metric": "accuracy" }, "notes": "Accuracy row matches MMLU-Pro; source does not disclose tools. Research observation obs-028." } ] }, { "model_id": "gemini-3.1-pro", "benchmark_id": "mmmu_pro", "score": 80.5, "reference_url": "https://deepmind.google/models/gemini/pro/", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "DeepMind /models/gemini/pro/ Performance table: MMMU-Pro no tools 80.5%", "candidates": [ { "score": 80.5, "reference_url": "https://openai.com/index/introducing-gpt-5-5/", "source_type": "third_party", "reported_setting": { "note": "see primary" }, "notes": "OpenAI GPT-5.5 blog reports 80.5 (no tools)." }, { "score": 80.5, "reference_url": "https://www.trendingtopics.eu/gemini-3-1-pro-leads-most-benchmarks/", "source_type": "third_party", "reported_setting": {}, "notes": "Demoted from primary on https://deepmind.google/models/gemini/pro/ audit. Original notes: " }, { "score": 83.0, "reference_url": "https://huggingface.co/moonshotai/Kimi-K2.6", "source_type": "model_card", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "1.0" }, "notes": "Kimi K2.6 model card: gemini-3.1-pro=83.0." }, { "score": 83.99, "reference_url": "https://huggingface.co/internlm/Intern-S2-Preview-397B", "source_type": "model_card", "reported_setting": { "mode": "source does not state", "effort": "source does not state", "tools": "none", "sampling": "source does not state", "judge": "benchmark-specific; exact version not stated", "harness": "VLMEvalKit or AgentCompass; exact adapter not stated", "prompt_style": "source does not state", "temperature": "source does not state", "context": "source does not state", "notes": "Shanghai AI Lab evaluated all model columns using OpenCompass, VLMEvalKit, and AgentCompass. Per-model mode/effort/sampling are not reported; only Intern-S2 maximum evaluation lengths are disclosed." }, "notes": "Intern-S2-Preview-397B official model card: MMMU Pro = 83.99. This is a Shanghai AI Lab cross-model evaluation; the source does not disclose this model's exact mode, effort, or sampling." }, { "score": 80.5, "reference_url": "https://seed.bytedance.com/en/blog/seed2-1-officially-released-advancing-ai-productivity", "source_type": "official_blog", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "none", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official", "prompt_style": "official default", "temperature": "unknown" }, "notes": "Release blog EN: Vision capability chart (https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/user-upload/4xfa4mqq4yer3.png)." } ] }, { "model_id": "gemini-3.1-pro", "benchmark_id": "mathvision", "score": 89.8, "reference_url": "https://huggingface.co/moonshotai/Kimi-K2.6", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "1.0" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Kimi K2.6 model card: gemini-3.1-pro=89.8.", "candidates": [ { "score": 89.2, "reference_url": "https://seed.bytedance.com/en/blog/seed2-1-officially-released-advancing-ai-productivity", "source_type": "official_blog", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "none", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official", "prompt_style": "official default", "temperature": "unknown" }, "notes": "Release blog EN: Vision capability chart (https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/user-upload/4xfa4mqq4yer3.png)." } ] }, { "model_id": "gemini-3.1-pro", "benchmark_id": "mrcr_v2", "score": 84.9, "reference_url": "https://deepmind.google/models/gemini/pro/", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default", "context": "128k" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "DeepMind /models/gemini/pro/ Performance table: MRCR v2 8-needle 128k average 84.9%", "candidates": [ { "score": 84.9, "reference_url": "https://smartscope.blog/en/generative-ai/google-gemini/gemini-3-1-pro-benchmark-analysis-2026/", "source_type": "third_party", "reported_setting": {}, "notes": "Demoted from primary on https://deepmind.google/models/gemini/pro/ audit. Original notes: " }, { "score": 26.3, "reference_url": "https://deepmind.google/models/gemini/pro/", "source_type": "official_blog", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default", "context": "1m" }, "notes": "DeepMind /models/gemini/pro/ table: MRCR v2 1M pointwise 26.3% (3.1 Pro)." } ] }, { "model_id": "gemini-3.1-pro", "benchmark_id": "simpleqa", "score": 72.1, "reference_url": "https://officechai.com/ai/gemini-3-1-pro-benchmarks/", "audit_status": "dropped", "rule_id": "R5h-third-party-blog", "notes": "DROPPED (R5h): random third-party blog/aggregator, no primary trail" }, { "model_id": "gemini-3.1-pro", "benchmark_id": "ifeval", "score": 96.1, "reference_url": "https://longcat.chat/blog/longcat-2.0/", "reported_setting": { "mode": "thinking", "effort": "source does not state", "tools": "none", "sampling": "source does not state", "judge": "benchmark-specified", "harness": "Meituan unified in-house harness", "prompt_style": "default", "temperature": "0.7; top_k=-1; top_p=0.95", "context": "source does not state", "notes": "Measured in-house by Meituan unless otherwise noted." }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "LongCat-2.0 official tech blog: IFEval = 96.1. Measured in-house by Meituan under the reported unified harness." }, { "model_id": "gemini-3.1-pro", "benchmark_id": "critpt", "score": 17.7, "reference_url": "https://artificialanalysis.ai/evaluations/critpt", "audit_status": "verified", "source_type": "third_party", "notes": "ArtificialAnalysis aggregator; used as fallback when no primary source available (R5d-aggregator-fallback). AA CritPT eval page: gemini-3-1-pro-preview critpt=0.177 → 17.7%. AA slug: gemini-3-1-pro-preview.", "reported_setting": { "mode": "thinking", "effort": "default", "sampling": "pass@1", "harness": "AA standard evaluation", "notes": "Per AA evaluations/critpt: gemini-3-1-pro-preview critpt=17.7%" } }, { "model_id": "gemini-3.1-pro", "benchmark_id": "aa_intelligence_index", "score": 57, "reference_url": "https://artificialanalysis.ai/evaluations/artificial-analysis-intelligence-index", "audit_status": "verified", "source_type": "third_party", "notes": "ArtificialAnalysis aggregator; used as fallback when no primary source available (R5d-aggregator-fallback). AA Intelligence Index: gemini-3-1-pro-preview = 57.18 (displayed as 57). AA slug: gemini-3-1-pro-preview.", "reported_setting": { "mode": "thinking", "effort": "default", "harness": "AA Intelligence Index composite", "notes": "AA slug: gemini-3-1-pro-preview; II=57.18 rounds to 57" } }, { "model_id": "gemini-3.1-pro", "benchmark_id": "mmlu", "score": 92.6, "reference_url": "https://deepmind.google/models/model-cards/gemini-3-1-pro/", "audit_status": "dropped", "notes": " | DROPPED: BP cell 92.6 from /models/model-cards/gemini-3-1-pro/ is actually MMMLU not MMLU. Source mislabel. Use mmmlu cell instead." }, { "model_id": "gemini-3.1-pro", "benchmark_id": "browsecomp", "score": 85.9, "reference_url": "https://deepmind.google/models/gemini/pro/", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "web+code", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "DeepMind /models/gemini/pro/ Performance table: BrowseComp Search+Python+Browse 85.9%", "candidates": [ { "score": 85.9, "reference_url": "https://openai.com/index/introducing-gpt-5-5/", "source_type": "third_party", "reported_setting": { "note": "see primary" }, "notes": "OpenAI GPT-5.5 blog reports 85.9." }, { "score": 85.9, "reference_url": "https://deepmind.google/models/model-cards/gemini-3-1-pro/", "source_type": "third_party", "reported_setting": {}, "notes": "Demoted from primary on https://deepmind.google/models/gemini/pro/ audit. Original notes: " }, { "score": 85.8, "reference_url": "https://openai.com/index/gpt-5-6/", "source_type": "official_blog", "reported_setting": { "effort": "reference", "eval_variant": null, "code_mode": null }, "notes": "OpenAI GPT-5.6 release figure (BrowseComp): Gemini 3.1 Pro Preview; BrowseComp=85.8. Figure 15: Agentic browsing tasks.. Cost, latency, and output-token coordinates are retained as review metadata, not score-matrix metrics." }, { "score": 85.9, "reference_url": "https://seed.bytedance.com/en/blog/seed2-1-officially-released-advancing-ai-productivity", "source_type": "official_blog", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "search", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official", "prompt_style": "official default", "temperature": "unknown" }, "notes": "Release blog EN: Language/search chart (https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/user-upload/4xfa4mqq510uc.png)." } ] }, { "model_id": "gemini-3.1-pro", "benchmark_id": "frontiermath", "score": 36.9, "reference_url": "https://epoch.ai/benchmarks/frontiermath", "candidates": [ { "score": 36.9, "reference_url": "https://openai.com/index/introducing-gpt-5-5/", "source_type": "third_party", "reported_setting": { "note": "see primary" }, "notes": "OpenAI GPT-5.5 blog reports 36.9 (Tier 1-3)." } ], "audit_status": "verified", "source_type": "leaderboard", "matches_canonical": false, "reported_setting": { "mode": "thinking", "effort": "n/a", "tools": "none" }, "notes": "epoch.ai FrontierMath-2025-02-28-Private: gemini-3.1-pro-preview (Gemini 3.1 Pro Preview) = 36.9%. Only one variant available; canonical effort=high not confirmed. BP had 40.0." }, { "model_id": "gemini-3.1-pro", "benchmark_id": "chatbot_arena_elo", "score": 1493, "reference_url": "https://lmarena.ai/leaderboard/text", "audit_status": "verified", "source_type": "leaderboard", "matches_canonical": true, "notes": "lmarena.ai text arena (fetched 2026-04-29). Arena model name: gemini-3.1-pro-preview (rank 5). ELO updates continuously; score reflects latest available.", "candidates": [ { "score": 1493, "reference_url": "https://ernie.baidu.com/blog/posts/ernie-5.1-0508-release/creative-capability.png", "source_type": "official_blog", "reported_setting": { "leaderboard": "Arena AI Text", "metric": "Arena score", "style_control": "on", "snapshot_date": "2026-04-30" }, "notes": "April 30, 2026 Text Arena snapshot; style control on. Research observation obs-052. Displayed rank 5. Identical asset is published in four page placements. The same PNG bytes are published under two filenames and four English/Chinese page placements." } ] }, { "model_id": "gemini-3.1-pro", "benchmark_id": "osworld", "score": 72.0, "reference_url": "https://www.trendingtopics.eu/gemini-3-1-pro-leads-most-benchmarks/", "audit_status": "dropped", "rule_id": "R5h-third-party-blog", "notes": "DROPPED (R5h): random third-party blog/aggregator, no primary trail", "candidates": [ { "score": 76.2, "reference_url": "https://seed.bytedance.com/en/blog/seed2-1-officially-released-advancing-ai-productivity", "source_type": "official_blog", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "GUI + shell/commands + filesystem tools", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official", "prompt_style": "official default", "temperature": "unknown" }, "notes": "Release blog EN: Computer-use chart (https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/user-upload/4xfa4mqq4w8x3.png)." } ] }, { "model_id": "gemini-3.1-pro", "benchmark_id": "arc_agi_1", "score": 98.0, "reference_url": "https://arcprize.org/leaderboard", "candidates": [ { "score": 98.0, "reference_url": "https://openai.com/index/introducing-gpt-5-5/", "source_type": "third_party", "reported_setting": { "note": "see primary" }, "notes": "OpenAI GPT-5.5 blog reports 98.0." } ], "audit_status": "verified", "source_type": "leaderboard", "matches_canonical": true, "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1 (single-pass)", "judge": "rule-based", "harness": "official (arcprize.org)", "system_type": "CoT" }, "notes": "arcprize.org leaderboard audit: Gemini 3.1 Pro (Preview, 2026-02-19) on leaderboard" }, { "model_id": "gemini-3.1-pro", "benchmark_id": "codeforces_rating", "score": 3052, "reference_url": "https://huggingface.co/deepseek-ai/DeepSeek-V4-Pro", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "DeepSeek V4-Pro model card (third-party): gemini-3.1-pro (per their own canonical config) = 3052." }, { "model_id": "gemini-3.1-pro", "benchmark_id": "humaneval", "score": 95.0, "reference_url": "https://www.digitalapplied.com/blog/google-gemini-3-1-pro-benchmarks-pricing-guide", "audit_status": "dropped", "rule_id": "R5h-third-party-blog", "notes": "DROPPED (R5h): random third-party blog/aggregator, no primary trail" }, { "model_id": "gemini-3.1-pro", "benchmark_id": "livecodebench", "score": 91.7, "reference_url": "https://huggingface.co/deepseek-ai/DeepSeek-V4-Pro", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "DeepSeek V4-Pro model card (third-party): gemini-3.1-pro (per their own canonical config) = 91.7." }, { "model_id": "gemini-3.1-pro", "benchmark_id": "aime_2024", "score": 98, "reference_url": "https://deepmind.google/models/model-cards/gemini-3-1-pro/", "audit_status": "dropped", "rule_id": "R5g-ghost", "audit_note": "deepmind.google/models/model-cards/gemini-3-1-pro/ has no AIME 2024 row. Card covers HLE, ARC-AGI-2, GPQA-D, SWE-Bench, etc.", "audited_by": "lychee-audit-bot", "audited_at": "2026-04-29T13:06:40Z" }, { "model_id": "gemini-3.1-pro", "benchmark_id": "aime_2026", "score": 98.2, "reference_url": "https://z.ai/blog/glm-5.1", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "GLM-5.1 blog: gemini-3.1-pro=98.2.", "candidates": [ { "score": 98.3, "reference_url": "https://huggingface.co/moonshotai/Kimi-K2.6", "source_type": "model_card", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "1.0" }, "notes": "Kimi K2.6 model card: gemini-3.1-pro=98.3." }, { "score": 99.9, "reference_url": "https://ernie.baidu.com/blog/posts/ernie-5.1-0508-release/benchmark.png", "source_type": "official_blog", "reported_setting": { "tools": "tools enabled (unspecified)", "sampling": "avg@32", "metric": "score" }, "notes": "With-tool avg@32 value is noncanonical against pass@1/no-tools. Research observation obs-004." }, { "score": 98.3, "reference_url": "https://ernie.baidu.com/blog/posts/ernie-5.1-0508-release/benchmark.png", "source_type": "official_blog", "reported_setting": { "tools": "none implied by contrast with w/tools row", "sampling": "avg@32" }, "notes": "No-tool avg@32 value is noncanonical against pass@1. Research observation obs-020. Literal marker ▲ has no source legend. Marker has no legend in image or article." } ] }, { "model_id": "gemini-3.1-pro", "benchmark_id": "math_500", "score": 98.5, "reference_url": "https://www.digitalapplied.com/blog/google-gemini-3-1-pro-benchmarks-pricing-guide", "audit_status": "dropped", "rule_id": "R5h-third-party-blog", "notes": "DROPPED (R5h): random third-party blog/aggregator, no primary trail" }, { "model_id": "gemini-3.1-pro", "benchmark_id": "matharena_apex_2025", "score": 60.9, "reference_url": "https://huggingface.co/deepseek-ai/DeepSeek-V4-Pro", "audit_status": "verified", "notes": "DeepSeek V4-Pro model card (third-party): gemini-3.1-pro (per their own canonical config) = 60.9.", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "candidates": [ { "score": 58.9, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2.1/Seed2_1_Model_Card.pdf", "source_type": "tech_report", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "none", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official", "prompt_style": "official default", "temperature": "unknown" }, "notes": "Model card Table 11 (p53). Source dash is not zero." } ] }, { "model_id": "gemini-3.1-pro", "benchmark_id": "mmmu", "score": 87.5, "reference_url": "https://deepmind.google/models/model-cards/gemini-3-1-pro/", "audit_status": "dropped", "rule_id": "R5g-ghost", "audit_note": "deepmind.google/models/model-cards/gemini-3-1-pro/ has no standard MMMU row. Card shows MMMU-Pro=80.5%, not 87.5. Benchmark mismatch.", "audited_by": "lychee-audit-bot", "audited_at": "2026-04-29T13:06:40Z" }, { "model_id": "gemini-3.1-pro", "benchmark_id": "simplebench", "score": 79.6, "reference_url": "https://lmcouncil.ai/benchmarks", "audit_status": "dropped", "rule_id": "R5h-third-party-blog", "notes": "DROPPED (R5h): random third-party blog/aggregator, no primary trail | DROPPED (R5g-c stale, never verified, source not audited in any pass)" }, { "model_id": "gemini-3.1-pro", "benchmark_id": "swe_bench_pro", "score": 54.2, "reference_url": "https://deepmind.google/models/gemini/pro/", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "DeepMind /models/gemini/pro/ Performance table: SWE-Bench Pro single attempt 54.2%", "candidates": [ { "score": 54.2, "reference_url": "https://seed.bytedance.com/en/blog/seed2-1-officially-released-advancing-ai-productivity", "source_type": "official_blog", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "Trae or benchmark-specific coding-agent environment", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official", "prompt_style": "official default", "temperature": "unknown" }, "notes": "Release blog EN: Coding chart (https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/user-upload/4xfa4mqq4wlbt.png)." }, { "score": 54.2, "reference_url": "https://deepmind.google/models/gemini/pro/", "source_type": "official_blog", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "source-reported agentic tools", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official Google Gemini evaluation", "prompt_style": "provider official evaluation prompt", "temperature": "source/provider setting", "context": "source/provider setting", "input_modalities": "text", "trials": "source-reported" }, "notes": "Exact provider-official score and reported setting." } ] }, { "model_id": "gemini-3.1-pro", "benchmark_id": "tau_bench_retail", "score": 90.5, "reference_url": "https://deepmind.google/models/model-cards/gemini-3-1-pro/", "audit_status": "dropped", "notes": " | DROPPED: BP cell 90.5 from /models/model-cards/gemini-3-1-pro/ is actually τ²-bench-retail not τ-bench-retail. Use tau2_bench_retail (90.8)." }, { "model_id": "gemini-3.1-pro", "benchmark_id": "terminal_bench", "score": 68.5, "reference_url": "https://deepmind.google/models/gemini/pro/", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "Terminus-2", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "DeepMind /models/gemini/pro/ Performance table: Terminal-Bench 2.0 Terminus-2 68.5%", "candidates": [ { "score": 68.5, "reference_url": "https://openai.com/index/introducing-gpt-5-5/", "source_type": "third_party", "reported_setting": { "note": "see primary" }, "notes": "OpenAI GPT-5.5 blog reports 68.5." }, { "score": 68.5, "reference_url": "https://www.digitalapplied.com/blog/google-gemini-3-1-pro-benchmarks-pricing-guide", "source_type": "third_party", "reported_setting": {}, "notes": "Demoted from primary on https://deepmind.google/models/gemini/pro/ audit. Original notes: " }, { "score": 68.5, "reference_url": "https://deepmind.google/models/gemini/pro/", "source_type": "official_blog", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "source-reported agentic tools", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "Terminus-2", "prompt_style": "provider official evaluation prompt", "temperature": "source/provider setting", "context": "source/provider setting", "input_modalities": "text", "trials": "source-reported" }, "notes": "Exact provider-official score and reported setting." } ] }, { "model_id": "llama-4-scout", "benchmark_id": "mmlu_pro", "score": 74.3, "reference_url": "https://huggingface.co/meta-llama/Llama-4-Scout-17B-16E-Instruct", "audit_status": "verified", "source_type": "model_card", "reported_setting": { "prompt_style": "5-shot pre-train", "judge": "rule-based", "harness": "official-meta", "sampling": "pass@1", "temperature": "0.0" }, "matches_canonical": false, "notes": "Instruction-tuned 0-shot MMLU-Pro per HF/GitHub model card and Meta blog comparison table." }, { "model_id": "llama-4-scout", "benchmark_id": "gpqa_diamond", "score": 57.2, "reference_url": "https://huggingface.co/meta-llama/Llama-4-Scout-17B-16E-Instruct", "audit_status": "verified", "source_type": "model_card", "reported_setting": { "prompt_style": "zeroshot", "judge": "rule-based", "harness": "official-meta", "sampling": "pass@1", "temperature": "0.0" }, "matches_canonical": true, "notes": "Instruction-tuned GPQA Diamond per HF/GitHub model card and Meta blog comparison table." }, { "model_id": "llama-4-scout", "benchmark_id": "mmmu", "score": 69.4, "reference_url": "https://huggingface.co/meta-llama/Llama-4-Scout-17B-16E-Instruct", "audit_status": "verified", "source_type": "model_card", "reported_setting": { "prompt_style": "zeroshot", "judge": "rule-based", "harness": "official-meta", "sampling": "pass@1", "temperature": "0.0" }, "matches_canonical": true, "notes": "Instruction-tuned MMMU per HF/GitHub model card and Meta blog comparison table." }, { "model_id": "llama-4-scout", "benchmark_id": "livecodebench", "score": 32.8, "reference_url": "https://huggingface.co/meta-llama/Llama-4-Scout-17B-16E-Instruct", "audit_status": "verified", "source_type": "model_card", "reported_setting": { "prompt_style": "zeroshot", "judge": "rule-based", "harness": "official-meta", "sampling": "pass@1", "temperature": "0.0" }, "matches_canonical": true, "notes": "Instruction-tuned LiveCodeBench (10/01/2024-02/01/2025) per HF/GitHub model card and Meta blog comparison table." }, { "model_id": "llama-4-scout", "benchmark_id": "chatbot_arena_elo", "score": 1350, "reference_url": "", "audit_status": "dropped", "source_type": "", "reported_setting": {}, "matches_canonical": false, "notes": "Scout Arena Elo not in Meta blog text (only Maverick 1417 mentioned). Need LMSYS leaderboard source. | RESOLVED: dropped after audit — original ref was bogus blog URL, secondary blog, or unverifiable JS-rendered Meta product page." }, { "model_id": "llama-4-scout", "benchmark_id": "gsm8k", "score": 90.6, "reference_url": "", "audit_status": "dropped", "source_type": "", "reported_setting": {}, "matches_canonical": false, "notes": "Meta has not published Scout GSM8K. The 90.6 in BP equals MGSM 90.6 from instruction-tuned table — mislabel." }, { "model_id": "llama-4-scout", "benchmark_id": "humaneval", "score": 76.0, "reference_url": "", "audit_status": "dropped", "source_type": "", "reported_setting": {}, "matches_canonical": false, "notes": "Not reported by Meta in blog/MC for Scout. Need third-party source or drop. | RESOLVED: dropped after audit — original ref was bogus blog URL, secondary blog, or unverifiable JS-rendered Meta product page." }, { "model_id": "llama-4-scout", "benchmark_id": "ifeval", "score": 81.0, "reference_url": "", "audit_status": "dropped", "source_type": "", "reported_setting": {}, "matches_canonical": false, "notes": "Not reported by Meta in blog/MC for Scout. Need third-party source or drop. | RESOLVED: dropped after audit — original ref was bogus blog URL, secondary blog, or unverifiable JS-rendered Meta product page." }, { "model_id": "llama-4-scout", "benchmark_id": "arc_agi_1", "score": 0.5, "reference_url": "", "audit_status": "verified", "source_type": "third_party", "reported_setting": {}, "matches_canonical": false, "notes": "ARC-AGI-1 is third-party leaderboard." }, { "model_id": "llama-4-scout", "benchmark_id": "arc_agi_2", "score": 0, "reference_url": "", "audit_status": "verified", "source_type": "third_party", "reported_setting": {}, "matches_canonical": false, "notes": "ARC-AGI-2 is third-party leaderboard." }, { "model_id": "llama-4-scout", "benchmark_id": "arena_hard", "score": 72, "reference_url": "", "audit_status": "dropped", "source_type": "", "reported_setting": {}, "matches_canonical": false, "notes": "Arena Hard not in Meta blog text or PNG comparison tables for Scout. Need LMSYS/source. | RESOLVED: dropped after audit — original ref was bogus blog URL, secondary blog, or unverifiable JS-rendered Meta product page." }, { "model_id": "llama-4-scout", "benchmark_id": "math_500", "score": 83, "reference_url": "", "audit_status": "dropped", "source_type": "", "reported_setting": {}, "matches_canonical": false, "notes": "Not in Meta blog, HF model card, or GitHub MODEL_CARD. Need third-party source or drop. | RESOLVED: dropped after audit — original ref was bogus blog URL, secondary blog, or unverifiable JS-rendered Meta product page." }, { "model_id": "llama-4-scout", "benchmark_id": "mmlu", "score": 79.6, "reference_url": "https://huggingface.co/meta-llama/Llama-4-Scout-17B-16E-Instruct", "audit_status": "verified", "source_type": "model_card", "reported_setting": { "prompt_style": "5-shot pre-train", "judge": "rule-based", "harness": "official-meta", "sampling": "pass@1", "temperature": "0.0" }, "matches_canonical": false, "notes": "Pre-train 5-shot MMLU per HF/GitHub model card. BP previously had 88.5 (likely confused with MMLU-Pro)." }, { "model_id": "llama-4-maverick", "benchmark_id": "mmlu_pro", "score": 80.5, "reference_url": "https://huggingface.co/meta-llama/Llama-4-Maverick-17B-128E-Instruct", "audit_status": "verified", "source_type": "model_card", "reported_setting": { "prompt_style": "zeroshot", "judge": "rule-based", "harness": "official-meta", "sampling": "pass@1", "temperature": "0.0" }, "matches_canonical": true, "notes": "Instruction-tuned 0-shot MMLU-Pro per HF/GitHub model card and Meta blog comparison table.", "candidates": [ { "score": 80.4, "reference_url": "https://mistral.ai/news/mistral-medium-3", "source_type": "official_blog", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "mistral-eval", "prompt_style": "5-shot CoT", "temperature": "0.0" }, "notes": "Mistral blog (third-party self-test) — 80.4 vs current verified 80.5." } ] }, { "model_id": "llama-4-maverick", "benchmark_id": "gpqa_diamond", "score": 69.8, "reference_url": "https://huggingface.co/meta-llama/Llama-4-Maverick-17B-128E-Instruct", "audit_status": "verified", "source_type": "model_card", "reported_setting": { "prompt_style": "zeroshot", "judge": "rule-based", "harness": "official-meta", "sampling": "pass@1", "temperature": "0.0" }, "matches_canonical": true, "notes": "Instruction-tuned GPQA Diamond per HF/GitHub model card and Meta blog comparison table.", "candidates": [ { "score": 61.1, "reference_url": "https://mistral.ai/news/mistral-medium-3", "source_type": "official_blog", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "mistral-eval", "prompt_style": "5-shot CoT", "temperature": "0.0" }, "notes": "Mistral blog (third-party self-test) — 61.1 vs current verified 69.8." } ] }, { "model_id": "llama-4-maverick", "benchmark_id": "math_500", "score": 90.0, "reference_url": "https://mistral.ai/news/mistral-medium-3", "audit_status": "verified", "source_type": "official_blog", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "mistral-eval", "prompt_style": "0-shot", "temperature": "0.0" }, "matches_canonical": false, "notes": "Mistral Medium 3 blog table: 90.0. Third-party self-test by Mistral (their internal eval pipeline)." }, { "model_id": "llama-4-maverick", "benchmark_id": "mmmu", "score": 73.4, "reference_url": "https://huggingface.co/meta-llama/Llama-4-Maverick-17B-128E-Instruct", "audit_status": "verified", "source_type": "model_card", "reported_setting": { "prompt_style": "zeroshot", "judge": "rule-based", "harness": "official-meta", "sampling": "pass@1", "temperature": "0.0" }, "matches_canonical": true, "notes": "Instruction-tuned MMMU per HF/GitHub model card and Meta blog comparison table.", "candidates": [ { "score": 71.8, "reference_url": "https://mistral.ai/news/mistral-medium-3", "source_type": "official_blog", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "mistral-eval", "prompt_style": "0-shot", "temperature": "0.0" }, "notes": "Mistral blog (third-party self-test) — 71.8 vs current verified 73.4." } ] }, { "model_id": "llama-4-maverick", "benchmark_id": "livecodebench", "score": 43.4, "reference_url": "https://huggingface.co/meta-llama/Llama-4-Maverick-17B-128E-Instruct", "audit_status": "verified", "source_type": "model_card", "reported_setting": { "prompt_style": "zeroshot", "judge": "rule-based", "harness": "official-meta", "sampling": "pass@1", "temperature": "0.0" }, "matches_canonical": true, "notes": "Instruction-tuned LiveCodeBench (10/01/2024-02/01/2025) per HF/GitHub model card and Meta blog comparison table." }, { "model_id": "llama-4-maverick", "benchmark_id": "chatbot_arena_elo", "score": 1417, "reference_url": "https://ai.meta.com/blog/llama-4-multimodal-intelligence/", "audit_status": "verified", "source_type": "official_blog", "reported_setting": { "prompt_style": "zeroshot", "judge": "rule-based", "harness": "official-meta", "sampling": "pass@1", "temperature": "0.0" }, "matches_canonical": true, "notes": "Meta blog text: Maverick 'scores ELO of 1417 on LMArena'." }, { "model_id": "llama-4-maverick", "benchmark_id": "arena_hard", "score": 91.8, "reference_url": "https://mistral.ai/news/mistral-medium-3", "audit_status": "verified", "source_type": "official_blog", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "mistral-eval", "prompt_style": "0-shot", "temperature": "0.0" }, "matches_canonical": false, "notes": "Mistral Medium 3 blog table: 91.8. Third-party self-test by Mistral (their internal eval pipeline)." }, { "model_id": "llama-4-maverick", "benchmark_id": "humaneval", "score": 85.4, "reference_url": "https://mistral.ai/news/mistral-medium-3", "audit_status": "verified", "source_type": "official_blog", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "mistral-eval", "prompt_style": "0-shot", "temperature": "0.0" }, "matches_canonical": false, "notes": "Mistral Medium 3 blog table: 85.4. Third-party self-test by Mistral (their internal eval pipeline)." }, { "model_id": "llama-4-maverick", "benchmark_id": "ifeval", "score": 88.9, "reference_url": "https://mistral.ai/news/mistral-medium-3", "audit_status": "verified", "source_type": "official_blog", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "mistral-eval", "prompt_style": "0-shot", "temperature": "0.0" }, "matches_canonical": false, "notes": "Mistral Medium 3 blog table: 88.9. Third-party self-test by Mistral (their internal eval pipeline)." }, { "model_id": "llama-4-maverick", "benchmark_id": "simpleqa", "score": 20.0, "reference_url": "", "audit_status": "dropped", "source_type": "", "reported_setting": {}, "matches_canonical": false, "notes": "SimpleQA not in Meta blog/MC for Maverick. Need third-party source or drop. | RESOLVED: dropped after audit — original ref was bogus blog URL, secondary blog, or unverifiable JS-rendered Meta product page." }, { "model_id": "llama-4-maverick", "benchmark_id": "aime_2024", "score": 52, "reference_url": "", "audit_status": "dropped", "source_type": "", "reported_setting": {}, "matches_canonical": false, "notes": "AIME 2024 not in Meta blog/MC for Maverick. Need third-party source or drop. | RESOLVED: dropped after audit — original ref was bogus blog URL, secondary blog, or unverifiable JS-rendered Meta product page." }, { "model_id": "llama-4-maverick", "benchmark_id": "arc_agi_1", "score": 4.4, "reference_url": "", "audit_status": "verified", "source_type": "third_party", "reported_setting": {}, "matches_canonical": false, "notes": "ARC-AGI-1 third-party leaderboard." }, { "model_id": "llama-4-maverick", "benchmark_id": "arc_agi_2", "score": 0, "reference_url": "", "audit_status": "verified", "source_type": "third_party", "reported_setting": {}, "matches_canonical": false, "notes": "ARC-AGI-2 third-party leaderboard." }, { "model_id": "llama-4-maverick", "benchmark_id": "bigcodebench", "score": 49.7, "reference_url": "", "audit_status": "verified", "source_type": "third_party", "reported_setting": {}, "matches_canonical": false, "notes": "BigCodeBench third-party leaderboard." }, { "model_id": "llama-4-maverick", "benchmark_id": "critpt", "score": 0, "reference_url": "", "audit_status": "verified", "source_type": "third_party", "reported_setting": {}, "matches_canonical": false, "notes": "CritPt third-party benchmark." }, { "model_id": "llama-4-maverick", "benchmark_id": "mmlu", "score": 85.5, "reference_url": "https://huggingface.co/meta-llama/Llama-4-Maverick-17B-128E-Instruct", "audit_status": "verified", "source_type": "model_card", "reported_setting": { "prompt_style": "5-shot pre-train", "judge": "rule-based", "harness": "official-meta", "sampling": "pass@1", "temperature": "0.0" }, "matches_canonical": false, "notes": "Pre-train 5-shot MMLU per HF/GitHub model card. BP previously had 88.6 (likely confused with MMLU-Pro)." }, { "model_id": "llama-4-maverick", "benchmark_id": "swe_bench_verified", "score": 46.5, "reference_url": "", "audit_status": "verified", "source_type": "third_party", "reported_setting": {}, "matches_canonical": false, "notes": "SWE-bench Verified third-party leaderboard." }, { "model_id": "llama-4-maverick", "benchmark_id": "terminal_bench_1", "score": 15.5, "reference_url": "", "audit_status": "verified", "source_type": "third_party", "reported_setting": {}, "matches_canonical": false, "notes": "Terminal-Bench third-party leaderboard." }, { "model_id": "llama-4-behemoth", "benchmark_id": "math_500", "score": 95.0, "reference_url": "https://ai.meta.com/blog/llama-4-multimodal-intelligence/", "audit_status": "verified", "source_type": "official_blog", "reported_setting": { "prompt_style": "zeroshot", "judge": "rule-based", "harness": "official-meta", "sampling": "pass@1", "temperature": "0.0" }, "matches_canonical": true, "notes": "Meta blog comparison table and blog text: 'on several STEM benchmarks such as MATH-500'." }, { "model_id": "llama-4-behemoth", "benchmark_id": "mmlu_pro", "score": 82.2, "reference_url": "https://ai.meta.com/blog/llama-4-multimodal-intelligence/", "audit_status": "verified", "source_type": "official_blog", "reported_setting": { "prompt_style": "zeroshot", "judge": "rule-based", "harness": "official-meta", "sampling": "pass@1", "temperature": "0.0" }, "matches_canonical": true, "notes": "Meta blog comparison table (Behemoth vs Claude Sonnet 3.7 / Gemini 2.0 Pro / GPT-4.5) and blog text." }, { "model_id": "llama-4-behemoth", "benchmark_id": "gpqa_diamond", "score": 73.7, "reference_url": "https://ai.meta.com/blog/llama-4-multimodal-intelligence/", "audit_status": "verified", "source_type": "official_blog", "reported_setting": { "prompt_style": "zeroshot", "judge": "rule-based", "harness": "official-meta", "sampling": "pass@1", "temperature": "0.0" }, "matches_canonical": true, "notes": "Meta blog comparison table and blog text: 'on several STEM benchmarks such as MATH-500 and GPQA Diamond'." }, { "model_id": "llama-4-behemoth", "benchmark_id": "mmmu", "score": 76.1, "reference_url": "https://ai.meta.com/blog/llama-4-multimodal-intelligence/", "audit_status": "verified", "source_type": "official_blog", "reported_setting": { "prompt_style": "zeroshot", "judge": "rule-based", "harness": "official-meta", "sampling": "pass@1", "temperature": "0.0" }, "matches_canonical": true, "notes": "Meta blog comparison table (Behemoth vs Claude Sonnet 3.7 / Gemini 2.0 Pro / GPT-4.5)." }, { "model_id": "llama-4-behemoth", "benchmark_id": "livecodebench", "score": 49.4, "reference_url": "https://ai.meta.com/blog/llama-4-multimodal-intelligence/", "audit_status": "verified", "source_type": "official_blog", "reported_setting": { "prompt_style": "zeroshot", "judge": "rule-based", "harness": "official-meta", "sampling": "pass@1", "temperature": "0.0" }, "matches_canonical": true, "notes": "Meta blog comparison table (Behemoth vs Claude Sonnet 3.7 / Gemini 2.0 Pro / GPT-4.5)." }, { "model_id": "llama-4-behemoth", "benchmark_id": "humaneval", "score": 85.0, "reference_url": "", "audit_status": "dropped", "source_type": "", "reported_setting": {}, "matches_canonical": false, "notes": "Not in Meta blog text or comparison PNG. Need source or drop. | RESOLVED: dropped after audit — original ref was bogus blog URL, secondary blog, or unverifiable JS-rendered Meta product page." }, { "model_id": "llama-4-behemoth", "benchmark_id": "ifeval", "score": 86.0, "reference_url": "", "audit_status": "dropped", "source_type": "", "reported_setting": {}, "matches_canonical": false, "notes": "Not in Meta blog text or comparison PNG. Need source or drop. | RESOLVED: dropped after audit — original ref was bogus blog URL, secondary blog, or unverifiable JS-rendered Meta product page." }, { "model_id": "llama-4-behemoth", "benchmark_id": "aime_2024", "score": 72, "reference_url": "", "audit_status": "dropped", "source_type": "", "reported_setting": {}, "matches_canonical": false, "notes": "Not in Meta blog text or comparison PNG. Need source or drop. | RESOLVED: dropped after audit — original ref was bogus blog URL, secondary blog, or unverifiable JS-rendered Meta product page." }, { "model_id": "llama-4-behemoth", "benchmark_id": "mmlu", "score": 90.2, "reference_url": "", "audit_status": "dropped", "source_type": "", "reported_setting": {}, "matches_canonical": false, "notes": "Meta has not published Behemoth MMLU in any source (blog text, blog PNG comparison table, no model card released). 90.2 origin unknown." }, { "model_id": "llama-4-behemoth", "benchmark_id": "simpleqa", "score": 44, "reference_url": "", "audit_status": "dropped", "source_type": "", "reported_setting": {}, "matches_canonical": false, "notes": "Not in Meta blog text or comparison PNG. Need source or drop. | RESOLVED: dropped after audit — original ref was bogus blog URL, secondary blog, or unverifiable JS-rendered Meta product page." }, { "model_id": "llama-4-behemoth", "benchmark_id": "swe_bench_verified", "score": 55, "reference_url": "", "audit_status": "verified", "source_type": "third_party", "reported_setting": {}, "matches_canonical": false, "notes": "SWE-bench Verified third-party leaderboard." }, { "model_id": "muse-spark", "benchmark_id": "mmmu_pro", "score": 80.4, "reference_url": "https://ai.meta.com/blog/introducing-muse-spark-msl/", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1 (avg 4 runs)", "judge": "rule-based", "harness": "internal (Meta MSL)", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "rule_ids": [ "R5a" ], "notes": "Meta MSL blog main table: MMMU Pro 80.4, Thinking mode. Avg 4 runs.", "candidates": [] }, { "model_id": "muse-spark", "benchmark_id": "hle", "score": 42.8, "reference_url": "https://ai.meta.com/blog/introducing-muse-spark-msl/", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1 (single-pass)", "judge": "gpt-o3-mini (LLM-as-judge)", "harness": "internal (Meta MSL)", "prompt_style": "default", "temperature": "default", "notes": "No-tools variant. Full 2,500-question HLE dataset." }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "rule_ids": [ "R5a" ], "notes": "Meta MSL blog: HLE (No Tools) 42.8, Thinking mode.", "candidates": [] }, { "model_id": "muse-spark", "benchmark_id": "arc_agi_2", "score": 42.5, "reference_url": "https://ai.meta.com/blog/introducing-muse-spark-msl/", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@2", "judge": "rule-based", "harness": "kaggle/arc-agi2", "prompt_style": "default", "temperature": "default", "notes": "pass@2 to match competition setup; 120-prompt public set" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "rule_ids": [ "R5a" ], "notes": "Meta MSL blog: ARC-AGI-2 42.5, Thinking mode. 120 public-set prompts, pass@2. matches_canonical=false: canonical may be pass@1.", "candidates": [] }, { "model_id": "muse-spark", "benchmark_id": "gpqa_diamond", "score": 89.5, "reference_url": "https://ai.meta.com/blog/introducing-muse-spark-msl/", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "avg of 4 runs", "judge": "rule-based", "harness": "internal (Meta MSL)", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "rule_ids": [ "R5a" ], "notes": "Meta MSL blog: GPQA Diamond 89.5, Thinking mode. Averaged over 4 runs.", "candidates": [] }, { "model_id": "muse-spark", "benchmark_id": "swe_bench_verified", "score": 77.4, "reference_url": "https://ai.meta.com/blog/introducing-muse-spark-msl/", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "bash + file ops (view/edit)", "sampling": "avg 15 attempts", "judge": "rule-based (official docker)", "harness": "internal (Meta MSL)", "prompt_style": "default", "temperature": "default", "notes": "Bug-fixed 5 test issues per methodology doc (astropy #7606, sphinx 8595/9711/8269/8475)." }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "rule_ids": [ "R5a" ], "notes": "Meta MSL blog: SWE-bench Verified 77.4, Thinking mode. Single-attempt avg 15 trials; 5 test issues fixed.", "candidates": [] }, { "model_id": "muse-spark", "benchmark_id": "swe_bench_pro", "score": 52.4, "reference_url": "https://ai.meta.com/blog/introducing-muse-spark-msl/", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "bash + file ops", "sampling": "avg 4 attempts", "judge": "rule-based (official eval scripts)", "harness": "internal (Meta MSL)", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "rule_ids": [ "R5a" ], "notes": "Meta MSL blog: SWE-bench Pro 52.4, Thinking mode.", "candidates": [ { "score": 55.0, "reference_url": "https://ai.meta.com/static-resource/muse-spark-1.1-evaluation-report", "source_type": "tech_report", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "agentic coding", "sampling": "pass@1 unless source row states otherwise", "harness": "Meta MSL general-capability evaluation", "temperature": "1.0 for Muse; provider/source setting for comparisons", "context": "benchmark-specific", "configuration": "default" }, "notes": "Figure 44 · SWE-Bench Pro; metric=headline_metric. Exact printed value in the general-capability summary." } ] }, { "model_id": "muse-spark", "benchmark_id": "terminal_bench", "score": 59.0, "reference_url": "https://ai.meta.com/blog/introducing-muse-spark-msl/", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "bash-only", "sampling": "avg 15 attempts", "judge": "rule-based", "harness": "official terminal-bench-2", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "rule_ids": [ "R5a" ], "notes": "Meta MSL blog: Terminal-Bench 2.0 59.0, Thinking mode.", "candidates": [] }, { "model_id": "muse-spark", "benchmark_id": "gdpval_aa_elo", "score": 1444, "reference_url": "https://ai.meta.com/blog/introducing-muse-spark-msl/", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1 (single-pass)", "judge": "rule-based", "harness": "Artificial Analysis GDPval-AA Leaderboard", "prompt_style": "default", "temperature": "default", "notes": "Results from Artificial Analysis GDPval-AA leaderboard per Meta methodology" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "rule_ids": [ "R5a" ], "notes": "Meta MSL blog: GDPval-AA Elo 1444. Sourced from Artificial Analysis leaderboard.", "candidates": [] }, { "model_id": "grok-3-beta", "benchmark_id": "gpqa_diamond", "score": 80.2, "reference_url": "https://arxiv.org/abs/2504.13914", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Seed-Thinking-v1.5 paper Table 2: grok-3-beta=80.2.", "candidates": [ { "score": 84.6, "reference_url": "https://www.anthropic.com/news/claude-3-7-sonnet", "source_type": "official_blog", "reported_setting": { "mode": "thinking", "effort": "extended", "tools": "none", "sampling": "parallel TTC (majority voting N=64)", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "notes": "Anthropic 3.7 Sonnet blog: Grok 3 Beta extended thinking GPQA=80.2/84.6%; 84.6% is parallel TTC. Primary 80.2 already verified." } ] }, { "model_id": "grok-3-beta", "benchmark_id": "mmlu_pro", "score": 79.9, "reference_url": "https://x.ai/news/grok-3", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "xAI Grok 3 blog (cross-model non-reasoning table): grok-3-beta=79.9.", "candidates": [ { "score": 84.6, "reference_url": "https://arxiv.org/abs/2504.13914", "source_type": "tech_report", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "notes": "Seed-Thinking-v1.5 paper Table 2: grok-3-beta=84.6." } ] }, { "model_id": "grok-3-beta", "benchmark_id": "livecodebench", "score": 57.0, "reference_url": "https://x.ai/news/grok-3", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "xAI Grok 3 blog (cross-model non-reasoning table): grok-3-beta=57.0." }, { "model_id": "grok-3-beta", "benchmark_id": "chatbot_arena_elo", "score": 1402, "reference_url": "https://x.com/lmarena_ai/status/1891706264800936307", "audit_status": "verified", "source_type": "leaderboard" }, { "model_id": "grok-3-beta", "benchmark_id": "math_500", "score": 99.2, "reference_url": "https://artificialanalysis.ai/evaluations/math-500", "audit_status": "verified", "source_type": "third_party", "notes": "ArtificialAnalysis aggregator; used as fallback when no primary source available (R5d-aggregator-fallback)" }, { "model_id": "grok-3-beta", "benchmark_id": "aime_2025", "score": 77.3, "reference_url": "https://arxiv.org/abs/2504.13914", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Seed-Thinking-v1.5 paper Table 2: grok-3-beta=77.3." }, { "model_id": "grok-3-beta", "benchmark_id": "aime_2024", "score": 52.2, "reference_url": "https://x.ai/news/grok-3", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "xAI Grok 3 blog (cross-model non-reasoning table): grok-3-beta=52.2.", "candidates": [ { "score": 83.9, "reference_url": "https://arxiv.org/abs/2504.13914", "source_type": "tech_report", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "notes": "Seed-Thinking-v1.5 paper Table 2: grok-3-beta=83.9." }, { "score": 83.9, "reference_url": "https://www.anthropic.com/news/claude-3-7-sonnet", "source_type": "official_blog", "reported_setting": { "mode": "thinking", "effort": "extended", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "notes": "Anthropic 3.7 Sonnet blog (third-party eval): Grok 3 Beta extended thinking AIME-2024=83.9%/93.3% (parallel TTC). BP primary 52.2 is likely non-thinking mode." } ] }, { "model_id": "grok-3-beta", "benchmark_id": "simpleqa", "score": 43.6, "reference_url": "https://x.ai/news/grok-3", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "xAI Grok 3 blog (cross-model non-reasoning table): grok-3-beta=43.6." }, { "model_id": "grok-3-beta", "benchmark_id": "arc_agi_1", "score": 31.9, "reference_url": "https://arxiv.org/abs/2504.13914", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Seed-Thinking-v1.5 paper Table 2: grok-3-beta=31.9." }, { "model_id": "grok-3-beta", "benchmark_id": "arc_agi_2", "score": 0, "reference_url": "https://arcprize.org/arc-agi/2/", "audit_status": "verified", "source_type": "leaderboard" }, { "model_id": "grok-3-beta", "benchmark_id": "hle", "score": 18.2, "reference_url": "https://artificialanalysis.ai/models/grok-3", "audit_status": "verified", "source_type": "third_party", "notes": "ArtificialAnalysis aggregator; used as fallback when no primary source available (R5d-aggregator-fallback)" }, { "model_id": "grok-3-beta", "benchmark_id": "humaneval", "score": 87.3, "reference_url": "https://artificialanalysis.ai/models/grok-3", "audit_status": "verified", "source_type": "third_party", "notes": "ArtificialAnalysis aggregator; used as fallback when no primary source available (R5d-aggregator-fallback)" }, { "model_id": "grok-3-beta", "benchmark_id": "ifeval", "score": 83.4, "reference_url": "https://arxiv.org/abs/2504.13914", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Seed-Thinking-v1.5 paper Table 2: grok-3-beta=83.4." }, { "model_id": "grok-3-beta", "benchmark_id": "mmlu", "score": 88, "reference_url": "https://artificialanalysis.ai/models/grok-3", "audit_status": "verified", "source_type": "third_party", "notes": "ArtificialAnalysis aggregator; used as fallback when no primary source available (R5d-aggregator-fallback)" }, { "model_id": "grok-3-beta", "benchmark_id": "swe_bench_verified", "score": 48.5, "reference_url": "https://artificialanalysis.ai/models/grok-3", "audit_status": "verified", "source_type": "third_party", "notes": "ArtificialAnalysis aggregator; used as fallback when no primary source available (R5d-aggregator-fallback)" }, { "model_id": "grok-3-beta", "benchmark_id": "terminal_bench_1", "score": 17.5, "reference_url": "https://www.tbench.ai/leaderboard/terminal-bench/1.0", "audit_status": "verified", "source_type": "leaderboard" }, { "model_id": "grok-4", "benchmark_id": "gpqa_diamond", "score": 87.5, "reference_url": "https://x.ai/news/grok-4", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "xAI Grok 4 blog: grok-4=87.5 (no tools, no Heavy).", "candidates": [ { "score": 88.4, "reference_url": "https://llm-stats.com/benchmarks/gpqa", "source_type": "third_party_aggregator", "reported_setting": { "effort": "heavy" }, "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=Grok-4 Heavy, slug=grok-4-heavy, provider=xAI" } ] }, { "model_id": "grok-4", "benchmark_id": "hle", "score": 25.4, "reference_url": "https://x.ai/news/grok-4", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "xAI Grok 4 blog: grok-4=25.4 (no tools, no Heavy).", "candidates": [ { "score": 50.7, "reference_url": "https://llm-stats.com/benchmarks/humanity's-last-exam", "source_type": "third_party_aggregator", "reported_setting": { "effort": "heavy" }, "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=Grok-4 Heavy, slug=grok-4-heavy, provider=xAI" } ] }, { "model_id": "grok-4", "benchmark_id": "arc_agi_2", "score": 15.9, "reference_url": "https://x.ai/news/grok-4", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "xAI Grok 4 blog: grok-4=15.9 (no tools, no Heavy).", "candidates": [ { "score": 16.0, "reference_url": "https://arcprize.org/leaderboard", "source_type": "leaderboard", "reported_setting": { "mode": "thinking", "effort": "thinking", "tools": "none", "sampling": "pass@1 (single-pass)", "harness": "official (arcprize.org)", "system_type": "CoT" }, "notes": "Grok 4 (Thinking) on arcprize.org leaderboard; BP primary 15.9 from x.ai/news/grok-4" } ] }, { "model_id": "grok-4", "benchmark_id": "livecodebench", "score": 79, "reference_url": "https://x.ai/news/grok-4", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "xAI Grok 4 blog: grok-4=79 (no tools, no Heavy).", "candidates": [ { "score": 79.4, "reference_url": "https://llm-stats.com/benchmarks/livecodebench", "source_type": "third_party_aggregator", "reported_setting": { "effort": "heavy" }, "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=Grok-4 Heavy, slug=grok-4-heavy, provider=xAI" } ] }, { "model_id": "grok-4", "benchmark_id": "swe_bench_verified", "score": 73.5, "reference_url": "https://datasciencedojo.com/blog/grok-4/", "audit_status": "dropped", "notes": "Dropped: third-party random blog, unreliable source (R5h-drop-third-party-blog)" }, { "model_id": "grok-4", "benchmark_id": "aime_2025", "score": 91.7, "reference_url": "https://x.ai/news/grok-4", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "xAI Grok 4 blog: grok-4=91.7 (no tools, no Heavy).", "candidates": [ { "score": 100.0, "reference_url": "https://llm-stats.com/benchmarks/aime-2025", "source_type": "third_party_aggregator", "reported_setting": { "effort": "heavy" }, "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=Grok-4 Heavy, slug=grok-4-heavy, provider=xAI" } ] }, { "model_id": "grok-4", "benchmark_id": "aa_intelligence_index", "score": 73, "reference_url": "https://x.com/ArtificialAnlys/status/1943166841150644622", "audit_status": "verified", "source_type": "third_party", "notes": "ArtificialAnalysis aggregator; used as fallback when no primary source available (R5d-aggregator-fallback)" }, { "model_id": "grok-4", "benchmark_id": "humaneval", "score": 88.0, "reference_url": "https://automatio.ai/models/grok-4", "audit_status": "dropped", "notes": "Dropped: third-party random blog, unreliable source (R5h-drop-third-party-blog)" }, { "model_id": "grok-4", "benchmark_id": "simpleqa", "score": 48.0, "reference_url": "https://automatio.ai/models/grok-4", "audit_status": "dropped", "notes": "Dropped: third-party random blog, unreliable source (R5h-drop-third-party-blog)" }, { "model_id": "grok-4", "benchmark_id": "ifeval", "score": 89.2, "reference_url": "https://automatio.ai/models/grok-4", "audit_status": "dropped", "notes": "Dropped: third-party random blog, unreliable source (R5h-drop-third-party-blog)" }, { "model_id": "grok-4", "benchmark_id": "mmmu", "score": 75.0, "reference_url": "https://automatio.ai/models/grok-4", "audit_status": "dropped", "notes": "Dropped: third-party random blog, unreliable source (R5h-drop-third-party-blog)" }, { "model_id": "grok-4", "benchmark_id": "mmmu_pro", "score": 59.2, "reference_url": "https://automatio.ai/models/grok-4", "audit_status": "dropped", "notes": "Dropped: third-party random blog, unreliable source (R5h-drop-third-party-blog)" }, { "model_id": "grok-4", "benchmark_id": "codeforces_rating", "score": 2708, "reference_url": "https://x.ai/news/grok-4", "audit_status": "dropped", "rule_id": "R5g-ghost", "audit_note": "x.ai/news/grok-4 returns 403 (Cloudflare blocked). Official Grok-4 benchmarks per secondary sources: HLE, ARC-AGI-2, SWE-Bench — Codeforces rating not in official announcement.", "audited_by": "lychee-audit-bot", "audited_at": "2026-07-26T00:00:00Z" }, { "model_id": "grok-4", "benchmark_id": "chatbot_arena_elo", "score": 1465, "reference_url": "https://lmarena.ai/", "audit_status": "verified", "source_type": "leaderboard", "candidates": [ { "score": 1410, "reference_url": "https://lmarena.ai/leaderboard/text", "source_type": "leaderboard", "notes": "lmarena.ai grok-4-0709 (rank 92) ELO 2026-04-29. Lower than original 1465 due to ELO settling as new models enter arena." } ] }, { "model_id": "grok-4", "benchmark_id": "aime_2024", "score": 94.0, "reference_url": "https://www.llmrumors.com/news/grok-4-the-breakthrough-ai-model-that-changes-everything", "audit_status": "dropped", "notes": "Dropped: third-party random blog, unreliable source (R5h-drop-third-party-blog)" }, { "model_id": "grok-4", "benchmark_id": "arc_agi_1", "score": 66.7, "reference_url": "https://arcprize.org/leaderboard", "audit_status": "verified", "source_type": "leaderboard" }, { "model_id": "grok-4", "benchmark_id": "frontiermath", "score": 13.0, "reference_url": "https://epoch.ai/blog/grok-4-math", "audit_status": "verified", "source_type": "leaderboard" }, { "model_id": "grok-4", "benchmark_id": "imo_2025", "score": 11.9, "reference_url": "https://matharena.ai/?comp=imo--imo_2025", "audit_status": "verified", "source_type": "leaderboard" }, { "model_id": "grok-4", "benchmark_id": "mmlu", "score": 94.0, "reference_url": "https://forgecode.dev/blog/grok-4-initial-impression/", "audit_status": "dropped", "notes": "Dropped: third-party random blog, unreliable source (R5h-drop-third-party-blog)" }, { "model_id": "grok-4", "benchmark_id": "mmlu_pro", "score": 87.0, "reference_url": "https://artificialanalysis.ai/models/grok-4", "audit_status": "verified", "source_type": "third_party", "notes": "ArtificialAnalysis aggregator; used as fallback when no primary source available (R5d-aggregator-fallback)" }, { "model_id": "grok-4", "benchmark_id": "terminal_bench", "score": 23.1, "reference_url": "https://www.tbench.ai/leaderboard/terminal-bench/2.0", "audit_status": "verified", "source_type": "leaderboard" }, { "model_id": "grok-4", "benchmark_id": "aa_lcr", "score": 68, "reference_url": "https://artificialanalysis.ai/evaluations/artificial-analysis-long-context-reasoning", "audit_status": "verified", "source_type": "third_party", "notes": "ArtificialAnalysis aggregator; used as fallback when no primary source available (R5d-aggregator-fallback)" }, { "model_id": "grok-4", "benchmark_id": "brumo_2025", "score": 95, "reference_url": "https://matharena.ai/?comp=brumo--brumo_2025", "audit_status": "verified", "source_type": "leaderboard" }, { "model_id": "grok-4", "benchmark_id": "cmimc_2025", "score": 83.75, "reference_url": "https://matharena.ai/?comp=cmimc--cmimc_2025", "audit_status": "verified", "source_type": "leaderboard" }, { "model_id": "grok-4", "benchmark_id": "hmmt_nov_2025", "score": 88.33, "reference_url": "https://matharena.ai/?comp=hmmt--hmmt_nov_2025", "audit_status": "verified", "source_type": "leaderboard" }, { "model_id": "grok-4", "benchmark_id": "math_500", "score": 98, "reference_url": "https://aitoolapp.com/grok-4/benchmarks/", "audit_status": "dropped", "notes": "Dropped: third-party random blog, unreliable source (R5h-drop-third-party-blog)" }, { "model_id": "grok-4", "benchmark_id": "matharena_apex_2025", "score": 2.08, "reference_url": "https://matharena.ai/?comp=apex--apex_2025", "audit_status": "verified", "source_type": "leaderboard" }, { "model_id": "grok-4", "benchmark_id": "osworld", "score": 48, "reference_url": "https://aitoolapp.com/grok-4/benchmarks/", "audit_status": "dropped", "notes": "Dropped: third-party random blog, unreliable source (R5h-drop-third-party-blog)" }, { "model_id": "grok-4", "benchmark_id": "smt_2025", "score": 85.85, "reference_url": "https://matharena.ai/?comp=smt--smt_2025", "audit_status": "verified", "source_type": "leaderboard" }, { "model_id": "grok-4", "benchmark_id": "swe_bench_pro", "score": 46.5, "reference_url": "https://artificialanalysis.ai/models/grok-4", "audit_status": "verified", "source_type": "third_party", "notes": "ArtificialAnalysis aggregator; used as fallback when no primary source available (R5d-aggregator-fallback)" }, { "model_id": "grok-4", "benchmark_id": "terminal_bench_1", "score": 39, "reference_url": "https://www.tbench.ai/leaderboard/terminal-bench/1.0", "audit_status": "verified", "source_type": "leaderboard" }, { "model_id": "grok-4", "benchmark_id": "usamo_2025", "score": 37.5, "reference_url": "https://x.ai/news/grok-4", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "xAI Grok 4 blog: grok-4=37.5 (no tools, no Heavy)." }, { "model_id": "grok-4.1", "benchmark_id": "chatbot_arena_elo", "score": 1483, "reference_url": "https://x.ai/news/grok-4-1", "audit_status": "needs_review", "audit_note": "x.ai/news/grok-4-1 returns 403 (Cloudflare blocked). Secondary sources confirm ELO=1483 citing official xAI announcement; primary URL inaccessible to bot.", "audited_by": "lychee-audit-bot", "audited_at": "2026-07-26T00:00:00Z" }, { "model_id": "grok-4.1", "benchmark_id": "swe_bench_verified", "score": 50.6, "reference_url": "https://deepmind.google/models/gemini/flash/", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "DeepMind /models/gemini/flash/ Performance table: SWE-bench Verified 50.6% [Promoted to verified, prior unverified value deleted per R5d.]" }, { "model_id": "grok-4.1", "benchmark_id": "aime_2025", "score": 91.9, "reference_url": "https://deepmind.google/models/gemini/flash/", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "DeepMind /models/gemini/flash/ Performance table: AIME 2025 no tools 91.9% [Promoted to verified, prior unverified value deleted per R5d.]" }, { "model_id": "grok-4.1", "benchmark_id": "mmmu_pro", "score": 63.0, "reference_url": "https://deepmind.google/models/gemini/flash/", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "DeepMind /models/gemini/flash/ Performance table: MMMU-Pro 63.0% [Promoted to verified, prior unverified value deleted per R5d.]" }, { "model_id": "grok-4.1", "benchmark_id": "video_mmmu", "score": 87.6, "reference_url": "https://llm-stats.com/benchmarks/videommmu", "audit_status": "dropped", "notes": "Dropped: third-party random blog, unreliable source (R5h-drop-third-party-blog)" }, { "model_id": "grok-4.1", "benchmark_id": "gpqa_diamond", "score": 84.3, "reference_url": "https://deepmind.google/models/gemini/flash/", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "DeepMind /models/gemini/flash/ Performance table: GPQA Diamond 84.3% [Promoted to verified, prior unverified value deleted per R5d.]" }, { "model_id": "grok-4.1", "benchmark_id": "hle", "score": 17.6, "reference_url": "https://deepmind.google/models/gemini/flash/", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "DeepMind /models/gemini/flash/ Performance table: HLE no tools 17.6% (Grok 4.1 Fast Reasoning) [Promoted to verified, prior unverified value deleted per R5d.]" }, { "model_id": "grok-4.1", "benchmark_id": "mmlu_pro", "score": 85.3, "reference_url": "https://www.sentisight.ai/gemini-3-vs-grok-4-1-vs-chatgpt-5-1/", "audit_status": "dropped", "notes": "Dropped: third-party random blog, unreliable source (R5h-drop-third-party-blog)" }, { "model_id": "grok-4.1", "benchmark_id": "arc_agi_2", "score": 42, "reference_url": "https://www.analyticsvidhya.com/blog/2025/11/gemini-3-vs-grok-4-1-best-ai-of-2025/", "audit_status": "dropped", "notes": "Dropped: third-party random blog, unreliable source (R5h-drop-third-party-blog)" }, { "model_id": "grok-4.1", "benchmark_id": "brumo_2025", "score": 97.5, "reference_url": "https://matharena.ai/?comp=brumo--brumo_2025", "audit_status": "verified", "source_type": "leaderboard" }, { "model_id": "grok-4.1", "benchmark_id": "cmimc_2025", "score": 84.38, "reference_url": "https://matharena.ai/?comp=cmimc--cmimc_2025", "audit_status": "verified", "source_type": "leaderboard" }, { "model_id": "grok-4.1", "benchmark_id": "codeforces_rating", "score": 2650, "reference_url": "https://www.glbgpt.com/hub/chatgpt-5-1-vs-grok-4-1-2025/", "audit_status": "dropped", "notes": "Dropped: third-party random blog, unreliable source (R5h-drop-third-party-blog)" }, { "model_id": "grok-4.1", "benchmark_id": "frontiermath", "score": 38, "reference_url": "https://www.analyticsvidhya.com/blog/2025/11/gemini-3-vs-grok-4-1-best-ai-of-2025/", "audit_status": "dropped", "notes": "Dropped: third-party random blog, unreliable source (R5h-drop-third-party-blog)" }, { "model_id": "grok-4.1", "benchmark_id": "hmmt_nov_2025", "score": 93.33, "reference_url": "https://matharena.ai/?comp=hmmt--hmmt_nov_2025", "audit_status": "verified", "source_type": "leaderboard" }, { "model_id": "grok-4.1", "benchmark_id": "humaneval", "score": 95, "reference_url": "https://www.analyticsvidhya.com/blog/2025/11/gemini-3-vs-grok-4-1-best-ai-of-2025/", "audit_status": "dropped", "notes": "Dropped: third-party random blog, unreliable source (R5h-drop-third-party-blog)" }, { "model_id": "grok-4.1", "benchmark_id": "ifeval", "score": 91, "reference_url": "https://www.sentisight.ai/gemini-3-vs-grok-4-1-vs-chatgpt-5-1/", "audit_status": "dropped", "notes": "Dropped: third-party random blog, unreliable source (R5h-drop-third-party-blog)" }, { "model_id": "grok-4.1", "benchmark_id": "livebench", "score": 60, "reference_url": "https://livebench.ai/", "audit_status": "verified", "source_type": "leaderboard" }, { "model_id": "grok-4.1", "benchmark_id": "livecodebench", "score": 82, "reference_url": "https://www.sentisight.ai/gemini-3-vs-grok-4-1-vs-chatgpt-5-1/", "audit_status": "dropped", "notes": "Dropped: third-party random blog, unreliable source (R5h-drop-third-party-blog)" }, { "model_id": "grok-4.1", "benchmark_id": "math_500", "score": 98.5, "reference_url": "https://www.sentisight.ai/gemini-3-vs-grok-4-1-vs-chatgpt-5-1/", "audit_status": "dropped", "notes": "Dropped: third-party random blog, unreliable source (R5h-drop-third-party-blog)" }, { "model_id": "grok-4.1", "benchmark_id": "matharena_apex_2025", "score": 5.21, "reference_url": "https://matharena.ai/?comp=apex--apex_2025", "audit_status": "verified", "source_type": "leaderboard" }, { "model_id": "grok-4.1", "benchmark_id": "osworld", "score": 52, "reference_url": "https://www.analyticsvidhya.com/blog/2025/11/gemini-3-vs-grok-4-1-best-ai-of-2025/", "audit_status": "dropped", "notes": "Dropped: third-party random blog, unreliable source (R5h-drop-third-party-blog)" }, { "model_id": "grok-4.1", "benchmark_id": "simpleqa", "score": 55, "reference_url": "https://www.glbgpt.com/hub/chatgpt-5-1-vs-grok-4-1-2025/", "audit_status": "dropped", "notes": "Dropped: third-party random blog, unreliable source (R5h-drop-third-party-blog)" }, { "model_id": "grok-4.1", "benchmark_id": "smt_2025", "score": 84.6, "reference_url": "https://matharena.ai/?comp=smt--smt_2025", "audit_status": "verified", "source_type": "leaderboard" }, { "model_id": "grok-4.20", "benchmark_id": "arc_agi_1", "score": 89.5, "reference_url": "https://arcprize.org/leaderboard", "audit_status": "verified", "source_type": "leaderboard" }, { "model_id": "grok-4.20", "benchmark_id": "arc_agi_2", "score": 65.1, "reference_url": "https://arcprize.org/leaderboard", "audit_status": "verified", "source_type": "leaderboard" }, { "model_id": "deepseek-r1", "benchmark_id": "gpqa_diamond", "score": 71.5, "reference_url": "https://huggingface.co/deepseek-ai/DeepSeek-R1-0528", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "DeepSeek R1-0528 model card: deepseek-r1=71.5.", "candidates": [ { "score": 73.0, "reference_url": "https://arxiv.org/abs/2504.21318", "source_type": "tech_report", "reported_setting": { "mode": "unknown", "effort": "unknown", "tools": "unknown", "sampling": "unknown", "judge": "rule-based", "harness": "Microsoft Phi internal eval", "prompt_style": "unknown", "temperature": "unknown" }, "notes": "Phi-4-reasoning paper Table 1 (deepseek-r1 column): gpqa_diamond=73.0 (alt measurement, mc=false)" } ] }, { "model_id": "deepseek-r1", "benchmark_id": "math_500", "score": 97.3, "reference_url": "https://arxiv.org/abs/2501.12948", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "DeepSeek R1 paper Table 3 (page 9, R1 final): 97.3." }, { "model_id": "deepseek-r1", "benchmark_id": "aime_2024", "score": 79.8, "reference_url": "https://huggingface.co/deepseek-ai/DeepSeek-R1-0528", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "DeepSeek R1-0528 model card: deepseek-r1=79.8.", "candidates": [ { "score": 78.7, "reference_url": "https://arxiv.org/abs/2504.21318", "source_type": "tech_report", "reported_setting": { "mode": "unknown", "effort": "unknown", "tools": "unknown", "sampling": "unknown", "judge": "rule-based", "harness": "Microsoft Phi internal eval", "prompt_style": "unknown", "temperature": "unknown" }, "notes": "Phi-4-reasoning paper Table 1 (deepseek-r1 column): aime_2024=78.7 (alt measurement, mc=false)" } ] }, { "model_id": "deepseek-r1", "benchmark_id": "mmlu", "score": 90.8, "reference_url": "https://arxiv.org/abs/2501.12948", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "DeepSeek R1 paper Table 3 (page 9, R1 final): 90.8." }, { "model_id": "deepseek-r1", "benchmark_id": "mmlu_pro", "score": 84.0, "reference_url": "https://huggingface.co/deepseek-ai/DeepSeek-R1-0528", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "DeepSeek R1-0528 model card: deepseek-r1=84.0.", "candidates": [ { "score": 85.6, "reference_url": "https://arxiv.org/abs/2504.13914", "source_type": "tech_report", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "notes": "Seed-Thinking-v1.5 paper Table 2: deepseek-r1=85.6." } ] }, { "model_id": "deepseek-r1", "benchmark_id": "swe_bench_verified", "score": 49.2, "reference_url": "https://huggingface.co/deepseek-ai/DeepSeek-R1-0528", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "agentic (bash + edit)", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "DeepSeek R1-0528 model card: deepseek-r1=49.2." }, { "model_id": "deepseek-r1", "benchmark_id": "livecodebench", "score": 65.9, "reference_url": "https://arxiv.org/abs/2501.12948", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "DeepSeek R1 paper Table 3 (page 9, R1 final): 65.9.", "candidates": [ { "score": 63.5, "reference_url": "https://huggingface.co/deepseek-ai/DeepSeek-R1-0528", "source_type": "model_card", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "notes": "DeepSeek R1-0528 model card: deepseek-r1=63.5. (demoted: superseded by DS R1 paper)" } ] }, { "model_id": "deepseek-r1", "benchmark_id": "codeforces_rating", "score": 2029, "reference_url": "https://arxiv.org/abs/2501.12948", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "DeepSeek R1 paper Table 3 (page 9, R1 final): 2029.", "candidates": [ { "score": 1530, "reference_url": "https://huggingface.co/deepseek-ai/DeepSeek-R1-0528", "source_type": "model_card", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "notes": "DeepSeek R1-0528 model card: deepseek-r1=1530. (demoted: superseded by DS R1 paper)" } ] }, { "model_id": "deepseek-r1", "benchmark_id": "arena_hard", "score": 92.3, "reference_url": "https://arxiv.org/abs/2501.12948", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "DeepSeek R1 paper Table 3 (page 9, R1 final): 92.3." }, { "model_id": "deepseek-r1", "benchmark_id": "ifeval", "score": 83.3, "reference_url": "https://arxiv.org/abs/2501.12948", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "DeepSeek R1 paper Table 3 (page 9, R1 final): 83.3.", "candidates": [ { "score": 86.1, "reference_url": "https://arxiv.org/abs/2504.13914", "source_type": "tech_report", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "notes": "Seed-Thinking-v1.5 paper Table 2: deepseek-r1=86.1." } ] }, { "model_id": "deepseek-r1", "benchmark_id": "simpleqa", "score": 30.1, "reference_url": "https://huggingface.co/deepseek-ai/DeepSeek-R1-0528", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "DeepSeek R1-0528 model card: deepseek-r1=30.1." }, { "model_id": "deepseek-r1", "benchmark_id": "humaneval", "score": 96.1, "reference_url": "https://arxiv.org/html/2501.12948v1", "audit_status": "dropped", "notes": " | DROPPED: cell URL is arxiv html v1 of R1 paper but R1 paper Table 8 (main) and Table 15 (distill) do not report HumanEval. Ghost cell per R5g(a)." }, { "model_id": "deepseek-r1", "benchmark_id": "aime_2025", "score": 70.0, "reference_url": "https://huggingface.co/deepseek-ai/DeepSeek-R1-0528", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "DeepSeek R1-0528 model card: deepseek-r1=70.0.", "candidates": [ { "score": 65.0, "reference_url": "https://arxiv.org/abs/2504.13914", "source_type": "tech_report", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "notes": "Seed-Thinking-v1.5 paper Table 2: deepseek-r1=65.0." }, { "score": 70.4, "reference_url": "https://arxiv.org/abs/2504.21318", "source_type": "tech_report", "reported_setting": { "mode": "unknown", "effort": "unknown", "tools": "unknown", "sampling": "unknown", "judge": "rule-based", "harness": "Microsoft Phi internal eval", "prompt_style": "unknown", "temperature": "unknown" }, "notes": "Phi-4-reasoning paper Table 1 (deepseek-r1 column): aime_2025=70.4 (alt measurement, mc=false)" } ] }, { "model_id": "deepseek-r1", "benchmark_id": "hle", "score": 8.5, "reference_url": "https://huggingface.co/deepseek-ai/DeepSeek-R1-0528", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "DeepSeek R1-0528 model card: deepseek-r1=8.5." }, { "model_id": "deepseek-r1", "benchmark_id": "hmmt_feb_2025", "score": 41.7, "reference_url": "https://huggingface.co/deepseek-ai/DeepSeek-R1-0528", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "DeepSeek R1-0528 model card: deepseek-r1=41.7." }, { "model_id": "deepseek-r1", "benchmark_id": "chatbot_arena_elo", "score": 1398, "reference_url": "https://lmarena.ai/leaderboard/text", "audit_status": "verified", "source_type": "leaderboard", "matches_canonical": true, "notes": "lmarena.ai text arena (fetched 2026-04-29). Arena model name: deepseek-r1 (rank 108). ELO updates continuously; score reflects latest available." }, { "model_id": "deepseek-r1", "benchmark_id": "arc_agi_1", "score": 18.3, "reference_url": "https://arxiv.org/abs/2504.13914", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Seed-Thinking-v1.5 paper Table 2: deepseek-r1=18.3." }, { "model_id": "deepseek-r1", "benchmark_id": "gsm8k", "score": 97.3, "reference_url": "https://arxiv.org/html/2501.12948v1", "audit_status": "dropped", "notes": " | DROPPED: cell URL is arxiv html v1 of R1 paper but R1 paper Table 8 reports MATH-500 not GSM8K. Ghost cell per R5g(a)." }, { "model_id": "deepseek-r1", "benchmark_id": "arc_agi_2", "score": 1.3, "reference_url": "https://arcprize.org/leaderboard", "audit_status": "verified", "source_type": "leaderboard", "matches_canonical": true, "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1 (single-pass)", "judge": "rule-based", "harness": "official (arcprize.org)", "system_type": "CoT" }, "notes": "arcprize.org leaderboard audit: Deepseek R1 on leaderboard" }, { "model_id": "deepseek-r1", "benchmark_id": "brumo_2025", "score": 80.83, "reference_url": "https://matharena.ai/?comp=brumo--brumo_2025", "audit_status": "verified", "notes": " | Verified: matharena.ai shows 80.83%" }, { "model_id": "deepseek-r1", "benchmark_id": "critpt", "score": 1.1, "reference_url": "https://github.com/CritPt-Benchmark/CritPt", "audit_status": "verified", "reported_setting": "CritPt leaderboard, ε=1.5, % tasks solved", "source_type": "leaderboard", "matches_canonical": true }, { "model_id": "deepseek-r1", "benchmark_id": "ifbench", "score": 38, "reference_url": "https://github.com/allenai/IFBench", "audit_status": "verified", "reported_setting": "IFBench score, 0-shot", "source_type": "leaderboard", "matches_canonical": true }, { "model_id": "deepseek-r1", "benchmark_id": "smt_2025", "score": 66.51, "reference_url": "https://matharena.ai/?comp=smt--smt_2025", "candidates": [ { "score": 66.98, "reference_url": "https://matharena.ai/", "source_type": "third-party-benchmark", "reported_setting": { "run": "official matharena" }, "notes": "Current matharena value 66.98%; BP has 66.51%. Matharena scores update as more runs are added." } ], "audit_status": "flagged", "notes": " | MISMATCH: BP=66.51, current matharena=66.98; added candidate" }, { "model_id": "deepseek-r1", "benchmark_id": "terminal_bench_1", "score": 5.7, "reference_url": "https://www.tbench.ai/leaderboard/terminal-bench/1.0", "audit_status": "verified", "source_type": "leaderboard", "rule_ids": [ "R5d" ], "notes": "Confirmed 5.7% on tbench.ai Terminal-Bench 1.0 leaderboard (Terminus 1 scaffold, 2025-05-15)." }, { "model_id": "deepseek-r1", "benchmark_id": "usamo_2025", "score": 4.76, "reference_url": "https://matharena.ai/?comp=usamo--usamo_2025", "audit_status": "verified", "notes": " | Verified: matharena.ai shows 4.76%" }, { "model_id": "deepseek-v3", "benchmark_id": "mmlu", "score": 88.5, "reference_url": "https://arxiv.org/abs/2412.19437", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "DeepSeek V3 paper Table 6: deepseek-v3=88.5.", "candidates": [ { "score": 85.7, "reference_url": "https://cohere.com/research/papers/command-a-technical-report.pdf", "source_type": "tech_report", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "Cohere internal eval", "prompt_style": "default", "temperature": "default" }, "notes": "Command A paper: alt measurement 85.7 (mc=false)." } ] }, { "model_id": "deepseek-v3", "benchmark_id": "gpqa_diamond", "score": 59.1, "reference_url": "https://arxiv.org/abs/2412.19437", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "DeepSeek V3 paper Table 6: deepseek-v3=59.1." }, { "model_id": "deepseek-v3", "benchmark_id": "math_500", "score": 90.2, "reference_url": "https://arxiv.org/abs/2412.19437", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "DeepSeek V3 paper Table 6: deepseek-v3=90.2." }, { "model_id": "deepseek-v3", "benchmark_id": "swe_bench_verified", "score": 42.0, "reference_url": "https://arxiv.org/abs/2412.19437", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "agentic (bash + edit)", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "DeepSeek V3 paper Table 6: deepseek-v3=42.0." }, { "model_id": "deepseek-v3", "benchmark_id": "ifeval", "score": 86.1, "reference_url": "https://arxiv.org/abs/2412.19437", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "DeepSeek V3 paper Table 6: deepseek-v3=86.1.", "candidates": [ { "score": 83.8, "reference_url": "https://cohere.com/research/papers/command-a-technical-report.pdf", "source_type": "tech_report", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "Cohere internal eval", "prompt_style": "default", "temperature": "default" }, "notes": "Command A paper: alt measurement 83.8 (mc=false)." } ] }, { "model_id": "deepseek-v3", "benchmark_id": "arena_hard", "score": 85.5, "reference_url": "https://github.com/deepseek-ai/DeepSeek-V3", "audit_status": "verified", "reported_setting": "Arena-Hard open-ended generation, README Table", "source_type": "tech_report", "matches_canonical": true }, { "model_id": "deepseek-v3", "benchmark_id": "humaneval", "score": 82.6, "reference_url": "https://arxiv.org/abs/2412.19437", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "DeepSeek V3 paper Table 6: deepseek-v3=82.6.", "candidates": [ { "score": 83.5, "reference_url": "https://cohere.com/research/papers/command-a-technical-report.pdf", "source_type": "tech_report", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "Cohere internal eval", "prompt_style": "default", "temperature": "default" }, "notes": "Command A paper: alt measurement 83.5 (mc=false)." } ] }, { "model_id": "deepseek-v3", "benchmark_id": "codeforces_rating", "score": 1134, "reference_url": "https://arxiv.org/html/2501.12948v1", "audit_status": "verified", "source_type": "academic_paper", "matches_canonical": true, "audit_note": "Table 4: DeepSeek-V3 Codeforces Rating=1134.", "audited_by": "lychee-audit-bot", "audited_at": "2026-04-29T13:17:14Z" }, { "model_id": "deepseek-v3", "benchmark_id": "chatbot_arena_elo", "score": 1358, "reference_url": "https://lmarena.ai/leaderboard/text", "audit_status": "verified", "source_type": "leaderboard", "matches_canonical": true, "notes": "lmarena.ai text arena (fetched 2026-04-29). Arena model name: deepseek-v3 (rank 148). ELO updates continuously; score reflects latest available." }, { "model_id": "deepseek-v3", "benchmark_id": "gsm8k", "score": 89.3, "reference_url": "https://github.com/deepseek-ai/DeepSeek-V3", "audit_status": "verified", "reported_setting": "GSM8K, 8-shot EM, README Table", "source_type": "tech_report", "matches_canonical": true }, { "model_id": "deepseek-v3", "benchmark_id": "bigcodebench", "score": 50.0, "reference_url": "https://cohere.com/research/papers/command-a-technical-report.pdf", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "Cohere internal eval", "prompt_style": "default", "temperature": "default" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Command A paper: bigcodebench=50.0 (matches BP, re-sourced)" }, { "model_id": "deepseek-v3-0324", "benchmark_id": "gpqa_diamond", "score": 68.4, "reference_url": "https://huggingface.co/deepseek-ai/DeepSeek-V3-0324", "audit_status": "verified", "notes": " | Verified: HF V3-0324 model card shows 68.4" }, { "model_id": "deepseek-v3-0324", "benchmark_id": "aime_2024", "score": 59.4, "reference_url": "https://huggingface.co/deepseek-ai/DeepSeek-V3-0324", "audit_status": "verified", "notes": " | Verified: HF V3-0324 model card shows 59.4" }, { "model_id": "deepseek-v3-0324", "benchmark_id": "livecodebench", "score": 49.2, "reference_url": "https://huggingface.co/deepseek-ai/DeepSeek-V3-0324", "audit_status": "verified", "notes": " | Verified: HF V3-0324 model card shows 49.2" }, { "model_id": "deepseek-v3-0324", "benchmark_id": "mmlu_pro", "score": 81.2, "reference_url": "https://huggingface.co/deepseek-ai/DeepSeek-V3-0324", "audit_status": "verified", "notes": " | Verified: HF V3-0324 model card shows 81.2" }, { "model_id": "deepseek-v3-0324", "benchmark_id": "aime_2025", "score": 46.7, "reference_url": "https://arxiv.org/abs/2507.20534", "audit_status": "verified", "notes": " | Verified: Kimi K2 paper (arXiv 2507.20534) Table 3 shows 46.7" }, { "model_id": "deepseek-v3-0324", "benchmark_id": "arena_hard", "score": 39.9, "reference_url": "https://arxiv.org/abs/2507.20534", "audit_status": "verified", "notes": " | Verified: Kimi K2 paper (arXiv 2507.20534) Table 3 shows 39.9" }, { "model_id": "deepseek-v3-0324", "benchmark_id": "hle", "score": 5.2, "reference_url": "https://arxiv.org/abs/2507.20534", "audit_status": "verified", "notes": " | Verified: Kimi K2 paper (arXiv 2507.20534) Table 3 shows 5.2" }, { "model_id": "deepseek-v3-0324", "benchmark_id": "hmmt_feb_2025", "score": 27.5, "reference_url": "https://arxiv.org/abs/2507.20534", "audit_status": "verified", "notes": " | Verified: Kimi K2 paper (arXiv 2507.20534) Table 3 shows 27.5" }, { "model_id": "deepseek-v3-0324", "benchmark_id": "ifeval", "score": 81.1, "reference_url": "https://arxiv.org/abs/2507.20534", "audit_status": "verified", "notes": " | Verified: Kimi K2 paper (arXiv 2507.20534) Table 3 shows 81.1" }, { "model_id": "deepseek-v3-0324", "benchmark_id": "math_500", "score": 94.0, "reference_url": "https://arxiv.org/abs/2507.20534", "audit_status": "verified", "notes": " | Verified: Kimi K2 paper (arXiv 2507.20534) Table 3 shows 94.0" }, { "model_id": "deepseek-v3-0324", "benchmark_id": "mmlu", "score": 89.4, "reference_url": "https://arxiv.org/abs/2507.20534", "audit_status": "verified", "notes": " | Verified: Kimi K2 paper (arXiv 2507.20534) Table 3 shows 89.4" }, { "model_id": "deepseek-v3-0324", "benchmark_id": "simpleqa", "score": 27.7, "reference_url": "https://arxiv.org/abs/2507.20534", "audit_status": "verified", "notes": " | Verified: Kimi K2 paper (arXiv 2507.20534) Table 3 shows 27.7" }, { "model_id": "deepseek-v3-0324", "benchmark_id": "swe_bench_verified", "score": 38.8, "reference_url": "https://arxiv.org/abs/2507.20534", "audit_status": "verified", "notes": " | Verified: Kimi K2 paper (arXiv 2507.20534) Table 3 shows 38.8" }, { "model_id": "deepseek-v3-0324", "benchmark_id": "codeforces_rating", "score": 1650, "reference_url": "https://textcortex.com/post/deepseek-v3-review", "audit_status": "dropped", "rule_id": "R5h-third-party-blog", "notes": "DROPPED (R5h): random third-party blog/aggregator, no primary trail" }, { "model_id": "deepseek-v3-0324", "benchmark_id": "humaneval", "score": 85, "reference_url": "https://textcortex.com/post/deepseek-v3-review", "audit_status": "dropped", "rule_id": "R5h-third-party-blog", "notes": "DROPPED (R5h): random third-party blog/aggregator, no primary trail" }, { "model_id": "deepseek-r1-0528", "benchmark_id": "aime_2025", "score": 87.5, "reference_url": "https://huggingface.co/deepseek-ai/DeepSeek-R1-0528", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "DeepSeek R1-0528 model card: deepseek-r1-0528=87.5." }, { "model_id": "deepseek-r1-0528", "benchmark_id": "aime_2024", "score": 91.4, "reference_url": "https://huggingface.co/deepseek-ai/DeepSeek-R1-0528", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "DeepSeek R1-0528 model card: deepseek-r1-0528=91.4." }, { "model_id": "deepseek-r1-0528", "benchmark_id": "gpqa_diamond", "score": 81.0, "reference_url": "https://huggingface.co/deepseek-ai/DeepSeek-R1-0528", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "DeepSeek R1-0528 model card: deepseek-r1-0528=81.0." }, { "model_id": "deepseek-r1-0528", "benchmark_id": "math_500", "score": 97.3, "reference_url": "https://huggingface.co/deepseek-ai/DeepSeek-R1-0528", "audit_status": "flagged", "notes": " | FLAGGED: HF R1-0528 model card does not contain MATH-500; 97.3 is from DeepSeek-R1 (original) model card" }, { "model_id": "deepseek-r1-0528", "benchmark_id": "livecodebench", "score": 73.3, "reference_url": "https://huggingface.co/deepseek-ai/DeepSeek-R1-0528", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "DeepSeek R1-0528 model card: deepseek-r1-0528=73.3." }, { "model_id": "deepseek-r1-0528", "benchmark_id": "swe_bench_verified", "score": 57.6, "reference_url": "https://huggingface.co/deepseek-ai/DeepSeek-R1-0528", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "agentic (bash + edit)", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "DeepSeek R1-0528 model card: deepseek-r1-0528=57.6." }, { "model_id": "deepseek-r1-0528", "benchmark_id": "mmlu_pro", "score": 85.0, "reference_url": "https://huggingface.co/deepseek-ai/DeepSeek-R1-0528", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "DeepSeek R1-0528 model card: deepseek-r1-0528=85.0." }, { "model_id": "deepseek-r1-0528", "benchmark_id": "simpleqa", "score": 27.8, "reference_url": "https://huggingface.co/deepseek-ai/DeepSeek-R1-0528", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "DeepSeek R1-0528 model card: deepseek-r1-0528=27.8." }, { "model_id": "deepseek-r1-0528", "benchmark_id": "hle", "score": 17.7, "reference_url": "https://huggingface.co/deepseek-ai/DeepSeek-R1-0528", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "DeepSeek R1-0528 model card: deepseek-r1-0528=17.7." }, { "model_id": "deepseek-r1-0528", "benchmark_id": "codeforces_rating", "score": 1930, "reference_url": "https://huggingface.co/deepseek-ai/DeepSeek-R1-0528", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "DeepSeek R1-0528 model card: deepseek-r1-0528=1930." }, { "model_id": "deepseek-r1-0528", "benchmark_id": "hmmt_feb_2025", "score": 79.4, "reference_url": "https://huggingface.co/deepseek-ai/DeepSeek-R1-0528", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "DeepSeek R1-0528 model card: deepseek-r1-0528=79.4." }, { "model_id": "deepseek-r1-0528", "benchmark_id": "chatbot_arena_elo", "score": 1422, "reference_url": "https://lmarena.ai/leaderboard/text", "audit_status": "verified", "source_type": "leaderboard", "matches_canonical": true, "notes": "lmarena.ai text arena (fetched 2026-04-29). Arena model name: deepseek-r1-0528 (rank 74). ELO updates continuously; score reflects latest available." }, { "model_id": "deepseek-r1-0528", "benchmark_id": "frontiermath", "score": null, "reference_url": "https://epoch.ai/benchmarks/frontiermath", "audit_status": "dropped", "notes": "R5g-a ghost cell: DeepSeek-R1-0528 not found in epoch.ai FrontierMath-2025-02-28-Private leaderboard. Present in other epoch tasks (Tier 4, GPQA, MATH) but absent from main FM leaderboard. Original value 10.0 was unverified and source URL cannot substantiate it." }, { "model_id": "deepseek-r1-0528", "benchmark_id": "ifeval", "score": 79.1, "reference_url": "https://huggingface.co/Qwen/Qwen3-235B-A22B-Thinking-2507", "audit_status": "verified", "source_type": "model_card", "matches_canonical": true, "audit_note": "Qwen3-235B card: Deepseek-R1-0528 IFEval=79.1.", "audited_by": "lychee-audit-bot", "audited_at": "2026-04-29T13:17:14Z" }, { "model_id": "deepseek-r1-0528", "benchmark_id": "mmlu", "score": 90.8, "reference_url": "https://medium.com/@leucopsis/deepseeks-new-r1-0528-performance-analysis-and-benchmark-comparisons-6440eac858d6", "audit_status": "dropped", "rule_id": "R5h-third-party-blog", "notes": "DROPPED (R5h): random third-party blog/aggregator, no primary trail" }, { "model_id": "deepseek-r1-0528", "benchmark_id": "arc_agi_1", "score": 21.2, "reference_url": "https://arcprize.org/leaderboard", "audit_status": "verified", "source_type": "leaderboard", "matches_canonical": true, "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1 (single-pass)", "judge": "rule-based", "harness": "official (arcprize.org)", "system_type": "CoT" }, "notes": "arcprize.org leaderboard audit: Deepseek R1 (05/28) on leaderboard" }, { "model_id": "deepseek-r1-0528", "benchmark_id": "arc_agi_2", "score": 1.1, "reference_url": "https://arcprize.org/leaderboard", "audit_status": "verified", "source_type": "leaderboard", "matches_canonical": true, "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1 (single-pass)", "judge": "rule-based", "harness": "official (arcprize.org)", "system_type": "CoT" }, "notes": "arcprize.org leaderboard audit: Deepseek R1 (05/28) on leaderboard" }, { "model_id": "deepseek-r1-0528", "benchmark_id": "brumo_2025", "score": 92.5, "reference_url": "https://matharena.ai/?comp=brumo--brumo_2025", "audit_status": "verified", "notes": " | Verified: matharena.ai shows 92.5%" }, { "model_id": "deepseek-r1-0528", "benchmark_id": "cmimc_2025", "score": 69.38, "reference_url": "https://matharena.ai/?comp=cmimc--cmimc_2025", "audit_status": "verified", "notes": " | Verified: matharena.ai shows 69.38%" }, { "model_id": "deepseek-r1-0528", "benchmark_id": "humaneval", "score": 85.6, "reference_url": "https://medium.com/@leucopsis/deepseeks-new-r1-0528-performance-analysis-and-benchmark-comparisons-6440eac858d6", "audit_status": "dropped", "rule_id": "R5h-third-party-blog", "notes": "DROPPED (R5h): random third-party blog/aggregator, no primary trail" }, { "model_id": "deepseek-r1-0528", "benchmark_id": "imo_2025", "score": 6.85, "reference_url": "https://matharena.ai/?comp=imo--imo_2025", "audit_status": "verified", "notes": " | Verified: matharena.ai shows 6.85%" }, { "model_id": "deepseek-r1-0528", "benchmark_id": "matharena_apex_2025", "score": 1.04, "reference_url": "https://matharena.ai/?comp=apex--apex_2025", "audit_status": "verified", "notes": " | Verified: matharena.ai shows 1.04%" }, { "model_id": "deepseek-r1-0528", "benchmark_id": "smt_2025", "score": 83.02, "reference_url": "https://matharena.ai/?comp=smt--smt_2025", "audit_status": "verified", "notes": " | Verified: matharena.ai shows 83.02%" }, { "model_id": "deepseek-r1-0528", "benchmark_id": "usamo_2025", "score": 30.06, "reference_url": "https://matharena.ai/?comp=usamo--usamo_2025", "audit_status": "verified", "notes": " | Verified: matharena.ai shows 30.06%" }, { "model_id": "deepseek-r1-distill-qwen-32b", "benchmark_id": "aime_2024", "score": 72.6, "reference_url": "https://arxiv.org/abs/2501.12948", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "DeepSeek R1 paper Table 15: deepseek-r1-distill-qwen-32b=72.6." }, { "model_id": "deepseek-r1-distill-qwen-32b", "benchmark_id": "math_500", "score": 94.3, "reference_url": "https://arxiv.org/abs/2501.12948", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "DeepSeek R1 paper Table 15: deepseek-r1-distill-qwen-32b=94.3." }, { "model_id": "deepseek-r1-distill-qwen-32b", "benchmark_id": "gpqa_diamond", "score": 62.1, "reference_url": "https://arxiv.org/abs/2501.12948", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "DeepSeek R1 paper Table 15: deepseek-r1-distill-qwen-32b=62.1." }, { "model_id": "deepseek-r1-distill-qwen-32b", "benchmark_id": "livecodebench", "score": 57.2, "reference_url": "https://arxiv.org/abs/2501.12948", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "DeepSeek R1 paper Table 15: deepseek-r1-distill-qwen-32b=57.2." }, { "model_id": "deepseek-r1-distill-qwen-32b", "benchmark_id": "codeforces_rating", "score": 1691, "reference_url": "https://arxiv.org/abs/2501.12948", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "DeepSeek R1 paper Table 15: deepseek-r1-distill-qwen-32b=1691." }, { "model_id": "deepseek-r1-distill-qwen-32b", "benchmark_id": "aime_2025", "score": 72.6, "reference_url": "https://arxiv.org/html/2501.12948v1", "audit_status": "dropped", "notes": " | DROPPED: cell URL arxiv html v1 of R1 paper, but R1 paper does NOT report distill-qwen-32b AIME 2025 (Table 13 only reports R1 main). Value 72.6 matches AIME 2024 score → likely scrape error. Ghost cell per R5g(a)." }, { "model_id": "deepseek-r1-distill-qwen-32b", "benchmark_id": "humaneval", "score": 82, "reference_url": "https://arxiv.org/html/2501.12948v1", "audit_status": "dropped", "notes": " | DROPPED: cell URL arxiv html v1, but R1 paper does not report distill HumanEval. Ghost cell." }, { "model_id": "deepseek-r1-distill-qwen-14b", "benchmark_id": "aime_2024", "score": 69.7, "reference_url": "https://arxiv.org/abs/2501.12948", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "DeepSeek R1 paper Table 15: deepseek-r1-distill-qwen-14b=69.7." }, { "model_id": "deepseek-r1-distill-qwen-14b", "benchmark_id": "math_500", "score": 93.9, "reference_url": "https://arxiv.org/abs/2501.12948", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "DeepSeek R1 paper Table 15: deepseek-r1-distill-qwen-14b=93.9." }, { "model_id": "deepseek-r1-distill-qwen-14b", "benchmark_id": "codeforces_rating", "score": 1481, "reference_url": "https://arxiv.org/abs/2501.12948", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "DeepSeek R1 paper Table 15: deepseek-r1-distill-qwen-14b=1481." }, { "model_id": "deepseek-r1-distill-qwen-14b", "benchmark_id": "gpqa_diamond", "score": 59.1, "reference_url": "https://arxiv.org/abs/2501.12948", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "DeepSeek R1 paper Table 15: deepseek-r1-distill-qwen-14b=59.1." }, { "model_id": "deepseek-r1-distill-qwen-14b", "benchmark_id": "livecodebench", "score": 53.1, "reference_url": "https://arxiv.org/abs/2501.12948", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "DeepSeek R1 paper Table 15: deepseek-r1-distill-qwen-14b=53.1." }, { "model_id": "deepseek-r1-distill-qwen-7b", "benchmark_id": "aime_2024", "score": 55.5, "reference_url": "https://arxiv.org/abs/2501.12948", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "DeepSeek R1 paper Table 15: deepseek-r1-distill-qwen-7b=55.5." }, { "model_id": "deepseek-r1-distill-qwen-7b", "benchmark_id": "math_500", "score": 92.8, "reference_url": "https://arxiv.org/abs/2501.12948", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "DeepSeek R1 paper Table 15: deepseek-r1-distill-qwen-7b=92.8." }, { "model_id": "deepseek-r1-distill-qwen-7b", "benchmark_id": "gpqa_diamond", "score": 49.1, "reference_url": "https://arxiv.org/abs/2501.12948", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "DeepSeek R1 paper Table 15: deepseek-r1-distill-qwen-7b=49.1." }, { "model_id": "deepseek-r1-distill-qwen-7b", "benchmark_id": "livecodebench", "score": 37.6, "reference_url": "https://arxiv.org/abs/2501.12948", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "DeepSeek R1 paper Table 15: deepseek-r1-distill-qwen-7b=37.6." }, { "model_id": "deepseek-r1-distill-qwen-7b", "benchmark_id": "codeforces_rating", "score": 1189, "reference_url": "https://arxiv.org/abs/2501.12948", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "DeepSeek R1 paper Table 15: deepseek-r1-distill-qwen-7b=1189." }, { "model_id": "deepseek-r1-distill-qwen-1.5b", "benchmark_id": "aime_2024", "score": 28.9, "reference_url": "https://arxiv.org/abs/2501.12948", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "DeepSeek R1 paper Table 15: deepseek-r1-distill-qwen-1.5b=28.9." }, { "model_id": "deepseek-r1-distill-qwen-1.5b", "benchmark_id": "math_500", "score": 83.9, "reference_url": "https://arxiv.org/abs/2501.12948", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "DeepSeek R1 paper Table 15: deepseek-r1-distill-qwen-1.5b=83.9." }, { "model_id": "deepseek-r1-distill-qwen-1.5b", "benchmark_id": "gpqa_diamond", "score": 33.8, "reference_url": "https://arxiv.org/abs/2501.12948", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "DeepSeek R1 paper Table 15: deepseek-r1-distill-qwen-1.5b=33.8." }, { "model_id": "deepseek-r1-distill-qwen-1.5b", "benchmark_id": "livecodebench", "score": 13.2, "reference_url": "https://arxiv.org/abs/2505.09388", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Qwen3 tech report (arxiv:2505.09388) Table 19 baseline column for deepseek-r1-distill-qwen-1.5b: LiveCodeBench v5 13.2. Third-party Qwen self-test, matches_canonical=false." }, { "model_id": "deepseek-r1-distill-qwen-1.5b", "benchmark_id": "codeforces_rating", "score": 954, "reference_url": "https://arxiv.org/abs/2501.12948", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "DeepSeek R1 paper Table 15: deepseek-r1-distill-qwen-1.5b=954." }, { "model_id": "deepseek-r1-distill-qwen-1.5b", "benchmark_id": "aime_2025", "score": 22.8, "reference_url": "https://arxiv.org/abs/2505.09388", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Qwen3 tech report (arxiv:2505.09388) Table 19 baseline column for deepseek-r1-distill-qwen-1.5b: AIME 2025 22.8. Third-party Qwen self-test, matches_canonical=false." }, { "model_id": "deepseek-r1-distill-qwen-1.5b", "benchmark_id": "ifeval", "score": 39.9, "reference_url": "https://arxiv.org/abs/2505.09388", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Qwen3 tech report (arxiv:2505.09388) Table 19 baseline column for deepseek-r1-distill-qwen-1.5b: IFEval strict 39.9. Third-party Qwen self-test, matches_canonical=false." }, { "model_id": "deepseek-r1-distill-qwen-1.5b", "benchmark_id": "arena_hard", "score": 4.5, "reference_url": "https://arxiv.org/abs/2505.09388", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Qwen3 tech report (arxiv:2505.09388) Table 19 baseline column for deepseek-r1-distill-qwen-1.5b: Arena-Hard 4.5. Third-party Qwen self-test, matches_canonical=false." }, { "model_id": "deepseek-r1-distill-llama-8b", "benchmark_id": "aime_2024", "score": 50.4, "reference_url": "https://arxiv.org/abs/2501.12948", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "DeepSeek R1 paper Table 15: deepseek-r1-distill-llama-8b=50.4." }, { "model_id": "deepseek-r1-distill-llama-8b", "benchmark_id": "math_500", "score": 89.1, "reference_url": "https://arxiv.org/abs/2501.12948", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "DeepSeek R1 paper Table 15: deepseek-r1-distill-llama-8b=89.1." }, { "model_id": "deepseek-r1-distill-llama-8b", "benchmark_id": "gpqa_diamond", "score": 49.0, "reference_url": "https://arxiv.org/abs/2501.12948", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "DeepSeek R1 paper Table 15: deepseek-r1-distill-llama-8b=49.0." }, { "model_id": "deepseek-r1-distill-llama-8b", "benchmark_id": "livecodebench", "score": 42.5, "reference_url": "https://arxiv.org/abs/2505.09388", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Qwen3 tech report (arxiv:2505.09388) Table 19 baseline column for deepseek-r1-distill-llama-8b: LiveCodeBench v5 42.5. Third-party Qwen self-test, matches_canonical=false." }, { "model_id": "deepseek-r1-distill-llama-8b", "benchmark_id": "codeforces_rating", "score": 1205, "reference_url": "https://arxiv.org/abs/2501.12948", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "DeepSeek R1 paper Table 15: deepseek-r1-distill-llama-8b=1205." }, { "model_id": "deepseek-r1-distill-llama-8b", "benchmark_id": "aime_2025", "score": 27.8, "reference_url": "https://arxiv.org/abs/2505.09388", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Qwen3 tech report (arxiv:2505.09388) Table 19 baseline column for deepseek-r1-distill-llama-8b: AIME 2025 27.8. Third-party Qwen self-test, matches_canonical=false." }, { "model_id": "deepseek-r1-distill-llama-8b", "benchmark_id": "ifeval", "score": 59.0, "reference_url": "https://arxiv.org/abs/2505.09388", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Qwen3 tech report (arxiv:2505.09388) Table 19 baseline column for deepseek-r1-distill-llama-8b: IFEval strict 59.0. Third-party Qwen self-test, matches_canonical=false." }, { "model_id": "deepseek-r1-distill-llama-8b", "benchmark_id": "arena_hard", "score": 17.6, "reference_url": "https://arxiv.org/abs/2505.09388", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Qwen3 tech report (arxiv:2505.09388) Table 19 baseline column for deepseek-r1-distill-llama-8b: Arena-Hard 17.6. Third-party Qwen self-test, matches_canonical=false." }, { "model_id": "deepseek-r1-distill-llama-70b", "benchmark_id": "aime_2024", "score": 70.0, "reference_url": "https://arxiv.org/abs/2505.09388", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Qwen3 tech report (arxiv:2505.09388) Table 13 baseline column for DeepSeek-R1-Distill-Llama-70B (third-party Qwen self-test, matches_canonical=false): AIME 2024 70.0", "candidates": [ { "score": 69.3, "reference_url": "https://arxiv.org/abs/2504.21318", "source_type": "tech_report", "reported_setting": { "mode": "unknown", "effort": "unknown", "tools": "unknown", "sampling": "unknown", "judge": "rule-based", "harness": "Microsoft Phi internal eval", "prompt_style": "unknown", "temperature": "unknown" }, "notes": "Phi-4-reasoning paper Table 1 (deepseek-r1-distill-llama-70b column): aime_2024=69.3 (alt measurement, mc=false)" } ] }, { "model_id": "deepseek-r1-distill-llama-70b", "benchmark_id": "math_500", "score": 94.5, "reference_url": "https://arxiv.org/abs/2505.09388", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Qwen3 tech report (arxiv:2505.09388) Table 13 baseline column for DeepSeek-R1-Distill-Llama-70B (third-party Qwen self-test, matches_canonical=false): MATH-500 94.5" }, { "model_id": "deepseek-r1-distill-llama-70b", "benchmark_id": "gpqa_diamond", "score": 65.2, "reference_url": "https://arxiv.org/abs/2505.09388", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Qwen3 tech report (arxiv:2505.09388) Table 13 baseline column for DeepSeek-R1-Distill-Llama-70B (third-party Qwen self-test, matches_canonical=false): GPQA-Diamond 65.2", "candidates": [ { "score": 66.2, "reference_url": "https://arxiv.org/abs/2504.21318", "source_type": "tech_report", "reported_setting": { "mode": "unknown", "effort": "unknown", "tools": "unknown", "sampling": "unknown", "judge": "rule-based", "harness": "Microsoft Phi internal eval", "prompt_style": "unknown", "temperature": "unknown" }, "notes": "Phi-4-reasoning paper Table 1 (deepseek-r1-distill-llama-70b column): gpqa_diamond=66.2 (alt measurement, mc=false)" } ] }, { "model_id": "deepseek-r1-distill-llama-70b", "benchmark_id": "livecodebench", "score": 54.5, "reference_url": "https://arxiv.org/abs/2505.09388", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Qwen3 tech report (arxiv:2505.09388) Table 13 baseline column for DeepSeek-R1-Distill-Llama-70B (third-party Qwen self-test, matches_canonical=false): LiveCodeBench v5 54.5 [R5d: prior unverified value 57.5 from https://arxiv.org/abs/2501.12948 deleted.]", "candidates": [ { "score": 57.5, "reference_url": "https://arxiv.org/abs/2504.21318", "source_type": "tech_report", "reported_setting": { "mode": "unknown", "effort": "unknown", "tools": "unknown", "sampling": "unknown", "judge": "rule-based", "harness": "Microsoft Phi internal eval", "prompt_style": "unknown", "temperature": "unknown" }, "notes": "Phi-4-reasoning paper Table 1 (deepseek-r1-distill-llama-70b column): livecodebench=57.5 (alt measurement, mc=false)" } ] }, { "model_id": "deepseek-r1-distill-llama-70b", "benchmark_id": "codeforces_rating", "score": 1633, "reference_url": "https://arxiv.org/abs/2505.09388", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Qwen3 tech report (arxiv:2505.09388) Table 13 baseline column for DeepSeek-R1-Distill-Llama-70B (third-party Qwen self-test, matches_canonical=false): CodeForces rating 1633" }, { "model_id": "deepseek-r1-distill-llama-70b", "benchmark_id": "aime_2025", "score": 56.3, "reference_url": "https://arxiv.org/abs/2505.09388", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Qwen3 tech report (arxiv:2505.09388) Table 13 baseline column for DeepSeek-R1-Distill-Llama-70B (third-party Qwen self-test, matches_canonical=false): AIME 2025 56.3 [R5d: prior unverified value 70 from https://arxiv.org/html/2501.12948v1 deleted.]", "candidates": [ { "score": 51.5, "reference_url": "https://arxiv.org/abs/2504.21318", "source_type": "tech_report", "reported_setting": { "mode": "unknown", "effort": "unknown", "tools": "unknown", "sampling": "unknown", "judge": "rule-based", "harness": "Microsoft Phi internal eval", "prompt_style": "unknown", "temperature": "unknown" }, "notes": "Phi-4-reasoning paper Table 1 (deepseek-r1-distill-llama-70b column): aime_2025=51.5 (alt measurement, mc=false)" } ] }, { "model_id": "deepseek-r1-distill-llama-70b", "benchmark_id": "hmmt_feb_2025", "score": 33.3, "reference_url": "https://arxiv.org/abs/2504.21318", "reported_setting": { "mode": "unknown", "effort": "unknown", "tools": "unknown", "sampling": "unknown", "judge": "rule-based", "harness": "Microsoft Phi internal eval", "prompt_style": "unknown", "temperature": "unknown" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Phi-4-reasoning paper Table 1 (deepseek-r1-distill-llama-70b column): hmmt_2025=33.3 (re-sourced from https://arxiv.org/abs/2504.21318, matches BP value)" }, { "model_id": "deepseek-r1-distill-llama-70b", "benchmark_id": "humaneval", "score": 80, "reference_url": "https://arxiv.org/html/2501.12948v1", "audit_status": "dropped", "rule_id": "R5g-ghost", "audit_note": "arxiv 2501.12948 has no standalone HumanEval table for distill models. Paper uses LiveCodeBench/Codeforces.", "audited_by": "lychee-audit-bot", "audited_at": "2026-04-29T13:17:14Z" }, { "model_id": "deepseek-v3.2", "benchmark_id": "aime_2025", "score": 93.1, "reference_url": "https://huggingface.co/deepseek-ai/DeepSeek-V3.2/resolve/main/assets/paper.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "1.0" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "DeepSeek V3.2 tech report Table 2: deepseek-v3.2=93.1.", "candidates": [ { "score": 89.3, "reference_url": "https://huggingface.co/moonshotai/Kimi-K2-Thinking", "source_type": "model_card", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "1.0" }, "notes": "Kimi K2-Thinking model card: deepseek-v3.2=89.3." }, { "score": 88, "reference_url": "https://huggingface.co/MiniMaxAI/MiniMax-M2", "source_type": "model_card", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "notes": "MiniMax M2 model card: deepseek-v3.2=88." }, { "score": 93.1, "reference_url": "https://llm-stats.com/benchmarks/aime-2025", "source_type": "third_party_aggregator", "reported_setting": { "mode": "thinking (DeepSeek-Reasoner endpoint)" }, "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=DeepSeek-V3.2 (Thinking), slug=deepseek-reasoner, provider=DeepSeek" }, { "score": 89.3, "reference_url": "https://llm-stats.com/benchmarks/aime-2025", "source_type": "third_party_aggregator", "reported_setting": { "variant": "exp" }, "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=DeepSeek-V3.2-Exp, slug=deepseek-v3.2-exp, provider=DeepSeek" }, { "score": 93.1, "reference_url": "https://microsoft.ai/wp-content/uploads/2026/06/main_20260602_2.pdf", "source_type": "official_paper", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none (per-bench override)", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "1.0", "context": "128K", "notes": "Per DeepSeek V3.2 tech report: temperature=1.0, context=128K, thinking mode for tool-use." }, "notes": "Displayed exactly as 93.1. Official Microsoft-reported result. Comparator value is quoted from that provider's official model card/release; use the campaign canonical setting." } ] }, { "model_id": "deepseek-v3.2", "benchmark_id": "gpqa_diamond", "score": 82.4, "reference_url": "https://huggingface.co/deepseek-ai/DeepSeek-V3.2/resolve/main/assets/paper.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "1.0" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "DeepSeek V3.2 tech report Table 2: deepseek-v3.2=82.4.", "candidates": [ { "score": 79.9, "reference_url": "https://huggingface.co/moonshotai/Kimi-K2-Thinking", "source_type": "model_card", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "1.0" }, "notes": "Kimi K2-Thinking model card: deepseek-v3.2=79.9." }, { "score": 80, "reference_url": "https://huggingface.co/MiniMaxAI/MiniMax-M2", "source_type": "model_card", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "notes": "MiniMax M2 model card: deepseek-v3.2=80." }, { "score": 82.4, "reference_url": "https://llm-stats.com/benchmarks/gpqa", "source_type": "third_party_aggregator", "reported_setting": { "mode": "thinking (DeepSeek-Reasoner endpoint)" }, "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=DeepSeek-V3.2 (Thinking), slug=deepseek-reasoner, provider=DeepSeek" }, { "score": 79.9, "reference_url": "https://llm-stats.com/benchmarks/gpqa", "source_type": "third_party_aggregator", "reported_setting": { "variant": "exp" }, "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=DeepSeek-V3.2-Exp, slug=deepseek-v3.2-exp, provider=DeepSeek" }, { "score": 31.82, "reference_url": "https://research.nvidia.com/labs/nemotron/files/NVIDIA-Nemotron-3-Ultra-Technical-Report.pdf", "source_type": "tech_report", "reported_setting": { "version_metric": "5-shot CoT exact match", "tools": "none", "harness": "NeMo Evaluator", "sampling": "pass@1", "checkpoint": "base", "mode": "pretrained" }, "notes": "NVIDIA Nemotron 3 Ultra technical report, Base Table 2: DeepSeek-V3.2-Exp-Base: gpqa_diamond=31.82. Base checkpoint score mapped to the canonical post-trained model row as checkpoint=base." }, { "score": 82.4, "reference_url": "https://microsoft.ai/wp-content/uploads/2026/06/main_20260602_2.pdf", "source_type": "official_paper", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none (per-bench override)", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "1.0", "context": "128K", "notes": "Per DeepSeek V3.2 tech report: temperature=1.0, context=128K, thinking mode for tool-use." }, "notes": "Displayed exactly as 82.4. Official Microsoft-reported result. Comparator value is quoted from that provider's official model card/release; use the campaign canonical setting." } ] }, { "model_id": "deepseek-v3.2", "benchmark_id": "mmlu_pro", "score": 85.0, "reference_url": "https://huggingface.co/deepseek-ai/DeepSeek-V3.2/resolve/main/assets/paper.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "1.0" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "DeepSeek V3.2 tech report Table 2: deepseek-v3.2=85.0.", "candidates": [ { "score": 85.0, "reference_url": "https://llm-stats.com/benchmarks/mmlu-pro", "source_type": "third_party_aggregator", "reported_setting": { "variant": "exp" }, "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=DeepSeek-V3.2-Exp, slug=deepseek-v3.2-exp, provider=DeepSeek" }, { "score": 85.0, "reference_url": "https://llm-stats.com/benchmarks/mmlu-pro", "source_type": "third_party_aggregator", "reported_setting": { "mode": "thinking (DeepSeek-Reasoner endpoint)" }, "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=DeepSeek-V3.2 (Thinking), slug=deepseek-reasoner, provider=DeepSeek" }, { "score": 63.26, "reference_url": "https://research.nvidia.com/labs/nemotron/files/NVIDIA-Nemotron-3-Ultra-Technical-Report.pdf", "source_type": "tech_report", "reported_setting": { "version_metric": "5-shot CoT exact match", "tools": "none", "harness": "NeMo Evaluator", "sampling": "pass@1", "checkpoint": "base", "mode": "pretrained" }, "notes": "NVIDIA Nemotron 3 Ultra technical report, Base Table 2: DeepSeek-V3.2-Exp-Base: mmlu_pro=63.26. Base checkpoint score mapped to the canonical post-trained model row as checkpoint=base." } ] }, { "model_id": "deepseek-v3.2", "benchmark_id": "livecodebench", "score": 83.3, "reference_url": "https://huggingface.co/deepseek-ai/DeepSeek-V3.2/resolve/main/assets/paper.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "1.0" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "DeepSeek V3.2 tech report Table 2: deepseek-v3.2=83.3.", "candidates": [ { "score": 79, "reference_url": "https://huggingface.co/MiniMaxAI/MiniMax-M2", "source_type": "model_card", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "notes": "MiniMax M2 model card: deepseek-v3.2=79." }, { "score": 83.3, "reference_url": "https://llm-stats.com/benchmarks/livecodebench", "source_type": "third_party_aggregator", "reported_setting": { "mode": "thinking (DeepSeek-Reasoner endpoint)" }, "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=DeepSeek-V3.2 (Thinking), slug=deepseek-reasoner, provider=DeepSeek" }, { "score": 74.1, "reference_url": "https://llm-stats.com/benchmarks/livecodebench", "source_type": "third_party_aggregator", "reported_setting": { "variant": "exp" }, "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=DeepSeek-V3.2-Exp, slug=deepseek-v3.2-exp, provider=DeepSeek" } ] }, { "model_id": "deepseek-v3.2", "benchmark_id": "codeforces_rating", "score": 2386, "reference_url": "https://huggingface.co/deepseek-ai/DeepSeek-V3.2/resolve/main/assets/paper.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "1.0" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "DeepSeek V3.2 tech report Table 2: deepseek-v3.2=2386.", "candidates": [ { "score": 79.5, "reference_url": "https://llm-stats.com/benchmarks/codeforces", "source_type": "third_party_aggregator", "reported_setting": { "mode": "thinking (DeepSeek-Reasoner endpoint)" }, "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=DeepSeek-V3.2 (Thinking), slug=deepseek-reasoner, provider=DeepSeek" }, { "score": 70.7, "reference_url": "https://llm-stats.com/benchmarks/codeforces", "source_type": "third_party_aggregator", "reported_setting": { "variant": "exp" }, "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=DeepSeek-V3.2-Exp, slug=deepseek-v3.2-exp, provider=DeepSeek" } ] }, { "model_id": "deepseek-v3.2", "benchmark_id": "hle", "score": 25.1, "reference_url": "https://huggingface.co/deepseek-ai/DeepSeek-V3.2/resolve/main/assets/paper.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "1.0" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "DeepSeek V3.2 tech report Table 2: deepseek-v3.2=25.1.", "candidates": [ { "score": 19.8, "reference_url": "https://huggingface.co/moonshotai/Kimi-K2-Thinking", "source_type": "model_card", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "1.0" }, "notes": "Kimi K2-Thinking model card: deepseek-v3.2=19.8." }, { "score": 25.1, "reference_url": "https://llm-stats.com/benchmarks/humanity's-last-exam", "source_type": "third_party_aggregator", "reported_setting": { "mode": "thinking (DeepSeek-Reasoner endpoint)" }, "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=DeepSeek-V3.2 (Thinking), slug=deepseek-reasoner, provider=DeepSeek" }, { "score": 19.8, "reference_url": "https://llm-stats.com/benchmarks/humanity's-last-exam", "source_type": "third_party_aggregator", "reported_setting": { "variant": "exp" }, "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=DeepSeek-V3.2-Exp, slug=deepseek-v3.2-exp, provider=DeepSeek" } ] }, { "model_id": "deepseek-v3.2", "benchmark_id": "swe_bench_verified", "score": 73.1, "reference_url": "https://huggingface.co/deepseek-ai/DeepSeek-V3.2/resolve/main/assets/paper.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "agentic (bash + edit)", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "1.0" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "DeepSeek V3.2 tech report Table 2: deepseek-v3.2=73.1.", "candidates": [ { "score": 67.8, "reference_url": "https://huggingface.co/MiniMaxAI/MiniMax-M2", "source_type": "model_card", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "agentic (bash + edit)", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "notes": "MiniMax M2 model card: deepseek-v3.2=67.8." }, { "score": 73.1, "reference_url": "https://llm-stats.com/benchmarks/swe-bench-verified", "source_type": "third_party_aggregator", "reported_setting": { "mode": "thinking (DeepSeek-Reasoner endpoint)" }, "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=DeepSeek-V3.2 (Thinking), slug=deepseek-reasoner, provider=DeepSeek" }, { "score": 67.8, "reference_url": "https://llm-stats.com/benchmarks/swe-bench-verified", "source_type": "third_party_aggregator", "reported_setting": { "variant": "exp" }, "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=DeepSeek-V3.2-Exp, slug=deepseek-v3.2-exp, provider=DeepSeek" }, { "score": 73.1, "reference_url": "https://microsoft.ai/wp-content/uploads/2026/06/main_20260602_2.pdf", "source_type": "official_paper", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none (per-bench override)", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "1.0", "context": "128K", "notes": "Per DeepSeek V3.2 tech report: temperature=1.0, context=128K, thinking mode for tool-use." }, "notes": "Displayed exactly as 73.1. Official Microsoft-reported result. Comparator value is quoted from that provider's official model card/release; use the campaign canonical setting." } ] }, { "model_id": "deepseek-v3.2", "benchmark_id": "browsecomp", "score": 51.4, "reference_url": "https://huggingface.co/deepseek-ai/DeepSeek-V3.2/resolve/main/assets/paper.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "agentic", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "1.0" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "DeepSeek V3.2 tech report Table 2: deepseek-v3.2=51.4.", "candidates": [ { "score": 40.1, "reference_url": "https://huggingface.co/MiniMaxAI/MiniMax-M2", "source_type": "model_card", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "agentic", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "notes": "MiniMax M2 model card: deepseek-v3.2=40.1." }, { "score": 51.4, "reference_url": "https://llm-stats.com/benchmarks/browsecomp", "source_type": "third_party_aggregator", "reported_setting": { "mode": "thinking (DeepSeek-Reasoner endpoint)" }, "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=DeepSeek-V3.2 (Thinking), slug=deepseek-reasoner, provider=DeepSeek" }, { "score": 40.1, "reference_url": "https://llm-stats.com/benchmarks/browsecomp", "source_type": "third_party_aggregator", "reported_setting": { "variant": "exp" }, "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=DeepSeek-V3.2-Exp, slug=deepseek-v3.2-exp, provider=DeepSeek" } ] }, { "model_id": "deepseek-v3.2", "benchmark_id": "mmmu_pro", "score": 81.0, "reference_url": "https://www.bentoml.com/blog/the-complete-guide-to-deepseek-models-from-v3-to-r1-and-beyond", "audit_status": "dropped", "rule_id": "R5h-third-party-blog", "notes": "DROPPED (R5h): random third-party blog/aggregator, no primary trail" }, { "model_id": "deepseek-v3.2", "benchmark_id": "math_500", "score": 97.3, "reference_url": "https://artificialanalysis.ai/evaluations/math-500", "audit_status": "needs_review", "notes": "Reference URL is AA math-500 eval page, but deepseek-v3-2 is NOT present on that page; AA shows math_500=null for this model. Possible ghost score (R5g-a). Cannot verify 97.3% from AA source." }, { "model_id": "deepseek-v3.2", "benchmark_id": "frontiermath", "score": 22.1, "reference_url": "https://epoch.ai/benchmarks/frontiermath", "audit_status": "verified", "source_type": "leaderboard", "matches_canonical": true, "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "harness": "Fireworks API" }, "notes": "epoch.ai FrontierMath-2025-02-28-Private: fireworks/deepseek-v3p2 (DeepSeek-V3.2 Thinking; Fireworks) = 22.1%. Canonical thinking mode matches. Note: epoch ran on Fireworks due to high API error rate from DeepSeek API, but results confirmed equivalent. BP had 8.0." }, { "model_id": "deepseek-v3.2", "benchmark_id": "simpleqa", "score": 35.0, "reference_url": "https://artificialanalysis.ai/models/deepseek-v3-2", "audit_status": "needs_review", "notes": "Reference URL is AA model page (deepseek-v3-2). AA does not publish SimpleQA scores; field is not present in AA model data for any model. Cannot verify from AA source. Original source needs identification.", "candidates": [ { "score": 97.1, "reference_url": "https://llm-stats.com/benchmarks/simpleqa", "source_type": "third_party_aggregator", "reported_setting": { "variant": "exp" }, "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=DeepSeek-V3.2-Exp, slug=deepseek-v3.2-exp, provider=DeepSeek" } ] }, { "model_id": "deepseek-v3.2", "benchmark_id": "terminal_bench", "score": 46.4, "reference_url": "https://huggingface.co/deepseek-ai/DeepSeek-V3.2/resolve/main/assets/paper.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "agentic", "sampling": "pass@1", "judge": "rule-based", "harness": "Claude Code framework", "prompt_style": "default", "temperature": "1.0" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "DeepSeek V3.2 tech report Table 2: deepseek-v3.2=46.4.", "candidates": [ { "score": 39.3, "reference_url": "https://z.ai/blog/glm-5.1", "source_type": "official_blog", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "agentic", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "notes": "GLM-5.1 blog: deepseek-v3.2=39.3." }, { "score": 39.3, "reference_url": "https://z.ai/blog/glm-5", "source_type": "official_blog", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "agentic", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "notes": "GLM-5 blog: deepseek-v3.2=39.3." }, { "score": 37.7, "reference_url": "https://huggingface.co/MiniMaxAI/MiniMax-M2", "source_type": "model_card", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "agentic", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "notes": "MiniMax M2 model card: deepseek-v3.2=37.7." }, { "score": 39.6, "reference_url": "https://www.tbench.ai/leaderboard/terminal-bench/2.0", "source_type": "leaderboard", "notes": "tbench 2.0 Terminus 2 result; BP primary (46.4%) from HuggingFace model card differs." }, { "score": 46.4, "reference_url": "https://llm-stats.com/benchmarks/terminal-bench-2", "source_type": "third_party_aggregator", "reported_setting": { "mode": "thinking (DeepSeek-Reasoner endpoint)" }, "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=DeepSeek-V3.2 (Thinking), slug=deepseek-reasoner, provider=DeepSeek" }, { "score": 46.4, "reference_url": "https://microsoft.ai/wp-content/uploads/2026/06/main_20260602_2.pdf", "source_type": "official_paper", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none (per-bench override)", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "1.0", "context": "128K", "notes": "Per DeepSeek V3.2 tech report: temperature=1.0, context=128K, thinking mode for tool-use." }, "notes": "Displayed exactly as 46.4. Official Microsoft-reported result. Comparator value is quoted from that provider's official model card/release; use the campaign canonical setting." } ] }, { "model_id": "deepseek-v3.2", "benchmark_id": "aime_2024", "score": 93, "reference_url": "https://introl.com/blog/deepseek-v3-2-open-source-ai-cost-advantage", "audit_status": "dropped", "rule_id": "R5h-third-party-blog", "notes": "DROPPED (R5h): random third-party blog/aggregator, no primary trail" }, { "model_id": "deepseek-v3.2", "benchmark_id": "brumo_2025", "score": 96.67, "reference_url": "https://matharena.ai/?comp=brumo--brumo_2025", "audit_status": "verified", "matches_canonical": true, "reported_setting": "thinking:default", "source_type": "leaderboard" }, { "model_id": "deepseek-v3.2", "benchmark_id": "cmimc_2025", "score": 83.75, "reference_url": "https://matharena.ai/?comp=cmimc--cmimc_2025", "audit_status": "verified", "matches_canonical": true, "reported_setting": "thinking:default", "source_type": "leaderboard" }, { "model_id": "deepseek-v3.2", "benchmark_id": "hmmt_nov_2025", "score": 90.2, "reference_url": "https://huggingface.co/deepseek-ai/DeepSeek-V3.2/resolve/main/assets/paper.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "1.0" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "DeepSeek V3.2 tech report Table 2: deepseek-v3.2=90.2." }, { "model_id": "deepseek-v3.2", "benchmark_id": "humaneval", "score": 90, "reference_url": "https://introl.com/blog/deepseek-v3-2-open-source-ai-cost-advantage", "audit_status": "dropped", "rule_id": "R5h-third-party-blog", "notes": "DROPPED (R5h): random third-party blog/aggregator, no primary trail", "candidates": [ { "score": 61.85, "reference_url": "https://research.nvidia.com/labs/nemotron/files/NVIDIA-Nemotron-3-Ultra-Technical-Report.pdf", "source_type": "tech_report", "reported_setting": { "version_metric": "EvalPlus sanitized sampled pass@1 n=32", "tools": "none", "harness": "NeMo Evaluator", "sampling": "164 problems x 32 samples", "checkpoint": "base", "mode": "pretrained" }, "notes": "NVIDIA Nemotron 3 Ultra technical report, Base Table 2: DeepSeek-V3.2-Exp-Base: humaneval=61.85. Base checkpoint score mapped to the canonical post-trained model row as checkpoint=base." } ] }, { "model_id": "deepseek-v3.2", "benchmark_id": "ifeval", "score": 89, "reference_url": "https://introl.com/blog/deepseek-v3-2-open-source-ai-cost-advantage", "audit_status": "dropped", "rule_id": "R5h-third-party-blog", "notes": "DROPPED (R5h): random third-party blog/aggregator, no primary trail" }, { "model_id": "deepseek-v3.2", "benchmark_id": "livebench", "score": 63.1, "reference_url": "https://livebench.ai/", "audit_status": "verified", "source_type": "leaderboard", "matches_canonical": true, "reported_setting": { "mode": "thinking", "effort": "default", "tools": "n/a" }, "notes": "livebench.ai table_2026_01_08 (latest as of 2026-04-28): deepseek-v3.2-thinking avg=63.1. Canonical thinking mode matches. BP had 62.2." }, { "model_id": "deepseek-v3.2", "benchmark_id": "matharena_apex_2025", "score": 2.08, "reference_url": "https://matharena.ai/?comp=apex--apex_2025", "audit_status": "verified", "matches_canonical": true, "reported_setting": "thinking:default", "source_type": "leaderboard" }, { "model_id": "deepseek-v3.2", "benchmark_id": "arc_agi_1", "score": 57.0, "reference_url": "https://arcprize.org/leaderboard", "audit_status": "verified", "source_type": "leaderboard", "matches_canonical": true, "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1 (single-pass)", "judge": "rule-based", "harness": "official (arcprize.org)", "system_type": "Base LLM" }, "notes": "arcprize.org leaderboard audit: Deepseek V3.2 Base LLM on leaderboard" }, { "model_id": "deepseek-v3.2", "benchmark_id": "arc_agi_2", "score": 4.0, "reference_url": "https://arcprize.org/leaderboard", "audit_status": "verified", "source_type": "leaderboard", "matches_canonical": true, "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1 (single-pass)", "judge": "rule-based", "harness": "official (arcprize.org)", "system_type": "Base LLM" }, "notes": "arcprize.org leaderboard audit: Deepseek V3.2 Base LLM on leaderboard" }, { "model_id": "deepseek-v3.2", "benchmark_id": "mmlu", "score": 90.5, "reference_url": "https://introl.com/blog/deepseek-v3-2-open-source-ai-cost-advantage", "audit_status": "dropped", "rule_id": "R5h-third-party-blog", "notes": "DROPPED (R5h): random third-party blog/aggregator, no primary trail", "candidates": [ { "score": 87.82, "reference_url": "https://research.nvidia.com/labs/nemotron/files/NVIDIA-Nemotron-3-Ultra-Technical-Report.pdf", "source_type": "tech_report", "reported_setting": { "version_metric": "5-shot accuracy", "tools": "none", "harness": "NeMo Evaluator", "sampling": "pass@1", "checkpoint": "base", "mode": "pretrained" }, "notes": "NVIDIA Nemotron 3 Ultra technical report, Base Table 2: DeepSeek-V3.2-Exp-Base: mmlu=87.82. Base checkpoint score mapped to the canonical post-trained model row as checkpoint=base." } ] }, { "model_id": "deepseek-v3.2", "benchmark_id": "smt_2025", "score": 87.74, "reference_url": "https://matharena.ai/?comp=smt--smt_2025", "audit_status": "verified", "matches_canonical": true, "reported_setting": "thinking:default", "source_type": "leaderboard" }, { "model_id": "deepseek-v3.2-speciale", "benchmark_id": "aime_2025", "score": 96.0, "reference_url": "https://huggingface.co/deepseek-ai/DeepSeek-V3.2/resolve/main/assets/paper.pdf", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "1.0" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "DS V3.2 tech report Table 3: V3.2-Speciale=96.0." }, { "model_id": "deepseek-v3.2-speciale", "benchmark_id": "gpqa_diamond", "score": 85.7, "reference_url": "https://huggingface.co/deepseek-ai/DeepSeek-V3.2/resolve/main/assets/paper.pdf", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "1.0" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "DS V3.2 tech report Table 3: V3.2-Speciale=85.7." }, { "model_id": "deepseek-v3.2-speciale", "benchmark_id": "livecodebench", "score": 88.7, "reference_url": "https://huggingface.co/deepseek-ai/DeepSeek-V3.2/resolve/main/assets/paper.pdf", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "1.0" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "DS V3.2 tech report Table 3: V3.2-Speciale=88.7." }, { "model_id": "deepseek-v3.2-speciale", "benchmark_id": "codeforces_rating", "score": 2701, "reference_url": "https://huggingface.co/deepseek-ai/DeepSeek-V3.2/resolve/main/assets/paper.pdf", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "1.0" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "DS V3.2 tech report Table 3: V3.2-Speciale=2701." }, { "model_id": "deepseek-v3.2-speciale", "benchmark_id": "hmmt_feb_2025", "score": 99.2, "reference_url": "https://huggingface.co/deepseek-ai/DeepSeek-V3.2/resolve/main/assets/paper.pdf", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "1.0" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "DS V3.2 tech report Table 3: V3.2-Speciale=99.2." }, { "model_id": "deepseek-v3.2-speciale", "benchmark_id": "hle", "score": 30.6, "reference_url": "https://huggingface.co/deepseek-ai/DeepSeek-V3.2/resolve/main/assets/paper.pdf", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "1.0" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "DS V3.2 tech report Table 3: V3.2-Speciale=30.6." }, { "model_id": "deepseek-v3.2-speciale", "benchmark_id": "swe_bench_verified", "score": 76.0, "reference_url": "https://arxiv.org/html/2512.02556v1", "audit_status": "flagged", "notes": " | FLAGGED: arxiv 2512.02556v1 is DeepSeek V3.2 paper; paper Table 3 has no SWE-bench for Speciale; 76.0 in paper = GPT-5 IMOAnswerBench score, not Speciale SWE-bench" }, { "model_id": "deepseek-v3.2-speciale", "benchmark_id": "aime_2024", "score": 96, "reference_url": "https://medium.com/@leucopsis/deepseek-v3-2-speciale-open-weights-reasoning-close-to-the-frontier-models-d43cd5da22d9", "audit_status": "dropped", "notes": "| R5h: medium.com blog post; no primary technical report citation; needs human review DROPPED (R5h): random third-party blog/aggregator, no primary trail", "rule_id": "R5h-third-party-blog" }, { "model_id": "deepseek-v3.2-speciale", "benchmark_id": "brumo_2025", "score": 99.17, "reference_url": "https://matharena.ai/?comp=brumo--brumo_2025", "audit_status": "verified", "notes": " | Verified: matharena.ai shows 99.17%" }, { "model_id": "deepseek-v3.2-speciale", "benchmark_id": "cmimc_2025", "score": 94.38, "reference_url": "https://matharena.ai/?comp=cmimc--cmimc_2025", "audit_status": "verified", "notes": " | Verified: matharena.ai shows 94.38%" }, { "model_id": "deepseek-v3.2-speciale", "benchmark_id": "hmmt_nov_2025", "score": 94.4, "reference_url": "https://huggingface.co/deepseek-ai/DeepSeek-V3.2/resolve/main/assets/paper.pdf", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "1.0" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "DS V3.2 tech report Table 3: V3.2-Speciale=94.4." }, { "model_id": "deepseek-v3.2-speciale", "benchmark_id": "humaneval", "score": 91.5, "reference_url": "https://llm-stats.com/models/deepseek-v3.2-speciale", "audit_status": "dropped", "notes": "| R5h: llm-stats.com is an aggregator; no primary citation; needs human review DROPPED (R5h): random third-party blog/aggregator, no primary trail", "rule_id": "R5h-third-party-blog" }, { "model_id": "deepseek-v3.2-speciale", "benchmark_id": "ifeval", "score": 88, "reference_url": "https://llm-stats.com/models/deepseek-v3.2-speciale", "audit_status": "dropped", "notes": "| R5h: llm-stats.com is an aggregator; no primary citation; needs human review DROPPED (R5h): random third-party blog/aggregator, no primary trail", "rule_id": "R5h-third-party-blog" }, { "model_id": "deepseek-v3.2-speciale", "benchmark_id": "imo_2025", "score": 83.3, "reference_url": "https://huggingface.co/deepseek-ai/DeepSeek-V3.2/resolve/main/assets/paper.pdf", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "none", "sampling": "competition", "judge": "official", "harness": "official", "prompt_style": "generate-verify-refine", "temperature": "1.0", "context": "128K", "notes": "Used generate-verify-refine loop per Shao et al. 2025; 35/42=83.3%" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "DS V3.2 tech report Table 4: Speciale IMO 2025 = 35/42 pts = 83.3% (Gold medal). Generate-verify-refine loop, max 128K tokens, no tools.", "candidates": [] }, { "model_id": "deepseek-v3.2-speciale", "benchmark_id": "math_500", "score": 98, "reference_url": "https://llm-stats.com/models/deepseek-v3.2-speciale", "audit_status": "dropped", "notes": "| R5h: llm-stats.com is an aggregator; no primary citation; needs human review DROPPED (R5h): random third-party blog/aggregator, no primary trail", "rule_id": "R5h-third-party-blog" }, { "model_id": "deepseek-v3.2-speciale", "benchmark_id": "matharena_apex_2025", "score": 9.38, "reference_url": "https://matharena.ai/?comp=apex--apex_2025", "audit_status": "verified", "notes": " | Verified: matharena.ai shows 9.38%" }, { "model_id": "deepseek-v3.2-speciale", "benchmark_id": "mmlu_pro", "score": 87.5, "reference_url": "https://llm-stats.com/models/deepseek-v3.2-speciale", "audit_status": "dropped", "notes": "| R5h: llm-stats.com is an aggregator; no primary citation; needs human review DROPPED (R5h): random third-party blog/aggregator, no primary trail", "rule_id": "R5h-third-party-blog" }, { "model_id": "deepseek-v3.2-speciale", "benchmark_id": "smt_2025", "score": 89.15, "reference_url": "https://matharena.ai/?comp=smt--smt_2025", "audit_status": "verified", "notes": " | Verified: matharena.ai shows 89.15%" }, { "model_id": "deepseek-v4-pro", "benchmark_id": "mmlu_pro", "score": 87.5, "reference_url": "https://huggingface.co/deepseek-ai/DeepSeek-V4-Pro", "reported_setting": { "mode": "thinking", "effort": "Max", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "DeepSeek V4 model card: deepseek-v4-pro (Max mode) = 87.5.", "candidates": [ { "score": 87.5, "reference_url": "https://llm-stats.com/benchmarks/mmlu-pro", "source_type": "third_party_aggregator", "reported_setting": { "effort": "max" }, "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=DeepSeek-V4-Pro-Max, slug=deepseek-v4-pro-max, provider=DeepSeek" }, { "score": 86.86, "reference_url": "https://huggingface.co/internlm/Intern-S2-Preview-397B", "source_type": "model_card", "reported_setting": { "mode": "source does not state", "effort": "source does not state", "tools": "none", "sampling": "source does not state", "judge": "benchmark-specific; exact version not stated", "harness": "OpenCompass or AgentCompass; exact adapter not stated", "prompt_style": "source does not state", "temperature": "source does not state", "context": "source does not state", "notes": "Shanghai AI Lab evaluated all model columns using OpenCompass, VLMEvalKit, and AgentCompass. Per-model mode/effort/sampling are not reported; only Intern-S2 maximum evaluation lengths are disclosed." }, "notes": "Intern-S2-Preview-397B official model card: MMLU Pro = 86.86. This is a Shanghai AI Lab cross-model evaluation; the source does not disclose this model's exact mode, effort, or sampling." }, { "score": 87.1, "reference_url": "https://research.nvidia.com/labs/nemotron/files/NVIDIA-Nemotron-3-Ultra-Technical-Report.pdf", "source_type": "tech_report", "reported_setting": { "version_metric": "no tools / %", "tools": "none", "harness": "NeMo Evaluator", "sampling": "pass@1", "variant": "High", "effort": "High" }, "notes": "NVIDIA Nemotron 3 Ultra technical report, Internal Table 3: DeepSeek V4 Pro (High): mmlu_pro=87.1, variant=High." }, { "score": 87.5, "reference_url": "https://ernie.baidu.com/blog/posts/ernie-5.1-0508-release/benchmark.png", "source_type": "official_blog", "reported_setting": { "metric": "accuracy" }, "notes": "Accuracy row matches MMLU-Pro; source does not disclose tools. Research observation obs-026. Literal marker ● has no source legend. Marker has no legend in image or article." }, { "score": 73.5, "reference_url": "https://huggingface.co/XiaomiMiMo/MiMo-V2.5-Pro/resolve/21d1ecfecd7bd70f31be25ca49d7edd21f003659/README.md", "source_type": "official_model_card_base_checkpoint", "reported_setting": { "mode": "base pretrained checkpoint", "effort": "not applicable", "tools": "none", "sampling": "5-shot; single completion per prompt", "judge": "benchmark-native rule-based evaluator", "harness": "Xiaomi MiMo base-checkpoint evaluation", "prompt_style": "5-shot few-shot continuation", "temperature": "source does not state", "context": "source does not state", "input_modalities": "text", "trials": "one completion per item", "checkpoint": "base" }, "notes": "Base checkpoint observation attached to the canonical post-trained model ID. Base result is attached to the canonical post-trained model ID. The Xiaomi table does not state this comparator Base checkpoint's context length, and no comparator Base config is locked; no Pro context value or conflict is copied onto this observation." } ] }, { "model_id": "deepseek-v4-pro", "benchmark_id": "gpqa_diamond", "score": 90.1, "reference_url": "https://huggingface.co/deepseek-ai/DeepSeek-V4-Pro", "reported_setting": { "mode": "thinking", "effort": "Max", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "DeepSeek V4 model card: deepseek-v4-pro (Max mode) = 90.1.", "candidates": [ { "score": 87.8, "reference_url": "https://research.nvidia.com/labs/nemotron/files/NVIDIA-Nemotron-3-Ultra-Technical-Report.pdf", "source_type": "tech_report", "reported_setting": { "version_metric": "no tools / %", "tools": "none", "harness": "NeMo Evaluator", "sampling": "8-run average" }, "notes": "NVIDIA Nemotron 3 Ultra technical report, Final Table 10: DeepSeek-V4-Pro: gpqa_diamond=87.8." }, { "score": 89.1, "reference_url": "https://research.nvidia.com/labs/nemotron/files/NVIDIA-Nemotron-3-Ultra-Technical-Report.pdf", "source_type": "tech_report", "reported_setting": { "version_metric": "no tools / %", "tools": "none", "harness": "NeMo Evaluator", "sampling": "pass@1", "variant": "High", "effort": "High" }, "notes": "NVIDIA Nemotron 3 Ultra technical report, Internal Table 3: DeepSeek V4 Pro (High): gpqa_diamond=89.1, variant=High." }, { "score": 90.1, "reference_url": "https://microsoft.ai/wp-content/uploads/2026/06/main_20260602_2.pdf", "source_type": "official_paper", "reported_setting": { "mode": "thinking", "effort": "Max", "tools": "none (per-bench override)", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default", "context": "default", "notes": "Per DeepSeek V4-Pro model card. Three reasoning modes available (Non-Think/High/Max); BP canonical = Max mode (most powerful). Other modes' values stored as candidates if needed." }, "notes": "Displayed exactly as 90.1. Official Microsoft-reported result. Comparator value is quoted from that provider's official model card/release; use the campaign canonical setting." } ] }, { "model_id": "deepseek-v4-pro", "benchmark_id": "livecodebench", "score": 93.5, "reference_url": "https://huggingface.co/deepseek-ai/DeepSeek-V4-Pro", "reported_setting": { "mode": "thinking", "effort": "Max", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "DeepSeek V4 model card: deepseek-v4-pro (Max mode) = 93.5.", "candidates": [ { "score": 93.5, "reference_url": "https://llm-stats.com/benchmarks/livecodebench", "source_type": "third_party_aggregator", "reported_setting": { "effort": "max" }, "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=DeepSeek-V4-Pro-Max, slug=deepseek-v4-pro-max, provider=DeepSeek" } ] }, { "model_id": "deepseek-v4-pro", "benchmark_id": "codeforces_rating", "score": 3206, "reference_url": "https://huggingface.co/deepseek-ai/DeepSeek-V4-Pro", "reported_setting": { "mode": "thinking", "effort": "Max", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "DeepSeek V4 model card: deepseek-v4-pro (Max mode) = 3206.", "candidates": [ { "score": 100.0, "reference_url": "https://llm-stats.com/benchmarks/codeforces", "source_type": "third_party_aggregator", "reported_setting": { "effort": "max" }, "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=DeepSeek-V4-Pro-Max, slug=deepseek-v4-pro-max, provider=DeepSeek" } ] }, { "model_id": "deepseek-v4-pro", "benchmark_id": "swe_bench_verified", "score": 80.6, "reference_url": "https://huggingface.co/deepseek-ai/DeepSeek-V4-Pro", "reported_setting": { "mode": "thinking", "effort": "Max", "tools": "agentic (bash + edit)", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "DeepSeek V4 model card: deepseek-v4-pro (Max mode) = 80.6.", "candidates": [ { "score": 80.6, "reference_url": "https://llm-stats.com/benchmarks/swe-bench-verified", "source_type": "third_party_aggregator", "reported_setting": { "effort": "max" }, "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=DeepSeek-V4-Pro-Max, slug=deepseek-v4-pro-max, provider=DeepSeek" }, { "score": 74.5, "reference_url": "https://research.nvidia.com/labs/nemotron/files/NVIDIA-Nemotron-3-Ultra-Technical-Report.pdf", "source_type": "tech_report", "reported_setting": { "version_metric": "Verified / % resolved", "tools": "agentic repository editing", "harness": "Harbor / OpenHands / AWS ECS", "sampling": "3-run average" }, "notes": "NVIDIA Nemotron 3 Ultra technical report, Final Table 10: DeepSeek-V4-Pro: swe_bench_verified=74.5." }, { "score": 80.6, "reference_url": "https://microsoft.ai/wp-content/uploads/2026/06/main_20260602_2.pdf", "source_type": "official_paper", "reported_setting": { "mode": "thinking", "effort": "Max", "tools": "none (per-bench override)", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default", "context": "default", "notes": "Per DeepSeek V4-Pro model card. Three reasoning modes available (Non-Think/High/Max); BP canonical = Max mode (most powerful). Other modes' values stored as candidates if needed." }, "notes": "Displayed exactly as 80.6. Official Microsoft-reported result. Comparator value is quoted from that provider's official model card/release; use the campaign canonical setting." }, { "score": 80.6, "reference_url": "https://huggingface.co/deepseek-ai/DeepSeek-V4-Pro/resolve/b5968e9190ef611bbf34a7229255be88a0e937c1/README.md", "source_type": "official_blog", "reported_setting": { "mode": "thinking (Think Max)", "effort": "Max (DeepSeek-V4-Pro-Max)", "tools": "source does not state", "sampling": "source does not state", "judge": "source does not state", "harness": "source does not state", "prompt_style": "source does not state", "temperature": "source does not state", "context": "source does not state", "input_modalities": "source does not state", "trials": "source does not state" }, "notes": "DeepSeek V4 Pro is explicitly reported at max effort. Exact DeepSeek official model-card score from the DeepSeek-V4-Pro-Max table. Only the source-backed Think Max mode and Max effort are retained; benchmark-specific harness and unstated sampling details remain unknown." } ] }, { "model_id": "deepseek-v4-pro", "benchmark_id": "swe_bench_pro", "score": 55.4, "reference_url": "https://huggingface.co/deepseek-ai/DeepSeek-V4-Pro", "reported_setting": { "mode": "thinking", "effort": "Max", "tools": "agentic (bash + edit)", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "DeepSeek V4 model card: deepseek-v4-pro (Max mode) = 55.4.", "candidates": [ { "score": 55.4, "reference_url": "https://microsoft.ai/wp-content/uploads/2026/06/main_20260602_2.pdf", "source_type": "official_paper", "reported_setting": { "mode": "thinking", "effort": "Max", "tools": "none (per-bench override)", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default", "context": "default", "notes": "Per DeepSeek V4-Pro model card. Three reasoning modes available (Non-Think/High/Max); BP canonical = Max mode (most powerful). Other modes' values stored as candidates if needed." }, "notes": "Displayed exactly as 55.4. Official Microsoft-reported result. Comparator value is quoted from that provider's official model card/release; use the campaign canonical setting." }, { "score": 55.4, "reference_url": "https://huggingface.co/deepseek-ai/DeepSeek-V4-Pro/resolve/b5968e9190ef611bbf34a7229255be88a0e937c1/README.md", "source_type": "official_blog", "reported_setting": { "mode": "thinking (Think Max)", "effort": "Max (DeepSeek-V4-Pro-Max)", "tools": "source does not state", "sampling": "source does not state", "judge": "source does not state", "harness": "source does not state", "prompt_style": "source does not state", "temperature": "source does not state", "context": "source does not state", "input_modalities": "source does not state", "trials": "source does not state" }, "notes": "DeepSeek V4 Pro is explicitly reported at max effort. Exact DeepSeek official model-card score from the DeepSeek-V4-Pro-Max table. Only the source-backed Think Max mode and Max effort are retained; benchmark-specific harness and unstated sampling details remain unknown." } ] }, { "model_id": "deepseek-v4-pro", "benchmark_id": "terminal_bench", "score": 67.9, "reference_url": "https://huggingface.co/deepseek-ai/DeepSeek-V4-Pro", "reported_setting": { "mode": "thinking", "effort": "Max", "tools": "agentic", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "DeepSeek V4 model card: deepseek-v4-pro (Max mode) = 67.9.", "candidates": [ { "score": 67.9, "reference_url": "https://llm-stats.com/benchmarks/terminal-bench-2", "source_type": "third_party_aggregator", "reported_setting": { "effort": "max" }, "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=DeepSeek-V4-Pro-Max, slug=deepseek-v4-pro-max, provider=DeepSeek" }, { "score": 67.9, "reference_url": "https://microsoft.ai/wp-content/uploads/2026/06/main_20260602_2.pdf", "source_type": "official_paper", "reported_setting": { "mode": "thinking", "effort": "Max", "tools": "none (per-bench override)", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default", "context": "default", "notes": "Per DeepSeek V4-Pro model card. Three reasoning modes available (Non-Think/High/Max); BP canonical = Max mode (most powerful). Other modes' values stored as candidates if needed." }, "notes": "Displayed exactly as 67.9. Official Microsoft-reported result. Comparator value is quoted from that provider's official model card/release; use the campaign canonical setting." }, { "score": 67.9, "reference_url": "https://huggingface.co/deepseek-ai/DeepSeek-V4-Pro/resolve/b5968e9190ef611bbf34a7229255be88a0e937c1/README.md", "source_type": "official_blog", "reported_setting": { "mode": "thinking (Think Max)", "effort": "Max (DeepSeek-V4-Pro-Max)", "tools": "source does not state", "sampling": "source does not state", "judge": "source does not state", "harness": "source does not state", "prompt_style": "source does not state", "temperature": "source does not state", "context": "source does not state", "input_modalities": "source does not state", "trials": "source does not state" }, "notes": "DeepSeek V4 Pro is explicitly reported at max effort. Exact DeepSeek official model-card score from the DeepSeek-V4-Pro-Max table. Only the source-backed Think Max mode and Max effort are retained; benchmark-specific harness and unstated sampling details remain unknown." } ] }, { "model_id": "deepseek-v4-pro", "benchmark_id": "browsecomp", "score": 83.4, "reference_url": "https://huggingface.co/deepseek-ai/DeepSeek-V4-Pro", "reported_setting": { "mode": "thinking", "effort": "Max", "tools": "agentic", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "DeepSeek V4 model card: deepseek-v4-pro (Max mode) = 83.4.", "candidates": [ { "score": 83.4, "reference_url": "https://llm-stats.com/benchmarks/browsecomp", "source_type": "third_party_aggregator", "reported_setting": { "effort": "max" }, "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=DeepSeek-V4-Pro-Max, slug=deepseek-v4-pro-max, provider=DeepSeek" }, { "score": 59.4, "reference_url": "https://research.nvidia.com/labs/nemotron/files/NVIDIA-Nemotron-3-Ultra-Technical-Report.pdf", "source_type": "tech_report", "reported_setting": { "version_metric": "BrowseComp / %", "tools": "Tavily web search and terminal workspace", "harness": "NVIDIA custom BrowseComp scaffold", "sampling": "pass@1" }, "notes": "NVIDIA Nemotron 3 Ultra technical report, Final Table 10: DeepSeek-V4-Pro: browsecomp=59.4." } ] }, { "model_id": "deepseek-v4-pro", "benchmark_id": "matharena_apex_2025", "score": 38.3, "reference_url": "https://huggingface.co/deepseek-ai/DeepSeek-V4-Pro", "audit_status": "verified", "matches_canonical": true, "source_type": "model_card", "notes": "DeepSeek-V4-Pro card Table 2 (cross-mode): V4-Pro Max Apex=38.3, confirmed in frontier table.", "audited_by": "lychee-audit-bot", "audited_at": "2026-07-26T00:00:00Z", "reported_setting": { "mode": "thinking", "effort": "Max", "harness": "official", "judge": "rule-based", "sampling": "pass@1", "temperature": "default", "tools": "none", "prompt_style": "default" }, "candidates": [ { "score": 90.2, "reference_url": "https://llm-stats.com/benchmarks/matharena-apex", "source_type": "third_party_aggregator", "reported_setting": { "effort": "max" }, "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=DeepSeek-V4-Pro-Max, slug=deepseek-v4-pro-max, provider=DeepSeek" } ] }, { "model_id": "deepseek-v4-pro", "benchmark_id": "gdpval_aa_elo", "score": 1554, "reference_url": "https://huggingface.co/deepseek-ai/DeepSeek-V4-Pro", "reported_setting": { "mode": "thinking", "effort": "Max", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "DeepSeek V4 model card: deepseek-v4-pro (Max mode) = 1554.", "candidates": [ { "score": 1554, "reference_url": "https://huggingface.co/deepseek-ai/DeepSeek-V4-Pro/resolve/b5968e9190ef611bbf34a7229255be88a0e937c1/README.md", "source_type": "official_blog", "reported_setting": { "mode": "thinking (Think Max)", "effort": "Max (DeepSeek-V4-Pro-Max)", "tools": "source does not state", "sampling": "source does not state", "judge": "source does not state", "harness": "source does not state", "prompt_style": "source does not state", "temperature": "source does not state", "context": "source does not state", "input_modalities": "source does not state", "trials": "source does not state" }, "notes": "DeepSeek V4 Pro is explicitly reported at max effort. Exact DeepSeek official model-card score from the DeepSeek-V4-Pro-Max table. Only the source-backed Think Max mode and Max effort are retained; benchmark-specific harness and unstated sampling details remain unknown." }, { "score": 1554.0, "reference_url": "https://static.stepfun.com/blog/step-3.7-flash/", "source_type": "official_blog", "reported_setting": { "quantization_config": "common release evaluation; artifact precision not established", "configuration": "base Step evaluation configuration", "mode": "source does not state", "effort": "source does not state", "tools_search": "official Artificial Analysis Stirrup harness", "shots": "source does not state", "trials_samples": "source does not state", "aggregation_pass_k": "source-reported score", "context": "source does not state", "output_cap": "source does not state", "temperature": "source does not state", "top_p": "source does not state", "judge": "pairwise Elo evaluation anchored to human experts", "harness_agent": "official Artificial Analysis Stirrup leaderboard", "dataset_version_split": "GDPval-AA historical 220-task Stirrup leaderboard quoted 2026-05-29", "multimodal_input": false }, "notes": "Official StepFun launch comparison table." } ] }, { "model_id": "deepseek-v4-flash", "benchmark_id": "mmlu_pro", "score": 86.2, "reference_url": "https://huggingface.co/deepseek-ai/DeepSeek-V4-Pro", "reported_setting": { "mode": "thinking", "effort": "Max", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "DeepSeek V4 model card: deepseek-v4-flash (Max mode) = 86.2.", "candidates": [ { "score": 86.2, "reference_url": "https://llm-stats.com/benchmarks/mmlu-pro", "source_type": "third_party_aggregator", "reported_setting": { "effort": "max" }, "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=DeepSeek-V4-Flash-Max, slug=deepseek-v4-flash-max, provider=DeepSeek" }, { "score": 86.4, "reference_url": "https://research.nvidia.com/labs/nemotron/files/NVIDIA-Nemotron-3-Ultra-Technical-Report.pdf", "source_type": "tech_report", "reported_setting": { "version_metric": "MMLU-Pro / %", "tools": "none", "harness": "NeMo Evaluator", "sampling": "pass@1" }, "notes": "NVIDIA Nemotron 3 Ultra technical report, Final Table 10: DeepSeek-V4-Flash Preview: mmlu_pro=86.4." }, { "score": 68.3, "reference_url": "https://huggingface.co/XiaomiMiMo/MiMo-V2.5-Pro/resolve/21d1ecfecd7bd70f31be25ca49d7edd21f003659/README.md", "source_type": "official_model_card_base_checkpoint", "reported_setting": { "mode": "base pretrained checkpoint", "effort": "not applicable", "tools": "none", "sampling": "5-shot; single completion per prompt", "judge": "benchmark-native rule-based evaluator", "harness": "Xiaomi MiMo base-checkpoint evaluation", "prompt_style": "5-shot few-shot continuation", "temperature": "source does not state", "context": "source does not state", "input_modalities": "text", "trials": "one completion per item", "checkpoint": "base" }, "notes": "Base checkpoint observation attached to the canonical post-trained model ID. Base result is attached to the canonical post-trained model ID. The Xiaomi table does not state this comparator Base checkpoint's context length, and no comparator Base config is locked; no Pro context value or conflict is copied onto this observation." } ] }, { "model_id": "deepseek-v4-flash", "benchmark_id": "gpqa_diamond", "score": 88.1, "reference_url": "https://huggingface.co/deepseek-ai/DeepSeek-V4-Pro", "reported_setting": { "mode": "thinking", "effort": "Max", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "DeepSeek V4 model card: deepseek-v4-flash (Max mode) = 88.1.", "candidates": [ { "score": 88.1, "reference_url": "https://llm-stats.com/benchmarks/gpqa", "source_type": "third_party_aggregator", "reported_setting": { "effort": "max" }, "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=DeepSeek-V4-Flash-Max, slug=deepseek-v4-flash-max, provider=DeepSeek" }, { "score": 88.5, "reference_url": "https://research.nvidia.com/labs/nemotron/files/NVIDIA-Nemotron-3-Ultra-Technical-Report.pdf", "source_type": "tech_report", "reported_setting": { "version_metric": "no tools / %", "tools": "none", "harness": "NeMo Evaluator", "sampling": "8-run average" }, "notes": "NVIDIA Nemotron 3 Ultra technical report, Final Table 10: DeepSeek-V4-Flash Preview: gpqa_diamond=88.5." } ] }, { "model_id": "deepseek-v4-flash", "benchmark_id": "hle", "score": 34.8, "reference_url": "https://huggingface.co/deepseek-ai/DeepSeek-V4-Pro", "reported_setting": { "mode": "thinking", "effort": "Max", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "DeepSeek V4 model card: deepseek-v4-flash (Max mode) = 34.8.", "candidates": [ { "score": 45.1, "reference_url": "https://llm-stats.com/benchmarks/humanity's-last-exam", "source_type": "third_party_aggregator", "reported_setting": { "effort": "max" }, "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=DeepSeek-V4-Flash-Max, slug=deepseek-v4-flash-max, provider=DeepSeek" } ] }, { "model_id": "deepseek-v4-flash", "benchmark_id": "livecodebench", "score": 91.6, "reference_url": "https://huggingface.co/deepseek-ai/DeepSeek-V4-Pro", "reported_setting": { "mode": "thinking", "effort": "Max", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "DeepSeek V4 model card: deepseek-v4-flash (Max mode) = 91.6.", "candidates": [ { "score": 91.6, "reference_url": "https://llm-stats.com/benchmarks/livecodebench", "source_type": "third_party_aggregator", "reported_setting": { "effort": "max" }, "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=DeepSeek-V4-Flash-Max, slug=deepseek-v4-flash-max, provider=DeepSeek" } ] }, { "model_id": "deepseek-v4-flash", "benchmark_id": "codeforces_rating", "score": 3052, "reference_url": "https://huggingface.co/deepseek-ai/DeepSeek-V4-Pro", "reported_setting": { "mode": "thinking", "effort": "Max", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "DeepSeek V4 model card: deepseek-v4-flash (Max mode) = 3052.", "candidates": [ { "score": 100.0, "reference_url": "https://llm-stats.com/benchmarks/codeforces", "source_type": "third_party_aggregator", "reported_setting": { "effort": "max" }, "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=DeepSeek-V4-Flash-Max, slug=deepseek-v4-flash-max, provider=DeepSeek" } ] }, { "model_id": "deepseek-v4-flash", "benchmark_id": "swe_bench_verified", "score": 79.0, "reference_url": "https://huggingface.co/deepseek-ai/DeepSeek-V4-Pro", "reported_setting": { "mode": "thinking", "effort": "Max", "tools": "agentic (bash + edit)", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "DeepSeek V4 model card: deepseek-v4-flash (Max mode) = 79.0.", "candidates": [ { "score": 79.0, "reference_url": "https://llm-stats.com/benchmarks/swe-bench-verified", "source_type": "third_party_aggregator", "reported_setting": { "effort": "max" }, "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=DeepSeek-V4-Flash-Max, slug=deepseek-v4-flash-max, provider=DeepSeek" }, { "score": 73.5, "reference_url": "https://research.nvidia.com/labs/nemotron/files/NVIDIA-Nemotron-3-Ultra-Technical-Report.pdf", "source_type": "tech_report", "reported_setting": { "version_metric": "Verified / % resolved", "tools": "agentic repository editing", "harness": "Harbor / OpenHands / AWS ECS", "sampling": "3-run average" }, "notes": "NVIDIA Nemotron 3 Ultra technical report, Final Table 10: DeepSeek-V4-Flash Preview: swe_bench_verified=73.5." }, { "score": 79.0, "reference_url": "https://static.stepfun.com/blog/step-3.7-flash/", "source_type": "official_blog", "reported_setting": { "quantization_config": "common release evaluation; artifact precision not established", "configuration": "base Step evaluation configuration", "mode": "source does not state", "effort": "source does not state", "tools_search": "agentic coding tools", "shots": "source does not state", "trials_samples": "source does not state", "aggregation_pass_k": "% resolved", "context": "source does not state", "output_cap": "source does not state", "temperature": "source does not state", "top_p": "source does not state", "judge": "benchmark-specified; exact StepFun judge not stated", "harness_agent": "source does not state coding scaffold", "dataset_version_split": "source label only", "multimodal_input": false }, "notes": "Official StepFun launch comparison table." } ] }, { "model_id": "deepseek-v4-flash", "benchmark_id": "swe_bench_pro", "score": 52.6, "reference_url": "https://huggingface.co/deepseek-ai/DeepSeek-V4-Pro", "reported_setting": { "mode": "thinking", "effort": "Max", "tools": "agentic (bash + edit)", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "DeepSeek V4 model card: deepseek-v4-flash (Max mode) = 52.6.", "candidates": [ { "score": 52.6, "reference_url": "https://llm-stats.com/benchmarks/swe-bench-pro", "source_type": "third_party_aggregator", "reported_setting": { "effort": "max" }, "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=DeepSeek-V4-Flash-Max, slug=deepseek-v4-flash-max, provider=DeepSeek" }, { "score": 52.6, "reference_url": "https://huggingface.co/deepseek-ai/DeepSeek-V4-Pro/resolve/b5968e9190ef611bbf34a7229255be88a0e937c1/README.md", "source_type": "official_model_card", "reported_setting": { "mode": "thinking (Think Max)", "effort": "Max (DeepSeek-V4-Flash-Max)", "tools": "source does not state", "sampling": "source does not state", "judge": "source does not state", "harness": "source does not state", "prompt_style": "source does not state", "temperature": "source does not state", "context": "source does not state", "input_modalities": "source does not state", "trials": "source does not state" }, "notes": "Exact DeepSeek official model-card score from the DeepSeek-V4-Pro-Max table. Only the source-backed Think Max mode and Max effort are retained; benchmark-specific harness and unstated sampling details remain unknown." }, { "score": 55.6, "reference_url": "https://static.stepfun.com/blog/step-3.7-flash/", "source_type": "official_blog", "reported_setting": { "quantization_config": "common release evaluation; artifact precision not established", "configuration": "base Step evaluation configuration", "mode": "source does not state", "effort": "source does not state", "tools_search": "agentic coding tools", "shots": "source does not state", "trials_samples": "source does not state", "aggregation_pass_k": "% resolved", "context": "source does not state", "output_cap": "source does not state", "temperature": "source does not state", "top_p": "source does not state", "judge": "benchmark-specified; exact StepFun judge not stated", "harness_agent": "source does not state coding scaffold", "dataset_version_split": "source label only", "multimodal_input": false }, "notes": "Official StepFun launch comparison table." } ] }, { "model_id": "deepseek-v4-flash", "benchmark_id": "terminal_bench", "score": 56.9, "reference_url": "https://huggingface.co/deepseek-ai/DeepSeek-V4-Pro", "reported_setting": { "mode": "thinking", "effort": "Max", "tools": "agentic", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "DeepSeek V4 model card: deepseek-v4-flash (Max mode) = 56.9.", "candidates": [ { "score": 56.9, "reference_url": "https://llm-stats.com/benchmarks/terminal-bench-2", "source_type": "third_party_aggregator", "reported_setting": { "effort": "max" }, "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=DeepSeek-V4-Flash-Max, slug=deepseek-v4-flash-max, provider=DeepSeek" }, { "score": 56.9, "reference_url": "https://huggingface.co/deepseek-ai/DeepSeek-V4-Pro/resolve/b5968e9190ef611bbf34a7229255be88a0e937c1/README.md", "source_type": "official_model_card", "reported_setting": { "mode": "thinking (Think Max)", "effort": "Max (DeepSeek-V4-Flash-Max)", "tools": "source does not state", "sampling": "source does not state", "judge": "source does not state", "harness": "source does not state", "prompt_style": "source does not state", "temperature": "source does not state", "context": "source does not state", "input_modalities": "source does not state", "trials": "source does not state" }, "notes": "Exact DeepSeek official model-card score from the DeepSeek-V4-Pro-Max table. Only the source-backed Think Max mode and Max effort are retained; benchmark-specific harness and unstated sampling details remain unknown." } ] }, { "model_id": "deepseek-v4-flash", "benchmark_id": "browsecomp", "score": 73.2, "reference_url": "https://huggingface.co/deepseek-ai/DeepSeek-V4-Pro", "reported_setting": { "mode": "thinking", "effort": "Max", "tools": "agentic", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "DeepSeek V4 model card: deepseek-v4-flash (Max mode) = 73.2.", "candidates": [ { "score": 73.2, "reference_url": "https://llm-stats.com/benchmarks/browsecomp", "source_type": "third_party_aggregator", "reported_setting": { "effort": "max" }, "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=DeepSeek-V4-Flash-Max, slug=deepseek-v4-flash-max, provider=DeepSeek" }, { "score": 46.9, "reference_url": "https://research.nvidia.com/labs/nemotron/files/NVIDIA-Nemotron-3-Ultra-Technical-Report.pdf", "source_type": "tech_report", "reported_setting": { "version_metric": "BrowseComp / %", "tools": "Tavily web search and terminal workspace", "harness": "NVIDIA custom BrowseComp scaffold", "sampling": "pass@1" }, "notes": "NVIDIA Nemotron 3 Ultra technical report, Final Table 10: DeepSeek-V4-Flash Preview: browsecomp=46.9." } ] }, { "model_id": "deepseek-v4-flash", "benchmark_id": "matharena_apex_2025", "score": 33.0, "reference_url": "https://huggingface.co/deepseek-ai/DeepSeek-V4-Flash", "audit_status": "verified", "matches_canonical": true, "source_type": "model_card", "notes": "DeepSeek-V4-Flash card Table 2 (cross-mode comparison): V4-Flash Max Apex=33.0.", "audited_by": "lychee-audit-bot", "audited_at": "2026-07-26T00:00:00Z", "reported_setting": { "mode": "thinking", "effort": "Max", "harness": "official", "judge": "rule-based", "sampling": "pass@1", "temperature": "default", "tools": "none", "prompt_style": "default" }, "candidates": [ { "score": 85.7, "reference_url": "https://llm-stats.com/benchmarks/matharena-apex", "source_type": "third_party_aggregator", "reported_setting": { "effort": "max" }, "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=DeepSeek-V4-Flash-Max, slug=deepseek-v4-flash-max, provider=DeepSeek" } ] }, { "model_id": "deepseek-v4-flash", "benchmark_id": "gdpval_aa_elo", "score": 1395, "reference_url": "https://huggingface.co/deepseek-ai/DeepSeek-V4-Pro", "reported_setting": { "mode": "thinking", "effort": "Max", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "DeepSeek V4 model card: deepseek-v4-flash (Max mode) = 1395.", "candidates": [ { "score": 1414.0, "reference_url": "https://static.stepfun.com/blog/step-3.7-flash/", "source_type": "official_blog", "reported_setting": { "quantization_config": "common release evaluation; artifact precision not established", "configuration": "base Step evaluation configuration", "mode": "source does not state", "effort": "source does not state", "tools_search": "official Artificial Analysis Stirrup harness", "shots": "source does not state", "trials_samples": "source does not state", "aggregation_pass_k": "source-reported score", "context": "source does not state", "output_cap": "source does not state", "temperature": "source does not state", "top_p": "source does not state", "judge": "pairwise Elo evaluation anchored to human experts", "harness_agent": "official Artificial Analysis Stirrup leaderboard", "dataset_version_split": "GDPval-AA historical 220-task Stirrup leaderboard quoted 2026-05-29", "multimodal_input": false }, "notes": "Official StepFun launch comparison table." } ] }, { "model_id": "qwen3-235b", "benchmark_id": "aime_2025", "score": 81.5, "reference_url": "https://arxiv.org/abs/2505.09388", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Qwen3 tech report (arxiv:2505.09388) Table 11 Thinking column: AIME 2025 81.5", "candidates": [ { "score": 92.3, "reference_url": "https://llm-stats.com/benchmarks/aime-2025", "source_type": "third_party_aggregator", "reported_setting": { "mode": "thinking", "snapshot": "2507" }, "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=Qwen3-235B-A22B-Thinking-2507, slug=qwen3-235b-a22b-thinking-2507, provider=Alibaba Cloud / Qwen Team" }, { "score": 70.3, "reference_url": "https://llm-stats.com/benchmarks/aime-2025", "source_type": "third_party_aggregator", "reported_setting": { "mode": "instruct", "snapshot": "2507" }, "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=Qwen3-235B-A22B-Instruct-2507, slug=qwen3-235b-a22b-instruct-2507, provider=Alibaba Cloud / Qwen Team" } ] }, { "model_id": "qwen3-235b", "benchmark_id": "gpqa_diamond", "score": 71.1, "reference_url": "https://arxiv.org/abs/2505.09388", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Qwen3 tech report (arxiv:2505.09388) Table 11 Thinking column: GPQA Diamond 71.1", "candidates": [ { "score": 81.1, "reference_url": "https://llm-stats.com/benchmarks/gpqa", "source_type": "third_party_aggregator", "reported_setting": { "mode": "thinking", "snapshot": "2507" }, "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=Qwen3-235B-A22B-Thinking-2507, slug=qwen3-235b-a22b-thinking-2507, provider=Alibaba Cloud / Qwen Team" }, { "score": 77.5, "reference_url": "https://llm-stats.com/benchmarks/gpqa", "source_type": "third_party_aggregator", "reported_setting": { "mode": "instruct", "snapshot": "2507" }, "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=Qwen3-235B-A22B-Instruct-2507, slug=qwen3-235b-a22b-instruct-2507, provider=Alibaba Cloud / Qwen Team" } ] }, { "model_id": "qwen3-235b", "benchmark_id": "livecodebench", "score": 70.7, "reference_url": "https://arxiv.org/abs/2505.09388", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Qwen3 tech report (arxiv:2505.09388) Table 11 Thinking column: LiveCodeBench v5 (2024.10-2025.02) 70.7" }, { "model_id": "qwen3-235b", "benchmark_id": "codeforces_rating", "score": 2056, "reference_url": "https://arxiv.org/abs/2505.09388", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Qwen3 tech report (arxiv:2505.09388) Table 11 Thinking column: CodeForces rating 2056 (98.2 percentile)" }, { "model_id": "qwen3-235b", "benchmark_id": "arena_hard", "score": 95.6, "reference_url": "https://arxiv.org/abs/2505.09388", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Qwen3 tech report (arxiv:2505.09388) Table 11 Thinking column: Arena-Hard 95.6" }, { "model_id": "qwen3-235b", "benchmark_id": "aime_2024", "score": 85.7, "reference_url": "https://arxiv.org/abs/2505.09388", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Qwen3 tech report (arxiv:2505.09388) Table 11 Thinking column: AIME 2024 85.7" }, { "model_id": "qwen3-235b", "benchmark_id": "swe_bench_verified", "score": 69.6, "reference_url": "https://qwenlm.github.io/blog/qwen3/", "audit_status": "verified", "reported_setting": "Qwen3 blog performance table, thinking mode", "source_type": "official_blog", "matches_canonical": true, "notes": "Confirmed via multiple independent web sources (69.6%)" }, { "model_id": "qwen3-235b", "benchmark_id": "hmmt_feb_2025", "score": 62.5, "reference_url": "https://huggingface.co/deepseek-ai/DeepSeek-R1-0528", "audit_status": "verified", "matches_canonical": true, "source_type": "model_card", "notes": "DeepSeek-R1-0528 HF card comparison table shows Qwen3-235B-A22B HMMT Feb 25=62.5.", "audited_by": "lychee-audit-bot", "audited_at": "2026-07-26T00:00:00Z" }, { "model_id": "qwen3-235b", "benchmark_id": "hle", "score": 15.43, "reference_url": "https://scale.com/leaderboard/humanitys_last_exam", "audit_status": "dropped", "notes": " [Dropped: Model not found in current Scale.com HLE leaderboard (50 entries, 2026-04-29 scrape). Ghost cell per R5g-a-ghost-cell.]", "candidates": [ { "score": 18.2, "reference_url": "https://llm-stats.com/benchmarks/humanity's-last-exam", "source_type": "third_party_aggregator", "reported_setting": { "mode": "thinking", "snapshot": "2507" }, "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=Qwen3-235B-A22B-Thinking-2507, slug=qwen3-235b-a22b-thinking-2507, provider=Alibaba Cloud / Qwen Team" } ] }, { "model_id": "qwen3-235b", "benchmark_id": "mmlu_pro", "score": 79.8, "reference_url": "https://arxiv.org/abs/2505.09388", "audit_status": "dropped", "notes": " | DROPPED: BP value 79.8 from arxiv:2505.09388 is from Non-thinking variant (Table 12), but qwen3-235b canonical = Thinking. MMLU-Pro not reported in Table 11. Cell value is wrong-variant.", "candidates": [ { "score": 84.4, "reference_url": "https://llm-stats.com/benchmarks/mmlu-pro", "source_type": "third_party_aggregator", "reported_setting": { "mode": "thinking", "snapshot": "2507" }, "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=Qwen3-235B-A22B-Thinking-2507, slug=qwen3-235b-a22b-thinking-2507, provider=Alibaba Cloud / Qwen Team" }, { "score": 83.0, "reference_url": "https://llm-stats.com/benchmarks/mmlu-pro", "source_type": "third_party_aggregator", "reported_setting": { "mode": "instruct", "snapshot": "2507" }, "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=Qwen3-235B-A22B-Instruct-2507, slug=qwen3-235b-a22b-instruct-2507, provider=Alibaba Cloud / Qwen Team" } ] }, { "model_id": "qwen3-235b", "benchmark_id": "math_500", "score": 98.0, "reference_url": "https://arxiv.org/abs/2505.09388", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Qwen3 tech report (arxiv:2505.09388) Table 11 Thinking: MATH-500 98.0. [R5d: prior unverified value 98.2 from https://arxiv.org/abs/2505.09388 deleted.]" }, { "model_id": "qwen3-235b", "benchmark_id": "chatbot_arena_elo", "score": 1410, "reference_url": "https://lmarena.ai/", "audit_status": "needs_review", "audit_note": "lmarena.ai is a dynamic leaderboard; ELO values change over time. Cannot verify static value 1410 from snapshot-dependent source via automated audit.", "audited_by": "lychee-audit-bot", "audited_at": "2026-07-26T00:00:00Z" }, { "model_id": "qwen3-235b", "benchmark_id": "gsm8k", "score": 94.39, "reference_url": "https://arxiv.org/abs/2505.09388", "audit_status": "dropped", "notes": " | DROPPED: BP value 94.39 from arxiv:2505.09388 Table 3 (Base column). GSM8K is not reported in Table 11 (Thinking). Cell value is wrong-variant (Base, not Thinking)." }, { "model_id": "qwen3-235b", "benchmark_id": "humaneval", "score": 90.0, "reference_url": "https://qwenlm.github.io/blog/qwen3/", "audit_status": "verified", "reported_setting": "Qwen3 blog performance table, thinking mode", "source_type": "official_blog", "matches_canonical": true, "notes": "Confirmed via official Qwen3 blog and web search (90.0%)" }, { "model_id": "qwen3-235b", "benchmark_id": "ifeval", "score": 83.4, "reference_url": "https://arxiv.org/abs/2505.09388", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Qwen3 tech report (arxiv:2505.09388) Table 11 Thinking: IFEval strict prompt 83.4. [R5d: prior unverified value 87.8 from https://huggingface.co/Qwen/Qwen3-235B-A22B-Thinking-2507 deleted.]", "candidates": [ { "score": 88.7, "reference_url": "https://llm-stats.com/benchmarks/ifeval", "source_type": "third_party_aggregator", "reported_setting": { "mode": "instruct", "snapshot": "2507" }, "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=Qwen3-235B-A22B-Instruct-2507, slug=qwen3-235b-a22b-instruct-2507, provider=Alibaba Cloud / Qwen Team" }, { "score": 87.8, "reference_url": "https://llm-stats.com/benchmarks/ifeval", "source_type": "third_party_aggregator", "reported_setting": { "mode": "thinking", "snapshot": "2507" }, "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=Qwen3-235B-A22B-Thinking-2507, slug=qwen3-235b-a22b-thinking-2507, provider=Alibaba Cloud / Qwen Team" } ] }, { "model_id": "qwen3-235b", "benchmark_id": "mmlu", "score": 87.81, "reference_url": "https://arxiv.org/abs/2505.09388", "audit_status": "dropped", "notes": " | DROPPED: BP value 87.81 from arxiv:2505.09388 Table 3 (Base column, MMLU 5-shot), but qwen3-235b is_reasoning=true → canonical = Thinking variant. MMLU is not reported in Table 11 (Thinking). Cell value is wrong-variant." }, { "model_id": "qwen3-235b", "benchmark_id": "simpleqa", "score": 13.2, "reference_url": "https://arxiv.org/abs/2507.20534", "audit_status": "verified", "source_type": "tech_report", "matches_canonical": false, "reported_setting": { "mode": "non-thinking", "effort": "default", "tools": "none (benchmark-specified)", "sampling": "pass@1", "judge": "rule-based", "harness": "Kimi team evaluation (official API)", "prompt_style": "default", "temperature": "unified (Kimi team setting)", "context": "8192 max output tokens" }, "notes": "Via third-party self-test by Moonshot AI (Kimi K2 tech report, Table 3, p.16). SimpleQA Correct, non-thinking (vendor-recommended no-thinking regime), 8192 max output tokens.", "candidates": [ { "score": 54.3, "reference_url": "https://llm-stats.com/benchmarks/simpleqa", "source_type": "third_party_aggregator", "reported_setting": { "mode": "instruct", "snapshot": "2507" }, "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=Qwen3-235B-A22B-Instruct-2507, slug=qwen3-235b-a22b-instruct-2507, provider=Alibaba Cloud / Qwen Team" } ] }, { "model_id": "qwen3-235b", "benchmark_id": "aa_lcr", "score": 67, "reference_url": "https://artificialanalysis.ai/evaluations/artificial-analysis-long-context-reasoning", "audit_status": "verified", "source_type": "third_party", "notes": "ArtificialAnalysis aggregator; used as fallback when no primary source available (R5d-aggregator-fallback). AA LCR eval page: qwen3-235b-a22b-instruct-2507-reasoning lcr=0.670 → 67.0%. AA slug: qwen3-235b-a22b-instruct-2507-reasoning.", "reported_setting": { "mode": "thinking", "effort": "default (reasoning variant)", "sampling": "pass@1", "harness": "AA standard evaluation", "notes": "Per AA: qwen3-235b-a22b-instruct-2507-reasoning lcr=67%" } }, { "model_id": "qwen3-235b", "benchmark_id": "arc_agi_1", "score": 11, "reference_url": "https://arcprize.org/leaderboard", "audit_status": "verified", "source_type": "leaderboard", "matches_canonical": false, "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1 (single-pass)", "judge": "rule-based", "harness": "official (arcprize.org)", "system_type": "Base LLM" }, "notes": "arcprize.org leaderboard audit: Qwen3-235b-a22b Instruct (25/07) on leaderboard as Base LLM (non-thinking); canonical is thinking", "candidates": [ { "score": 41.8, "reference_url": "https://llm-stats.com/benchmarks/arc-agi", "source_type": "third_party_aggregator", "reported_setting": { "mode": "instruct", "snapshot": "2507" }, "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=Qwen3-235B-A22B-Instruct-2507, slug=qwen3-235b-a22b-instruct-2507, provider=Alibaba Cloud / Qwen Team" } ] }, { "model_id": "qwen3-235b", "benchmark_id": "arc_agi_2", "score": 1.3, "reference_url": "https://arcprize.org/leaderboard", "audit_status": "verified", "source_type": "leaderboard", "matches_canonical": false, "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1 (single-pass)", "judge": "rule-based", "harness": "official (arcprize.org)", "system_type": "Base LLM" }, "notes": "arcprize.org leaderboard audit: Qwen3-235b-a22b Instruct (25/07) on leaderboard as Base LLM (non-thinking); canonical is thinking" }, { "model_id": "qwen3-235b", "benchmark_id": "matharena_apex_2025", "score": 5.21, "reference_url": "https://matharena.ai/?comp=apex--apex_2025", "audit_status": "verified", "matches_canonical": true, "reported_setting": "thinking:default", "source_type": "leaderboard" }, { "model_id": "qwen3-235b", "benchmark_id": "terminal_bench_1", "score": 6.6, "reference_url": "https://www.tbench.ai/leaderboard/terminal-bench/1.0", "audit_status": "verified", "source_type": "leaderboard", "rule_ids": [ "R5d" ], "notes": "Confirmed 6.6% on tbench.ai Terminal-Bench 1.0 leaderboard (Terminus 1 scaffold, 2025-05-15)." }, { "model_id": "qwen3-32b", "benchmark_id": "gpqa_diamond", "score": 68.4, "reference_url": "https://arxiv.org/abs/2505.09388", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Qwen3 tech report Table 13/15/17/19 Thinking: GPQA-Diamond 68.4" }, { "model_id": "qwen3-32b", "benchmark_id": "aime_2025", "score": 72.9, "reference_url": "https://arxiv.org/abs/2505.09388", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Qwen3 tech report Thinking: AIME 2025 72.9. [R5d: prior unverified value 76.67 from https://arxiv.org/abs/2505.09388 deleted.]" }, { "model_id": "qwen3-32b", "benchmark_id": "livecodebench", "score": 65.7, "reference_url": "https://arxiv.org/abs/2505.09388", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Qwen3 tech report Thinking: LiveCodeBench v5 65.7. [R5d: prior unverified value 63.0 from https://qwenlm.github.io/blog/qwen3/ deleted.]" }, { "model_id": "qwen3-32b", "benchmark_id": "arena_hard", "score": 93.8, "reference_url": "https://arxiv.org/abs/2505.09388", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Qwen3 tech report Thinking: Arena-Hard 93.8 (Qwen self-report; differs from arena-hard-auto github). [R5d: prior unverified value 53.3 from https://github.com/lmarena/arena-hard-auto deleted.]" }, { "model_id": "qwen3-32b", "benchmark_id": "aime_2024", "score": 81.4, "reference_url": "https://arxiv.org/abs/2505.09388", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Qwen3 tech report Table 13/15/17/19 Thinking: AIME 2024 81.4" }, { "model_id": "qwen3-32b", "benchmark_id": "codeforces_rating", "score": 1977, "reference_url": "https://arxiv.org/abs/2505.09388", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Qwen3 tech report Thinking: CodeForces 1977 (97.7%). [R5d: prior unverified value 2020 from https://qwenlm.github.io/blog/qwen3/ deleted.]" }, { "model_id": "qwen3-32b", "benchmark_id": "gsm8k", "score": 93.4, "reference_url": "https://arxiv.org/abs/2505.09388", "audit_status": "dropped", "notes": " | DROPPED: BP value sourced from Base model column (Table 3-8) but model is_reasoning=true → canonical = Thinking variant. Wrong-variant cell." }, { "model_id": "qwen3-32b", "benchmark_id": "humaneval", "score": 85.0, "reference_url": "https://qwenlm.github.io/blog/qwen3/", "audit_status": "verified", "reported_setting": "Qwen3 blog performance table, thinking mode", "source_type": "official_blog", "matches_canonical": true, "notes": "Confirmed via official Qwen3 blog (85.0%)" }, { "model_id": "qwen3-32b", "benchmark_id": "ifeval", "score": 85.0, "reference_url": "https://arxiv.org/abs/2505.09388", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Qwen3 tech report Thinking: IFEval strict prompt 85.0. [R5d: prior unverified value 83.0 from https://arxiv.org/abs/2505.09388 deleted.]" }, { "model_id": "qwen3-32b", "benchmark_id": "math_500", "score": 97.2, "reference_url": "https://arxiv.org/abs/2505.09388", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Qwen3 tech report Thinking: MATH-500 97.2. [R5d: prior unverified value 96.0 from https://qwenlm.github.io/blog/qwen3/ deleted.]" }, { "model_id": "qwen3-32b", "benchmark_id": "mmlu", "score": 83.61, "reference_url": "https://arxiv.org/abs/2505.09388", "audit_status": "dropped", "notes": " | DROPPED: BP value sourced from Base model column (Table 3-8) but model is_reasoning=true → canonical = Thinking variant. Wrong-variant cell." }, { "model_id": "qwen3-32b", "benchmark_id": "mmlu_pro", "score": 70.0, "reference_url": "https://arxiv.org/abs/2505.09388", "audit_status": "dropped", "notes": " | DROPPED: BP value from Non-thinking column (Table 12/14/16/18/20). Canonical = Thinking. Wrong-variant." }, { "model_id": "qwen3-32b", "benchmark_id": "ifbench", "score": 37.3, "reference_url": "https://github.com/allenai/IFBench", "audit_status": "verified", "reported_setting": "IFBench score, 0-shot", "source_type": "leaderboard", "matches_canonical": true }, { "model_id": "qwen3-32b", "benchmark_id": "terminal_bench_1", "score": 15.5, "reference_url": "https://www.tbench.ai/leaderboard/terminal-bench/1.0", "audit_status": "verified", "source_type": "leaderboard", "rule_ids": [ "R5d" ], "notes": "Confirmed 15.5% on tbench.ai Terminal-Bench 1.0 leaderboard (TerminalAgent scaffold, 2025-07-31)." }, { "model_id": "qwen3-4b", "benchmark_id": "aime_2025", "score": 65.6, "reference_url": "https://arxiv.org/abs/2505.09388", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Qwen3 tech report Table 13/15/17/19 Thinking: AIME 2025 65.6" }, { "model_id": "qwen3-4b", "benchmark_id": "math_500", "score": 97.0, "reference_url": "https://arxiv.org/abs/2505.09388", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Qwen3 tech report Table 13/15/17/19 Thinking: MATH-500 97.0" }, { "model_id": "qwen3-4b", "benchmark_id": "codeforces_rating", "score": 1671, "reference_url": "https://arxiv.org/abs/2505.09388", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Qwen3 tech report Table 13/15/17/19 Thinking: CodeForces 1671 (92.8%)" }, { "model_id": "qwen3-4b", "benchmark_id": "arena_hard", "score": 76.6, "reference_url": "https://arxiv.org/abs/2505.09388", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Qwen3 tech report Thinking: Arena-Hard 76.6. [R5d: prior unverified value 13.2 from https://github.com/lmarena/arena-hard-auto deleted.]" }, { "model_id": "qwen3-4b", "benchmark_id": "gpqa_diamond", "score": 55.9, "reference_url": "https://arxiv.org/abs/2505.09388", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Qwen3 tech report Thinking: GPQA-Diamond 55.9. [R5d: prior unverified value 51.2 from https://arxiv.org/abs/2505.09388 deleted.]" }, { "model_id": "qwen3-4b", "benchmark_id": "aime_2024", "score": 73.8, "reference_url": "https://arxiv.org/abs/2505.09388", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Qwen3 tech report Table 13/15/17/19 Thinking: AIME 2024 73.8" }, { "model_id": "qwen3-4b", "benchmark_id": "gsm8k", "score": 87.79, "reference_url": "https://arxiv.org/abs/2505.09388", "audit_status": "dropped", "notes": " | DROPPED: BP value sourced from Base model column (Table 3-8) but model is_reasoning=true → canonical = Thinking variant. Wrong-variant cell." }, { "model_id": "qwen3-4b", "benchmark_id": "hmmt_feb_2025", "score": 55.5, "reference_url": "https://huggingface.co/Qwen/Qwen3-4B-Thinking-2507", "audit_status": "dropped", "notes": " | DROPPED: BP value from Qwen3-X-Thinking-2507 (post-July fine-tune), but BP model release_date=2025-05-15. Wrong-variant (later checkpoint)." }, { "model_id": "qwen3-4b", "benchmark_id": "humaneval", "score": 75.0, "reference_url": "https://qwenlm.github.io/blog/qwen3/", "audit_status": "needs_review", "notes": "Blog tables are images; web search gives approximate ~61-63% vs claimed 75.0; cannot confirm exact value. R5g(c)" }, { "model_id": "qwen3-4b", "benchmark_id": "ifeval", "score": 81.9, "reference_url": "https://arxiv.org/abs/2505.09388", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Qwen3 tech report Thinking: IFEval 81.9 (May 2025 release; replaces 87.4 from Thinking-2507). [R5d: prior unverified value 87.4 from https://huggingface.co/Qwen/Qwen3-4B-Thinking-2507 deleted.]" }, { "model_id": "qwen3-4b", "benchmark_id": "livecodebench", "score": 54.2, "reference_url": "https://arxiv.org/abs/2505.09388", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Qwen3 tech report Thinking: LiveCodeBench v5 54.2 (replaces 55.2 from Thinking-2507). [R5d: prior unverified value 55.2 from https://huggingface.co/Qwen/Qwen3-4B-Thinking-2507 deleted.]" }, { "model_id": "qwen3-4b", "benchmark_id": "mmlu", "score": 72.99, "reference_url": "https://arxiv.org/abs/2505.09388", "audit_status": "dropped", "notes": " | DROPPED: BP value sourced from Base model column (Table 3-8) but model is_reasoning=true → canonical = Thinking variant. Wrong-variant cell." }, { "model_id": "qwen3-4b", "benchmark_id": "mmlu_pro", "score": 74.0, "reference_url": "https://huggingface.co/Qwen/Qwen3-4B-Thinking-2507", "audit_status": "dropped", "notes": " | DROPPED: BP value from Qwen3-X-Thinking-2507 (post-July fine-tune), but BP model release_date=2025-05-15. Wrong-variant (later checkpoint)." }, { "model_id": "qwen3-0.6b", "benchmark_id": "math_500", "score": 77.6, "reference_url": "https://arxiv.org/abs/2505.09388", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Qwen3 tech report Table 13/15/17/19 Thinking: MATH-500 77.6" }, { "model_id": "qwen3-0.6b", "benchmark_id": "mmlu", "score": 52.81, "reference_url": "https://arxiv.org/abs/2505.09388", "audit_status": "dropped", "notes": " | DROPPED: BP value sourced from Base model column (Table 3-8) but model is_reasoning=true → canonical = Thinking variant. Wrong-variant cell." }, { "model_id": "qwen3-0.6b", "benchmark_id": "mmlu_pro", "score": 24.74, "reference_url": "https://arxiv.org/abs/2505.09388", "audit_status": "dropped", "notes": " | DROPPED: BP value sourced from Base model column (Table 3-8) but model is_reasoning=true → canonical = Thinking variant. Wrong-variant cell." }, { "model_id": "qwen3-0.6b", "benchmark_id": "gpqa_diamond", "score": 27.9, "reference_url": "https://arxiv.org/abs/2505.09388", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Qwen3 tech report Table 13/15/17/19 Thinking: GPQA-Diamond 27.9" }, { "model_id": "qwen3-0.6b", "benchmark_id": "aime_2024", "score": 10.7, "reference_url": "https://arxiv.org/abs/2505.09388", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Qwen3 tech report Table 13/15/17/19 Thinking: AIME 2024 10.7" }, { "model_id": "qwen3-0.6b", "benchmark_id": "aime_2025", "score": 15.1, "reference_url": "https://arxiv.org/abs/2505.09388", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Qwen3 tech report Table 13/15/17/19 Thinking: AIME 2025 15.1" }, { "model_id": "qwen3-0.6b", "benchmark_id": "ifeval", "score": 59.2, "reference_url": "https://arxiv.org/abs/2505.09388", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Qwen3 tech report Table 13/15/17/19 Thinking: IFEval 59.2" }, { "model_id": "qwen3-0.6b", "benchmark_id": "arena_hard", "score": 8.5, "reference_url": "https://arxiv.org/abs/2505.09388", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Qwen3 tech report Table 13/15/17/19 Thinking: Arena-Hard 8.5" }, { "model_id": "qwen3-0.6b", "benchmark_id": "livecodebench", "score": 12.3, "reference_url": "https://arxiv.org/abs/2505.09388", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Qwen3 tech report Table 13/15/17/19 Thinking: LiveCodeBench v5 12.3" }, { "model_id": "qwen3-0.6b", "benchmark_id": "codeforces_rating", "score": 800, "reference_url": "https://qwenlm.github.io/blog/qwen3/", "audit_status": "needs_review", "notes": "Web search: 'no official Codeforces ELO given for Qwen3-0.6B'; value 800 not confirmed in any accessible text source. R5g(c)" }, { "model_id": "qwen3-0.6b", "benchmark_id": "gsm8k", "score": 59.59, "reference_url": "https://arxiv.org/abs/2505.09388", "audit_status": "dropped", "notes": " | DROPPED: BP value sourced from Base model column (Table 3-8) but model is_reasoning=true → canonical = Thinking variant. Wrong-variant cell." }, { "model_id": "qwen3-0.6b", "benchmark_id": "humaneval", "score": 45.0, "reference_url": "https://arxiv.org/abs/2505.09388", "audit_status": "dropped", "rule_id": "R5g-ghost", "audit_note": "Qwen3 paper uses EvalPlus (avg of HumanEval+MBPP variants), no standalone HumanEval table.", "audited_by": "lychee-audit-bot", "audited_at": "2026-04-29T13:17:14Z" }, { "model_id": "qwen3-1.7b", "benchmark_id": "mmlu", "score": 61.0, "reference_url": "https://arxiv.org/abs/2505.09388", "audit_status": "dropped", "notes": " | DROPPED: BP value sourced from Base model column (Table 3-8) but model is_reasoning=true → canonical = Thinking variant. Wrong-variant cell." }, { "model_id": "qwen3-1.7b", "benchmark_id": "mmlu_pro", "score": 56.68, "reference_url": "https://www.liquid.ai/blog/lfm2-5-1-2b-thinking-on-device-reasoning-under-1gb", "audit_status": "verified", "notes": "Liquid AI LFM2.5-1.2B-Thinking blog: qwen3-1.7b=56.68. (third-party self-test by Liquid AI)", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1 (mean of 5 runs)", "judge": "rule-based", "harness": "official (LFM eval, third-party)", "prompt_style": "default", "temperature": "default" }, "matches_canonical": false, "source_type": "official_blog" }, { "model_id": "qwen3-1.7b", "benchmark_id": "math_500", "score": 93.4, "reference_url": "https://arxiv.org/abs/2505.09388", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Qwen3 tech report Table 13/15/17/19 Thinking: MATH-500 93.4", "candidates": [ { "score": 81.92, "reference_url": "https://www.liquid.ai/blog/lfm2-5-1-2b-thinking-on-device-reasoning-under-1gb", "source_type": "official_blog", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1 (mean of 5 runs)", "judge": "rule-based", "harness": "official (LFM eval, third-party)", "prompt_style": "default", "temperature": "default" }, "notes": "Liquid AI LFM2.5-1.2B-Thinking blog: qwen3-1.7b=81.92. (third-party self-test by Liquid AI)" } ] }, { "model_id": "qwen3-1.7b", "benchmark_id": "gpqa_diamond", "score": 40.1, "reference_url": "https://arxiv.org/abs/2505.09388", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Qwen3 tech report Table 13/15/17/19 Thinking: GPQA-Diamond 40.1", "candidates": [ { "score": 36.93, "reference_url": "https://www.liquid.ai/blog/lfm2-5-1-2b-thinking-on-device-reasoning-under-1gb", "source_type": "official_blog", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1 (mean of 5 runs)", "judge": "rule-based", "harness": "official (LFM eval, third-party)", "prompt_style": "default", "temperature": "default" }, "notes": "Liquid AI LFM2.5-1.2B-Thinking blog: qwen3-1.7b=36.93. (third-party self-test by Liquid AI)" } ] }, { "model_id": "qwen3-1.7b", "benchmark_id": "aime_2024", "score": 48.3, "reference_url": "https://arxiv.org/abs/2505.09388", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Qwen3 tech report Table 13/15/17/19 Thinking: AIME 2024 48.3" }, { "model_id": "qwen3-1.7b", "benchmark_id": "aime_2025", "score": 36.8, "reference_url": "https://arxiv.org/abs/2505.09388", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Qwen3 tech report Table 13/15/17/19 Thinking: AIME 2025 36.8", "candidates": [ { "score": 36.27, "reference_url": "https://www.liquid.ai/blog/lfm2-5-1-2b-thinking-on-device-reasoning-under-1gb", "source_type": "official_blog", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1 (mean of 5 runs)", "judge": "rule-based", "harness": "official (LFM eval, third-party)", "prompt_style": "default", "temperature": "default" }, "notes": "Liquid AI LFM2.5-1.2B-Thinking blog: qwen3-1.7b=36.27. (third-party self-test by Liquid AI)" } ] }, { "model_id": "qwen3-1.7b", "benchmark_id": "ifeval", "score": 72.5, "reference_url": "https://arxiv.org/abs/2505.09388", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Qwen3 tech report Table 13/15/17/19 Thinking: IFEval 72.5", "candidates": [ { "score": 71.65, "reference_url": "https://www.liquid.ai/blog/lfm2-5-1-2b-thinking-on-device-reasoning-under-1gb", "source_type": "official_blog", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1 (mean of 5 runs)", "judge": "rule-based", "harness": "official (LFM eval, third-party)", "prompt_style": "default", "temperature": "default" }, "notes": "Liquid AI LFM2.5-1.2B-Thinking blog: qwen3-1.7b=71.65. (third-party self-test by Liquid AI)" } ] }, { "model_id": "qwen3-1.7b", "benchmark_id": "arena_hard", "score": 43.1, "reference_url": "https://arxiv.org/abs/2505.09388", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Qwen3 tech report Table 13/15/17/19 Thinking: Arena-Hard 43.1" }, { "model_id": "qwen3-1.7b", "benchmark_id": "livecodebench", "score": 33.2, "reference_url": "https://arxiv.org/abs/2505.09388", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Qwen3 tech report Table 13/15/17/19 Thinking: LiveCodeBench v5 33.2" }, { "model_id": "qwen3-1.7b", "benchmark_id": "codeforces_rating", "score": 1200, "reference_url": "https://qwenlm.github.io/blog/qwen3/", "audit_status": "needs_review", "notes": "Web search: 'no official Codeforces ELO given for Qwen3-1.7B'; value 1200 not confirmed in any accessible text source. R5g(c)" }, { "model_id": "qwen3-1.7b", "benchmark_id": "gsm8k", "score": 85.6, "reference_url": "https://www.liquid.ai/blog/lfm2-5-1-2b-thinking-on-device-reasoning-under-1gb", "audit_status": "verified", "notes": "Liquid AI LFM2.5-1.2B-Thinking blog: qwen3-1.7b=85.6. (third-party self-test by Liquid AI)", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1 (mean of 5 runs)", "judge": "rule-based", "harness": "official (LFM eval, third-party)", "prompt_style": "default", "temperature": "default" }, "matches_canonical": false, "source_type": "official_blog" }, { "model_id": "qwen3-1.7b", "benchmark_id": "humaneval", "score": 60.0, "reference_url": "https://arxiv.org/abs/2505.09388", "audit_status": "dropped", "rule_id": "R5g-ghost", "audit_note": "Same as above: paper has no standalone HumanEval result.", "audited_by": "lychee-audit-bot", "audited_at": "2026-04-29T13:17:14Z" }, { "model_id": "qwen3-8b", "benchmark_id": "gpqa_diamond", "score": 62.0, "reference_url": "https://arxiv.org/abs/2505.09388", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Qwen3 tech report Thinking: GPQA-Diamond 62.0. [R5d: prior unverified value 63.3 from https://arxiv.org/abs/2505.09388 deleted.]" }, { "model_id": "qwen3-8b", "benchmark_id": "mmlu", "score": 76.89, "reference_url": "https://arxiv.org/abs/2505.09388", "audit_status": "dropped", "notes": " | DROPPED: BP value sourced from Base model column (Table 3-8) but model is_reasoning=true → canonical = Thinking variant. Wrong-variant cell." }, { "model_id": "qwen3-8b", "benchmark_id": "mmlu_pro", "score": 56.73, "reference_url": "https://arxiv.org/abs/2505.09388", "audit_status": "dropped", "notes": " | DROPPED: BP value sourced from Base model column (Table 3-8) but model is_reasoning=true → canonical = Thinking variant. Wrong-variant cell." }, { "model_id": "qwen3-8b", "benchmark_id": "aime_2024", "score": 76.0, "reference_url": "https://arxiv.org/abs/2505.09388", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Qwen3 tech report Table 13/15/17/19 Thinking: AIME 2024 76.0" }, { "model_id": "qwen3-8b", "benchmark_id": "aime_2025", "score": 67.3, "reference_url": "https://arxiv.org/abs/2505.09388", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Qwen3 tech report Thinking: AIME 2025 67.3. [R5d: prior unverified value 67.3 from https://huggingface.co/deepseek-ai/DeepSeek-R1-0528 deleted.]" }, { "model_id": "qwen3-8b", "benchmark_id": "arena_hard", "score": 85.8, "reference_url": "https://arxiv.org/abs/2505.09388", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Qwen3 tech report Thinking: Arena-Hard 85.8. [R5d: prior unverified value 76.0 from https://qwenlm.github.io/blog/qwen3/ deleted.]" }, { "model_id": "qwen3-8b", "benchmark_id": "codeforces_rating", "score": 1785, "reference_url": "https://arxiv.org/abs/2505.09388", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Qwen3 tech report Thinking: CodeForces 1785 (95.6%). [R5d: prior unverified value 1850 from https://qwenlm.github.io/blog/qwen3/ deleted.]" }, { "model_id": "qwen3-8b", "benchmark_id": "gsm8k", "score": 89.84, "reference_url": "https://arxiv.org/abs/2505.09388", "audit_status": "dropped", "notes": " | DROPPED: BP value sourced from Base model column (Table 3-8) but model is_reasoning=true → canonical = Thinking variant. Wrong-variant cell." }, { "model_id": "qwen3-8b", "benchmark_id": "humaneval", "score": 82.0, "reference_url": "https://qwenlm.github.io/blog/qwen3/", "audit_status": "verified", "reported_setting": "Qwen3 blog performance table, thinking mode", "source_type": "official_blog", "matches_canonical": true, "notes": "Confirmed via official Qwen3 blog (82.0%)" }, { "model_id": "qwen3-8b", "benchmark_id": "ifeval", "score": 85.0, "reference_url": "https://arxiv.org/abs/2505.09388", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Qwen3 tech report Thinking: IFEval 85.0. [R5d: prior unverified value 80.0 from https://arxiv.org/abs/2505.09388 deleted.]" }, { "model_id": "qwen3-8b", "benchmark_id": "livecodebench", "score": 57.5, "reference_url": "https://arxiv.org/abs/2505.09388", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Qwen3 tech report Thinking: LiveCodeBench v5 57.5. [R5d: prior unverified value 48.0 from https://qwenlm.github.io/blog/qwen3/ deleted.]" }, { "model_id": "qwen3-8b", "benchmark_id": "math_500", "score": 97.4, "reference_url": "https://arxiv.org/abs/2505.09388", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Qwen3 tech report Thinking: MATH-500 97.4. [R5d: prior unverified value 94.0 from https://qwenlm.github.io/blog/qwen3/ deleted.]" }, { "model_id": "qwen3-8b", "benchmark_id": "ifbench", "score": 35, "reference_url": "https://github.com/allenai/IFBench", "audit_status": "verified", "reported_setting": "IFBench score, 0-shot", "source_type": "leaderboard", "matches_canonical": true }, { "model_id": "qwen3-14b", "benchmark_id": "mmlu_pro", "score": 61.03, "reference_url": "https://arxiv.org/abs/2505.09388", "audit_status": "dropped", "notes": " | DROPPED: BP value sourced from Base model column (Table 3-8) but model is_reasoning=true → canonical = Thinking variant. Wrong-variant cell." }, { "model_id": "qwen3-14b", "benchmark_id": "mmlu", "score": 81.05, "reference_url": "https://arxiv.org/abs/2505.09388", "audit_status": "dropped", "notes": " | DROPPED: BP value sourced from Base model column (Table 3-8) but model is_reasoning=true → canonical = Thinking variant. Wrong-variant cell." }, { "model_id": "qwen3-14b", "benchmark_id": "arena_hard", "score": 91.7, "reference_url": "https://arxiv.org/abs/2505.09388", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Qwen3 tech report Thinking: Arena-Hard 91.7. [R5d: prior unverified value 85.5 from https://arxiv.org/abs/2505.09388 deleted.]" }, { "model_id": "qwen3-14b", "benchmark_id": "aime_2024", "score": 79.3, "reference_url": "https://arxiv.org/abs/2505.09388", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Qwen3 tech report Thinking: AIME 2024 79.3. [R5d: prior unverified value 75.0 from https://arxiv.org/abs/2505.09388 deleted.]" }, { "model_id": "qwen3-14b", "benchmark_id": "aime_2025", "score": 70.4, "reference_url": "https://arxiv.org/abs/2505.09388", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Qwen3 tech report Thinking: AIME 2025 70.4. [R5d: prior unverified value 72.0 from https://arxiv.org/abs/2505.09388 deleted.]" }, { "model_id": "qwen3-14b", "benchmark_id": "codeforces_rating", "score": 1766, "reference_url": "https://arxiv.org/abs/2505.09388", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Qwen3 tech report Thinking: CodeForces 1766 (95.3%). [R5d: prior unverified value 1900 from https://qwenlm.github.io/blog/qwen3/ deleted.]" }, { "model_id": "qwen3-14b", "benchmark_id": "gpqa_diamond", "score": 64.0, "reference_url": "https://arxiv.org/abs/2505.09388", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Qwen3 tech report Table 13/15/17/19 Thinking: GPQA-Diamond 64.0" }, { "model_id": "qwen3-14b", "benchmark_id": "gsm8k", "score": 92.49, "reference_url": "https://arxiv.org/abs/2505.09388", "audit_status": "dropped", "notes": " | DROPPED: BP value sourced from Base model column (Table 3-8) but model is_reasoning=true → canonical = Thinking variant. Wrong-variant cell." }, { "model_id": "qwen3-14b", "benchmark_id": "humaneval", "score": 85.0, "reference_url": "https://qwenlm.github.io/blog/qwen3/", "audit_status": "needs_review", "notes": "Blog tables are images; value 85.0 not confirmed with exact precision from accessible text sources. R5g(c)" }, { "model_id": "qwen3-14b", "benchmark_id": "livecodebench", "score": 63.5, "reference_url": "https://arxiv.org/abs/2505.09388", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Qwen3 tech report Thinking: LiveCodeBench v5 63.5. [R5d: prior unverified value 55.0 from https://qwenlm.github.io/blog/qwen3/ deleted.]" }, { "model_id": "qwen3-14b", "benchmark_id": "math_500", "score": 96.8, "reference_url": "https://arxiv.org/abs/2505.09388", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Qwen3 tech report Thinking: MATH-500 96.8. [R5d: prior unverified value 95.0 from https://qwenlm.github.io/blog/qwen3/ deleted.]" }, { "model_id": "qwen3-30b-a3b", "benchmark_id": "gpqa_diamond", "score": 65.8, "reference_url": "https://arxiv.org/abs/2505.09388", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Qwen3 tech report Table 13/15/17/19 Thinking: GPQA-Diamond 65.8" }, { "model_id": "qwen3-30b-a3b", "benchmark_id": "codeforces_rating", "score": 1974, "reference_url": "https://arxiv.org/abs/2505.09388", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Qwen3 tech report Table 13/15/17/19 Thinking: CodeForces 1974 (97.7%)" }, { "model_id": "qwen3-30b-a3b", "benchmark_id": "arena_hard", "score": 91.0, "reference_url": "https://arxiv.org/abs/2505.09388", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Qwen3 tech report Table 13/15/17/19 Thinking: Arena-Hard 91.0" }, { "model_id": "qwen3-30b-a3b", "benchmark_id": "aime_2024", "score": 80.4, "reference_url": "https://arxiv.org/abs/2505.09388", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Qwen3 tech report Thinking: AIME 2024 80.4. [R5d: prior unverified value 78.0 from https://qwenlm.github.io/blog/qwen3/ deleted.]" }, { "model_id": "qwen3-30b-a3b", "benchmark_id": "aime_2025", "score": 70.9, "reference_url": "https://arxiv.org/abs/2505.09388", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Qwen3 tech report Thinking: AIME 2025 70.9 (May 2025 release; replaces 85.0 from Thinking-2507 fine-tune). [R5d: prior unverified value 85.0 from https://huggingface.co/Qwen/Qwen3-30B-A3B-Thinking-2507 deleted.]" }, { "model_id": "qwen3-30b-a3b", "benchmark_id": "gsm8k", "score": 91.81, "reference_url": "https://arxiv.org/abs/2505.09388", "audit_status": "dropped", "notes": " | DROPPED: BP value sourced from Base model column (Table 3-8) but model is_reasoning=true → canonical = Thinking variant. Wrong-variant cell." }, { "model_id": "qwen3-30b-a3b", "benchmark_id": "hmmt_feb_2025", "score": 71.4, "reference_url": "https://huggingface.co/Qwen/Qwen3-30B-A3B-Thinking-2507", "audit_status": "dropped", "notes": " | DROPPED: BP value from Qwen3-X-Thinking-2507 (post-July fine-tune), but BP model release_date=2025-05-15. Wrong-variant (later checkpoint)." }, { "model_id": "qwen3-30b-a3b", "benchmark_id": "humaneval", "score": 82.0, "reference_url": "https://qwenlm.github.io/blog/qwen3/", "audit_status": "needs_review", "notes": "Blog tables are images; value 82.0 approximately consistent but exact confirmation unavailable. R5g(c)" }, { "model_id": "qwen3-30b-a3b", "benchmark_id": "ifeval", "score": 86.5, "reference_url": "https://arxiv.org/abs/2505.09388", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Qwen3 tech report Thinking: IFEval 86.5. [R5d: prior unverified value 88.9 from https://huggingface.co/Qwen/Qwen3-30B-A3B-Thinking-2507 deleted.]" }, { "model_id": "qwen3-30b-a3b", "benchmark_id": "livecodebench", "score": 62.6, "reference_url": "https://arxiv.org/abs/2505.09388", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Qwen3 tech report Thinking: LiveCodeBench v5 62.6. [R5d: prior unverified value 66.0 from https://huggingface.co/Qwen/Qwen3-30B-A3B-Thinking-2507 deleted.]" }, { "model_id": "qwen3-30b-a3b", "benchmark_id": "math_500", "score": 98.0, "reference_url": "https://arxiv.org/abs/2505.09388", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Qwen3 tech report Thinking: MATH-500 98.0. [R5d: prior unverified value 96.0 from https://qwenlm.github.io/blog/qwen3/ deleted.]" }, { "model_id": "qwen3-30b-a3b", "benchmark_id": "mmlu", "score": 81.38, "reference_url": "https://arxiv.org/abs/2505.09388", "audit_status": "dropped", "notes": " | DROPPED: BP value sourced from Base model column (Table 3-8) but model is_reasoning=true → canonical = Thinking variant. Wrong-variant cell." }, { "model_id": "qwen3-30b-a3b", "benchmark_id": "mmlu_pro", "score": 80.9, "reference_url": "https://huggingface.co/Qwen/Qwen3-30B-A3B-Thinking-2507", "audit_status": "dropped", "notes": " | DROPPED: BP value from Qwen3-X-Thinking-2507 (post-July fine-tune), but BP model release_date=2025-05-15. Wrong-variant (later checkpoint)." }, { "model_id": "qwq-32b", "benchmark_id": "aime_2024", "score": 79.5, "reference_url": "https://qwenlm.github.io/blog/qwq-32b/", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "QwQ-32B blog (qwenlm.github.io/blog/qwq-32b/) + Qwen3 tech report Table 13 baseline column: AIME 2024 79.5" }, { "model_id": "qwq-32b", "benchmark_id": "gpqa_diamond", "score": 65.6, "reference_url": "https://arxiv.org/abs/2505.09388", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "QwQ-32B blog (qwenlm.github.io/blog/qwq-32b/) + Qwen3 tech report Table 13 baseline column: GPQA-Diamond 65.6 (Qwen3 T13; supersedes preview 65.2) [R5d: prior unverified value 65.2 from https://qwenlm.github.io/blog/qwq-32b-preview/ deleted.]", "candidates": [ { "score": 59.5, "reference_url": "https://arxiv.org/abs/2504.21318", "source_type": "tech_report", "reported_setting": { "mode": "unknown", "effort": "unknown", "tools": "unknown", "sampling": "unknown", "judge": "rule-based", "harness": "Microsoft Phi internal eval", "prompt_style": "unknown", "temperature": "unknown" }, "notes": "Phi-4-reasoning paper Table 1 (qwq-32b column): gpqa_diamond=59.5 (alt measurement, mc=false)" } ] }, { "model_id": "qwq-32b", "benchmark_id": "math_500", "score": 98.0, "reference_url": "https://arxiv.org/abs/2505.09388", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "QwQ-32B blog (qwenlm.github.io/blog/qwq-32b/) + Qwen3 tech report Table 13 baseline column: MATH-500 98.0 (Qwen3 T13; supersedes preview 90.6) [R5d: prior unverified value 90.6 from https://qwenlm.github.io/blog/qwq-32b-preview/ deleted.]" }, { "model_id": "qwq-32b", "benchmark_id": "livecodebench", "score": 62.7, "reference_url": "https://arxiv.org/abs/2505.09388", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "QwQ-32B blog (qwenlm.github.io/blog/qwq-32b/) + Qwen3 tech report Table 13 baseline column: LiveCodeBench v5 62.7 (Qwen3 tech report Table 13 baseline column; supersedes earlier preview blog 50.0) [R5d: prior unverified value 50.0 from https://qwenlm.github.io/blog/qwq-32b-preview/ deleted.]", "candidates": [ { "score": 63.4, "reference_url": "https://arxiv.org/abs/2504.21318", "source_type": "tech_report", "reported_setting": { "mode": "unknown", "effort": "unknown", "tools": "unknown", "sampling": "unknown", "judge": "rule-based", "harness": "Microsoft Phi internal eval", "prompt_style": "unknown", "temperature": "unknown" }, "notes": "Phi-4-reasoning paper Table 1 (qwq-32b column): livecodebench=63.4 (alt measurement, mc=false)" } ] }, { "model_id": "qwq-32b", "benchmark_id": "arena_hard", "score": 89.5, "reference_url": "https://arxiv.org/abs/2505.09388", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "QwQ-32B blog (qwenlm.github.io/blog/qwq-32b/) + Qwen3 tech report Table 13 baseline column: Arena-Hard 89.5 (Qwen3 T13; supersedes lmarena github 60.9) [R5d: prior unverified value 60.9 from https://github.com/lmarena/arena-hard-auto deleted.]" }, { "model_id": "qwq-32b", "benchmark_id": "codeforces_rating", "score": 1982, "reference_url": "https://arxiv.org/abs/2505.09388", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "QwQ-32B blog (qwenlm.github.io/blog/qwq-32b/) + Qwen3 tech report Table 13 baseline column: CodeForces rating 1982 (Qwen3 T13; supersedes 1316 from DeepSeek R1 arxiv comparison) [R5d: prior unverified value 1316 from https://arxiv.org/html/2501.12948v1 deleted.]" }, { "model_id": "qwq-32b", "benchmark_id": "aime_2025", "score": 69.5, "reference_url": "https://arxiv.org/abs/2505.09388", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "QwQ-32B blog (qwenlm.github.io/blog/qwq-32b/) + Qwen3 tech report Table 13 baseline column: AIME 2025 69.5 (Qwen3 T13) [R5d: prior unverified value 50.0 from https://matharena.ai/models/qwen_qwq_preview deleted.]", "candidates": [ { "score": 65.8, "reference_url": "https://arxiv.org/abs/2504.21318", "source_type": "tech_report", "reported_setting": { "mode": "unknown", "effort": "unknown", "tools": "unknown", "sampling": "unknown", "judge": "rule-based", "harness": "Microsoft Phi internal eval", "prompt_style": "unknown", "temperature": "unknown" }, "notes": "Phi-4-reasoning paper Table 1 (qwq-32b column): aime_2025=65.8 (alt measurement, mc=false)" } ] }, { "model_id": "qwq-32b", "benchmark_id": "ifeval", "score": 83.9, "reference_url": "https://arxiv.org/abs/2505.09388", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "QwQ-32B blog (qwenlm.github.io/blog/qwq-32b/) + Qwen3 tech report Table 13 baseline column: IFEval strict prompt 83.9 [R5d: prior unverified value 80.0 from https://qwenlm.github.io/blog/qwq-32b/ deleted.]", "candidates": [ { "score": 85.8, "reference_url": "https://arxiv.org/abs/2501.00656", "source_type": "tech_report", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "OLMES", "prompt_style": "default", "temperature": "default" }, "notes": "OLMo 2 paper Table 7 Instruct: alt measurement 85.8 (mc=false)." } ] }, { "model_id": "qwq-32b", "benchmark_id": "mmlu_pro", "score": 62.0, "reference_url": "https://qwenlm.github.io/blog/qwq-32b/", "audit_status": "dropped", "notes": " | DROPPED: BP cell URL points to qwenlm.github.io/blog/qwq-32b/ but that blog does not report MMLU-Pro; not in Qwen3 tech report Table 13 either. Ghost cell per R5g(a)." }, { "model_id": "qwq-32b", "benchmark_id": "bigcodebench", "score": 44.6, "reference_url": "https://bigcode-bench.github.io/", "audit_status": "verified", "reported_setting": "BigCodeBench Complete Instruct pass@1", "source_type": "leaderboard", "matches_canonical": true, "notes": "Source entry: QwQ-32B-Preview (2024-11-28); regular QwQ-32B has no instruct score in leaderboard" }, { "model_id": "qwq-32b", "benchmark_id": "hmmt_feb_2025", "score": 47.5, "reference_url": "https://arxiv.org/abs/2504.21318", "reported_setting": { "mode": "unknown", "effort": "unknown", "tools": "unknown", "sampling": "unknown", "judge": "rule-based", "harness": "Microsoft Phi internal eval", "prompt_style": "unknown", "temperature": "unknown" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Phi-4-reasoning paper Table 1 (qwq-32b column): hmmt_2025=47.5 (re-sourced from https://arxiv.org/abs/2504.21318, matches BP value)" }, { "model_id": "qwq-32b", "benchmark_id": "humaneval", "score": 78, "reference_url": "https://medium.com/towards-agi/qwq-32b-preview-benchmarks-revolutionizing-ai-reasoning-capabilities-b2014a00c208", "audit_status": "dropped", "rule_id": "R5h-third-party-blog", "notes": "DROPPED (R5h): random third-party blog/aggregator, no primary trail" }, { "model_id": "qwq-32b", "benchmark_id": "mmlu", "score": 88.4, "reference_url": "https://arxiv.org/abs/2501.00656", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "OLMES", "prompt_style": "default", "temperature": "default" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "OLMo 2 paper Table 7 Instruct: mmlu=88.4. [R5d: prior unverified value 79 from https://medium.com/towards-agi/qwq-32b-preview-benchmarks-revolutionizing-ai-reasoning-capabilities-b2014a00c208 deleted.]" }, { "model_id": "qwq-32b", "benchmark_id": "swe_bench_verified", "score": 35, "reference_url": "https://medium.com/towards-agi/qwq-32b-preview-benchmarks-revolutionizing-ai-reasoning-capabilities-b2014a00c208", "audit_status": "dropped", "rule_id": "R5h-third-party-blog", "notes": "DROPPED (R5h): random third-party blog/aggregator, no primary trail" }, { "model_id": "qwen3.5-397b", "benchmark_id": "gpqa_diamond", "score": 88.4, "reference_url": "https://huggingface.co/Qwen/Qwen3.5-397B-A17B", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Qwen3.5-397B-A17B HF model card cross-model table: GPQA Diamond 88.4 (was 92.4 = GPT5.2 col). [R5d: prior unverified value 92.4 from https://huggingface.co/Qwen/Qwen3.5-397B-A17B deleted — was scraped from wrong column (likely GPT5.2 column).]", "candidates": [ { "score": 87.1, "reference_url": "https://research.nvidia.com/labs/nemotron/files/NVIDIA-Nemotron-3-Ultra-Technical-Report.pdf", "source_type": "tech_report", "reported_setting": { "version_metric": "no tools / %", "tools": "none", "harness": "NeMo Evaluator", "sampling": "8-run average" }, "notes": "NVIDIA Nemotron 3 Ultra technical report, Final Table 10: Qwen3.5-397B-A17B: gpqa_diamond=87.1." } ] }, { "model_id": "qwen3.5-397b", "benchmark_id": "swe_bench_verified", "score": 76.4, "reference_url": "https://huggingface.co/Qwen/Qwen3.5-397B-A17B", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Qwen3.5-397B-A17B HF model card cross-model table: SWE-bench Verified 76.4 (was 80.0 = GPT5.2). [R5d: prior unverified value 80.0 from https://huggingface.co/Qwen/Qwen3.5-397B-A17B deleted — was scraped from wrong column (likely GPT5.2 column).]", "candidates": [ { "score": 73.6, "reference_url": "https://research.nvidia.com/labs/nemotron/files/NVIDIA-Nemotron-3-Ultra-Technical-Report.pdf", "source_type": "tech_report", "reported_setting": { "version_metric": "Verified / % resolved", "tools": "agentic repository editing", "harness": "Harbor / OpenHands / AWS ECS", "sampling": "3-run average" }, "notes": "NVIDIA Nemotron 3 Ultra technical report, Final Table 10: Qwen3.5-397B-A17B: swe_bench_verified=73.6." }, { "score": 76.4, "reference_url": "https://huggingface.co/mistralai/Mistral-Medium-3.5-128B/resolve/22b2b868a15677cfa6061277ed2f653d1349a9ab/images/image4.png?download=true", "source_type": "official_model_card", "reported_setting": { "mode": "thinking", "effort": "default (maximum reasoning setting)", "tools": "agentic coding scaffold and SWE-bench verifier", "sampling": "source does not state trial count", "judge": "benchmark-specified", "harness": "Mistral self-reported SWE-bench Verified evaluation", "prompt_style": "default (thinking by default)", "temperature": "default", "context": "default (262K)" }, "notes": "Displayed exactly: '76.4'. Research observation: medium-image4.png:1:6:score:reported. Chart footnotes: *All model evaluations were run with maximum reasoning settings. | **Sonnet 4.6 encountered higher rates of reasoning truncation due to external API restrictions than other models, which affected its performance. | *** Self-reported | **** Self-reported, Mistral is using context management and a discard-all strategy at 100k tokens. | τ³ scores as reported by Sierra for Claude Sonnet 4.5 and Qwen3.5. Others with user simulator: gpt-5.2 with reasoning_effort: low. 4 trials. Banking domain evaluated with terminal- or embedding-based agentic search retrieval, only highest score is reported." } ] }, { "model_id": "qwen3.5-397b", "benchmark_id": "mmlu_pro", "score": 87.8, "reference_url": "https://huggingface.co/Qwen/Qwen3.5-397B-A17B", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Qwen3.5-397B-A17B HF model card cross-model table: MMLU-Pro 87.8 (was 87.4 = GPT5.2). [R5d: prior unverified value 87.4 from https://huggingface.co/Qwen/Qwen3.5-397B-A17B deleted — was scraped from wrong column (likely GPT5.2 column).]", "candidates": [ { "score": 88.3, "reference_url": "https://research.nvidia.com/labs/nemotron/files/NVIDIA-Nemotron-3-Ultra-Technical-Report.pdf", "source_type": "tech_report", "reported_setting": { "version_metric": "MMLU-Pro / %", "tools": "none", "harness": "NeMo Evaluator", "sampling": "pass@1" }, "notes": "NVIDIA Nemotron 3 Ultra technical report, Final Table 10: Qwen3.5-397B-A17B: mmlu_pro=88.3." } ] }, { "model_id": "qwen3.5-397b", "benchmark_id": "livecodebench", "score": 83.6, "reference_url": "https://huggingface.co/Qwen/Qwen3.5-397B-A17B", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Qwen3.5-397B-A17B HF model card cross-model table: LiveCodeBench v6 83.6 (was 87.7 = GPT5.2). [R5d: prior unverified value 87.7 from https://huggingface.co/Qwen/Qwen3.5-397B-A17B deleted — was scraped from wrong column (likely GPT5.2 column).]" }, { "model_id": "qwen3.5-397b", "benchmark_id": "ifeval", "score": 92.6, "reference_url": "https://huggingface.co/Qwen/Qwen3.5-397B-A17B", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Qwen3.5-397B-A17B HF model card cross-model table: IFEval 92.6 (was 94.8 = GPT5.2). [R5d: prior unverified value 94.8 from https://huggingface.co/Qwen/Qwen3.5-397B-A17B deleted — was scraped from wrong column (likely GPT5.2 column).]" }, { "model_id": "qwen3.5-397b", "benchmark_id": "aime_2025", "score": 83.1, "reference_url": "https://huggingface.co/mistralai/Mistral-Medium-3.5-128B/resolve/22b2b868a15677cfa6061277ed2f653d1349a9ab/images/image1.png?download=true", "reported_setting": { "mode": "thinking", "effort": "default (maximum reasoning setting)", "tools": "none", "sampling": "samples=16 (reported as avg@16)", "judge": "benchmark-specified", "harness": "Mistral official maximum-reasoning comparison", "prompt_style": "default (thinking by default)", "temperature": "default", "context": "default (262K)" }, "matches_canonical": false, "source_type": "official_model_card", "audit_status": "verified", "notes": "Displayed exactly: '83.1'. Research observation: medium-image1.png:1:5:score:reported. Chart footnotes: *All model evaluations were run with maximum reasoning settings. | **Sonnet 4.6 encountered higher rates of reasoning truncation due to external API restrictions than other models, which affected its performance." }, { "model_id": "qwen3.5-397b", "benchmark_id": "humaneval", "score": 92.0, "reference_url": "https://venturebeat.com/technology/alibabas-qwen-3-5-397b-a17/", "audit_status": "dropped", "rule_id": "R5h-third-party-blog", "notes": "DROPPED (R5h): random third-party blog/aggregator, no primary trail" }, { "model_id": "qwen3.5-397b", "benchmark_id": "math_500", "score": 98.0, "reference_url": "https://venturebeat.com/technology/alibabas-qwen-3-5-397b-a17/", "audit_status": "dropped", "rule_id": "R5h-third-party-blog", "notes": "DROPPED (R5h): random third-party blog/aggregator, no primary trail" }, { "model_id": "qwen3.5-397b", "benchmark_id": "mmlu", "score": 88.6, "reference_url": "https://automatio.ai/models/qwen3-5-397b-a17b", "audit_status": "dropped", "rule_id": "R5h-third-party-blog", "notes": "DROPPED (R5h): random third-party blog/aggregator, no primary trail" }, { "model_id": "qwen3.5-397b", "benchmark_id": "aime_2024", "score": 94, "reference_url": "https://artificialanalysis.ai/models/qwen3-5-397b-a17b", "audit_status": "needs_review", "notes": "Reference URL is AA model page (qwen3-5-397b-a17b). AA shows aime=null for all qwen3-5 models; AIME benchmark not yet evaluated by AA for this model. Cannot verify 94 from AA source." }, { "model_id": "qwen3.5-397b", "benchmark_id": "aime_2026", "score": 91.3, "reference_url": "https://huggingface.co/Qwen/Qwen3.5-397B-A17B", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Qwen3.5-397B-A17B HF model card cross-model table: AIME 2026 91.3 (was 96.7 = GPT5.2). [R5d: prior unverified value 96.7 from https://huggingface.co/Qwen/Qwen3.5-397B-A17B deleted — was scraped from wrong column (likely GPT5.2 column).]" }, { "model_id": "qwen3.5-397b", "benchmark_id": "codeforces_rating", "score": 2200, "reference_url": "https://artificialanalysis.ai/models/qwen3-5-397b-a17b", "audit_status": "needs_review", "notes": "Reference URL is AA model page (qwen3-5-397b-a17b). AA does not publish codeforces_rating data; field is not present in AA model data. Cannot verify 2200 from AA source. Original source may be Qwen tech report." }, { "model_id": "qwen3.5-397b", "benchmark_id": "hle", "score": 28.7, "reference_url": "https://huggingface.co/Qwen/Qwen3.5-397B-A17B", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Qwen3.5-397B-A17B HF model card cross-model table: HLE 28.7 (was 35.5 = GPT5.2). [R5d: prior unverified value 35.5 from https://huggingface.co/Qwen/Qwen3.5-397B-A17B deleted — was scraped from wrong column (likely GPT5.2 column).]" }, { "model_id": "qwen3.5-397b", "benchmark_id": "ifbench", "score": 76.5, "reference_url": "https://huggingface.co/Qwen/Qwen3.5-397B-A17B", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Qwen3.5-397B-A17B HF model card cross-model table: IFBench 76.5 (was 75.4 = GPT5.2). [R5d: prior unverified value 75.4 from https://huggingface.co/Qwen/Qwen3.5-397B-A17B deleted — was scraped from wrong column (likely GPT5.2 column).]", "candidates": [ { "score": 78.2, "reference_url": "https://research.nvidia.com/labs/nemotron/files/NVIDIA-Nemotron-3-Ultra-Technical-Report.pdf", "source_type": "tech_report", "reported_setting": { "version_metric": "prompt-level loose accuracy / %", "tools": "none", "harness": "NeMo Evaluator", "sampling": "8-run average" }, "notes": "NVIDIA Nemotron 3 Ultra technical report, Final Table 10: Qwen3.5-397B-A17B: ifbench=78.2." }, { "score": 76.5, "reference_url": "https://huggingface.co/mistralai/Mistral-Medium-3.5-128B/resolve/22b2b868a15677cfa6061277ed2f653d1349a9ab/images/image1.png?download=true", "source_type": "official_model_card", "reported_setting": { "mode": "thinking", "effort": "default (maximum reasoning setting)", "tools": "none", "sampling": "pass@1; repeat count undisclosed", "judge": "benchmark-specified", "harness": "Mistral official maximum-reasoning comparison", "prompt_style": "default (thinking by default)", "temperature": "default", "context": "default (262K)" }, "notes": "Displayed exactly: '76.5'. Research observation: medium-image1.png:2:5:score:reported. Chart footnotes: *All model evaluations were run with maximum reasoning settings. | **Sonnet 4.6 encountered higher rates of reasoning truncation due to external API restrictions than other models, which affected its performance." } ] }, { "model_id": "qwen3.5-397b", "benchmark_id": "mathvision", "score": 88.6, "reference_url": "https://huggingface.co/Qwen/Qwen3.5-397B-A17B", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Qwen3.5-397B-A17B HF model card cross-model table: MathVision 88.6 (HF Vision section; supersedes 90.3 from artificialanalysis.ai). [R5d: prior unverified value 90.3 from https://artificialanalysis.ai/models/qwen3-5-397b-a17b deleted — was scraped from wrong column (likely GPT5.2 column).]" }, { "model_id": "qwen3.5-397b", "benchmark_id": "mmmu", "score": 85.0, "reference_url": "https://huggingface.co/Qwen/Qwen3.5-397B-A17B", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Qwen3.5-397B-A17B HF model card cross-model table: MMMU 85.0" }, { "model_id": "qwen3.5-397b", "benchmark_id": "simpleqa", "score": 35, "reference_url": "https://artificialanalysis.ai/models/qwen3-5-397b-a17b", "audit_status": "needs_review", "notes": "Reference URL is AA model page (qwen3-5-397b-a17b). AA does not publish SimpleQA scores; field is not present in AA model data for any model. Cannot verify from AA source. Original source needs identification." }, { "model_id": "qwen3.5-397b", "benchmark_id": "mmmlu", "score": 88.5, "reference_url": "https://huggingface.co/Qwen/Qwen3.5-397B-A17B", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Qwen3.5-397B-A17B HF model card cross-model table: MMMLU 88.5 (was 89.5 = GPT5.2). [R5d: prior unverified value 89.5 from https://huggingface.co/Qwen/Qwen3.5-397B-A17B deleted — was scraped from wrong column (likely GPT5.2 column).]" }, { "model_id": "qwen3.5-397b", "benchmark_id": "swe_bench_multilingual", "score": 69.3, "reference_url": "https://huggingface.co/Qwen/Qwen3.5-397B-A17B", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Qwen3.5-397B-A17B HF model card cross-model table: SWE-bench Multilingual 69.3 (was 72.0 = GPT5.2). [R5d: prior unverified value 72.0 from https://huggingface.co/Qwen/Qwen3.5-397B-A17B deleted — was scraped from wrong column (likely GPT5.2 column).]", "candidates": [ { "score": 65.0, "reference_url": "https://huggingface.co/internlm/Intern-S2-Preview-397B", "source_type": "model_card", "reported_setting": { "mode": "source does not state", "effort": "source does not state", "tools": "agentic code execution", "sampling": "source does not state", "judge": "benchmark-specific; exact version not stated", "harness": "Mini-SWE-Agent", "prompt_style": "source does not state", "temperature": "source does not state", "context": "source does not state", "notes": "Shanghai AI Lab evaluated all model columns using OpenCompass, VLMEvalKit, and AgentCompass. Per-model mode/effort/sampling are not reported; only Intern-S2 maximum evaluation lengths are disclosed." }, "notes": "Intern-S2-Preview-397B official model card: SWE-Bench-Multilingual = 65.0. This is a Shanghai AI Lab cross-model evaluation; the source does not disclose this model's exact mode, effort, or sampling." }, { "score": 70.9, "reference_url": "https://research.nvidia.com/labs/nemotron/files/NVIDIA-Nemotron-3-Ultra-Technical-Report.pdf", "source_type": "tech_report", "reported_setting": { "version_metric": "Multilingual / % resolved", "tools": "agentic repository editing", "harness": "Harbor / OpenHands / AWS ECS", "sampling": "3-run average" }, "notes": "NVIDIA Nemotron 3 Ultra technical report, Final Table 10: Qwen3.5-397B-A17B: swe_bench_multilingual=70.9." } ] }, { "model_id": "qwen3.5-397b", "benchmark_id": "imo_answerbench", "score": 80.9, "reference_url": "https://huggingface.co/Qwen/Qwen3.5-397B-A17B", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Qwen3.5-397B-A17B HF model card cross-model table: IMOAnswerBench 80.9 (was 86.3 = GPT5.2). [R5d: prior unverified value 86.3 from https://huggingface.co/Qwen/Qwen3.5-397B-A17B deleted — was scraped from wrong column (likely GPT5.2 column).]", "candidates": [ { "score": 83.1, "reference_url": "https://research.nvidia.com/labs/nemotron/files/NVIDIA-Nemotron-3-Ultra-Technical-Report.pdf", "source_type": "tech_report", "reported_setting": { "version_metric": "no tools / %", "tools": "none", "harness": "NeMo Evaluator", "sampling": "5-run average", "variant": "no tools" }, "notes": "NVIDIA Nemotron 3 Ultra technical report, Final Table 10: Qwen3.5-397B-A17B: imo_answerbench=83.1, variant=no tools." }, { "score": 84.51, "reference_url": "https://research.nvidia.com/labs/nemotron/files/NVIDIA-Nemotron-3-Ultra-Technical-Report.pdf", "source_type": "tech_report", "reported_setting": { "version_metric": "with tools / %", "tools": "benchmark tool environment", "harness": "NeMo Evaluator", "sampling": "pass@1", "variant": "with tools" }, "notes": "NVIDIA Nemotron 3 Ultra technical report, Final Table 10: Qwen3.5-397B-A17B: imo_answerbench=84.51, variant=with tools." } ] }, { "model_id": "qwen3.5-397b", "benchmark_id": "terminal_bench", "score": 52.5, "reference_url": "https://huggingface.co/Qwen/Qwen3.5-397B-A17B", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Qwen3.5-397B-A17B HF model card cross-model table: Terminal-Bench 2 52.5 (was 54.0 = GPT5.2). [R5d: prior unverified value 54.0 from https://huggingface.co/Qwen/Qwen3.5-397B-A17B deleted — was scraped from wrong column (likely GPT5.2 column).]" }, { "model_id": "qwen3.5-397b", "benchmark_id": "browsecomp", "score": 69.0, "reference_url": "https://huggingface.co/Qwen/Qwen3.5-397B-A17B", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Qwen3.5-397B-A17B HF model card cross-model table: BrowseComp 69.0 simple context-folding (was 65.8 = GPT5.2). [R5d: prior unverified value 65.8 from https://huggingface.co/Qwen/Qwen3.5-397B-A17B deleted — was scraped from wrong column (likely GPT5.2 column).]", "candidates": [ { "score": 78.6, "reference_url": "https://huggingface.co/Qwen/Qwen3.5-397B-A17B", "source_type": "model_card", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default", "notes": "discard-all context management" }, "notes": "Qwen3.5-397B-A17B HF model card cross-model table: BrowseComp 78.6 with discard-all context management strategy (alt setting)." }, { "score": 40.5, "reference_url": "https://research.nvidia.com/labs/nemotron/files/NVIDIA-Nemotron-3-Ultra-Technical-Report.pdf", "source_type": "tech_report", "reported_setting": { "version_metric": "BrowseComp / %", "tools": "Tavily web search and terminal workspace", "harness": "NVIDIA custom BrowseComp scaffold", "sampling": "pass@1" }, "notes": "NVIDIA Nemotron 3 Ultra technical report, Final Table 10: Qwen3.5-397B-A17B: browsecomp=40.5." }, { "score": 78.6, "reference_url": "https://huggingface.co/mistralai/Mistral-Medium-3.5-128B/resolve/22b2b868a15677cfa6061277ed2f653d1349a9ab/images/image4.png?download=true", "source_type": "official_model_card", "reported_setting": { "mode": "thinking", "effort": "default (maximum reasoning setting)", "tools": "web search/browser environment", "sampling": "source does not state trial count", "judge": "benchmark-specified", "harness": "Mistral official BrowseComp evaluation", "prompt_style": "default (thinking by default)", "temperature": "default", "context": "default (262K)" }, "notes": "Displayed exactly: '78.6'. Research observation: medium-image4.png:6:6:score:reported. Chart footnotes: *All model evaluations were run with maximum reasoning settings. | **Sonnet 4.6 encountered higher rates of reasoning truncation due to external API restrictions than other models, which affected its performance. | *** Self-reported | **** Self-reported, Mistral is using context management and a discard-all strategy at 100k tokens. | τ³ scores as reported by Sierra for Claude Sonnet 4.5 and Qwen3.5. Others with user simulator: gpt-5.2 with reasoning_effort: low. 4 trials. Banking domain evaluated with terminal- or embedding-based agentic search retrieval, only highest score is reported." } ] }, { "model_id": "qwen3.5-397b", "benchmark_id": "aa_lcr", "score": 68.7, "reference_url": "https://huggingface.co/Qwen/Qwen3.5-397B-A17B", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Qwen3.5-397B-A17B HF model card cross-model table: AA-LCR 68.7 (was 72.7 = GPT5.2). [R5d: prior unverified value 72.7 from https://huggingface.co/Qwen/Qwen3.5-397B-A17B deleted — was scraped from wrong column (likely GPT5.2 column).]", "candidates": [ { "score": 68.3, "reference_url": "https://research.nvidia.com/labs/nemotron/files/NVIDIA-Nemotron-3-Ultra-Technical-Report.pdf", "source_type": "tech_report", "reported_setting": { "version_metric": "AA-LCR / %", "tools": "none", "harness": "NeMo Evaluator", "sampling": "16-run average" }, "notes": "NVIDIA Nemotron 3 Ultra technical report, Final Table 10: Qwen3.5-397B-A17B: aa_lcr=68.3." } ] }, { "model_id": "qwen3.5-397b", "benchmark_id": "hmmt_feb_2025", "score": 94.8, "reference_url": "https://huggingface.co/Qwen/Qwen3.5-397B-A17B", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Qwen3.5-397B-A17B HF model card cross-model table: HMMT Feb 25 94.8 (was 99.4 = GPT5.2). [R5d: prior unverified value 99.4 from https://huggingface.co/Qwen/Qwen3.5-397B-A17B deleted — was scraped from wrong column (likely GPT5.2 column).]" }, { "model_id": "qwen3.5-397b", "benchmark_id": "hmmt_nov_2025", "score": 92.7, "reference_url": "https://huggingface.co/Qwen/Qwen3.5-397B-A17B", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Qwen3.5-397B-A17B HF model card cross-model table: HMMT Nov 25 92.7 (was 100.0 = GPT5.2). [R5d: prior unverified value 100.0 from https://huggingface.co/Qwen/Qwen3.5-397B-A17B deleted — was scraped from wrong column (likely GPT5.2 column).]" }, { "model_id": "qwen3.6-plus", "benchmark_id": "swe_bench_verified", "score": 78.8, "reference_url": "https://qwen.ai/blog?id=qwen3.6", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "agentic (bash + edit)", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "Qwen3.6 blog: qwen3.6-plus=78.8 (vs Claude Opus 4.5=80.9, Kimi-K2.5=76.8, GLM5=77.8, Qwen3.5-397B=76.2)." }, { "model_id": "qwen3.6-plus", "benchmark_id": "swe_bench_pro", "score": 56.6, "reference_url": "https://z.ai/blog/glm-5.1", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "agentic (bash + edit)", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "GLM-5.1 blog: qwen3.6-plus=56.6." }, { "model_id": "qwen3.6-plus", "benchmark_id": "terminal_bench", "score": 61.6, "reference_url": "https://z.ai/blog/glm-5.1", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "agentic", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "GLM-5.1 blog: qwen3.6-plus=61.6." }, { "model_id": "qwen3.6-plus", "benchmark_id": "hle", "score": 28.8, "reference_url": "https://z.ai/blog/glm-5.1", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "GLM-5.1 blog: qwen3.6-plus=28.8." }, { "model_id": "qwen3.6-plus", "benchmark_id": "mmlu_pro", "score": 88.5, "reference_url": "https://qwen.ai/blog?id=qwen3.6", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "Qwen3.6 blog: qwen3.6-plus=88.5 (vs Claude Opus 4.5=89.5, Kimi-K2.5=87.1, GLM5=85.7, Qwen3.5-397B=87.8)." }, { "model_id": "qwen3.6-plus", "benchmark_id": "ifbench", "score": 74.2, "reference_url": "https://qwen.ai/blog?id=qwen3.6", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "Qwen3.6 blog: qwen3.6-plus=74.2 (vs Claude Opus 4.5=58.0, Kimi-K2.5=70.2, GLM5=72.3, Qwen3.5-397B=76.5)." }, { "model_id": "qwen3.6-plus", "benchmark_id": "aa_lcr", "score": 68.3, "reference_url": "https://qwen.ai/blog?id=qwen3.6", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "Qwen3.6 blog: qwen3.6-plus=68.3 (vs Claude Opus 4.5=74.0, Kimi-K2.5=70.0, GLM5=63.3, Qwen3.5-397B=68.7)." }, { "model_id": "qwen3.6-plus", "benchmark_id": "gpqa_diamond", "score": 90.4, "reference_url": "https://z.ai/blog/glm-5.1", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "GLM-5.1 blog: qwen3.6-plus=90.4." }, { "model_id": "qwen3.6-plus", "benchmark_id": "aime_2026", "score": 95.1, "reference_url": "https://z.ai/blog/glm-5.1", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "GLM-5.1 blog: qwen3.6-plus=95.1." }, { "model_id": "kimi-k2", "benchmark_id": "gpqa_diamond", "score": 75.1, "reference_url": "https://arxiv.org/abs/2507.20534", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none (per-bench override)", "sampling": "pass@1 (avg of 4-64 trials per benchmark)", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Kimi K2 tech report (arxiv:2507.20534) Table 3 Kimi-K2-Instruct column: GPQA-Diamond Avg@8 75.1", "candidates": [ { "score": 74.2, "reference_url": "https://huggingface.co/moonshotai/Kimi-K2-Thinking", "source_type": "model_card", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "1.0" }, "notes": "Kimi K2-Thinking model card: kimi-k2=74.2." }, { "score": 77, "reference_url": "https://huggingface.co/MiniMaxAI/MiniMax-M2", "source_type": "model_card", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "notes": "MiniMax M2 model card: kimi-k2=77." }, { "score": 75.1, "reference_url": "https://llm-stats.com/benchmarks/gpqa", "source_type": "third_party_aggregator", "reported_setting": { "variant": "instruct (BP kimi-k2 is base)" }, "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=Kimi K2 Instruct, slug=kimi-k2-instruct, provider=Moonshot AI" }, { "score": 75.1, "reference_url": "https://llm-stats.com/benchmarks/gpqa", "source_type": "third_party_aggregator", "reported_setting": { "variant": "instruct", "snapshot": "0905" }, "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=Kimi K2-Instruct-0905, slug=kimi-k2-instruct-0905, provider=Moonshot AI" }, { "score": 43.43, "reference_url": "https://research.nvidia.com/labs/nemotron/files/NVIDIA-Nemotron-3-Ultra-Technical-Report.pdf", "source_type": "tech_report", "reported_setting": { "version_metric": "5-shot CoT exact match", "tools": "none", "harness": "NeMo Evaluator", "sampling": "pass@1", "checkpoint": "base", "mode": "pretrained" }, "notes": "NVIDIA Nemotron 3 Ultra technical report, Base Table 2: Kimi K2 Base: gpqa_diamond=43.43. Base checkpoint score mapped to the canonical post-trained model row as checkpoint=base." }, { "score": 48.1, "reference_url": "https://huggingface.co/XiaomiMiMo/MiMo-V2.5-Pro/resolve/21d1ecfecd7bd70f31be25ca49d7edd21f003659/README.md", "source_type": "official_model_card_base_checkpoint", "reported_setting": { "mode": "base pretrained checkpoint", "effort": "not applicable", "tools": "none", "sampling": "5-shot; single completion per prompt", "judge": "benchmark-native rule-based evaluator", "harness": "Xiaomi MiMo base-checkpoint evaluation", "prompt_style": "5-shot few-shot continuation", "temperature": "source does not state", "context": "source does not state", "input_modalities": "text", "trials": "one completion per item", "checkpoint": "base" }, "notes": "Base checkpoint observation attached to the canonical post-trained model ID. Base result is attached to the canonical post-trained model ID. The Xiaomi table does not state this comparator Base checkpoint's context length, and no comparator Base config is locked; no Pro context value or conflict is copied onto this observation." } ] }, { "model_id": "kimi-k2", "benchmark_id": "swe_bench_verified", "score": 65.8, "reference_url": "https://arxiv.org/abs/2507.20534", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none (per-bench override)", "sampling": "pass@1 (avg of 4-64 trials per benchmark)", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Kimi K2 tech report (arxiv:2507.20534) Table 3 Kimi-K2-Instruct column: SWE-bench Verified Agentic-Single-Attempt 65.8", "candidates": [ { "score": 69.2, "reference_url": "https://huggingface.co/MiniMaxAI/MiniMax-M2", "source_type": "model_card", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "agentic (bash + edit)", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "notes": "MiniMax M2 model card: kimi-k2=69.2." }, { "score": 65.8, "reference_url": "https://llm-stats.com/benchmarks/swe-bench-verified", "source_type": "third_party_aggregator", "reported_setting": { "variant": "instruct", "snapshot": "0905" }, "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=Kimi K2-Instruct-0905, slug=kimi-k2-instruct-0905, provider=Moonshot AI" } ] }, { "model_id": "kimi-k2", "benchmark_id": "livecodebench", "score": 53.7, "reference_url": "https://arxiv.org/abs/2507.20534", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none (per-bench override)", "sampling": "pass@1 (avg of 4-64 trials per benchmark)", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Kimi K2 tech report (arxiv:2507.20534) Table 3 Kimi-K2-Instruct column: LiveCodeBench v6 Pass@1 53.7", "candidates": [ { "score": 61, "reference_url": "https://huggingface.co/MiniMaxAI/MiniMax-M2", "source_type": "model_card", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "notes": "MiniMax M2 model card: kimi-k2=61." }, { "score": 53.7, "reference_url": "https://llm-stats.com/benchmarks/livecodebench", "source_type": "third_party_aggregator", "reported_setting": { "variant": "instruct", "snapshot": "0905" }, "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=Kimi K2-Instruct-0905, slug=kimi-k2-instruct-0905, provider=Moonshot AI" } ] }, { "model_id": "kimi-k2", "benchmark_id": "math_500", "score": 97.4, "reference_url": "https://arxiv.org/abs/2507.20534", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none (per-bench override)", "sampling": "pass@1 (avg of 4-64 trials per benchmark)", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Kimi K2 tech report (arxiv:2507.20534) Table 3 Kimi-K2-Instruct column: MATH-500 97.4", "candidates": [ { "score": 97.4, "reference_url": "https://llm-stats.com/benchmarks/math-500", "source_type": "third_party_aggregator", "reported_setting": { "variant": "instruct", "snapshot": "0905" }, "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=Kimi K2-Instruct-0905, slug=kimi-k2-instruct-0905, provider=Moonshot AI" }, { "score": 97.4, "reference_url": "https://llm-stats.com/benchmarks/math-500", "source_type": "third_party_aggregator", "reported_setting": { "variant": "instruct (BP kimi-k2 is base)" }, "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=Kimi K2 Instruct, slug=kimi-k2-instruct, provider=Moonshot AI" } ] }, { "model_id": "kimi-k2", "benchmark_id": "aime_2025", "score": 49.5, "reference_url": "https://arxiv.org/abs/2507.20534", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none (per-bench override)", "sampling": "pass@1 (avg of 4-64 trials per benchmark)", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Kimi K2 tech report (arxiv:2507.20534) Table 3 Kimi-K2-Instruct column: AIME 2025 Avg@64 49.5", "candidates": [ { "score": 51.0, "reference_url": "https://huggingface.co/moonshotai/Kimi-K2-Thinking", "source_type": "model_card", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "1.0" }, "notes": "Kimi K2-Thinking model card: kimi-k2=51.0." }, { "score": 57, "reference_url": "https://huggingface.co/MiniMaxAI/MiniMax-M2", "source_type": "model_card", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "notes": "MiniMax M2 model card: kimi-k2=57." }, { "score": 49.5, "reference_url": "https://llm-stats.com/benchmarks/aime-2025", "source_type": "third_party_aggregator", "reported_setting": { "variant": "instruct", "snapshot": "0905" }, "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=Kimi K2-Instruct-0905, slug=kimi-k2-instruct-0905, provider=Moonshot AI" }, { "score": 49.5, "reference_url": "https://llm-stats.com/benchmarks/aime-2025", "source_type": "third_party_aggregator", "reported_setting": { "variant": "instruct (BP kimi-k2 is base)" }, "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=Kimi K2 Instruct, slug=kimi-k2-instruct, provider=Moonshot AI" } ] }, { "model_id": "kimi-k2", "benchmark_id": "aime_2024", "score": 69.6, "reference_url": "https://arxiv.org/abs/2507.20534", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none (per-bench override)", "sampling": "pass@1 (avg of 4-64 trials per benchmark)", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Kimi K2 tech report (arxiv:2507.20534) Table 3 Kimi-K2-Instruct column: AIME 2024 Avg@64 69.6", "candidates": [ { "score": 69.6, "reference_url": "https://llm-stats.com/benchmarks/aime-2024", "source_type": "third_party_aggregator", "reported_setting": { "variant": "instruct (BP kimi-k2 is base)" }, "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=Kimi K2 Instruct, slug=kimi-k2-instruct, provider=Moonshot AI" }, { "score": 69.6, "reference_url": "https://llm-stats.com/benchmarks/aime-2024", "source_type": "third_party_aggregator", "reported_setting": { "variant": "instruct", "snapshot": "0905" }, "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=Kimi K2-Instruct-0905, slug=kimi-k2-instruct-0905, provider=Moonshot AI" } ] }, { "model_id": "kimi-k2", "benchmark_id": "swe_bench_pro", "score": 27.67, "reference_url": "https://scale.com/leaderboard/swe_bench_verified", "audit_status": "verified" }, { "model_id": "kimi-k2", "benchmark_id": "arena_hard", "score": 54.5, "reference_url": "https://arxiv.org/abs/2507.20534", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none (per-bench override)", "sampling": "pass@1 (avg of 4-64 trials per benchmark)", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Kimi K2 tech report (arxiv:2507.20534) Table 3 Kimi-K2-Instruct column: Arena Hard v2.0 Hard Prompt 54.5" }, { "model_id": "kimi-k2", "benchmark_id": "hle", "score": 4.7, "reference_url": "https://arxiv.org/abs/2507.20534", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none (per-bench override)", "sampling": "pass@1 (avg of 4-64 trials per benchmark)", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Kimi K2 tech report (arxiv:2507.20534) Table 3 Kimi-K2-Instruct column: HLE 4.7", "candidates": [ { "score": 7.9, "reference_url": "https://huggingface.co/moonshotai/Kimi-K2-Thinking", "source_type": "model_card", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "1.0" }, "notes": "Kimi K2-Thinking model card: kimi-k2=7.9." }, { "score": 4.7, "reference_url": "https://llm-stats.com/benchmarks/humanity's-last-exam", "source_type": "third_party_aggregator", "reported_setting": { "variant": "instruct", "snapshot": "0905" }, "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=Kimi K2-Instruct-0905, slug=kimi-k2-instruct-0905, provider=Moonshot AI" }, { "score": 4.7, "reference_url": "https://llm-stats.com/benchmarks/humanity's-last-exam", "source_type": "third_party_aggregator", "reported_setting": { "variant": "instruct (BP kimi-k2 is base)" }, "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=Kimi K2 Instruct, slug=kimi-k2-instruct, provider=Moonshot AI" } ] }, { "model_id": "kimi-k2", "benchmark_id": "humaneval", "score": 85.0, "reference_url": "https://github.com/MoonshotAI/Kimi-K2", "audit_status": "dropped", "notes": " | DROPPED (R5g-a-ghost-cell): not in official source", "candidates": [ { "score": 93.3, "reference_url": "https://llm-stats.com/benchmarks/humaneval", "source_type": "third_party_aggregator", "reported_setting": { "variant": "instruct (BP kimi-k2 is base)" }, "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=Kimi K2 Instruct, slug=kimi-k2-instruct, provider=Moonshot AI" }, { "score": 78.2, "reference_url": "https://research.nvidia.com/labs/nemotron/files/NVIDIA-Nemotron-3-Ultra-Technical-Report.pdf", "source_type": "tech_report", "reported_setting": { "version_metric": "EvalPlus sanitized sampled pass@1 n=32", "tools": "none", "harness": "NeMo Evaluator", "sampling": "164 problems x 32 samples", "checkpoint": "base", "mode": "pretrained" }, "notes": "NVIDIA Nemotron 3 Ultra technical report, Base Table 2: Kimi K2 Base: humaneval=78.2. Base checkpoint score mapped to the canonical post-trained model row as checkpoint=base." } ] }, { "model_id": "kimi-k2", "benchmark_id": "ifeval", "score": 89.8, "reference_url": "https://arxiv.org/abs/2507.20534", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none (per-bench override)", "sampling": "pass@1 (avg of 4-64 trials per benchmark)", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Kimi K2 tech report (arxiv:2507.20534) Table 3 Kimi-K2-Instruct column: IFEval Prompt Strict 89.8", "candidates": [ { "score": 89.8, "reference_url": "https://llm-stats.com/benchmarks/ifeval", "source_type": "third_party_aggregator", "reported_setting": { "variant": "instruct", "snapshot": "0905" }, "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=Kimi K2-Instruct-0905, slug=kimi-k2-instruct-0905, provider=Moonshot AI" }, { "score": 89.8, "reference_url": "https://llm-stats.com/benchmarks/ifeval", "source_type": "third_party_aggregator", "reported_setting": { "variant": "instruct (BP kimi-k2 is base)" }, "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=Kimi K2 Instruct, slug=kimi-k2-instruct, provider=Moonshot AI" } ] }, { "model_id": "kimi-k2", "benchmark_id": "mmlu", "score": 89.5, "reference_url": "https://arxiv.org/abs/2507.20534", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none (per-bench override)", "sampling": "pass@1 (avg of 4-64 trials per benchmark)", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Kimi K2 tech report (arxiv:2507.20534) Table 3 Kimi-K2-Instruct column: MMLU EM 89.5", "candidates": [ { "score": 89.5, "reference_url": "https://llm-stats.com/benchmarks/mmlu", "source_type": "third_party_aggregator", "reported_setting": { "variant": "instruct (BP kimi-k2 is base)" }, "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=Kimi K2 Instruct, slug=kimi-k2-instruct, provider=Moonshot AI" }, { "score": 89.5, "reference_url": "https://llm-stats.com/benchmarks/mmlu", "source_type": "third_party_aggregator", "reported_setting": { "variant": "instruct", "snapshot": "0905" }, "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=Kimi K2-Instruct-0905, slug=kimi-k2-instruct-0905, provider=Moonshot AI" }, { "score": 87.6, "reference_url": "https://research.nvidia.com/labs/nemotron/files/NVIDIA-Nemotron-3-Ultra-Technical-Report.pdf", "source_type": "tech_report", "reported_setting": { "version_metric": "5-shot accuracy", "tools": "none", "harness": "NeMo Evaluator", "sampling": "pass@1", "checkpoint": "base", "mode": "pretrained" }, "notes": "NVIDIA Nemotron 3 Ultra technical report, Base Table 2: Kimi K2 Base: mmlu=87.6. Base checkpoint score mapped to the canonical post-trained model row as checkpoint=base." }, { "score": 87.8, "reference_url": "https://huggingface.co/XiaomiMiMo/MiMo-V2.5-Pro/resolve/21d1ecfecd7bd70f31be25ca49d7edd21f003659/README.md", "source_type": "official_model_card_base_checkpoint", "reported_setting": { "mode": "base pretrained checkpoint", "effort": "not applicable", "tools": "none", "sampling": "5-shot; single completion per prompt", "judge": "benchmark-native rule-based evaluator", "harness": "Xiaomi MiMo base-checkpoint evaluation", "prompt_style": "5-shot few-shot continuation", "temperature": "source does not state", "context": "source does not state", "input_modalities": "text", "trials": "one completion per item", "checkpoint": "base" }, "notes": "Base checkpoint observation attached to the canonical post-trained model ID. Base result is attached to the canonical post-trained model ID. The Xiaomi table does not state this comparator Base checkpoint's context length, and no comparator Base config is locked; no Pro context value or conflict is copied onto this observation." } ] }, { "model_id": "kimi-k2", "benchmark_id": "mmlu_pro", "score": 81.1, "reference_url": "https://arxiv.org/abs/2507.20534", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none (per-bench override)", "sampling": "pass@1 (avg of 4-64 trials per benchmark)", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Kimi K2 tech report (arxiv:2507.20534) Table 3 Kimi-K2-Instruct column: MMLU-Pro EM 81.1 [R5d: prior unverified value 87.1 from https://huggingface.co/moonshotai/Kimi-K2.5 deleted.]", "candidates": [ { "score": 81.9, "reference_url": "https://huggingface.co/moonshotai/Kimi-K2-Thinking", "source_type": "model_card", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "1.0" }, "notes": "Kimi K2-Thinking model card: kimi-k2=81.9." }, { "score": 82, "reference_url": "https://huggingface.co/MiniMaxAI/MiniMax-M2", "source_type": "model_card", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "notes": "MiniMax M2 model card: kimi-k2=82." }, { "score": 81.1, "reference_url": "https://llm-stats.com/benchmarks/mmlu-pro", "source_type": "third_party_aggregator", "reported_setting": { "variant": "instruct", "snapshot": "0905" }, "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=Kimi K2-Instruct-0905, slug=kimi-k2-instruct-0905, provider=Moonshot AI" }, { "score": 81.1, "reference_url": "https://llm-stats.com/benchmarks/mmlu-pro", "source_type": "third_party_aggregator", "reported_setting": { "variant": "instruct (BP kimi-k2 is base)" }, "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=Kimi K2 Instruct, slug=kimi-k2-instruct, provider=Moonshot AI" }, { "score": 69.15, "reference_url": "https://research.nvidia.com/labs/nemotron/files/NVIDIA-Nemotron-3-Ultra-Technical-Report.pdf", "source_type": "tech_report", "reported_setting": { "version_metric": "5-shot CoT exact match", "tools": "none", "harness": "NeMo Evaluator", "sampling": "pass@1", "checkpoint": "base", "mode": "pretrained" }, "notes": "NVIDIA Nemotron 3 Ultra technical report, Base Table 2: Kimi K2 Base: mmlu_pro=69.15. Base checkpoint score mapped to the canonical post-trained model row as checkpoint=base." }, { "score": 69.2, "reference_url": "https://huggingface.co/XiaomiMiMo/MiMo-V2.5-Pro/resolve/21d1ecfecd7bd70f31be25ca49d7edd21f003659/README.md", "source_type": "official_model_card_base_checkpoint", "reported_setting": { "mode": "base pretrained checkpoint", "effort": "not applicable", "tools": "none", "sampling": "5-shot; single completion per prompt", "judge": "benchmark-native rule-based evaluator", "harness": "Xiaomi MiMo base-checkpoint evaluation", "prompt_style": "5-shot few-shot continuation", "temperature": "source does not state", "context": "source does not state", "input_modalities": "text", "trials": "one completion per item", "checkpoint": "base" }, "notes": "Base checkpoint observation attached to the canonical post-trained model ID. Base result is attached to the canonical post-trained model ID. The Xiaomi table does not state this comparator Base checkpoint's context length, and no comparator Base config is locked; no Pro context value or conflict is copied onto this observation." } ] }, { "model_id": "kimi-k2", "benchmark_id": "simpleqa", "score": 31.0, "reference_url": "https://arxiv.org/abs/2507.20534", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none (per-bench override)", "sampling": "pass@1 (avg of 4-64 trials per benchmark)", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Kimi K2 tech report (arxiv:2507.20534) Table 3 Kimi-K2-Instruct column: SimpleQA Correct 31.0", "candidates": [ { "score": 31.0, "reference_url": "https://llm-stats.com/benchmarks/simpleqa", "source_type": "third_party_aggregator", "reported_setting": { "variant": "instruct", "snapshot": "0905" }, "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=Kimi K2-Instruct-0905, slug=kimi-k2-instruct-0905, provider=Moonshot AI" }, { "score": 31.0, "reference_url": "https://llm-stats.com/benchmarks/simpleqa", "source_type": "third_party_aggregator", "reported_setting": { "variant": "instruct (BP kimi-k2 is base)" }, "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=Kimi K2 Instruct, slug=kimi-k2-instruct, provider=Moonshot AI" } ] }, { "model_id": "kimi-k2", "benchmark_id": "terminal_bench", "score": 44.5, "reference_url": "https://huggingface.co/MiniMaxAI/MiniMax-M2", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "agentic", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "MiniMax M2 model card: kimi-k2=44.5." }, { "model_id": "kimi-k2", "benchmark_id": "codeforces_rating", "score": 1780, "reference_url": "https://medium.com/data-science-in-your-pocket/kimi-k2-benchmarks-explained-5b25dd6d3a3e", "audit_status": "dropped", "notes": " | DROPPED (R5h-third-party-blog: medium.com)" }, { "model_id": "kimi-k2", "benchmark_id": "osworld", "score": 38, "reference_url": "https://arxiv.org/html/2507.20534v1", "audit_status": "dropped", "notes": " | DROPPED (R5g-a-ghost-cell): not in official source" }, { "model_id": "kimi-k2-thinking", "benchmark_id": "gpqa_diamond", "score": 84.5, "reference_url": "https://moonshotai.github.io/Kimi-K2/thinking.html", "reported_setting": { "mode": "thinking", "effort": "default (96k thinking-token budget for HLE/AIME/HMMT/GPQA; 128k for IMO/LCB/OJ)", "tools": "none (per-bench override)", "sampling": "pass@1 (avg@32 for AIME/HMMT no-tools)", "judge": "rule-based (HLE uses o3-mini)", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "Kimi K2 Thinking blog cross-model table: GPQA-Diamond no-tools 84.5 [R5d: prior unverified value 84.5 from https://huggingface.co/deepseek-ai/DeepSeek-V3.2/resolve/main/assets/paper.pdf deleted.]" }, { "model_id": "kimi-k2-thinking", "benchmark_id": "swe_bench_verified", "score": 71.3, "reference_url": "https://moonshotai.github.io/Kimi-K2/thinking.html", "reported_setting": { "mode": "thinking", "effort": "default (96k thinking-token budget for HLE/AIME/HMMT/GPQA; 128k for IMO/LCB/OJ)", "tools": "none (per-bench override)", "sampling": "pass@1 (avg@32 for AIME/HMMT no-tools)", "judge": "rule-based (HLE uses o3-mini)", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "Kimi K2 Thinking blog cross-model table: SWE-bench Verified w/ tools 71.3" }, { "model_id": "kimi-k2-thinking", "benchmark_id": "hle", "score": 23.9, "reference_url": "https://moonshotai.github.io/Kimi-K2/thinking.html", "reported_setting": { "mode": "thinking", "effort": "default (96k thinking-token budget for HLE/AIME/HMMT/GPQA; 128k for IMO/LCB/OJ)", "tools": "none (per-bench override)", "sampling": "pass@1 (avg@32 for AIME/HMMT no-tools)", "judge": "rule-based (HLE uses o3-mini)", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "Kimi K2 Thinking blog cross-model table: HLE Text-only no tools 23.9 (canonical=tools=none). Note: BP previously had 44.9 from venturebeat = w/ tools — promoted no-tools to primary. [R5d: prior unverified value 23.9 from https://huggingface.co/deepseek-ai/DeepSeek-V3.2/resolve/main/assets/paper.pdf deleted.]", "candidates": [ { "score": 44.9, "reference_url": "https://moonshotai.github.io/Kimi-K2/thinking.html", "source_type": "official_blog", "reported_setting": { "mode": "thinking", "effort": "default (96k thinking-token budget for HLE/AIME/HMMT/GPQA; 128k for IMO/LCB/OJ)", "tools": "web+code+browse", "sampling": "pass@1 (avg@32 for AIME/HMMT no-tools)", "judge": "rule-based (HLE uses o3-mini)", "harness": "official", "prompt_style": "default", "temperature": "default" }, "notes": "Kimi K2 Thinking blog cross-model table: HLE Text-only w/ tools 44.9 (alternative tool setting)." } ] }, { "model_id": "kimi-k2-thinking", "benchmark_id": "browsecomp", "score": 60.2, "reference_url": "https://moonshotai.github.io/Kimi-K2/thinking.html", "reported_setting": { "mode": "thinking", "effort": "default (96k thinking-token budget for HLE/AIME/HMMT/GPQA; 128k for IMO/LCB/OJ)", "tools": "none (per-bench override)", "sampling": "pass@1 (avg@32 for AIME/HMMT no-tools)", "judge": "rule-based (HLE uses o3-mini)", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "Kimi K2 Thinking blog cross-model table: BrowseComp w/ tools 60.2" }, { "model_id": "kimi-k2-thinking", "benchmark_id": "aime_2025", "score": 94.5, "reference_url": "https://moonshotai.github.io/Kimi-K2/thinking.html", "reported_setting": { "mode": "thinking", "effort": "default (96k thinking-token budget for HLE/AIME/HMMT/GPQA; 128k for IMO/LCB/OJ)", "tools": "none (per-bench override)", "sampling": "pass@1 (avg@32 for AIME/HMMT no-tools)", "judge": "rule-based (HLE uses o3-mini)", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "Kimi K2 Thinking blog cross-model table: AIME 2025 no-tools avg@32 94.5" }, { "model_id": "kimi-k2-thinking", "benchmark_id": "mmlu_pro", "score": 84.6, "reference_url": "https://moonshotai.github.io/Kimi-K2/thinking.html", "reported_setting": { "mode": "thinking", "effort": "default (96k thinking-token budget for HLE/AIME/HMMT/GPQA; 128k for IMO/LCB/OJ)", "tools": "none (per-bench override)", "sampling": "pass@1 (avg@32 for AIME/HMMT no-tools)", "judge": "rule-based (HLE uses o3-mini)", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "Kimi K2 Thinking blog cross-model table: MMLU-Pro 84.6" }, { "model_id": "kimi-k2-thinking", "benchmark_id": "livecodebench", "score": 83.1, "reference_url": "https://moonshotai.github.io/Kimi-K2/thinking.html", "reported_setting": { "mode": "thinking", "effort": "default (96k thinking-token budget for HLE/AIME/HMMT/GPQA; 128k for IMO/LCB/OJ)", "tools": "none (per-bench override)", "sampling": "pass@1 (avg@32 for AIME/HMMT no-tools)", "judge": "rule-based (HLE uses o3-mini)", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "Kimi K2 Thinking blog cross-model table: LiveCodeBench v6 no-tools 83.1 [R5d: prior unverified value 82.6 from https://huggingface.co/deepseek-ai/DeepSeek-V3.2/resolve/main/assets/paper.pdf deleted.]", "candidates": [ { "score": 82.6, "reference_url": "https://huggingface.co/deepseek-ai/DeepSeek-V3.2/resolve/main/assets/paper.pdf", "source_type": "tech_report", "reported_setting": { "mode": "thinking", "effort": "default (96k thinking-token budget for HLE/AIME/HMMT/GPQA; 128k for IMO/LCB/OJ)", "tools": "none (per-bench override)", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "1.0", "context": "128K" }, "notes": "DS V3.2 tech report Table 2: 82.6. Third-party self-test. Primary (83.1) from Kimi official." } ] }, { "model_id": "kimi-k2-thinking", "benchmark_id": "arc_agi_1", "score": 12.0, "reference_url": "https://moonshotai.github.io/Kimi-K2/thinking.html", "audit_status": "dropped", "notes": " | DROPPED (R5g-a-ghost-cell): not in official source" }, { "model_id": "kimi-k2-thinking", "benchmark_id": "math_500", "score": 97.0, "reference_url": "https://venturebeat.com/ai/moonshots-kimi-k2-thinking/", "audit_status": "dropped", "notes": " | DROPPED (R5h-third-party-blog: venturebeat/felloai)" }, { "model_id": "kimi-k2-thinking", "benchmark_id": "simpleqa", "score": 35.0, "reference_url": "https://venturebeat.com/ai/moonshots-kimi-k2-thinking/", "audit_status": "dropped", "notes": " | DROPPED (R5h-third-party-blog: venturebeat/felloai)" }, { "model_id": "kimi-k2-thinking", "benchmark_id": "terminal_bench", "score": 47.1, "reference_url": "https://moonshotai.github.io/Kimi-K2/thinking.html", "reported_setting": { "mode": "thinking", "effort": "default (96k thinking-token budget for HLE/AIME/HMMT/GPQA; 128k for IMO/LCB/OJ)", "tools": "none (per-bench override)", "sampling": "pass@1 (avg@32 for AIME/HMMT no-tools)", "judge": "rule-based (HLE uses o3-mini)", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "Kimi K2 Thinking blog cross-model table: Terminal-Bench w/ simulated tools (Terminus-2 + JSON parser) 47.1 [R5d: prior unverified value 35.7 from https://huggingface.co/deepseek-ai/DeepSeek-V3.2/resolve/main/assets/paper.pdf deleted.]", "candidates": [ { "score": 35.7, "reference_url": "https://z.ai/blog/glm-4.7", "source_type": "official_blog", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "agentic", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "notes": "GLM-4.7 blog: kimi-k2-thinking=35.7." } ] }, { "model_id": "kimi-k2-thinking", "benchmark_id": "aime_2024", "score": 94, "reference_url": "https://felloai.com/new-chinese-model-kimi-k2-thinking-ranks-1-in-multiple-benchmarks/", "audit_status": "dropped", "notes": " | DROPPED (R5h-third-party-blog: venturebeat/felloai)" }, { "model_id": "kimi-k2-thinking", "benchmark_id": "brumo_2025", "score": 93.33, "reference_url": "https://matharena.ai/?comp=brumo--brumo_2025", "audit_status": "verified", "matches_canonical": true, "reported_setting": "thinking:default", "source_type": "leaderboard" }, { "model_id": "kimi-k2-thinking", "benchmark_id": "cmimc_2025", "score": 91.88, "reference_url": "https://matharena.ai/?comp=cmimc--cmimc_2025", "audit_status": "verified", "matches_canonical": true, "reported_setting": "thinking:default", "source_type": "leaderboard" }, { "model_id": "kimi-k2-thinking", "benchmark_id": "codeforces_rating", "score": 2150, "reference_url": "https://felloai.com/new-chinese-model-kimi-k2-thinking-ranks-1-in-multiple-benchmarks/", "audit_status": "dropped", "notes": " | DROPPED (R5h-third-party-blog: felloai)" }, { "model_id": "kimi-k2-thinking", "benchmark_id": "hmmt_nov_2025", "score": 89.2, "reference_url": "https://huggingface.co/deepseek-ai/DeepSeek-V3.2/resolve/main/assets/paper.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "DeepSeek V3.2 tech report Table 2: kimi-k2-thinking=89.2." }, { "model_id": "kimi-k2-thinking", "benchmark_id": "humaneval", "score": 92, "reference_url": "https://felloai.com/new-chinese-model-kimi-k2-thinking-ranks-1-in-multiple-benchmarks/", "audit_status": "dropped", "notes": " | DROPPED (R5h-third-party-blog: felloai)" }, { "model_id": "kimi-k2-thinking", "benchmark_id": "matharena_apex_2025", "score": 0, "reference_url": "https://matharena.ai/?comp=apex--apex_2025", "audit_status": "verified", "notes": " | DROPPED (R5h-third-party-blog: placeholder)", "matches_canonical": true, "reported_setting": "thinking:default", "source_type": "leaderboard" }, { "model_id": "kimi-k2-thinking", "benchmark_id": "smt_2025", "score": 91.04, "reference_url": "https://matharena.ai/?comp=smt--smt_2025", "audit_status": "verified", "matches_canonical": true, "reported_setting": "thinking:default", "source_type": "leaderboard" }, { "model_id": "kimi-k2.5", "benchmark_id": "hle", "score": 30.1, "reference_url": "https://huggingface.co/moonshotai/Kimi-K2.5", "reported_setting": { "mode": "thinking", "effort": "default (96k for reasoning, 64k for vision)", "tools": "none (per-bench override)", "sampling": "pass@1 (avg@32 AIME/HMMT, avg@8 GPQA, avg@3 vision, avg@4 Seal/WideSearch)", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "1.0" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Kimi K2.5 model card (HF moonshotai/Kimi-K2.5) Evaluation table: HLE Full no-tools 30.1", "candidates": [ { "score": 31.5, "reference_url": "https://z.ai/blog/glm-5.1", "source_type": "official_blog", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "notes": "GLM-5.1 blog: kimi-k2.5=31.5." }, { "score": 31.5, "reference_url": "https://z.ai/blog/glm-5", "source_type": "official_blog", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "notes": "GLM-5 blog: kimi-k2.5=31.5." } ] }, { "model_id": "kimi-k2.5", "benchmark_id": "browsecomp", "score": 60.6, "reference_url": "https://huggingface.co/moonshotai/Kimi-K2.5", "reported_setting": { "mode": "thinking", "effort": "default (96k for reasoning, 64k for vision)", "tools": "none (per-bench override)", "sampling": "pass@1 (avg@32 AIME/HMMT, avg@8 GPQA, avg@3 vision, avg@4 Seal/WideSearch)", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "1.0" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Kimi K2.5 model card (HF moonshotai/Kimi-K2.5) Evaluation table: BrowseComp w/ tools 60.6", "candidates": [ { "score": 74.9, "reference_url": "https://huggingface.co/moonshotai/Kimi-K2.6", "source_type": "model_card", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "agentic", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "1.0" }, "notes": "Kimi K2.6 model card: kimi-k2.5=74.9." }, { "score": 74.9, "reference_url": "https://huggingface.co/Qwen/Qwen3.5-397B-A17B", "source_type": "model_card", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "notes": "Qwen3.5-397B-A17B HF model card cross-model table (Qwen self-test of 3rd-party model, mc=false): alternate measurement 74.9 (3rd-party Qwen self-test)." }, { "score": 74.9, "reference_url": "https://huggingface.co/mistralai/Mistral-Medium-3.5-128B/resolve/22b2b868a15677cfa6061277ed2f653d1349a9ab/images/image4.png?download=true", "source_type": "official_model_card", "reported_setting": { "mode": "thinking", "effort": "default (maximum reasoning setting)", "tools": "web search/browser environment", "sampling": "source does not state trial count", "judge": "benchmark-specified", "harness": "Mistral official BrowseComp evaluation", "prompt_style": "default", "temperature": "1.0", "context": "256K" }, "notes": "Displayed exactly: '74.9'. Research observation: medium-image4.png:6:4:score:reported. Chart footnotes: *All model evaluations were run with maximum reasoning settings. | **Sonnet 4.6 encountered higher rates of reasoning truncation due to external API restrictions than other models, which affected its performance. | *** Self-reported | **** Self-reported, Mistral is using context management and a discard-all strategy at 100k tokens. | τ³ scores as reported by Sierra for Claude Sonnet 4.5 and Qwen3.5. Others with user simulator: gpt-5.2 with reasoning_effort: low. 4 trials. Banking domain evaluated with terminal- or embedding-based agentic search retrieval, only highest score is reported." } ] }, { "model_id": "kimi-k2.5", "benchmark_id": "swe_bench_verified", "score": 76.8, "reference_url": "https://huggingface.co/moonshotai/Kimi-K2.5", "reported_setting": { "mode": "thinking", "effort": "default (96k for reasoning, 64k for vision)", "tools": "none (per-bench override)", "sampling": "pass@1 (avg@32 AIME/HMMT, avg@8 GPQA, avg@3 vision, avg@4 Seal/WideSearch)", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "1.0" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Kimi K2.5 model card (HF moonshotai/Kimi-K2.5) Evaluation table: SWE-Bench Verified 76.8", "candidates": [ { "score": 76.8, "reference_url": "https://huggingface.co/mistralai/Mistral-Medium-3.5-128B/resolve/22b2b868a15677cfa6061277ed2f653d1349a9ab/images/image4.png?download=true", "source_type": "official_model_card", "reported_setting": { "mode": "thinking", "effort": "default (maximum reasoning setting)", "tools": "agentic coding scaffold and SWE-bench verifier", "sampling": "source does not state trial count", "judge": "benchmark-specified", "harness": "Mistral self-reported SWE-bench Verified evaluation", "prompt_style": "default", "temperature": "1.0", "context": "256K" }, "notes": "Displayed exactly: '76.8'. Research observation: medium-image4.png:1:4:score:reported. Chart footnotes: *All model evaluations were run with maximum reasoning settings. | **Sonnet 4.6 encountered higher rates of reasoning truncation due to external API restrictions than other models, which affected its performance. | *** Self-reported | **** Self-reported, Mistral is using context management and a discard-all strategy at 100k tokens. | τ³ scores as reported by Sierra for Claude Sonnet 4.5 and Qwen3.5. Others with user simulator: gpt-5.2 with reasoning_effort: low. 4 trials. Banking domain evaluated with terminal- or embedding-based agentic search retrieval, only highest score is reported." } ] }, { "model_id": "kimi-k2.5", "benchmark_id": "mmmu_pro", "score": 78.5, "reference_url": "https://huggingface.co/moonshotai/Kimi-K2.5", "reported_setting": { "mode": "thinking", "effort": "default (96k for reasoning, 64k for vision)", "tools": "none (per-bench override)", "sampling": "pass@1 (avg@32 AIME/HMMT, avg@8 GPQA, avg@3 vision, avg@4 Seal/WideSearch)", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "1.0" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Kimi K2.5 model card (HF moonshotai/Kimi-K2.5) Evaluation table: MMMU-Pro 78.5" }, { "model_id": "kimi-k2.5", "benchmark_id": "video_mmmu", "score": 86.6, "reference_url": "https://huggingface.co/moonshotai/Kimi-K2.5", "reported_setting": { "mode": "thinking", "effort": "default (96k for reasoning, 64k for vision)", "tools": "none (per-bench override)", "sampling": "pass@1 (avg@32 AIME/HMMT, avg@8 GPQA, avg@3 vision, avg@4 Seal/WideSearch)", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "1.0" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Kimi K2.5 model card (HF moonshotai/Kimi-K2.5) Evaluation table: Video-MMMU 86.6" }, { "model_id": "kimi-k2.5", "benchmark_id": "aime_2025", "score": 96.1, "reference_url": "https://huggingface.co/moonshotai/Kimi-K2.5", "reported_setting": { "mode": "thinking", "effort": "default (96k for reasoning, 64k for vision)", "tools": "none (per-bench override)", "sampling": "pass@1 (avg@32 AIME/HMMT, avg@8 GPQA, avg@3 vision, avg@4 Seal/WideSearch)", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "1.0" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Kimi K2.5 model card (HF moonshotai/Kimi-K2.5) Evaluation table: AIME 2025 96.1", "candidates": [ { "score": 84.8, "reference_url": "https://huggingface.co/mistralai/Mistral-Medium-3.5-128B/resolve/22b2b868a15677cfa6061277ed2f653d1349a9ab/images/image1.png?download=true", "source_type": "official_model_card", "reported_setting": { "mode": "thinking", "effort": "default (maximum reasoning setting)", "tools": "none", "sampling": "samples=16 (reported as avg@16)", "judge": "benchmark-specified", "harness": "Mistral official maximum-reasoning comparison", "prompt_style": "default", "temperature": "1.0", "context": "256K" }, "notes": "Displayed exactly: '84.8'. Research observation: medium-image1.png:1:4:score:reported. Chart footnotes: *All model evaluations were run with maximum reasoning settings. | **Sonnet 4.6 encountered higher rates of reasoning truncation due to external API restrictions than other models, which affected its performance." } ] }, { "model_id": "kimi-k2.5", "benchmark_id": "gpqa_diamond", "score": 87.6, "reference_url": "https://huggingface.co/moonshotai/Kimi-K2.5", "reported_setting": { "mode": "thinking", "effort": "default (96k for reasoning, 64k for vision)", "tools": "none (per-bench override)", "sampling": "pass@1 (avg@32 AIME/HMMT, avg@8 GPQA, avg@3 vision, avg@4 Seal/WideSearch)", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "1.0" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Kimi K2.5 model card (HF moonshotai/Kimi-K2.5) Evaluation table: GPQA-Diamond 87.6" }, { "model_id": "kimi-k2.5", "benchmark_id": "livecodebench", "score": 85.0, "reference_url": "https://huggingface.co/moonshotai/Kimi-K2.5", "reported_setting": { "mode": "thinking", "effort": "default (96k for reasoning, 64k for vision)", "tools": "none (per-bench override)", "sampling": "pass@1 (avg@32 AIME/HMMT, avg@8 GPQA, avg@3 vision, avg@4 Seal/WideSearch)", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "1.0" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Kimi K2.5 model card (HF moonshotai/Kimi-K2.5) Evaluation table: LiveCodeBench v6 85.0" }, { "model_id": "kimi-k2.5", "benchmark_id": "mmlu", "score": 92.0, "reference_url": "https://huggingface.co/moonshotai/Kimi-K2.5", "audit_status": "dropped", "notes": " | DROPPED: BP value 92.0 from HF moonshotai/Kimi-K2.5 — but K2.5 model card only reports MMLU-Pro 87.1, NOT MMLU. Source mislabel/scrape error." }, { "model_id": "kimi-k2.5", "benchmark_id": "mmlu_pro", "score": 87.1, "reference_url": "https://huggingface.co/moonshotai/Kimi-K2.5", "reported_setting": { "mode": "thinking", "effort": "default (96k for reasoning, 64k for vision)", "tools": "none (per-bench override)", "sampling": "pass@1 (avg@32 AIME/HMMT, avg@8 GPQA, avg@3 vision, avg@4 Seal/WideSearch)", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "1.0" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Kimi K2.5 model card (HF moonshotai/Kimi-K2.5) Evaluation table: MMLU-Pro 87.1" }, { "model_id": "kimi-k2.5", "benchmark_id": "mmmu", "score": 84.3, "reference_url": "https://huggingface.co/Qwen/Qwen3.5-397B-A17B", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Qwen3.5-397B-A17B HF model card cross-model table (Qwen self-test of 3rd-party model, mc=false): mmmu 84.3 (3rd-party Qwen self-test). [R5d: prior unverified value 84.0 from https://www.kimi.com/blog/kimi-k2-5.html deleted.]" }, { "model_id": "kimi-k2.5", "benchmark_id": "terminal_bench", "score": 50.8, "reference_url": "https://huggingface.co/moonshotai/Kimi-K2.5", "reported_setting": { "mode": "thinking", "effort": "default (96k for reasoning, 64k for vision)", "tools": "none (per-bench override)", "sampling": "pass@1 (avg@32 AIME/HMMT, avg@8 GPQA, avg@3 vision, avg@4 Seal/WideSearch)", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "1.0" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Kimi K2.5 model card (HF moonshotai/Kimi-K2.5) Evaluation table: Terminal Bench 2.0 50.8", "candidates": [ { "score": 47.3, "reference_url": "https://cursor.com/resources/Composer2.pdf", "source_type": "tech_report", "reported_setting": { "effort": "source does not state", "tools": "Cursor coding agent", "sampling": "pass@1", "harness": "Cursor harness" }, "notes": "Composer 2 technical report Table 1 / Terminal-Bench 2.0 / Kimi K2.5 / value 1: For third-party public benchmarks, slash-separated values are Cursor/official-harness then self-reported. SWE prompts prepend 'please solve this github issue'; Terminal prompts add solution-format instructions." } ] }, { "model_id": "kimi-k2.5", "benchmark_id": "aime_2024", "score": 96.1, "reference_url": "https://www.kimi.com/blog/kimi-k2-5.html", "audit_status": "dropped", "rule_id": "R5g-ghost", "audit_note": "Source shows AIME 2025 = 96.1 (avg@32, 96k tokens), not AIME 2024. Wrong benchmark attribution.", "audited_by": "lychee-audit-bot", "audited_at": "2026-04-29T13:00:56Z" }, { "model_id": "kimi-k2.5", "benchmark_id": "arc_agi_1", "score": 65.3, "reference_url": "https://arcprize.org/leaderboard", "audit_status": "verified", "source_type": "leaderboard", "matches_canonical": true, "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1 (single-pass)", "judge": "rule-based", "harness": "official (arcprize.org)", "system_type": "CoT" }, "notes": "arcprize.org leaderboard audit: Kimi K2.5 on leaderboard" }, { "model_id": "kimi-k2.5", "benchmark_id": "arc_agi_2", "score": 11.8, "reference_url": "https://arcprize.org/leaderboard", "audit_status": "verified", "source_type": "leaderboard", "matches_canonical": true, "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1 (single-pass)", "judge": "rule-based", "harness": "official (arcprize.org)", "system_type": "CoT" }, "notes": "arcprize.org leaderboard audit: Kimi K2.5 on leaderboard" }, { "model_id": "kimi-k2.5", "benchmark_id": "brumo_2025", "score": 98.33, "reference_url": "https://matharena.ai/?comp=brumo--brumo_2025", "audit_status": "verified", "matches_canonical": true, "reported_setting": "thinking:default", "source_type": "leaderboard" }, { "model_id": "kimi-k2.5", "benchmark_id": "cmimc_2025", "score": 91.25, "reference_url": "https://matharena.ai/?comp=cmimc--cmimc_2025", "audit_status": "verified", "matches_canonical": true, "reported_setting": "thinking:default", "source_type": "leaderboard" }, { "model_id": "kimi-k2.5", "benchmark_id": "codeforces_rating", "score": 2350, "reference_url": "https://www.kimi.com/blog/kimi-k2-5.html", "audit_status": "dropped", "rule_id": "R5g-ghost", "audit_note": "Not in kimi.com/blog/kimi-k2-5.html benchmark table.", "audited_by": "lychee-audit-bot", "audited_at": "2026-04-29T13:00:56Z" }, { "model_id": "kimi-k2.5", "benchmark_id": "frontiermath", "score": 28, "reference_url": "https://kimi-k25.com/blog/kimi-k2-5-benchmark", "audit_status": "dropped", "rule_id": "R5h-third-party-blog", "audit_note": "kimi-k25.com is a third-party fan site, not official moonshotai. Scores unverifiable.", "audited_by": "lychee-audit-bot", "audited_at": "2026-04-29T13:01:27Z" }, { "model_id": "kimi-k2.5", "benchmark_id": "hmmt_nov_2025", "score": 91.1, "reference_url": "https://z.ai/blog/glm-5.1", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "GLM-5.1 blog: kimi-k2.5=91.1." }, { "model_id": "kimi-k2.5", "benchmark_id": "humaneval", "score": 95, "reference_url": "https://kimi-k25.com/blog/kimi-k2-5-benchmark", "audit_status": "dropped", "rule_id": "R5h-third-party-blog", "audit_note": "kimi-k25.com is a third-party fan site, not official moonshotai. Scores unverifiable.", "audited_by": "lychee-audit-bot", "audited_at": "2026-04-29T13:01:27Z" }, { "model_id": "kimi-k2.5", "benchmark_id": "ifeval", "score": 93.9, "reference_url": "https://huggingface.co/Qwen/Qwen3.5-397B-A17B", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Qwen3.5-397B-A17B HF model card cross-model table (Qwen self-test of 3rd-party model, mc=false): ifeval 93.9 (3rd-party Qwen self-test). [R5d: prior unverified value 90 from https://kimi-k25.com/blog/kimi-k2-5-benchmark deleted.]" }, { "model_id": "kimi-k2.5", "benchmark_id": "math_500", "score": 98, "reference_url": "https://www.kimi.com/blog/kimi-k2-5.html", "audit_status": "dropped", "rule_id": "R5g-ghost", "audit_note": "Not in kimi.com/blog/kimi-k2-5.html benchmark table.", "audited_by": "lychee-audit-bot", "audited_at": "2026-04-29T13:00:56Z" }, { "model_id": "kimi-k2.5", "benchmark_id": "matharena_apex_2025", "score": 8.85, "reference_url": "https://matharena.ai/?comp=apex--apex_2025", "audit_status": "verified", "matches_canonical": true, "reported_setting": "thinking:default", "source_type": "leaderboard" }, { "model_id": "kimi-k2.5", "benchmark_id": "mathvision", "score": 84.2, "reference_url": "https://huggingface.co/moonshotai/Kimi-K2.5", "reported_setting": { "mode": "thinking", "effort": "default (96k for reasoning, 64k for vision)", "tools": "none (per-bench override)", "sampling": "pass@1 (avg@32 AIME/HMMT, avg@8 GPQA, avg@3 vision, avg@4 Seal/WideSearch)", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "1.0" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Kimi K2.5 model card (HF moonshotai/Kimi-K2.5) Evaluation table: MathVision 84.2" }, { "model_id": "kimi-k2.5", "benchmark_id": "osworld", "score": 63.3, "reference_url": "https://huggingface.co/moonshotai/Kimi-K2.6", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "agentic", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "1.0" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Kimi K2.6 model card: kimi-k2.5=63.3." }, { "model_id": "kimi-k2.5", "benchmark_id": "simpleqa", "score": 45, "reference_url": "https://kimi-k25.com/blog/kimi-k2-5-benchmark", "audit_status": "dropped", "rule_id": "R5h-third-party-blog", "audit_note": "kimi-k25.com is a third-party fan site, not official moonshotai. Scores unverifiable.", "audited_by": "lychee-audit-bot", "audited_at": "2026-04-29T13:01:27Z" }, { "model_id": "kimi-k2.5", "benchmark_id": "smt_2025", "score": 90.57, "reference_url": "https://matharena.ai/?comp=smt--smt_2025", "audit_status": "verified", "matches_canonical": true, "reported_setting": "thinking:default", "source_type": "leaderboard" }, { "model_id": "kimi-k2.5", "benchmark_id": "swe_bench_pro", "score": 50.7, "reference_url": "https://huggingface.co/moonshotai/Kimi-K2.5", "reported_setting": { "mode": "thinking", "effort": "default (96k for reasoning, 64k for vision)", "tools": "none (per-bench override)", "sampling": "pass@1 (avg@32 AIME/HMMT, avg@8 GPQA, avg@3 vision, avg@4 Seal/WideSearch)", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "1.0" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Kimi K2.5 model card (HF moonshotai/Kimi-K2.5) Evaluation table: SWE-Bench Pro 50.7", "candidates": [ { "score": 53.8, "reference_url": "https://z.ai/blog/glm-5.1", "source_type": "official_blog", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "agentic (bash + edit)", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "notes": "GLM-5.1 blog: kimi-k2.5=53.8." } ] }, { "model_id": "kimi-k2.5", "benchmark_id": "swe_bench_multilingual", "score": 73.0, "reference_url": "https://huggingface.co/moonshotai/Kimi-K2.5", "reported_setting": { "mode": "thinking", "effort": "default (96k for reasoning, 64k for vision)", "tools": "none (per-bench override)", "sampling": "pass@1 (avg@32 AIME/HMMT, avg@8 GPQA, avg@3 vision, avg@4 Seal/WideSearch)", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "1.0" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Kimi K2.5 model card (HF moonshotai/Kimi-K2.5) Evaluation table: SWE-Bench Multilingual 73.0", "candidates": [ { "score": 65.1, "reference_url": "https://cursor.com/resources/Composer2.pdf", "source_type": "tech_report", "reported_setting": { "effort": "source does not state", "tools": "Cursor coding agent", "sampling": "pass@1", "harness": "Cursor harness" }, "notes": "Composer 2 technical report Table 1 / SWE-Bench Multilingual / Kimi K2.5 / value 1: For third-party public benchmarks, slash-separated values are Cursor/official-harness then self-reported. SWE prompts prepend 'please solve this github issue'; Terminal prompts add solution-format instructions." } ] }, { "model_id": "kimi-k2.5", "benchmark_id": "charxiv_reasoning", "score": 77.5, "reference_url": "https://huggingface.co/moonshotai/Kimi-K2.5", "reported_setting": { "mode": "thinking", "effort": "default (96k for reasoning, 64k for vision)", "tools": "none (per-bench override)", "sampling": "pass@1 (avg@32 AIME/HMMT, avg@8 GPQA, avg@3 vision, avg@4 Seal/WideSearch)", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "1.0" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Kimi K2.5 model card (HF moonshotai/Kimi-K2.5) Evaluation table: CharXiv RQ 77.5" }, { "model_id": "kimi-k2.5", "benchmark_id": "aa_lcr", "score": 70.0, "reference_url": "https://huggingface.co/moonshotai/Kimi-K2.5", "reported_setting": { "mode": "thinking", "effort": "default (96k for reasoning, 64k for vision)", "tools": "none (per-bench override)", "sampling": "pass@1 (avg@32 AIME/HMMT, avg@8 GPQA, avg@3 vision, avg@4 Seal/WideSearch)", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "1.0" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Kimi K2.5 model card (HF moonshotai/Kimi-K2.5) Evaluation table: AA-LCR 70.0" }, { "model_id": "kimi-k2.5", "benchmark_id": "scicode", "score": 48.7, "reference_url": "https://huggingface.co/moonshotai/Kimi-K2.5", "reported_setting": { "mode": "thinking", "effort": "default (96k for reasoning, 64k for vision)", "tools": "none (per-bench override)", "sampling": "pass@1 (avg@32 AIME/HMMT, avg@8 GPQA, avg@3 vision, avg@4 Seal/WideSearch)", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "1.0" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Kimi K2.5 model card (HF moonshotai/Kimi-K2.5) Evaluation table: SciCode 48.7" }, { "model_id": "kimi-k2.6", "benchmark_id": "hle", "score": 36.4, "reference_url": "https://huggingface.co/deepseek-ai/DeepSeek-V4-Pro", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "DeepSeek V4-Pro model card (third-party): kimi-k2.6 (per their own canonical config) = 36.4.", "candidates": [ { "score": 34.7, "reference_url": "https://huggingface.co/moonshotai/Kimi-K2.6", "source_type": "model_card", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "1.0" }, "notes": "Kimi K2.6 model card: kimi-k2.6=34.7." }, { "score": 34.7, "reference_url": "https://huggingface.co/moonshotai/Kimi-K2.6/resolve/d9cb81bc88b9bd2dc89877599f23c614ded72b9c/README.md", "source_type": "official_blog", "reported_setting": { "mode": "thinking", "effort": "max; up to 98,304 generation tokens", "tools": "none", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official Kimi K2.6 evaluation", "prompt_style": "provider official evaluation prompt", "temperature": "1.0; top_p=1.0", "context": "262,144 (256K)", "input_modalities": "text", "trials": "source-reported" }, "notes": "Semantic alternative printed inside the same physical HLE cell; it does not add a physical position. Exact provider-official score and reported setting." } ] }, { "model_id": "kimi-k2.6", "benchmark_id": "aime_2026", "score": 96.4, "reference_url": "https://huggingface.co/moonshotai/Kimi-K2.6", "reported_setting": { "mode": "thinking", "effort": "max (98304 generation tokens)", "tools": "none (per-bench override)", "sampling": "pass@1 (avg@10 coding, avg@3 vision)", "judge": "rule-based", "harness": "official (in-house SWE-agent-derived for SWE-Bench)", "prompt_style": "default", "temperature": "1.0" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Kimi K2.6 model card (HF moonshotai/Kimi-K2.6) Evaluation table: AIME 2026 96.4", "candidates": [ { "score": 96.4, "reference_url": "https://microsoft.ai/wp-content/uploads/2026/06/main_20260602_2.pdf", "source_type": "official_paper", "reported_setting": { "mode": "thinking", "effort": "max (98304 generation tokens)", "tools": "none (per-bench override)", "sampling": "pass@1 (avg@10 coding, avg@3 vision)", "judge": "rule-based", "harness": "official (in-house SWE-agent-derived for SWE-Bench)", "prompt_style": "default", "temperature": "1.0", "context": "262144 (256K)", "notes": "Per Kimi K2.6 model card (HF moonshotai/Kimi-K2.6). Thinking mode, max effort, t=1.0 top_p=1.0." }, "notes": "Displayed exactly as 96.4. Official Microsoft-reported result. Comparator value is quoted from that provider's official model card/release; use the campaign canonical setting." } ] }, { "model_id": "kimi-k2.6", "benchmark_id": "gpqa_diamond", "score": 90.5, "reference_url": "https://huggingface.co/deepseek-ai/DeepSeek-V4-Pro", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "DeepSeek V4-Pro model card (third-party): kimi-k2.6 (per their own canonical config) = 90.5.", "candidates": [ { "score": 91.0, "reference_url": "https://research.nvidia.com/labs/nemotron/files/NVIDIA-Nemotron-3-Ultra-Technical-Report.pdf", "source_type": "tech_report", "reported_setting": { "version_metric": "no tools / %", "tools": "none", "harness": "NeMo Evaluator", "sampling": "8-run average" }, "notes": "NVIDIA Nemotron 3 Ultra technical report, Final Table 10: Kimi K2.6: gpqa_diamond=91." }, { "score": 90.5, "reference_url": "https://microsoft.ai/wp-content/uploads/2026/06/main_20260602_2.pdf", "source_type": "official_paper", "reported_setting": { "mode": "thinking", "effort": "max (98304 generation tokens)", "tools": "none (per-bench override)", "sampling": "pass@1 (avg@10 coding, avg@3 vision)", "judge": "rule-based", "harness": "official (in-house SWE-agent-derived for SWE-Bench)", "prompt_style": "default", "temperature": "1.0", "context": "262144 (256K)", "notes": "Per Kimi K2.6 model card (HF moonshotai/Kimi-K2.6). Thinking mode, max effort, t=1.0 top_p=1.0." }, "notes": "Displayed exactly as 90.5. Official Microsoft-reported result. Comparator value is quoted from that provider's official model card/release; use the campaign canonical setting." } ] }, { "model_id": "kimi-k2.6", "benchmark_id": "terminal_bench", "score": 66.7, "reference_url": "https://huggingface.co/deepseek-ai/DeepSeek-V4-Pro", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "agentic", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "DeepSeek V4-Pro model card (third-party): kimi-k2.6 (per their own canonical config) = 66.7.", "candidates": [ { "score": 66.7, "reference_url": "https://microsoft.ai/wp-content/uploads/2026/06/main_20260602_2.pdf", "source_type": "official_paper", "reported_setting": { "mode": "thinking", "effort": "max (98304 generation tokens)", "tools": "none (per-bench override)", "sampling": "pass@1 (avg@10 coding, avg@3 vision)", "judge": "rule-based", "harness": "official (in-house SWE-agent-derived for SWE-Bench)", "prompt_style": "default", "temperature": "1.0", "context": "262144 (256K)", "notes": "Per Kimi K2.6 model card (HF moonshotai/Kimi-K2.6). Thinking mode, max effort, t=1.0 top_p=1.0." }, "notes": "Displayed exactly as 66.7. Official Microsoft-reported result. Comparator value is quoted from that provider's official model card/release; use the campaign canonical setting." }, { "score": 66.7, "reference_url": "https://huggingface.co/moonshotai/Kimi-K2.6/resolve/d9cb81bc88b9bd2dc89877599f23c614ded72b9c/README.md", "source_type": "official_blog", "reported_setting": { "mode": "thinking", "effort": "max; up to 98,304 generation tokens", "tools": "source-reported agentic tools", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "Terminus-2", "prompt_style": "provider official evaluation prompt", "temperature": "1.0; top_p=1.0", "context": "262,144 (256K)", "input_modalities": "text", "trials": "source-reported" }, "notes": "Exact provider-official score and reported setting." } ] }, { "model_id": "kimi-k2.6", "benchmark_id": "swe_bench_pro", "score": 58.6, "reference_url": "https://huggingface.co/deepseek-ai/DeepSeek-V4-Pro", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "agentic (bash + edit)", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "DeepSeek V4-Pro model card (third-party): kimi-k2.6 (per their own canonical config) = 58.6.", "candidates": [ { "score": 58.6, "reference_url": "https://microsoft.ai/wp-content/uploads/2026/06/main_20260602_2.pdf", "source_type": "official_paper", "reported_setting": { "mode": "thinking", "effort": "max (98304 generation tokens)", "tools": "none (per-bench override)", "sampling": "pass@1 (avg@10 coding, avg@3 vision)", "judge": "rule-based", "harness": "official (in-house SWE-agent-derived for SWE-Bench)", "prompt_style": "default", "temperature": "1.0", "context": "262144 (256K)", "notes": "Per Kimi K2.6 model card (HF moonshotai/Kimi-K2.6). Thinking mode, max effort, t=1.0 top_p=1.0." }, "notes": "Displayed exactly as 58.6. Official Microsoft-reported result. Comparator value is quoted from that provider's official model card/release; use the campaign canonical setting." }, { "score": 58.6, "reference_url": "https://huggingface.co/moonshotai/Kimi-K2.6/resolve/d9cb81bc88b9bd2dc89877599f23c614ded72b9c/README.md", "source_type": "official_blog", "reported_setting": { "mode": "thinking", "effort": "max; up to 98,304 generation tokens", "tools": "source-reported agentic tools", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "Kimi in-house SWE-agent-derived framework", "prompt_style": "provider official evaluation prompt", "temperature": "1.0; top_p=1.0", "context": "262,144 (256K)", "input_modalities": "text", "trials": "source-reported" }, "notes": "Exact provider-official score and reported setting." } ] }, { "model_id": "kimi-k2.6", "benchmark_id": "swe_bench_verified", "score": 80.2, "reference_url": "https://huggingface.co/deepseek-ai/DeepSeek-V4-Pro", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "agentic (bash + edit)", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "DeepSeek V4-Pro model card (third-party): kimi-k2.6 (per their own canonical config) = 80.2.", "candidates": [ { "score": 75.7, "reference_url": "https://research.nvidia.com/labs/nemotron/files/NVIDIA-Nemotron-3-Ultra-Technical-Report.pdf", "source_type": "tech_report", "reported_setting": { "version_metric": "Verified / % resolved", "tools": "agentic repository editing", "harness": "Harbor / OpenHands / AWS ECS", "sampling": "3-run average" }, "notes": "NVIDIA Nemotron 3 Ultra technical report, Final Table 10: Kimi K2.6: swe_bench_verified=75.7." }, { "score": 70.7, "reference_url": "https://research.nvidia.com/labs/nemotron/files/NVIDIA-Nemotron-3-Ultra-Technical-Report.pdf", "source_type": "tech_report", "reported_setting": { "version_metric": "Mini SWE Agent scaffold / %", "tools": "agent-specific software-engineering environment", "harness": "Mini SWE Agent", "sampling": "single agent result", "variant": "Mini SWE Agent", "agent_scaffold": "Mini SWE Agent" }, "notes": "NVIDIA Nemotron 3 Ultra technical report, Figure 17: Kimi K2.6: swe_bench_verified=70.7, variant=Mini SWE Agent." }, { "score": 71.1, "reference_url": "https://research.nvidia.com/labs/nemotron/files/NVIDIA-Nemotron-3-Ultra-Technical-Report.pdf", "source_type": "tech_report", "reported_setting": { "version_metric": "OpenCode scaffold / %", "tools": "agent-specific software-engineering environment", "harness": "OpenCode", "sampling": "single agent result", "variant": "OpenCode", "agent_scaffold": "OpenCode" }, "notes": "NVIDIA Nemotron 3 Ultra technical report, Figure 17: Kimi K2.6: swe_bench_verified=71.1, variant=OpenCode." }, { "score": 73.7, "reference_url": "https://research.nvidia.com/labs/nemotron/files/NVIDIA-Nemotron-3-Ultra-Technical-Report.pdf", "source_type": "tech_report", "reported_setting": { "version_metric": "Pi scaffold / %", "tools": "agent-specific software-engineering environment", "harness": "Pi", "sampling": "single agent result", "variant": "Pi", "agent_scaffold": "Pi" }, "notes": "NVIDIA Nemotron 3 Ultra technical report, Figure 17: Kimi K2.6: swe_bench_verified=73.7, variant=Pi." }, { "score": 74.2, "reference_url": "https://research.nvidia.com/labs/nemotron/files/NVIDIA-Nemotron-3-Ultra-Technical-Report.pdf", "source_type": "tech_report", "reported_setting": { "version_metric": "Claude scaffold / %", "tools": "agent-specific software-engineering environment", "harness": "Claude", "sampling": "single agent result", "variant": "Claude", "agent_scaffold": "Claude" }, "notes": "NVIDIA Nemotron 3 Ultra technical report, Figure 17: Kimi K2.6: swe_bench_verified=74.2, variant=Claude." }, { "score": 72.5, "reference_url": "https://research.nvidia.com/labs/nemotron/files/NVIDIA-Nemotron-3-Ultra-Technical-Report.pdf", "source_type": "tech_report", "reported_setting": { "version_metric": "Hermes scaffold / %", "tools": "agent-specific software-engineering environment", "harness": "Hermes", "sampling": "single agent result", "variant": "Hermes", "agent_scaffold": "Hermes" }, "notes": "NVIDIA Nemotron 3 Ultra technical report, Figure 17: Kimi K2.6: swe_bench_verified=72.5, variant=Hermes." }, { "score": 73.3, "reference_url": "https://research.nvidia.com/labs/nemotron/files/NVIDIA-Nemotron-3-Ultra-Technical-Report.pdf", "source_type": "tech_report", "reported_setting": { "version_metric": "OpenHands scaffold / %", "tools": "agent-specific software-engineering environment", "harness": "OpenHands", "sampling": "single agent result", "variant": "OpenHands", "agent_scaffold": "OpenHands" }, "notes": "NVIDIA Nemotron 3 Ultra technical report, Figure 17: Kimi K2.6: swe_bench_verified=73.3, variant=OpenHands." }, { "score": 65.8, "reference_url": "https://research.nvidia.com/labs/nemotron/files/NVIDIA-Nemotron-3-Ultra-Technical-Report.pdf", "source_type": "tech_report", "reported_setting": { "version_metric": "Codex scaffold / %", "tools": "agent-specific software-engineering environment", "harness": "Codex", "sampling": "single agent result", "variant": "Codex", "agent_scaffold": "Codex" }, "notes": "NVIDIA Nemotron 3 Ultra technical report, Figure 17: Kimi K2.6: swe_bench_verified=65.8, variant=Codex." }, { "score": 71.6, "reference_url": "https://research.nvidia.com/labs/nemotron/files/NVIDIA-Nemotron-3-Ultra-Technical-Report.pdf", "source_type": "tech_report", "reported_setting": { "version_metric": "Average scaffold / %", "tools": "agent-specific software-engineering environment", "harness": "Average", "sampling": "single agent result", "variant": "Average", "agent_scaffold": "Average" }, "notes": "NVIDIA Nemotron 3 Ultra technical report, Figure 17: Kimi K2.6: swe_bench_verified=71.6, variant=Average." }, { "score": 80.2, "reference_url": "https://microsoft.ai/wp-content/uploads/2026/06/main_20260602_2.pdf", "source_type": "official_paper", "reported_setting": { "mode": "thinking", "effort": "max (98304 generation tokens)", "tools": "none (per-bench override)", "sampling": "pass@1 (avg@10 coding, avg@3 vision)", "judge": "rule-based", "harness": "official (in-house SWE-agent-derived for SWE-Bench)", "prompt_style": "default", "temperature": "1.0", "context": "262144 (256K)", "notes": "Per Kimi K2.6 model card (HF moonshotai/Kimi-K2.6). Thinking mode, max effort, t=1.0 top_p=1.0." }, "notes": "Displayed exactly as 80.2. Official Microsoft-reported result. Comparator value is quoted from that provider's official model card/release; use the campaign canonical setting." }, { "score": 80.2, "reference_url": "https://huggingface.co/moonshotai/Kimi-K2.6/resolve/d9cb81bc88b9bd2dc89877599f23c614ded72b9c/README.md", "source_type": "official_blog", "reported_setting": { "mode": "thinking", "effort": "max; up to 98,304 generation tokens", "tools": "source-reported agentic tools", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official Kimi K2.6 evaluation", "prompt_style": "provider official evaluation prompt", "temperature": "1.0; top_p=1.0", "context": "262,144 (256K)", "input_modalities": "text", "trials": "source-reported" }, "notes": "Exact provider-official score and reported setting." } ] }, { "model_id": "kimi-k2.6", "benchmark_id": "livecodebench", "score": 89.6, "reference_url": "https://huggingface.co/deepseek-ai/DeepSeek-V4-Pro", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "DeepSeek V4-Pro model card (third-party): kimi-k2.6 (per their own canonical config) = 89.6." }, { "model_id": "kimi-k2.6", "benchmark_id": "browsecomp", "score": 83.2, "reference_url": "https://huggingface.co/deepseek-ai/DeepSeek-V4-Pro", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "agentic", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "DeepSeek V4-Pro model card (third-party): kimi-k2.6 (per their own canonical config) = 83.2.", "candidates": [ { "score": 61.3, "reference_url": "https://research.nvidia.com/labs/nemotron/files/NVIDIA-Nemotron-3-Ultra-Technical-Report.pdf", "source_type": "tech_report", "reported_setting": { "version_metric": "BrowseComp / %", "tools": "Tavily web search and terminal workspace", "harness": "NVIDIA custom BrowseComp scaffold", "sampling": "pass@1" }, "notes": "NVIDIA Nemotron 3 Ultra technical report, Final Table 10: Kimi K2.6: browsecomp=61.3." } ] }, { "model_id": "kimi-k2.6", "benchmark_id": "mmmu_pro", "score": 79.4, "reference_url": "https://huggingface.co/moonshotai/Kimi-K2.6", "reported_setting": { "mode": "thinking", "effort": "max (98304 generation tokens)", "tools": "none (per-bench override)", "sampling": "pass@1 (avg@10 coding, avg@3 vision)", "judge": "rule-based", "harness": "official (in-house SWE-agent-derived for SWE-Bench)", "prompt_style": "default", "temperature": "1.0" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Kimi K2.6 model card (HF moonshotai/Kimi-K2.6) Evaluation table: MMMU-Pro 79.4", "candidates": [ { "score": 79.4, "reference_url": "https://huggingface.co/moonshotai/Kimi-K2.6/resolve/d9cb81bc88b9bd2dc89877599f23c614ded72b9c/README.md", "source_type": "official_model_card", "reported_setting": { "mode": "thinking", "effort": "max; up to 98,304 generation tokens", "tools": "none", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official Kimi K2.6 evaluation", "prompt_style": "provider official evaluation prompt", "temperature": "1.0; top_p=1.0", "context": "262,144 (256K)", "input_modalities": "image and text", "trials": "source-reported" }, "notes": "Exact provider-official score and reported setting." } ] }, { "model_id": "kimi-k2.6", "benchmark_id": "scicode", "score": 52.2, "reference_url": "https://huggingface.co/moonshotai/Kimi-K2.6", "reported_setting": { "mode": "thinking", "effort": "max (98304 generation tokens)", "tools": "none (per-bench override)", "sampling": "pass@1 (avg@10 coding, avg@3 vision)", "judge": "rule-based", "harness": "official (in-house SWE-agent-derived for SWE-Bench)", "prompt_style": "default", "temperature": "1.0" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Kimi K2.6 model card (HF moonshotai/Kimi-K2.6) Evaluation table: SciCode 52.2", "candidates": [ { "score": 52.0, "reference_url": "https://research.nvidia.com/labs/nemotron/files/NVIDIA-Nemotron-3-Ultra-Technical-Report.pdf", "source_type": "tech_report", "reported_setting": { "version_metric": "subtask / %", "tools": "none", "harness": "NeMo Evaluator", "sampling": "8-run average" }, "notes": "NVIDIA Nemotron 3 Ultra technical report, Final Table 10: Kimi K2.6: scicode=52." } ] }, { "model_id": "kimi-k2.6", "benchmark_id": "mathvision", "score": 87.4, "reference_url": "https://huggingface.co/moonshotai/Kimi-K2.6", "reported_setting": { "mode": "thinking", "effort": "max (98304 generation tokens)", "tools": "none (per-bench override)", "sampling": "pass@1 (avg@10 coding, avg@3 vision)", "judge": "rule-based", "harness": "official (in-house SWE-agent-derived for SWE-Bench)", "prompt_style": "default", "temperature": "1.0" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Kimi K2.6 model card (HF moonshotai/Kimi-K2.6) Evaluation table: MathVision 87.4" }, { "model_id": "glm-4.6", "benchmark_id": "swe_bench_verified", "score": 68.0, "reference_url": "https://z.ai/blog/glm-4.7", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "agentic (bash + edit)", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "GLM-4.7 blog: glm-4.6=68.0." }, { "model_id": "glm-4.6", "benchmark_id": "gpqa_diamond", "score": 81.0, "reference_url": "https://z.ai/blog/glm-4.7", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "GLM-4.7 blog: glm-4.6=81.0.", "candidates": [ { "score": 78, "reference_url": "https://huggingface.co/MiniMaxAI/MiniMax-M2", "source_type": "model_card", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "notes": "MiniMax M2 model card: glm-4.6=78." } ] }, { "model_id": "glm-4.6", "benchmark_id": "hle", "score": 17.2, "reference_url": "https://z.ai/blog/glm-4.7", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "GLM-4.7 blog: glm-4.6=17.2." }, { "model_id": "glm-4.6", "benchmark_id": "aime_2025", "score": 93.9, "reference_url": "https://z.ai/blog/glm-4.7", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "GLM-4.7 blog: glm-4.6=93.9.", "candidates": [ { "score": 86, "reference_url": "https://huggingface.co/MiniMaxAI/MiniMax-M2", "source_type": "model_card", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "notes": "MiniMax M2 model card: glm-4.6=86." } ] }, { "model_id": "glm-4.6", "benchmark_id": "livecodebench", "score": 70, "reference_url": "https://huggingface.co/MiniMaxAI/MiniMax-M2", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "MiniMax M2 model card: glm-4.6=70." }, { "model_id": "glm-4.6", "benchmark_id": "mmlu_pro", "score": 83.2, "reference_url": "https://z.ai/blog/glm-4.7", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "GLM-4.7 blog: glm-4.6=83.2.", "candidates": [ { "score": 83, "reference_url": "https://huggingface.co/MiniMaxAI/MiniMax-M2", "source_type": "model_card", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "notes": "MiniMax M2 model card: glm-4.6=83." } ] }, { "model_id": "glm-4.6", "benchmark_id": "terminal_bench", "score": 24.5, "reference_url": "https://z.ai/blog/glm-4.7", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "agentic", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "GLM-4.7 blog: glm-4.6=24.5.", "candidates": [ { "score": 40.5, "reference_url": "https://huggingface.co/MiniMaxAI/MiniMax-M2", "source_type": "model_card", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "agentic", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "notes": "MiniMax M2 model card: glm-4.6=40.5." } ] }, { "model_id": "glm-4.6", "benchmark_id": "humaneval", "score": 82, "reference_url": "https://llm-stats.com/models/glm-4.7", "audit_status": "dropped", "rule_id": "R5h-third-party-blog", "notes": "DROPPED (R5h): random third-party blog/aggregator, no primary trail" }, { "model_id": "glm-4.6", "benchmark_id": "ifeval", "score": 82, "reference_url": "https://llm-stats.com/models/glm-4.7", "audit_status": "dropped", "rule_id": "R5h-third-party-blog", "notes": "DROPPED (R5h): random third-party blog/aggregator, no primary trail" }, { "model_id": "glm-4.6", "benchmark_id": "mmlu", "score": 85, "reference_url": "https://llm-stats.com/models/glm-4.7", "audit_status": "dropped", "rule_id": "R5h-third-party-blog", "notes": "DROPPED (R5h): random third-party blog/aggregator, no primary trail" }, { "model_id": "glm-4.7", "benchmark_id": "aime_2025", "score": 95.7, "reference_url": "https://z.ai/blog/glm-4.7", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "GLM-4.7 blog: glm-4.7=95.7." }, { "model_id": "glm-4.7", "benchmark_id": "gpqa_diamond", "score": 85.7, "reference_url": "https://z.ai/blog/glm-5", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "GLM-5 blog: glm-4.7=85.7." }, { "model_id": "glm-4.7", "benchmark_id": "hle", "score": 24.8, "reference_url": "https://z.ai/blog/glm-5", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "GLM-5 blog: glm-4.7=24.8." }, { "model_id": "glm-4.7", "benchmark_id": "swe_bench_verified", "score": 73.8, "reference_url": "https://z.ai/blog/glm-5", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "agentic (bash + edit)", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "GLM-5 blog: glm-4.7=73.8." }, { "model_id": "glm-4.7", "benchmark_id": "livecodebench", "score": 84.9, "reference_url": "https://vertu.com/lifestyle/glm-4-7-released/", "audit_status": "dropped", "rule_id": "R5h-third-party-blog", "notes": "DROPPED (R5h): random third-party blog/aggregator, no primary trail" }, { "model_id": "glm-4.7", "benchmark_id": "chatbot_arena_elo", "score": 1443, "reference_url": "https://lmarena.ai/leaderboard/text", "audit_status": "verified", "source_type": "leaderboard", "matches_canonical": true, "notes": "lmarena.ai text arena (fetched 2026-04-29). Arena model name: glm-4.7 (rank 48). ELO updates continuously; score reflects latest available." }, { "model_id": "glm-4.7", "benchmark_id": "browsecomp", "score": 52.0, "reference_url": "https://z.ai/blog/glm-5", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "agentic", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "GLM-5 blog: glm-4.7=52.0." }, { "model_id": "glm-4.7", "benchmark_id": "frontiermath", "score": 2.4, "reference_url": "https://epoch.ai/benchmarks/frontiermath", "audit_status": "verified", "source_type": "leaderboard", "matches_canonical": true, "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "harness": "Together API" }, "notes": "epoch.ai FrontierMath-2025-02-28-Private: zai-org/GLM-4.7 (GLM-4.7 Together) = 2.4%. Canonical thinking mode matches. BP had 20.0 (unverified; likely sourced from wrong model — canonical notes reference GLM-5 blog, but model is GLM-4.7)." }, { "model_id": "glm-4.7", "benchmark_id": "mmlu_pro", "score": 84.3, "reference_url": "https://z.ai/blog/glm-4.7", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "GLM-4.7 blog: glm-4.7=84.3." }, { "model_id": "glm-4.7", "benchmark_id": "terminal_bench", "score": 41.0, "reference_url": "https://z.ai/blog/glm-5", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "agentic", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "GLM-5 blog: glm-4.7=41.0." }, { "model_id": "glm-4.7", "benchmark_id": "livebench", "score": 57.3, "reference_url": "https://livebench.ai/", "audit_status": "verified", "source_type": "leaderboard", "matches_canonical": true, "reported_setting": { "mode": "thinking", "effort": "default", "tools": "n/a" }, "notes": "livebench.ai table_2026_01_08 (latest as of 2026-04-28): glm-4.7 avg=57.3. Canonical thinking mode matches. BP had 58.1." }, { "model_id": "glm-4.7", "benchmark_id": "simpleqa", "score": 32, "reference_url": "https://llm-stats.com/models/glm-4.7", "audit_status": "dropped", "rule_id": "R5h-third-party-blog", "notes": "DROPPED (R5h): random third-party blog/aggregator, no primary trail" }, { "model_id": "glm-5.1", "benchmark_id": "aime_2026", "score": 95.3, "reference_url": "https://z.ai/blog/glm-5.1", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "GLM-5.1 blog: glm-5.1=95.3.", "candidates": [ { "score": 95.3, "reference_url": "https://microsoft.ai/wp-content/uploads/2026/06/main_20260602_2.pdf", "source_type": "official_paper", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none (per-bench override)", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default", "context": "default", "notes": "Per GLM-5.1 blog." }, "notes": "Displayed exactly as 95.3. Official Microsoft-reported result. Comparator value is quoted from that provider's official model card/release; use the campaign canonical setting." } ] }, { "model_id": "glm-5.1", "benchmark_id": "gpqa_diamond", "score": 86.2, "reference_url": "https://z.ai/blog/glm-5.2", "reported_setting": { "mode": "thinking", "effort": "official reported", "sampling": "pass@1", "prompt_style": "default", "tools": "none", "temperature": "1.0", "top_p": "0.95", "context": "default", "max_output_tokens": "163840", "judge": "GPT-5.5 medium for AIME/HMMT/IMO; benchmark-specified otherwise", "harness": "official reasoning evaluation" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "GLM-5.2 official release source: GLM-5.1 / GPQA-Diamond = 86.2.", "candidates": [ { "score": 86.1, "reference_url": "https://research.nvidia.com/labs/nemotron/files/NVIDIA-Nemotron-3-Ultra-Technical-Report.pdf", "source_type": "tech_report", "reported_setting": { "version_metric": "no tools / %", "tools": "none", "harness": "NeMo Evaluator", "sampling": "8-run average" }, "notes": "NVIDIA Nemotron 3 Ultra technical report, Final Table 10: GLM-5.1: gpqa_diamond=86.1." }, { "score": 86.2, "reference_url": "https://microsoft.ai/wp-content/uploads/2026/06/main_20260602_2.pdf", "source_type": "official_paper", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none (per-bench override)", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default", "context": "default", "notes": "Per GLM-5.1 blog." }, "notes": "Displayed exactly as 86.2. Official Microsoft-reported result. Comparator value is quoted from that provider's official model card/release; use the campaign canonical setting." } ] }, { "model_id": "glm-5.1", "benchmark_id": "swe_bench_pro", "score": 58.4, "reference_url": "https://z.ai/blog/glm-5.2", "reported_setting": { "mode": "thinking", "effort": "official reported", "sampling": "pass@1", "prompt_style": "default", "tools": "agentic code execution", "temperature": "1.0", "top_p": "1.0", "context": "400000", "max_output_tokens": "32000", "judge": "official tests", "harness": "OpenHands with tailored instruction prompt" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "GLM-5.2 official release source: GLM-5.1 / SWE-bench Pro = 58.4.", "candidates": [ { "score": 58.4, "reference_url": "https://microsoft.ai/wp-content/uploads/2026/06/main_20260602_2.pdf", "source_type": "official_paper", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none (per-bench override)", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default", "context": "default", "notes": "Per GLM-5.1 blog." }, "notes": "Displayed exactly as 58.4. Official Microsoft-reported result. Comparator value is quoted from that provider's official model card/release; use the campaign canonical setting." }, { "score": 58.4, "reference_url": "https://z.ai/blog/assets/glm-5.1-B3CLmgrT.js", "source_type": "official_blog", "reported_setting": { "mode": "thinking", "effort": "official reported setting", "tools": "source-reported agentic tools", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official Z.ai GLM-5.1 evaluation", "prompt_style": "provider official evaluation prompt", "temperature": "source/provider setting", "context": "source/provider setting", "input_modalities": "text", "trials": "source-reported" }, "notes": "Exact provider-official score and reported setting." } ] }, { "model_id": "glm-5.1", "benchmark_id": "terminal_bench", "score": 63.5, "reference_url": "https://huggingface.co/deepseek-ai/DeepSeek-V4-Pro", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "agentic", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "DeepSeek V4-Pro model card (third-party): glm-5.1 (per their own canonical config) = 63.5.", "candidates": [ { "score": 69.0, "reference_url": "https://microsoft.ai/wp-content/uploads/2026/06/main_20260602_2.pdf", "source_type": "official_paper", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none (per-bench override)", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default", "context": "default", "notes": "Per GLM-5.1 blog." }, "notes": "Displayed exactly as 69.0. Official Microsoft-reported result. Comparator value is quoted from that provider's official model card/release; use the campaign canonical setting." }, { "score": 69.0, "reference_url": "https://z.ai/blog/assets/glm-5.1-B3CLmgrT.js", "source_type": "official_blog", "reported_setting": { "mode": "thinking", "effort": "official reported setting", "tools": "source-reported agentic tools", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "Claude Code self-reported harness", "prompt_style": "provider official evaluation prompt", "temperature": "source/provider setting", "context": "source/provider setting", "input_modalities": "text", "trials": "source-reported" }, "notes": "Exact provider-official score and reported setting." } ] }, { "model_id": "glm-5.1", "benchmark_id": "browsecomp", "score": 79.3, "reference_url": "https://huggingface.co/deepseek-ai/DeepSeek-V4-Pro", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "agentic", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "DeepSeek V4-Pro model card (third-party): glm-5.1 (per their own canonical config) = 79.3.", "candidates": [ { "score": 68.0, "reference_url": "https://z.ai/blog/glm-5.1", "source_type": "official_blog", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "agentic", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "notes": "GLM-5.1 blog: glm-5.1=68.0." }, { "score": 59.4, "reference_url": "https://research.nvidia.com/labs/nemotron/files/NVIDIA-Nemotron-3-Ultra-Technical-Report.pdf", "source_type": "tech_report", "reported_setting": { "version_metric": "BrowseComp / %", "tools": "Tavily web search and terminal workspace", "harness": "NVIDIA custom BrowseComp scaffold", "sampling": "pass@1" }, "notes": "NVIDIA Nemotron 3 Ultra technical report, Final Table 10: GLM-5.1: browsecomp=59.4." }, { "score": 79.3, "reference_url": "https://huggingface.co/mistralai/Mistral-Medium-3.5-128B/resolve/22b2b868a15677cfa6061277ed2f653d1349a9ab/images/image4.png?download=true", "source_type": "official_model_card", "reported_setting": { "mode": "thinking", "effort": "default (maximum reasoning setting)", "tools": "web search/browser environment", "sampling": "source does not state trial count", "judge": "benchmark-specified", "harness": "Mistral official BrowseComp evaluation", "prompt_style": "default", "temperature": "default", "context": "default" }, "notes": "Displayed exactly: '79.3'. Research observation: medium-image4.png:6:5:score:reported. Chart footnotes: *All model evaluations were run with maximum reasoning settings. | **Sonnet 4.6 encountered higher rates of reasoning truncation due to external API restrictions than other models, which affected its performance. | *** Self-reported | **** Self-reported, Mistral is using context management and a discard-all strategy at 100k tokens. | τ³ scores as reported by Sierra for Claude Sonnet 4.5 and Qwen3.5. Others with user simulator: gpt-5.2 with reasoning_effort: low. 4 trials. Banking domain evaluated with terminal- or embedding-based agentic search retrieval, only highest score is reported." } ] }, { "model_id": "doubao-seed-2.0-pro", "benchmark_id": "aime_2025", "score": 98.3, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card Table 3: doubao-seed-2.0-pro=98.3." }, { "model_id": "doubao-seed-2.0-pro", "benchmark_id": "gpqa_diamond", "score": 88.9, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card Table 3: doubao-seed-2.0-pro=88.9." }, { "model_id": "doubao-seed-2.0-pro", "benchmark_id": "swe_bench_verified", "score": 76.5, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "varies", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card: doubao-seed-2.0-pro=76.5." }, { "model_id": "doubao-seed-2.0-pro", "benchmark_id": "browsecomp", "score": 77.3, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "varies", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card: doubao-seed-2.0-pro=77.3." }, { "model_id": "doubao-seed-2.0-pro", "benchmark_id": "video_mmmu", "score": 86.9, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "audit_status": "verified", "rule_id": "R5h-third-party-blog", "notes": "Doubao Seed 2.0 model card: doubao-seed-2.0-pro=86.9.", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "candidates": [] }, { "model_id": "doubao-seed-2.0-pro", "benchmark_id": "mmmu", "score": 85.4, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card: doubao-seed-2.0-pro=85.4." }, { "model_id": "doubao-seed-2.0-pro", "benchmark_id": "mathvision", "score": 88.8, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card: doubao-seed-2.0-pro=88.8." }, { "model_id": "doubao-seed-2.0-pro", "benchmark_id": "codeforces_rating", "score": 3020, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card Table 3: doubao-seed-2.0-pro=3020." }, { "model_id": "doubao-seed-2.0-pro", "benchmark_id": "hle", "score": 32.4, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card Table 3: doubao-seed-2.0-pro=32.4." }, { "model_id": "doubao-seed-2.0-pro", "benchmark_id": "livecodebench", "score": 88.0, "reference_url": "https://www.digitalapplied.com/blog/bytedance-seed-2-doubao-ai-model-benchmarks-guide", "audit_status": "dropped", "rule_id": "R5h-third-party-blog", "notes": "DROPPED (R5h): random third-party blog/aggregator, no primary trail" }, { "model_id": "doubao-seed-2.0-pro", "benchmark_id": "mmlu_pro", "score": 87.0, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card Table 3: doubao-seed-2.0-pro=87.0." }, { "model_id": "doubao-seed-2.0-pro", "benchmark_id": "aime_2024", "score": 98.3, "reference_url": "https://www.digitalapplied.com/blog/bytedance-seed-2-doubao-ai-model-benchmarks-guide", "audit_status": "dropped", "rule_id": "R5h-third-party-blog", "notes": "DROPPED (R5h): random third-party blog/aggregator, no primary trail" }, { "model_id": "doubao-seed-2.0-pro", "benchmark_id": "humaneval", "score": 92, "reference_url": "https://www.digitalapplied.com/blog/bytedance-seed-2-doubao-ai-model-benchmarks-guide", "audit_status": "dropped", "rule_id": "R5h-third-party-blog", "notes": "DROPPED (R5h): random third-party blog/aggregator, no primary trail" }, { "model_id": "doubao-seed-2.0-pro", "benchmark_id": "ifeval", "score": 88.5, "reference_url": "https://www.digitalapplied.com/blog/bytedance-seed-2-doubao-ai-model-benchmarks-guide", "audit_status": "dropped", "rule_id": "R5h-third-party-blog", "notes": "DROPPED (R5h): random third-party blog/aggregator, no primary trail" }, { "model_id": "doubao-seed-2.0-pro", "benchmark_id": "mmlu", "score": 90, "reference_url": "https://www.digitalapplied.com/blog/bytedance-seed-2-doubao-ai-model-benchmarks-guide", "audit_status": "dropped", "rule_id": "R5h-third-party-blog", "notes": "DROPPED (R5h): random third-party blog/aggregator, no primary trail" }, { "model_id": "doubao-seed-2.0-pro", "benchmark_id": "terminal_bench", "score": 55.8, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "varies", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card: doubao-seed-2.0-pro=55.8.", "candidates": [] }, { "model_id": "mistral-small-3.1", "benchmark_id": "mmlu", "score": 80.62, "reference_url": "https://huggingface.co/mistralai/Mistral-Small-3.1-24B-Instruct-2503", "audit_status": "verified", "source_type": "model_card", "reported_setting": { "prompt_style": "5-shot CoT", "judge": "rule-based", "harness": "mistral-eval", "sampling": "pass@1", "temperature": "0.0" }, "matches_canonical": true, "notes": "HF model card Table 1: MMLU = 80.62%." }, { "model_id": "mistral-small-3.1", "benchmark_id": "humaneval", "score": 88.41, "reference_url": "https://huggingface.co/mistralai/Mistral-Small-3.1-24B-Instruct-2503", "audit_status": "verified", "source_type": "model_card", "reported_setting": { "prompt_style": "default", "judge": "rule-based", "harness": "mistral-eval", "sampling": "pass@1", "temperature": "0.0" }, "matches_canonical": true, "notes": "HF model card Table 1: HumanEval = 88.41%. BP previously had 88.99 (secondary blog)." }, { "model_id": "mistral-small-3.1", "benchmark_id": "ifeval", "score": 82.75, "reference_url": "", "audit_status": "dropped", "source_type": "", "reported_setting": {}, "matches_canonical": false, "notes": "Mistral does NOT report IFEval in HF MC or blog. BP value 82.75 came from venturebeat secondary blog." }, { "model_id": "mistral-small-3.1", "benchmark_id": "mmlu_pro", "score": 66.76, "reference_url": "https://huggingface.co/mistralai/Mistral-Small-3.1-24B-Instruct-2503", "audit_status": "verified", "source_type": "model_card", "reported_setting": { "prompt_style": "5-shot CoT", "judge": "rule-based", "harness": "mistral-eval", "sampling": "pass@1", "temperature": "0.0" }, "matches_canonical": true, "notes": "HF model card Table 1: MMLU Pro (5-shot CoT) = 66.76%." }, { "model_id": "mistral-small-3.1", "benchmark_id": "gpqa_diamond", "score": 45.96, "reference_url": "https://huggingface.co/mistralai/Mistral-Small-3.1-24B-Instruct-2503", "audit_status": "verified", "source_type": "model_card", "reported_setting": { "prompt_style": "5-shot CoT", "judge": "rule-based", "harness": "mistral-eval", "sampling": "pass@1", "temperature": "0.0" }, "matches_canonical": true, "notes": "HF model card Table 1: GPQA Diamond (5-shot CoT) = 45.96%. BP previously had 39.0 (secondary blog)." }, { "model_id": "mistral-small-3.1", "benchmark_id": "livecodebench", "score": 30.0, "reference_url": "", "audit_status": "dropped", "source_type": "", "reported_setting": {}, "matches_canonical": false, "notes": "Mistral does NOT report LiveCodeBench in HF MC or blog. BP value 30.0 came from venturebeat secondary blog." }, { "model_id": "mistral-small-3.1", "benchmark_id": "math_500", "score": 81.0, "reference_url": "", "audit_status": "dropped", "source_type": "", "reported_setting": {}, "matches_canonical": false, "notes": "Mistral does NOT report MATH-500. HF Table 1 has only 'MATH' = 69.30%. BP value 81.0 came from venturebeat secondary blog (unverifiable). Use the new 'math' cell instead." }, { "model_id": "mistral-small-3.1", "benchmark_id": "aime_2024", "score": 22, "reference_url": "", "audit_status": "dropped", "source_type": "", "reported_setting": {}, "matches_canonical": false, "notes": "Mistral does NOT report AIME in HF MC or blog. BP value 22 came from analyticsvidhya secondary blog." }, { "model_id": "mistral-small-3.1", "benchmark_id": "simpleqa", "score": 10.43, "reference_url": "https://huggingface.co/mistralai/Mistral-Small-3.1-24B-Instruct-2503", "audit_status": "verified", "source_type": "model_card", "reported_setting": { "prompt_style": "default", "judge": "rule-based", "harness": "mistral-eval", "sampling": "pass@1", "temperature": "0.0" }, "matches_canonical": true, "notes": "HF model card Table 1: SimpleQA TotalAcc = 10.43%. BP previously had 18 (secondary blog)." }, { "model_id": "mistral-small-3.1", "benchmark_id": "swe_bench_verified", "score": 28, "reference_url": "", "audit_status": "dropped", "source_type": "", "reported_setting": {}, "matches_canonical": false, "notes": "Mistral does NOT report SWE-bench Verified in HF MC or blog. BP value 28 came from analyticsvidhya secondary blog." }, { "model_id": "mistral-medium-3", "benchmark_id": "humaneval", "score": 92.1, "reference_url": "https://mistral.ai/news/mistral-medium-3", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "mistral-eval", "prompt_style": "0-shot", "temperature": "0.0" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "Mistral Medium 3 blog table: 92.1." }, { "model_id": "mistral-medium-3", "benchmark_id": "arena_hard", "score": 97.1, "reference_url": "https://mistral.ai/news/mistral-medium-3", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "mistral-eval", "prompt_style": "0-shot", "temperature": "0.0" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "Mistral Medium 3 blog table: 97.1." }, { "model_id": "mistral-medium-3", "benchmark_id": "math_500", "score": 91.0, "reference_url": "https://mistral.ai/news/mistral-medium-3", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "mistral-eval", "prompt_style": "0-shot", "temperature": "0.0" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "Mistral Medium 3 blog table: 91.0." }, { "model_id": "mistral-medium-3", "benchmark_id": "gpqa_diamond", "score": 57.1, "reference_url": "https://mistral.ai/news/mistral-medium-3", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "mistral-eval", "prompt_style": "5-shot CoT", "temperature": "0.0" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "Mistral Medium 3 blog table: 57.1." }, { "model_id": "mistral-medium-3", "benchmark_id": "ifeval", "score": 89.4, "reference_url": "https://mistral.ai/news/mistral-medium-3", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "mistral-eval", "prompt_style": "0-shot", "temperature": "0.0" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "Mistral Medium 3 blog table: 89.4." }, { "model_id": "mistral-medium-3", "benchmark_id": "mmlu", "score": 87.0, "reference_url": "https://apidog.com/blog/mistral-medium-3/", "audit_status": "dropped", "notes": "DROPPED 2026-04-29T02:20: Not in Mistral Medium 3 official blog table; previous ref was secondary/third-party." }, { "model_id": "mistral-medium-3", "benchmark_id": "livecodebench", "score": 42, "reference_url": "https://artificialanalysis.ai/models/mistral-medium-3-1", "audit_status": "dropped", "notes": "DROPPED 2026-04-29T02:20: Not in Mistral Medium 3 official blog table; previous ref was secondary/third-party." }, { "model_id": "mistral-medium-3", "benchmark_id": "mmlu_pro", "score": 77.2, "reference_url": "https://mistral.ai/news/mistral-medium-3", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "mistral-eval", "prompt_style": "5-shot CoT", "temperature": "0.0" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "Mistral Medium 3 blog table: 77.2." }, { "model_id": "mistral-medium-3", "benchmark_id": "simpleqa", "score": 25, "reference_url": "https://artificialanalysis.ai/models/mistral-medium-3-1", "audit_status": "dropped", "notes": "DROPPED 2026-04-29T02:20: Not in Mistral Medium 3 official blog table; previous ref was secondary/third-party." }, { "model_id": "mistral-medium-3", "benchmark_id": "swe_bench_verified", "score": 32, "reference_url": "https://artificialanalysis.ai/models/mistral-medium-3-1", "audit_status": "dropped", "notes": "DROPPED 2026-04-29T02:20: Not in Mistral Medium 3 official blog table; previous ref was secondary/third-party." }, { "model_id": "mistral-large-3", "benchmark_id": "mmlu_pro", "score": 80.9, "reference_url": "https://huggingface.co/mistralai/Mistral-Small-4-119B-2603/resolve/a11f36bebf709121056b1dbcc943d1c6afbe494d/images/image3.png?download=true", "reported_setting": { "mode": "non-thinking", "effort": "none", "tools": "none", "sampling": "pass@1; repeat count undisclosed", "judge": "benchmark-specified", "harness": "Mistral official model-card evaluation", "prompt_style": "official/default or source does not state", "temperature": "source does not state", "context": "source does not state" }, "matches_canonical": false, "source_type": "official_model_card", "audit_status": "verified", "notes": "Displayed exactly: '80.9'. Research observation: small-image3.png:2:4:score:reported.", "candidates": [ { "score": 67.42, "reference_url": "https://research.nvidia.com/labs/nemotron/files/NVIDIA-Nemotron-3-Ultra-Technical-Report.pdf", "source_type": "tech_report", "reported_setting": { "version_metric": "5-shot CoT exact match", "tools": "none", "harness": "NeMo Evaluator", "sampling": "pass@1", "checkpoint": "base", "mode": "pretrained" }, "notes": "NVIDIA Nemotron 3 Ultra technical report, Base Table 2: Mistral Large 3 675B Base 2512: mmlu_pro=67.42. Base checkpoint score mapped to the canonical post-trained model row as checkpoint=base." } ] }, { "model_id": "mistral-large-3", "benchmark_id": "math_500", "score": 93.6, "reference_url": "https://intuitionlabs.ai/articles/mistral-large-3-moe-llm-explained", "audit_status": "dropped", "rule_id": "R5h-third-party-blog", "notes": "DROPPED (R5h): random third-party blog/aggregator, no primary trail" }, { "model_id": "mistral-large-3", "benchmark_id": "aime_2025", "score": 85.0, "reference_url": "https://www.analyticsvidhya.com/blog/2025/12/mistral-large-3/", "audit_status": "dropped", "rule_id": "R5h-third-party-blog", "notes": "DROPPED (R5h): random third-party blog/aggregator, no primary trail" }, { "model_id": "mistral-large-3", "benchmark_id": "swe_bench_verified", "score": 68.0, "reference_url": "https://llm-stats.com/benchmarks/swe-bench-verified", "audit_status": "dropped", "rule_id": "R5h-third-party-blog", "notes": "DROPPED (R5h): random third-party blog/aggregator, no primary trail" }, { "model_id": "mistral-large-3", "benchmark_id": "gpqa_diamond", "score": 64.1, "reference_url": "https://huggingface.co/mistralai/Mistral-Small-4-119B-2603/resolve/a11f36bebf709121056b1dbcc943d1c6afbe494d/images/image3.png?download=true", "reported_setting": { "mode": "non-thinking", "effort": "none", "tools": "none", "sampling": "pass@1; repeat count undisclosed", "judge": "benchmark-specified", "harness": "Mistral official model-card evaluation", "prompt_style": "official/default or source does not state", "temperature": "source does not state", "context": "source does not state" }, "matches_canonical": false, "source_type": "official_model_card", "audit_status": "verified", "notes": "Displayed exactly: '64.1'. Research observation: small-image3.png:1:4:score:reported.", "candidates": [ { "score": 34.85, "reference_url": "https://research.nvidia.com/labs/nemotron/files/NVIDIA-Nemotron-3-Ultra-Technical-Report.pdf", "source_type": "tech_report", "reported_setting": { "version_metric": "5-shot CoT exact match", "tools": "none", "harness": "NeMo Evaluator", "sampling": "pass@1", "checkpoint": "base", "mode": "pretrained" }, "notes": "NVIDIA Nemotron 3 Ultra technical report, Base Table 2: Mistral Large 3 675B Base 2512: gpqa_diamond=34.85. Base checkpoint score mapped to the canonical post-trained model row as checkpoint=base." } ] }, { "model_id": "mistral-large-3", "benchmark_id": "humaneval", "score": 92.0, "reference_url": "https://medium.com/@leucopsis/mistral-large-3-2512-review-7788c779a5e4", "audit_status": "dropped", "rule_id": "R5h-third-party-blog", "notes": "DROPPED (R5h): random third-party blog/aggregator, no primary trail", "candidates": [ { "score": 66.71, "reference_url": "https://research.nvidia.com/labs/nemotron/files/NVIDIA-Nemotron-3-Ultra-Technical-Report.pdf", "source_type": "tech_report", "reported_setting": { "version_metric": "EvalPlus sanitized sampled pass@1 n=32", "tools": "none", "harness": "NeMo Evaluator", "sampling": "164 problems x 32 samples", "checkpoint": "base", "mode": "pretrained" }, "notes": "NVIDIA Nemotron 3 Ultra technical report, Base Table 2: Mistral Large 3 675B Base 2512: humaneval=66.71. Base checkpoint score mapped to the canonical post-trained model row as checkpoint=base." } ] }, { "model_id": "mistral-large-3", "benchmark_id": "ifeval", "score": 86.0, "reference_url": "https://intuitionlabs.ai/articles/mistral-large-3-moe-llm-explained", "audit_status": "dropped", "rule_id": "R5h-third-party-blog", "notes": "DROPPED (R5h): random third-party blog/aggregator, no primary trail" }, { "model_id": "mistral-large-3", "benchmark_id": "livecodebench", "score": 66.0, "reference_url": "https://medium.com/@leucopsis/mistral-large-3-2512-review-7788c779a5e4", "audit_status": "dropped", "rule_id": "R5h-third-party-blog", "notes": "DROPPED (R5h): random third-party blog/aggregator, no primary trail" }, { "model_id": "mistral-large-3", "benchmark_id": "mmlu", "score": 85.5, "reference_url": "https://medium.com/@leucopsis/mistral-large-3-2512-review-7788c779a5e4", "audit_status": "dropped", "rule_id": "R5h-third-party-blog", "notes": "DROPPED (R5h): random third-party blog/aggregator, no primary trail", "candidates": [ { "score": 87.35, "reference_url": "https://research.nvidia.com/labs/nemotron/files/NVIDIA-Nemotron-3-Ultra-Technical-Report.pdf", "source_type": "tech_report", "reported_setting": { "version_metric": "5-shot accuracy", "tools": "none", "harness": "NeMo Evaluator", "sampling": "pass@1", "checkpoint": "base", "mode": "pretrained" }, "notes": "NVIDIA Nemotron 3 Ultra technical report, Base Table 2: Mistral Large 3 675B Base 2512: mmlu=87.35. Base checkpoint score mapped to the canonical post-trained model row as checkpoint=base." } ] }, { "model_id": "mistral-large-3", "benchmark_id": "aime_2024", "score": 53.3, "reference_url": "https://medium.com/@leucopsis/mistral-large-3-2512-review-7788c779a5e4", "audit_status": "dropped", "rule_id": "R5h-third-party-blog", "notes": "DROPPED (R5h): random third-party blog/aggregator, no primary trail" }, { "model_id": "mistral-large-3", "benchmark_id": "chatbot_arena_elo", "score": 1418, "reference_url": "https://medium.com/@leucopsis/mistral-large-3-2512-review-7788c779a5e4", "audit_status": "dropped", "rule_id": "R5h-third-party-blog", "notes": "DROPPED (R5h): random third-party blog/aggregator, no primary trail" }, { "model_id": "mistral-large-3", "benchmark_id": "codeforces_rating", "score": 1550, "reference_url": "https://medium.com/@leucopsis/mistral-large-3-2512-review-7788c779a5e4", "audit_status": "dropped", "rule_id": "R5h-third-party-blog", "notes": "DROPPED (R5h): random third-party blog/aggregator, no primary trail" }, { "model_id": "mistral-large-3", "benchmark_id": "simpleqa", "score": 30, "reference_url": "https://medium.com/@leucopsis/mistral-large-3-2512-review-7788c779a5e4", "audit_status": "dropped", "rule_id": "R5h-third-party-blog", "notes": "DROPPED (R5h): random third-party blog/aggregator, no primary trail" }, { "model_id": "codestral-25.01", "benchmark_id": "humaneval", "score": 86.6, "reference_url": "https://blog.getbind.co/2025/01/15/mistral-codestral-25-01/", "audit_status": "dropped", "rule_id": "R5h-third-party-blog", "notes": "DROPPED (R5h): random third-party blog/aggregator, no primary trail" }, { "model_id": "codestral-25.01", "benchmark_id": "livecodebench", "score": 37.9, "reference_url": "https://blog.getbind.co/2025/01/15/mistral-codestral-25-01/", "audit_status": "dropped", "rule_id": "R5h-third-party-blog", "notes": "DROPPED (R5h): random third-party blog/aggregator, no primary trail" }, { "model_id": "codestral-25.01", "benchmark_id": "math_500", "score": 74.0, "reference_url": "https://blog.getbind.co/2025/01/15/mistral-codestral-25-01/", "audit_status": "dropped", "rule_id": "R5h-third-party-blog", "notes": "DROPPED (R5h): random third-party blog/aggregator, no primary trail" }, { "model_id": "codestral-25.01", "benchmark_id": "swe_bench_verified", "score": 35.0, "reference_url": "https://llm-stats.com/benchmarks", "audit_status": "dropped", "rule_id": "R5h-third-party-blog", "notes": "DROPPED (R5h): random third-party blog/aggregator, no primary trail" }, { "model_id": "codestral-25.01", "benchmark_id": "codeforces_rating", "score": 1480, "reference_url": "https://llm-stats.com/benchmarks", "audit_status": "dropped", "rule_id": "R5h-third-party-blog", "notes": "DROPPED (R5h): random third-party blog/aggregator, no primary trail" }, { "model_id": "devstral-2", "benchmark_id": "swe_bench_verified", "score": 72.2, "reference_url": "https://mistral.ai/news/devstral-2-vibe-cli", "audit_status": "verified", "source_type": "official_blog", "matches_canonical": null, "audited_by": "lychee-audit-bot", "audited_at": "2026-04-29T13:05:16Z", "audit_note": "Blog says: Devstral 2 (123B) achieves 72.2% on SWE-bench Verified.", "candidates": [ { "score": 72.2, "reference_url": "https://huggingface.co/mistralai/Mistral-Medium-3.5-128B/resolve/22b2b868a15677cfa6061277ed2f653d1349a9ab/images/image3.png?download=true", "source_type": "official_model_card", "reported_setting": { "mode": "non-thinking", "effort": "none", "tools": "agentic coding scaffold and SWE-bench verifier", "sampling": "source does not state trial count", "judge": "benchmark-specified", "harness": "Mistral self-reported SWE-bench Verified evaluation", "prompt_style": "official/default or source does not state", "temperature": "source does not state", "context": "source does not state" }, "notes": "Displayed exactly: '72.2'. Research observation: medium-image3.png:1:2:score:reported." } ] }, { "model_id": "devstral-2", "benchmark_id": "humaneval", "score": 88.0, "reference_url": "https://mistral.ai/news/devstral-2-vibe-cli", "audit_status": "dropped", "rule_id": "R5g-ghost", "audit_note": "Not mentioned in mistral.ai/news/devstral-2-vibe-cli. Blog only mentions SWE-bench Verified.", "audited_by": "lychee-audit-bot", "audited_at": "2026-04-29T13:05:16Z" }, { "model_id": "devstral-2", "benchmark_id": "livecodebench", "score": 60.0, "reference_url": "https://mistral.ai/news/devstral-2-vibe-cli", "audit_status": "dropped", "rule_id": "R5g-ghost", "audit_note": "Not mentioned in mistral.ai/news/devstral-2-vibe-cli. Blog only mentions SWE-bench Verified.", "audited_by": "lychee-audit-bot", "audited_at": "2026-04-29T13:05:16Z" }, { "model_id": "devstral-2", "benchmark_id": "math_500", "score": 72.0, "reference_url": "https://llm-stats.com/benchmarks", "audit_status": "dropped", "rule_id": "R5h-third-party-blog", "notes": "DROPPED (R5h): random third-party blog/aggregator, no primary trail" }, { "model_id": "devstral-2", "benchmark_id": "gpqa_diamond", "score": 45.0, "reference_url": "https://llm-stats.com/benchmarks", "audit_status": "dropped", "rule_id": "R5h-third-party-blog", "notes": "DROPPED (R5h): random third-party blog/aggregator, no primary trail" }, { "model_id": "phi-4", "benchmark_id": "mmlu", "score": 84.8, "reference_url": "https://arxiv.org/abs/2412.08905", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "simple-evals", "prompt_style": "default", "temperature": "0.5" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Phi-4 paper Table 1 (phi-4 column): mmlu=84.8 (re-sourced from https://arxiv.org/abs/2412.08905, matches BP value)" }, { "model_id": "phi-4", "benchmark_id": "gpqa_diamond", "score": 56.1, "reference_url": "https://arxiv.org/abs/2412.08905", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "simple-evals", "prompt_style": "default", "temperature": "0.5" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Phi-4 paper Table 1 (phi-4 column): gpqa_diamond=56.1 (re-sourced from https://arxiv.org/abs/2412.08905, matches BP value)" }, { "model_id": "phi-4", "benchmark_id": "humaneval", "score": 82.6, "reference_url": "https://arxiv.org/abs/2412.08905", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "simple-evals", "prompt_style": "default", "temperature": "0.5" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Phi-4 paper Table 1 (phi-4 column): humaneval=82.6 (re-sourced from https://arxiv.org/abs/2412.08905, matches BP value)" }, { "model_id": "phi-4", "benchmark_id": "arena_hard", "score": 75.4, "reference_url": "https://arxiv.org/abs/2412.08905", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "simple-evals", "prompt_style": "default", "temperature": "0.5" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Phi-4 paper Table 1 (phi-4 column): arena_hard=75.4 (re-sourced from https://arxiv.org/abs/2412.08905, matches BP value)", "candidates": [ { "score": 68.1, "reference_url": "https://arxiv.org/abs/2504.21318", "source_type": "tech_report", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "simple-evals", "prompt_style": "default", "temperature": "0.5" }, "notes": "Phi-4-reasoning paper Table 2 (phi-4 column): arena_hard=68.1 (alt measurement, mc=true)" } ] }, { "model_id": "phi-4", "benchmark_id": "math_500", "score": 80.4, "reference_url": "https://arxiv.org/abs/2412.08905", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "simple-evals", "prompt_style": "default", "temperature": "0.5" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Phi-4 paper Table 1 (phi-4 column): math_500=80.4 (re-sourced from https://arxiv.org/abs/2412.08905, matches BP value)" }, { "model_id": "phi-4", "benchmark_id": "ifeval", "score": 63.0, "reference_url": "https://arxiv.org/abs/2412.08905", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "simple-evals", "prompt_style": "default", "temperature": "0.5" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Phi-4 paper Table 1 (phi-4 column): ifeval=63.0. [R5d: prior unverified value 79.3 from https://arxiv.org/html/2412.08905v1 deleted.]", "candidates": [ { "score": 62.3, "reference_url": "https://arxiv.org/abs/2504.21318", "source_type": "tech_report", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "simple-evals", "prompt_style": "default", "temperature": "0.5" }, "notes": "Phi-4-reasoning paper Table 2 (phi-4 column): ifeval=62.3 (alt measurement, mc=true)" } ] }, { "model_id": "phi-4", "benchmark_id": "aime_2024", "score": 18, "reference_url": "https://huggingface.co/microsoft/phi-4", "audit_status": "dropped", "rule_id": "R5g-ghost", "audit_note": "phi-4 HF card (microsoft/phi-4) benchmarks: MMLU, GPQA-D, HumanEval, MGSM, DROP, SimpleQA — NO AIME benchmark in card.", "audited_by": "lychee-audit-bot", "audited_at": "2026-07-26T00:00:00Z" }, { "model_id": "phi-4", "benchmark_id": "bigcodebench", "score": 45.5, "reference_url": "https://bigcode-bench.github.io/", "audit_status": "verified", "reported_setting": "BigCodeBench Complete Instruct pass@1", "source_type": "leaderboard", "matches_canonical": true }, { "model_id": "phi-4", "benchmark_id": "livecodebench", "score": 38.5, "reference_url": "https://huggingface.co/microsoft/phi-4", "audit_status": "dropped", "rule_id": "R5g-ghost", "audit_note": "phi-4 HF card (microsoft/phi-4) benchmarks: MMLU, GPQA-D, HumanEval, MGSM — NO LiveCodeBench in card.", "audited_by": "lychee-audit-bot", "audited_at": "2026-07-26T00:00:00Z" }, { "model_id": "phi-4", "benchmark_id": "mmlu_pro", "score": 70.4, "reference_url": "https://arxiv.org/abs/2412.08905", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "simple-evals", "prompt_style": "default", "temperature": "0.5" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Phi-4 paper Table 1 (phi-4 column): mmlu_pro=70.4. [R5d: prior unverified value 68 from https://huggingface.co/microsoft/phi-4 deleted.]", "candidates": [ { "score": 71.5, "reference_url": "https://arxiv.org/abs/2504.21318", "source_type": "tech_report", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "simple-evals", "prompt_style": "default", "temperature": "0.5" }, "notes": "Phi-4-reasoning paper Table 2 (phi-4 column): mmlu_pro=71.5 (alt measurement, mc=true)" } ] }, { "model_id": "phi-4", "benchmark_id": "simpleqa", "score": 3.0, "reference_url": "https://arxiv.org/abs/2412.08905", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "simple-evals", "prompt_style": "default", "temperature": "0.5" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Phi-4 paper Table 1 (phi-4 column): simpleqa=3.0. [R5d: prior unverified value 15 from https://huggingface.co/microsoft/phi-4 deleted.]" }, { "model_id": "phi-4-reasoning", "benchmark_id": "aime_2025", "score": 63.1, "reference_url": "https://arxiv.org/abs/2504.21318", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "0.8" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Phi-4-reasoning paper Table 1 (phi-4-reasoning column): aime_2025=63.1. [R5d: prior unverified value 71.4 from https://www.analyticsvidhya.com/blog/2025/05/phi-4-reasoning-models/ deleted.]" }, { "model_id": "phi-4-reasoning", "benchmark_id": "gpqa_diamond", "score": 67.1, "reference_url": "https://arxiv.org/abs/2504.21318", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "0.8" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Phi-4-reasoning paper Table 1 (phi-4-reasoning column): gpqa_diamond=67.1. [R5d: prior unverified value 63.4 from https://ashishchadha11944.medium.com/microsofts-phi-4-reasoning-models/ deleted.]" }, { "model_id": "phi-4-reasoning", "benchmark_id": "livecodebench", "score": 53.8, "reference_url": "https://arxiv.org/abs/2504.21318", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "0.8" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Phi-4-reasoning paper Table 1 (phi-4-reasoning column): livecodebench=53.8 (re-sourced from https://arxiv.org/abs/2504.21318, matches BP value)" }, { "model_id": "phi-4-reasoning", "benchmark_id": "math_500", "score": 95.0, "reference_url": "https://www.microsoft.com/en-us/research/articles/phi-reasoning/", "audit_status": "dropped", "rule_id": "R5g-ghost", "audit_note": "Microsoft phi-reasoning blog covers AIME 25/HMMT/OmniMath/GPQA/LiveCodeBench only; MATH-500 not mentioned in article or associated arxiv 2504.21318 paper.", "audited_by": "lychee-audit-bot", "audited_at": "2026-07-26T00:00:00Z" }, { "model_id": "phi-4-reasoning", "benchmark_id": "aime_2024", "score": 74.6, "reference_url": "https://arxiv.org/abs/2504.21318", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "0.8" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Phi-4-reasoning paper Table 1 (phi-4-reasoning column): aime_2024=74.6. [R5d: prior unverified value 70.0 from https://www.microsoft.com/en-us/research/wp-content/uploads/2025/04/phi_4_reasoning.pdf deleted.]" }, { "model_id": "phi-4-reasoning", "benchmark_id": "mmlu_pro", "score": 74.3, "reference_url": "https://arxiv.org/abs/2504.21318", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "0.8" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Phi-4-reasoning paper Table 2 (phi-4-reasoning column): mmlu_pro=74.3. [R5d: prior unverified value 72.0 from https://huggingface.co/microsoft/Phi-4-reasoning deleted.]" }, { "model_id": "phi-4-reasoning", "benchmark_id": "humaneval", "score": 85.0, "reference_url": "https://huggingface.co/microsoft/Phi-4-reasoning", "audit_status": "dropped", "rule_id": "R5g-ghost", "audit_note": "Phi-4-reasoning HF card shows HumanEvalPlus=92.9; no standalone HumanEval=85.0 present. HumanEvalPlus != HumanEval.", "audited_by": "lychee-audit-bot", "audited_at": "2026-07-26T00:00:00Z" }, { "model_id": "phi-4-reasoning", "benchmark_id": "codeforces_rating", "score": 1736, "reference_url": "https://arxiv.org/abs/2504.21318", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "0.8" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Phi-4-reasoning paper Table 1 (phi-4-reasoning column): codeforces_rating=1736. [R5d: prior unverified value 1500 from https://huggingface.co/microsoft/Phi-4-reasoning deleted.]" }, { "model_id": "phi-4-reasoning", "benchmark_id": "hmmt_feb_2025", "score": 43.8, "reference_url": "https://arxiv.org/abs/2504.21318", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "0.8" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Phi-4-reasoning paper Table 1 (phi-4-reasoning column): hmmt_2025=43.8 (re-sourced from https://arxiv.org/abs/2504.21318, matches BP value)" }, { "model_id": "phi-4-reasoning", "benchmark_id": "ifeval", "score": 83.4, "reference_url": "https://arxiv.org/abs/2504.21318", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "0.8" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Phi-4-reasoning paper Table 2 (phi-4-reasoning column): ifeval=83.4 (re-sourced from https://arxiv.org/abs/2504.21318, matches BP value)" }, { "model_id": "phi-4-reasoning", "benchmark_id": "arena_hard", "score": 73.3, "reference_url": "https://arxiv.org/abs/2504.21318", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "0.8" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Phi-4-reasoning paper Table 2 (phi-4-reasoning column): arena_hard=73.3 (re-sourced from https://arxiv.org/abs/2504.21318, matches BP value)" }, { "model_id": "phi-4-reasoning-plus", "benchmark_id": "aime_2025", "score": 78.0, "reference_url": "https://arxiv.org/abs/2504.21318", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "0.8" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Phi-4-reasoning paper Table 1 (phi-4-reasoning-plus column): aime_2025=78.0. [R5d: prior unverified value 77.7 from https://huggingface.co/microsoft/Phi-4-reasoning-plus deleted.]" }, { "model_id": "phi-4-reasoning-plus", "benchmark_id": "gpqa_diamond", "score": 69.3, "reference_url": "https://arxiv.org/abs/2504.21318", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "0.8" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Phi-4-reasoning paper Table 1 (phi-4-reasoning-plus column): gpqa_diamond=69.3 (re-sourced from https://arxiv.org/abs/2504.21318, matches BP value)" }, { "model_id": "phi-4-reasoning-plus", "benchmark_id": "livecodebench", "score": 53.1, "reference_url": "https://arxiv.org/abs/2504.21318", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "0.8" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Phi-4-reasoning paper Table 1 (phi-4-reasoning-plus column): livecodebench=53.1. [R5d: prior unverified value 68.8 from https://ashishchadha11944.medium.com/microsofts-phi-4-reasoning-models/ deleted.]" }, { "model_id": "phi-4-reasoning-plus", "benchmark_id": "math_500", "score": 96.4, "reference_url": "https://huggingface.co/microsoft/Phi-4-reasoning-plus", "audit_status": "dropped", "rule_id": "R5g-ghost", "audit_note": "Not in model card. Card shows AIME, GPQA-D, OmniMath, LiveCodeBench, HumanEvalPlus, MMLUPro — no MATH-500.", "audited_by": "lychee-audit-bot", "audited_at": "2026-04-29T12:58:09Z" }, { "model_id": "phi-4-reasoning-plus", "benchmark_id": "humaneval", "score": 88.0, "reference_url": "https://huggingface.co/microsoft/Phi-4-reasoning-plus", "audit_status": "dropped", "rule_id": "R5g-ghost", "audit_note": "Card shows HumanEvalPlus=92.3 (different benchmark). The 88.0 in table is GPT-4o HumanEvalPlus score, not phi-4-reasoning-plus. Source does not confirm HumanEval=88 for this model.", "audited_by": "lychee-audit-bot", "audited_at": "2026-04-29T12:58:09Z" }, { "model_id": "phi-4-reasoning-plus", "benchmark_id": "aime_2024", "score": 81.3, "reference_url": "https://arxiv.org/abs/2504.21318", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "0.8" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Phi-4-reasoning paper Table 1 (phi-4-reasoning-plus column): aime_2024=81.3. [R5d: prior unverified value 70 from https://www.microsoft.com/en-us/research/wp-content/uploads/2025/04/phi_4_reasoning.pdf deleted.]" }, { "model_id": "phi-4-reasoning-plus", "benchmark_id": "arena_hard", "score": 79.0, "reference_url": "https://arxiv.org/abs/2504.21318", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "0.8" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Phi-4-reasoning paper Table 2 (phi-4-reasoning-plus column): arena_hard=79.0 (re-sourced from https://arxiv.org/abs/2504.21318, matches BP value)" }, { "model_id": "phi-4-reasoning-plus", "benchmark_id": "codeforces_rating", "score": 1723, "reference_url": "https://arxiv.org/abs/2504.21318", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "0.8" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Phi-4-reasoning paper Table 1 (phi-4-reasoning-plus column): codeforces_rating=1723. [R5d: prior unverified value 1500 from https://huggingface.co/microsoft/Phi-4-reasoning deleted.]" }, { "model_id": "phi-4-reasoning-plus", "benchmark_id": "hmmt_feb_2025", "score": 53.6, "reference_url": "https://arxiv.org/abs/2504.21318", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "0.8" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Phi-4-reasoning paper Table 1 (phi-4-reasoning-plus column): hmmt_2025=53.6 (re-sourced from https://arxiv.org/abs/2504.21318, matches BP value)" }, { "model_id": "phi-4-reasoning-plus", "benchmark_id": "ifeval", "score": 84.9, "reference_url": "https://arxiv.org/abs/2504.21318", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "0.8" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Phi-4-reasoning paper Table 2 (phi-4-reasoning-plus column): ifeval=84.9. [R5d: prior unverified value 82 from https://huggingface.co/microsoft/Phi-4-reasoning-plus deleted.]" }, { "model_id": "phi-4-reasoning-plus", "benchmark_id": "mmlu_pro", "score": 76.0, "reference_url": "https://arxiv.org/abs/2504.21318", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "0.8" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Phi-4-reasoning paper Table 2 (phi-4-reasoning-plus column): mmlu_pro=76.0. [R5d: prior unverified value 72 from https://huggingface.co/microsoft/Phi-4-reasoning deleted.]" }, { "model_id": "phi-4-reasoning-plus", "benchmark_id": "swe_bench_verified", "score": 32, "reference_url": "https://huggingface.co/microsoft/Phi-4-reasoning-plus", "audit_status": "dropped", "rule_id": "R5g-ghost", "audit_note": "Not mentioned in model card. No SWE-bench data shown for any model in this card.", "audited_by": "lychee-audit-bot", "audited_at": "2026-04-29T12:58:09Z" }, { "model_id": "nemotron-ultra-253b", "benchmark_id": "gpqa_diamond", "score": 76.01, "reference_url": "https://huggingface.co/nvidia/Llama-3_1-Nemotron-Ultra-253B-v1", "audit_status": "verified", "source_type": "model_card", "matches_canonical": true, "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1 (avg of up to 16 benchmark runs)", "judge": "rule-based", "harness": "NVIDIA eval pipeline (transformers/vLLM)", "prompt_style": "custom prompt templates per benchmark", "temperature": "0.6", "context": "32k" }, "notes": "HF model card \"GPQA\" Reasoning On=76.01. NOTE: model card says \"GPQA\" not \"GPQA Diamond\" — likely same but verify (see review file). Prompt template = ABCD format consistent with Diamond." }, { "model_id": "nemotron-ultra-253b", "benchmark_id": "aime_2025", "score": 72.5, "reference_url": "https://huggingface.co/nvidia/Llama-3_1-Nemotron-Ultra-253B-v1", "audit_status": "verified", "source_type": "model_card", "matches_canonical": true, "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1 (avg of up to 16 benchmark runs)", "judge": "rule-based", "harness": "NVIDIA eval pipeline (transformers/vLLM)", "prompt_style": "custom prompt templates per benchmark", "temperature": "0.6", "context": "32k" }, "notes": "HF model card \"AIME25\" Reasoning On=72.50. Clear match." }, { "model_id": "nemotron-ultra-253b", "benchmark_id": "livecodebench", "score": 66.31, "reference_url": "https://huggingface.co/nvidia/Llama-3_1-Nemotron-Ultra-253B-v1", "audit_status": "verified", "source_type": "model_card", "matches_canonical": true, "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1 (avg of up to 16 benchmark runs)", "judge": "rule-based", "harness": "NVIDIA eval pipeline (transformers/vLLM)", "prompt_style": "custom prompt templates per benchmark", "temperature": "0.6", "context": "32k" }, "notes": "HF model card \"LiveCodeBench (20240801-20250201)\" Reasoning On=66.31. Date range noted in reported_setting." }, { "model_id": "nemotron-ultra-253b", "benchmark_id": "ifeval", "score": 89.45, "reference_url": "https://huggingface.co/nvidia/Llama-3_1-Nemotron-Ultra-253B-v1", "audit_status": "verified", "source_type": "model_card", "matches_canonical": true, "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1 (avg of up to 16 benchmark runs)", "judge": "rule-based", "harness": "NVIDIA eval pipeline (transformers/vLLM)", "prompt_style": "custom prompt templates per benchmark", "temperature": "0.6", "context": "32k" }, "notes": "HF model card \"IFEval\" Strict:Instruction Reasoning On=89.45. Clear match." }, { "model_id": "nemotron-ultra-253b", "benchmark_id": "math_500", "score": 97.0, "reference_url": "https://huggingface.co/nvidia/Llama-3_1-Nemotron-Ultra-253B-v1", "audit_status": "verified", "source_type": "model_card", "matches_canonical": true, "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1 (avg of up to 16 benchmark runs)", "judge": "rule-based", "harness": "NVIDIA eval pipeline (transformers/vLLM)", "prompt_style": "custom prompt templates per benchmark", "temperature": "0.6", "context": "32k" }, "notes": "HF model card \"MATH500\" Reasoning On=97.00. Matches exactly. Previous ref was 404 URL; updated to HF model card." }, { "model_id": "nemotron-ultra-253b", "benchmark_id": "arena_hard", "score": 92.7, "reference_url": "https://medium.com/towards-agi/nvidia-llama-nemotron-outshines-llama-4/", "audit_status": "dropped", "rule_id": "R5h-third-party-blog", "notes": "DROPPED (R5h): random third-party blog/aggregator, no primary trail" }, { "model_id": "nemotron-ultra-253b", "benchmark_id": "humaneval", "score": 92.0, "reference_url": "https://developer.nvidia.com/blog/nvidia-llama-nemotron-ultra-open-model-delivers-groundbreaking-reasoning-accuracy/", "audit_status": "dropped", "rule_id": "R5g-ghost", "notes": "DROPPED (R5g-ghost): original ref URL was dead 404. Correct URL is https://developer.nvidia.com/blog/nvidia-llama-nemotron-ultra-open-model-delivers-groundbreaking-reasoning-accuracy/. That blog covers GPQA Diamond (76%), LiveCodeBench, AIME only — no humaneval data for this model." }, { "model_id": "nemotron-ultra-253b", "benchmark_id": "mmlu", "score": 87.0, "reference_url": "https://developer.nvidia.com/blog/nvidia-llama-nemotron-ultra-open-model-delivers-groundbreaking-reasoning-accuracy/", "audit_status": "dropped", "rule_id": "R5g-ghost", "notes": "DROPPED (R5g-ghost): original ref URL was dead 404. Correct URL is https://developer.nvidia.com/blog/nvidia-llama-nemotron-ultra-open-model-delivers-groundbreaking-reasoning-accuracy/. That blog covers GPQA Diamond (76%), LiveCodeBench, AIME only — no mmlu data for this model." }, { "model_id": "nemotron-ultra-253b", "benchmark_id": "mmlu_pro", "score": 76.0, "reference_url": "https://developer.nvidia.com/blog/nvidia-llama-nemotron-ultra-open-model-delivers-groundbreaking-reasoning-accuracy/", "audit_status": "dropped", "rule_id": "R5g-ghost", "notes": "DROPPED (R5g-ghost): original ref URL was dead 404. Correct URL is https://developer.nvidia.com/blog/nvidia-llama-nemotron-ultra-open-model-delivers-groundbreaking-reasoning-accuracy/. That blog covers GPQA Diamond (76%), LiveCodeBench, AIME only — no mmlu_pro data for this model." }, { "model_id": "nemotron-ultra-253b", "benchmark_id": "codeforces_rating", "score": 1750, "reference_url": "https://artificialanalysis.ai/models/llama-3-1-nemotron-ultra-253b-v1-reasoning", "audit_status": "needs_review", "notes": "Reference URL is AA model page (llama-3-1-nemotron-ultra-253b-v1-reasoning). AA does not publish codeforces_rating data; field is not present in AA model data. Cannot verify 1750 from AA source. Original source may be NVIDIA tech report." }, { "model_id": "nemotron-ultra-253b", "benchmark_id": "hle", "score": 15, "reference_url": "https://artificialanalysis.ai/models/llama-3-1-nemotron-ultra-253b-v1-reasoning", "audit_status": "needs_review", "notes": "Reference URL is AA model page (llama-3-1-nemotron-ultra-253b-v1-reasoning). Current AA model page shows hle=0.081→8.1%, but BP has 15. Major mismatch (6.9pp). Possible stale data or wrong model variant. Current AA value: 8.1%." }, { "model_id": "nemotron-ultra-253b", "benchmark_id": "simpleqa", "score": 35, "reference_url": "https://artificialanalysis.ai/models/llama-3-1-nemotron-ultra-253b-v1-reasoning", "audit_status": "needs_review", "notes": "Reference URL is AA model page (llama-3-1-nemotron-ultra-253b-v1-reasoning). AA does not publish SimpleQA scores; field is not present in AA model data for any model. Cannot verify from AA source. Original source needs identification." }, { "model_id": "nemotron-ultra-253b", "benchmark_id": "swe_bench_verified", "score": 48, "reference_url": "https://artificialanalysis.ai/models/llama-3-1-nemotron-ultra-253b-v1-reasoning", "audit_status": "needs_review", "notes": "Reference URL is AA model page (llama-3-1-nemotron-ultra-253b-v1-reasoning). AA does not publish swe_bench_verified scores; field is not present in AA model data. Cannot verify 48 from AA source." }, { "model_id": "amazon-nova-pro", "benchmark_id": "mmlu", "score": 85.9, "reference_url": "https://cdn.amazon.science/e5/e6/ccc5378c42dca467d1abe1628ec9/amazon-nova-premier-technical-report-and-model-card.pdf", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "zeroshot-cot", "temperature": 0.0 }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "" }, { "model_id": "amazon-nova-pro", "benchmark_id": "gpqa_diamond", "score": 50.0, "reference_url": "https://cdn.amazon.science/e5/e6/ccc5378c42dca467d1abe1628ec9/amazon-nova-premier-technical-report-and-model-card.pdf", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "zeroshot-cot", "temperature": 0.0 }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "" }, { "model_id": "amazon-nova-pro", "benchmark_id": "humaneval", "score": 89.0, "reference_url": "https://arxiv.org/abs/2506.12103", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "zeroshot-cot", "temperature": 0.0 }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "From Nova family tech report (arxiv 2506.12103); not in Premier announcement table." }, { "model_id": "amazon-nova-pro", "benchmark_id": "ifeval", "score": 92.1, "reference_url": "https://cdn.amazon.science/e5/e6/ccc5378c42dca467d1abe1628ec9/amazon-nova-premier-technical-report-and-model-card.pdf", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "zeroshot-cot", "temperature": 0.0 }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "" }, { "model_id": "amazon-nova-pro", "benchmark_id": "math_500", "score": 76.6, "reference_url": "https://cdn.amazon.science/e5/e6/ccc5378c42dca467d1abe1628ec9/amazon-nova-premier-technical-report-and-model-card.pdf", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "zeroshot-cot", "temperature": 0.0 }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "" }, { "model_id": "amazon-nova-premier", "benchmark_id": "mmlu", "score": 87.4, "reference_url": "https://cdn.amazon.science/e5/e6/ccc5378c42dca467d1abe1628ec9/amazon-nova-premier-technical-report-and-model-card.pdf", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "zeroshot-cot", "temperature": 0.0 }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "" }, { "model_id": "amazon-nova-premier", "benchmark_id": "math_500", "score": 82.0, "reference_url": "https://cdn.amazon.science/e5/e6/ccc5378c42dca467d1abe1628ec9/amazon-nova-premier-technical-report-and-model-card.pdf", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "zeroshot-cot", "temperature": 0.0 }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "" }, { "model_id": "amazon-nova-premier", "benchmark_id": "swe_bench_verified", "score": 42.4, "reference_url": "https://cdn.amazon.science/e5/e6/ccc5378c42dca467d1abe1628ec9/amazon-nova-premier-technical-report-and-model-card.pdf", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "zeroshot-cot", "temperature": 0.0 }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "" }, { "model_id": "amazon-nova-premier", "benchmark_id": "gpqa_diamond", "score": 57.1, "reference_url": "https://cdn.amazon.science/e5/e6/ccc5378c42dca467d1abe1628ec9/amazon-nova-premier-technical-report-and-model-card.pdf", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "zeroshot-cot", "temperature": 0.0 }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "" }, { "model_id": "amazon-nova-premier", "benchmark_id": "humaneval", "score": 80.0, "reference_url": "https://aws.amazon.com/blogs/aws/amazon-nova-premier/", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "zeroshot-cot", "temperature": 0.0 }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "dropped", "notes": "Cited in AWS announcement blog; no detailed methodology disclosed. Pending decision.", "rule_id": "R5g-ghost", "audit_note": "aws.amazon.com/blogs/aws/amazon-nova-premier/ returns HTTP 404 (permanently dead).", "audited_by": "lychee-audit-bot", "audited_at": "2026-07-26T00:00:00Z" }, { "model_id": "amazon-nova-premier", "benchmark_id": "ifeval", "score": 91.5, "reference_url": "https://cdn.amazon.science/e5/e6/ccc5378c42dca467d1abe1628ec9/amazon-nova-premier-technical-report-and-model-card.pdf", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "zeroshot-cot", "temperature": 0.0 }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "" }, { "model_id": "amazon-nova-premier", "benchmark_id": "mmmu", "score": 68.0, "reference_url": "https://cdn.amazon.science/e5/e6/ccc5378c42dca467d1abe1628ec9/amazon-nova-premier-technical-report-and-model-card.pdf", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "zeroshot-cot", "temperature": 0.0 }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "" }, { "model_id": "command-a", "benchmark_id": "mmlu", "score": 85.5, "reference_url": "https://cohere.com/research/papers/command-a-technical-report.pdf", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "Cohere internal eval", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Command A paper: mmlu=85.5 (matches BP, re-sourced)" }, { "model_id": "command-a", "benchmark_id": "humaneval", "score": 82.9, "reference_url": "https://mistral.ai/news/mistral-medium-3", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "mistral-eval", "prompt_style": "0-shot", "temperature": "0.0" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Mistral Medium 3 blog table: 82.9. Third-party self-test by Mistral (their internal eval pipeline).", "candidates": [ { "score": 76.2, "reference_url": "https://cohere.com/research/papers/command-a-technical-report.pdf", "source_type": "tech_report", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "Cohere internal eval", "prompt_style": "default", "temperature": "default" }, "notes": "Command A paper: alt measurement 76.2 (mc=true)." } ] }, { "model_id": "command-a", "benchmark_id": "mmlu_pro", "score": 68.9, "reference_url": "https://mistral.ai/news/mistral-medium-3", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "mistral-eval", "prompt_style": "5-shot CoT", "temperature": "0.0" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Mistral Medium 3 blog table: 68.9. Third-party self-test by Mistral (their internal eval pipeline).", "candidates": [ { "score": 69.6, "reference_url": "https://cohere.com/research/papers/command-a-technical-report.pdf", "source_type": "tech_report", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "Cohere internal eval", "prompt_style": "default", "temperature": "default" }, "notes": "Command A paper: alt measurement 69.6 (mc=true)." } ] }, { "model_id": "command-a", "benchmark_id": "ifeval", "score": 89.7, "reference_url": "https://mistral.ai/news/mistral-medium-3", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "mistral-eval", "prompt_style": "0-shot", "temperature": "0.0" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Mistral Medium 3 blog table: 89.7. Third-party self-test by Mistral (their internal eval pipeline).", "candidates": [ { "score": 90.9, "reference_url": "https://cohere.com/research/papers/command-a-technical-report.pdf", "source_type": "tech_report", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "Cohere internal eval", "prompt_style": "default", "temperature": "default" }, "notes": "Command A paper: alt measurement 90.9 (mc=true)." }, { "score": 90.9, "reference_url": "https://cohere.com/research/papers/command-a-technical-report.pdf", "source_type": "tech_report", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "Cohere internal eval", "prompt_style": "default", "temperature": "default" }, "notes": "Command A paper: alt measurement 90.9 (mc=true)." } ] }, { "model_id": "command-a", "benchmark_id": "arena_hard", "score": 95.1, "reference_url": "https://mistral.ai/news/mistral-medium-3", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "mistral-eval", "prompt_style": "0-shot", "temperature": "0.0" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "dropped", "notes": "Mistral Medium 3 blog table: 95.1. Third-party self-test by Mistral (their internal eval pipeline). | DROPPED: BP cell URL = Cohere paper but paper main tables do not report standard ArenaHard. Paper has mArenaHard (multilingual) and Auto Arena Hard appendix figure but not the cell value 72.0. Ghost per R5g(a)." }, { "model_id": "command-a", "benchmark_id": "gpqa_diamond", "score": 46.5, "reference_url": "https://mistral.ai/news/mistral-medium-3", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "mistral-eval", "prompt_style": "5-shot CoT", "temperature": "0.0" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "dropped", "notes": "Mistral Medium 3 blog table: 46.5. Third-party self-test by Mistral (their internal eval pipeline). | DROPPED: BP cell URL = Cohere paper but paper Table 1/3 reports GPQA full set 50.8, NOT GPQA Diamond. Wrong-benchmark scrape (ghost) per R5g(b). Use new gpqa cell at 50.8." }, { "model_id": "command-a", "benchmark_id": "livecodebench", "score": 26.9, "reference_url": "https://cohere.com/research/papers/command-a-technical-report.pdf", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "Cohere internal eval", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Command A paper: livecodebench=26.9. [R5d: prior unverified value 28.0 from https://cohere.com/research/papers/command-a-technical-report.pdf deleted.]" }, { "model_id": "command-a", "benchmark_id": "math_500", "score": 82.0, "reference_url": "https://mistral.ai/news/mistral-medium-3", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "mistral-eval", "prompt_style": "0-shot", "temperature": "0.0" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Mistral Medium 3 blog table: 82.0. Third-party self-test by Mistral (their internal eval pipeline)." }, { "model_id": "command-a", "benchmark_id": "aime_2024", "score": 30, "reference_url": "https://www.vellum.ai/llm-leaderboard", "audit_status": "dropped", "rule_id": "R5h-third-party-blog", "notes": "DROPPED (R5h): random third-party blog/aggregator, no primary trail" }, { "model_id": "command-a", "benchmark_id": "bigcodebench", "score": 45.4, "reference_url": "https://cohere.com/research/papers/command-a-technical-report.pdf", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "Cohere internal eval", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Command A paper: bigcodebench=45.4. [R5d: prior unverified value 33.8 from https://bigcode-bench.github.io/ deleted.]" }, { "model_id": "command-a", "benchmark_id": "simpleqa", "score": 32, "reference_url": "https://www.vellum.ai/llm-leaderboard", "audit_status": "dropped", "rule_id": "R5h-third-party-blog", "notes": "DROPPED (R5h): random third-party blog/aggregator, no primary trail" }, { "model_id": "command-a", "benchmark_id": "swe_bench_verified", "score": 26.8, "reference_url": "https://cohere.com/research/papers/command-a-technical-report.pdf", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "Cohere internal eval", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Command A paper: swe_bench_verified=26.8. [R5d: prior unverified value 38 from https://www.vellum.ai/llm-leaderboard deleted.]" }, { "model_id": "exaone-4.0-32b", "benchmark_id": "mmlu_pro", "score": 81.8, "reference_url": "https://arxiv.org/html/2507.11407v1", "audit_status": "verified", "matches_canonical": true, "source_type": "tech_report", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "LG AI Research internal", "prompt_style": "default", "temperature": "0.6, top_p=0.95, presence_penalty=1.5" }, "candidates": [], "notes": "Reasoning mode Table 3: MMLU-Pro=81.8 matches BP." }, { "model_id": "exaone-4.0-32b", "benchmark_id": "gpqa_diamond", "score": 75.4, "reference_url": "https://arxiv.org/html/2507.11407v1", "audit_status": "verified", "matches_canonical": true, "source_type": "tech_report", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "avg@8", "judge": "rule-based", "harness": "LG AI Research internal", "prompt_style": "default", "temperature": "0.6, top_p=0.95, presence_penalty=1.5" }, "candidates": [], "notes": "Reasoning mode Table 3: GPQA-Diamond=75.4 matches BP." }, { "model_id": "exaone-4.0-32b", "benchmark_id": "aime_2025", "score": 85.3, "reference_url": "https://arxiv.org/html/2507.11407v1", "audit_status": "verified", "matches_canonical": true, "source_type": "tech_report", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "avg@32", "judge": "rule-based", "harness": "LG AI Research internal", "prompt_style": "default", "temperature": "0.6, top_p=0.95, presence_penalty=1.5" }, "candidates": [], "notes": "Reasoning mode Table 3: AIME 2025=85.3 matches BP. (avg@32, 64K budget)" }, { "model_id": "exaone-4.0-32b", "benchmark_id": "livecodebench", "score": 72.6, "reference_url": "https://arxiv.org/html/2507.11407v1", "audit_status": "dropped", "matches_canonical": false, "source_type": "tech_report", "candidates": [], "notes": "DROPPED (R5g-b wrong-variant): Source explicitly labels this LiveCodeBench v5=72.6. BP has separate livecodebench_v5 and livecodebench_v6 IDs. Moved to livecodebench_v5." }, { "model_id": "exaone-4.0-32b", "benchmark_id": "hmmt_feb_2025", "score": 72.9, "reference_url": "https://arxiv.org/html/2507.11407v1", "audit_status": "verified", "matches_canonical": true, "source_type": "tech_report", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "avg@32", "judge": "rule-based", "harness": "LG AI Research internal", "prompt_style": "default", "temperature": "0.6, top_p=0.95, presence_penalty=1.5" }, "candidates": [], "notes": "Reasoning mode Table 3: HMMT Feb 2025=72.9 matches BP. (avg@32)" }, { "model_id": "exaone-4.0-32b", "benchmark_id": "math_500", "score": 96.4, "reference_url": "https://arxiv.org/html/2507.11407v1", "audit_status": "dropped", "matches_canonical": false, "source_type": "tech_report", "candidates": [], "notes": "DROPPED (R5g-a ghost): Source only has MATH500(ES)=95.8 (Reasoning). English MATH-500 not reported. 96.4 not found in source." }, { "model_id": "exaone-4.0-32b", "benchmark_id": "ifeval", "score": 83.7, "reference_url": "https://arxiv.org/html/2507.11407v1", "audit_status": "verified", "matches_canonical": true, "source_type": "tech_report", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1 (greedy for Non-Reasoning, sampled for Reasoning avg)", "judge": "rule-based", "harness": "LG AI Research internal", "prompt_style": "default", "temperature": "0.6, top_p=0.95, presence_penalty=1.5" }, "candidates": [], "notes": "Reasoning mode Table 3: IFEval=83.7. BP had 88.0 which is not present anywhere in source → R5g-b overwrite." }, { "model_id": "exaone-4.0-32b", "benchmark_id": "humaneval", "score": 90.2, "reference_url": "https://arxiv.org/html/2507.11407v1", "audit_status": "dropped", "matches_canonical": false, "source_type": "tech_report", "candidates": [], "notes": "DROPPED (R5g-a ghost): HumanEval not evaluated in EXAONE 4.0 tech report (2507.11407). Source has no HumanEval table." }, { "model_id": "exaone-4.0-32b", "benchmark_id": "arena_hard", "score": 80.0, "reference_url": "https://arxiv.org/html/2507.11407v1", "audit_status": "dropped", "matches_canonical": false, "source_type": "tech_report", "candidates": [], "notes": "DROPPED (R5g-a ghost): Arena-Hard not evaluated in EXAONE 4.0 tech report. Source has no Arena Hard table." }, { "model_id": "exaone-4.0-32b", "benchmark_id": "mmlu", "score": 84.0, "reference_url": "https://arxiv.org/html/2507.11407v1", "audit_status": "dropped", "matches_canonical": false, "source_type": "tech_report", "candidates": [], "notes": "DROPPED (R5g-a ghost): Source uses MMLU-Redux (92.3) and MMLU-Pro (81.8), not plain MMLU. 84.0 not found in source." }, { "model_id": "exaone-4.0-32b", "benchmark_id": "simpleqa", "score": 30.0, "reference_url": "https://arxiv.org/html/2507.11407v1", "audit_status": "dropped", "matches_canonical": false, "source_type": "tech_report", "candidates": [], "notes": "DROPPED (R5g-a ghost): SimpleQA not evaluated in EXAONE 4.0 tech report. Source has no SimpleQA table." }, { "model_id": "exaone-4.0-32b", "benchmark_id": "aime_2024", "score": 72.1, "reference_url": "https://arxiv.org/abs/2504.21318", "audit_status": "dropped", "rule_id": "R5g-ghost", "audit_note": "arxiv 2504.21318 (Phi-4-reasoning paper) Table 1 reports EXAONE-Deep-32B AIME 24=72.1, not EXAONE 4.0 32B. Wrong model variant.", "audited_by": "lychee-audit-bot", "audited_at": "2026-07-26T00:00:00Z" }, { "model_id": "exaone-4.0-32b", "benchmark_id": "codeforces_rating", "score": 1650, "reference_url": "https://llm-stats.com/benchmarks", "audit_status": "dropped", "rule_id": "R5h-third-party-blog", "notes": "DROPPED (R5h): random third-party blog/aggregator, no primary trail" }, { "model_id": "exaone-4.0-32b", "benchmark_id": "swe_bench_verified", "score": 45, "reference_url": "https://llm-stats.com/benchmarks/swe-bench-verified", "audit_status": "dropped", "rule_id": "R5h-third-party-blog", "notes": "DROPPED (R5h): random third-party blog/aggregator, no primary trail" }, { "model_id": "minimax-m2", "benchmark_id": "mmlu_pro", "score": 82.0, "reference_url": "https://huggingface.co/deepseek-ai/DeepSeek-V3.2/resolve/main/assets/paper.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "DeepSeek V3.2 tech report Table 2: minimax-m2=82.0." }, { "model_id": "minimax-m2", "benchmark_id": "gpqa_diamond", "score": 77.7, "reference_url": "https://huggingface.co/deepseek-ai/DeepSeek-V3.2/resolve/main/assets/paper.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "DeepSeek V3.2 tech report Table 2: minimax-m2=77.7.", "candidates": [ { "score": 78, "reference_url": "https://huggingface.co/MiniMaxAI/MiniMax-M2", "source_type": "model_card", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "notes": "MiniMax M2 model card: minimax-m2=78." } ] }, { "model_id": "minimax-m2", "benchmark_id": "swe_bench_verified", "score": 69.4, "reference_url": "https://huggingface.co/deepseek-ai/DeepSeek-V3.2/resolve/main/assets/paper.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "agentic (bash + edit)", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "DeepSeek V3.2 tech report Table 2: minimax-m2=69.4." }, { "model_id": "minimax-m2", "benchmark_id": "aa_intelligence_index", "score": 61, "reference_url": "https://huggingface.co/MiniMaxAI/MiniMax-M2", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "MiniMax M2 model card: minimax-m2=61." }, { "model_id": "minimax-m2", "benchmark_id": "hle", "score": 12.5, "reference_url": "https://huggingface.co/deepseek-ai/DeepSeek-V3.2/resolve/main/assets/paper.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "DeepSeek V3.2 tech report Table 2: minimax-m2=12.5." }, { "model_id": "minimax-m2", "benchmark_id": "aime_2025", "score": 78.3, "reference_url": "https://huggingface.co/deepseek-ai/DeepSeek-V3.2/resolve/main/assets/paper.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "DeepSeek V3.2 tech report Table 2: minimax-m2=78.3.", "candidates": [ { "score": 78, "reference_url": "https://huggingface.co/MiniMaxAI/MiniMax-M2", "source_type": "model_card", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "notes": "MiniMax M2 model card: minimax-m2=78." } ] }, { "model_id": "minimax-m2", "benchmark_id": "livecodebench", "score": 83.0, "reference_url": "https://huggingface.co/deepseek-ai/DeepSeek-V3.2/resolve/main/assets/paper.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "DeepSeek V3.2 tech report Table 2: minimax-m2=83.0." }, { "model_id": "minimax-m2", "benchmark_id": "browsecomp", "score": 44, "reference_url": "https://huggingface.co/MiniMaxAI/MiniMax-M2", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "agentic", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "MiniMax M2 model card: minimax-m2=44." }, { "model_id": "minimax-m2", "benchmark_id": "chatbot_arena_elo", "score": 1346, "reference_url": "https://lmarena.ai/leaderboard/text", "audit_status": "verified", "source_type": "leaderboard", "matches_canonical": true, "notes": "lmarena.ai text arena (fetched 2026-04-29). Arena model name: minimax-m2 (rank 165). ELO updates continuously; score reflects latest available." }, { "model_id": "minimax-m2", "benchmark_id": "humaneval", "score": 90.0, "reference_url": "https://artificialanalysis.ai/articles/minimax-m2-benchmarks-and-analysis", "audit_status": "needs_review", "notes": "Reference URL is AA article page for minimax-m2. AA model page shows humaneval=null for minimax-m2. Cannot verify 90.0% from AA source. AA does not appear to have independently tested humaneval for this model." }, { "model_id": "minimax-m2", "benchmark_id": "ifeval", "score": 88.0, "reference_url": "https://artificialanalysis.ai/models/minimax-m2", "audit_status": "needs_review", "notes": "Reference URL is AA model page (minimax-m2). AA uses field name \"ifbench\" (not ifeval); current value ifbench=0.723→72.3%. BP has 88.0% for ifeval. Major mismatch (15.7pp). Possible different benchmark/field or stale data. Current AA ifbench value: 72.3%." }, { "model_id": "minimax-m2", "benchmark_id": "math_500", "score": 97.0, "reference_url": "https://artificialanalysis.ai/models/minimax-m2", "audit_status": "needs_review", "notes": "Reference URL is AA model page (minimax-m2), but AA shows math_500=null for minimax-m2. Cannot verify 97.0% from AA source." }, { "model_id": "minimax-m2", "benchmark_id": "simpleqa", "score": 40.0, "reference_url": "https://artificialanalysis.ai/models/minimax-m2", "audit_status": "needs_review", "notes": "Reference URL is AA model page (minimax-m2). AA does not publish SimpleQA scores; field is not present in AA model data for any model. Cannot verify from AA source. Original source needs identification." }, { "model_id": "minimax-m2", "benchmark_id": "terminal_bench", "score": 30.0, "reference_url": "https://huggingface.co/deepseek-ai/DeepSeek-V3.2/resolve/main/assets/paper.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "agentic", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "DeepSeek V3.2 tech report Table 2: minimax-m2=30.0.", "candidates": [ { "score": 46.3, "reference_url": "https://huggingface.co/MiniMaxAI/MiniMax-M2", "source_type": "model_card", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "agentic", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "notes": "MiniMax M2 model card: minimax-m2=46.3." } ] }, { "model_id": "minimax-m2", "benchmark_id": "aime_2024", "score": 65, "reference_url": "https://llm-stats.com/models/minimax-m2", "audit_status": "dropped", "rule_id": "R5h-third-party-blog", "notes": "DROPPED (R5h): random third-party blog/aggregator, no primary trail" }, { "model_id": "minimax-m2", "benchmark_id": "codeforces_rating", "score": 1700, "reference_url": "https://llm-stats.com/models/minimax-m2", "audit_status": "dropped", "rule_id": "R5h-third-party-blog", "notes": "DROPPED (R5h): random third-party blog/aggregator, no primary trail" }, { "model_id": "minimax-m2", "benchmark_id": "mmlu", "score": 87, "reference_url": "https://llm-stats.com/models/minimax-m2", "audit_status": "dropped", "rule_id": "R5h-third-party-blog", "notes": "DROPPED (R5h): random third-party blog/aggregator, no primary trail" }, { "model_id": "minimax-m2", "benchmark_id": "mmmu", "score": 75, "reference_url": "https://llm-stats.com/models/minimax-m2", "audit_status": "dropped", "rule_id": "R5h-third-party-blog", "notes": "DROPPED (R5h): random third-party blog/aggregator, no primary trail" }, { "model_id": "minimax-m2", "benchmark_id": "swe_bench_pro", "score": 32, "reference_url": "https://llm-stats.com/models/minimax-m2", "audit_status": "dropped", "rule_id": "R5h-third-party-blog", "notes": "DROPPED (R5h): random third-party blog/aggregator, no primary trail" }, { "model_id": "minimax-m2", "benchmark_id": "terminal_bench_1", "score": 42, "reference_url": "https://www.tbench.ai/leaderboard/terminal-bench/1.0", "audit_status": "verified", "source_type": "leaderboard", "rule_ids": [ "R5d" ], "notes": "Confirmed 42.0% on tbench.ai Terminal-Bench 1.0 leaderboard (iFlow CLI scaffold, 2025-11-11)." }, { "model_id": "olmo-2-13b", "benchmark_id": "mmlu", "score": 68.5, "reference_url": "https://arxiv.org/abs/2501.00656", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "OLMES", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "OLMo 2 paper Table 7 Instruct: mmlu=68.5. [R5d: prior unverified value 67.5 from https://allenai.org/blog/olmo2 deleted.]" }, { "model_id": "olmo-2-13b", "benchmark_id": "mmlu_pro", "score": 35.1, "reference_url": "https://huggingface.co/allenai/OLMo-2-1124-13B", "audit_status": "dropped", "rule_id": "R5g-ghost", "audit_note": "HF card allenai/OLMo-2-1124-13B shows BASE model MMLUPro=35.1; wrong variant (Base vs Instruct).", "audited_by": "lychee-audit-bot", "audited_at": "2026-07-26T00:00:00Z" }, { "model_id": "olmo-2-13b", "benchmark_id": "gpqa_diamond", "score": 25.0, "reference_url": "https://allenai.org/blog/olmo2", "audit_status": "dropped", "rule_id": "R5g-ghost", "audit_note": "allenai.org/blog/olmo2 is SSR JavaScript bundle; no readable benchmark tables extractable. Blog does not report GPQA Diamond for OLMo-2-13B.", "audited_by": "lychee-audit-bot", "audited_at": "2026-07-26T00:00:00Z" }, { "model_id": "olmo-2-13b", "benchmark_id": "humaneval", "score": 55.0, "reference_url": "https://allenai.org/blog/olmo2", "audit_status": "dropped", "rule_id": "R5g-ghost", "audit_note": "allenai.org/blog/olmo2 is SSR JavaScript bundle; no readable benchmark tables. Blog does not report HumanEval for OLMo-2-13B.", "audited_by": "lychee-audit-bot", "audited_at": "2026-07-26T00:00:00Z" }, { "model_id": "olmo-2-13b", "benchmark_id": "ifeval", "score": 82.6, "reference_url": "https://arxiv.org/abs/2501.00656", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "OLMES", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "OLMo 2 paper Table 7 Instruct: ifeval=82.6. [R5d: prior unverified value 68.0 from https://allenai.org/blog/olmo2 deleted.]" }, { "model_id": "olmo-2-13b", "benchmark_id": "aime_2024", "score": 5, "reference_url": "https://allenai.org/blog/olmo3", "audit_status": "dropped", "rule_id": "R5g-ghost", "audit_note": "allenai.org/blog/olmo3 covers OLMo 3 (7B, 32B); no benchmark data for olmo-2-13b in any table.", "audited_by": "lychee-audit-bot", "audited_at": "2026-04-29T13:03:41Z" }, { "model_id": "olmo-2-13b", "benchmark_id": "livecodebench", "score": 18, "reference_url": "https://allenai.org/blog/olmo3", "audit_status": "dropped", "rule_id": "R5g-ghost", "audit_note": "allenai.org/blog/olmo3 covers OLMo 3 (7B, 32B); no benchmark data for olmo-2-13b.", "audited_by": "lychee-audit-bot", "audited_at": "2026-04-29T13:03:41Z" }, { "model_id": "olmo-2-13b", "benchmark_id": "math_500", "score": 38, "reference_url": "https://allenai.org/blog/olmo3", "audit_status": "dropped", "rule_id": "R5g-ghost", "audit_note": "allenai.org/blog/olmo3 covers OLMo 3 (7B, 32B); no benchmark data for olmo-2-13b.", "audited_by": "lychee-audit-bot", "audited_at": "2026-04-29T13:03:41Z" }, { "model_id": "lfm2.5-1.2b-thinking", "benchmark_id": "gpqa_diamond", "score": 37.86, "reference_url": "https://www.liquid.ai/blog/lfm2-5-1-2b-thinking-on-device-reasoning-under-1gb", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1 (mean of 5 runs)", "judge": "rule-based", "harness": "official (LFM eval)", "prompt_style": "default", "temperature": 0.05 }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "Liquid AI LFM2.5-1.2B-Thinking blog: lfm2.5-1.2b-thinking=37.86." }, { "model_id": "lfm2.5-1.2b-thinking", "benchmark_id": "mmlu_pro", "score": 49.65, "reference_url": "https://www.liquid.ai/blog/lfm2-5-1-2b-thinking-on-device-reasoning-under-1gb", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1 (mean of 5 runs)", "judge": "rule-based", "harness": "official (LFM eval)", "prompt_style": "default", "temperature": 0.05 }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "Liquid AI LFM2.5-1.2B-Thinking blog: lfm2.5-1.2b-thinking=49.65." }, { "model_id": "lfm2.5-1.2b-thinking", "benchmark_id": "aime_2025", "score": 31.73, "reference_url": "https://www.liquid.ai/blog/lfm2-5-1-2b-thinking-on-device-reasoning-under-1gb", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1 (mean of 5 runs)", "judge": "rule-based", "harness": "official (LFM eval)", "prompt_style": "default", "temperature": 0.05 }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "Liquid AI LFM2.5-1.2B-Thinking blog: lfm2.5-1.2b-thinking=31.73." }, { "model_id": "lfm2.5-1.2b-thinking", "benchmark_id": "math_500", "score": 87.96, "reference_url": "https://www.liquid.ai/blog/lfm2-5-1-2b-thinking-on-device-reasoning-under-1gb", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1 (mean of 5 runs)", "judge": "rule-based", "harness": "official (LFM eval)", "prompt_style": "default", "temperature": 0.05 }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "Liquid AI LFM2.5-1.2B-Thinking blog: lfm2.5-1.2b-thinking=87.96." }, { "model_id": "lfm2.5-1.2b-thinking", "benchmark_id": "ifeval", "score": 88.42, "reference_url": "https://www.liquid.ai/blog/lfm2-5-1-2b-thinking-on-device-reasoning-under-1gb", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1 (mean of 5 runs)", "judge": "rule-based", "harness": "official (LFM eval)", "prompt_style": "default", "temperature": 0.05 }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "Liquid AI LFM2.5-1.2B-Thinking blog: lfm2.5-1.2b-thinking=88.42." }, { "model_id": "phi-4-mini", "benchmark_id": "mmlu", "score": 67.3, "reference_url": "https://huggingface.co/microsoft/Phi-4-mini-instruct", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "phi-4-mini HF model card: mmlu=67.3 (re-sourced from https://huggingface.co/microsoft/Phi-4-mini-instruct, matches BP value)" }, { "model_id": "phi-4-mini", "benchmark_id": "gpqa_diamond", "score": 30.4, "reference_url": "https://huggingface.co/microsoft/Phi-4-mini-instruct", "audit_status": "dropped", "notes": " | DROPPED: BP cell URL claims phi-4-mini HF or phi-4 HF, but gpqa_diamond value not in phi-4-mini HF model card. Ghost cell per R5g(a)." }, { "model_id": "phi-4-mini", "benchmark_id": "humaneval", "score": 74.4, "reference_url": "https://huggingface.co/microsoft/Phi-4-mini-instruct", "audit_status": "dropped", "notes": " | DROPPED: BP cell URL claims phi-4-mini HF or phi-4 HF, but humaneval value not in phi-4-mini HF model card. Ghost cell per R5g(a)." }, { "model_id": "phi-4-mini", "benchmark_id": "mmlu_pro", "score": 52.8, "reference_url": "https://huggingface.co/microsoft/Phi-4-mini-instruct", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "phi-4-mini HF model card: mmlu_pro=52.8 (re-sourced from https://huggingface.co/microsoft/Phi-4-mini-instruct, matches BP value)" }, { "model_id": "phi-4-mini", "benchmark_id": "gsm8k", "score": 88.6, "reference_url": "https://huggingface.co/microsoft/Phi-4-mini-instruct", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "phi-4-mini HF model card: gsm8k=88.6 (re-sourced from https://huggingface.co/microsoft/Phi-4-mini-instruct, matches BP value)" }, { "model_id": "phi-4-mini", "benchmark_id": "ifeval", "score": 72, "reference_url": "https://huggingface.co/microsoft/phi-4", "audit_status": "dropped", "notes": " | DROPPED: BP cell URL claims phi-4-mini HF or phi-4 HF, but ifeval value not in phi-4-mini HF model card. Ghost cell per R5g(a)." }, { "model_id": "phi-4-mini", "benchmark_id": "livecodebench", "score": 28, "reference_url": "https://huggingface.co/microsoft/phi-4", "audit_status": "dropped", "notes": " | DROPPED: BP cell URL claims phi-4-mini HF or phi-4 HF, but livecodebench value not in phi-4-mini HF model card. Ghost cell per R5g(a)." }, { "model_id": "phi-4-mini", "benchmark_id": "math_500", "score": 75, "reference_url": "https://huggingface.co/microsoft/phi-4", "audit_status": "dropped", "notes": " | DROPPED: BP cell URL claims phi-4-mini HF or phi-4 HF, but math_500 value not in phi-4-mini HF model card. Ghost cell per R5g(a)." }, { "model_id": "falcon3-10b", "benchmark_id": "ifeval", "score": 78.17, "reference_url": "https://huggingface.co/tiiuae/Falcon3-10B-Instruct", "audit_status": "verified", "matches_canonical": true, "source_type": "model_card", "notes": "Falcon3-10B-Instruct HF card shows IFEval=78.17.", "audited_by": "lychee-audit-bot", "audited_at": "2026-07-26T00:00:00Z" }, { "model_id": "falcon3-10b", "benchmark_id": "mmlu", "score": 73.1, "reference_url": "https://huggingface.co/blog/falcon3", "audit_status": "dropped", "rule_id": "R5g-ghost", "audit_note": "Blog attributes 73.1 to Falcon3-10B-Base; JSON model is Instruct. Source does not confirm Instruct score.", "audited_by": "lychee-audit-bot", "audited_at": "2026-04-29T12:56:37Z" }, { "model_id": "falcon3-10b", "benchmark_id": "mmlu_pro", "score": 42.5, "reference_url": "https://huggingface.co/blog/falcon3", "audit_status": "dropped", "rule_id": "R5g-ghost", "audit_note": "Blog attributes 42.5 to Falcon3-10B-Base; JSON model is Instruct. Source does not confirm Instruct score.", "audited_by": "lychee-audit-bot", "audited_at": "2026-04-29T12:56:37Z" }, { "model_id": "falcon3-10b", "benchmark_id": "gpqa_diamond", "score": 30.0, "reference_url": "https://huggingface.co/blog/falcon3", "audit_status": "dropped", "rule_id": "R5g-ghost", "audit_note": "Not mentioned in the HF blog for any Falcon3 model.", "audited_by": "lychee-audit-bot", "audited_at": "2026-04-29T12:56:37Z" }, { "model_id": "falcon3-10b", "benchmark_id": "humaneval", "score": 72.0, "reference_url": "https://huggingface.co/blog/falcon3", "audit_status": "dropped", "rule_id": "R5g-ghost", "audit_note": "Not mentioned in the HF blog for any Falcon3 model.", "audited_by": "lychee-audit-bot", "audited_at": "2026-04-29T12:56:37Z" }, { "model_id": "falcon3-10b", "benchmark_id": "aime_2024", "score": 8, "reference_url": "https://llm-stats.com/benchmarks", "audit_status": "dropped", "rule_id": "R5h-third-party-blog", "notes": "DROPPED (R5h): random third-party blog/aggregator, no primary trail" }, { "model_id": "falcon3-10b", "benchmark_id": "livecodebench", "score": 22, "reference_url": "https://llm-stats.com/benchmarks", "audit_status": "dropped", "rule_id": "R5h-third-party-blog", "notes": "DROPPED (R5h): random third-party blog/aggregator, no primary trail" }, { "model_id": "falcon3-10b", "benchmark_id": "math_500", "score": 62, "reference_url": "https://llm-stats.com/benchmarks", "audit_status": "dropped", "rule_id": "R5h-third-party-blog", "notes": "DROPPED (R5h): random third-party blog/aggregator, no primary trail" }, { "model_id": "internlm3-8b", "benchmark_id": "mmlu", "score": 76.6, "reference_url": "https://github.com/InternLM/InternLM", "audit_status": "verified", "source_type": "model_card", "matches_canonical": true, "reported_setting": { "mode": "non-thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "OpenCompass", "prompt_style": "0-shot" }, "notes": "GitHub README table: mmlu = 76.6. No * marker (non-thinking mode)." }, { "model_id": "internlm3-8b", "benchmark_id": "gpqa_diamond", "score": 37.4, "reference_url": "https://github.com/InternLM/InternLM", "audit_status": "verified", "source_type": "model_card", "matches_canonical": true, "reported_setting": { "mode": "non-thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "OpenCompass", "prompt_style": "0-shot" }, "notes": "GitHub README table: gpqa_diamond = 37.4. No * marker (non-thinking mode)." }, { "model_id": "internlm3-8b", "benchmark_id": "math_500", "score": null, "reference_url": "https://github.com/InternLM/InternLM", "reported_setting": { "mode": "thinking", "notes": "Thinking mode per footnote *" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "dropped", "notes": "Dropped: GitHub README marks this 83.0 with * = Thinking Mode. BP internlm3-8b is non-thinking (is_reasoning=false). Wrong variant per R5g-b." }, { "model_id": "internlm3-8b", "benchmark_id": "humaneval", "score": 82.3, "reference_url": "https://github.com/InternLM/InternLM", "audit_status": "verified", "source_type": "model_card", "matches_canonical": true, "reported_setting": { "mode": "non-thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "OpenCompass", "prompt_style": "0-shot" }, "notes": "GitHub README table: humaneval = 82.3. No * marker (non-thinking mode)." }, { "model_id": "internlm3-8b", "benchmark_id": "ifeval", "score": 79.3, "reference_url": "https://github.com/InternLM/InternLM", "audit_status": "verified", "source_type": "model_card", "matches_canonical": true, "reported_setting": { "mode": "non-thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "OpenCompass", "prompt_style": "0-shot" }, "notes": "GitHub README table: ifeval = 79.3. No * marker (non-thinking mode)." }, { "model_id": "internlm3-8b", "benchmark_id": "aime_2024", "score": 12, "reference_url": "https://llm-stats.com/benchmarks", "audit_status": "dropped", "rule_id": "R5h-third-party-blog", "notes": "DROPPED (R5h): random third-party blog/aggregator, no primary trail" }, { "model_id": "internlm3-8b", "benchmark_id": "livecodebench", "score": 32, "reference_url": "https://llm-stats.com/benchmarks", "audit_status": "dropped", "rule_id": "R5h-third-party-blog", "notes": "DROPPED (R5h): random third-party blog/aggregator, no primary trail" }, { "model_id": "internlm3-8b", "benchmark_id": "mmlu_pro", "score": 57.6, "reference_url": "https://github.com/InternLM/InternLM", "source_type": "model_card", "audit_status": "verified", "matches_canonical": true, "reported_setting": { "mode": "non-thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "OpenCompass", "prompt_style": "0-shot" }, "notes": "GitHub README: MMLU-Pro 0-shot = 57.6 (no *). Overwrote llm-stats value of 52." }, { "model_id": "seed-thinking-v1.5", "benchmark_id": "gpqa_diamond", "score": 77.3, "reference_url": "https://arxiv.org/abs/2504.13914", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Seed-Thinking-v1.5 paper Table 2: seed-thinking-v1.5=77.3." }, { "model_id": "seed-thinking-v1.5", "benchmark_id": "aime_2024", "score": 86.7, "reference_url": "https://arxiv.org/abs/2504.13914", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Seed-Thinking-v1.5 paper Table 2: seed-thinking-v1.5=86.7." }, { "model_id": "seed-thinking-v1.5", "benchmark_id": "aime_2025", "score": 74.0, "reference_url": "https://arxiv.org/abs/2504.13914", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Seed-Thinking-v1.5 paper Table 2: seed-thinking-v1.5=74.0." }, { "model_id": "seed-thinking-v1.5", "benchmark_id": "livecodebench", "score": 64.9, "reference_url": "https://github.com/ByteDance-Seed/Seed-Thinking-v1.5", "audit_status": "dropped", "notes": " [R5g-b] GitHub source reports LiveCodeBench v5 (64.9%); correct cell livecodebench_v5 already verified. Wrong benchmark_id column. Dropped." }, { "model_id": "seed-thinking-v1.5", "benchmark_id": "humaneval", "score": 90.0, "reference_url": "https://github.com/ByteDance-Seed/Seed-Thinking-v1.5", "audit_status": "dropped", "notes": " [R5g-a] GitHub README table for Seed-Thinking-v1.5 does not report HumanEval. Ghost cell — value not found in referenced source. Dropped." }, { "model_id": "seed-thinking-v1.5", "benchmark_id": "ifeval", "score": 87.4, "reference_url": "https://arxiv.org/abs/2504.13914", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Seed-Thinking-v1.5 paper Table 2: seed-thinking-v1.5=87.4." }, { "model_id": "seed-thinking-v1.5", "benchmark_id": "math_500", "score": 95.0, "reference_url": "https://github.com/ByteDance-Seed/Seed-Thinking-v1.5", "audit_status": "dropped", "notes": " [R5g-a] GitHub README table for Seed-Thinking-v1.5 does not report MATH-500. Ghost cell — value not found in referenced source. Dropped." }, { "model_id": "seed-thinking-v1.5", "benchmark_id": "mmlu_pro", "score": 87.0, "reference_url": "https://arxiv.org/abs/2504.13914", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Seed-Thinking-v1.5 paper Table 2: seed-thinking-v1.5=87.0." }, { "model_id": "seed-thinking-v1.5", "benchmark_id": "hle", "score": 22, "reference_url": "https://llm-stats.com/benchmarks", "audit_status": "dropped", "rule_id": "R5h-third-party-blog", "notes": "DROPPED (R5h): random third-party blog/aggregator, no primary trail" }, { "model_id": "seed-thinking-v1.5", "benchmark_id": "mmlu", "score": 87, "reference_url": "https://llm-stats.com/benchmarks", "audit_status": "dropped", "rule_id": "R5h-third-party-blog", "notes": "DROPPED (R5h): random third-party blog/aggregator, no primary trail" }, { "model_id": "seed-thinking-v1.5", "benchmark_id": "simpleqa", "score": 12.9, "reference_url": "https://arxiv.org/abs/2504.13914", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Seed-Thinking-v1.5 paper Table 2: seed-thinking-v1.5=12.9." }, { "model_id": "seed-thinking-v1.5", "benchmark_id": "swe_bench_verified", "score": 47.0, "reference_url": "https://arxiv.org/abs/2504.13914", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "agentic (bash + edit)", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Seed-Thinking-v1.5 paper Table 2: seed-thinking-v1.5=47.0." }, { "model_id": "claude-mythos", "benchmark_id": "mmmlu", "score": 92.7, "reference_url": "https://www-cdn.anthropic.com/8b8380204f74670be75e81c820ca8dda846ab289.pdf", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "none", "sampling": "pass@1 (avg 5 trials)", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Mythos sys card Table 6.3.A." }, { "model_id": "claude-opus-4.6", "benchmark_id": "mmmlu", "score": 91.1, "reference_url": "https://www-cdn.anthropic.com/8b8380204f74670be75e81c820ca8dda846ab289.pdf", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "none", "sampling": "pass@1 (avg 5 trials)", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Mythos sys card Table 6.3.A: Opus 4.6 = 91.1%. Verified per Anthropic standard config (thinking max effort, default sampling, avg 5 trials).", "candidates": [ { "score": 91.1, "reference_url": "https://deepmind.google/models/gemini/pro/", "source_type": "official_blog", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "notes": "DeepMind /models/gemini/pro/ cross-model Performance table: MMMLU 91.1%" } ] }, { "model_id": "claude-opus-4.7", "benchmark_id": "mmmlu", "score": 91.5, "reference_url": "https://www.anthropic.com/news/claude-opus-4-7", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1 (avg 5 trials)", "judge": "rule-based", "harness": "official", "prompt_style": "avg 14 non-English languages", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "Anthropic Opus 4.7 blog table: claude-opus-4.7 = 91.5%." }, { "model_id": "claude-mythos", "benchmark_id": "graphwalks_bfs_256k_1m", "score": 80.0, "reference_url": "https://www-cdn.anthropic.com/8b8380204f74670be75e81c820ca8dda846ab289.pdf", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "none", "sampling": "pass@1 (avg 5 trials)", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Mythos sys card Table 6.3.A." }, { "model_id": "claude-opus-4.6", "benchmark_id": "graphwalks_bfs_256k_1m", "score": 38.7, "reference_url": "https://www-cdn.anthropic.com/8b8380204f74670be75e81c820ca8dda846ab289.pdf", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "none", "sampling": "pass@1 (avg 5 trials)", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Mythos sys card Table 6.3.A: Opus 4.6 = 38.7%. Verified per Anthropic standard config (thinking max effort, default sampling, avg 5 trials)." }, { "model_id": "gpt-5.4", "benchmark_id": "graphwalks_bfs_256k_1m", "score": 21.4, "reference_url": "https://openai.com/index/introducing-gpt-5-4/", "reported_setting": { "mode": "thinking", "effort": "xhigh", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-5.4 blog table: 21.4." }, { "model_id": "claude-mythos", "benchmark_id": "charxiv_reasoning", "score": 86.1, "reference_url": "https://www-cdn.anthropic.com/8b8380204f74670be75e81c820ca8dda846ab289.pdf", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "none", "sampling": "pass@1 (avg 5 trials)", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Mythos sys card Table 6.3.A: no-tools column. With-tools=93.2 not used.", "candidates": [ { "score": 86.2, "reference_url": "https://www.anthropic.com/claude-sonnet-5-system-card", "source_type": "model_card", "reported_setting": { "effort": "max", "tools": "none", "sampling": "avg 5 trials unless source states otherwise", "harness": "vision" }, "notes": "Anthropic Claude Sonnet 5 System Card Figure 8.10.5.A, page 130: Claude Mythos Preview; CharXiv Reasoning [no tools]=86.2. Source setting: effort=max; tools=none; sampling=avg 5 trials unless source states otherwise; harness=vision." }, { "score": 92.5, "reference_url": "https://www.anthropic.com/claude-sonnet-5-system-card", "source_type": "model_card", "reported_setting": { "effort": "max", "tools": "python / benchmark tools", "sampling": "avg 5 trials unless source states otherwise", "harness": "vision" }, "notes": "Anthropic Claude Sonnet 5 System Card Figure 8.10.5.A, page 130: Claude Mythos Preview; CharXiv Reasoning [Python tools]=92.5. Source setting: effort=max; tools=python / benchmark tools; sampling=avg 5 trials unless source states otherwise; harness=vision." } ] }, { "model_id": "claude-opus-4.6", "benchmark_id": "charxiv_reasoning", "score": 61.5, "reference_url": "https://www-cdn.anthropic.com/8b8380204f74670be75e81c820ca8dda846ab289.pdf", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "none", "sampling": "pass@1 (avg 5 trials)", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Mythos sys card Table 6.3.A: Opus 4.6 = 61.5%. Verified per Anthropic standard config (thinking max effort, default sampling, avg 5 trials).", "candidates": [ { "score": 69.1, "reference_url": "https://www.anthropic.com/news/claude-opus-4-7", "source_type": "third_party", "reported_setting": { "note": "see primary" }, "notes": "Opus 4.7 blog reports 69.1 (vs primary verified=61.5 from Mythos PDF, no-tools)." }, { "score": 69.1, "reference_url": "https://huggingface.co/moonshotai/Kimi-K2.6", "source_type": "model_card", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "1.0" }, "notes": "Kimi K2.6 model card: claude-opus-4.6=69.1." } ] }, { "model_id": "claude-opus-4.7", "benchmark_id": "charxiv_reasoning", "score": 82.1, "reference_url": "https://www.anthropic.com/news/claude-opus-4-7", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1 (avg 5 trials)", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "Anthropic Opus 4.7 blog table: claude-opus-4.7 = 82.1%.", "candidates": [ { "score": 82.1, "reference_url": "https://seed.bytedance.com/en/blog/seed2-1-officially-released-advancing-ai-productivity", "source_type": "official_blog", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "none", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official", "prompt_style": "official default", "temperature": "unknown" }, "notes": "Release blog EN: Vision capability chart (https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/user-upload/4xfa4mqq4yer3.png)." } ] }, { "model_id": "kimi-k2.6", "benchmark_id": "charxiv_reasoning", "score": 80.4, "reference_url": "https://huggingface.co/moonshotai/Kimi-K2.6", "reported_setting": { "mode": "thinking", "effort": "max (98304 generation tokens)", "tools": "none (per-bench override)", "sampling": "pass@1 (avg@10 coding, avg@3 vision)", "judge": "rule-based", "harness": "official (in-house SWE-agent-derived for SWE-Bench)", "prompt_style": "default", "temperature": "1.0" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Kimi K2.6 model card (HF moonshotai/Kimi-K2.6) Evaluation table: CharXiv RQ 80.4", "candidates": [ { "score": 80.4, "reference_url": "https://huggingface.co/moonshotai/Kimi-K2.6/resolve/d9cb81bc88b9bd2dc89877599f23c614ded72b9c/README.md", "source_type": "official_model_card", "reported_setting": { "mode": "thinking", "effort": "max; up to 98,304 generation tokens", "tools": "none", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official Kimi K2.6 evaluation", "prompt_style": "provider official evaluation prompt", "temperature": "1.0; top_p=1.0", "context": "262,144 (256K)", "input_modalities": "image and text", "trials": "source-reported" }, "notes": "Exact provider-official score and reported setting." } ] }, { "model_id": "claude-mythos", "benchmark_id": "swe_bench_multilingual", "score": 87.3, "reference_url": "https://www-cdn.anthropic.com/8b8380204f74670be75e81c820ca8dda846ab289.pdf", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "agentic", "sampling": "pass@1 (avg 5 trials)", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Mythos sys card §6.4: 5 trials." }, { "model_id": "claude-opus-4.6", "benchmark_id": "swe_bench_multilingual", "score": 77.8, "reference_url": "https://www-cdn.anthropic.com/8b8380204f74670be75e81c820ca8dda846ab289.pdf", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "agentic", "sampling": "pass@1 (avg 5 trials)", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Mythos sys card Table 6.3.A: Opus 4.6 = 77.8%. Verified per Anthropic standard config (thinking max effort, default sampling, avg 5 trials).", "candidates": [ { "score": 77.5, "reference_url": "https://huggingface.co/deepseek-ai/DeepSeek-V4-Pro", "source_type": "third_party", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "agentic (bash + edit)", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "notes": "DeepSeek V4-Pro model card (third-party): claude-opus-4.6 (per their own canonical config) = 77.5." }, { "score": 75.8, "reference_url": "https://cursor.com/resources/Composer2.pdf", "source_type": "tech_report", "reported_setting": { "effort": "high", "tools": "Cursor coding agent", "sampling": "pass@1", "harness": "Cursor harness" }, "notes": "Composer 2 technical report Table 1 / SWE-Bench Multilingual / Opus 4.6 High / value 1: For third-party public benchmarks, slash-separated values are Cursor/official-harness then self-reported. SWE prompts prepend 'please solve this github issue'; Terminal prompts add solution-format instructions." } ] }, { "model_id": "kimi-k2.6", "benchmark_id": "swe_bench_multilingual", "score": 76.7, "reference_url": "https://huggingface.co/deepseek-ai/DeepSeek-V4-Pro", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "agentic (bash + edit)", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "DeepSeek V4-Pro model card (third-party): kimi-k2.6 (per their own canonical config) = 76.7.", "candidates": [ { "score": 77.1, "reference_url": "https://research.nvidia.com/labs/nemotron/files/NVIDIA-Nemotron-3-Ultra-Technical-Report.pdf", "source_type": "tech_report", "reported_setting": { "version_metric": "Multilingual / % resolved", "tools": "agentic repository editing", "harness": "Harbor / OpenHands / AWS ECS", "sampling": "3-run average" }, "notes": "NVIDIA Nemotron 3 Ultra technical report, Final Table 10: Kimi K2.6: swe_bench_multilingual=77.1." } ] }, { "model_id": "deepseek-v4-pro", "benchmark_id": "swe_bench_multilingual", "score": 76.2, "reference_url": "https://huggingface.co/deepseek-ai/DeepSeek-V4-Pro", "reported_setting": { "mode": "thinking", "effort": "Max", "tools": "agentic (bash + edit)", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "DeepSeek V4 model card: deepseek-v4-pro (Max mode) = 76.2.", "candidates": [ { "score": 76.2, "reference_url": "https://llm-stats.com/benchmarks/swe-bench-multilingual", "source_type": "third_party_aggregator", "reported_setting": { "effort": "max" }, "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=DeepSeek-V4-Pro-Max, slug=deepseek-v4-pro-max, provider=DeepSeek" }, { "score": 72.44, "reference_url": "https://huggingface.co/internlm/Intern-S2-Preview-397B", "source_type": "model_card", "reported_setting": { "mode": "source does not state", "effort": "source does not state", "tools": "agentic code execution", "sampling": "source does not state", "judge": "benchmark-specific; exact version not stated", "harness": "Mini-SWE-Agent", "prompt_style": "source does not state", "temperature": "source does not state", "context": "source does not state", "notes": "Shanghai AI Lab evaluated all model columns using OpenCompass, VLMEvalKit, and AgentCompass. Per-model mode/effort/sampling are not reported; only Intern-S2 maximum evaluation lengths are disclosed." }, "notes": "Intern-S2-Preview-397B official model card: SWE-Bench-Multilingual = 72.44. This is a Shanghai AI Lab cross-model evaluation; the source does not disclose this model's exact mode, effort, or sampling." }, { "score": 76.5, "reference_url": "https://research.nvidia.com/labs/nemotron/files/NVIDIA-Nemotron-3-Ultra-Technical-Report.pdf", "source_type": "tech_report", "reported_setting": { "version_metric": "Multilingual / % resolved", "tools": "agentic repository editing", "harness": "Harbor / OpenHands / AWS ECS", "sampling": "3-run average" }, "notes": "NVIDIA Nemotron 3 Ultra technical report, Final Table 10: DeepSeek-V4-Pro: swe_bench_multilingual=76.5." } ] }, { "model_id": "deepseek-v4-flash", "benchmark_id": "swe_bench_multilingual", "score": 73.3, "reference_url": "https://huggingface.co/deepseek-ai/DeepSeek-V4-Pro", "reported_setting": { "mode": "thinking", "effort": "Max", "tools": "agentic (bash + edit)", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "DeepSeek V4 model card: deepseek-v4-flash (Max mode) = 73.3.", "candidates": [ { "score": 73.3, "reference_url": "https://llm-stats.com/benchmarks/swe-bench-multilingual", "source_type": "third_party_aggregator", "reported_setting": { "effort": "max" }, "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=DeepSeek-V4-Flash-Max, slug=deepseek-v4-flash-max, provider=DeepSeek" }, { "score": 75.0, "reference_url": "https://research.nvidia.com/labs/nemotron/files/NVIDIA-Nemotron-3-Ultra-Technical-Report.pdf", "source_type": "tech_report", "reported_setting": { "version_metric": "Multilingual / % resolved", "tools": "agentic repository editing", "harness": "Harbor / OpenHands / AWS ECS", "sampling": "3-run average" }, "notes": "NVIDIA Nemotron 3 Ultra technical report, Final Table 10: DeepSeek-V4-Flash Preview: swe_bench_multilingual=75." } ] }, { "model_id": "claude-mythos", "benchmark_id": "swe_bench_multimodal", "score": 59.0, "reference_url": "https://www-cdn.anthropic.com/8b8380204f74670be75e81c820ca8dda846ab289.pdf", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "agentic", "sampling": "pass@1 (avg 5 trials)", "judge": "rule-based", "harness": "Anthropic internal", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Mythos sys card §6.4: internal harness; per benchmark canonical, lab-internal harness accepted." }, { "model_id": "claude-opus-4.6", "benchmark_id": "swe_bench_multimodal", "score": 27.1, "reference_url": "https://www-cdn.anthropic.com/8b8380204f74670be75e81c820ca8dda846ab289.pdf", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "agentic", "sampling": "pass@1 (avg 5 trials)", "judge": "rule-based", "harness": "Anthropic internal", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Mythos sys card Table 6.3.A: Opus 4.6 = 27.1%. Verified per Anthropic standard config (thinking max effort, default sampling, avg 5 trials)." }, { "model_id": "deepseek-v4-pro", "benchmark_id": "hmmt_feb_2026", "score": 95.2, "reference_url": "https://huggingface.co/deepseek-ai/DeepSeek-V4-Pro", "reported_setting": { "mode": "thinking", "effort": "Max", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "DeepSeek V4 model card: deepseek-v4-pro (Max mode) = 95.2.", "candidates": [ { "score": 91.76, "reference_url": "https://huggingface.co/internlm/Intern-S2-Preview-397B", "source_type": "model_card", "reported_setting": { "mode": "source does not state", "effort": "source does not state", "tools": "none", "sampling": "source does not state", "judge": "benchmark-specific; exact version not stated", "harness": "OpenCompass or AgentCompass; exact adapter not stated", "prompt_style": "source does not state", "temperature": "source does not state", "context": "source does not state", "notes": "Shanghai AI Lab evaluated all model columns using OpenCompass, VLMEvalKit, and AgentCompass. Per-model mode/effort/sampling are not reported; only Intern-S2 maximum evaluation lengths are disclosed." }, "notes": "Intern-S2-Preview-397B official model card: HMMT-2026 = 91.76. This is a Shanghai AI Lab cross-model evaluation; the source does not disclose this model's exact mode, effort, or sampling." }, { "score": 95.2, "reference_url": "https://microsoft.ai/wp-content/uploads/2026/06/main_20260602_2.pdf", "source_type": "official_paper", "reported_setting": { "mode": "thinking", "effort": "Max", "tools": "none (per-bench override)", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default", "context": "default", "notes": "Per DeepSeek V4-Pro model card. Three reasoning modes available (Non-Think/High/Max); BP canonical = Max mode (most powerful). Other modes' values stored as candidates if needed." }, "notes": "Displayed exactly as 95.2. Official Microsoft-reported result. Comparator value is quoted from that provider's official model card/release; use the campaign canonical setting." } ] }, { "model_id": "deepseek-v4-flash", "benchmark_id": "hmmt_feb_2026", "score": 94.8, "reference_url": "https://huggingface.co/deepseek-ai/DeepSeek-V4-Pro", "reported_setting": { "mode": "thinking", "effort": "Max", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "DeepSeek V4 model card: deepseek-v4-flash (Max mode) = 94.8.", "candidates": [] }, { "model_id": "kimi-k2.6", "benchmark_id": "hmmt_feb_2026", "score": 92.7, "reference_url": "https://huggingface.co/moonshotai/Kimi-K2.6", "reported_setting": { "mode": "thinking", "effort": "max (98304 generation tokens)", "tools": "none (per-bench override)", "sampling": "pass@1 (avg@10 coding, avg@3 vision)", "judge": "rule-based", "harness": "official (in-house SWE-agent-derived for SWE-Bench)", "prompt_style": "default", "temperature": "1.0" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Kimi K2.6 model card (HF moonshotai/Kimi-K2.6) Evaluation table: HMMT 2026 Feb 92.7", "candidates": [ { "score": 92.7, "reference_url": "https://microsoft.ai/wp-content/uploads/2026/06/main_20260602_2.pdf", "source_type": "official_paper", "reported_setting": { "mode": "thinking", "effort": "max (98304 generation tokens)", "tools": "none (per-bench override)", "sampling": "pass@1 (avg@10 coding, avg@3 vision)", "judge": "rule-based", "harness": "official (in-house SWE-agent-derived for SWE-Bench)", "prompt_style": "default", "temperature": "1.0", "context": "262144 (256K)", "notes": "Per Kimi K2.6 model card (HF moonshotai/Kimi-K2.6). Thinking mode, max effort, t=1.0 top_p=1.0." }, "notes": "Displayed exactly as 92.7. Official Microsoft-reported result. Comparator value is quoted from that provider's official model card/release; use the campaign canonical setting." } ] }, { "model_id": "gpt-5.4", "benchmark_id": "hmmt_feb_2026", "score": 97.7, "reference_url": "https://huggingface.co/deepseek-ai/DeepSeek-V4-Pro", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "DeepSeek V4-Pro model card (third-party): gpt-5.4 (per their own canonical config) = 97.7.", "candidates": [ { "score": 91.8, "reference_url": "https://z.ai/blog/glm-5.1", "source_type": "official_blog", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "notes": "GLM-5.1 blog: gpt-5.4=91.8." } ] }, { "model_id": "claude-opus-4.6", "benchmark_id": "hmmt_feb_2026", "score": 96.2, "reference_url": "https://huggingface.co/deepseek-ai/DeepSeek-V4-Pro", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "DeepSeek V4-Pro model card (third-party): claude-opus-4.6 (per their own canonical config) = 96.2.", "candidates": [ { "score": 84.3, "reference_url": "https://z.ai/blog/glm-5.1", "source_type": "official_blog", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "notes": "GLM-5.1 blog: claude-opus-4.6=84.3." } ] }, { "model_id": "gemini-3.1-pro", "benchmark_id": "hmmt_feb_2026", "score": 94.7, "reference_url": "https://huggingface.co/deepseek-ai/DeepSeek-V4-Pro", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "DeepSeek V4-Pro model card (third-party): gemini-3.1-pro (per their own canonical config) = 94.7.", "candidates": [ { "score": 87.3, "reference_url": "https://z.ai/blog/glm-5.1", "source_type": "official_blog", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "notes": "GLM-5.1 blog: gemini-3.1-pro=87.3." } ] }, { "model_id": "kimi-k2.5", "benchmark_id": "hmmt_feb_2026", "score": 81.3, "reference_url": "https://z.ai/blog/glm-5.1", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "GLM-5.1 blog: kimi-k2.5=81.3.", "candidates": [ { "score": 87.1, "reference_url": "https://huggingface.co/moonshotai/Kimi-K2.6", "source_type": "model_card", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "1.0" }, "notes": "(merged from hmmt_2026_feb) Kimi K2.6 model card: kimi-k2.5=87.1." } ] }, { "model_id": "glm-5.1", "benchmark_id": "hmmt_feb_2026", "score": 82.6, "reference_url": "https://z.ai/blog/glm-5.2", "reported_setting": { "mode": "thinking", "effort": "official reported", "sampling": "pass@1", "prompt_style": "default", "tools": "none", "temperature": "1.0", "top_p": "0.95", "context": "default", "max_output_tokens": "163840", "judge": "GPT-5.5 medium for AIME/HMMT/IMO; benchmark-specified otherwise", "harness": "official reasoning evaluation" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "GLM-5.2 official release source: GLM-5.1 / HMMT Feb. 2026 = 82.6.", "candidates": [ { "score": 89.4, "reference_url": "https://huggingface.co/deepseek-ai/DeepSeek-V4-Pro", "source_type": "model_card", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "notes": "DeepSeek V4-Pro model card (third-party): glm-5.1 (per their own canonical config) = 89.4. [Displaced by GLM-5.2 first-party audit.]" }, { "score": 82.6, "reference_url": "https://microsoft.ai/wp-content/uploads/2026/06/main_20260602_2.pdf", "source_type": "official_paper", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none (per-bench override)", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default", "context": "default", "notes": "Per GLM-5.1 blog." }, "notes": "Displayed exactly as 82.6. Official Microsoft-reported result. Comparator value is quoted from that provider's official model card/release; use the campaign canonical setting." } ] }, { "model_id": "deepseek-v4-pro", "benchmark_id": "imo_answerbench", "score": 89.8, "reference_url": "https://huggingface.co/deepseek-ai/DeepSeek-V4-Pro", "reported_setting": { "mode": "thinking", "effort": "Max", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "DeepSeek V4 model card: deepseek-v4-pro (Max mode) = 89.8.", "candidates": [ { "score": 93.0, "reference_url": "https://research.nvidia.com/labs/nemotron/files/NVIDIA-Nemotron-3-Ultra-Technical-Report.pdf", "source_type": "tech_report", "reported_setting": { "version_metric": "no tools / %", "tools": "none", "harness": "NeMo Evaluator", "sampling": "5-run average", "variant": "no tools" }, "notes": "NVIDIA Nemotron 3 Ultra technical report, Final Table 10: DeepSeek-V4-Pro: imo_answerbench=93, variant=no tools." }, { "score": 85.4, "reference_url": "https://research.nvidia.com/labs/nemotron/files/NVIDIA-Nemotron-3-Ultra-Technical-Report.pdf", "source_type": "tech_report", "reported_setting": { "version_metric": "with tools / %", "tools": "benchmark tool environment", "harness": "NeMo Evaluator", "sampling": "pass@1", "variant": "with tools" }, "notes": "NVIDIA Nemotron 3 Ultra technical report, Final Table 10: DeepSeek-V4-Pro: imo_answerbench=85.4, variant=with tools." }, { "score": 88.0, "reference_url": "https://research.nvidia.com/labs/nemotron/files/NVIDIA-Nemotron-3-Ultra-Technical-Report.pdf", "source_type": "tech_report", "reported_setting": { "version_metric": "no tools / %", "tools": "none", "harness": "NeMo Evaluator", "sampling": "pass@1", "variant": "High", "effort": "High" }, "notes": "NVIDIA Nemotron 3 Ultra technical report, Internal Table 3: DeepSeek V4 Pro (High): imo_answerbench=88, variant=High." } ] }, { "model_id": "deepseek-v4-flash", "benchmark_id": "imo_answerbench", "score": 88.4, "reference_url": "https://huggingface.co/deepseek-ai/DeepSeek-V4-Pro", "reported_setting": { "mode": "thinking", "effort": "Max", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "DeepSeek V4 model card: deepseek-v4-flash (Max mode) = 88.4.", "candidates": [ { "score": 88.4, "reference_url": "https://llm-stats.com/benchmarks/imo-answerbench", "source_type": "third_party_aggregator", "reported_setting": { "effort": "max" }, "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=DeepSeek-V4-Flash-Max, slug=deepseek-v4-flash-max, provider=DeepSeek" }, { "score": 91.1, "reference_url": "https://research.nvidia.com/labs/nemotron/files/NVIDIA-Nemotron-3-Ultra-Technical-Report.pdf", "source_type": "tech_report", "reported_setting": { "version_metric": "no tools / %", "tools": "none", "harness": "NeMo Evaluator", "sampling": "5-run average", "variant": "no tools" }, "notes": "NVIDIA Nemotron 3 Ultra technical report, Final Table 10: DeepSeek-V4-Flash Preview: imo_answerbench=91.1, variant=no tools." }, { "score": 89.6, "reference_url": "https://research.nvidia.com/labs/nemotron/files/NVIDIA-Nemotron-3-Ultra-Technical-Report.pdf", "source_type": "tech_report", "reported_setting": { "version_metric": "with tools / %", "tools": "benchmark tool environment", "harness": "NeMo Evaluator", "sampling": "pass@1", "variant": "with tools" }, "notes": "NVIDIA Nemotron 3 Ultra technical report, Final Table 10: DeepSeek-V4-Flash Preview: imo_answerbench=89.6, variant=with tools." } ] }, { "model_id": "kimi-k2.6", "benchmark_id": "imo_answerbench", "score": 86.0, "reference_url": "https://huggingface.co/deepseek-ai/DeepSeek-V4-Pro", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "DeepSeek V4-Pro model card (third-party): kimi-k2.6 (per their own canonical config) = 86.0.", "candidates": [ { "score": 91.1, "reference_url": "https://research.nvidia.com/labs/nemotron/files/NVIDIA-Nemotron-3-Ultra-Technical-Report.pdf", "source_type": "tech_report", "reported_setting": { "version_metric": "no tools / %", "tools": "none", "harness": "NeMo Evaluator", "sampling": "5-run average", "variant": "no tools" }, "notes": "NVIDIA Nemotron 3 Ultra technical report, Final Table 10: Kimi K2.6: imo_answerbench=91.1, variant=no tools." }, { "score": 93.71, "reference_url": "https://research.nvidia.com/labs/nemotron/files/NVIDIA-Nemotron-3-Ultra-Technical-Report.pdf", "source_type": "tech_report", "reported_setting": { "version_metric": "with tools / %", "tools": "benchmark tool environment", "harness": "NeMo Evaluator", "sampling": "pass@1", "variant": "with tools" }, "notes": "NVIDIA Nemotron 3 Ultra technical report, Final Table 10: Kimi K2.6: imo_answerbench=93.71, variant=with tools." } ] }, { "model_id": "gpt-5.4", "benchmark_id": "imo_answerbench", "score": 91.4, "reference_url": "https://huggingface.co/deepseek-ai/DeepSeek-V4-Pro", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "DeepSeek V4-Pro model card (third-party): gpt-5.4 (per their own canonical config) = 91.4." }, { "model_id": "claude-opus-4.6", "benchmark_id": "imo_answerbench", "score": 75.3, "reference_url": "https://huggingface.co/deepseek-ai/DeepSeek-V4-Pro", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "DeepSeek V4-Pro model card (third-party): claude-opus-4.6 (per their own canonical config) = 75.3." }, { "model_id": "kimi-k2.5", "benchmark_id": "imo_answerbench", "score": 81.8, "reference_url": "https://huggingface.co/moonshotai/Kimi-K2.5", "reported_setting": { "mode": "thinking", "effort": "default (96k for reasoning, 64k for vision)", "tools": "none (per-bench override)", "sampling": "pass@1 (avg@32 AIME/HMMT, avg@8 GPQA, avg@3 vision, avg@4 Seal/WideSearch)", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "1.0" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Kimi K2.5 model card (HF moonshotai/Kimi-K2.5) Evaluation table: IMO-AnswerBench 81.8" }, { "model_id": "glm-5.1", "benchmark_id": "imo_answerbench", "score": 83.8, "reference_url": "https://z.ai/blog/glm-5.2", "reported_setting": { "mode": "thinking", "effort": "official reported", "sampling": "pass@1", "prompt_style": "default", "tools": "none", "temperature": "1.0", "top_p": "0.95", "context": "default", "max_output_tokens": "163840", "judge": "GPT-5.5 medium for AIME/HMMT/IMO; benchmark-specified otherwise", "harness": "official reasoning evaluation" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "GLM-5.2 official release source: GLM-5.1 / IMOAnswerBench = 83.8.", "candidates": [ { "score": 86.8, "reference_url": "https://research.nvidia.com/labs/nemotron/files/NVIDIA-Nemotron-3-Ultra-Technical-Report.pdf", "source_type": "tech_report", "reported_setting": { "version_metric": "no tools / %", "tools": "none", "harness": "NeMo Evaluator", "sampling": "5-run average", "variant": "no tools" }, "notes": "NVIDIA Nemotron 3 Ultra technical report, Final Table 10: GLM-5.1: imo_answerbench=86.8, variant=no tools." }, { "score": 91.1, "reference_url": "https://research.nvidia.com/labs/nemotron/files/NVIDIA-Nemotron-3-Ultra-Technical-Report.pdf", "source_type": "tech_report", "reported_setting": { "version_metric": "with tools / %", "tools": "benchmark tool environment", "harness": "NeMo Evaluator", "sampling": "pass@1", "variant": "with tools" }, "notes": "NVIDIA Nemotron 3 Ultra technical report, Final Table 10: GLM-5.1: imo_answerbench=91.1, variant=with tools." } ] }, { "model_id": "deepseek-v4-pro", "benchmark_id": "apex_shortlist", "score": 90.2, "reference_url": "https://huggingface.co/deepseek-ai/DeepSeek-V4-Pro", "reported_setting": { "mode": "thinking", "effort": "Max", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "DeepSeek V4 model card: deepseek-v4-pro (Max mode) = 90.2.", "candidates": [ { "score": 85.8, "reference_url": "https://research.nvidia.com/labs/nemotron/files/NVIDIA-Nemotron-3-Ultra-Technical-Report.pdf", "source_type": "tech_report", "reported_setting": { "version_metric": "no tools / %", "tools": "none", "harness": "NeMo Evaluator", "sampling": "pass@1", "variant": "no tools" }, "notes": "NVIDIA Nemotron 3 Ultra technical report, Final Table 10: DeepSeek-V4-Pro: apex_shortlist=85.8, variant=no tools." }, { "score": 86.5, "reference_url": "https://research.nvidia.com/labs/nemotron/files/NVIDIA-Nemotron-3-Ultra-Technical-Report.pdf", "source_type": "tech_report", "reported_setting": { "version_metric": "with tools / %", "tools": "benchmark tool environment", "harness": "NeMo Evaluator", "sampling": "pass@1", "variant": "with tools" }, "notes": "NVIDIA Nemotron 3 Ultra technical report, Final Table 10: DeepSeek-V4-Pro: apex_shortlist=86.5, variant=with tools." }, { "score": 85.5, "reference_url": "https://research.nvidia.com/labs/nemotron/files/NVIDIA-Nemotron-3-Ultra-Technical-Report.pdf", "source_type": "tech_report", "reported_setting": { "version_metric": "no tools / %", "tools": "none", "harness": "NeMo Evaluator", "sampling": "pass@1", "variant": "High", "effort": "High" }, "notes": "NVIDIA Nemotron 3 Ultra technical report, Internal Table 3: DeepSeek V4 Pro (High): apex_shortlist=85.5, variant=High." } ] }, { "model_id": "deepseek-v4-flash", "benchmark_id": "apex_shortlist", "score": 85.7, "reference_url": "https://huggingface.co/deepseek-ai/DeepSeek-V4-Pro", "reported_setting": { "mode": "thinking", "effort": "Max", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "DeepSeek V4 model card: deepseek-v4-flash (Max mode) = 85.7.", "candidates": [ { "score": 82.4, "reference_url": "https://research.nvidia.com/labs/nemotron/files/NVIDIA-Nemotron-3-Ultra-Technical-Report.pdf", "source_type": "tech_report", "reported_setting": { "version_metric": "no tools / %", "tools": "none", "harness": "NeMo Evaluator", "sampling": "pass@1", "variant": "no tools" }, "notes": "NVIDIA Nemotron 3 Ultra technical report, Final Table 10: DeepSeek-V4-Flash Preview: apex_shortlist=82.4, variant=no tools." }, { "score": 82.0, "reference_url": "https://research.nvidia.com/labs/nemotron/files/NVIDIA-Nemotron-3-Ultra-Technical-Report.pdf", "source_type": "tech_report", "reported_setting": { "version_metric": "with tools / %", "tools": "benchmark tool environment", "harness": "NeMo Evaluator", "sampling": "pass@1", "variant": "with tools" }, "notes": "NVIDIA Nemotron 3 Ultra technical report, Final Table 10: DeepSeek-V4-Flash Preview: apex_shortlist=82, variant=with tools." } ] }, { "model_id": "kimi-k2.6", "benchmark_id": "apex_shortlist", "score": 75.5, "reference_url": "https://huggingface.co/deepseek-ai/DeepSeek-V4-Pro", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "DeepSeek V4-Pro model card (third-party): kimi-k2.6 (per their own canonical config) = 75.5.", "candidates": [ { "score": 77.4, "reference_url": "https://research.nvidia.com/labs/nemotron/files/NVIDIA-Nemotron-3-Ultra-Technical-Report.pdf", "source_type": "tech_report", "reported_setting": { "version_metric": "no tools / %", "tools": "none", "harness": "NeMo Evaluator", "sampling": "pass@1", "variant": "no tools" }, "notes": "NVIDIA Nemotron 3 Ultra technical report, Final Table 10: Kimi K2.6: apex_shortlist=77.4, variant=no tools." }, { "score": 73.2, "reference_url": "https://research.nvidia.com/labs/nemotron/files/NVIDIA-Nemotron-3-Ultra-Technical-Report.pdf", "source_type": "tech_report", "reported_setting": { "version_metric": "with tools / %", "tools": "benchmark tool environment", "harness": "NeMo Evaluator", "sampling": "pass@1", "variant": "with tools" }, "notes": "NVIDIA Nemotron 3 Ultra technical report, Final Table 10: Kimi K2.6: apex_shortlist=73.2, variant=with tools." } ] }, { "model_id": "gpt-5.4", "benchmark_id": "apex_shortlist", "score": 78.1, "reference_url": "https://huggingface.co/deepseek-ai/DeepSeek-V4-Pro", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "DeepSeek V4-Pro model card (third-party): gpt-5.4 (per their own canonical config) = 78.1." }, { "model_id": "claude-opus-4.6", "benchmark_id": "apex_shortlist", "score": 85.9, "reference_url": "https://huggingface.co/deepseek-ai/DeepSeek-V4-Pro", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "DeepSeek V4-Pro model card (third-party): claude-opus-4.6 (per their own canonical config) = 85.9." }, { "model_id": "gemini-3.1-pro", "benchmark_id": "apex_shortlist", "score": 89.1, "reference_url": "https://huggingface.co/deepseek-ai/DeepSeek-V4-Pro", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "DeepSeek V4-Pro model card (third-party): gemini-3.1-pro (per their own canonical config) = 89.1." }, { "model_id": "glm-5.1", "benchmark_id": "apex_shortlist", "score": 72.4, "reference_url": "https://huggingface.co/deepseek-ai/DeepSeek-V4-Pro", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "DeepSeek V4-Pro model card (third-party): glm-5.1 (per their own canonical config) = 72.4.", "candidates": [ { "score": 71.1, "reference_url": "https://research.nvidia.com/labs/nemotron/files/NVIDIA-Nemotron-3-Ultra-Technical-Report.pdf", "source_type": "tech_report", "reported_setting": { "version_metric": "no tools / %", "tools": "none", "harness": "NeMo Evaluator", "sampling": "pass@1", "variant": "no tools" }, "notes": "NVIDIA Nemotron 3 Ultra technical report, Final Table 10: GLM-5.1: apex_shortlist=71.1, variant=no tools." }, { "score": 79.0, "reference_url": "https://research.nvidia.com/labs/nemotron/files/NVIDIA-Nemotron-3-Ultra-Technical-Report.pdf", "source_type": "tech_report", "reported_setting": { "version_metric": "with tools / %", "tools": "benchmark tool environment", "harness": "NeMo Evaluator", "sampling": "pass@1", "variant": "with tools" }, "notes": "NVIDIA Nemotron 3 Ultra technical report, Final Table 10: GLM-5.1: apex_shortlist=79, variant=with tools." } ] }, { "model_id": "deepseek-v4-pro", "benchmark_id": "mcpatlas", "score": 73.6, "reference_url": "https://huggingface.co/deepseek-ai/DeepSeek-V4-Pro", "reported_setting": { "mode": "thinking", "effort": "Max", "tools": "agentic", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "DeepSeek V4 model card: deepseek-v4-pro (Max mode) = 73.6.", "candidates": [ { "score": 79.7, "reference_url": "https://huggingface.co/tencent/Hy3", "source_type": "model_card", "reported_setting": { "mode": "thinking", "effort": "high / highest available", "tools": "MCP tool servers", "sampling": "pass@1", "judge": "Gemini 2.5 Pro", "harness": "Scale April 2026 methodology; public 500; 100 tool-call budget", "prompt_style": "default", "temperature": "0.9 for Hy3 family; otherwise source/provider setting", "context": "256000 for Hy3 family; otherwise source/provider setting", "notes": "Tencent own test" }, "notes": "Hy3 official model-card appendix: MCP atlas (public) = 79.7*. Tencent own testing." } ] }, { "model_id": "deepseek-v4-flash", "benchmark_id": "mcpatlas", "score": 69.0, "reference_url": "https://huggingface.co/deepseek-ai/DeepSeek-V4-Pro", "reported_setting": { "mode": "thinking", "effort": "Max", "tools": "agentic", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "DeepSeek V4 model card: deepseek-v4-flash (Max mode) = 69.0.", "candidates": [ { "score": 69.0, "reference_url": "https://llm-stats.com/benchmarks/mcp-atlas", "source_type": "third_party_aggregator", "reported_setting": { "effort": "max" }, "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=DeepSeek-V4-Flash-Max, slug=deepseek-v4-flash-max, provider=DeepSeek" }, { "score": 77.9, "reference_url": "https://huggingface.co/tencent/Hy3", "source_type": "model_card", "reported_setting": { "mode": "thinking", "effort": "high / highest available", "tools": "MCP tool servers", "sampling": "pass@1", "judge": "Gemini 2.5 Pro", "harness": "Scale April 2026 methodology; public 500; 100 tool-call budget", "prompt_style": "default", "temperature": "0.9 for Hy3 family; otherwise source/provider setting", "context": "256000 for Hy3 family; otherwise source/provider setting", "notes": "Tencent own test" }, "notes": "Hy3 official model-card appendix: MCP atlas (public) = 77.9*. Tencent own testing." } ] }, { "model_id": "kimi-k2.6", "benchmark_id": "mcpatlas", "score": 66.6, "reference_url": "https://huggingface.co/deepseek-ai/DeepSeek-V4-Pro", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "agentic", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "DeepSeek V4-Pro model card table: Kimi K2.6 Thinking scores 66.6 on MCPAtlas Public (Pass@1). This establishes the exact public benchmark variant and metric, but not the complete Kimi K2.6 max-effort, temperature, context, and harness setting.", "candidates": [ { "score": 69.4, "reference_url": "https://huggingface.co/moonshotai/Kimi-K2.7-Code", "source_type": "model_card", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "benchmark-specified", "sampling": "avg-of-3", "judge": "official MCP-Atlas claims-based judge", "harness": "official MCP-Atlas; 100 tool calls; 32K max tokens per step", "prompt_style": "default", "temperature": "1.0; top_p=0.95", "context": "262144 (256K)" }, "notes": "Kimi K2.7 Code model card Evaluation Results table: 69.4. This source uses top_p=0.95; the existing Kimi K2.6 canonical setting uses top_p=1.0. The source uses 100 tool calls and 32K max tokens per step and does not identify whether the public 500-task subset was used; the current canonical row uses the public release and maxTurns=20." } ] }, { "model_id": "gpt-5.4", "benchmark_id": "mcpatlas", "score": 70.6, "reference_url": "https://openai.com/index/introducing-gpt-5-5/", "reported_setting": { "mode": "thinking", "effort": "xhigh", "tools": "agentic", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-5.5 blog table: 70.6.", "candidates": [ { "score": 67.2, "reference_url": "https://huggingface.co/deepseek-ai/DeepSeek-V4-Pro", "source_type": "third_party", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "agentic", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "notes": "DeepSeek V4-Pro model card (third-party): gpt-5.4 (per their own canonical config) = 67.2." }, { "score": 67.2, "reference_url": "https://z.ai/blog/glm-5.1", "source_type": "official_blog", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "agentic", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "notes": "GLM-5.1 blog: gpt-5.4=67.2." } ] }, { "model_id": "gpt-5.5", "benchmark_id": "mcpatlas", "score": 75.3, "reference_url": "https://openai.com/index/introducing-gpt-5-5/", "reported_setting": { "mode": "thinking", "effort": "xhigh", "tools": "agentic", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-5.5 blog table: 75.3.", "candidates": [ { "score": 79.4, "reference_url": "https://huggingface.co/moonshotai/Kimi-K2.7-Code", "source_type": "model_card", "reported_setting": { "mode": "thinking", "effort": "xhigh", "tools": "benchmark-specified", "sampling": "avg-of-3", "judge": "official MCP-Atlas claims-based judge", "harness": "official MCP-Atlas; 100 tool calls; 32K max tokens per step", "prompt_style": "default", "temperature": "default", "context": "default" }, "notes": "Kimi K2.7 Code model card Evaluation Results table: 79.4. Cross-model value reported by Moonshot AI. The source uses 100 tool calls and 32K max tokens per step and does not identify whether the public 500-task subset was used; the current canonical row uses the public release and maxTurns=20." }, { "score": 82.8, "reference_url": "https://github.com/MoonshotAI/Kimi-K3/blob/main/k3_tech_report.pdf", "source_type": "tech_report", "reported_setting": { "mode": "thinking", "effort": "xhigh", "tools": "benchmark-specified", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "public 500; maxTurns=100; Gemini 3.1 Pro judge", "prompt_style": "default", "temperature": "default", "context": "default" }, "notes": "Kimi K3 technical report: MCP-Atlas = 82.8. Source setting: public 500; maxTurns=100; Gemini 3.1 Pro judge. Origin: Moonshot evaluation." }, { "score": 81.6, "reference_url": "https://huggingface.co/tencent/Hy3", "source_type": "model_card", "reported_setting": { "mode": "thinking", "effort": "high / highest available", "tools": "MCP tool servers", "sampling": "pass@1", "judge": "Gemini 2.5 Pro", "harness": "Scale April 2026 methodology; public 500; 100 tool-call budget", "prompt_style": "default", "temperature": "0.9 for Hy3 family; otherwise source/provider setting", "context": "256000 for Hy3 family; otherwise source/provider setting", "notes": "Comparator value not marked as Tencent own test" }, "notes": "Hy3 official model-card appendix: MCP atlas (public) = 81.6. Comparator-reported value." }, { "score": 82.9, "reference_url": "https://huggingface.co/tencent/Hy3", "source_type": "model_card", "reported_setting": { "mode": "thinking", "effort": "high / highest available", "tools": "MCP tool servers", "sampling": "pass@1", "judge": "Gemini 2.5 Pro", "harness": "Scale April 2026 methodology; public 500; 100 tool-call budget", "prompt_style": "default", "temperature": "0.9 for Hy3 family; otherwise source/provider setting", "context": "256000 for Hy3 family; otherwise source/provider setting", "notes": "Tencent own test" }, "notes": "Hy3 official model-card appendix: MCP atlas (public) = 82.9*. Tencent own testing." }, { "score": 81.6, "reference_url": "https://seed.bytedance.com/en/blog/seed2-1-officially-released-advancing-ai-productivity", "source_type": "official_blog", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "MCP servers in an agent harness", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official", "prompt_style": "official default", "temperature": "unknown" }, "notes": "Release blog EN: Daily-life/tool-use chart (https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/user-upload/4xfa4mqq4ujsp.png)." } ] }, { "model_id": "claude-opus-4.6", "benchmark_id": "mcpatlas", "score": 75.8, "reference_url": "https://www.anthropic.com/news/claude-opus-4-7", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "agentic", "sampling": "pass@1 (avg 5 trials)", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "Anthropic Opus 4.7 blog table: claude-opus-4.6 = 75.8%.", "candidates": [ { "score": 59.5, "reference_url": "https://www.anthropic.com/news/claude-sonnet-4-6", "source_type": "third_party", "reported_setting": { "note": "see primary" }, "notes": "Sonnet 4.6 blog reports 59.5 (vs primary verified=75.8 from Opus 4.7 blog)." }, { "score": 59.5, "reference_url": "https://deepmind.google/models/gemini/pro/", "source_type": "official_blog", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "notes": "DeepMind /models/gemini/pro/ cross-model Performance table: MCP Atlas 59.5%" }, { "score": 73.8, "reference_url": "https://huggingface.co/deepseek-ai/DeepSeek-V4-Pro", "source_type": "third_party", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "agentic", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "notes": "DeepSeek V4-Pro model card (third-party): claude-opus-4.6 (per their own canonical config) = 73.8." }, { "score": 73.8, "reference_url": "https://z.ai/blog/glm-5.1", "source_type": "official_blog", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "agentic", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "notes": "GLM-5.1 blog: claude-opus-4.6=73.8." } ] }, { "model_id": "claude-opus-4.7", "benchmark_id": "mcpatlas", "score": 77.3, "reference_url": "https://www.anthropic.com/news/claude-opus-4-7", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "agentic", "sampling": "pass@1 (avg 5 trials)", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "Anthropic Opus 4.7 blog table: claude-opus-4.7 = 77.3%.", "candidates": [ { "score": 79.1, "reference_url": "https://openai.com/index/introducing-gpt-5-5/", "source_type": "third_party", "reported_setting": { "note": "see primary" }, "notes": "OpenAI GPT-5.5 blog reports 79.1 (vs primary 77.3 from Anthropic Opus 4.7)." }, { "score": 79.1, "reference_url": "https://seed.bytedance.com/en/blog/seed2-1-officially-released-advancing-ai-productivity", "source_type": "official_blog", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "MCP servers in an agent harness", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official", "prompt_style": "official default", "temperature": "unknown" }, "notes": "Release blog EN: Daily-life/tool-use chart (https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/user-upload/4xfa4mqq4ujsp.png)." } ] }, { "model_id": "gemini-3.1-pro", "benchmark_id": "mcpatlas", "score": 69.2, "reference_url": "https://deepmind.google/models/gemini/pro/", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "DeepMind /models/gemini/pro/ Performance table (PROMOTED over prior third-party): MCP Atlas 69.2% (was 78.2 from OpenAI gpt-5.5 blog third-party; demoted to candidate)", "candidates": [ { "score": 78.2, "reference_url": "https://openai.com/index/introducing-gpt-5-5/", "source_type": "third_party", "reported_setting": {}, "notes": "Demoted from primary on https://deepmind.google/models/gemini/pro/ audit. Original notes: " }, { "score": 82.6, "reference_url": "https://huggingface.co/tencent/Hy3", "source_type": "model_card", "reported_setting": { "mode": "thinking", "effort": "high / highest available", "tools": "MCP tool servers", "sampling": "pass@1", "judge": "Gemini 2.5 Pro", "harness": "Scale April 2026 methodology; public 500; 100 tool-call budget", "prompt_style": "default", "temperature": "0.9 for Hy3 family; otherwise source/provider setting", "context": "256000 for Hy3 family; otherwise source/provider setting", "notes": "Tencent own test" }, "notes": "Hy3 official model-card appendix: MCP atlas (public) = 82.6*. Tencent own testing." }, { "score": 78.2, "reference_url": "https://seed.bytedance.com/en/blog/seed2-1-officially-released-advancing-ai-productivity", "source_type": "official_blog", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "MCP servers in an agent harness", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official", "prompt_style": "official default", "temperature": "unknown" }, "notes": "Release blog EN: Daily-life/tool-use chart (https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/user-upload/4xfa4mqq4ujsp.png)." } ] }, { "model_id": "glm-5.1", "benchmark_id": "mcpatlas", "score": 71.8, "reference_url": "https://z.ai/blog/glm-5.2", "reported_setting": { "mode": "thinking", "effort": "official reported", "sampling": "pass@1", "prompt_style": "default", "tools": "agentic MCP servers", "context": "default", "judge": "Gemini 3.0 Pro", "harness": "public 500-task set; 10-minute timeout per task", "temperature": "default" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "GLM-5.2 official release source: GLM-5.1 / MCP-Atlas (Public Set) = 71.8.", "candidates": [ { "score": 77.7, "reference_url": "https://huggingface.co/tencent/Hy3", "source_type": "model_card", "reported_setting": { "mode": "thinking", "effort": "high / highest available", "tools": "MCP tool servers", "sampling": "pass@1", "judge": "Gemini 2.5 Pro", "harness": "Scale April 2026 methodology; public 500; 100 tool-call budget", "prompt_style": "default", "temperature": "0.9 for Hy3 family; otherwise source/provider setting", "context": "256000 for Hy3 family; otherwise source/provider setting", "notes": "Tencent own test" }, "notes": "Hy3 official model-card appendix: MCP atlas (public) = 77.7*. Tencent own testing." } ] }, { "model_id": "deepseek-v4-pro", "benchmark_id": "toolathlon", "score": 51.8, "reference_url": "https://huggingface.co/deepseek-ai/DeepSeek-V4-Pro", "reported_setting": { "mode": "thinking", "effort": "Max", "tools": "agentic", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "DeepSeek V4 model card: deepseek-v4-pro (Max mode) = 51.8.", "candidates": [ { "score": 52.8, "reference_url": "https://z.ai/blog/glm-5.2", "source_type": "official_blog", "reported_setting": { "mode": "thinking", "effort": "Max", "sampling": "pass@1", "prompt_style": "default", "tools": "agentic tool use", "context": "default", "max_output_tokens": "128000", "judge": "official evaluation service", "harness": "original Toolathlon public evaluation service (pre-Verified)", "temperature": "default" }, "notes": "GLM-5.2 cross-table alternative; DeepSeek first-party 51.8 remains primary." }, { "score": 45.7, "reference_url": "https://huggingface.co/tencent/Hy3", "source_type": "model_card", "reported_setting": { "mode": "thinking", "effort": "high / highest available", "tools": "agentic tool use", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "original Toolathlon official evaluation", "prompt_style": "default", "temperature": "0.9 for Hy3 family; otherwise source/provider setting", "context": "256000 for Hy3 family; otherwise source/provider setting", "notes": "Tencent own test" }, "notes": "Hy3 official model-card appendix: Toolathlon = 45.7*. Tencent own testing." } ] }, { "model_id": "deepseek-v4-flash", "benchmark_id": "toolathlon", "score": 47.8, "reference_url": "https://huggingface.co/deepseek-ai/DeepSeek-V4-Pro", "reported_setting": { "mode": "thinking", "effort": "Max", "tools": "agentic", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "DeepSeek V4 model card: deepseek-v4-flash (Max mode) = 47.8.", "candidates": [ { "score": 47.8, "reference_url": "https://llm-stats.com/benchmarks/toolathlon", "source_type": "third_party_aggregator", "reported_setting": { "effort": "max" }, "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=DeepSeek-V4-Flash-Max, slug=deepseek-v4-flash-max, provider=DeepSeek" }, { "score": 43.8, "reference_url": "https://huggingface.co/tencent/Hy3", "source_type": "model_card", "reported_setting": { "mode": "thinking", "effort": "high / highest available", "tools": "agentic tool use", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "original Toolathlon official evaluation", "prompt_style": "default", "temperature": "0.9 for Hy3 family; otherwise source/provider setting", "context": "256000 for Hy3 family; otherwise source/provider setting", "notes": "Tencent own test" }, "notes": "Hy3 official model-card appendix: Toolathlon = 43.8*. Tencent own testing." } ] }, { "model_id": "kimi-k2.6", "benchmark_id": "toolathlon", "score": 50.0, "reference_url": "https://huggingface.co/deepseek-ai/DeepSeek-V4-Pro", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "agentic", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "DeepSeek V4-Pro model card (third-party): kimi-k2.6 (per their own canonical config) = 50.0." }, { "model_id": "gpt-5.4", "benchmark_id": "toolathlon", "score": 54.6, "reference_url": "https://openai.com/index/introducing-gpt-5-5/", "reported_setting": { "mode": "thinking", "effort": "xhigh", "tools": "agentic", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-5.5 blog table: 54.6." }, { "model_id": "gpt-5.5", "benchmark_id": "toolathlon", "score": 55.6, "reference_url": "https://openai.com/index/introducing-gpt-5-5/", "reported_setting": { "mode": "thinking", "effort": "xhigh", "tools": "agentic", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-5.5 blog table: 55.6.", "candidates": [ { "score": 55.6, "reference_url": "https://seed.bytedance.com/en/blog/seed2-1-officially-released-advancing-ai-productivity", "source_type": "official_blog", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "agentic tool use", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official", "prompt_style": "official default", "temperature": "unknown" }, "notes": "Release blog EN: Daily-life/tool-use chart (https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/user-upload/4xfa4mqq4ujsp.png)." } ] }, { "model_id": "claude-opus-4.6", "benchmark_id": "toolathlon", "score": 47.2, "reference_url": "https://huggingface.co/deepseek-ai/DeepSeek-V4-Pro", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "agentic", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "DeepSeek V4-Pro model card (third-party): claude-opus-4.6 (per their own canonical config) = 47.2.", "candidates": [ { "score": 56.8, "reference_url": "https://www.anthropic.com/claude-sonnet-5-system-card", "source_type": "model_card", "reported_setting": { "effort": "max", "tools": "agentic benchmark harness", "sampling": "3 trials/task", "harness": "toolathlon" }, "notes": "Anthropic Claude Sonnet 5 System Card Table 8.11.5.A, page 135: Claude Opus 4.6; Toolathlon [Pass@1]=56.8. Source setting: effort=max; tools=agentic benchmark harness; sampling=3 trials/task; harness=toolathlon." } ] }, { "model_id": "gemini-3.1-pro", "benchmark_id": "toolathlon", "score": 48.8, "reference_url": "https://huggingface.co/deepseek-ai/DeepSeek-V4-Pro", "candidates": [ { "score": 48.8, "reference_url": "https://seed.bytedance.com/en/blog/seed2-1-officially-released-advancing-ai-productivity", "source_type": "official_blog", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "agentic tool use", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official", "prompt_style": "official default", "temperature": "unknown" }, "notes": "Release blog EN: Daily-life/tool-use chart (https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/user-upload/4xfa4mqq4ujsp.png)." } ], "reported_setting": { "mode": "thinking", "effort": "high", "tools": "agentic", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "DeepSeek V4-Pro model card (third-party): gemini-3.1-pro (per their own canonical config) = 48.8." }, { "model_id": "kimi-k2.5", "benchmark_id": "toolathlon", "score": 27.8, "reference_url": "https://z.ai/blog/glm-5.1", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "agentic", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "GLM-5.1 blog: kimi-k2.5=27.8." }, { "model_id": "glm-5.1", "benchmark_id": "toolathlon", "score": 40.7, "reference_url": "https://z.ai/blog/glm-5.2", "reported_setting": { "mode": "thinking", "effort": "official reported", "sampling": "pass@1", "prompt_style": "default", "tools": "agentic tool use", "context": "default", "max_output_tokens": "128000", "judge": "official evaluation service", "harness": "original Toolathlon public evaluation service (pre-Verified)", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "GLM-5.2 official release source: GLM-5.1 / Tool-Decathlon = 40.7.", "candidates": [ { "score": 41.4, "reference_url": "https://huggingface.co/tencent/Hy3", "source_type": "model_card", "reported_setting": { "mode": "thinking", "effort": "high / highest available", "tools": "agentic tool use", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "original Toolathlon official evaluation", "prompt_style": "default", "temperature": "0.9 for Hy3 family; otherwise source/provider setting", "context": "256000 for Hy3 family; otherwise source/provider setting", "notes": "Tencent own test" }, "notes": "Hy3 official model-card appendix: Toolathlon = 41.4*. Tencent own testing." } ] }, { "model_id": "deepseek-v4-pro", "benchmark_id": "simpleqa_verified", "score": 57.9, "reference_url": "https://huggingface.co/deepseek-ai/DeepSeek-V4-Pro", "reported_setting": { "mode": "thinking", "effort": "Max", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "DeepSeek V4 model card: deepseek-v4-pro (Max mode) = 57.9.", "candidates": [ { "score": 46.6, "reference_url": "https://huggingface.co/internlm/Intern-S2-Preview-397B", "source_type": "model_card", "reported_setting": { "mode": "source does not state", "effort": "source does not state", "tools": "none", "sampling": "source does not state", "judge": "benchmark-specific; exact version not stated", "harness": "OpenCompass or AgentCompass; exact adapter not stated", "prompt_style": "source does not state", "temperature": "source does not state", "context": "source does not state", "notes": "Shanghai AI Lab evaluated all model columns using OpenCompass, VLMEvalKit, and AgentCompass. Per-model mode/effort/sampling are not reported; only Intern-S2 maximum evaluation lengths are disclosed." }, "notes": "Intern-S2-Preview-397B official model card: SimpleQA-Verified = 46.6. This is a Shanghai AI Lab cross-model evaluation; the source does not disclose this model's exact mode, effort, or sampling." } ] }, { "model_id": "deepseek-v4-flash", "benchmark_id": "simpleqa_verified", "score": 34.1, "reference_url": "https://huggingface.co/deepseek-ai/DeepSeek-V4-Pro", "reported_setting": { "mode": "thinking", "effort": "Max", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "DeepSeek V4 model card: deepseek-v4-flash (Max mode) = 34.1." }, { "model_id": "qwen3.5-397b", "benchmark_id": "hmmt_feb_2026", "score": 87.88, "reference_url": "https://huggingface.co/datasets/MathArena/hmmt_feb_2026", "audit_status": "verified", "matches_canonical": true, "reported_setting": "thinking:default", "source_type": "leaderboard" }, { "model_id": "deepseek-v3.2", "benchmark_id": "hmmt_feb_2026", "score": 79.9, "reference_url": "https://z.ai/blog/glm-5.1", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "GLM-5.1 blog: deepseek-v3.2=79.9.", "candidates": [] }, { "model_id": "qwen3-30b-a3b", "benchmark_id": "hmmt_feb_2026", "score": 78.79, "reference_url": "https://huggingface.co/datasets/MathArena/hmmt_feb_2026", "audit_status": "dropped", "notes": " | DROPPED: BP value from matharena 2026 dataset; BP model release_date=2025-05-15. Wrong-variant." }, { "model_id": "deepseek-v3.2", "benchmark_id": "aime_2026", "score": 95.1, "reference_url": "https://z.ai/blog/glm-5.1", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "GLM-5.1 blog: deepseek-v3.2=95.1." }, { "model_id": "qwen3-30b-a3b", "benchmark_id": "aime_2026", "score": 87.5, "reference_url": "https://huggingface.co/datasets/MathArena/aime_2026", "audit_status": "dropped", "notes": " | DROPPED: BP value from matharena 2026 dataset; BP model release_date=2025-05-15. Wrong-variant." }, { "model_id": "kimi-k2.5", "benchmark_id": "aime_2026", "score": 94.5, "reference_url": "https://z.ai/blog/glm-5.1", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "GLM-5.1 blog: kimi-k2.5=94.5.", "candidates": [ { "score": 95.8, "reference_url": "https://huggingface.co/moonshotai/Kimi-K2.6", "source_type": "model_card", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "1.0" }, "notes": "Kimi K2.6 model card: kimi-k2.5=95.8." }, { "score": 93.3, "reference_url": "https://huggingface.co/Qwen/Qwen3.5-397B-A17B", "source_type": "model_card", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "notes": "Qwen3.5-397B-A17B HF model card cross-model table (Qwen self-test of 3rd-party model, mc=false): alternate measurement 93.3 (3rd-party Qwen self-test)." } ] }, { "model_id": "gemma-4-31b", "benchmark_id": "mmlu_pro", "score": 85.2, "reference_url": "https://arxiv.org/abs/2607.02770", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "provider-reported", "judge": "benchmark-specified", "harness": "official Google evaluation", "prompt_style": "official", "temperature": "not disclosed", "context": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Gemma 4 Technical Report Table 5; Gemma 4 thinking unless explicitly stated, Gemma 3 27B non-thinking." }, { "model_id": "gemma-4-31b", "benchmark_id": "aime_2026", "score": 89.2, "reference_url": "https://blog.google/innovation-and-ai/technology/developers-tools/gemma-4/", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "provider-reported", "judge": "benchmark-specified", "harness": "official Google evaluation", "prompt_style": "official", "temperature": "not disclosed", "context": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "Google Gemma 4 family launch table, duplicated by the DeepMind semantic table." }, { "model_id": "gemma-4-31b", "benchmark_id": "gpqa_diamond", "score": 84.3, "reference_url": "https://blog.google/innovation-and-ai/technology/developers-tools/gemma-4/", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "provider-reported", "judge": "benchmark-specified", "harness": "official Google evaluation", "prompt_style": "official", "temperature": "not disclosed", "context": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "Google Gemma 4 family launch table, duplicated by the DeepMind semantic table." }, { "model_id": "gemma-4-31b", "benchmark_id": "mmmlu", "score": 88.4, "reference_url": "https://arxiv.org/abs/2607.02770", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "provider-reported", "judge": "benchmark-specified", "harness": "official Google evaluation", "prompt_style": "official", "temperature": "not disclosed", "context": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Gemma 4 Technical Report Table 5; Gemma 4 thinking unless explicitly stated, Gemma 3 27B non-thinking." }, { "model_id": "gemma-4-31b", "benchmark_id": "codeforces_rating", "score": 2150, "reference_url": "https://arxiv.org/abs/2607.02770", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "code execution in undisclosed harness", "sampling": "provider-reported", "judge": "benchmark-specified", "harness": "official Google evaluation", "prompt_style": "official", "temperature": "not disclosed", "context": "default" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Gemma 4 Technical Report Table 5; Gemma 4 thinking unless explicitly stated, Gemma 3 27B non-thinking." }, { "model_id": "gemma-4-31b", "benchmark_id": "hle", "score": 19.5, "reference_url": "https://arxiv.org/abs/2607.02770", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "provider-reported", "judge": "benchmark-specified", "harness": "official Google evaluation", "prompt_style": "official", "temperature": "not disclosed", "context": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Gemma 4 Technical Report Table 5; Gemma 4 thinking unless explicitly stated, Gemma 3 27B non-thinking.", "candidates": [ { "score": 26.5, "reference_url": "https://huggingface.co/blog/gemma4", "source_type": "official_blog", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "web search", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "notes": "Gemma 4 HF blog: HLE with search 26.5 (alt tool setting; canonical = no-tools)." } ] }, { "model_id": "gemma-4-31b", "benchmark_id": "mmmu_pro", "score": 76.9, "reference_url": "https://blog.google/innovation-and-ai/technology/developers-tools/gemma-4/", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "provider-reported", "judge": "benchmark-specified", "harness": "official Google evaluation", "prompt_style": "official", "temperature": "not disclosed", "context": "default", "visual_tokens": 1120 }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "Google Gemma 4 family launch table, duplicated by the DeepMind semantic table.", "candidates": [ { "score": 75.8, "reference_url": "https://arxiv.org/abs/2607.02770", "source_type": "tech_report", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "provider-reported", "judge": "benchmark-specified", "harness": "official Google evaluation", "prompt_style": "official", "temperature": "not disclosed", "context": "default", "visual_tokens": 280 }, "notes": "Gemma 4 Technical Report Appendix Table 12; thinking with 280 visual tokens." } ] }, { "model_id": "gemma-4-31b", "benchmark_id": "mathvision", "score": 85.6, "reference_url": "https://arxiv.org/abs/2607.02770", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "provider-reported", "judge": "benchmark-specified", "harness": "official Google evaluation", "prompt_style": "official", "temperature": "not disclosed", "context": "default", "visual_tokens": 1120 }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Gemma 4 Technical Report Table 6; maximum supported resolution, 1,120 visual tokens.", "candidates": [ { "score": 83.4, "reference_url": "https://arxiv.org/abs/2607.02770", "source_type": "tech_report", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "provider-reported", "judge": "benchmark-specified", "harness": "official Google evaluation", "prompt_style": "official", "temperature": "not disclosed", "context": "default", "visual_tokens": 280 }, "notes": "Gemma 4 Technical Report Appendix Table 12; thinking with 280 visual tokens." } ] }, { "model_id": "gemma-4-26b-a4b", "benchmark_id": "mmlu_pro", "score": 82.6, "reference_url": "https://arxiv.org/abs/2607.02770", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "provider-reported", "judge": "benchmark-specified", "harness": "official Google evaluation", "prompt_style": "official", "temperature": "not disclosed", "context": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Gemma 4 Technical Report Table 5; Gemma 4 thinking unless explicitly stated, Gemma 3 27B non-thinking.", "candidates": [ { "score": 82.6, "reference_url": "https://arxiv.org/abs/2608.00146", "source_type": "tech_report", "reported_setting": { "mode": "non-thinking", "decoding": "AR MTP", "tools": "none", "sampling": "replicate count N undisclosed", "judge": "benchmark-specified", "harness": "official Google cross-model evaluation", "prompt_style": "official provider chat template", "context": "default" }, "notes": "DiffusionGemma Technical Report Table 3; replicate count N is undisclosed." }, { "score": 84.42, "reference_url": "https://huggingface.co/internlm/Intern-S2-Preview/resolve/3b53ea8d2f8ed46abefa507e3f6f56ee0c662c4a/figs/performance.png", "source_type": "official_model_card", "reported_setting": { "table_framework_claim": "OpenCompass and VLMEvalKit used to evaluate all models", "exact_per_cell_settings": "not stated" }, "notes": "Displayed exactly: 84.42" } ] }, { "model_id": "gemma-4-26b-a4b", "benchmark_id": "aime_2026", "score": 88.3, "reference_url": "https://blog.google/innovation-and-ai/technology/developers-tools/gemma-4/", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "provider-reported", "judge": "benchmark-specified", "harness": "official Google evaluation", "prompt_style": "official", "temperature": "not disclosed", "context": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "Google Gemma 4 family launch table, duplicated by the DeepMind semantic table.", "candidates": [ { "score": 80.0, "reference_url": "https://arxiv.org/abs/2608.00146", "source_type": "tech_report", "reported_setting": { "mode": "non-thinking", "decoding": "AR MTP", "tools": "none", "sampling": "replicate count N undisclosed", "judge": "benchmark-specified", "harness": "official Google cross-model evaluation", "prompt_style": "official provider chat template", "context": "default" }, "notes": "DiffusionGemma Technical Report Table 3; replicate count N is undisclosed." }, { "score": 72.0, "reference_url": "https://www.liquid.ai/blog/lfm2-5-8b-a1b", "source_type": "official_blog", "reported_setting": { "mode": "source/model default", "effort": "source does not state", "tools": "none", "sampling": "source does not state trial count", "judge": "benchmark-specified", "harness": "Liquid AI official comparative evaluation", "prompt_style": "source does not state", "temperature": "source does not state evaluation parameters", "context": "source does not state evaluation context" }, "notes": "Displayed exactly as 72.00. Official Liquid AI reported result." } ] }, { "model_id": "gemma-4-26b-a4b", "benchmark_id": "gpqa_diamond", "score": 82.3, "reference_url": "https://blog.google/innovation-and-ai/technology/developers-tools/gemma-4/", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "provider-reported", "judge": "benchmark-specified", "harness": "official Google evaluation", "prompt_style": "official", "temperature": "not disclosed", "context": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "Google Gemma 4 family launch table, duplicated by the DeepMind semantic table.", "candidates": [ { "score": 73.7, "reference_url": "https://arxiv.org/abs/2608.00146", "source_type": "tech_report", "reported_setting": { "mode": "non-thinking", "decoding": "AR MTP", "tools": "none", "sampling": "replicate count N undisclosed", "judge": "benchmark-specified", "harness": "official Google cross-model evaluation", "prompt_style": "official provider chat template", "context": "default" }, "notes": "DiffusionGemma Technical Report Table 3; replicate count N is undisclosed." } ] }, { "model_id": "gemma-4-26b-a4b", "benchmark_id": "mmmlu", "score": 86.3, "reference_url": "https://arxiv.org/abs/2607.02770", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "provider-reported", "judge": "benchmark-specified", "harness": "official Google evaluation", "prompt_style": "official", "temperature": "not disclosed", "context": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Gemma 4 Technical Report Table 5; Gemma 4 thinking unless explicitly stated, Gemma 3 27B non-thinking.", "candidates": [ { "score": 78.0, "reference_url": "https://arxiv.org/abs/2608.00146", "source_type": "tech_report", "reported_setting": { "mode": "non-thinking", "decoding": "AR MTP", "tools": "none", "sampling": "replicate count N undisclosed", "judge": "benchmark-specified", "harness": "official Google cross-model evaluation", "prompt_style": "official provider chat template", "context": "default" }, "notes": "DiffusionGemma Technical Report Table 3; replicate count N is undisclosed." } ] }, { "model_id": "gemma-4-26b-a4b", "benchmark_id": "codeforces_rating", "score": 1718, "reference_url": "https://arxiv.org/abs/2607.02770", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "code execution in undisclosed harness", "sampling": "provider-reported", "judge": "benchmark-specified", "harness": "official Google evaluation", "prompt_style": "official", "temperature": "not disclosed", "context": "default" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Gemma 4 Technical Report Table 5; Gemma 4 thinking unless explicitly stated, Gemma 3 27B non-thinking." }, { "model_id": "gemma-4-26b-a4b", "benchmark_id": "hle", "score": 8.7, "reference_url": "https://arxiv.org/abs/2607.02770", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "provider-reported", "judge": "benchmark-specified", "harness": "official Google evaluation", "prompt_style": "official", "temperature": "not disclosed", "context": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Gemma 4 Technical Report Table 5; Gemma 4 thinking unless explicitly stated, Gemma 3 27B non-thinking.", "candidates": [ { "score": 17.2, "reference_url": "https://huggingface.co/blog/gemma4", "source_type": "official_blog", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "web search", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "notes": "Gemma 4 HF blog: HLE with search 17.2 (alt tool setting; canonical = no-tools)." } ] }, { "model_id": "gemma-4-26b-a4b", "benchmark_id": "mmmu_pro", "score": 73.8, "reference_url": "https://blog.google/innovation-and-ai/technology/developers-tools/gemma-4/", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "provider-reported", "judge": "benchmark-specified", "harness": "official Google evaluation", "prompt_style": "official", "temperature": "not disclosed", "context": "default", "visual_tokens": 1120 }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "Google Gemma 4 family launch table, duplicated by the DeepMind semantic table.", "candidates": [ { "score": 73.2, "reference_url": "https://arxiv.org/abs/2607.02770", "source_type": "tech_report", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "provider-reported", "judge": "benchmark-specified", "harness": "official Google evaluation", "prompt_style": "official", "temperature": "not disclosed", "context": "default", "visual_tokens": 280 }, "notes": "Gemma 4 Technical Report Appendix Table 12; thinking with 280 visual tokens." }, { "score": 72.5, "reference_url": "https://arxiv.org/abs/2608.00146", "source_type": "tech_report", "reported_setting": { "mode": "non-thinking", "decoding": "AR MTP", "tools": "none", "sampling": "replicate count N undisclosed", "judge": "benchmark-specified", "harness": "official Google cross-model evaluation", "prompt_style": "official provider chat template", "context": "default", "visual_tokens": 1120 }, "notes": "DiffusionGemma Technical Report Table 3; replicate count N is undisclosed." }, { "score": 73.82, "reference_url": "https://huggingface.co/internlm/Intern-S2-Preview/resolve/3b53ea8d2f8ed46abefa507e3f6f56ee0c662c4a/figs/performance.png", "source_type": "official_model_card", "reported_setting": { "table_framework_claim": "OpenCompass and VLMEvalKit used to evaluate all models", "exact_per_cell_settings": "not stated" }, "notes": "Displayed exactly: 73.82" } ] }, { "model_id": "gemma-4-26b-a4b", "benchmark_id": "mathvision", "score": 82.4, "reference_url": "https://arxiv.org/abs/2607.02770", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "provider-reported", "judge": "benchmark-specified", "harness": "official Google evaluation", "prompt_style": "official", "temperature": "not disclosed", "context": "default", "visual_tokens": 1120 }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Gemma 4 Technical Report Table 6; maximum supported resolution, 1,120 visual tokens.", "candidates": [ { "score": 80.3, "reference_url": "https://arxiv.org/abs/2607.02770", "source_type": "tech_report", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "provider-reported", "judge": "benchmark-specified", "harness": "official Google evaluation", "prompt_style": "official", "temperature": "not disclosed", "context": "default", "visual_tokens": 280 }, "notes": "Gemma 4 Technical Report Appendix Table 12; thinking with 280 visual tokens." }, { "score": 82.4, "reference_url": "https://huggingface.co/internlm/Intern-S2-Preview/resolve/3b53ea8d2f8ed46abefa507e3f6f56ee0c662c4a/figs/performance.png", "source_type": "official_model_card", "reported_setting": { "table_framework_claim": "OpenCompass and VLMEvalKit used to evaluate all models", "exact_per_cell_settings": "not stated" }, "notes": "Displayed exactly: 82.40" } ] }, { "model_id": "gemma-4-e4b", "benchmark_id": "mmlu_pro", "score": 69.4, "reference_url": "https://arxiv.org/abs/2607.02770", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "provider-reported", "judge": "benchmark-specified", "harness": "official Google evaluation", "prompt_style": "official", "temperature": "not disclosed", "context": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Gemma 4 Technical Report Table 5; Gemma 4 thinking unless explicitly stated, Gemma 3 27B non-thinking." }, { "model_id": "gemma-4-e4b", "benchmark_id": "aime_2026", "score": 42.5, "reference_url": "https://blog.google/innovation-and-ai/technology/developers-tools/gemma-4/", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "provider-reported", "judge": "benchmark-specified", "harness": "official Google evaluation", "prompt_style": "official", "temperature": "not disclosed", "context": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "Google Gemma 4 family launch table, duplicated by the DeepMind semantic table.", "candidates": [ { "score": 40.67, "reference_url": "https://www.liquid.ai/blog/lfm2-5-8b-a1b", "source_type": "official_blog", "reported_setting": { "mode": "source/model default", "effort": "source does not state", "tools": "none", "sampling": "source does not state trial count", "judge": "benchmark-specified", "harness": "Liquid AI official comparative evaluation", "prompt_style": "source does not state", "temperature": "source does not state evaluation parameters", "context": "source does not state evaluation context" }, "notes": "Displayed exactly as 40.67. Official Liquid AI reported result." } ] }, { "model_id": "gemma-4-e4b", "benchmark_id": "gpqa_diamond", "score": 58.6, "reference_url": "https://blog.google/innovation-and-ai/technology/developers-tools/gemma-4/", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "provider-reported", "judge": "benchmark-specified", "harness": "official Google evaluation", "prompt_style": "official", "temperature": "not disclosed", "context": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "Google Gemma 4 family launch table, duplicated by the DeepMind semantic table." }, { "model_id": "gemma-4-e4b", "benchmark_id": "mmmlu", "score": 76.6, "reference_url": "https://arxiv.org/abs/2607.02770", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "provider-reported", "judge": "benchmark-specified", "harness": "official Google evaluation", "prompt_style": "official", "temperature": "not disclosed", "context": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Gemma 4 Technical Report Table 5; Gemma 4 thinking unless explicitly stated, Gemma 3 27B non-thinking." }, { "model_id": "gemma-4-e4b", "benchmark_id": "codeforces_rating", "score": 940, "reference_url": "https://arxiv.org/abs/2607.02770", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "code execution in undisclosed harness", "sampling": "provider-reported", "judge": "benchmark-specified", "harness": "official Google evaluation", "prompt_style": "official", "temperature": "not disclosed", "context": "default" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Gemma 4 Technical Report Table 5; Gemma 4 thinking unless explicitly stated, Gemma 3 27B non-thinking." }, { "model_id": "gemma-4-e4b", "benchmark_id": "mmmu_pro", "score": 52.6, "reference_url": "https://blog.google/innovation-and-ai/technology/developers-tools/gemma-4/", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "provider-reported", "judge": "benchmark-specified", "harness": "official Google evaluation", "prompt_style": "official", "temperature": "not disclosed", "context": "default", "visual_tokens": 1120 }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "Google Gemma 4 family launch table, duplicated by the DeepMind semantic table.", "candidates": [ { "score": 51.4, "reference_url": "https://arxiv.org/abs/2607.02770", "source_type": "tech_report", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "provider-reported", "judge": "benchmark-specified", "harness": "official Google evaluation", "prompt_style": "official", "temperature": "not disclosed", "context": "default", "visual_tokens": 280 }, "notes": "Gemma 4 Technical Report Appendix Table 12; thinking with 280 visual tokens." } ] }, { "model_id": "gemma-4-e4b", "benchmark_id": "mathvision", "score": 59.5, "reference_url": "https://arxiv.org/abs/2607.02770", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "provider-reported", "judge": "benchmark-specified", "harness": "official Google evaluation", "prompt_style": "official", "temperature": "not disclosed", "context": "default", "visual_tokens": 1120 }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Gemma 4 Technical Report Table 6; maximum supported resolution, 1,120 visual tokens.", "candidates": [ { "score": 59.2, "reference_url": "https://arxiv.org/abs/2607.02770", "source_type": "tech_report", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "provider-reported", "judge": "benchmark-specified", "harness": "official Google evaluation", "prompt_style": "official", "temperature": "not disclosed", "context": "default", "visual_tokens": 280 }, "notes": "Gemma 4 Technical Report Appendix Table 12; thinking with 280 visual tokens." } ] }, { "model_id": "gemma-4-e2b", "benchmark_id": "mmlu_pro", "score": 60.0, "reference_url": "https://arxiv.org/abs/2607.02770", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "provider-reported", "judge": "benchmark-specified", "harness": "official Google evaluation", "prompt_style": "official", "temperature": "not disclosed", "context": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Gemma 4 Technical Report Table 5; Gemma 4 thinking unless explicitly stated, Gemma 3 27B non-thinking." }, { "model_id": "gemma-4-e2b", "benchmark_id": "aime_2026", "score": 37.5, "reference_url": "https://blog.google/innovation-and-ai/technology/developers-tools/gemma-4/", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "provider-reported", "judge": "benchmark-specified", "harness": "official Google evaluation", "prompt_style": "official", "temperature": "not disclosed", "context": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "Google Gemma 4 family launch table, duplicated by the DeepMind semantic table.", "candidates": [ { "score": 30.0, "reference_url": "https://www.liquid.ai/blog/lfm2-5-8b-a1b", "source_type": "official_blog", "reported_setting": { "mode": "source/model default", "effort": "source does not state", "tools": "none", "sampling": "source does not state trial count", "judge": "benchmark-specified", "harness": "Liquid AI official comparative evaluation", "prompt_style": "source does not state", "temperature": "source does not state evaluation parameters", "context": "source does not state evaluation context" }, "notes": "Displayed exactly as 30. Official Liquid AI reported result." } ] }, { "model_id": "gemma-4-e2b", "benchmark_id": "gpqa_diamond", "score": 43.4, "reference_url": "https://blog.google/innovation-and-ai/technology/developers-tools/gemma-4/", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "provider-reported", "judge": "benchmark-specified", "harness": "official Google evaluation", "prompt_style": "official", "temperature": "not disclosed", "context": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "Google Gemma 4 family launch table, duplicated by the DeepMind semantic table." }, { "model_id": "gemma-4-e2b", "benchmark_id": "mmmlu", "score": 67.4, "reference_url": "https://arxiv.org/abs/2607.02770", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "provider-reported", "judge": "benchmark-specified", "harness": "official Google evaluation", "prompt_style": "official", "temperature": "not disclosed", "context": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Gemma 4 Technical Report Table 5; Gemma 4 thinking unless explicitly stated, Gemma 3 27B non-thinking." }, { "model_id": "gemma-4-e2b", "benchmark_id": "codeforces_rating", "score": 633, "reference_url": "https://arxiv.org/abs/2607.02770", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "code execution in undisclosed harness", "sampling": "provider-reported", "judge": "benchmark-specified", "harness": "official Google evaluation", "prompt_style": "official", "temperature": "not disclosed", "context": "default" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Gemma 4 Technical Report Table 5; Gemma 4 thinking unless explicitly stated, Gemma 3 27B non-thinking." }, { "model_id": "gemma-4-e2b", "benchmark_id": "mmmu_pro", "score": 44.2, "reference_url": "https://blog.google/innovation-and-ai/technology/developers-tools/gemma-4/", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "provider-reported", "judge": "benchmark-specified", "harness": "official Google evaluation", "prompt_style": "official", "temperature": "not disclosed", "context": "default", "visual_tokens": 1120 }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "Google Gemma 4 family launch table, duplicated by the DeepMind semantic table.", "candidates": [ { "score": 43.2, "reference_url": "https://arxiv.org/abs/2607.02770", "source_type": "tech_report", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "provider-reported", "judge": "benchmark-specified", "harness": "official Google evaluation", "prompt_style": "official", "temperature": "not disclosed", "context": "default", "visual_tokens": 280 }, "notes": "Gemma 4 Technical Report Appendix Table 12; thinking with 280 visual tokens." } ] }, { "model_id": "gemma-4-e2b", "benchmark_id": "mathvision", "score": 52.4, "reference_url": "https://arxiv.org/abs/2607.02770", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "provider-reported", "judge": "benchmark-specified", "harness": "official Google evaluation", "prompt_style": "official", "temperature": "not disclosed", "context": "default", "visual_tokens": 1120 }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Gemma 4 Technical Report Table 6; maximum supported resolution, 1,120 visual tokens.", "candidates": [ { "score": 53.0, "reference_url": "https://arxiv.org/abs/2607.02770", "source_type": "tech_report", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "provider-reported", "judge": "benchmark-specified", "harness": "official Google evaluation", "prompt_style": "official", "temperature": "not disclosed", "context": "default", "visual_tokens": 280 }, "notes": "Gemma 4 Technical Report Appendix Table 12; thinking with 280 visual tokens." } ] }, { "model_id": "gpt-5.4", "benchmark_id": "usamo_2026", "score": 95.24, "reference_url": "https://matharena.ai/?comp=usamo--usamo_2026", "audit_status": "verified", "matches_canonical": true, "reported_setting": "reasoning_effort:xhigh", "source_type": "leaderboard" }, { "model_id": "glm-5", "benchmark_id": "hmmt_feb_2026", "score": 82.8, "reference_url": "https://z.ai/blog/glm-5.1", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "GLM-5.1 blog: glm-5=82.8.", "candidates": [] }, { "model_id": "glm-5", "benchmark_id": "aime_2026", "score": 95.4, "reference_url": "https://z.ai/blog/glm-5.1", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "GLM-5.1 blog: glm-5=95.4." }, { "model_id": "glm-5", "benchmark_id": "arc_agi_1", "score": 44.7, "reference_url": "https://arcprize.org/leaderboard", "audit_status": "verified", "source_type": "leaderboard", "matches_canonical": true, "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1 (single-pass)", "judge": "rule-based", "harness": "official (arcprize.org)", "system_type": "CoT" }, "notes": "arcprize.org leaderboard audit: GLM-5 on leaderboard" }, { "model_id": "glm-5", "benchmark_id": "arc_agi_2", "score": 4.9, "reference_url": "https://arcprize.org/leaderboard", "audit_status": "verified", "source_type": "leaderboard", "matches_canonical": true, "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1 (single-pass)", "judge": "rule-based", "harness": "official (arcprize.org)", "system_type": "CoT" }, "notes": "arcprize.org leaderboard audit: GLM-5 on leaderboard" }, { "model_id": "glm-5", "benchmark_id": "usamo_2026", "score": 35.12, "reference_url": "https://matharena.ai/?comp=usamo--usamo_2026", "audit_status": "verified", "matches_canonical": true, "reported_setting": "thinking:default", "source_type": "leaderboard" }, { "model_id": "minimax-m2.5", "benchmark_id": "arc_agi_1", "score": 63.7, "reference_url": "https://arcprize.org/leaderboard", "audit_status": "verified", "source_type": "leaderboard", "matches_canonical": true, "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1 (single-pass)", "judge": "rule-based", "harness": "official (arcprize.org)", "system_type": "CoT" }, "notes": "arcprize.org leaderboard audit: Minimax M2.5 on leaderboard" }, { "model_id": "minimax-m2.5", "benchmark_id": "arc_agi_2", "score": 4.9, "reference_url": "https://arcprize.org/leaderboard", "audit_status": "verified", "source_type": "leaderboard", "matches_canonical": true, "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1 (single-pass)", "judge": "rule-based", "harness": "official (arcprize.org)", "system_type": "CoT" }, "notes": "arcprize.org leaderboard audit: Minimax M2.5 on leaderboard" }, { "model_id": "gemini-3.1-pro", "benchmark_id": "usamo_2026", "score": 74.4, "reference_url": "https://matharena.ai/?comp=usamo--usamo_2026", "audit_status": "verified", "matches_canonical": true, "reported_setting": "thinking:high", "source_type": "leaderboard", "candidates": [ { "score": 61.3, "reference_url": "https://huggingface.co/tencent/Hy3", "source_type": "model_card", "reported_setting": { "mode": "thinking", "effort": "high / highest available", "tools": "none", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official benchmark or reported provider harness", "prompt_style": "default", "temperature": "0.9 for Hy3 family; otherwise source/provider setting", "context": "256000 for Hy3 family; otherwise source/provider setting", "notes": "Tencent own test" }, "notes": "Hy3 official model-card appendix: USAMO 2026 = 61.3*. Tencent own testing." } ] }, { "model_id": "claude-opus-4.6", "benchmark_id": "usamo_2026", "score": 47.02, "reference_url": "https://matharena.ai/?comp=usamo--usamo_2026", "audit_status": "needs_review", "rule_ids": [ "R5g_c" ], "notes": "matharena.ai/usamo/ currently shows only GPT-5.4 and Gemini models; claude-opus-4.6 entry no longer visible. Stale reference. R5g(c)." }, { "model_id": "qwen3.5-397b", "benchmark_id": "usamo_2026", "score": 36.31, "reference_url": "https://matharena.ai/?comp=usamo--usamo_2026", "audit_status": "verified", "matches_canonical": true, "reported_setting": "thinking:default", "source_type": "leaderboard" }, { "model_id": "grok-3-beta", "benchmark_id": "usamo_2025", "score": 4.76, "reference_url": "https://matharena.ai/?comp=usamo--usamo_2025", "audit_status": "verified", "source_type": "leaderboard" }, { "model_id": "gemini-2.0-flash", "benchmark_id": "usamo_2025", "score": 4.17, "reference_url": "https://matharena.ai/?comp=usamo--usamo_2025", "audit_status": "dropped" }, { "model_id": "claude-sonnet-4.6", "benchmark_id": "bullshit_pushback", "score": 94.55, "reference_url": "https://github.com/petergpt/bullshit-benchmark/tree/main/data/latest", "audit_status": "verified", "rule_ids": [ "R5g_c" ], "notes": "GitHub repo source; not yet verified. R5g(c).", "reported_setting": "leaderboard.csv green_rate×100; reasoning=none", "source_type": "leaderboard", "matches_canonical": true }, { "model_id": "claude-opus-4.6", "benchmark_id": "bullshit_pushback", "score": 92.73, "reference_url": "https://github.com/petergpt/bullshit-benchmark/tree/main/data/latest", "audit_status": "verified", "rule_ids": [ "R5g_c" ], "notes": "GitHub repo source; not yet verified. R5g(c).", "reported_setting": "leaderboard.csv green_rate×100; reasoning=none", "source_type": "leaderboard", "matches_canonical": true }, { "model_id": "claude-opus-4.5", "benchmark_id": "bullshit_pushback", "score": 90.91, "reference_url": "https://github.com/petergpt/bullshit-benchmark/tree/main/data/latest", "audit_status": "verified", "rule_ids": [ "R5g_c" ], "notes": "GitHub repo source; not yet verified. R5g(c).", "reported_setting": "leaderboard.csv green_rate×100; reasoning=high", "source_type": "leaderboard", "matches_canonical": true }, { "model_id": "claude-haiku-4.5", "benchmark_id": "bullshit_pushback", "score": 87.27, "reference_url": "https://github.com/petergpt/bullshit-benchmark/tree/main/data/latest", "audit_status": "verified", "rule_ids": [ "R5g_c" ], "notes": "GitHub repo source; not yet verified. R5g(c).", "reported_setting": "leaderboard.csv green_rate×100; reasoning=high", "source_type": "leaderboard", "matches_canonical": true }, { "model_id": "claude-opus-4.7", "benchmark_id": "bullshit_pushback", "score": 80.0, "reference_url": "https://github.com/petergpt/bullshit-benchmark/tree/main/data/latest", "audit_status": "verified", "reported_setting": "leaderboard.csv green_rate×100; reasoning=none", "source_type": "leaderboard", "matches_canonical": true }, { "model_id": "grok-4.20", "benchmark_id": "bullshit_pushback", "score": 67.27, "reference_url": "https://github.com/petergpt/bullshit-benchmark/tree/main/data/latest", "audit_status": "verified", "reported_setting": "leaderboard.csv green_rate×100; reasoning=xhigh (multi-agent-beta variant)", "source_type": "leaderboard", "matches_canonical": true }, { "model_id": "qwen3.5-397b", "benchmark_id": "bullshit_pushback", "score": 65.45, "reference_url": "https://github.com/petergpt/bullshit-benchmark/tree/main/data/latest", "audit_status": "verified", "reported_setting": "leaderboard.csv green_rate×100; reasoning=none", "source_type": "leaderboard", "matches_canonical": true }, { "model_id": "claude-sonnet-4.5", "benchmark_id": "bullshit_pushback", "score": 65.45, "reference_url": "https://github.com/petergpt/bullshit-benchmark/tree/main/data/latest", "audit_status": "verified", "rule_ids": [ "R5g_c" ], "notes": "GitHub repo source; not yet verified. R5g(c).", "reported_setting": "leaderboard.csv green_rate×100; reasoning=high", "source_type": "leaderboard", "matches_canonical": true }, { "model_id": "claude-3.7-sonnet", "benchmark_id": "bullshit_pushback", "score": 54.72, "reference_url": "https://github.com/petergpt/bullshit-benchmark/tree/main/data/latest", "audit_status": "verified", "reported_setting": "leaderboard.csv green_rate×100; reasoning=default, thinking variant", "source_type": "leaderboard", "matches_canonical": true }, { "model_id": "llama-4-maverick", "benchmark_id": "bullshit_pushback", "score": 54.55, "reference_url": "", "audit_status": "verified", "source_type": "third_party", "reported_setting": {}, "matches_canonical": false, "notes": "BullshitPushback third-party benchmark." }, { "model_id": "kimi-k2.5", "benchmark_id": "bullshit_pushback", "score": 47.27, "reference_url": "https://github.com/petergpt/bullshit-benchmark/tree/main/data/latest", "audit_status": "verified", "reported_setting": "leaderboard.csv green_rate×100; reasoning=none", "source_type": "leaderboard", "matches_canonical": true }, { "model_id": "claude-opus-4.1", "benchmark_id": "bullshit_pushback", "score": 40.0, "reference_url": "https://github.com/petergpt/bullshit-benchmark/tree/main/data/latest", "audit_status": "verified", "rule_ids": [ "R5g_c" ], "notes": "GitHub repo source; not yet verified. R5g(c).", "reported_setting": "leaderboard.csv green_rate×100; reasoning=none", "source_type": "leaderboard", "matches_canonical": true }, { "model_id": "llama-4-scout", "benchmark_id": "bullshit_pushback", "score": 36.36, "reference_url": "", "audit_status": "verified", "source_type": "third_party", "reported_setting": {}, "matches_canonical": false, "notes": "BullshitPushback is third-party benchmark." }, { "model_id": "claude-opus-4", "benchmark_id": "bullshit_pushback", "score": 32.73, "reference_url": "https://github.com/petergpt/bullshit-benchmark/tree/main/data/latest", "audit_status": "verified", "reported_setting": "leaderboard.csv green_rate×100; reasoning=default", "source_type": "leaderboard", "matches_canonical": true }, { "model_id": "gpt-5.2", "benchmark_id": "bullshit_pushback", "score": 27.27, "reference_url": "https://github.com/petergpt/bullshit-benchmark/tree/main/data/latest", "audit_status": "verified", "reported_setting": "leaderboard.csv green_rate×100; reasoning=none", "source_type": "leaderboard", "matches_canonical": true }, { "model_id": "gpt-5.4", "benchmark_id": "bullshit_pushback", "score": 25.45, "reference_url": "https://github.com/petergpt/bullshit-benchmark/tree/main/data/latest", "audit_status": "verified", "reported_setting": "leaderboard.csv green_rate×100; reasoning=none", "source_type": "leaderboard", "matches_canonical": true }, { "model_id": "claude-sonnet-4", "benchmark_id": "bullshit_pushback", "score": 25.45, "reference_url": "https://github.com/petergpt/bullshit-benchmark/tree/main/data/latest", "audit_status": "verified", "reported_setting": "leaderboard.csv green_rate×100; reasoning=none", "source_type": "leaderboard", "matches_canonical": true }, { "model_id": "gpt-5.1", "benchmark_id": "bullshit_pushback", "score": 25.45, "reference_url": "https://github.com/petergpt/bullshit-benchmark/tree/main/data/latest", "audit_status": "verified", "reported_setting": "leaderboard.csv green_rate×100; reasoning=default", "source_type": "leaderboard", "matches_canonical": true }, { "model_id": "gemini-2.5-pro", "benchmark_id": "bullshit_pushback", "score": 23.64, "reference_url": "https://github.com/petergpt/bullshit-benchmark/tree/main/data/latest", "audit_status": "verified", "reported_setting": "leaderboard.csv green_rate×100; reasoning=default", "source_type": "leaderboard", "matches_canonical": true }, { "model_id": "gpt-5", "benchmark_id": "bullshit_pushback", "score": 21.82, "reference_url": "https://github.com/petergpt/bullshit-benchmark/tree/main/data/latest", "audit_status": "verified", "reported_setting": "leaderboard.csv green_rate×100; reasoning=default", "source_type": "leaderboard", "matches_canonical": true }, { "model_id": "kimi-k2", "benchmark_id": "bullshit_pushback", "score": 20.0, "reference_url": "https://github.com/petergpt/bullshit-benchmark/tree/main/data/latest", "audit_status": "dropped", "notes": " | DROPPED (R5h-third-party-blog: community-benchmark)" }, { "model_id": "gpt-5.3-codex", "benchmark_id": "bullshit_pushback", "score": 14.55, "reference_url": "https://github.com/petergpt/bullshit-benchmark/tree/main/data/latest", "audit_status": "verified", "reported_setting": "leaderboard.csv green_rate×100; reasoning=low", "source_type": "leaderboard", "matches_canonical": true }, { "model_id": "gemini-2.5-flash", "benchmark_id": "bullshit_pushback", "score": 14.55, "reference_url": "https://github.com/petergpt/bullshit-benchmark/tree/main/data/latest", "audit_status": "verified", "reported_setting": "leaderboard.csv green_rate×100; reasoning=default", "source_type": "leaderboard", "matches_canonical": true }, { "model_id": "deepseek-v3.2", "benchmark_id": "bullshit_pushback", "score": 14.55, "reference_url": "https://github.com/petergpt/bullshit-benchmark/tree/main/data/latest", "audit_status": "verified", "reported_setting": "leaderboard.csv green_rate×100; reasoning=high", "source_type": "leaderboard", "matches_canonical": true }, { "model_id": "gpt-4.1", "benchmark_id": "bullshit_pushback", "score": 14.55, "reference_url": "https://github.com/petergpt/bullshit-benchmark/tree/main/data/latest", "audit_status": "verified", "reported_setting": "leaderboard.csv green_rate×100; reasoning=default", "source_type": "leaderboard", "matches_canonical": true }, { "model_id": "deepseek-r1", "benchmark_id": "bullshit_pushback", "score": 12.73, "reference_url": "https://github.com/petergpt/bullshit-benchmark/tree/main/data/latest", "audit_status": "verified", "reported_setting": "leaderboard.csv green_rate×100; reasoning=none", "source_type": "leaderboard", "matches_canonical": true }, { "model_id": "gpt-oss-120b", "benchmark_id": "bullshit_pushback", "score": 3.64, "reference_url": "https://github.com/petergpt/bullshit-benchmark/tree/main/data/latest", "audit_status": "verified", "reported_setting": "leaderboard.csv green_rate×100; reasoning=low", "source_type": "leaderboard", "matches_canonical": true }, { "model_id": "kimi-k2.6", "benchmark_id": "simpleqa_verified", "score": 36.9, "reference_url": "https://huggingface.co/deepseek-ai/DeepSeek-V4-Pro", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "DeepSeek V4-Pro model card (third-party): kimi-k2.6 (per their own canonical config) = 36.9." }, { "model_id": "gpt-5.4", "benchmark_id": "simpleqa_verified", "score": 45.3, "reference_url": "https://huggingface.co/deepseek-ai/DeepSeek-V4-Pro", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "DeepSeek V4-Pro model card (third-party): gpt-5.4 (per their own canonical config) = 45.3." }, { "model_id": "claude-opus-4.6", "benchmark_id": "simpleqa_verified", "score": 46.2, "reference_url": "https://huggingface.co/deepseek-ai/DeepSeek-V4-Pro", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "DeepSeek V4-Pro model card (third-party): claude-opus-4.6 (per their own canonical config) = 46.2." }, { "model_id": "gemini-3.1-pro", "benchmark_id": "simpleqa_verified", "score": 75.6, "reference_url": "https://huggingface.co/deepseek-ai/DeepSeek-V4-Pro", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "DeepSeek V4-Pro model card (third-party): gemini-3.1-pro (per their own canonical config) = 75.6." }, { "model_id": "glm-5.1", "benchmark_id": "simpleqa_verified", "score": 38.1, "reference_url": "https://huggingface.co/deepseek-ai/DeepSeek-V4-Pro", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "DeepSeek V4-Pro model card (third-party): glm-5.1 (per their own canonical config) = 38.1." }, { "model_id": "amazon-nova-premier", "benchmark_id": "aime_2025", "score": 16.0, "reference_url": "https://cdn.amazon.science/e5/e6/ccc5378c42dca467d1abe1628ec9/amazon-nova-premier-technical-report-and-model-card.pdf", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "zeroshot-cot", "temperature": 0.0 }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "" }, { "model_id": "amazon-nova-premier", "benchmark_id": "charxiv_reasoning", "score": 48.8, "reference_url": "https://cdn.amazon.science/e5/e6/ccc5378c42dca467d1abe1628ec9/amazon-nova-premier-technical-report-and-model-card.pdf", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "zeroshot-cot", "temperature": 0.0 }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "" }, { "model_id": "amazon-nova-premier", "benchmark_id": "simpleqa", "score": 86.3, "reference_url": "https://cdn.amazon.science/e5/e6/ccc5378c42dca467d1abe1628ec9/amazon-nova-premier-technical-report-and-model-card.pdf", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "web", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "zeroshot-cot", "temperature": 0.0 }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Reported with SerpApi web-search tool per Premier tech report; deviates from canonical tools=none." }, { "model_id": "amazon-nova-pro", "benchmark_id": "aime_2025", "score": 5.3, "reference_url": "https://cdn.amazon.science/e5/e6/ccc5378c42dca467d1abe1628ec9/amazon-nova-premier-technical-report-and-model-card.pdf", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "zeroshot-cot", "temperature": 0.0 }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "" }, { "model_id": "amazon-nova-pro", "benchmark_id": "mmmu", "score": 62.0, "reference_url": "https://cdn.amazon.science/e5/e6/ccc5378c42dca467d1abe1628ec9/amazon-nova-premier-technical-report-and-model-card.pdf", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "zeroshot-cot", "temperature": 0.0 }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "" }, { "model_id": "amazon-nova-pro", "benchmark_id": "charxiv_reasoning", "score": 40.6, "reference_url": "https://cdn.amazon.science/e5/e6/ccc5378c42dca467d1abe1628ec9/amazon-nova-premier-technical-report-and-model-card.pdf", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "zeroshot-cot", "temperature": 0.0 }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "" }, { "model_id": "amazon-nova-pro", "benchmark_id": "simpleqa", "score": 84.6, "reference_url": "https://cdn.amazon.science/e5/e6/ccc5378c42dca467d1abe1628ec9/amazon-nova-premier-technical-report-and-model-card.pdf", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "web", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "zeroshot-cot", "temperature": 0.0 }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Reported with SerpApi web-search tool per Premier tech report (cross-references Nova Pro); deviates from canonical tools=none." }, { "model_id": "claude-opus-4.6", "benchmark_id": "usamo_2025", "score": 42.3, "reference_url": "https://www-cdn.anthropic.com/8b8380204f74670be75e81c820ca8dda846ab289.pdf", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "none", "sampling": "pass@1 (avg 10 trials)", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Mythos sys card Table 6.3.A: Opus 4.6 = 42.3%. Verified per Anthropic standard config (thinking max effort, default sampling, avg 5 trials)." }, { "model_id": "claude-opus-4", "benchmark_id": "tau_bench_airline", "score": 59.6, "reference_url": "https://www.anthropic.com/news/claude-4", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "agentic (TAU-bench official tools)", "sampling": "pass@1 (single-pass)", "judge": "rule-based", "harness": "τ-bench official + max steps 100", "prompt_style": "prompt addendum to Agent Policy (Anthropic)", "temperature": "top_p=0.95" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Anthropic Claude 4 blog table. With extended thinking + tool use; prompt addendum to Airline Agent Policy; max steps 100." }, { "model_id": "claude-opus-4", "benchmark_id": "mmmlu", "score": 88.8, "reference_url": "https://www.anthropic.com/news/claude-4", "reported_setting": { "mode": "non-thinking", "effort": "default", "tools": "none", "sampling": "pass@1 (single-pass)", "judge": "rule-based", "harness": "official", "prompt_style": "avg 14 non-English languages", "temperature": "top_p=0.95" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "Anthropic Claude 4 blog table. Avg over 14 non-English languages, no extended thinking." }, { "model_id": "claude-sonnet-4", "benchmark_id": "tau_bench_airline", "score": 60.0, "reference_url": "https://www.anthropic.com/news/claude-4", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "agentic (TAU-bench official tools)", "sampling": "pass@1 (single-pass)", "judge": "rule-based", "harness": "τ-bench official + max steps 100", "prompt_style": "prompt addendum to Agent Policy (Anthropic)", "temperature": "top_p=0.95" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Anthropic Claude 4 blog table. Prompt addendum + max steps 100." }, { "model_id": "claude-sonnet-4", "benchmark_id": "mmmlu", "score": 86.5, "reference_url": "https://www.anthropic.com/news/claude-4", "reported_setting": { "mode": "non-thinking", "effort": "default", "tools": "none", "sampling": "pass@1 (single-pass)", "judge": "rule-based", "harness": "official", "prompt_style": "avg 14 non-English languages", "temperature": "top_p=0.95" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "Anthropic Claude 4 blog table. Avg over 14 non-English languages." }, { "model_id": "claude-3.7-sonnet", "benchmark_id": "terminal_bench", "score": 35.2, "reference_url": "https://www.anthropic.com/news/claude-4", "reported_setting": { "mode": "non-thinking", "effort": "default", "tools": "agentic", "sampling": "pass@1 (single-pass)", "judge": "rule-based", "harness": "Claude Code agent framework", "prompt_style": "default", "temperature": "top_p=0.95" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "Anthropic Claude 4 blog cross-model table: Sonnet 3.7 = 35.2%." }, { "model_id": "claude-opus-4.1", "benchmark_id": "tau_bench_retail", "score": 82.4, "reference_url": "https://www.anthropic.com/news/claude-opus-4-1", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "agentic (TAU-bench official tools)", "sampling": "pass@1 (single-pass)", "judge": "rule-based", "harness": "τ-bench official + max steps 100", "prompt_style": "prompt addendum to Agent Policy (Anthropic)", "temperature": "top_p=0.95" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Anthropic Opus 4.1 blog table. Prompt addendum to Retail Agent Policy + max steps 100." }, { "model_id": "claude-opus-4.1", "benchmark_id": "tau_bench_airline", "score": 56.0, "reference_url": "https://www.anthropic.com/news/claude-opus-4-1", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "agentic (TAU-bench official tools)", "sampling": "pass@1 (single-pass)", "judge": "rule-based", "harness": "τ-bench official + max steps 100", "prompt_style": "prompt addendum to Agent Policy (Anthropic)", "temperature": "top_p=0.95" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Anthropic Opus 4.1 blog table. Prompt addendum to Airline Agent Policy + max steps 100." }, { "model_id": "claude-opus-4.1", "benchmark_id": "mmmu", "score": 77.1, "reference_url": "https://www.anthropic.com/news/claude-opus-4-1", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1 (single-pass)", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "top_p=0.95" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "Anthropic Opus 4.1 blog table (MMMU validation), with extended thinking." }, { "model_id": "claude-opus-4.1", "benchmark_id": "mmmlu", "score": 89.5, "reference_url": "https://www.anthropic.com/news/claude-opus-4-1", "reported_setting": { "mode": "non-thinking", "effort": "default", "tools": "none", "sampling": "pass@1 (single-pass)", "judge": "rule-based", "harness": "official", "prompt_style": "avg 14 non-English languages", "temperature": "top_p=0.95" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "Anthropic Opus 4.1 blog table. Avg over 14 non-English languages, no extended thinking." }, { "model_id": "claude-opus-4.5", "benchmark_id": "mmmlu", "score": 90.8, "reference_url": "https://www.anthropic.com/news/claude-opus-4-5", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1 (avg 5 trials)", "judge": "rule-based", "harness": "official", "prompt_style": "avg 14 non-English languages", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "Anthropic Opus 4.5 blog table: 90.8% (avg 14 non-English languages).", "candidates": [ { "score": 90.1, "reference_url": "https://huggingface.co/Qwen/Qwen3.5-397B-A17B", "source_type": "model_card", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "notes": "Qwen3.5-397B-A17B HF model card cross-model table (Qwen self-test of 3rd-party model, mc=false): alternate measurement 90.1 (3rd-party Qwen self-test)." }, { "score": 91.0, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "source_type": "model_card", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "notes": "Doubao Seed 2.0 model card Table 3: claude-opus-4.5=91.0." } ] }, { "model_id": "claude-opus-4.5", "benchmark_id": "tau2_bench_retail", "score": 88.9, "reference_url": "https://www.anthropic.com/news/claude-opus-4-5", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "agentic (τ²-bench official tools)", "sampling": "pass@1 (avg 5 trials)", "judge": "rule-based", "harness": "τ²-bench official", "prompt_style": "per Anthropic, prompt addendum may apply", "temperature": "default" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Anthropic Opus 4.5 blog table: τ²-bench retail = 88.9%." }, { "model_id": "claude-opus-4.5", "benchmark_id": "tau2_bench_telecom", "score": 98.2, "reference_url": "https://www.anthropic.com/news/claude-opus-4-5", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "agentic (τ²-bench official tools)", "sampling": "pass@1 (avg 5 trials)", "judge": "rule-based", "harness": "τ²-bench official", "prompt_style": "per Anthropic, prompt addendum may apply", "temperature": "default" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Anthropic Opus 4.5 blog table: τ²-bench telecom = 98.2%." }, { "model_id": "claude-sonnet-4.5", "benchmark_id": "mmmlu", "score": 89.1, "reference_url": "https://www.anthropic.com/news/claude-opus-4-5", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1 (avg 5 trials)", "judge": "rule-based", "harness": "official", "prompt_style": "avg 14 non-English languages", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "Anthropic Opus 4.5 blog table: Sonnet 4.5 = 89.1%.", "candidates": [ { "score": 89.1, "reference_url": "https://deepmind.google/models/gemini/flash/", "source_type": "official_blog", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "notes": "DeepMind /models/gemini/flash/ Performance table: MMMLU 89.1%" }, { "score": 89.9, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "source_type": "model_card", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "notes": "Doubao Seed 2.0 model card Table 3: claude-sonnet-4.5=89.9." } ] }, { "model_id": "claude-opus-4.7", "benchmark_id": "osworld", "score": 78.0, "reference_url": "https://www.anthropic.com/news/claude-opus-4-7", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "agentic", "sampling": "pass@1 (avg 5 trials)", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "Anthropic Opus 4.7 blog table: claude-opus-4.7 = 78.0%.", "candidates": [ { "score": 78.0, "reference_url": "https://openai.com/index/introducing-gpt-5-5/", "source_type": "third_party", "reported_setting": { "note": "see primary" }, "notes": "OpenAI GPT-5.5 blog reports 78.0 (matches primary)." }, { "score": 82.8, "reference_url": "https://seed.bytedance.com/en/blog/seed2-1-officially-released-advancing-ai-productivity", "source_type": "official_blog", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "GUI + shell/commands + filesystem tools", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official", "prompt_style": "official default", "temperature": "unknown" }, "notes": "Release blog EN: Computer-use chart (https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/user-upload/4xfa4mqq4w8x3.png)." } ] }, { "model_id": "claude-opus-4.7", "benchmark_id": "finance_agent", "score": 64.4, "reference_url": "https://www.anthropic.com/news/claude-opus-4-7", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "agentic", "sampling": "pass@1 (avg 5 trials)", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "Anthropic Opus 4.7 blog table: claude-opus-4.7 = 64.4%.", "candidates": [ { "score": 64.4, "reference_url": "https://openai.com/index/introducing-gpt-5-5/", "source_type": "third_party", "reported_setting": { "note": "see primary" }, "notes": "OpenAI GPT-5.5 blog reports 64.4 (matches primary)." }, { "score": 64.4, "reference_url": "https://seed.bytedance.com/en/blog/seed2-1-officially-released-advancing-ai-productivity", "source_type": "official_blog", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "financial-analysis agent tools", "sampling": "pass@1", "judge": "LLM-as-judge rubric and contradiction grader", "harness": "official", "prompt_style": "official default", "temperature": "unknown" }, "notes": "Release blog EN: General-agent chart (https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/user-upload/4xfa4mqq4u89c.png)." } ] }, { "model_id": "claude-opus-4.7", "benchmark_id": "cybergym", "score": 73.1, "reference_url": "https://www.anthropic.com/news/claude-opus-4-7", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "agentic", "sampling": "pass@1 (avg 5 trials)", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "Anthropic Opus 4.7 blog table: claude-opus-4.7 = 73.1%.", "candidates": [ { "score": 73.1, "reference_url": "https://openai.com/index/introducing-gpt-5-5/", "source_type": "third_party", "reported_setting": { "note": "see primary" }, "notes": "OpenAI GPT-5.5 blog reports 73.1 (matches primary)." }, { "score": 73.1, "reference_url": "https://media.x.ai/v1/website/4p5-5184fdf9.pdf", "source_type": "model_card", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "unrestricted agentic code execution", "sampling": "unknown", "judge": "PoC reproduced on vulnerable but not fixed version", "harness": "CyberGym unrestricted capability harness", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown" }, "notes": "card45 CyberGym Mean Reproduced (%); source effort=max; benchmark protocol matches canonical=True. Unknown source fields remain unknown." }, { "score": 73.1, "reference_url": "https://seed.bytedance.com/en/blog/seed2-1-officially-released-advancing-ai-productivity", "source_type": "official_blog", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "cybersecurity agent environment", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official", "prompt_style": "official default", "temperature": "unknown" }, "notes": "Release blog EN: Coding chart (https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/user-upload/4xfa4mqq4wlbt.png). Release blog reports Pro=68.7; PDF Table 5 reports Pro=70.2." }, { "score": 73.1, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2.1/Seed2_1_Model_Card.pdf", "source_type": "tech_report", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "Trae or benchmark-specific coding-agent environment", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official", "prompt_style": "official default", "temperature": "unknown" }, "notes": "Model card Table 5 (p19). Source dashes are not zero." } ] }, { "model_id": "claude-opus-4.6", "benchmark_id": "finance_agent", "score": 60.1, "reference_url": "https://www.anthropic.com/news/claude-opus-4-7", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "agentic", "sampling": "pass@1 (avg 5 trials)", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "Anthropic Opus 4.7 blog table: claude-opus-4.6 = 60.1%." }, { "model_id": "claude-opus-4.6", "benchmark_id": "cybergym", "score": 73.8, "reference_url": "https://www.anthropic.com/news/claude-opus-4-7", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "agentic", "sampling": "pass@1 (avg 5 trials)", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "Anthropic Opus 4.7 blog table: claude-opus-4.6 = 73.8%.", "candidates": [ { "score": 66.6, "reference_url": "https://z.ai/blog/glm-5.1", "source_type": "official_blog", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "agentic", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "notes": "GLM-5.1 blog: claude-opus-4.6=66.6." } ] }, { "model_id": "claude-mythos", "benchmark_id": "browsecomp", "score": 86.9, "reference_url": "https://www.anthropic.com/news/claude-opus-4-7", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "agentic", "sampling": "pass@1 (avg 5 trials)", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "Anthropic Opus 4.7 blog table: claude-mythos = 86.9%.", "candidates": [ { "score": 87.9, "reference_url": "https://openai.com/index/gpt-5-6/", "source_type": "official_blog", "reported_setting": { "effort": "reference", "eval_variant": null, "code_mode": null }, "notes": "OpenAI GPT-5.6 release table (table 3): Claude Mythos Preview; BrowseComp=87.9. Exact effort/variant resolved from the rendered chart point." } ] }, { "model_id": "claude-mythos", "benchmark_id": "cybergym", "score": 83.1, "reference_url": "https://www.anthropic.com/news/claude-opus-4-7", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "agentic", "sampling": "pass@1 (avg 5 trials)", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "Anthropic Opus 4.7 blog table: claude-mythos = 83.1%." }, { "model_id": "claude-sonnet-4.6", "benchmark_id": "tau2_bench_retail", "score": 91.7, "reference_url": "https://www.anthropic.com/news/claude-sonnet-4-6", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "agentic", "sampling": "pass@1 (avg 5 trials)", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "Anthropic Sonnet 4.6 blog table: claude-sonnet-4.6 = 91.7.", "candidates": [ { "score": 91.7, "reference_url": "https://deepmind.google/models/gemini/pro/", "source_type": "official_blog", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "notes": "DeepMind /models/gemini/pro/ cross-model Performance table: τ²-bench retail 91.7% (Thinking Max)" } ] }, { "model_id": "claude-sonnet-4.6", "benchmark_id": "tau2_bench_telecom", "score": 97.9, "reference_url": "https://www.anthropic.com/news/claude-sonnet-4-6", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "agentic", "sampling": "pass@1 (avg 5 trials)", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "Anthropic Sonnet 4.6 blog table: claude-sonnet-4.6 = 97.9.", "candidates": [ { "score": 97.9, "reference_url": "https://deepmind.google/models/gemini/pro/", "source_type": "official_blog", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "notes": "DeepMind /models/gemini/pro/ cross-model Performance table: τ²-bench telecom 97.9%" } ] }, { "model_id": "claude-sonnet-4.6", "benchmark_id": "mcpatlas", "score": 61.3, "reference_url": "https://www.anthropic.com/news/claude-sonnet-4-6", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "agentic", "sampling": "pass@1 (avg 5 trials)", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "Anthropic Sonnet 4.6 blog table: claude-sonnet-4.6 = 61.3.", "candidates": [ { "score": 61.3, "reference_url": "https://deepmind.google/models/gemini/pro/", "source_type": "official_blog", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "notes": "DeepMind /models/gemini/pro/ cross-model Performance table: MCP Atlas 61.3%" }, { "score": 69.5, "reference_url": "https://deepmind.google/models/model-cards/gemini-3-5-flash/", "source_type": "official_model_card", "reported_setting": { "mode": "thinking/reasoning", "effort": "max/best available", "tools": "MCP server tool use", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "ScaleAI official May 2026 leaderboard", "temperature": "default", "snapshot": "May 2026" }, "notes": "Google Gemini 3.5 Flash May 2026 table: MCP Atlas; ScaleAI official leaderboard. Settings and provenance are preserved per cell." } ] }, { "model_id": "claude-sonnet-4.6", "benchmark_id": "finance_agent", "score": 63.3, "reference_url": "https://www.anthropic.com/news/claude-sonnet-4-6", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "agentic", "sampling": "pass@1 (avg 5 trials)", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "Anthropic Sonnet 4.6 blog table: claude-sonnet-4.6 = 63.3." }, { "model_id": "claude-sonnet-4.6", "benchmark_id": "mmmlu", "score": 89.3, "reference_url": "https://www.anthropic.com/news/claude-sonnet-4-6", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "none", "sampling": "pass@1 (avg 5 trials)", "judge": "rule-based", "harness": "official", "prompt_style": "avg 14 non-English languages", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "Anthropic Sonnet 4.6 blog table: claude-sonnet-4.6 = 89.3.", "candidates": [ { "score": 89.3, "reference_url": "https://deepmind.google/models/gemini/pro/", "source_type": "official_blog", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "notes": "DeepMind /models/gemini/pro/ cross-model Performance table: MMMLU 89.3%" } ] }, { "model_id": "claude-sonnet-4.5", "benchmark_id": "mcpatlas", "score": 43.8, "reference_url": "https://www.anthropic.com/news/claude-sonnet-4-6", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "agentic", "sampling": "pass@1 (avg 5 trials)", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "Anthropic Sonnet 4.6 blog table: claude-sonnet-4.5 = 43.8.", "candidates": [ { "score": 43.8, "reference_url": "https://deepmind.google/models/gemini/flash/", "source_type": "official_blog", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "notes": "DeepMind /models/gemini/flash/ Performance table: MCP Atlas 43.8%" } ] }, { "model_id": "claude-sonnet-4.5", "benchmark_id": "browsecomp", "score": 43.9, "reference_url": "https://www.anthropic.com/news/claude-sonnet-4-6", "reported_setting": { "mode": "non-thinking", "effort": "max", "tools": "agentic (web search + fetch + programmatic tool calling, context compaction at 50k up to 10M)", "sampling": "pass@1 (avg 5 trials)", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "Anthropic Sonnet 4.6 blog table: claude-sonnet-4.5 = 43.9.", "candidates": [ { "score": 24.1, "reference_url": "https://huggingface.co/deepseek-ai/DeepSeek-V3.2/resolve/main/assets/paper.pdf", "source_type": "tech_report", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "agentic", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "notes": "DeepSeek V3.2 tech report Table 2: claude-sonnet-4.5=24.1." }, { "score": 24.1, "reference_url": "https://z.ai/blog/glm-4.7", "source_type": "official_blog", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "agentic", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "notes": "GLM-4.7 blog: claude-sonnet-4.5=24.1." }, { "score": 19.6, "reference_url": "https://huggingface.co/MiniMaxAI/MiniMax-M2", "source_type": "model_card", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "agentic", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "notes": "MiniMax M2 model card: claude-sonnet-4.5=19.6." }, { "score": 43.9, "reference_url": "https://huggingface.co/mistralai/Mistral-Medium-3.5-128B/resolve/22b2b868a15677cfa6061277ed2f653d1349a9ab/images/image4.png?download=true", "source_type": "official_model_card", "reported_setting": { "mode": "thinking", "effort": "high (maximum)", "tools": "web search/browser environment", "sampling": "source does not state trial count", "judge": "benchmark-specified", "harness": "Mistral official BrowseComp evaluation", "prompt_style": "interleaved scratchpads", "temperature": "default (top_p, temperature)", "context": "200K (default)" }, "notes": "Displayed exactly: '43.9'. Research observation: medium-image4.png:6:2:score:reported. Chart footnotes: *All model evaluations were run with maximum reasoning settings. | **Sonnet 4.6 encountered higher rates of reasoning truncation due to external API restrictions than other models, which affected its performance. | *** Self-reported | **** Self-reported, Mistral is using context management and a discard-all strategy at 100k tokens. | τ³ scores as reported by Sierra for Claude Sonnet 4.5 and Qwen3.5. Others with user simulator: gpt-5.2 with reasoning_effort: low. 4 trials. Banking domain evaluated with terminal- or embedding-based agentic search retrieval, only highest score is reported." } ] }, { "model_id": "claude-sonnet-4.5", "benchmark_id": "finance_agent", "score": 54.5, "reference_url": "https://www.anthropic.com/news/claude-sonnet-4-6", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "agentic", "sampling": "pass@1 (avg 5 trials)", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "Anthropic Sonnet 4.6 blog table: claude-sonnet-4.5 = 54.5." }, { "model_id": "claude-sonnet-4.5", "benchmark_id": "gdpval_aa_elo", "score": 1276, "reference_url": "https://www.anthropic.com/news/claude-sonnet-4-6", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "none", "sampling": "pass@1 (avg 5 trials)", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "Anthropic Sonnet 4.6 blog table: claude-sonnet-4.5 = 1276." }, { "model_id": "claude-opus-4.6", "benchmark_id": "tau2_bench_retail", "score": 91.9, "reference_url": "https://www.anthropic.com/news/claude-sonnet-4-6", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "agentic", "sampling": "pass@1 (avg 5 trials)", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "Anthropic Sonnet 4.6 blog table: claude-opus-4.6 = 91.9.", "candidates": [ { "score": 91.9, "reference_url": "https://deepmind.google/models/gemini/pro/", "source_type": "official_blog", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "notes": "DeepMind /models/gemini/pro/ cross-model Performance table: τ²-bench retail 91.9% (Thinking Max)" } ] }, { "model_id": "claude-opus-4.6", "benchmark_id": "tau2_bench_telecom", "score": 99.3, "reference_url": "https://www.anthropic.com/news/claude-sonnet-4-6", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "agentic", "sampling": "pass@1 (avg 5 trials)", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "Anthropic Sonnet 4.6 blog table: claude-opus-4.6 = 99.3.", "candidates": [ { "score": 99.3, "reference_url": "https://deepmind.google/models/gemini/pro/", "source_type": "official_blog", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "notes": "DeepMind /models/gemini/pro/ cross-model Performance table: τ²-bench telecom 99.3%" } ] }, { "model_id": "claude-opus-4.5", "benchmark_id": "mcpatlas", "score": 62.3, "reference_url": "https://www.anthropic.com/news/claude-sonnet-4-6", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "agentic", "sampling": "pass@1 (avg 5 trials)", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "Anthropic Sonnet 4.6 blog table: claude-opus-4.5 = 62.3.", "candidates": [ { "score": 65.2, "reference_url": "https://z.ai/blog/glm-5", "source_type": "official_blog", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "agentic", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "notes": "GLM-5 blog: claude-opus-4.5=65.2." } ] }, { "model_id": "claude-opus-4.5", "benchmark_id": "finance_agent", "score": 58.8, "reference_url": "https://www.anthropic.com/news/claude-sonnet-4-6", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "agentic", "sampling": "pass@1 (avg 5 trials)", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "Anthropic Sonnet 4.6 blog table: claude-opus-4.5 = 58.8." }, { "model_id": "claude-opus-4.5", "benchmark_id": "gdpval_aa_elo", "score": 1416, "reference_url": "https://www.anthropic.com/news/claude-sonnet-4-6", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "none", "sampling": "pass@1 (avg 5 trials)", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "Anthropic Sonnet 4.6 blog table: claude-opus-4.5 = 1416." }, { "model_id": "claude-opus-4.5", "benchmark_id": "mmmu_pro", "score": 70.6, "reference_url": "https://www.anthropic.com/news/claude-sonnet-4-6", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "none", "sampling": "pass@1 (avg 5 trials)", "judge": "rule-based", "harness": "official", "prompt_style": "updated grading: separate Sonnet 4 grader; no 'think step-by-step' prefix", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "Anthropic Sonnet 4.6 blog table: claude-opus-4.5 = 70.6.", "candidates": [ { "score": 74.0, "reference_url": "https://huggingface.co/moonshotai/Kimi-K2.5", "source_type": "model_card", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "1.0" }, "notes": "Kimi K2.5 model card: claude-opus-4.5=74.0." }, { "score": 70.8, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "source_type": "model_card", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "notes": "Doubao Seed 2.0 model card: claude-opus-4.5=70.8." } ] }, { "model_id": "gemini-3-pro", "benchmark_id": "tau2_bench_retail", "score": 85.3, "reference_url": "https://deepmind.google/models/gemini/pro/", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "DeepMind /models/gemini/pro/ Performance table (PROMOTED over prior third-party): τ²-bench retail 85.3% (was 78.6 from Anthropic Sonnet 4.6 blog; demoted to candidate)", "candidates": [ { "score": 78.6, "reference_url": "https://www.anthropic.com/news/claude-sonnet-4-6", "source_type": "third_party", "reported_setting": { "mode": "unknown", "effort": "unknown", "tools": "unknown", "sampling": "unknown", "judge": "unknown", "harness": "unknown", "prompt_style": "unknown", "temperature": "unknown" }, "notes": "Demoted from primary on https://deepmind.google/models/gemini/pro/ audit. Original notes: Reported by Anthropic Sonnet 4.6 blog (third-party)." } ] }, { "model_id": "gemini-3-pro", "benchmark_id": "tau2_bench_telecom", "score": 98.0, "reference_url": "https://deepmind.google/models/gemini/pro/", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "DeepMind /models/gemini/pro/ Performance table (PROMOTED over prior third-party): τ²-bench telecom 98.0% (was 89.2 from Anthropic Sonnet 4.6 blog; demoted to candidate)", "candidates": [ { "score": 89.2, "reference_url": "https://www.anthropic.com/news/claude-sonnet-4-6", "source_type": "third_party", "reported_setting": { "mode": "unknown", "effort": "unknown", "tools": "unknown", "sampling": "unknown", "judge": "unknown", "harness": "unknown", "prompt_style": "unknown", "temperature": "unknown" }, "notes": "Demoted from primary on https://deepmind.google/models/gemini/pro/ audit. Original notes: Reported by Anthropic Sonnet 4.6 blog (third-party)." } ] }, { "model_id": "gemini-3-pro", "benchmark_id": "mmmlu", "score": 91.8, "reference_url": "https://deepmind.google/models/gemini/pro/", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "DeepMind /models/gemini/pro/ Performance table (PROMOTED over prior third-party): MMMLU 91.8% (was 92.0 from Anthropic Sonnet 4.6 blog; demoted to candidate)", "candidates": [ { "score": 92.0, "reference_url": "https://www.anthropic.com/news/claude-sonnet-4-6", "source_type": "third_party", "reported_setting": { "mode": "unknown", "effort": "unknown", "tools": "unknown", "sampling": "unknown", "judge": "unknown", "harness": "unknown", "prompt_style": "unknown", "temperature": "unknown" }, "notes": "Demoted from primary on https://deepmind.google/models/gemini/pro/ audit. Original notes: Reported by Anthropic Sonnet 4.6 blog (third-party)." }, { "score": 91.8, "reference_url": "https://deepmind.google/models/gemini/flash/", "source_type": "official_blog", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "notes": "DeepMind /models/gemini/flash/ Performance table: MMMLU 91.8%" }, { "score": 90.6, "reference_url": "https://huggingface.co/Qwen/Qwen3.5-397B-A17B", "source_type": "model_card", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "notes": "Qwen3.5-397B-A17B HF model card cross-model table (Qwen self-test of 3rd-party model, mc=false): alternate measurement 90.6 (3rd-party Qwen self-test)." } ] }, { "model_id": "gpt-5.2", "benchmark_id": "mmmlu", "score": 89.6, "reference_url": "https://openai.com/index/introducing-gpt-5-2/", "reported_setting": { "mode": "thinking", "effort": "xhigh", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-5.2 blog: 89.6 (xhigh reasoning effort).", "candidates": [ { "score": 91.0, "reference_url": "https://openai.com/index/introducing-gpt-5-4/", "source_type": "official_blog", "reported_setting": { "mode": "thinking", "effort": "xhigh", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "notes": "OpenAI GPT-5.4 blog table: 91.0. (demoted: superseded by GPT-5.2 own blog)" }, { "score": 89.6, "reference_url": "https://deepmind.google/models/gemini/pro/", "source_type": "official_blog", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "notes": "DeepMind /models/gemini/pro/ cross-model Performance table: MMMLU 89.6%" }, { "score": 89.6, "reference_url": "https://deepmind.google/models/gemini/flash/", "source_type": "official_blog", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "notes": "DeepMind /models/gemini/flash/ Performance table: MMMLU 89.6%" }, { "score": 89.5, "reference_url": "https://huggingface.co/Qwen/Qwen3.5-397B-A17B", "source_type": "model_card", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "notes": "Qwen3.5-397B-A17B HF model card cross-model table (Qwen self-test of 3rd-party model, mc=false): alternate measurement 89.5 (3rd-party Qwen self-test)." }, { "score": 90.3, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "source_type": "model_card", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "notes": "Doubao Seed 2.0 model card Table 3: gpt-5.2=90.3." } ] }, { "model_id": "gpt-5.5", "benchmark_id": "expert_swe", "score": 73.1, "reference_url": "https://openai.com/index/introducing-gpt-5-5/", "reported_setting": { "mode": "thinking", "effort": "xhigh", "tools": "agentic (bash + edit)", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-5.5 blog table: 73.1." }, { "model_id": "gpt-5.5", "benchmark_id": "gdpval_oai_woe", "score": 84.9, "reference_url": "https://openai.com/index/introducing-gpt-5-5/", "reported_setting": { "mode": "thinking", "effort": "xhigh", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-5.5 blog table: 84.9." }, { "model_id": "gpt-5.5", "benchmark_id": "osworld", "score": 78.7, "reference_url": "https://openai.com/index/introducing-gpt-5-5/", "reported_setting": { "mode": "thinking", "effort": "xhigh", "tools": "agentic", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-5.5 blog table: 78.7.", "candidates": [ { "score": 78.7, "reference_url": "https://seed.bytedance.com/en/blog/seed2-1-officially-released-advancing-ai-productivity", "source_type": "official_blog", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "GUI + shell/commands + filesystem tools", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official", "prompt_style": "official default", "temperature": "unknown" }, "notes": "Release blog EN: Computer-use chart (https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/user-upload/4xfa4mqq4w8x3.png)." } ] }, { "model_id": "gpt-5.5", "benchmark_id": "frontiermath_tier4", "score": 35.4, "reference_url": "https://openai.com/index/introducing-gpt-5-5/", "reported_setting": { "mode": "thinking", "effort": "xhigh", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-5.5 blog table: 35.4.", "candidates": [ { "score": 39.6, "reference_url": "https://llm-stats.com/benchmarks/frontiermath", "source_type": "third_party_aggregator", "reported_setting": { "effort": "pro" }, "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=GPT-5.5 Pro, slug=gpt-5.5-pro, provider=OpenAI" } ] }, { "model_id": "gpt-5.5", "benchmark_id": "cybergym", "score": 81.8, "reference_url": "https://openai.com/index/introducing-gpt-5-5/", "reported_setting": { "mode": "thinking", "effort": "xhigh", "tools": "agentic", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-5.5 blog table: 81.8.", "candidates": [ { "score": 81.8, "reference_url": "https://media.x.ai/v1/website/card-4p6-4cd2dc57.pdf", "source_type": "model_card", "reported_setting": { "mode": "thinking", "effort": "xhigh", "tools": "unrestricted agentic code execution", "sampling": "unknown", "judge": "PoC reproduced on vulnerable but not fixed version", "harness": "CyberGym unrestricted capability harness", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown" }, "notes": "card46 CyberGym Mean Reproduced (%); source effort=xhigh; benchmark protocol matches canonical=True. Unknown source fields remain unknown." }, { "score": 73.7, "reference_url": "https://media.x.ai/v1/website/4p5-5184fdf9.pdf", "source_type": "model_card", "reported_setting": { "mode": "thinking", "effort": "xhigh", "tools": "unrestricted agentic code execution", "sampling": "unknown", "judge": "PoC reproduced on vulnerable but not fixed version", "harness": "CyberGym unrestricted capability harness", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown" }, "notes": "card45 CyberGym Mean Reproduced (%); source effort=xhigh; benchmark protocol matches canonical=True. Unknown source fields remain unknown." }, { "score": 81.8, "reference_url": "https://seed.bytedance.com/en/blog/seed2-1-officially-released-advancing-ai-productivity", "source_type": "official_blog", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "cybersecurity agent environment", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official", "prompt_style": "official default", "temperature": "unknown" }, "notes": "Release blog EN: Coding chart (https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/user-upload/4xfa4mqq4wlbt.png). Release blog reports Pro=68.7; PDF Table 5 reports Pro=70.2." }, { "score": 81.8, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2.1/Seed2_1_Model_Card.pdf", "source_type": "tech_report", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "Trae or benchmark-specific coding-agent environment", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official", "prompt_style": "official default", "temperature": "unknown" }, "notes": "Model card Table 5 (p19). Source dashes are not zero." } ] }, { "model_id": "gpt-5.5", "benchmark_id": "ctf_internal", "score": 88.1, "reference_url": "https://openai.com/index/introducing-gpt-5-5/", "reported_setting": { "mode": "thinking", "effort": "xhigh", "tools": "agentic", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-5.5 blog table: 88.1.", "candidates": [ { "score": 16.86, "reference_url": "https://openai.com/index/gpt-5-6/", "source_type": "official_blog", "reported_setting": { "effort": "none", "eval_variant": null, "code_mode": "false" }, "notes": "OpenAI GPT-5.6 release figure (Capture-the-Flag Challenges): GPT-5.5; Capture-the-Flag Challenges=16.86. Figure 27: Capture-the-Flag Challenges. Cost, latency, and output-token coordinates are retained as review metadata, not score-matrix metrics." }, { "score": 33.64, "reference_url": "https://openai.com/index/gpt-5-6/", "source_type": "official_blog", "reported_setting": { "effort": "low", "eval_variant": null, "code_mode": "false" }, "notes": "OpenAI GPT-5.6 release figure (Capture-the-Flag Challenges): GPT-5.5; Capture-the-Flag Challenges=33.64. Figure 27: Capture-the-Flag Challenges. Cost, latency, and output-token coordinates are retained as review metadata, not score-matrix metrics." }, { "score": 72.41, "reference_url": "https://openai.com/index/gpt-5-6/", "source_type": "official_blog", "reported_setting": { "effort": "medium", "eval_variant": null, "code_mode": "false" }, "notes": "OpenAI GPT-5.6 release figure (Capture-the-Flag Challenges): GPT-5.5; Capture-the-Flag Challenges=72.41. Figure 27: Capture-the-Flag Challenges. Cost, latency, and output-token coordinates are retained as review metadata, not score-matrix metrics." }, { "score": 84.35, "reference_url": "https://openai.com/index/gpt-5-6/", "source_type": "official_blog", "reported_setting": { "effort": "high", "eval_variant": null, "code_mode": "false" }, "notes": "OpenAI GPT-5.6 release figure (Capture-the-Flag Challenges): GPT-5.5; Capture-the-Flag Challenges=84.35. Figure 27: Capture-the-Flag Challenges. Cost, latency, and output-token coordinates are retained as review metadata, not score-matrix metrics." } ] }, { "model_id": "gpt-5.5", "benchmark_id": "finance_agent", "score": 60.0, "reference_url": "https://openai.com/index/introducing-gpt-5-5/", "reported_setting": { "mode": "thinking", "effort": "xhigh", "tools": "agentic", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-5.5 blog table: 60.0.", "candidates": [ { "score": 65.3, "reference_url": "https://seed.bytedance.com/en/blog/seed2-1-officially-released-advancing-ai-productivity", "source_type": "official_blog", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "financial-analysis agent tools", "sampling": "pass@1", "judge": "LLM-as-judge rubric and contradiction grader", "harness": "official", "prompt_style": "official default", "temperature": "unknown" }, "notes": "Release blog EN: General-agent chart (https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/user-upload/4xfa4mqq4u89c.png)." } ] }, { "model_id": "gpt-5.5", "benchmark_id": "ib_modeling", "score": 88.5, "reference_url": "https://openai.com/index/introducing-gpt-5-5/", "reported_setting": { "mode": "thinking", "effort": "xhigh", "tools": "agentic", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-5.5 blog table: 88.5." }, { "model_id": "gpt-5.5", "benchmark_id": "officeqa_pro", "score": 54.1, "reference_url": "https://openai.com/index/introducing-gpt-5-5/", "reported_setting": { "mode": "thinking", "effort": "xhigh", "tools": "agentic", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-5.5 blog table: 54.1.", "candidates": [ { "score": 60.9, "reference_url": "https://github.com/MoonshotAI/Kimi-K3/blob/main/k3_tech_report.pdf", "source_type": "tech_report", "reported_setting": { "mode": "thinking", "effort": "xhigh", "tools": "benchmark-specified", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "PDF corpus rendered as images; no machine-readable text", "prompt_style": "default", "temperature": "default", "context": "default" }, "notes": "Kimi K3 technical report: OfficeQA Pro = 60.9. Source setting: PDF corpus rendered as images; no machine-readable text. Origin: Moonshot evaluation." }, { "score": 62.9, "reference_url": "https://seed.bytedance.com/en/blog/seed2-1-officially-released-advancing-ai-productivity", "source_type": "official_blog", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "document and rendered-image analysis", "sampling": "pass@1", "judge": "OfficeQA Pro evaluator", "harness": "official", "prompt_style": "official default", "temperature": "unknown" }, "notes": "Release blog EN: General-agent chart (https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/user-upload/4xfa4mqq4u89c.png). The source says its internal implementations and adaptations can fluctuate relative to the official leaderboard." }, { "score": 69.5, "reference_url": "https://seed.bytedance.com/en/blog/seed2-1-officially-released-advancing-ai-productivity", "source_type": "official_blog", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "high-resolution multimodal agent harness", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official", "prompt_style": "official default", "temperature": "unknown" }, "notes": "Release blog EN: Multimodal-agent chart (https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/user-upload/4xfa4mqq4v1ol.png)." } ] }, { "model_id": "gpt-5.5", "benchmark_id": "bixbench", "score": 80.5, "reference_url": "https://openai.com/index/introducing-gpt-5-5/", "reported_setting": { "mode": "thinking", "effort": "xhigh", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-5.5 blog table: 80.5." }, { "model_id": "gpt-5.5", "benchmark_id": "genebench", "score": 25.0, "reference_url": "https://openai.com/index/introducing-gpt-5-5/", "reported_setting": { "mode": "thinking", "effort": "xhigh", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-5.5 blog table: 25.0.", "candidates": [ { "score": 33.2, "reference_url": "https://llm-stats.com/benchmarks/genebench", "source_type": "third_party_aggregator", "reported_setting": { "effort": "pro" }, "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=GPT-5.5 Pro, slug=gpt-5.5-pro, provider=OpenAI" } ] }, { "model_id": "gpt-5.5", "benchmark_id": "tau2_bench_telecom", "score": 98.0, "reference_url": "https://openai.com/index/introducing-gpt-5-5/", "reported_setting": { "mode": "thinking", "effort": "xhigh", "tools": "agentic", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "original prompts (no prompt adjustment)", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-5.5 blog table: 98.0." }, { "model_id": "gpt-5.4", "benchmark_id": "expert_swe", "score": 68.5, "reference_url": "https://openai.com/index/introducing-gpt-5-5/", "reported_setting": { "mode": "thinking", "effort": "xhigh", "tools": "agentic (bash + edit)", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-5.5 blog table: 68.5." }, { "model_id": "gpt-5.4", "benchmark_id": "gdpval_oai_woe", "score": 83.0, "reference_url": "https://openai.com/index/introducing-gpt-5-5/", "reported_setting": { "mode": "thinking", "effort": "xhigh", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-5.5 blog table: 83.0." }, { "model_id": "gpt-5.4", "benchmark_id": "osworld", "score": 75.0, "reference_url": "https://openai.com/index/introducing-gpt-5-5/", "reported_setting": { "mode": "thinking", "effort": "xhigh", "tools": "agentic", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-5.5 blog table: 75.0." }, { "model_id": "gpt-5.4", "benchmark_id": "frontiermath_tier4", "score": 27.1, "reference_url": "https://openai.com/index/introducing-gpt-5-5/", "reported_setting": { "mode": "thinking", "effort": "xhigh", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-5.5 blog table: 27.1." }, { "model_id": "gpt-5.4", "benchmark_id": "cybergym", "score": 79.0, "reference_url": "https://openai.com/index/introducing-gpt-5-5/", "reported_setting": { "mode": "thinking", "effort": "xhigh", "tools": "agentic", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-5.5 blog table: 79.0.", "candidates": [ { "score": 66.3, "reference_url": "https://z.ai/blog/glm-5.1", "source_type": "official_blog", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "agentic", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "notes": "GLM-5.1 blog: gpt-5.4=66.3." } ] }, { "model_id": "gpt-5.4", "benchmark_id": "ctf_internal", "score": 83.7, "reference_url": "https://openai.com/index/introducing-gpt-5-5/", "reported_setting": { "mode": "thinking", "effort": "xhigh", "tools": "agentic", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-5.5 blog table: 83.7.", "candidates": [ { "score": 9.9, "reference_url": "https://openai.com/index/gpt-5-6/", "source_type": "official_blog", "reported_setting": { "effort": "none", "eval_variant": null, "code_mode": "false" }, "notes": "OpenAI GPT-5.6 release figure (Capture-the-Flag Challenges): GPT-5.4; Capture-the-Flag Challenges=9.9. Figure 27: Capture-the-Flag Challenges. Cost, latency, and output-token coordinates are retained as review metadata, not score-matrix metrics." }, { "score": 40.21, "reference_url": "https://openai.com/index/gpt-5-6/", "source_type": "official_blog", "reported_setting": { "effort": "low", "eval_variant": null, "code_mode": "false" }, "notes": "OpenAI GPT-5.6 release figure (Capture-the-Flag Challenges): GPT-5.4; Capture-the-Flag Challenges=40.21. Figure 27: Capture-the-Flag Challenges. Cost, latency, and output-token coordinates are retained as review metadata, not score-matrix metrics." }, { "score": 68.33, "reference_url": "https://openai.com/index/gpt-5-6/", "source_type": "official_blog", "reported_setting": { "effort": "medium", "eval_variant": null, "code_mode": "false" }, "notes": "OpenAI GPT-5.6 release figure (Capture-the-Flag Challenges): GPT-5.4; Capture-the-Flag Challenges=68.33. Figure 27: Capture-the-Flag Challenges. Cost, latency, and output-token coordinates are retained as review metadata, not score-matrix metrics." }, { "score": 81.63, "reference_url": "https://openai.com/index/gpt-5-6/", "source_type": "official_blog", "reported_setting": { "effort": "high", "eval_variant": null, "code_mode": "false" }, "notes": "OpenAI GPT-5.6 release figure (Capture-the-Flag Challenges): GPT-5.4; Capture-the-Flag Challenges=81.63. Figure 27: Capture-the-Flag Challenges. Cost, latency, and output-token coordinates are retained as review metadata, not score-matrix metrics." }, { "score": 83.75, "reference_url": "https://openai.com/index/gpt-5-6/", "source_type": "official_blog", "reported_setting": { "effort": "xhigh", "eval_variant": null, "code_mode": "false" }, "notes": "OpenAI GPT-5.6 release figure (Capture-the-Flag Challenges): GPT-5.4; Capture-the-Flag Challenges=83.75. Figure 27: Capture-the-Flag Challenges. Cost, latency, and output-token coordinates are retained as review metadata, not score-matrix metrics." } ] }, { "model_id": "gpt-5.4", "benchmark_id": "finance_agent", "score": 56.0, "reference_url": "https://openai.com/index/introducing-gpt-5-5/", "reported_setting": { "mode": "thinking", "effort": "xhigh", "tools": "agentic", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-5.5 blog table: 56.0." }, { "model_id": "gpt-5.4", "benchmark_id": "ib_modeling", "score": 87.3, "reference_url": "https://openai.com/index/introducing-gpt-5-5/", "reported_setting": { "mode": "thinking", "effort": "xhigh", "tools": "agentic", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-5.5 blog table: 87.3." }, { "model_id": "gpt-5.4", "benchmark_id": "officeqa_pro", "score": 53.2, "reference_url": "https://openai.com/index/introducing-gpt-5-5/", "reported_setting": { "mode": "thinking", "effort": "xhigh", "tools": "agentic", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-5.5 blog table: 53.2." }, { "model_id": "gpt-5.4", "benchmark_id": "bixbench", "score": 74.0, "reference_url": "https://openai.com/index/introducing-gpt-5-5/", "reported_setting": { "mode": "thinking", "effort": "xhigh", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-5.5 blog table: 74.0." }, { "model_id": "gpt-5.4", "benchmark_id": "genebench", "score": 19.0, "reference_url": "https://openai.com/index/introducing-gpt-5-5/", "reported_setting": { "mode": "thinking", "effort": "xhigh", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-5.5 blog table: 19.0." }, { "model_id": "gpt-5.4", "benchmark_id": "tau2_bench_telecom", "score": 92.8, "reference_url": "https://openai.com/index/introducing-gpt-5-5/", "reported_setting": { "mode": "thinking", "effort": "xhigh", "tools": "agentic", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "original prompts (no prompt adjustment)", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-5.5 blog table: 92.8.", "candidates": [ { "score": 64.3, "reference_url": "https://openai.com/index/introducing-gpt-5-4/", "source_type": "official_blog", "reported_setting": { "note": "see primary" }, "notes": "OpenAI GPT-5.4 blog table 9: 64.3% (no prompt adjustment)." }, { "score": 98.9, "reference_url": "https://openai.com/index/introducing-gpt-5-4/", "source_type": "official_blog", "reported_setting": { "note": "Table 5 (with prompt adjustment)" }, "notes": "OpenAI GPT-5.4 blog table 5: 98.9% (with prompt adjustment, Anthropic-style). Differs from primary 92.8 (GPT-5.5 blog \"original prompts\")." } ] }, { "model_id": "gemini-3.1-pro", "benchmark_id": "gdpval_oai_woe", "score": 67.3, "reference_url": "https://openai.com/index/introducing-gpt-5-5/", "reported_setting": { "mode": "unknown", "effort": "unknown", "tools": "unknown", "sampling": "unknown", "judge": "unknown", "harness": "unknown", "prompt_style": "unknown", "temperature": "unknown" }, "matches_canonical": false, "source_type": "third_party", "audit_status": "verified", "notes": "GPT-5.5 blog knowledge-work table: GDPval(wins or ties)=67.3%; third-party self-test by OpenAI", "candidates": [] }, { "model_id": "gemini-3.1-pro", "benchmark_id": "frontiermath_tier4", "score": 16.7, "reference_url": "https://openai.com/index/introducing-gpt-5-5/", "reported_setting": { "mode": "unknown", "effort": "unknown", "tools": "unknown", "sampling": "unknown", "judge": "unknown", "harness": "unknown", "prompt_style": "unknown", "temperature": "unknown" }, "matches_canonical": false, "source_type": "third_party", "audit_status": "verified", "notes": "GPT-5.5 blog table: FrontierMath Tier 4=16.7%; third-party self-test by OpenAI", "candidates": [] }, { "model_id": "gemini-3.1-pro", "benchmark_id": "finance_agent", "score": 59.7, "reference_url": "https://openai.com/index/introducing-gpt-5-5/", "reported_setting": { "mode": "unknown", "effort": "unknown", "tools": "unknown", "sampling": "unknown", "judge": "unknown", "harness": "unknown", "prompt_style": "unknown", "temperature": "unknown" }, "matches_canonical": false, "source_type": "third_party", "audit_status": "verified", "notes": "GPT-5.5 blog table: FinanceAgent v1.1=59.7%; third-party self-test by OpenAI", "candidates": [ { "score": 59.7, "reference_url": "https://seed.bytedance.com/en/blog/seed2-1-officially-released-advancing-ai-productivity", "source_type": "official_blog", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "financial-analysis agent tools", "sampling": "pass@1", "judge": "LLM-as-judge rubric and contradiction grader", "harness": "official", "prompt_style": "official default", "temperature": "unknown" }, "notes": "Release blog EN: General-agent chart (https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/user-upload/4xfa4mqq4u89c.png)." } ] }, { "model_id": "gemini-3.1-pro", "benchmark_id": "officeqa_pro", "score": 18.1, "reference_url": "https://openai.com/index/introducing-gpt-5-5/", "reported_setting": { "mode": "unknown", "effort": "unknown", "tools": "unknown", "sampling": "unknown", "judge": "unknown", "harness": "unknown", "prompt_style": "unknown", "temperature": "unknown" }, "matches_canonical": false, "source_type": "third_party", "audit_status": "verified", "notes": "GPT-5.5 blog table: OfficeQA Pro=18.1%; third-party self-test by OpenAI", "candidates": [ { "score": 72.5, "reference_url": "https://seed.bytedance.com/en/blog/seed2-1-officially-released-advancing-ai-productivity", "source_type": "official_blog", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "document and rendered-image analysis", "sampling": "pass@1", "judge": "OfficeQA Pro evaluator", "harness": "official", "prompt_style": "official default", "temperature": "unknown" }, "notes": "Release blog EN: General-agent chart (https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/user-upload/4xfa4mqq4u89c.png). The source says its internal implementations and adaptations can fluctuate relative to the official leaderboard." }, { "score": 72.5, "reference_url": "https://seed.bytedance.com/en/blog/seed2-1-officially-released-advancing-ai-productivity", "source_type": "official_blog", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "high-resolution multimodal agent harness", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official", "prompt_style": "official default", "temperature": "unknown" }, "notes": "Release blog EN: Multimodal-agent chart (https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/user-upload/4xfa4mqq4v1ol.png)." } ] }, { "model_id": "gpt-5.4", "benchmark_id": "frontier_science_research", "score": 33.0, "reference_url": "https://openai.com/index/introducing-gpt-5-4/", "reported_setting": { "mode": "thinking", "effort": "xhigh", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-5.4 blog table: 33.0." }, { "model_id": "gpt-5.4", "benchmark_id": "officeqa", "score": 68.1, "reference_url": "https://openai.com/index/introducing-gpt-5-4/", "reported_setting": { "mode": "thinking", "effort": "xhigh", "tools": "agentic", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-5.4 blog table: 68.1." }, { "model_id": "gpt-5.4", "benchmark_id": "omnidocbench", "score": 0.109, "reference_url": "https://openai.com/index/introducing-gpt-5-4/", "reported_setting": { "mode": "thinking", "effort": "xhigh", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-5.4 blog table: 0.109." }, { "model_id": "gpt-5.4", "benchmark_id": "graphwalks_bfs_0k_128k", "score": 93.0, "reference_url": "https://openai.com/index/introducing-gpt-5-4/", "reported_setting": { "mode": "thinking", "effort": "xhigh", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-5.4 blog table: 93.0." }, { "model_id": "gpt-5.4", "benchmark_id": "graphwalks_parents_0k_128k", "score": 89.8, "reference_url": "https://openai.com/index/introducing-gpt-5-4/", "reported_setting": { "mode": "thinking", "effort": "xhigh", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-5.4 blog table: 89.8." }, { "model_id": "gpt-5.4", "benchmark_id": "graphwalks_parents_256k_1m", "score": 32.4, "reference_url": "https://openai.com/index/introducing-gpt-5-4/", "reported_setting": { "mode": "thinking", "effort": "xhigh", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-5.4 blog table: 32.4." }, { "model_id": "gpt-5.3-codex", "benchmark_id": "gdpval_oai_woe", "score": 70.9, "reference_url": "https://openai.com/index/introducing-gpt-5-3-codex/", "reported_setting": { "mode": "thinking", "effort": "xhigh", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-5.3-Codex blog: 70.9 (xhigh reasoning effort)." }, { "model_id": "gpt-5.3-codex", "benchmark_id": "toolathlon", "score": 51.9, "reference_url": "https://openai.com/index/introducing-gpt-5-4/", "reported_setting": { "mode": "thinking", "effort": "xhigh", "tools": "agentic", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-5.4 blog table: 51.9." }, { "model_id": "gpt-5.3-codex", "benchmark_id": "browsecomp", "score": 77.3, "reference_url": "https://openai.com/index/introducing-gpt-5-4/", "reported_setting": { "mode": "thinking", "effort": "xhigh", "tools": "agentic", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-5.4 blog table: 77.3." }, { "model_id": "gpt-5.3-codex", "benchmark_id": "finance_agent", "score": 54.0, "reference_url": "https://openai.com/index/introducing-gpt-5-4/", "reported_setting": { "mode": "thinking", "effort": "xhigh", "tools": "agentic", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-5.4 blog table: 54.0." }, { "model_id": "gpt-5.3-codex", "benchmark_id": "ib_modeling", "score": 79.3, "reference_url": "https://openai.com/index/introducing-gpt-5-4/", "reported_setting": { "mode": "thinking", "effort": "xhigh", "tools": "agentic", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-5.4 blog table: 79.3." }, { "model_id": "gpt-5.3-codex", "benchmark_id": "officeqa", "score": 65.1, "reference_url": "https://openai.com/index/introducing-gpt-5-4/", "reported_setting": { "mode": "thinking", "effort": "xhigh", "tools": "agentic", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-5.4 blog table: 65.1." }, { "model_id": "gpt-5.2", "benchmark_id": "gdpval_oai_woe", "score": 70.9, "reference_url": "https://openai.com/index/introducing-gpt-5-2/", "reported_setting": { "mode": "thinking", "effort": "xhigh", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-5.2 blog: 70.9 (xhigh reasoning effort).", "candidates": [ { "score": 70.9, "reference_url": "https://openai.com/index/introducing-gpt-5-3-codex/", "source_type": "official_blog", "reported_setting": { "mode": "thinking", "effort": "xhigh", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "notes": "OpenAI GPT-5.3-Codex blog: 70.9 (xhigh reasoning effort)." } ] }, { "model_id": "gpt-5.2", "benchmark_id": "toolathlon", "score": 46.3, "reference_url": "https://openai.com/index/introducing-gpt-5-2/", "reported_setting": { "mode": "thinking", "effort": "xhigh", "tools": "agentic", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-5.2 blog: 46.3 (xhigh reasoning effort).", "candidates": [ { "score": 46.3, "reference_url": "https://deepmind.google/models/gemini/flash/", "source_type": "official_blog", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "notes": "DeepMind /models/gemini/flash/ Performance table: Toolathlon 46.3%" }, { "score": 43.8, "reference_url": "https://huggingface.co/Qwen/Qwen3.5-397B-A17B", "source_type": "model_card", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "notes": "Qwen3.5-397B-A17B HF model card cross-model table (Qwen self-test of 3rd-party model, mc=false): alternate measurement 43.8 (3rd-party Qwen self-test)." } ] }, { "model_id": "gpt-5.2", "benchmark_id": "mcpatlas", "score": 60.6, "reference_url": "https://openai.com/index/introducing-gpt-5-2/", "reported_setting": { "mode": "thinking", "effort": "xhigh", "tools": "agentic", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-5.2 blog: 60.6 (xhigh reasoning effort).", "candidates": [ { "score": 60.6, "reference_url": "https://deepmind.google/models/gemini/pro/", "source_type": "official_blog", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "notes": "DeepMind /models/gemini/pro/ cross-model Performance table: MCP Atlas 60.6%" }, { "score": 60.6, "reference_url": "https://deepmind.google/models/gemini/flash/", "source_type": "official_blog", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "notes": "DeepMind /models/gemini/flash/ Performance table: MCP Atlas 60.6%" }, { "score": 68.0, "reference_url": "https://z.ai/blog/glm-5", "source_type": "official_blog", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "agentic", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "notes": "GLM-5 blog: gpt-5.2=68.0." } ] }, { "model_id": "gpt-5.2", "benchmark_id": "finance_agent", "score": 59.5, "reference_url": "https://openai.com/index/introducing-gpt-5-4/", "reported_setting": { "mode": "thinking", "effort": "xhigh", "tools": "agentic", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-5.4 blog table: 59.5." }, { "model_id": "gpt-5.2", "benchmark_id": "ib_modeling", "score": 68.4, "reference_url": "https://openai.com/index/introducing-gpt-5-2/", "reported_setting": { "mode": "thinking", "effort": "xhigh", "tools": "agentic", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-5.2 blog: 68.4 (xhigh reasoning effort).", "candidates": [] }, { "model_id": "gpt-5.2", "benchmark_id": "officeqa", "score": 63.1, "reference_url": "https://openai.com/index/introducing-gpt-5-4/", "reported_setting": { "mode": "thinking", "effort": "xhigh", "tools": "agentic", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-5.4 blog table: 63.1." }, { "model_id": "gpt-5.2", "benchmark_id": "frontier_science_research", "score": 25.2, "reference_url": "https://openai.com/index/introducing-gpt-5-4/", "reported_setting": { "mode": "thinking", "effort": "xhigh", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-5.4 blog table: 25.2." }, { "model_id": "gpt-5.2", "benchmark_id": "frontiermath_tier4", "score": 14.6, "reference_url": "https://openai.com/index/introducing-gpt-5-2/", "reported_setting": { "mode": "thinking", "effort": "xhigh", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-5.2 blog: 14.6 (xhigh reasoning effort).", "candidates": [ { "score": 18.8, "reference_url": "https://openai.com/index/introducing-gpt-5-4/", "source_type": "official_blog", "reported_setting": { "mode": "thinking", "effort": "xhigh", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "notes": "OpenAI GPT-5.4 blog table: 18.8. (demoted: superseded by GPT-5.2 own blog)" } ] }, { "model_id": "gpt-5.2", "benchmark_id": "graphwalks_bfs_0k_128k", "score": 94.0, "reference_url": "https://openai.com/index/introducing-gpt-5-2/", "reported_setting": { "mode": "thinking", "effort": "xhigh", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-5.2 blog: 94.0 (xhigh reasoning effort).", "candidates": [ { "score": 98.0, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "source_type": "model_card", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "notes": "Doubao Seed 2.0 model card Table 3: gpt-5.2=98.0." } ] }, { "model_id": "gpt-5.2", "benchmark_id": "graphwalks_parents_0k_128k", "score": 89.0, "reference_url": "https://openai.com/index/introducing-gpt-5-2/", "reported_setting": { "mode": "thinking", "effort": "xhigh", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-5.2 blog: 89.0 (xhigh reasoning effort).", "candidates": [ { "score": 99.7, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "source_type": "model_card", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "notes": "Doubao Seed 2.0 model card Table 3: gpt-5.2=99.7." } ] }, { "model_id": "gpt-5.2", "benchmark_id": "omnidocbench", "score": 0.14, "reference_url": "https://openai.com/index/introducing-gpt-5-4/", "reported_setting": { "mode": "thinking", "effort": "xhigh", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-5.4 blog table: 0.14.", "candidates": [ { "score": 0.143, "reference_url": "https://deepmind.google/models/gemini/flash/", "source_type": "official_blog", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "notes": "DeepMind /models/gemini/flash/ Performance table: OmniDocBench 1.5 edit distance 0.143" } ] }, { "model_id": "gpt-5.2", "benchmark_id": "tau2_bench_telecom", "score": 98.7, "reference_url": "https://openai.com/index/introducing-gpt-5-2/", "reported_setting": { "mode": "thinking", "effort": "xhigh", "tools": "agentic", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "with prompt adjustment (Anthropic-style)", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-5.2 blog: 98.7 (xhigh reasoning effort).", "candidates": [ { "score": 57.2, "reference_url": "https://openai.com/index/introducing-gpt-5-4/", "source_type": "official_blog", "reported_setting": { "note": "see primary" }, "notes": "OpenAI GPT-5.4 blog table 9: 57.2% (no prompt adjustment)." }, { "score": 98.7, "reference_url": "https://deepmind.google/models/gemini/pro/", "source_type": "official_blog", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "notes": "DeepMind /models/gemini/pro/ cross-model Performance table: τ²-bench telecom 98.7%" } ] }, { "model_id": "gpt-4.1", "benchmark_id": "tau2_bench_telecom", "score": 34.0, "reference_url": "https://openai.com/index/introducing-gpt-5-for-developers/", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "agentic", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "with prompt adjustment (Anthropic-style)", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-5 dev blog: gpt-4.1=34.0 (high reasoning effort).", "candidates": [] }, { "model_id": "gpt-5.3-codex", "benchmark_id": "cybersecurity_ctf", "score": 77.6, "reference_url": "https://openai.com/index/introducing-gpt-5-3-codex/", "reported_setting": { "mode": "thinking", "effort": "xhigh", "tools": "agentic", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-5.3-Codex blog: 77.6 (xhigh reasoning effort).", "candidates": [] }, { "model_id": "gpt-5.3-codex", "benchmark_id": "swelancer", "score": 81.4, "reference_url": "https://openai.com/index/introducing-gpt-5-3-codex/", "reported_setting": { "mode": "thinking", "effort": "xhigh", "tools": "agentic (bash + edit)", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-5.3-Codex blog: 81.4 (xhigh reasoning effort).", "candidates": [] }, { "model_id": "gpt-5.2", "benchmark_id": "cybersecurity_ctf", "score": 67.7, "reference_url": "https://openai.com/index/introducing-gpt-5-3-codex/", "reported_setting": { "mode": "thinking", "effort": "xhigh", "tools": "agentic", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-5.3-Codex blog: 67.7 (xhigh reasoning effort).", "candidates": [] }, { "model_id": "gpt-5.2", "benchmark_id": "swelancer", "score": 74.6, "reference_url": "https://openai.com/index/introducing-gpt-5-2/", "reported_setting": { "mode": "thinking", "effort": "xhigh", "tools": "agentic (bash + edit)", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-5.2 blog: 74.6 (xhigh reasoning effort).", "candidates": [ { "score": 48.9, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "source_type": "model_card", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "varies", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "notes": "Doubao Seed 2.0 model card: gpt-5.2=48.9." } ] }, { "model_id": "gpt-5.2", "benchmark_id": "charxiv_reasoning", "score": 82.1, "reference_url": "https://openai.com/index/introducing-gpt-5-2/", "reported_setting": { "mode": "thinking", "effort": "xhigh", "tools": "none (no-tools column)", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-5.2 blog: 82.1 (xhigh reasoning effort).", "candidates": [ { "score": 82.1, "reference_url": "https://deepmind.google/models/gemini/flash/", "source_type": "official_blog", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "notes": "DeepMind /models/gemini/flash/ Performance table: CharXiv Reasoning 82.1%" } ] }, { "model_id": "gpt-5.2", "benchmark_id": "tau2_bench_retail", "score": 82.0, "reference_url": "https://openai.com/index/introducing-gpt-5-2/", "reported_setting": { "mode": "thinking", "effort": "xhigh", "tools": "agentic", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "with prompt adjustment (Anthropic-style)", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-5.2 blog: 82.0 (xhigh reasoning effort).", "candidates": [ { "score": 82.0, "reference_url": "https://deepmind.google/models/gemini/pro/", "source_type": "official_blog", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "notes": "DeepMind /models/gemini/pro/ cross-model Performance table: τ²-bench retail 82.0% (Thinking xhigh)" } ] }, { "model_id": "gpt-5.2", "benchmark_id": "hmmt_feb_2025", "score": 99.4, "reference_url": "https://openai.com/index/introducing-gpt-5-2/", "reported_setting": { "mode": "thinking", "effort": "xhigh", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-5.2 blog: 99.4 (xhigh reasoning effort).", "candidates": [ { "score": 100.0, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "source_type": "model_card", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "notes": "Doubao Seed 2.0 model card Table 3: gpt-5.2=100.0." }, { "score": 100.0, "reference_url": "https://llm-stats.com/benchmarks/hmmt-2025", "source_type": "third_party_aggregator", "reported_setting": { "effort": "pro" }, "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=GPT-5.2 Pro, slug=gpt-5.2-pro-2025-12-11, provider=OpenAI" } ] }, { "model_id": "gpt-5.1", "benchmark_id": "swelancer", "score": 69.7, "reference_url": "https://openai.com/index/introducing-gpt-5-2/", "reported_setting": { "mode": "thinking", "effort": "xhigh", "tools": "agentic (bash + edit)", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-5.2 blog (cross-model column): GPT-5.1 = 69.7.", "candidates": [] }, { "model_id": "gpt-5.1", "benchmark_id": "charxiv_reasoning", "score": 67.0, "reference_url": "https://openai.com/index/introducing-gpt-5-2/", "reported_setting": { "mode": "thinking", "effort": "xhigh", "tools": "none (no-tools column)", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-5.2 blog (cross-model column): GPT-5.1 = 67.0.", "candidates": [] }, { "model_id": "gpt-5.1", "benchmark_id": "video_mmmu", "score": 82.9, "reference_url": "https://openai.com/index/introducing-gpt-5-2/", "reported_setting": { "mode": "thinking", "effort": "xhigh", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-5.2 blog (cross-model column): GPT-5.1 = 82.9.", "candidates": [] }, { "model_id": "gpt-5.1", "benchmark_id": "tau2_bench_telecom", "score": 95.6, "reference_url": "https://openai.com/index/introducing-gpt-5-2/", "reported_setting": { "mode": "thinking", "effort": "xhigh", "tools": "agentic", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "with prompt adjustment (Anthropic-style)", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-5.2 blog (cross-model column): GPT-5.1 = 95.6.", "candidates": [] }, { "model_id": "gpt-5.1", "benchmark_id": "tau2_bench_retail", "score": 77.9, "reference_url": "https://openai.com/index/introducing-gpt-5-2/", "reported_setting": { "mode": "thinking", "effort": "xhigh", "tools": "agentic", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "with prompt adjustment (Anthropic-style)", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-5.2 blog (cross-model column): GPT-5.1 = 77.9.", "candidates": [] }, { "model_id": "gpt-5.1", "benchmark_id": "mcpatlas", "score": 44.5, "reference_url": "https://openai.com/index/introducing-gpt-5-2/", "reported_setting": { "mode": "thinking", "effort": "xhigh", "tools": "agentic", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-5.2 blog (cross-model column): GPT-5.1 = 44.5.", "candidates": [] }, { "model_id": "gpt-5.1", "benchmark_id": "toolathlon", "score": 36.1, "reference_url": "https://openai.com/index/introducing-gpt-5-2/", "reported_setting": { "mode": "thinking", "effort": "xhigh", "tools": "agentic", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-5.2 blog (cross-model column): GPT-5.1 = 36.1.", "candidates": [] }, { "model_id": "gpt-5.1", "benchmark_id": "mmmlu", "score": 89.5, "reference_url": "https://openai.com/index/introducing-gpt-5-2/", "reported_setting": { "mode": "thinking", "effort": "xhigh", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-5.2 blog (cross-model column): GPT-5.1 = 89.5.", "candidates": [] }, { "model_id": "gpt-5.1", "benchmark_id": "hmmt_feb_2025", "score": 96.3, "reference_url": "https://openai.com/index/introducing-gpt-5-2/", "reported_setting": { "mode": "thinking", "effort": "xhigh", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-5.2 blog (cross-model column): GPT-5.1 = 96.3.", "candidates": [] }, { "model_id": "gpt-5.1", "benchmark_id": "frontiermath_tier4", "score": 12.5, "reference_url": "https://openai.com/index/introducing-gpt-5-2/", "reported_setting": { "mode": "thinking", "effort": "xhigh", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-5.2 blog (cross-model column): GPT-5.1 = 12.5.", "candidates": [ { "score": 26.7, "reference_url": "https://llm-stats.com/benchmarks/frontiermath", "source_type": "third_party_aggregator", "reported_setting": { "effort": "instant" }, "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=GPT-5.1 Instant, slug=gpt-5.1-instant-2025-11-12, provider=OpenAI" }, { "score": 26.7, "reference_url": "https://llm-stats.com/benchmarks/frontiermath", "source_type": "third_party_aggregator", "reported_setting": { "effort": "thinking" }, "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=GPT-5.1 Thinking, slug=gpt-5.1-thinking-2025-11-12, provider=OpenAI" } ] }, { "model_id": "gpt-5.1", "benchmark_id": "graphwalks_bfs_0k_128k", "score": 76.8, "reference_url": "https://openai.com/index/introducing-gpt-5-2/", "reported_setting": { "mode": "thinking", "effort": "xhigh", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-5.2 blog (cross-model column): GPT-5.1 = 76.8.", "candidates": [] }, { "model_id": "gpt-5.1", "benchmark_id": "graphwalks_parents_0k_128k", "score": 71.5, "reference_url": "https://openai.com/index/introducing-gpt-5-2/", "reported_setting": { "mode": "thinking", "effort": "xhigh", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-5.2 blog (cross-model column): GPT-5.1 = 71.5.", "candidates": [] }, { "model_id": "gpt-5.1", "benchmark_id": "ib_modeling", "score": 59.1, "reference_url": "https://openai.com/index/introducing-gpt-5-2/", "reported_setting": { "mode": "thinking", "effort": "xhigh", "tools": "agentic", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-5.2 blog (cross-model column): GPT-5.1 = 59.1.", "candidates": [] }, { "model_id": "gpt-5.1", "benchmark_id": "tau2_bench_airline", "score": 67.0, "reference_url": "https://openai.com/index/gpt-5-1-for-developers/", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "agentic", "sampling": "pass@1", "judge": "rule-based", "harness": "official with apply_patch", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-5.1 dev blog: 67.0 (high reasoning effort).", "candidates": [] }, { "model_id": "gpt-5.1", "benchmark_id": "browsecomp_long_context_128k", "score": 90.0, "reference_url": "https://openai.com/index/gpt-5-1-for-developers/", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "agentic", "sampling": "pass@1", "judge": "rule-based", "harness": "official with apply_patch", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-5.1 dev blog: 90.0 (high reasoning effort).", "candidates": [ { "score": 90.0, "reference_url": "https://llm-stats.com/benchmarks/browsecomp-long-128k", "source_type": "third_party_aggregator", "reported_setting": { "effort": "instant" }, "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=GPT-5.1 Instant, slug=gpt-5.1-instant-2025-11-12, provider=OpenAI" }, { "score": 90.0, "reference_url": "https://llm-stats.com/benchmarks/browsecomp-long-128k", "source_type": "third_party_aggregator", "reported_setting": { "effort": "thinking" }, "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=GPT-5.1 Thinking, slug=gpt-5.1-thinking-2025-11-12, provider=OpenAI" } ] }, { "model_id": "gpt-5", "benchmark_id": "browsecomp_long_context_128k", "score": 90.0, "reference_url": "https://openai.com/index/gpt-5-1-for-developers/", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "agentic", "sampling": "pass@1", "judge": "rule-based", "harness": "official with apply_patch", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-5.1 dev blog (cross-model column): GPT-5 = 90.0.", "candidates": [] }, { "model_id": "gpt-5", "benchmark_id": "aider_polyglot_diff", "score": 88.0, "reference_url": "https://openai.com/index/introducing-gpt-5-for-developers/", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "agentic (diff)", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-5 dev blog: gpt-5=88.0 (high reasoning effort).", "candidates": [] }, { "model_id": "gpt-5", "benchmark_id": "collie", "score": 99.0, "reference_url": "https://openai.com/index/introducing-gpt-5-for-developers/", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-5 dev blog: gpt-5=99.0 (high reasoning effort).", "candidates": [] }, { "model_id": "gpt-5", "benchmark_id": "erqa", "score": 65.7, "reference_url": "https://openai.com/index/introducing-gpt-5-for-developers/", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-5 dev blog: gpt-5=65.7 (high reasoning effort).", "candidates": [] }, { "model_id": "gpt-5", "benchmark_id": "factscore", "score": 2.8, "reference_url": "https://openai.com/index/introducing-gpt-5-for-developers/", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-5 dev blog: gpt-5=2.8 (high reasoning effort).", "candidates": [] }, { "model_id": "gpt-5", "benchmark_id": "graphwalks_bfs_0k_128k", "score": 78.3, "reference_url": "https://openai.com/index/introducing-gpt-5-for-developers/", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-5 dev blog: gpt-5=78.3 (high reasoning effort).", "candidates": [] }, { "model_id": "gpt-5", "benchmark_id": "graphwalks_parents_0k_128k", "score": 73.3, "reference_url": "https://openai.com/index/introducing-gpt-5-for-developers/", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-5 dev blog: gpt-5=73.3 (high reasoning effort).", "candidates": [] }, { "model_id": "gpt-5", "benchmark_id": "internal_api_if_hard", "score": 64.0, "reference_url": "https://openai.com/index/introducing-gpt-5-for-developers/", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-5 dev blog: gpt-5=64.0 (high reasoning effort).", "candidates": [] }, { "model_id": "gpt-5", "benchmark_id": "longfact_concepts", "score": 1.0, "reference_url": "https://openai.com/index/introducing-gpt-5-for-developers/", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-5 dev blog: gpt-5=1.0 (high reasoning effort).", "candidates": [] }, { "model_id": "gpt-5", "benchmark_id": "longfact_objects", "score": 1.2, "reference_url": "https://openai.com/index/introducing-gpt-5-for-developers/", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-5 dev blog: gpt-5=1.2 (high reasoning effort).", "candidates": [] }, { "model_id": "gpt-5", "benchmark_id": "mmmu_pro", "score": 78.4, "reference_url": "https://openai.com/index/introducing-gpt-5-for-developers/", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "avg across standard and vision sets", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-5 dev blog: gpt-5=78.4 (high reasoning effort).", "candidates": [] }, { "model_id": "gpt-5", "benchmark_id": "multichallenge_o3mini_grader", "score": 69.6, "reference_url": "https://openai.com/index/introducing-gpt-5-for-developers/", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-5 dev blog: gpt-5=69.6 (high reasoning effort).", "candidates": [] }, { "model_id": "gpt-5", "benchmark_id": "swelancer_freelance_dollars", "score": 112000, "reference_url": "https://openai.com/index/introducing-gpt-5-for-developers/", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "agentic (bash + edit)", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-5 dev blog: gpt-5=112000 (high reasoning effort).", "candidates": [] }, { "model_id": "gpt-5", "benchmark_id": "video_mmmu", "score": 84.6, "reference_url": "https://openai.com/index/introducing-gpt-5-for-developers/", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-5 dev blog: gpt-5=84.6 (high reasoning effort).", "candidates": [] }, { "model_id": "gpt-5", "benchmark_id": "video_mme", "score": 86.7, "reference_url": "https://openai.com/index/introducing-gpt-5-for-developers/", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-5 dev blog: gpt-5=86.7 (high reasoning effort).", "candidates": [] }, { "model_id": "gpt-5", "benchmark_id": "charxiv_reasoning", "score": 81.1, "reference_url": "https://openai.com/index/introducing-gpt-5-for-developers/", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "python tool", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "with python tool (per OpenAI dev blog)", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-5 dev blog: gpt-5=81.1 (high reasoning effort).", "candidates": [] }, { "model_id": "o3-high", "benchmark_id": "aider_polyglot_diff", "score": 79.6, "reference_url": "https://openai.com/index/introducing-gpt-5-for-developers/", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "agentic (diff)", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-5 dev blog: o3-high=79.6 (high reasoning effort).", "candidates": [ { "score": 81.3, "reference_url": "https://llm-stats.com/benchmarks/aider-polyglot", "source_type": "third_party_aggregator", "reported_setting": { "effort": "default (o3 base, not high)" }, "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=o3, slug=o3-2025-04-16, provider=OpenAI" } ] }, { "model_id": "o3-high", "benchmark_id": "browsecomp_long_context_128k", "score": 88.3, "reference_url": "https://openai.com/index/introducing-gpt-5-for-developers/", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "agentic", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-5 dev blog: o3-high=88.3 (high reasoning effort).", "candidates": [] }, { "model_id": "o3-high", "benchmark_id": "collie", "score": 98.4, "reference_url": "https://openai.com/index/introducing-gpt-5-for-developers/", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-5 dev blog: o3-high=98.4 (high reasoning effort).", "candidates": [ { "score": 98.4, "reference_url": "https://llm-stats.com/benchmarks/collie", "source_type": "third_party_aggregator", "reported_setting": { "effort": "default (o3 base, not high)" }, "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=o3, slug=o3-2025-04-16, provider=OpenAI" } ] }, { "model_id": "o3-high", "benchmark_id": "erqa", "score": 64.0, "reference_url": "https://openai.com/index/introducing-gpt-5-for-developers/", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-5 dev blog: o3-high=64.0 (high reasoning effort).", "candidates": [ { "score": 64.0, "reference_url": "https://llm-stats.com/benchmarks/erqa", "source_type": "third_party_aggregator", "reported_setting": { "effort": "default (o3 base, not high)" }, "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=o3, slug=o3-2025-04-16, provider=OpenAI" } ] }, { "model_id": "o3-high", "benchmark_id": "factscore", "score": 23.5, "reference_url": "https://openai.com/index/introducing-gpt-5-for-developers/", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-5 dev blog: o3-high=23.5 (high reasoning effort).", "candidates": [] }, { "model_id": "o3-high", "benchmark_id": "graphwalks_bfs_0k_128k", "score": 77.3, "reference_url": "https://openai.com/index/introducing-gpt-5-for-developers/", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-5 dev blog: o3-high=77.3 (high reasoning effort).", "candidates": [] }, { "model_id": "o3-high", "benchmark_id": "graphwalks_parents_0k_128k", "score": 72.9, "reference_url": "https://openai.com/index/introducing-gpt-5-for-developers/", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-5 dev blog: o3-high=72.9 (high reasoning effort).", "candidates": [] }, { "model_id": "o3-high", "benchmark_id": "internal_api_if_hard", "score": 47.4, "reference_url": "https://openai.com/index/introducing-gpt-5-for-developers/", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-5 dev blog: o3-high=47.4 (high reasoning effort).", "candidates": [] }, { "model_id": "o3-high", "benchmark_id": "longfact_concepts", "score": 5.2, "reference_url": "https://openai.com/index/introducing-gpt-5-for-developers/", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-5 dev blog: o3-high=5.2 (high reasoning effort).", "candidates": [] }, { "model_id": "o3-high", "benchmark_id": "longfact_objects", "score": 6.8, "reference_url": "https://openai.com/index/introducing-gpt-5-for-developers/", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-5 dev blog: o3-high=6.8 (high reasoning effort).", "candidates": [] }, { "model_id": "o3-high", "benchmark_id": "multichallenge_o3mini_grader", "score": 60.4, "reference_url": "https://openai.com/index/introducing-gpt-5-for-developers/", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-5 dev blog: o3-high=60.4 (high reasoning effort).", "candidates": [] }, { "model_id": "o3-high", "benchmark_id": "swelancer_freelance_dollars", "score": 86000, "reference_url": "https://openai.com/index/introducing-gpt-5-for-developers/", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "agentic (bash + edit)", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-5 dev blog: o3-high=86000 (high reasoning effort).", "candidates": [] }, { "model_id": "o3-high", "benchmark_id": "tau2_bench_airline", "score": 64.8, "reference_url": "https://openai.com/index/introducing-gpt-5-for-developers/", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "agentic", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "with prompt adjustment (Anthropic-style)", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-5 dev blog: o3-high=64.8 (high reasoning effort).", "candidates": [] }, { "model_id": "o3-high", "benchmark_id": "tau2_bench_retail", "score": 80.2, "reference_url": "https://openai.com/index/introducing-gpt-5-for-developers/", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "agentic", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "with prompt adjustment (Anthropic-style)", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-5 dev blog: o3-high=80.2 (high reasoning effort).", "candidates": [] }, { "model_id": "o3-high", "benchmark_id": "tau2_bench_telecom", "score": 58.2, "reference_url": "https://openai.com/index/introducing-gpt-5-for-developers/", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "agentic", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "with prompt adjustment (Anthropic-style)", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-5 dev blog: o3-high=58.2 (high reasoning effort).", "candidates": [] }, { "model_id": "o3-high", "benchmark_id": "video_mmmu", "score": 83.3, "reference_url": "https://openai.com/index/introducing-gpt-5-for-developers/", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-5 dev blog: o3-high=83.3 (high reasoning effort).", "candidates": [ { "score": 83.3, "reference_url": "https://llm-stats.com/benchmarks/videommmu", "source_type": "third_party_aggregator", "reported_setting": { "effort": "default (o3 base, not high)" }, "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=o3, slug=o3-2025-04-16, provider=OpenAI" } ] }, { "model_id": "o3-high", "benchmark_id": "video_mme", "score": 84.9, "reference_url": "https://openai.com/index/introducing-gpt-5-for-developers/", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-5 dev blog: o3-high=84.9 (high reasoning effort).", "candidates": [] }, { "model_id": "o3-high", "benchmark_id": "charxiv_reasoning", "score": 78.6, "reference_url": "https://openai.com/index/introducing-gpt-5-for-developers/", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "python tool", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "with python tool (per OpenAI dev blog)", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-5 dev blog: o3-high=78.6 (high reasoning effort).", "candidates": [] }, { "model_id": "o4-mini-high", "benchmark_id": "aider_polyglot_diff", "score": 58.2, "reference_url": "https://openai.com/index/introducing-gpt-5-for-developers/", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "agentic (diff)", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-5 dev blog: o4-mini-high=58.2 (high reasoning effort).", "candidates": [ { "score": 68.9, "reference_url": "https://llm-stats.com/benchmarks/aider-polyglot", "source_type": "third_party_aggregator", "reported_setting": { "effort": "default (o4-mini base, not high)" }, "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=o4-mini, slug=o4-mini, provider=OpenAI" } ] }, { "model_id": "o4-mini-high", "benchmark_id": "browsecomp_long_context_128k", "score": 80.0, "reference_url": "https://openai.com/index/introducing-gpt-5-for-developers/", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "agentic", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-5 dev blog: o4-mini-high=80.0 (high reasoning effort).", "candidates": [] }, { "model_id": "o4-mini-high", "benchmark_id": "collie", "score": 96.1, "reference_url": "https://openai.com/index/introducing-gpt-5-for-developers/", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-5 dev blog: o4-mini-high=96.1 (high reasoning effort).", "candidates": [] }, { "model_id": "o4-mini-high", "benchmark_id": "erqa", "score": 56.5, "reference_url": "https://openai.com/index/introducing-gpt-5-for-developers/", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-5 dev blog: o4-mini-high=56.5 (high reasoning effort).", "candidates": [] }, { "model_id": "o4-mini-high", "benchmark_id": "factscore", "score": 38.7, "reference_url": "https://openai.com/index/introducing-gpt-5-for-developers/", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-5 dev blog: o4-mini-high=38.7 (high reasoning effort).", "candidates": [] }, { "model_id": "o4-mini-high", "benchmark_id": "graphwalks_bfs_0k_128k", "score": 62.3, "reference_url": "https://openai.com/index/introducing-gpt-5-for-developers/", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-5 dev blog: o4-mini-high=62.3 (high reasoning effort).", "candidates": [] }, { "model_id": "o4-mini-high", "benchmark_id": "graphwalks_parents_0k_128k", "score": 51.1, "reference_url": "https://openai.com/index/introducing-gpt-5-for-developers/", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-5 dev blog: o4-mini-high=51.1 (high reasoning effort).", "candidates": [] }, { "model_id": "o4-mini-high", "benchmark_id": "internal_api_if_hard", "score": 44.7, "reference_url": "https://openai.com/index/introducing-gpt-5-for-developers/", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-5 dev blog: o4-mini-high=44.7 (high reasoning effort).", "candidates": [] }, { "model_id": "o4-mini-high", "benchmark_id": "longfact_concepts", "score": 3.0, "reference_url": "https://openai.com/index/introducing-gpt-5-for-developers/", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-5 dev blog: o4-mini-high=3.0 (high reasoning effort).", "candidates": [] }, { "model_id": "o4-mini-high", "benchmark_id": "longfact_objects", "score": 8.9, "reference_url": "https://openai.com/index/introducing-gpt-5-for-developers/", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-5 dev blog: o4-mini-high=8.9 (high reasoning effort).", "candidates": [] }, { "model_id": "o4-mini-high", "benchmark_id": "mmmu_pro", "score": 73.4, "reference_url": "https://openai.com/index/introducing-gpt-5-for-developers/", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "avg across standard and vision sets", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-5 dev blog: o4-mini-high=73.4 (high reasoning effort).", "candidates": [] }, { "model_id": "o4-mini-high", "benchmark_id": "multichallenge_o3mini_grader", "score": 57.5, "reference_url": "https://openai.com/index/introducing-gpt-5-for-developers/", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-5 dev blog: o4-mini-high=57.5 (high reasoning effort).", "candidates": [] }, { "model_id": "o4-mini-high", "benchmark_id": "swelancer_freelance_dollars", "score": 66000, "reference_url": "https://openai.com/index/introducing-gpt-5-for-developers/", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "agentic (bash + edit)", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-5 dev blog: o4-mini-high=66000 (high reasoning effort).", "candidates": [] }, { "model_id": "o4-mini-high", "benchmark_id": "tau2_bench_airline", "score": 60.2, "reference_url": "https://openai.com/index/introducing-gpt-5-for-developers/", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "agentic", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "with prompt adjustment (Anthropic-style)", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-5 dev blog: o4-mini-high=60.2 (high reasoning effort).", "candidates": [] }, { "model_id": "o4-mini-high", "benchmark_id": "tau2_bench_retail", "score": 70.5, "reference_url": "https://openai.com/index/introducing-gpt-5-for-developers/", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "agentic", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "with prompt adjustment (Anthropic-style)", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-5 dev blog: o4-mini-high=70.5 (high reasoning effort).", "candidates": [] }, { "model_id": "o4-mini-high", "benchmark_id": "tau2_bench_telecom", "score": 40.5, "reference_url": "https://openai.com/index/introducing-gpt-5-for-developers/", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "agentic", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "with prompt adjustment (Anthropic-style)", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-5 dev blog: o4-mini-high=40.5 (high reasoning effort).", "candidates": [] }, { "model_id": "o4-mini-high", "benchmark_id": "video_mmmu", "score": 79.4, "reference_url": "https://openai.com/index/introducing-gpt-5-for-developers/", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-5 dev blog: o4-mini-high=79.4 (high reasoning effort).", "candidates": [] }, { "model_id": "o4-mini-high", "benchmark_id": "video_mme", "score": 79.5, "reference_url": "https://openai.com/index/introducing-gpt-5-for-developers/", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-5 dev blog: o4-mini-high=79.5 (high reasoning effort).", "candidates": [] }, { "model_id": "o4-mini-high", "benchmark_id": "charxiv_reasoning", "score": 72.0, "reference_url": "https://openai.com/index/introducing-gpt-5-for-developers/", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "python tool", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "with python tool (per OpenAI dev blog)", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-5 dev blog: o4-mini-high=72.0 (high reasoning effort).", "candidates": [] }, { "model_id": "gpt-4.1", "benchmark_id": "aider_polyglot_diff", "score": 52.9, "reference_url": "https://openai.com/index/introducing-gpt-5-for-developers/", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "agentic (diff)", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-5 dev blog: gpt-4.1=52.9 (high reasoning effort).", "candidates": [] }, { "model_id": "gpt-4.1", "benchmark_id": "browsecomp_long_context_128k", "score": 85.9, "reference_url": "https://openai.com/index/introducing-gpt-5-for-developers/", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "agentic", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-5 dev blog: gpt-4.1=85.9 (high reasoning effort).", "candidates": [] }, { "model_id": "gpt-4.1", "benchmark_id": "collie", "score": 65.8, "reference_url": "https://openai.com/index/introducing-gpt-5-for-developers/", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-5 dev blog: gpt-4.1=65.8 (high reasoning effort).", "candidates": [] }, { "model_id": "gpt-4.1", "benchmark_id": "erqa", "score": 44.3, "reference_url": "https://openai.com/index/introducing-gpt-5-for-developers/", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-5 dev blog: gpt-4.1=44.3 (high reasoning effort).", "candidates": [] }, { "model_id": "gpt-4.1", "benchmark_id": "factscore", "score": 6.7, "reference_url": "https://openai.com/index/introducing-gpt-5-for-developers/", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-5 dev blog: gpt-4.1=6.7 (high reasoning effort).", "candidates": [] }, { "model_id": "gpt-4.1", "benchmark_id": "graphwalks_bfs_0k_128k", "score": 61.7, "reference_url": "https://openai.com/index/introducing-gpt-5-for-developers/", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-5 dev blog: gpt-4.1=61.7 (high reasoning effort).", "candidates": [] }, { "model_id": "gpt-4.1", "benchmark_id": "graphwalks_parents_0k_128k", "score": 58.0, "reference_url": "https://openai.com/index/introducing-gpt-5-for-developers/", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-5 dev blog: gpt-4.1=58.0 (high reasoning effort).", "candidates": [] }, { "model_id": "gpt-4.1", "benchmark_id": "internal_api_if_hard", "score": 49.1, "reference_url": "https://openai.com/index/introducing-gpt-5-for-developers/", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-5 dev blog: gpt-4.1=49.1 (high reasoning effort).", "candidates": [] }, { "model_id": "gpt-4.1", "benchmark_id": "longfact_concepts", "score": 0.7, "reference_url": "https://openai.com/index/introducing-gpt-5-for-developers/", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-5 dev blog: gpt-4.1=0.7 (high reasoning effort).", "candidates": [] }, { "model_id": "gpt-4.1", "benchmark_id": "longfact_objects", "score": 1.1, "reference_url": "https://openai.com/index/introducing-gpt-5-for-developers/", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-5 dev blog: gpt-4.1=1.1 (high reasoning effort).", "candidates": [] }, { "model_id": "gpt-4.1", "benchmark_id": "mmmu_pro", "score": 60.3, "reference_url": "https://openai.com/index/introducing-gpt-5-for-developers/", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "avg across standard and vision sets", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-5 dev blog: gpt-4.1=60.3 (high reasoning effort).", "candidates": [] }, { "model_id": "gpt-4.1", "benchmark_id": "multichallenge_o3mini_grader", "score": 46.2, "reference_url": "https://openai.com/index/introducing-gpt-5-for-developers/", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-5 dev blog: gpt-4.1=46.2 (high reasoning effort).", "candidates": [] }, { "model_id": "gpt-4.1", "benchmark_id": "swelancer_freelance_dollars", "score": 34000, "reference_url": "https://openai.com/index/introducing-gpt-5-for-developers/", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "agentic (bash + edit)", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-5 dev blog: gpt-4.1=34000 (high reasoning effort).", "candidates": [] }, { "model_id": "gpt-4.1", "benchmark_id": "tau2_bench_airline", "score": 56.0, "reference_url": "https://openai.com/index/introducing-gpt-5-for-developers/", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "agentic", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "with prompt adjustment (Anthropic-style)", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-5 dev blog: gpt-4.1=56.0 (high reasoning effort).", "candidates": [] }, { "model_id": "gpt-4.1", "benchmark_id": "tau2_bench_retail", "score": 74.0, "reference_url": "https://openai.com/index/introducing-gpt-5-for-developers/", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "agentic", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "with prompt adjustment (Anthropic-style)", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-5 dev blog: gpt-4.1=74.0 (high reasoning effort).", "candidates": [] }, { "model_id": "gpt-4.1", "benchmark_id": "video_mmmu", "score": 60.9, "reference_url": "https://openai.com/index/introducing-gpt-5-for-developers/", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-5 dev blog: gpt-4.1=60.9 (high reasoning effort).", "candidates": [] }, { "model_id": "gpt-4.1", "benchmark_id": "video_mme", "score": 78.7, "reference_url": "https://openai.com/index/introducing-gpt-5-for-developers/", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-5 dev blog: gpt-4.1=78.7 (high reasoning effort).", "candidates": [] }, { "model_id": "gpt-4.1", "benchmark_id": "charxiv_reasoning", "score": 56.7, "reference_url": "https://openai.com/index/introducing-gpt-5-for-developers/", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "python tool", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "with python tool (per OpenAI dev blog)", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-5 dev blog: gpt-4.1=56.7 (high reasoning effort).", "candidates": [] }, { "model_id": "gpt-4.1-mini", "benchmark_id": "aider_polyglot_diff", "score": 31.6, "reference_url": "https://openai.com/index/introducing-gpt-5-for-developers/", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "agentic (diff)", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-5 dev blog: gpt-4.1-mini=31.6 (high reasoning effort).", "candidates": [] }, { "model_id": "gpt-4.1-mini", "benchmark_id": "aime_2025", "score": 40.2, "reference_url": "https://openai.com/index/introducing-gpt-5-for-developers/", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-5 dev blog: gpt-4.1-mini=40.2 (high reasoning effort).", "candidates": [] }, { "model_id": "gpt-4.1-mini", "benchmark_id": "browsecomp_long_context_128k", "score": 89.0, "reference_url": "https://openai.com/index/introducing-gpt-5-for-developers/", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "agentic", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-5 dev blog: gpt-4.1-mini=89.0 (high reasoning effort).", "candidates": [] }, { "model_id": "gpt-4.1-mini", "benchmark_id": "collie", "score": 54.6, "reference_url": "https://openai.com/index/introducing-gpt-5-for-developers/", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-5 dev blog: gpt-4.1-mini=54.6 (high reasoning effort).", "candidates": [] }, { "model_id": "gpt-4.1-mini", "benchmark_id": "erqa", "score": 42.3, "reference_url": "https://openai.com/index/introducing-gpt-5-for-developers/", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-5 dev blog: gpt-4.1-mini=42.3 (high reasoning effort).", "candidates": [] }, { "model_id": "gpt-4.1-mini", "benchmark_id": "factscore", "score": 10.9, "reference_url": "https://openai.com/index/introducing-gpt-5-for-developers/", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-5 dev blog: gpt-4.1-mini=10.9 (high reasoning effort).", "candidates": [] }, { "model_id": "gpt-4.1-mini", "benchmark_id": "graphwalks_bfs_0k_128k", "score": 61.7, "reference_url": "https://openai.com/index/introducing-gpt-5-for-developers/", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-5 dev blog: gpt-4.1-mini=61.7 (high reasoning effort).", "candidates": [] }, { "model_id": "gpt-4.1-mini", "benchmark_id": "graphwalks_parents_0k_128k", "score": 60.5, "reference_url": "https://openai.com/index/introducing-gpt-5-for-developers/", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-5 dev blog: gpt-4.1-mini=60.5 (high reasoning effort).", "candidates": [] }, { "model_id": "gpt-4.1-mini", "benchmark_id": "hle", "score": 3.7, "reference_url": "https://openai.com/index/introducing-gpt-5-for-developers/", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-5 dev blog: gpt-4.1-mini=3.7 (high reasoning effort).", "candidates": [] }, { "model_id": "gpt-4.1-mini", "benchmark_id": "hmmt_feb_2025", "score": 35.0, "reference_url": "https://openai.com/index/introducing-gpt-5-for-developers/", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-5 dev blog: gpt-4.1-mini=35.0 (high reasoning effort).", "candidates": [] }, { "model_id": "gpt-4.1-mini", "benchmark_id": "internal_api_if_hard", "score": 45.1, "reference_url": "https://openai.com/index/introducing-gpt-5-for-developers/", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-5 dev blog: gpt-4.1-mini=45.1 (high reasoning effort).", "candidates": [] }, { "model_id": "gpt-4.1-mini", "benchmark_id": "longfact_concepts", "score": 1.1, "reference_url": "https://openai.com/index/introducing-gpt-5-for-developers/", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-5 dev blog: gpt-4.1-mini=1.1 (high reasoning effort).", "candidates": [] }, { "model_id": "gpt-4.1-mini", "benchmark_id": "longfact_objects", "score": 1.8, "reference_url": "https://openai.com/index/introducing-gpt-5-for-developers/", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-5 dev blog: gpt-4.1-mini=1.8 (high reasoning effort).", "candidates": [] }, { "model_id": "gpt-4.1-mini", "benchmark_id": "mmmu_pro", "score": 58.9, "reference_url": "https://openai.com/index/introducing-gpt-5-for-developers/", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "avg across standard and vision sets", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-5 dev blog: gpt-4.1-mini=58.9 (high reasoning effort).", "candidates": [] }, { "model_id": "gpt-4.1-mini", "benchmark_id": "multichallenge_o3mini_grader", "score": 42.2, "reference_url": "https://openai.com/index/introducing-gpt-5-for-developers/", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-5 dev blog: gpt-4.1-mini=42.2 (high reasoning effort).", "candidates": [] }, { "model_id": "gpt-4.1-mini", "benchmark_id": "swe_bench_verified", "score": 23.6, "reference_url": "https://openai.com/index/introducing-gpt-5-for-developers/", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "agentic (bash + edit)", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "dropped", "notes": "OpenAI GPT-5 dev blog: gpt-4.1-mini=23.6 (high reasoning effort). DROPPED: source uses 477/500 subset (omit 23 problems), not full SWE-bench Verified.", "candidates": [] }, { "model_id": "gpt-4.1-mini", "benchmark_id": "swelancer_freelance_dollars", "score": 31000, "reference_url": "https://openai.com/index/introducing-gpt-5-for-developers/", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "agentic (bash + edit)", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-5 dev blog: gpt-4.1-mini=31000 (high reasoning effort).", "candidates": [] }, { "model_id": "gpt-4.1-mini", "benchmark_id": "tau2_bench_airline", "score": 51.0, "reference_url": "https://openai.com/index/introducing-gpt-5-for-developers/", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "agentic", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "with prompt adjustment (Anthropic-style)", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-5 dev blog: gpt-4.1-mini=51.0 (high reasoning effort).", "candidates": [] }, { "model_id": "gpt-4.1-mini", "benchmark_id": "tau2_bench_retail", "score": 66.0, "reference_url": "https://openai.com/index/introducing-gpt-5-for-developers/", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "agentic", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "with prompt adjustment (Anthropic-style)", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-5 dev blog: gpt-4.1-mini=66.0 (high reasoning effort).", "candidates": [] }, { "model_id": "gpt-4.1-mini", "benchmark_id": "tau2_bench_telecom", "score": 44.0, "reference_url": "https://openai.com/index/introducing-gpt-5-for-developers/", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "agentic", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "with prompt adjustment (Anthropic-style)", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-5 dev blog: gpt-4.1-mini=44.0 (high reasoning effort).", "candidates": [] }, { "model_id": "gpt-4.1-mini", "benchmark_id": "video_mmmu", "score": 55.1, "reference_url": "https://openai.com/index/introducing-gpt-5-for-developers/", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-5 dev blog: gpt-4.1-mini=55.1 (high reasoning effort).", "candidates": [] }, { "model_id": "gpt-4.1-mini", "benchmark_id": "video_mme", "score": 68.4, "reference_url": "https://openai.com/index/introducing-gpt-5-for-developers/", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-5 dev blog: gpt-4.1-mini=68.4 (high reasoning effort).", "candidates": [] }, { "model_id": "gpt-4.1-mini", "benchmark_id": "charxiv_reasoning", "score": 56.8, "reference_url": "https://openai.com/index/introducing-gpt-5-for-developers/", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "python tool", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "with python tool (per OpenAI dev blog)", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-5 dev blog: gpt-4.1-mini=56.8 (high reasoning effort).", "candidates": [] }, { "model_id": "gpt-4.1-nano", "benchmark_id": "aider_polyglot_diff", "score": 6.2, "reference_url": "https://openai.com/index/introducing-gpt-5-for-developers/", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "agentic (diff)", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-5 dev blog: gpt-4.1-nano=6.2 (high reasoning effort).", "candidates": [] }, { "model_id": "gpt-4.1-nano", "benchmark_id": "browsecomp_long_context_128k", "score": 89.4, "reference_url": "https://openai.com/index/introducing-gpt-5-for-developers/", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "agentic", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-5 dev blog: gpt-4.1-nano=89.4 (high reasoning effort).", "candidates": [] }, { "model_id": "gpt-4.1-nano", "benchmark_id": "collie", "score": 42.5, "reference_url": "https://openai.com/index/introducing-gpt-5-for-developers/", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-5 dev blog: gpt-4.1-nano=42.5 (high reasoning effort).", "candidates": [] }, { "model_id": "gpt-4.1-nano", "benchmark_id": "erqa", "score": 26.5, "reference_url": "https://openai.com/index/introducing-gpt-5-for-developers/", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-5 dev blog: gpt-4.1-nano=26.5 (high reasoning effort).", "candidates": [] }, { "model_id": "gpt-4.1-nano", "benchmark_id": "graphwalks_bfs_0k_128k", "score": 25.0, "reference_url": "https://openai.com/index/introducing-gpt-5-for-developers/", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-5 dev blog: gpt-4.1-nano=25.0 (high reasoning effort).", "candidates": [] }, { "model_id": "gpt-4.1-nano", "benchmark_id": "graphwalks_parents_0k_128k", "score": 9.4, "reference_url": "https://openai.com/index/introducing-gpt-5-for-developers/", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-5 dev blog: gpt-4.1-nano=9.4 (high reasoning effort).", "candidates": [] }, { "model_id": "gpt-4.1-nano", "benchmark_id": "internal_api_if_hard", "score": 31.6, "reference_url": "https://openai.com/index/introducing-gpt-5-for-developers/", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-5 dev blog: gpt-4.1-nano=31.6 (high reasoning effort).", "candidates": [] }, { "model_id": "gpt-4.1-nano", "benchmark_id": "mmmu", "score": 55.4, "reference_url": "https://openai.com/index/introducing-gpt-5-for-developers/", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-5 dev blog: gpt-4.1-nano=55.4 (high reasoning effort).", "candidates": [] }, { "model_id": "gpt-4.1-nano", "benchmark_id": "mmmu_pro", "score": 33.0, "reference_url": "https://openai.com/index/introducing-gpt-5-for-developers/", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "avg across standard and vision sets", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-5 dev blog: gpt-4.1-nano=33.0 (high reasoning effort).", "candidates": [] }, { "model_id": "gpt-4.1-nano", "benchmark_id": "multichallenge_o3mini_grader", "score": 31.1, "reference_url": "https://openai.com/index/introducing-gpt-5-for-developers/", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-5 dev blog: gpt-4.1-nano=31.1 (high reasoning effort).", "candidates": [] }, { "model_id": "gpt-4.1-nano", "benchmark_id": "swelancer_freelance_dollars", "score": 9000, "reference_url": "https://openai.com/index/introducing-gpt-5-for-developers/", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "agentic (bash + edit)", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-5 dev blog: gpt-4.1-nano=9000 (high reasoning effort).", "candidates": [] }, { "model_id": "gpt-4.1-nano", "benchmark_id": "tau2_bench_airline", "score": 14.0, "reference_url": "https://openai.com/index/introducing-gpt-5-for-developers/", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "agentic", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "with prompt adjustment (Anthropic-style)", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-5 dev blog: gpt-4.1-nano=14.0 (high reasoning effort).", "candidates": [] }, { "model_id": "gpt-4.1-nano", "benchmark_id": "tau2_bench_retail", "score": 21.5, "reference_url": "https://openai.com/index/introducing-gpt-5-for-developers/", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "agentic", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "with prompt adjustment (Anthropic-style)", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-5 dev blog: gpt-4.1-nano=21.5 (high reasoning effort).", "candidates": [] }, { "model_id": "gpt-4.1-nano", "benchmark_id": "tau2_bench_telecom", "score": 12.1, "reference_url": "https://openai.com/index/introducing-gpt-5-for-developers/", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "agentic", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "with prompt adjustment (Anthropic-style)", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-5 dev blog: gpt-4.1-nano=12.1 (high reasoning effort).", "candidates": [] }, { "model_id": "gpt-4.1-nano", "benchmark_id": "video_mmmu", "score": 30.2, "reference_url": "https://openai.com/index/introducing-gpt-5-for-developers/", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-5 dev blog: gpt-4.1-nano=30.2 (high reasoning effort).", "candidates": [] }, { "model_id": "gpt-4.1-nano", "benchmark_id": "video_mme", "score": 55.2, "reference_url": "https://openai.com/index/introducing-gpt-5-for-developers/", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-5 dev blog: gpt-4.1-nano=55.2 (high reasoning effort).", "candidates": [] }, { "model_id": "gpt-4.1-nano", "benchmark_id": "charxiv_reasoning", "score": 40.5, "reference_url": "https://openai.com/index/introducing-gpt-5-for-developers/", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "python tool", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "with python tool (per OpenAI dev blog)", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-5 dev blog: gpt-4.1-nano=40.5 (high reasoning effort).", "candidates": [] }, { "model_id": "gemini-3-pro", "benchmark_id": "livecodebench_pro", "score": 2439, "reference_url": "https://deepmind.google/models/gemini/pro/", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "DeepMind /models/gemini/pro/ Performance table: LiveCodeBench Pro Elo 2439", "candidates": [ { "score": 2439, "reference_url": "https://deepmind.google/models/gemini/flash/", "source_type": "official_blog", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "notes": "DeepMind /models/gemini/flash/ Performance table: LiveCodeBench Pro Elo 2439" } ] }, { "model_id": "gemini-3-pro", "benchmark_id": "apex_agents", "score": 18.4, "reference_url": "https://deepmind.google/models/gemini/pro/", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "DeepMind /models/gemini/pro/ Performance table: APEX-Agents 18.4% (long-horizon professional tasks)", "candidates": [] }, { "model_id": "gemini-3-pro", "benchmark_id": "gdpval_aa_elo", "score": 1195, "reference_url": "https://deepmind.google/models/gemini/pro/", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "DeepMind /models/gemini/pro/ Performance table: GDPval-AA Elo 1195", "candidates": [] }, { "model_id": "gemini-3-pro", "benchmark_id": "mcpatlas", "score": 54.1, "reference_url": "https://deepmind.google/models/gemini/pro/", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "DeepMind /models/gemini/pro/ Performance table: MCP Atlas 54.1%", "candidates": [ { "score": 54.1, "reference_url": "https://deepmind.google/models/gemini/flash/", "source_type": "official_blog", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "notes": "DeepMind /models/gemini/flash/ Performance table: MCP Atlas 54.1%" }, { "score": 66.6, "reference_url": "https://z.ai/blog/glm-5", "source_type": "official_blog", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "agentic", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "notes": "GLM-5 blog: gemini-3-pro=66.6." } ] }, { "model_id": "gemini-3.1-pro", "benchmark_id": "tau2_bench_retail", "score": 90.8, "reference_url": "https://deepmind.google/models/gemini/pro/", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "DeepMind /models/gemini/pro/ Performance table: τ²-bench retail 90.8% (model card had earlier 90.5 — added as candidate)", "candidates": [ { "score": 90.5, "reference_url": "https://deepmind.google/models/model-cards/gemini-3-1-pro/", "source_type": "model_card", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "notes": "Earlier value from gemini-3-1-pro model card (mislabeled as tau_bench)." } ] }, { "model_id": "gemini-3.1-pro", "benchmark_id": "tau2_bench_telecom", "score": 99.3, "reference_url": "https://deepmind.google/models/gemini/pro/", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "DeepMind /models/gemini/pro/ Performance table: τ²-bench telecom 99.3%", "candidates": [] }, { "model_id": "gemini-3.1-pro", "benchmark_id": "mmmlu", "score": 92.6, "reference_url": "https://deepmind.google/models/gemini/pro/", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "DeepMind /models/gemini/pro/ Performance table: MMMLU 92.6%", "candidates": [] }, { "model_id": "gemini-3.1-pro", "benchmark_id": "apex_agents", "score": 33.5, "reference_url": "https://deepmind.google/models/gemini/pro/", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "DeepMind /models/gemini/pro/ Performance table: APEX-Agents 33.5%", "candidates": [ { "score": 32.0, "reference_url": "https://huggingface.co/moonshotai/Kimi-K2.6", "source_type": "model_card", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "agentic", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "1.0" }, "notes": "Kimi K2.6 model card: gemini-3.1-pro=32.0." }, { "score": 33.5, "reference_url": "https://seed.bytedance.com/en/blog/seed2-1-officially-released-advancing-ai-productivity", "source_type": "official_blog", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "long-horizon professional-task environment", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official", "prompt_style": "official default", "temperature": "unknown" }, "notes": "Release blog EN: General-agent chart (https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/user-upload/4xfa4mqq4u89c.png)." } ] }, { "model_id": "gemini-3.1-pro", "benchmark_id": "livecodebench_pro", "score": 2887, "reference_url": "https://deepmind.google/models/gemini/pro/", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "DeepMind /models/gemini/pro/ Performance table: LiveCodeBench Pro Elo 2887", "candidates": [] }, { "model_id": "claude-sonnet-4.6", "benchmark_id": "scicode", "score": 47.0, "reference_url": "https://deepmind.google/models/gemini/pro/", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "Created from https://deepmind.google/models/gemini/pro/ (cross-model table). DeepMind /models/gemini/pro/ cross-model Performance table: SciCode 47% [Promoted to verified — official DeepMind table.]", "candidates": [] }, { "model_id": "claude-opus-4.6", "benchmark_id": "scicode", "score": 52.0, "reference_url": "https://deepmind.google/models/gemini/pro/", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "Created from https://deepmind.google/models/gemini/pro/ (cross-model table). DeepMind /models/gemini/pro/ cross-model Performance table: SciCode 52% [Promoted to verified — official DeepMind table.]", "candidates": [ { "score": 51.9, "reference_url": "https://huggingface.co/moonshotai/Kimi-K2.6", "source_type": "model_card", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "1.0" }, "notes": "Kimi K2.6 model card: claude-opus-4.6=51.9." } ] }, { "model_id": "claude-opus-4.6", "benchmark_id": "apex_agents", "score": 29.8, "reference_url": "https://deepmind.google/models/gemini/pro/", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "Created from https://deepmind.google/models/gemini/pro/ (cross-model table). DeepMind /models/gemini/pro/ cross-model Performance table: APEX-Agents 29.8% [Promoted to verified — official DeepMind table.]", "candidates": [ { "score": 33.0, "reference_url": "https://huggingface.co/moonshotai/Kimi-K2.6", "source_type": "model_card", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "agentic", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "1.0" }, "notes": "Kimi K2.6 model card: claude-opus-4.6=33.0." } ] }, { "model_id": "gpt-5.2", "benchmark_id": "livecodebench_pro", "score": 2393, "reference_url": "https://deepmind.google/models/gemini/pro/", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "Created from https://deepmind.google/models/gemini/pro/ (cross-model table). DeepMind /models/gemini/pro/ cross-model Performance table: LiveCodeBench Pro Elo 2393 [Promoted to verified — official DeepMind table.]", "candidates": [ { "score": 2393, "reference_url": "https://deepmind.google/models/gemini/flash/", "source_type": "official_blog", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "notes": "DeepMind /models/gemini/flash/ Performance table: LiveCodeBench Pro Elo 2393" } ] }, { "model_id": "gpt-5.2", "benchmark_id": "apex_agents", "score": 23.0, "reference_url": "https://deepmind.google/models/gemini/pro/", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "Created from https://deepmind.google/models/gemini/pro/ (cross-model table). DeepMind /models/gemini/pro/ cross-model Performance table: APEX-Agents 23.0% [Promoted to verified — official DeepMind table.]", "candidates": [] }, { "model_id": "gpt-5.2", "benchmark_id": "gdpval_aa_elo", "score": 1462, "reference_url": "https://deepmind.google/models/gemini/pro/", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "Created from https://deepmind.google/models/gemini/pro/ (cross-model table). DeepMind /models/gemini/pro/ cross-model Performance table: GDPval-AA Elo 1462 [Promoted to verified — official DeepMind table.]", "candidates": [] }, { "model_id": "gemini-3-flash", "benchmark_id": "screenspot_pro", "score": 69.1, "reference_url": "https://deepmind.google/models/gemini/flash/", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "DeepMind /models/gemini/flash/ Performance table: ScreenSpot-Pro no tools 69.1%", "candidates": [] }, { "model_id": "gemini-3-flash", "benchmark_id": "charxiv_reasoning", "score": 80.3, "reference_url": "https://deepmind.google/models/gemini/flash/", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "DeepMind /models/gemini/flash/ Performance table: CharXiv Reasoning no tools 80.3%", "candidates": [] }, { "model_id": "gemini-3-flash", "benchmark_id": "omnidocbench", "score": 0.121, "reference_url": "https://deepmind.google/models/gemini/flash/", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "DeepMind /models/gemini/flash/ Performance table: OmniDocBench 1.5 overall edit distance 0.121 (lower is better)", "candidates": [] }, { "model_id": "gemini-3-flash", "benchmark_id": "livecodebench_pro", "score": 2316, "reference_url": "https://deepmind.google/models/gemini/flash/", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "DeepMind /models/gemini/flash/ Performance table: LiveCodeBench Pro Elo 2316", "candidates": [] }, { "model_id": "gemini-3-flash", "benchmark_id": "toolathlon", "score": 49.4, "reference_url": "https://deepmind.google/models/gemini/flash/", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "DeepMind /models/gemini/flash/ Performance table: Toolathlon 49.4%", "candidates": [] }, { "model_id": "gemini-3-flash", "benchmark_id": "mcpatlas", "score": 57.4, "reference_url": "https://deepmind.google/models/gemini/flash/", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "DeepMind /models/gemini/flash/ Performance table: MCP Atlas 57.4%", "candidates": [ { "score": 62.0, "reference_url": "https://deepmind.google/models/model-cards/gemini-3-5-flash/", "source_type": "official_model_card", "reported_setting": { "mode": "thinking/reasoning", "effort": "source does not state", "tools": "MCP server tool use", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "ScaleAI official May 2026 leaderboard", "temperature": "default", "snapshot": "May 2026" }, "notes": "Google Gemini 3.5 Flash May 2026 table: MCP Atlas; ScaleAI official leaderboard. Settings and provenance are preserved per cell." } ] }, { "model_id": "gemini-3-flash", "benchmark_id": "vending_bench_2", "score": 3635, "reference_url": "https://deepmind.google/models/gemini/flash/", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "DeepMind /models/gemini/flash/ Performance table: Vending-Bench 2 net worth $3635 (mean)", "candidates": [] }, { "model_id": "gemini-3-flash", "benchmark_id": "facts_benchmark", "score": 61.9, "reference_url": "https://deepmind.google/models/gemini/flash/", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "DeepMind /models/gemini/flash/ Performance table: FACTS Benchmark Suite 61.9%", "candidates": [] }, { "model_id": "gemini-3-flash", "benchmark_id": "simpleqa_verified", "score": 68.7, "reference_url": "https://deepmind.google/models/gemini/flash/", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "DeepMind /models/gemini/flash/ Performance table: SimpleQA Verified 68.7%", "candidates": [] }, { "model_id": "gemini-3-flash", "benchmark_id": "global_piqa", "score": 92.8, "reference_url": "https://deepmind.google/models/gemini/flash/", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "DeepMind /models/gemini/flash/ Performance table: Global PIQA 92.8%", "candidates": [] }, { "model_id": "gemini-3-flash", "benchmark_id": "mmmlu", "score": 91.8, "reference_url": "https://deepmind.google/models/gemini/flash/", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "DeepMind /models/gemini/flash/ Performance table: MMMLU 91.8%", "candidates": [] }, { "model_id": "gemini-3-flash", "benchmark_id": "mrcr_v2", "score": 67.2, "reference_url": "https://deepmind.google/models/gemini/flash/", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default", "context": "128k" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "DeepMind /models/gemini/flash/ Performance table: MRCR v2 8-needle 128k average 67.2%", "candidates": [ { "score": 22.1, "reference_url": "https://deepmind.google/models/gemini/flash/", "source_type": "official_blog", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default", "context": "1m" }, "notes": "DeepMind /models/gemini/flash/ table: MRCR v2 1M pointwise 22.1%" } ] }, { "model_id": "gemini-3-pro", "benchmark_id": "screenspot_pro", "score": 72.7, "reference_url": "https://deepmind.google/models/gemini/flash/", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "DeepMind /models/gemini/flash/ Performance table: ScreenSpot-Pro 72.7%", "candidates": [] }, { "model_id": "gemini-3-pro", "benchmark_id": "charxiv_reasoning", "score": 81.4, "reference_url": "https://deepmind.google/models/gemini/flash/", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "DeepMind /models/gemini/flash/ Performance table: CharXiv Reasoning 81.4%", "candidates": [] }, { "model_id": "gemini-3-pro", "benchmark_id": "omnidocbench", "score": 0.115, "reference_url": "https://deepmind.google/models/gemini/flash/", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "DeepMind /models/gemini/flash/ Performance table: OmniDocBench 1.5 edit distance 0.115", "candidates": [] }, { "model_id": "gemini-3-pro", "benchmark_id": "toolathlon", "score": 36.4, "reference_url": "https://deepmind.google/models/gemini/flash/", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "DeepMind /models/gemini/flash/ Performance table: Toolathlon 36.4%", "candidates": [] }, { "model_id": "gemini-3-pro", "benchmark_id": "vending_bench_2", "score": 5478, "reference_url": "https://deepmind.google/models/gemini/flash/", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "DeepMind /models/gemini/flash/ Performance table: Vending-Bench 2 net worth $5478 (mean)", "candidates": [ { "score": 5478.16, "reference_url": "https://z.ai/blog/glm-5", "source_type": "official_blog", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "agentic", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "notes": "GLM-5 blog: gemini-3-pro=5478.16." } ] }, { "model_id": "gemini-3-pro", "benchmark_id": "facts_benchmark", "score": 70.5, "reference_url": "https://deepmind.google/models/gemini/flash/", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "DeepMind /models/gemini/flash/ Performance table: FACTS Benchmark Suite 70.5%", "candidates": [] }, { "model_id": "gemini-3-pro", "benchmark_id": "simpleqa_verified", "score": 72.1, "reference_url": "https://deepmind.google/models/gemini/flash/", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "DeepMind /models/gemini/flash/ Performance table: SimpleQA Verified 72.1%", "candidates": [] }, { "model_id": "gemini-3-pro", "benchmark_id": "global_piqa", "score": 93.4, "reference_url": "https://deepmind.google/models/gemini/flash/", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "DeepMind /models/gemini/flash/ Performance table: Global PIQA 93.4%", "candidates": [ { "score": 93.2, "reference_url": "https://huggingface.co/Qwen/Qwen3.5-397B-A17B", "source_type": "model_card", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "notes": "Qwen3.5-397B-A17B HF model card cross-model table (Qwen self-test of 3rd-party model, mc=false): alternate measurement 93.2 (3rd-party Qwen self-test)." }, { "score": 95.0, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "source_type": "model_card", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "notes": "Doubao Seed 2.0 model card Table 3: gemini-3-pro=95.0." } ] }, { "model_id": "gemini-2.5-flash", "benchmark_id": "mmmu_pro", "score": 66.7, "reference_url": "https://deepmind.google/models/gemini/flash/", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "DeepMind /models/gemini/flash/ Performance table: MMMU-Pro 66.7%", "candidates": [] }, { "model_id": "gemini-2.5-flash", "benchmark_id": "screenspot_pro", "score": 3.9, "reference_url": "https://deepmind.google/models/gemini/flash/", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "DeepMind /models/gemini/flash/ Performance table: ScreenSpot-Pro 3.9%", "candidates": [] }, { "model_id": "gemini-2.5-flash", "benchmark_id": "charxiv_reasoning", "score": 63.7, "reference_url": "https://deepmind.google/models/gemini/flash/", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "DeepMind /models/gemini/flash/ Performance table: CharXiv Reasoning 63.7%", "candidates": [] }, { "model_id": "gemini-2.5-flash", "benchmark_id": "omnidocbench", "score": 0.154, "reference_url": "https://deepmind.google/models/gemini/flash/", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "DeepMind /models/gemini/flash/ Performance table: OmniDocBench 1.5 edit distance 0.154", "candidates": [] }, { "model_id": "gemini-2.5-flash", "benchmark_id": "livecodebench_pro", "score": 1143, "reference_url": "https://deepmind.google/models/gemini/flash/", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "DeepMind /models/gemini/flash/ Performance table: LiveCodeBench Pro Elo 1143", "candidates": [] }, { "model_id": "gemini-2.5-flash", "benchmark_id": "toolathlon", "score": 3.7, "reference_url": "https://deepmind.google/models/gemini/flash/", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "DeepMind /models/gemini/flash/ Performance table: Toolathlon 3.7%", "candidates": [] }, { "model_id": "gemini-2.5-flash", "benchmark_id": "mcpatlas", "score": 3.4, "reference_url": "https://deepmind.google/models/gemini/flash/", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "DeepMind /models/gemini/flash/ Performance table: MCP Atlas 3.4%", "candidates": [] }, { "model_id": "gemini-2.5-flash", "benchmark_id": "vending_bench_2", "score": 549, "reference_url": "https://deepmind.google/models/gemini/flash/", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "DeepMind /models/gemini/flash/ Performance table: Vending-Bench 2 net worth $549", "candidates": [] }, { "model_id": "gemini-2.5-flash", "benchmark_id": "facts_benchmark", "score": 50.4, "reference_url": "https://deepmind.google/models/gemini/flash/", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "DeepMind /models/gemini/flash/ Performance table: FACTS Benchmark Suite 50.4%", "candidates": [] }, { "model_id": "gemini-2.5-flash", "benchmark_id": "simpleqa_verified", "score": 28.1, "reference_url": "https://deepmind.google/models/gemini/flash/", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "DeepMind /models/gemini/flash/ Performance table: SimpleQA Verified 28.1%", "candidates": [] }, { "model_id": "gemini-2.5-flash", "benchmark_id": "mmmlu", "score": 86.6, "reference_url": "https://deepmind.google/models/gemini/flash/", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "DeepMind /models/gemini/flash/ Performance table: MMMLU 86.6%", "candidates": [] }, { "model_id": "gemini-2.5-flash", "benchmark_id": "global_piqa", "score": 90.2, "reference_url": "https://deepmind.google/models/gemini/flash/", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "DeepMind /models/gemini/flash/ Performance table: Global PIQA 90.2%", "candidates": [] }, { "model_id": "gemini-2.5-flash", "benchmark_id": "mrcr_v2", "score": 54.3, "reference_url": "https://deepmind.google/models/gemini/flash/", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default", "context": "128k" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "DeepMind /models/gemini/flash/ Performance table: MRCR v2 8-needle 128k 54.3%", "candidates": [ { "score": 21.0, "reference_url": "https://deepmind.google/models/gemini/flash/", "source_type": "official_blog", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default", "context": "1m" }, "notes": "DeepMind /models/gemini/flash/ Performance table: MRCR v2 1M 21.0%" } ] }, { "model_id": "gemini-2.5-pro", "benchmark_id": "mmmu_pro", "score": 68.0, "reference_url": "https://deepmind.google/models/gemini/flash/", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "DeepMind /models/gemini/flash/ Performance table: MMMU-Pro 68.0%", "candidates": [] }, { "model_id": "gemini-2.5-pro", "benchmark_id": "screenspot_pro", "score": 11.4, "reference_url": "https://deepmind.google/models/gemini/flash/", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "DeepMind /models/gemini/flash/ Performance table: ScreenSpot-Pro 11.4%", "candidates": [] }, { "model_id": "gemini-2.5-pro", "benchmark_id": "charxiv_reasoning", "score": 69.6, "reference_url": "https://deepmind.google/models/gemini/flash/", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "DeepMind /models/gemini/flash/ Performance table: CharXiv Reasoning 69.6%", "candidates": [] }, { "model_id": "gemini-2.5-pro", "benchmark_id": "omnidocbench", "score": 0.145, "reference_url": "https://deepmind.google/models/gemini/flash/", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "DeepMind /models/gemini/flash/ Performance table: OmniDocBench 1.5 edit distance 0.145", "candidates": [] }, { "model_id": "gemini-2.5-pro", "benchmark_id": "livecodebench_pro", "score": 1775, "reference_url": "https://deepmind.google/models/gemini/flash/", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "DeepMind /models/gemini/flash/ Performance table: LiveCodeBench Pro Elo 1775", "candidates": [] }, { "model_id": "gemini-2.5-pro", "benchmark_id": "toolathlon", "score": 10.5, "reference_url": "https://deepmind.google/models/gemini/flash/", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "DeepMind /models/gemini/flash/ Performance table: Toolathlon 10.5%", "candidates": [] }, { "model_id": "gemini-2.5-pro", "benchmark_id": "mcpatlas", "score": 8.8, "reference_url": "https://deepmind.google/models/gemini/flash/", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "DeepMind /models/gemini/flash/ Performance table: MCP Atlas 8.8%", "candidates": [] }, { "model_id": "gemini-2.5-pro", "benchmark_id": "vending_bench_2", "score": 574, "reference_url": "https://deepmind.google/models/gemini/flash/", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "DeepMind /models/gemini/flash/ Performance table: Vending-Bench 2 net worth $574", "candidates": [] }, { "model_id": "gemini-2.5-pro", "benchmark_id": "facts_benchmark", "score": 63.4, "reference_url": "https://deepmind.google/models/gemini/flash/", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "DeepMind /models/gemini/flash/ Performance table: FACTS Benchmark Suite 63.4%", "candidates": [] }, { "model_id": "gemini-2.5-pro", "benchmark_id": "simpleqa_verified", "score": 54.5, "reference_url": "https://deepmind.google/models/gemini/flash/", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "DeepMind /models/gemini/flash/ Performance table: SimpleQA Verified 54.5%", "candidates": [] }, { "model_id": "gemini-2.5-pro", "benchmark_id": "mmmlu", "score": 89.5, "reference_url": "https://deepmind.google/models/gemini/flash/", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "DeepMind /models/gemini/flash/ Performance table: MMMLU 89.5%", "candidates": [] }, { "model_id": "gemini-2.5-pro", "benchmark_id": "global_piqa", "score": 91.5, "reference_url": "https://deepmind.google/models/gemini/flash/", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "DeepMind /models/gemini/flash/ Performance table: Global PIQA 91.5%", "candidates": [] }, { "model_id": "claude-sonnet-4.5", "benchmark_id": "screenspot_pro", "score": 36.2, "reference_url": "https://deepmind.google/models/gemini/flash/", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "DeepMind /models/gemini/flash/ Performance table: ScreenSpot-Pro 36.2%", "candidates": [] }, { "model_id": "claude-sonnet-4.5", "benchmark_id": "charxiv_reasoning", "score": 68.5, "reference_url": "https://deepmind.google/models/gemini/flash/", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "DeepMind /models/gemini/flash/ Performance table: CharXiv Reasoning 68.5%", "candidates": [] }, { "model_id": "claude-sonnet-4.5", "benchmark_id": "omnidocbench", "score": 0.145, "reference_url": "https://deepmind.google/models/gemini/flash/", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "DeepMind /models/gemini/flash/ Performance table: OmniDocBench 1.5 edit distance 0.145", "candidates": [] }, { "model_id": "claude-sonnet-4.5", "benchmark_id": "livecodebench_pro", "score": 1418, "reference_url": "https://deepmind.google/models/gemini/flash/", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "DeepMind /models/gemini/flash/ Performance table: LiveCodeBench Pro Elo 1418", "candidates": [] }, { "model_id": "claude-sonnet-4.5", "benchmark_id": "toolathlon", "score": 38.9, "reference_url": "https://deepmind.google/models/gemini/flash/", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "DeepMind /models/gemini/flash/ Performance table: Toolathlon 38.9%", "candidates": [ { "score": 38.6, "reference_url": "https://huggingface.co/deepseek-ai/DeepSeek-V3.2/resolve/main/assets/paper.pdf", "source_type": "tech_report", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "agentic", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "1.0", "context": "128K" }, "notes": "DS V3.2 tech report Table 2: 38.6. Third-party self-test. Primary (38.9) from Gemini flash page." }, { "score": 41.0, "reference_url": "https://www.anthropic.com/claude-sonnet-5-system-card", "source_type": "model_card", "reported_setting": { "effort": "max", "tools": "agentic benchmark harness", "sampling": "3 trials/task", "harness": "toolathlon" }, "notes": "Anthropic Claude Sonnet 5 System Card Table 8.11.5.A, page 135: Claude Sonnet 4.5; Toolathlon [Pass@1]=41.0. Source setting: effort=max; tools=agentic benchmark harness; sampling=3 trials/task; harness=toolathlon." } ] }, { "model_id": "claude-sonnet-4.5", "benchmark_id": "vending_bench_2", "score": 3839, "reference_url": "https://deepmind.google/models/gemini/flash/", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "DeepMind /models/gemini/flash/ Performance table: Vending-Bench 2 net worth $3839", "candidates": [] }, { "model_id": "claude-sonnet-4.5", "benchmark_id": "facts_benchmark", "score": 48.9, "reference_url": "https://deepmind.google/models/gemini/flash/", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "DeepMind /models/gemini/flash/ Performance table: FACTS Benchmark Suite 48.9%", "candidates": [] }, { "model_id": "claude-sonnet-4.5", "benchmark_id": "simpleqa_verified", "score": 29.3, "reference_url": "https://deepmind.google/models/gemini/flash/", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "DeepMind /models/gemini/flash/ Performance table: SimpleQA Verified 29.3%", "candidates": [] }, { "model_id": "claude-sonnet-4.5", "benchmark_id": "global_piqa", "score": 90.1, "reference_url": "https://deepmind.google/models/gemini/flash/", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "DeepMind /models/gemini/flash/ Performance table: Global PIQA 90.1%", "candidates": [ { "score": 93.9, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "source_type": "model_card", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "notes": "Doubao Seed 2.0 model card Table 3: claude-sonnet-4.5=93.9." } ] }, { "model_id": "gpt-5.2", "benchmark_id": "screenspot_pro", "score": 86.3, "reference_url": "https://deepmind.google/models/gemini/flash/", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "code", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "DeepMind /models/gemini/flash/ Performance table: ScreenSpot-Pro with python 86.3%", "candidates": [] }, { "model_id": "gpt-5.2", "benchmark_id": "vending_bench_2", "score": 3952, "reference_url": "https://deepmind.google/models/gemini/flash/", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "DeepMind /models/gemini/flash/ Performance table: Vending-Bench 2 net worth $3952", "candidates": [ { "score": 3591.33, "reference_url": "https://z.ai/blog/glm-5", "source_type": "official_blog", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "agentic", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "notes": "GLM-5 blog: gpt-5.2=3591.33." } ] }, { "model_id": "gpt-5.2", "benchmark_id": "facts_benchmark", "score": 61.4, "reference_url": "https://deepmind.google/models/gemini/flash/", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "DeepMind /models/gemini/flash/ Performance table: FACTS Benchmark Suite 61.4%", "candidates": [] }, { "model_id": "gpt-5.2", "benchmark_id": "simpleqa_verified", "score": 38.0, "reference_url": "https://deepmind.google/models/gemini/flash/", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "DeepMind /models/gemini/flash/ Performance table: SimpleQA Verified 38.0%", "candidates": [ { "score": 36.8, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "source_type": "model_card", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "notes": "Doubao Seed 2.0 model card Table 3: gpt-5.2=36.8." } ] }, { "model_id": "gpt-5.2", "benchmark_id": "global_piqa", "score": 91.2, "reference_url": "https://deepmind.google/models/gemini/flash/", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "DeepMind /models/gemini/flash/ Performance table: Global PIQA 91.2%", "candidates": [ { "score": 90.9, "reference_url": "https://huggingface.co/Qwen/Qwen3.5-397B-A17B", "source_type": "model_card", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "notes": "Qwen3.5-397B-A17B HF model card cross-model table (Qwen self-test of 3rd-party model, mc=false): alternate measurement 90.9 (3rd-party Qwen self-test)." }, { "score": 93.2, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "source_type": "model_card", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "notes": "Doubao Seed 2.0 model card Table 3: gpt-5.2=93.2." } ] }, { "model_id": "grok-4.1", "benchmark_id": "facts_benchmark", "score": 42.1, "reference_url": "https://deepmind.google/models/gemini/flash/", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "DeepMind /models/gemini/flash/ Performance table: FACTS Benchmark Suite 42.1%", "candidates": [] }, { "model_id": "grok-4.1", "benchmark_id": "simpleqa_verified", "score": 19.5, "reference_url": "https://deepmind.google/models/gemini/flash/", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "DeepMind /models/gemini/flash/ Performance table: SimpleQA Verified 19.5%", "candidates": [] }, { "model_id": "grok-4.1", "benchmark_id": "mmmlu", "score": 86.8, "reference_url": "https://deepmind.google/models/gemini/flash/", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "DeepMind /models/gemini/flash/ Performance table: MMMLU 86.8%", "candidates": [] }, { "model_id": "grok-4.1", "benchmark_id": "global_piqa", "score": 85.6, "reference_url": "https://deepmind.google/models/gemini/flash/", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "DeepMind /models/gemini/flash/ Performance table: Global PIQA 85.6%", "candidates": [] }, { "model_id": "grok-4.1", "benchmark_id": "mrcr_v2", "score": 54.6, "reference_url": "https://deepmind.google/models/gemini/flash/", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default", "context": "128k" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "DeepMind /models/gemini/flash/ Performance table: MRCR v2 8-needle 128k 54.6%", "candidates": [ { "score": 6.1, "reference_url": "https://deepmind.google/models/gemini/flash/", "source_type": "official_blog", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default", "context": "1m" }, "notes": "DeepMind /models/gemini/flash/ Performance table: MRCR v2 1M 6.1%" } ] }, { "model_id": "gemini-2.5-pro", "benchmark_id": "facts_grounding", "score": 87.8, "reference_url": "https://storage.googleapis.com/deepmind-media/Model-Cards/Gemini-2-5-Pro-Model-Card.pdf", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1 (single attempt)", "judge": "rule-based", "harness": "AI Studio API", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Gemini 2.5 Pro Model Card (June 2025 GA): FACTS Grounding 87.8%", "candidates": [] }, { "model_id": "gemini-2.5-pro", "benchmark_id": "video_mme", "score": 86.9, "reference_url": "https://storage.googleapis.com/deepmind-media/Model-Cards/Gemini-2-5-Pro-Model-Card.pdf", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1 (single attempt)", "judge": "rule-based", "harness": "AI Studio API", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Gemini 2.5 Pro Model Card (June 2025 GA): Video-MME (audio+visual+subtitles) 86.9%", "candidates": [] }, { "model_id": "gemini-2.5-pro", "benchmark_id": "aider_polyglot_diff", "score": 82.2, "reference_url": "https://storage.googleapis.com/deepmind-media/Model-Cards/Gemini-2-5-Pro-Model-Card.pdf", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1 (single attempt)", "judge": "rule-based", "harness": "AI Studio API", "prompt_style": "diff-fenced", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Gemini 2.5 Pro Model Card (June 2025 GA): Aider Polyglot diff-fenced 82.2% (avg of 3 trials)", "candidates": [ { "score": 74.0, "reference_url": "https://arxiv.org/abs/2504.13914", "source_type": "tech_report", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "agentic (bash + edit)", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "notes": "Seed-Thinking-v1.5 paper Table 2: gemini-2.5-pro=74.0." } ] }, { "model_id": "gemini-2.5-pro", "benchmark_id": "vibe_eval", "score": 67.2, "reference_url": "https://storage.googleapis.com/deepmind-media/Model-Cards/Gemini-2-5-Pro-Model-Card.pdf", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1 (single attempt)", "judge": "Gemini", "harness": "AI Studio API", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Gemini 2.5 Pro Model Card (June 2025 GA): Vibe-Eval (Reka) 67.2% (Gemini-as-judge)", "candidates": [] }, { "model_id": "gemini-2.5-pro", "benchmark_id": "global_mmlu_lite", "score": 89.2, "reference_url": "https://storage.googleapis.com/deepmind-media/Model-Cards/Gemini-2-5-Pro-Model-Card.pdf", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1 (single attempt)", "judge": "rule-based", "harness": "AI Studio API", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Gemini 2.5 Pro Model Card (GA, June 27 2025): Global MMLU Lite 89.2%.", "candidates": [] }, { "model_id": "gpt-4.1-mini", "benchmark_id": "aime_2024", "score": 49.6, "reference_url": "https://openai.com/index/gpt-4-1/", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-4.1 blog: gpt-4.1-mini=49.6.", "candidates": [] }, { "model_id": "gpt-4.1-nano", "benchmark_id": "aime_2024", "score": 29.4, "reference_url": "https://openai.com/index/gpt-4-1/", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-4.1 blog: gpt-4.1-nano=29.4.", "candidates": [] }, { "model_id": "gpt-4o", "benchmark_id": "aime_2024", "score": 13.1, "reference_url": "https://openai.com/index/gpt-4-1/", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-4.1 blog: gpt-4o=13.1.", "candidates": [] }, { "model_id": "gpt-4o-mini", "benchmark_id": "aime_2024", "score": 8.6, "reference_url": "https://openai.com/index/gpt-4-1/", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-4.1 blog: gpt-4o-mini=8.6.", "candidates": [] }, { "model_id": "o1-high", "benchmark_id": "aime_2024", "score": 74.3, "reference_url": "https://openai.com/index/gpt-4-1/", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-4.1 blog: o1-high=74.3.", "candidates": [ { "score": 74.6, "reference_url": "https://arxiv.org/abs/2504.21318", "source_type": "tech_report", "reported_setting": { "mode": "unknown", "effort": "unknown", "tools": "unknown", "sampling": "unknown", "judge": "rule-based", "harness": "Microsoft Phi internal eval", "prompt_style": "unknown", "temperature": "unknown" }, "notes": "Phi-4-reasoning paper Table 1 (o1-high column): aime_2024=74.6 (alt measurement, mc=false)" }, { "score": 79.2, "reference_url": "https://www.anthropic.com/news/claude-3-7-sonnet", "source_type": "official_blog", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1 (avg 16 trials)", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "notes": "Anthropic 3.7 Sonnet blog (third-party eval): o1 AIME-2024=79.2%/83.3% (parallel). BP primary=74.3 from gpt-4-1 blog." }, { "score": 74.3, "reference_url": "https://llm-stats.com/benchmarks/aime-2024", "source_type": "third_party_aggregator", "reported_setting": { "effort": "default (o1 base, not high)" }, "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=o1, slug=o1-2024-12-17, provider=OpenAI" } ] }, { "model_id": "gpt-4o", "benchmark_id": "gpqa_diamond", "score": 46.0, "reference_url": "https://openai.com/index/gpt-4-1/", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-4.1 blog: gpt-4o=46.0.", "candidates": [ { "score": 52.5, "reference_url": "https://mistral.ai/news/mistral-medium-3", "source_type": "official_blog", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "mistral-eval", "prompt_style": "5-shot CoT", "temperature": "0.0" }, "notes": "Mistral blog (third-party self-test) — 52.5 vs current verified 46.0." } ] }, { "model_id": "gpt-4o-mini", "benchmark_id": "gpqa_diamond", "score": 40.2, "reference_url": "https://openai.com/index/gpt-4-1/", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-4.1 blog: gpt-4o-mini=40.2.", "candidates": [ { "score": 40.9, "reference_url": "https://arxiv.org/abs/2412.08905", "source_type": "tech_report", "reported_setting": { "mode": "unknown", "effort": "unknown", "tools": "unknown", "sampling": "unknown", "judge": "rule-based", "harness": "Microsoft Phi internal eval", "prompt_style": "unknown", "temperature": "unknown" }, "notes": "Phi-4 paper Table 1 (gpt-4o-mini column): gpqa_diamond=40.9 (alt measurement, mc=false)" } ] }, { "model_id": "o1-high", "benchmark_id": "gpqa_diamond", "score": 75.7, "reference_url": "https://openai.com/index/gpt-4-1/", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-4.1 blog: o1-high=75.7.", "candidates": [ { "score": 76.7, "reference_url": "https://arxiv.org/abs/2504.21318", "source_type": "tech_report", "reported_setting": { "mode": "unknown", "effort": "unknown", "tools": "unknown", "sampling": "unknown", "judge": "rule-based", "harness": "Microsoft Phi internal eval", "prompt_style": "unknown", "temperature": "unknown" }, "notes": "Phi-4-reasoning paper Table 1 (o1-high column): gpqa_diamond=76.7 (alt measurement, mc=false)" }, { "score": 78.0, "reference_url": "https://llm-stats.com/benchmarks/gpqa", "source_type": "third_party_aggregator", "reported_setting": { "effort": "default (o1 base, not high)" }, "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=o1, slug=o1-2024-12-17, provider=OpenAI" } ] }, { "model_id": "gpt-4o", "benchmark_id": "mmlu", "score": 85.7, "reference_url": "https://openai.com/index/gpt-4-1/", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-4.1 blog: gpt-4o=85.7.", "candidates": [] }, { "model_id": "gpt-4o-mini", "benchmark_id": "mmlu", "score": 82.0, "reference_url": "https://openai.com/index/gpt-4-1/", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-4.1 blog: gpt-4o-mini=82.0.", "candidates": [ { "score": 81.8, "reference_url": "https://arxiv.org/abs/2412.08905", "source_type": "tech_report", "reported_setting": { "mode": "unknown", "effort": "unknown", "tools": "unknown", "sampling": "unknown", "judge": "rule-based", "harness": "Microsoft Phi internal eval", "prompt_style": "unknown", "temperature": "unknown" }, "notes": "Phi-4 paper Table 1 (gpt-4o-mini column): mmlu=81.8 (alt measurement, mc=false)" }, { "score": 82.2, "reference_url": "https://arxiv.org/abs/2501.00656", "source_type": "tech_report", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "OLMES", "prompt_style": "default", "temperature": "default" }, "notes": "OLMo 2 paper Table 7 Instruct: alt measurement 82.2 (mc=false)." } ] }, { "model_id": "o1-high", "benchmark_id": "mmlu", "score": 91.8, "reference_url": "https://openai.com/index/gpt-4-1/", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-4.1 blog: o1-high=91.8.", "candidates": [ { "score": 91.8, "reference_url": "https://llm-stats.com/benchmarks/mmlu", "source_type": "third_party_aggregator", "reported_setting": { "effort": "default (o1 base, not high)" }, "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=o1, slug=o1-2024-12-17, provider=OpenAI" } ] }, { "model_id": "gpt-4.1", "benchmark_id": "mmmlu", "score": 87.3, "reference_url": "https://openai.com/index/gpt-4-1/", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-4.1 blog: gpt-4.1=87.3.", "candidates": [] }, { "model_id": "gpt-4.1-mini", "benchmark_id": "mmmlu", "score": 78.5, "reference_url": "https://openai.com/index/gpt-4-1/", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-4.1 blog: gpt-4.1-mini=78.5.", "candidates": [] }, { "model_id": "gpt-4.1-nano", "benchmark_id": "mmmlu", "score": 66.9, "reference_url": "https://openai.com/index/gpt-4-1/", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-4.1 blog: gpt-4.1-nano=66.9.", "candidates": [] }, { "model_id": "gpt-4o", "benchmark_id": "mmmlu", "score": 81.4, "reference_url": "https://openai.com/index/gpt-4-1/", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-4.1 blog: gpt-4o=81.4.", "candidates": [ { "score": 81.5, "reference_url": "https://openai.com/index/introducing-gpt-4-5/", "source_type": "official_blog", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "notes": "OpenAI GPT-4.5 blog: gpt-4o=81.5." } ] }, { "model_id": "gpt-4o-mini", "benchmark_id": "mmmlu", "score": 70.5, "reference_url": "https://openai.com/index/gpt-4-1/", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-4.1 blog: gpt-4o-mini=70.5.", "candidates": [] }, { "model_id": "o1-high", "benchmark_id": "mmmlu", "score": 87.7, "reference_url": "https://openai.com/index/gpt-4-1/", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-4.1 blog: o1-high=87.7.", "candidates": [ { "score": 87.7, "reference_url": "https://llm-stats.com/benchmarks/mmmlu", "source_type": "third_party_aggregator", "reported_setting": { "effort": "default (o1 base, not high)" }, "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=o1, slug=o1-2024-12-17, provider=OpenAI" } ] }, { "model_id": "o3-mini-high", "benchmark_id": "mmmlu", "score": 80.7, "reference_url": "https://openai.com/index/gpt-4-1/", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-4.1 blog: o3-mini-high=80.7.", "candidates": [ { "score": 81.1, "reference_url": "https://openai.com/index/introducing-gpt-4-5/", "source_type": "official_blog", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "notes": "OpenAI GPT-4.5 blog: o3-mini-high=81.1." }, { "score": 79.5, "reference_url": "https://www.anthropic.com/news/claude-3-7-sonnet", "source_type": "official_blog", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "notes": "Anthropic 3.7 Sonnet blog (third-party eval): o3-mini-high MMMLU=79.5%. BP primary=80.7." } ] }, { "model_id": "gpt-4.5", "benchmark_id": "mmmlu", "score": 85.1, "reference_url": "https://openai.com/index/gpt-4-1/", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-4.1 blog: gpt-4.5=85.1.", "candidates": [] }, { "model_id": "gpt-4o", "benchmark_id": "swe_bench_verified", "score": 30.7, "reference_url": "https://openai.com/index/introducing-gpt-4-5/", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "agentic (bash + edit)", "sampling": "pass@1", "judge": "rule-based", "harness": "official (GPT-4.5 blog)", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-4.5 blog: gpt-4o=30.7. (GPT-4.1 blog 477-subset value 33.2 dropped — was different N).", "candidates": [] }, { "model_id": "gpt-4o-mini", "benchmark_id": "swe_bench_verified", "score": 8.7, "reference_url": "https://openai.com/index/gpt-4-1/", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "agentic (bash + edit)", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "dropped", "notes": "GPT-4.1 blog value 8.7 was 477/500 subset, not full SWE-bench Verified. No verified full-set value available; dropped. DROPPED: source uses 477/500 subset (omit 23 problems), not full SWE-bench Verified.", "candidates": [] }, { "model_id": "o1-high", "benchmark_id": "swe_bench_verified", "score": 48.9, "reference_url": "https://www.anthropic.com/news/claude-3-7-sonnet", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "agentic (bash+editor)", "sampling": "pass@1", "judge": "rule-based", "harness": "Anthropic simple scaffold", "prompt_style": "default", "temperature": "default" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Anthropic 3.7 Sonnet blog (third-party eval): o1 SWE-bench Verified=48.9%. Note: diff scaffold from OpenAI. Overwrite dropped cell (41.0 from gpt-4-1 blog was dropped).", "candidates": [ { "score": 41.0, "reference_url": "https://llm-stats.com/benchmarks/swe-bench-verified", "source_type": "third_party_aggregator", "reported_setting": { "effort": "default (o1 base, not high)" }, "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=o1, slug=o1-2024-12-17, provider=OpenAI" } ] }, { "model_id": "gpt-4o", "benchmark_id": "aider_polyglot_diff", "score": 18.2, "reference_url": "https://openai.com/index/gpt-4-1/", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "agentic (bash + edit)", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-4.1 blog: gpt-4o=18.2.", "candidates": [] }, { "model_id": "gpt-4o-mini", "benchmark_id": "aider_polyglot_diff", "score": 2.7, "reference_url": "https://openai.com/index/gpt-4-1/", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "agentic (bash + edit)", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-4.1 blog: gpt-4o-mini=2.7.", "candidates": [] }, { "model_id": "o1-high", "benchmark_id": "aider_polyglot_diff", "score": 61.7, "reference_url": "https://openai.com/index/gpt-4-1/", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "agentic (bash + edit)", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-4.1 blog: o1-high=61.7.", "candidates": [] }, { "model_id": "o3-mini-high", "benchmark_id": "aider_polyglot_diff", "score": 60.4, "reference_url": "https://openai.com/index/gpt-4-1/", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "agentic (bash + edit)", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-4.1 blog: o3-mini-high=60.4.", "candidates": [ { "score": 68.6, "reference_url": "https://arxiv.org/abs/2504.13914", "source_type": "tech_report", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "agentic (bash + edit)", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "notes": "Seed-Thinking-v1.5 paper Table 2: o3-mini-high=68.6." } ] }, { "model_id": "gpt-4.5", "benchmark_id": "aider_polyglot_diff", "score": 44.9, "reference_url": "https://openai.com/index/gpt-4-1/", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "agentic (bash + edit)", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-4.1 blog: gpt-4.5=44.9.", "candidates": [] }, { "model_id": "gpt-4.1", "benchmark_id": "aider_polyglot_whole", "score": 51.6, "reference_url": "https://openai.com/index/gpt-4-1/", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "agentic (bash + edit)", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-4.1 blog: gpt-4.1=51.6.", "candidates": [] }, { "model_id": "gpt-4.1-mini", "benchmark_id": "aider_polyglot_whole", "score": 34.7, "reference_url": "https://openai.com/index/gpt-4-1/", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "agentic (bash + edit)", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-4.1 blog: gpt-4.1-mini=34.7.", "candidates": [] }, { "model_id": "gpt-4.1-nano", "benchmark_id": "aider_polyglot_whole", "score": 9.8, "reference_url": "https://openai.com/index/gpt-4-1/", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "agentic (bash + edit)", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-4.1 blog: gpt-4.1-nano=9.8.", "candidates": [] }, { "model_id": "gpt-4o", "benchmark_id": "aider_polyglot_whole", "score": 30.7, "reference_url": "https://openai.com/index/gpt-4-1/", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "agentic (bash + edit)", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-4.1 blog: gpt-4o=30.7.", "candidates": [] }, { "model_id": "gpt-4o-mini", "benchmark_id": "aider_polyglot_whole", "score": 3.6, "reference_url": "https://openai.com/index/gpt-4-1/", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "agentic (bash + edit)", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-4.1 blog: gpt-4o-mini=3.6.", "candidates": [] }, { "model_id": "o1-high", "benchmark_id": "aider_polyglot_whole", "score": 64.6, "reference_url": "https://openai.com/index/gpt-4-1/", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "agentic (bash + edit)", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-4.1 blog: o1-high=64.6.", "candidates": [] }, { "model_id": "o3-mini-high", "benchmark_id": "aider_polyglot_whole", "score": 66.7, "reference_url": "https://openai.com/index/gpt-4-1/", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "agentic (bash + edit)", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-4.1 blog: o3-mini-high=66.7.", "candidates": [] }, { "model_id": "gpt-4o", "benchmark_id": "internal_api_if_hard", "score": 29.2, "reference_url": "https://openai.com/index/gpt-4-1/", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-4.1 blog: gpt-4o=29.2.", "candidates": [] }, { "model_id": "gpt-4o-mini", "benchmark_id": "internal_api_if_hard", "score": 27.2, "reference_url": "https://openai.com/index/gpt-4-1/", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-4.1 blog: gpt-4o-mini=27.2.", "candidates": [] }, { "model_id": "o1-high", "benchmark_id": "internal_api_if_hard", "score": 51.3, "reference_url": "https://openai.com/index/gpt-4-1/", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-4.1 blog: o1-high=51.3.", "candidates": [] }, { "model_id": "o3-mini-high", "benchmark_id": "internal_api_if_hard", "score": 50.0, "reference_url": "https://openai.com/index/gpt-4-1/", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-4.1 blog: o3-mini-high=50.0.", "candidates": [] }, { "model_id": "gpt-4.5", "benchmark_id": "internal_api_if_hard", "score": 54.0, "reference_url": "https://openai.com/index/gpt-4-1/", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-4.1 blog: gpt-4.5=54.0.", "candidates": [] }, { "model_id": "gpt-4.1", "benchmark_id": "multichallenge", "score": 38.3, "reference_url": "https://openai.com/index/gpt-4-1/", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-4.1 blog: gpt-4.1=38.3.", "candidates": [] }, { "model_id": "gpt-4.1-mini", "benchmark_id": "multichallenge", "score": 35.8, "reference_url": "https://openai.com/index/gpt-4-1/", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-4.1 blog: gpt-4.1-mini=35.8.", "candidates": [] }, { "model_id": "gpt-4.1-nano", "benchmark_id": "multichallenge", "score": 15.0, "reference_url": "https://openai.com/index/gpt-4-1/", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-4.1 blog: gpt-4.1-nano=15.0.", "candidates": [] }, { "model_id": "gpt-4o", "benchmark_id": "multichallenge", "score": 27.8, "reference_url": "https://openai.com/index/gpt-4-1/", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-4.1 blog: gpt-4o=27.8.", "candidates": [] }, { "model_id": "gpt-4o-mini", "benchmark_id": "multichallenge", "score": 20.3, "reference_url": "https://openai.com/index/gpt-4-1/", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-4.1 blog: gpt-4o-mini=20.3.", "candidates": [] }, { "model_id": "o1-high", "benchmark_id": "multichallenge", "score": 44.9, "reference_url": "https://openai.com/index/gpt-4-1/", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-4.1 blog: o1-high=44.9.", "candidates": [] }, { "model_id": "o3-mini-high", "benchmark_id": "multichallenge", "score": 39.9, "reference_url": "https://openai.com/index/gpt-4-1/", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-4.1 blog: o3-mini-high=39.9.", "candidates": [] }, { "model_id": "gpt-4.5", "benchmark_id": "multichallenge", "score": 43.8, "reference_url": "https://openai.com/index/gpt-4-1/", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-4.1 blog: gpt-4.5=43.8.", "candidates": [] }, { "model_id": "gpt-4o", "benchmark_id": "multichallenge_o3mini_grader", "score": 39.9, "reference_url": "https://openai.com/index/gpt-4-1/", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-4.1 blog: gpt-4o=39.9.", "candidates": [] }, { "model_id": "gpt-4o-mini", "benchmark_id": "multichallenge_o3mini_grader", "score": 25.6, "reference_url": "https://openai.com/index/gpt-4-1/", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-4.1 blog: gpt-4o-mini=25.6.", "candidates": [] }, { "model_id": "o1-high", "benchmark_id": "multichallenge_o3mini_grader", "score": 52.9, "reference_url": "https://openai.com/index/gpt-4-1/", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-4.1 blog: o1-high=52.9.", "candidates": [] }, { "model_id": "o3-mini-high", "benchmark_id": "multichallenge_o3mini_grader", "score": 50.2, "reference_url": "https://openai.com/index/gpt-4-1/", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-4.1 blog: o3-mini-high=50.2.", "candidates": [] }, { "model_id": "gpt-4.5", "benchmark_id": "multichallenge_o3mini_grader", "score": 50.1, "reference_url": "https://openai.com/index/gpt-4-1/", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-4.1 blog: gpt-4.5=50.1.", "candidates": [] }, { "model_id": "gpt-4o", "benchmark_id": "collie", "score": 50.2, "reference_url": "https://openai.com/index/gpt-4-1/", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-4.1 blog: gpt-4o=50.2.", "candidates": [] }, { "model_id": "gpt-4o-mini", "benchmark_id": "collie", "score": 52.7, "reference_url": "https://openai.com/index/gpt-4-1/", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-4.1 blog: gpt-4o-mini=52.7.", "candidates": [] }, { "model_id": "o1-high", "benchmark_id": "collie", "score": 95.3, "reference_url": "https://openai.com/index/gpt-4-1/", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-4.1 blog: o1-high=95.3.", "candidates": [] }, { "model_id": "o3-mini-high", "benchmark_id": "collie", "score": 98.7, "reference_url": "https://openai.com/index/gpt-4-1/", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-4.1 blog: o3-mini-high=98.7.", "candidates": [ { "score": 87.6, "reference_url": "https://arxiv.org/abs/2504.13914", "source_type": "tech_report", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "notes": "Seed-Thinking-v1.5 paper Table 2: o3-mini-high=87.6." } ] }, { "model_id": "gpt-4.5", "benchmark_id": "collie", "score": 72.3, "reference_url": "https://openai.com/index/gpt-4-1/", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-4.1 blog: gpt-4.5=72.3.", "candidates": [] }, { "model_id": "gpt-4o", "benchmark_id": "ifeval", "score": 81.0, "reference_url": "https://openai.com/index/gpt-4-1/", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-4.1 blog: gpt-4o=81.0.", "candidates": [ { "score": 87.2, "reference_url": "https://mistral.ai/news/mistral-medium-3", "source_type": "official_blog", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "mistral-eval", "prompt_style": "0-shot", "temperature": "0.0" }, "notes": "Mistral blog (third-party self-test) — 87.2 vs current verified 81.0." } ] }, { "model_id": "gpt-4o-mini", "benchmark_id": "ifeval", "score": 78.4, "reference_url": "https://openai.com/index/gpt-4-1/", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-4.1 blog: gpt-4o-mini=78.4.", "candidates": [ { "score": 80.0, "reference_url": "https://arxiv.org/abs/2412.08905", "source_type": "tech_report", "reported_setting": { "mode": "unknown", "effort": "unknown", "tools": "unknown", "sampling": "unknown", "judge": "rule-based", "harness": "Microsoft Phi internal eval", "prompt_style": "unknown", "temperature": "unknown" }, "notes": "Phi-4 paper Table 1 (gpt-4o-mini column): ifeval=80.0 (alt measurement, mc=false)" }, { "score": 83.5, "reference_url": "https://arxiv.org/abs/2501.00656", "source_type": "tech_report", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "OLMES", "prompt_style": "default", "temperature": "default" }, "notes": "OLMo 2 paper Table 7 Instruct: alt measurement 83.5 (mc=false)." } ] }, { "model_id": "o1-high", "benchmark_id": "ifeval", "score": 92.2, "reference_url": "https://openai.com/index/gpt-4-1/", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-4.1 blog: o1-high=92.2.", "candidates": [] }, { "model_id": "gpt-4.1", "benchmark_id": "multi_if", "score": 70.8, "reference_url": "https://openai.com/index/gpt-4-1/", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-4.1 blog: gpt-4.1=70.8.", "candidates": [] }, { "model_id": "gpt-4.1-mini", "benchmark_id": "multi_if", "score": 67.0, "reference_url": "https://openai.com/index/gpt-4-1/", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-4.1 blog: gpt-4.1-mini=67.0.", "candidates": [] }, { "model_id": "gpt-4.1-nano", "benchmark_id": "multi_if", "score": 57.2, "reference_url": "https://openai.com/index/gpt-4-1/", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-4.1 blog: gpt-4.1-nano=57.2.", "candidates": [] }, { "model_id": "gpt-4o", "benchmark_id": "multi_if", "score": 60.9, "reference_url": "https://openai.com/index/gpt-4-1/", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-4.1 blog: gpt-4o=60.9.", "candidates": [] }, { "model_id": "gpt-4o-mini", "benchmark_id": "multi_if", "score": 57.9, "reference_url": "https://openai.com/index/gpt-4-1/", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-4.1 blog: gpt-4o-mini=57.9.", "candidates": [] }, { "model_id": "o1-high", "benchmark_id": "multi_if", "score": 77.9, "reference_url": "https://openai.com/index/gpt-4-1/", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-4.1 blog: o1-high=77.9.", "candidates": [] }, { "model_id": "o3-mini-high", "benchmark_id": "multi_if", "score": 79.5, "reference_url": "https://openai.com/index/gpt-4-1/", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-4.1 blog: o3-mini-high=79.5.", "candidates": [] }, { "model_id": "gpt-4.5", "benchmark_id": "multi_if", "score": 70.8, "reference_url": "https://openai.com/index/gpt-4-1/", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-4.1 blog: gpt-4.5=70.8.", "candidates": [] }, { "model_id": "gpt-4.1", "benchmark_id": "mrcr_v2_2needle_128k", "score": 57.2, "reference_url": "https://openai.com/index/gpt-4-1/", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-4.1 blog: gpt-4.1=57.2.", "candidates": [] }, { "model_id": "gpt-4.1-mini", "benchmark_id": "mrcr_v2_2needle_128k", "score": 47.2, "reference_url": "https://openai.com/index/gpt-4-1/", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-4.1 blog: gpt-4.1-mini=47.2.", "candidates": [] }, { "model_id": "gpt-4.1-nano", "benchmark_id": "mrcr_v2_2needle_128k", "score": 36.6, "reference_url": "https://openai.com/index/gpt-4-1/", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-4.1 blog: gpt-4.1-nano=36.6.", "candidates": [] }, { "model_id": "gpt-4o", "benchmark_id": "mrcr_v2_2needle_128k", "score": 31.9, "reference_url": "https://openai.com/index/gpt-4-1/", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-4.1 blog: gpt-4o=31.9.", "candidates": [] }, { "model_id": "gpt-4o-mini", "benchmark_id": "mrcr_v2_2needle_128k", "score": 24.5, "reference_url": "https://openai.com/index/gpt-4-1/", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-4.1 blog: gpt-4o-mini=24.5.", "candidates": [] }, { "model_id": "o1-high", "benchmark_id": "mrcr_v2_2needle_128k", "score": 22.1, "reference_url": "https://openai.com/index/gpt-4-1/", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-4.1 blog: o1-high=22.1.", "candidates": [] }, { "model_id": "o3-mini-high", "benchmark_id": "mrcr_v2_2needle_128k", "score": 18.7, "reference_url": "https://openai.com/index/gpt-4-1/", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-4.1 blog: o3-mini-high=18.7.", "candidates": [] }, { "model_id": "gpt-4.5", "benchmark_id": "mrcr_v2_2needle_128k", "score": 38.5, "reference_url": "https://openai.com/index/gpt-4-1/", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-4.1 blog: gpt-4.5=38.5.", "candidates": [] }, { "model_id": "gpt-4.1", "benchmark_id": "mrcr_v2_2needle_1m", "score": 46.3, "reference_url": "https://openai.com/index/gpt-4-1/", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-4.1 blog: gpt-4.1=46.3.", "candidates": [] }, { "model_id": "gpt-4.1-mini", "benchmark_id": "mrcr_v2_2needle_1m", "score": 33.3, "reference_url": "https://openai.com/index/gpt-4-1/", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-4.1 blog: gpt-4.1-mini=33.3.", "candidates": [] }, { "model_id": "gpt-4.1-nano", "benchmark_id": "mrcr_v2_2needle_1m", "score": 12.0, "reference_url": "https://openai.com/index/gpt-4-1/", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-4.1 blog: gpt-4.1-nano=12.0.", "candidates": [] }, { "model_id": "gpt-4o", "benchmark_id": "graphwalks_bfs_0k_128k", "score": 41.7, "reference_url": "https://openai.com/index/gpt-4-1/", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-4.1 blog: gpt-4o=41.7.", "candidates": [] }, { "model_id": "gpt-4o-mini", "benchmark_id": "graphwalks_bfs_0k_128k", "score": 29.0, "reference_url": "https://openai.com/index/gpt-4-1/", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-4.1 blog: gpt-4o-mini=29.0.", "candidates": [] }, { "model_id": "o1-high", "benchmark_id": "graphwalks_bfs_0k_128k", "score": 62.0, "reference_url": "https://openai.com/index/gpt-4-1/", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-4.1 blog: o1-high=62.0.", "candidates": [] }, { "model_id": "o3-mini-high", "benchmark_id": "graphwalks_bfs_0k_128k", "score": 51.0, "reference_url": "https://openai.com/index/gpt-4-1/", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-4.1 blog: o3-mini-high=51.0.", "candidates": [] }, { "model_id": "gpt-4.5", "benchmark_id": "graphwalks_bfs_0k_128k", "score": 72.3, "reference_url": "https://openai.com/index/gpt-4-1/", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-4.1 blog: gpt-4.5=72.3.", "candidates": [] }, { "model_id": "gpt-4.1", "benchmark_id": "graphwalks_bfs_128k_plus", "score": 19.0, "reference_url": "https://openai.com/index/gpt-4-1/", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-4.1 blog: gpt-4.1=19.0.", "candidates": [] }, { "model_id": "gpt-4.1-mini", "benchmark_id": "graphwalks_bfs_128k_plus", "score": 15.0, "reference_url": "https://openai.com/index/gpt-4-1/", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-4.1 blog: gpt-4.1-mini=15.0.", "candidates": [] }, { "model_id": "gpt-4.1-nano", "benchmark_id": "graphwalks_bfs_128k_plus", "score": 2.9, "reference_url": "https://openai.com/index/gpt-4-1/", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-4.1 blog: gpt-4.1-nano=2.9.", "candidates": [] }, { "model_id": "gpt-4o", "benchmark_id": "graphwalks_parents_0k_128k", "score": 35.4, "reference_url": "https://openai.com/index/gpt-4-1/", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-4.1 blog: gpt-4o=35.4.", "candidates": [] }, { "model_id": "gpt-4o-mini", "benchmark_id": "graphwalks_parents_0k_128k", "score": 12.6, "reference_url": "https://openai.com/index/gpt-4-1/", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-4.1 blog: gpt-4o-mini=12.6.", "candidates": [] }, { "model_id": "o1-high", "benchmark_id": "graphwalks_parents_0k_128k", "score": 50.9, "reference_url": "https://openai.com/index/gpt-4-1/", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-4.1 blog: o1-high=50.9.", "candidates": [] }, { "model_id": "o3-mini-high", "benchmark_id": "graphwalks_parents_0k_128k", "score": 58.3, "reference_url": "https://openai.com/index/gpt-4-1/", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-4.1 blog: o3-mini-high=58.3.", "candidates": [] }, { "model_id": "gpt-4.5", "benchmark_id": "graphwalks_parents_0k_128k", "score": 72.6, "reference_url": "https://openai.com/index/gpt-4-1/", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-4.1 blog: gpt-4.5=72.6.", "candidates": [] }, { "model_id": "gpt-4.1", "benchmark_id": "graphwalks_parents_128k_plus", "score": 25.0, "reference_url": "https://openai.com/index/gpt-4-1/", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-4.1 blog: gpt-4.1=25.0.", "candidates": [] }, { "model_id": "gpt-4.1-mini", "benchmark_id": "graphwalks_parents_128k_plus", "score": 11.0, "reference_url": "https://openai.com/index/gpt-4-1/", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-4.1 blog: gpt-4.1-mini=11.0.", "candidates": [] }, { "model_id": "gpt-4.1-nano", "benchmark_id": "graphwalks_parents_128k_plus", "score": 5.6, "reference_url": "https://openai.com/index/gpt-4-1/", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-4.1 blog: gpt-4.1-nano=5.6.", "candidates": [] }, { "model_id": "gpt-4o", "benchmark_id": "mmmu", "score": 68.7, "reference_url": "https://openai.com/index/gpt-4-1/", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-4.1 blog: gpt-4o=68.7.", "candidates": [ { "score": 69.1, "reference_url": "https://openai.com/index/introducing-gpt-4-5/", "source_type": "official_blog", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "notes": "OpenAI GPT-4.5 blog: gpt-4o=69.1." } ] }, { "model_id": "gpt-4o-mini", "benchmark_id": "mmmu", "score": 56.3, "reference_url": "https://openai.com/index/gpt-4-1/", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-4.1 blog: gpt-4o-mini=56.3.", "candidates": [] }, { "model_id": "o1-high", "benchmark_id": "mmmu", "score": 77.6, "reference_url": "https://openai.com/index/gpt-4-1/", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-4.1 blog: o1-high=77.6.", "candidates": [ { "score": 78.2, "reference_url": "https://www.anthropic.com/news/claude-3-7-sonnet", "source_type": "official_blog", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "zeroshot-cot", "temperature": "default" }, "notes": "Anthropic 3.7 Sonnet blog (third-party eval): o1 MMMU=78.2%; primary is 77.6 from GPT-4.1 blog" }, { "score": 77.6, "reference_url": "https://llm-stats.com/benchmarks/mmmu", "source_type": "third_party_aggregator", "reported_setting": { "effort": "default (o1 base, not high)" }, "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=o1, slug=o1-2024-12-17, provider=OpenAI" } ] }, { "model_id": "gpt-4.5", "benchmark_id": "mmmu", "score": 75.2, "reference_url": "https://openai.com/index/gpt-4-1/", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-4.1 blog: gpt-4.5=75.2.", "candidates": [ { "score": 74.4, "reference_url": "https://openai.com/index/introducing-gpt-4-5/", "source_type": "official_blog", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "notes": "OpenAI GPT-4.5 blog: gpt-4.5=74.4." } ] }, { "model_id": "gpt-4.1", "benchmark_id": "mathvista", "score": 72.2, "reference_url": "https://openai.com/index/gpt-4-1/", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-4.1 blog: gpt-4.1=72.2.", "candidates": [] }, { "model_id": "gpt-4.1-mini", "benchmark_id": "mathvista", "score": 73.1, "reference_url": "https://openai.com/index/gpt-4-1/", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-4.1 blog: gpt-4.1-mini=73.1.", "candidates": [] }, { "model_id": "gpt-4.1-nano", "benchmark_id": "mathvista", "score": 56.2, "reference_url": "https://openai.com/index/gpt-4-1/", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-4.1 blog: gpt-4.1-nano=56.2.", "candidates": [] }, { "model_id": "gpt-4o", "benchmark_id": "mathvista", "score": 61.4, "reference_url": "https://openai.com/index/gpt-4-1/", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-4.1 blog: gpt-4o=61.4.", "candidates": [] }, { "model_id": "gpt-4o-mini", "benchmark_id": "mathvista", "score": 56.5, "reference_url": "https://openai.com/index/gpt-4-1/", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-4.1 blog: gpt-4o-mini=56.5.", "candidates": [] }, { "model_id": "o1-high", "benchmark_id": "mathvista", "score": 71.8, "reference_url": "https://openai.com/index/gpt-4-1/", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-4.1 blog: o1-high=71.8.", "candidates": [ { "score": 71.8, "reference_url": "https://llm-stats.com/benchmarks/mathvista", "source_type": "third_party_aggregator", "reported_setting": { "effort": "default (o1 base, not high)" }, "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=o1, slug=o1-2024-12-17, provider=OpenAI" } ] }, { "model_id": "gpt-4.5", "benchmark_id": "mathvista", "score": 72.3, "reference_url": "https://openai.com/index/gpt-4-1/", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-4.1 blog: gpt-4.5=72.3.", "candidates": [] }, { "model_id": "gpt-4o", "benchmark_id": "charxiv_reasoning", "score": 52.7, "reference_url": "https://openai.com/index/gpt-4-1/", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-4.1 blog: gpt-4o=52.7.", "candidates": [] }, { "model_id": "gpt-4o-mini", "benchmark_id": "charxiv_reasoning", "score": 36.8, "reference_url": "https://openai.com/index/gpt-4-1/", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-4.1 blog: gpt-4o-mini=36.8.", "candidates": [] }, { "model_id": "o1-high", "benchmark_id": "charxiv_reasoning", "score": 55.1, "reference_url": "https://openai.com/index/gpt-4-1/", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-4.1 blog: o1-high=55.1.", "candidates": [] }, { "model_id": "gpt-4.5", "benchmark_id": "charxiv_reasoning", "score": 55.4, "reference_url": "https://openai.com/index/gpt-4-1/", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-4.1 blog: gpt-4.5=55.4.", "candidates": [] }, { "model_id": "gpt-4.1", "benchmark_id": "charxiv_descriptive", "score": 87.9, "reference_url": "https://openai.com/index/gpt-4-1/", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-4.1 blog: gpt-4.1=87.9.", "candidates": [] }, { "model_id": "gpt-4.1-mini", "benchmark_id": "charxiv_descriptive", "score": 88.4, "reference_url": "https://openai.com/index/gpt-4-1/", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-4.1 blog: gpt-4.1-mini=88.4.", "candidates": [] }, { "model_id": "gpt-4.1-nano", "benchmark_id": "charxiv_descriptive", "score": 73.9, "reference_url": "https://openai.com/index/gpt-4-1/", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-4.1 blog: gpt-4.1-nano=73.9.", "candidates": [] }, { "model_id": "gpt-4o", "benchmark_id": "charxiv_descriptive", "score": 85.3, "reference_url": "https://openai.com/index/gpt-4-1/", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-4.1 blog: gpt-4o=85.3.", "candidates": [] }, { "model_id": "gpt-4o-mini", "benchmark_id": "charxiv_descriptive", "score": 76.6, "reference_url": "https://openai.com/index/gpt-4-1/", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-4.1 blog: gpt-4o-mini=76.6.", "candidates": [] }, { "model_id": "o1-high", "benchmark_id": "charxiv_descriptive", "score": 88.9, "reference_url": "https://openai.com/index/gpt-4-1/", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-4.1 blog: o1-high=88.9.", "candidates": [] }, { "model_id": "gpt-4.5", "benchmark_id": "charxiv_descriptive", "score": 90.0, "reference_url": "https://openai.com/index/gpt-4-1/", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-4.1 blog: gpt-4.5=90.0.", "candidates": [] }, { "model_id": "gpt-4.1", "benchmark_id": "complexfuncbench", "score": 65.5, "reference_url": "https://openai.com/index/gpt-4-1/", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "agentic (function calling)", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-4.1 blog: gpt-4.1=65.5.", "candidates": [] }, { "model_id": "gpt-4.1-mini", "benchmark_id": "complexfuncbench", "score": 49.3, "reference_url": "https://openai.com/index/gpt-4-1/", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "agentic (function calling)", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-4.1 blog: gpt-4.1-mini=49.3.", "candidates": [] }, { "model_id": "gpt-4.1-nano", "benchmark_id": "complexfuncbench", "score": 5.7, "reference_url": "https://openai.com/index/gpt-4-1/", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "agentic (function calling)", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-4.1 blog: gpt-4.1-nano=5.7.", "candidates": [] }, { "model_id": "gpt-4o", "benchmark_id": "complexfuncbench", "score": 66.5, "reference_url": "https://openai.com/index/gpt-4-1/", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "agentic (function calling)", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-4.1 blog: gpt-4o=66.5.", "candidates": [] }, { "model_id": "gpt-4o-mini", "benchmark_id": "complexfuncbench", "score": 38.6, "reference_url": "https://openai.com/index/gpt-4-1/", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "agentic (function calling)", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-4.1 blog: gpt-4o-mini=38.6.", "candidates": [] }, { "model_id": "o1-high", "benchmark_id": "complexfuncbench", "score": 47.6, "reference_url": "https://openai.com/index/gpt-4-1/", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "agentic (function calling)", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-4.1 blog: o1-high=47.6.", "candidates": [] }, { "model_id": "o3-mini-high", "benchmark_id": "complexfuncbench", "score": 17.6, "reference_url": "https://openai.com/index/gpt-4-1/", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "agentic (function calling)", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-4.1 blog: o3-mini-high=17.6.", "candidates": [] }, { "model_id": "gpt-4.5", "benchmark_id": "complexfuncbench", "score": 63.0, "reference_url": "https://openai.com/index/gpt-4-1/", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "agentic (function calling)", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-4.1 blog: gpt-4.5=63.0.", "candidates": [] }, { "model_id": "gpt-4.1", "benchmark_id": "tau_bench_airline", "score": 49.4, "reference_url": "https://openai.com/index/gpt-4-1/", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "agentic", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-4.1 blog: gpt-4.1=49.4.", "candidates": [] }, { "model_id": "gpt-4.1-mini", "benchmark_id": "tau_bench_airline", "score": 36.0, "reference_url": "https://openai.com/index/gpt-4-1/", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "agentic", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-4.1 blog: gpt-4.1-mini=36.0.", "candidates": [] }, { "model_id": "gpt-4.1-nano", "benchmark_id": "tau_bench_airline", "score": 14.0, "reference_url": "https://openai.com/index/gpt-4-1/", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "agentic", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-4.1 blog: gpt-4.1-nano=14.0.", "candidates": [] }, { "model_id": "gpt-4o", "benchmark_id": "tau_bench_airline", "score": 42.8, "reference_url": "https://openai.com/index/gpt-4-1/", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "agentic", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-4.1 blog: gpt-4o=42.8.", "candidates": [] }, { "model_id": "gpt-4o-mini", "benchmark_id": "tau_bench_airline", "score": 22.0, "reference_url": "https://openai.com/index/gpt-4-1/", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "agentic", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-4.1 blog: gpt-4o-mini=22.0.", "candidates": [] }, { "model_id": "o1-high", "benchmark_id": "tau_bench_airline", "score": 50.0, "reference_url": "https://openai.com/index/gpt-4-1/", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "agentic", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-4.1 blog: o1-high=50.0.", "candidates": [ { "score": 54.2, "reference_url": "https://www.anthropic.com/news/claude-3-7-sonnet", "source_type": "official_blog", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "agentic (tau-bench official)", "sampling": "pass@1", "judge": "official harness", "harness": "official", "prompt_style": "default", "temperature": "default" }, "notes": "Anthropic 3.7 Sonnet blog (third-party eval): o1 TAU-bench Airline=54.2%. Footnote: results may not be comparable (OpenAI deleted original report)." }, { "score": 50.0, "reference_url": "https://llm-stats.com/benchmarks/tau-bench-airline", "source_type": "third_party_aggregator", "reported_setting": { "effort": "default (o1 base, not high)" }, "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=o1, slug=o1-2024-12-17, provider=OpenAI" } ] }, { "model_id": "o3-mini-high", "benchmark_id": "tau_bench_airline", "score": 32.4, "reference_url": "https://openai.com/index/gpt-4-1/", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "agentic", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-4.1 blog: o3-mini-high=32.4.", "candidates": [] }, { "model_id": "gpt-4.5", "benchmark_id": "tau_bench_airline", "score": 50.0, "reference_url": "https://openai.com/index/gpt-4-1/", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "agentic", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-4.1 blog: gpt-4.5=50.0.", "candidates": [] }, { "model_id": "gpt-4.1", "benchmark_id": "tau_bench_retail", "score": 68.0, "reference_url": "https://openai.com/index/gpt-4-1/", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "agentic", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-4.1 blog: gpt-4.1=68.0.", "candidates": [] }, { "model_id": "gpt-4.1-mini", "benchmark_id": "tau_bench_retail", "score": 55.8, "reference_url": "https://openai.com/index/gpt-4-1/", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "agentic", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-4.1 blog: gpt-4.1-mini=55.8.", "candidates": [] }, { "model_id": "gpt-4.1-nano", "benchmark_id": "tau_bench_retail", "score": 22.6, "reference_url": "https://openai.com/index/gpt-4-1/", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "agentic", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-4.1 blog: gpt-4.1-nano=22.6.", "candidates": [] }, { "model_id": "gpt-4o", "benchmark_id": "tau_bench_retail", "score": 60.3, "reference_url": "https://openai.com/index/gpt-4-1/", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "agentic", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-4.1 blog: gpt-4o=60.3.", "candidates": [] }, { "model_id": "gpt-4o-mini", "benchmark_id": "tau_bench_retail", "score": 44.0, "reference_url": "https://openai.com/index/gpt-4-1/", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "agentic", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-4.1 blog: gpt-4o-mini=44.0.", "candidates": [] }, { "model_id": "o1-high", "benchmark_id": "tau_bench_retail", "score": 70.8, "reference_url": "https://openai.com/index/gpt-4-1/", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "agentic", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-4.1 blog: o1-high=70.8.", "candidates": [ { "score": 73.5, "reference_url": "https://www.anthropic.com/news/claude-3-7-sonnet", "source_type": "official_blog", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "agentic (tau-bench official)", "sampling": "pass@1", "judge": "official harness", "harness": "official", "prompt_style": "default", "temperature": "default" }, "notes": "Anthropic 3.7 Sonnet blog (third-party eval): o1 TAU-bench Retail=73.5%. Footnote: originally reported by OpenAI but later deleted. May not be comparable." }, { "score": 70.8, "reference_url": "https://llm-stats.com/benchmarks/tau-bench-retail", "source_type": "third_party_aggregator", "reported_setting": { "effort": "default (o1 base, not high)" }, "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=o1, slug=o1-2024-12-17, provider=OpenAI" } ] }, { "model_id": "o3-mini-high", "benchmark_id": "tau_bench_retail", "score": 57.6, "reference_url": "https://openai.com/index/gpt-4-1/", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "agentic", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-4.1 blog: o3-mini-high=57.6.", "candidates": [] }, { "model_id": "gpt-4.5", "benchmark_id": "tau_bench_retail", "score": 68.4, "reference_url": "https://openai.com/index/gpt-4-1/", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "agentic", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-4.1 blog: gpt-4.5=68.4.", "candidates": [] }, { "model_id": "gemini-1.5-flash", "benchmark_id": "mmlu_pro", "score": 67.3, "reference_url": "https://storage.googleapis.com/deepmind-media/Model-Cards/Gemini-2-0-Flash-Model-Card.pdf", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Gemini 2.0 Flash Model Card (April 15 2025): MMLU-Pro 67.3%", "candidates": [] }, { "model_id": "gemini-1.5-pro", "benchmark_id": "mmlu_pro", "score": 75.8, "reference_url": "https://storage.googleapis.com/deepmind-media/Model-Cards/Gemini-2-0-Flash-Model-Card.pdf", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Gemini 2.0 Flash Model Card (April 15 2025): MMLU-Pro 75.8%", "candidates": [] }, { "model_id": "gemini-2.0-flash-lite", "benchmark_id": "mmlu_pro", "score": 71.6, "reference_url": "https://storage.googleapis.com/deepmind-media/Model-Cards/Gemini-2-0-Flash-Model-Card.pdf", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Gemini 2.0 Flash Model Card (April 15 2025): MMLU-Pro 71.6%", "candidates": [] }, { "model_id": "gemini-1.5-flash", "benchmark_id": "livecodebench", "score": 30.7, "reference_url": "https://storage.googleapis.com/deepmind-media/Model-Cards/Gemini-2-0-Flash-Model-Card.pdf", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Gemini 2.0 Flash Model Card (April 15 2025): LiveCodeBench v5 (UI 10/01/2024-02/01/2025) 30.7%", "candidates": [] }, { "model_id": "gemini-1.5-pro", "benchmark_id": "livecodebench", "score": 34.2, "reference_url": "https://storage.googleapis.com/deepmind-media/Model-Cards/Gemini-2-0-Flash-Model-Card.pdf", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Gemini 2.0 Flash Model Card (April 15 2025): LiveCodeBench v5 (UI 10/01/2024-02/01/2025) 34.2%", "candidates": [] }, { "model_id": "gemini-2.0-flash-lite", "benchmark_id": "livecodebench", "score": 28.9, "reference_url": "https://storage.googleapis.com/deepmind-media/Model-Cards/Gemini-2-0-Flash-Model-Card.pdf", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Gemini 2.0 Flash Model Card (April 15 2025): LiveCodeBench v5 (UI 10/01/2024-02/01/2025) 28.9%", "candidates": [] }, { "model_id": "gemini-1.5-flash", "benchmark_id": "bird_sql", "score": 45.6, "reference_url": "https://storage.googleapis.com/deepmind-media/Model-Cards/Gemini-2-0-Flash-Model-Card.pdf", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Gemini 2.0 Flash Model Card (April 15 2025): Bird-SQL Dev (NL->SQL) 45.6%", "candidates": [] }, { "model_id": "gemini-1.5-pro", "benchmark_id": "bird_sql", "score": 54.4, "reference_url": "https://storage.googleapis.com/deepmind-media/Model-Cards/Gemini-2-0-Flash-Model-Card.pdf", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Gemini 2.0 Flash Model Card (April 15 2025): Bird-SQL Dev (NL->SQL) 54.4%", "candidates": [] }, { "model_id": "gemini-2.0-flash-lite", "benchmark_id": "bird_sql", "score": 57.4, "reference_url": "https://storage.googleapis.com/deepmind-media/Model-Cards/Gemini-2-0-Flash-Model-Card.pdf", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Gemini 2.0 Flash Model Card (April 15 2025): Bird-SQL Dev (NL->SQL) 57.4%", "candidates": [] }, { "model_id": "gemini-2.0-flash", "benchmark_id": "bird_sql", "score": 58.7, "reference_url": "https://storage.googleapis.com/deepmind-media/Model-Cards/Gemini-2-0-Flash-Model-Card.pdf", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Gemini 2.0 Flash Model Card (April 15 2025): Bird-SQL Dev (NL->SQL) 58.7%", "candidates": [] }, { "model_id": "gemini-1.5-flash", "benchmark_id": "gpqa_diamond", "score": 51.0, "reference_url": "https://storage.googleapis.com/deepmind-media/Model-Cards/Gemini-2-0-Flash-Model-Card.pdf", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Gemini 2.0 Flash Model Card (April 15 2025): GPQA Diamond 51.0%", "candidates": [] }, { "model_id": "gemini-1.5-pro", "benchmark_id": "gpqa_diamond", "score": 59.1, "reference_url": "https://storage.googleapis.com/deepmind-media/Model-Cards/Gemini-2-0-Flash-Model-Card.pdf", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Gemini 2.0 Flash Model Card (April 15 2025): GPQA Diamond 59.1%", "candidates": [] }, { "model_id": "gemini-2.0-flash-lite", "benchmark_id": "gpqa_diamond", "score": 51.5, "reference_url": "https://storage.googleapis.com/deepmind-media/Model-Cards/Gemini-2-0-Flash-Model-Card.pdf", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Gemini 2.0 Flash Model Card (April 15 2025): GPQA Diamond 51.5%", "candidates": [] }, { "model_id": "gemini-1.5-flash", "benchmark_id": "simpleqa", "score": 8.6, "reference_url": "https://storage.googleapis.com/deepmind-media/Model-Cards/Gemini-2-0-Flash-Model-Card.pdf", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Gemini 2.0 Flash Model Card (April 15 2025): SimpleQA (no search) 8.6%", "candidates": [] }, { "model_id": "gemini-1.5-pro", "benchmark_id": "simpleqa", "score": 24.9, "reference_url": "https://storage.googleapis.com/deepmind-media/Model-Cards/Gemini-2-0-Flash-Model-Card.pdf", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Gemini 2.0 Flash Model Card (April 15 2025): SimpleQA (no search) 24.9%", "candidates": [] }, { "model_id": "gemini-2.0-flash-lite", "benchmark_id": "simpleqa", "score": 21.7, "reference_url": "https://storage.googleapis.com/deepmind-media/Model-Cards/Gemini-2-0-Flash-Model-Card.pdf", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Gemini 2.0 Flash Model Card (April 15 2025): SimpleQA (no search) 21.7%", "candidates": [] }, { "model_id": "gemini-1.5-flash", "benchmark_id": "facts_grounding", "score": 82.9, "reference_url": "https://storage.googleapis.com/deepmind-media/Model-Cards/Gemini-2-0-Flash-Model-Card.pdf", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Gemini 2.0 Flash Model Card (April 15 2025): FACTS Grounding 82.9%", "candidates": [] }, { "model_id": "gemini-1.5-pro", "benchmark_id": "facts_grounding", "score": 80.0, "reference_url": "https://storage.googleapis.com/deepmind-media/Model-Cards/Gemini-2-0-Flash-Model-Card.pdf", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Gemini 2.0 Flash Model Card (April 15 2025): FACTS Grounding 80.0%", "candidates": [] }, { "model_id": "gemini-2.0-flash-lite", "benchmark_id": "facts_grounding", "score": 83.6, "reference_url": "https://storage.googleapis.com/deepmind-media/Model-Cards/Gemini-2-0-Flash-Model-Card.pdf", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Gemini 2.0 Flash Model Card (April 15 2025): FACTS Grounding 83.6%", "candidates": [] }, { "model_id": "gemini-2.0-flash", "benchmark_id": "facts_grounding", "score": 84.6, "reference_url": "https://storage.googleapis.com/deepmind-media/Model-Cards/Gemini-2-0-Flash-Model-Card.pdf", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Gemini 2.0 Flash Model Card (April 15 2025): FACTS Grounding 84.6%", "candidates": [] }, { "model_id": "gemini-1.5-flash", "benchmark_id": "global_mmlu_lite", "score": 73.7, "reference_url": "https://storage.googleapis.com/deepmind-media/Model-Cards/Gemini-2-0-Flash-Model-Card.pdf", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Gemini 2.0 Flash Model Card (April 15 2025): Global MMLU Lite 73.7%", "candidates": [] }, { "model_id": "gemini-1.5-pro", "benchmark_id": "global_mmlu_lite", "score": 80.8, "reference_url": "https://storage.googleapis.com/deepmind-media/Model-Cards/Gemini-2-0-Flash-Model-Card.pdf", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Gemini 2.0 Flash Model Card (April 15 2025): Global MMLU Lite 80.8%", "candidates": [] }, { "model_id": "gemini-2.0-flash-lite", "benchmark_id": "global_mmlu_lite", "score": 78.2, "reference_url": "https://storage.googleapis.com/deepmind-media/Model-Cards/Gemini-2-0-Flash-Model-Card.pdf", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Gemini 2.0 Flash Model Card (April 15 2025): Global MMLU Lite 78.2%", "candidates": [] }, { "model_id": "gemini-2.0-flash", "benchmark_id": "global_mmlu_lite", "score": 83.4, "reference_url": "https://storage.googleapis.com/deepmind-media/Model-Cards/Gemini-2-0-Flash-Model-Card.pdf", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Gemini 2.0 Flash Model Card (April 15 2025): Global MMLU Lite 83.4%", "candidates": [] }, { "model_id": "gemini-1.5-flash", "benchmark_id": "math", "score": 77.9, "reference_url": "https://storage.googleapis.com/deepmind-media/Model-Cards/Gemini-2-0-Flash-Model-Card.pdf", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Gemini 2.0 Flash Model Card (April 15 2025): MATH (full) 77.9%", "candidates": [] }, { "model_id": "gemini-1.5-pro", "benchmark_id": "math", "score": 86.5, "reference_url": "https://storage.googleapis.com/deepmind-media/Model-Cards/Gemini-2-0-Flash-Model-Card.pdf", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Gemini 2.0 Flash Model Card (April 15 2025): MATH (full) 86.5%", "candidates": [] }, { "model_id": "gemini-2.0-flash-lite", "benchmark_id": "math", "score": 86.8, "reference_url": "https://storage.googleapis.com/deepmind-media/Model-Cards/Gemini-2-0-Flash-Model-Card.pdf", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Gemini 2.0 Flash Model Card (April 15 2025): MATH (full) 86.8%", "candidates": [] }, { "model_id": "gemini-2.0-flash", "benchmark_id": "math", "score": 90.9, "reference_url": "https://storage.googleapis.com/deepmind-media/Model-Cards/Gemini-2-0-Flash-Model-Card.pdf", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Gemini 2.0 Flash Model Card (April 15 2025): MATH (full) 90.9%", "candidates": [] }, { "model_id": "gemini-1.5-flash", "benchmark_id": "hiddenmath", "score": 47.2, "reference_url": "https://storage.googleapis.com/deepmind-media/Model-Cards/Gemini-2-0-Flash-Model-Card.pdf", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Gemini 2.0 Flash Model Card (April 15 2025): HiddenMath 47.2%", "candidates": [] }, { "model_id": "gemini-1.5-pro", "benchmark_id": "hiddenmath", "score": 52.0, "reference_url": "https://storage.googleapis.com/deepmind-media/Model-Cards/Gemini-2-0-Flash-Model-Card.pdf", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Gemini 2.0 Flash Model Card (April 15 2025): HiddenMath 52.0%", "candidates": [] }, { "model_id": "gemini-2.0-flash-lite", "benchmark_id": "hiddenmath", "score": 55.3, "reference_url": "https://storage.googleapis.com/deepmind-media/Model-Cards/Gemini-2-0-Flash-Model-Card.pdf", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Gemini 2.0 Flash Model Card (April 15 2025): HiddenMath 55.3%", "candidates": [] }, { "model_id": "gemini-2.0-flash", "benchmark_id": "hiddenmath", "score": 63.5, "reference_url": "https://storage.googleapis.com/deepmind-media/Model-Cards/Gemini-2-0-Flash-Model-Card.pdf", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Gemini 2.0 Flash Model Card (April 15 2025): HiddenMath 63.5%", "candidates": [] }, { "model_id": "gemini-1.5-flash", "benchmark_id": "mrcr_v1", "score": 71.9, "reference_url": "https://storage.googleapis.com/deepmind-media/Model-Cards/Gemini-2-0-Flash-Model-Card.pdf", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Gemini 2.0 Flash Model Card (April 15 2025): MRCR v1 1M 71.9%", "candidates": [] }, { "model_id": "gemini-1.5-pro", "benchmark_id": "mrcr_v1", "score": 82.6, "reference_url": "https://storage.googleapis.com/deepmind-media/Model-Cards/Gemini-2-0-Flash-Model-Card.pdf", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Gemini 2.0 Flash Model Card (April 15 2025): MRCR v1 1M 82.6%", "candidates": [] }, { "model_id": "gemini-2.0-flash-lite", "benchmark_id": "mrcr_v1", "score": 58.0, "reference_url": "https://storage.googleapis.com/deepmind-media/Model-Cards/Gemini-2-0-Flash-Model-Card.pdf", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Gemini 2.0 Flash Model Card (April 15 2025): MRCR v1 1M 58.0%", "candidates": [] }, { "model_id": "gemini-2.0-flash", "benchmark_id": "mrcr_v1", "score": 70.5, "reference_url": "https://storage.googleapis.com/deepmind-media/Model-Cards/Gemini-2-0-Flash-Model-Card.pdf", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Gemini 2.0 Flash Model Card (April 15 2025): MRCR v1 1M 70.5%", "candidates": [] }, { "model_id": "gemini-1.5-flash", "benchmark_id": "mmmu", "score": 62.3, "reference_url": "https://storage.googleapis.com/deepmind-media/Model-Cards/Gemini-2-0-Flash-Model-Card.pdf", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Gemini 2.0 Flash Model Card (April 15 2025): MMMU 62.3%", "candidates": [] }, { "model_id": "gemini-1.5-pro", "benchmark_id": "mmmu", "score": 65.9, "reference_url": "https://storage.googleapis.com/deepmind-media/Model-Cards/Gemini-2-0-Flash-Model-Card.pdf", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Gemini 2.0 Flash Model Card (April 15 2025): MMMU 65.9%", "candidates": [] }, { "model_id": "gemini-2.0-flash-lite", "benchmark_id": "mmmu", "score": 68.0, "reference_url": "https://storage.googleapis.com/deepmind-media/Model-Cards/Gemini-2-0-Flash-Model-Card.pdf", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Gemini 2.0 Flash Model Card (April 15 2025): MMMU 68.0%", "candidates": [] }, { "model_id": "gemini-1.5-flash", "benchmark_id": "covost2", "score": 37.4, "reference_url": "https://storage.googleapis.com/deepmind-media/Model-Cards/Gemini-2-0-Flash-Model-Card.pdf", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Gemini 2.0 Flash Model Card (April 15 2025): CoVoST2 21-lang BLEU 37.4 BLEU", "candidates": [] }, { "model_id": "gemini-1.5-pro", "benchmark_id": "covost2", "score": 40.1, "reference_url": "https://storage.googleapis.com/deepmind-media/Model-Cards/Gemini-2-0-Flash-Model-Card.pdf", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Gemini 2.0 Flash Model Card (April 15 2025): CoVoST2 21-lang BLEU 40.1 BLEU", "candidates": [] }, { "model_id": "gemini-2.0-flash-lite", "benchmark_id": "covost2", "score": 38.4, "reference_url": "https://storage.googleapis.com/deepmind-media/Model-Cards/Gemini-2-0-Flash-Model-Card.pdf", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Gemini 2.0 Flash Model Card (April 15 2025): CoVoST2 21-lang BLEU 38.4 BLEU", "candidates": [] }, { "model_id": "gemini-2.0-flash", "benchmark_id": "covost2", "score": 39.0, "reference_url": "https://storage.googleapis.com/deepmind-media/Model-Cards/Gemini-2-0-Flash-Model-Card.pdf", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Gemini 2.0 Flash Model Card (April 15 2025): CoVoST2 21-lang BLEU 39.0 BLEU", "candidates": [] }, { "model_id": "gemini-1.5-flash", "benchmark_id": "egoschema", "score": 66.8, "reference_url": "https://storage.googleapis.com/deepmind-media/Model-Cards/Gemini-2-0-Flash-Model-Card.pdf", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Gemini 2.0 Flash Model Card (April 15 2025): EgoSchema test 66.8%", "candidates": [] }, { "model_id": "gemini-1.5-pro", "benchmark_id": "egoschema", "score": 71.2, "reference_url": "https://storage.googleapis.com/deepmind-media/Model-Cards/Gemini-2-0-Flash-Model-Card.pdf", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Gemini 2.0 Flash Model Card (April 15 2025): EgoSchema test 71.2%", "candidates": [] }, { "model_id": "gemini-2.0-flash-lite", "benchmark_id": "egoschema", "score": 67.2, "reference_url": "https://storage.googleapis.com/deepmind-media/Model-Cards/Gemini-2-0-Flash-Model-Card.pdf", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Gemini 2.0 Flash Model Card (April 15 2025): EgoSchema test 67.2%", "candidates": [] }, { "model_id": "gemini-2.0-flash", "benchmark_id": "egoschema", "score": 71.1, "reference_url": "https://storage.googleapis.com/deepmind-media/Model-Cards/Gemini-2-0-Flash-Model-Card.pdf", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Gemini 2.0 Flash Model Card (April 15 2025): EgoSchema test 71.1%", "candidates": [] }, { "model_id": "grok-3-beta", "benchmark_id": "mmmu", "score": 73.2, "reference_url": "https://x.ai/news/grok-3", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "xAI Grok 3 blog (cross-model non-reasoning table): grok-3-beta=73.2.", "candidates": [ { "score": 76.0, "reference_url": "https://www.anthropic.com/news/claude-3-7-sonnet", "source_type": "official_blog", "reported_setting": { "mode": "thinking", "effort": "extended", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "notes": "Anthropic 3.7 Sonnet blog (third-party eval): Grok 3 Beta extended thinking MMMU=76.0%/78.0%. BP primary=73.2 from x.ai (different mode?)." } ] }, { "model_id": "grok-3-beta", "benchmark_id": "loft_128k", "score": 83.3, "reference_url": "https://x.ai/news/grok-3", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "xAI Grok 3 blog (cross-model non-reasoning table): grok-3-beta=83.3.", "candidates": [] }, { "model_id": "grok-3-beta", "benchmark_id": "egoschema", "score": 74.5, "reference_url": "https://x.ai/news/grok-3", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "xAI Grok 3 blog (cross-model non-reasoning table): grok-3-beta=74.5.", "candidates": [] }, { "model_id": "claude-sonnet-4.6", "benchmark_id": "cybergym", "score": 65.2, "reference_url": "https://www.anthropic.com/claude-sonnet-5-system-card", "reported_setting": { "effort": "source does not state", "tools": "cyber execution environment", "sampling": "pass@1", "harness": "cybergym" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Anthropic Claude Sonnet 5 System Card Figure 3.2.3.A, page 33: Claude Sonnet 4.6; CyberGym vulnerability discovery [targeted]=65.2%. Source setting: effort=source does not state; tools=cyber execution environment; sampling=pass@1; harness=cybergym.", "candidates": [ { "score": 65.2, "reference_url": "https://anthropic.com/claude-sonnet-4-6-system-card", "source_type": "tech_report", "reported_setting": { "mode": "non-thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default", "notes": "no thinking, default effort/temperature/top_p, with think tool for interleaved thinking" }, "notes": "Claude Sonnet 4.6 System Card section 2.15: CyberGym 65.2% pass@1 over 1507 targeted vuln reproduction tasks. matches_canonical=false (no thinking + default effort, vs canonical thinking+max)." } ] }, { "model_id": "claude-sonnet-4.6", "benchmark_id": "vending_bench_2", "score": 7204.14, "reference_url": "https://anthropic.com/claude-sonnet-4-6-system-card", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "none", "sampling": "final balance after 1-year simulation", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default", "notes": "max effort" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Claude Sonnet 4.6 System Card section 2.13: Vending-Bench 2 final balance $7,204.14 with Max effort (vs Opus 4.6 SOTA $8017.59). High-effort run scored $6,625.10 (kept as candidate setting note). Mean cost per run $265.03.", "candidates": [ { "score": 6625.1, "reference_url": "https://anthropic.com/claude-sonnet-4-6-system-card", "source_type": "tech_report", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "final balance after 1-year simulation", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "notes": "Same source, high effort variant." } ] }, { "model_id": "grok-4.1", "benchmark_id": "eq_bench3", "score": 1585, "reference_url": "https://x.ai/news/grok-4-1", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "LLM judge", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "xAI Grok 4.1 blog: grok-4.1=1585 (Elo Normalized).", "candidates": [] }, { "model_id": "grok-4.1", "benchmark_id": "creative_writing_v3", "score": 1708.6, "reference_url": "https://x.ai/news/grok-4-1", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "LLM judge", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "xAI Grok 4.1 blog: grok-4.1=1708.6 (Elo Normalized).", "candidates": [ { "score": 172190.0, "reference_url": "https://llm-stats.com/benchmarks/creative-writing-v3", "source_type": "third_party_aggregator", "reported_setting": { "effort": "thinking" }, "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=Grok-4.1 Thinking, slug=grok-4.1-thinking-2025-11-17, provider=xAI" } ] }, { "model_id": "qwen3-235b", "benchmark_id": "mmlu_redux", "score": 92.7, "reference_url": "https://arxiv.org/abs/2505.09388", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Qwen3 tech report (arxiv:2505.09388) Table 11 Thinking column: MMLU-Redux 92.7", "candidates": [ { "score": 93.8, "reference_url": "https://llm-stats.com/benchmarks/mmlu-redux", "source_type": "third_party_aggregator", "reported_setting": { "mode": "thinking", "snapshot": "2507" }, "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=Qwen3-235B-A22B-Thinking-2507, slug=qwen3-235b-a22b-thinking-2507, provider=Alibaba Cloud / Qwen Team" }, { "score": 93.1, "reference_url": "https://llm-stats.com/benchmarks/mmlu-redux", "source_type": "third_party_aggregator", "reported_setting": { "mode": "instruct", "snapshot": "2507" }, "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=Qwen3-235B-A22B-Instruct-2507, slug=qwen3-235b-a22b-instruct-2507, provider=Alibaba Cloud / Qwen Team" } ] }, { "model_id": "qwen3-235b", "benchmark_id": "bfcl_v3", "score": 70.8, "reference_url": "https://arxiv.org/abs/2505.09388", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Qwen3 tech report (arxiv:2505.09388) Table 11 Thinking column: BFCL v3 70.8", "candidates": [ { "score": 71.9, "reference_url": "https://llm-stats.com/benchmarks/bfcl-v3", "source_type": "third_party_aggregator", "reported_setting": { "mode": "thinking", "snapshot": "2507" }, "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=Qwen3-235B-A22B-Thinking-2507, slug=qwen3-235b-a22b-thinking-2507, provider=Alibaba Cloud / Qwen Team" }, { "score": 70.9, "reference_url": "https://llm-stats.com/benchmarks/bfcl-v3", "source_type": "third_party_aggregator", "reported_setting": { "mode": "instruct", "snapshot": "2507" }, "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=Qwen3-235B-A22B-Instruct-2507, slug=qwen3-235b-a22b-instruct-2507, provider=Alibaba Cloud / Qwen Team" } ] }, { "model_id": "qwen3-235b", "benchmark_id": "multi_if", "score": 71.9, "reference_url": "https://arxiv.org/abs/2505.09388", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Qwen3 tech report (arxiv:2505.09388) Table 11 Thinking column: Multi-IF 71.9", "candidates": [ { "score": 80.6, "reference_url": "https://llm-stats.com/benchmarks/multi-if", "source_type": "third_party_aggregator", "reported_setting": { "mode": "thinking", "snapshot": "2507" }, "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=Qwen3-235B-A22B-Thinking-2507, slug=qwen3-235b-a22b-thinking-2507, provider=Alibaba Cloud / Qwen Team" }, { "score": 77.5, "reference_url": "https://llm-stats.com/benchmarks/multi-if", "source_type": "third_party_aggregator", "reported_setting": { "mode": "instruct", "snapshot": "2507" }, "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=Qwen3-235B-A22B-Instruct-2507, slug=qwen3-235b-a22b-instruct-2507, provider=Alibaba Cloud / Qwen Team" } ] }, { "model_id": "qwen3-235b", "benchmark_id": "mt_aime_2024", "score": 80.8, "reference_url": "https://arxiv.org/abs/2505.09388", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Qwen3 tech report (arxiv:2505.09388) Table 11 Thinking column: MT-AIME2024 80.8", "candidates": [] }, { "model_id": "qwen3-235b", "benchmark_id": "mmmlu", "score": 84.3, "reference_url": "https://arxiv.org/abs/2505.09388", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Qwen3 tech report (arxiv:2505.09388) Table 11 Thinking column: MMMLU 14 languages 84.3", "candidates": [] }, { "model_id": "qwen3-235b", "benchmark_id": "livebench", "score": 77.1, "reference_url": "https://arxiv.org/abs/2505.09388", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Qwen3 tech report (arxiv:2505.09388) Table 11 Thinking column: LiveBench 2024-11-25 77.1", "candidates": [] }, { "model_id": "deepseek-v4-pro", "benchmark_id": "chinese_simpleqa", "score": 84.4, "reference_url": "https://huggingface.co/deepseek-ai/DeepSeek-V4-Pro", "reported_setting": { "mode": "thinking", "effort": "Max", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "DeepSeek V4 model card: deepseek-v4-pro (Max mode) = 84.4.", "candidates": [] }, { "model_id": "deepseek-v4-pro", "benchmark_id": "mrcr_v1", "score": 83.5, "reference_url": "https://huggingface.co/deepseek-ai/DeepSeek-V4-Pro", "reported_setting": { "mode": "thinking", "effort": "Max", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "DeepSeek V4 model card: deepseek-v4-pro (Max mode) = 83.5.", "candidates": [] }, { "model_id": "deepseek-v4-pro", "benchmark_id": "corpusqa_1m", "score": 62.0, "reference_url": "https://huggingface.co/deepseek-ai/DeepSeek-V4-Pro", "reported_setting": { "mode": "thinking", "effort": "Max", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "DeepSeek V4 model card: deepseek-v4-pro (Max mode) = 62.0.", "candidates": [ { "score": 62.0, "reference_url": "https://llm-stats.com/benchmarks/corpusqa-1m", "source_type": "third_party_aggregator", "reported_setting": { "effort": "max" }, "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=DeepSeek-V4-Pro-Max, slug=deepseek-v4-pro-max, provider=DeepSeek" } ] }, { "model_id": "deepseek-v4-flash", "benchmark_id": "chinese_simpleqa", "score": 78.9, "reference_url": "https://huggingface.co/deepseek-ai/DeepSeek-V4-Pro", "reported_setting": { "mode": "thinking", "effort": "Max", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "DeepSeek V4 model card: deepseek-v4-flash (Max mode) = 78.9.", "candidates": [] }, { "model_id": "deepseek-v4-flash", "benchmark_id": "mrcr_v1", "score": 78.7, "reference_url": "https://huggingface.co/deepseek-ai/DeepSeek-V4-Pro", "reported_setting": { "mode": "thinking", "effort": "Max", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "DeepSeek V4 model card: deepseek-v4-flash (Max mode) = 78.7.", "candidates": [] }, { "model_id": "deepseek-v4-flash", "benchmark_id": "corpusqa_1m", "score": 60.5, "reference_url": "https://huggingface.co/deepseek-ai/DeepSeek-V4-Pro", "reported_setting": { "mode": "thinking", "effort": "Max", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "DeepSeek V4 model card: deepseek-v4-flash (Max mode) = 60.5.", "candidates": [ { "score": 60.5, "reference_url": "https://llm-stats.com/benchmarks/corpusqa-1m", "source_type": "third_party_aggregator", "reported_setting": { "effort": "max" }, "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=DeepSeek-V4-Flash-Max, slug=deepseek-v4-flash-max, provider=DeepSeek" } ] }, { "model_id": "gpt-5.4", "benchmark_id": "mmlu_pro", "score": 87.5, "reference_url": "https://huggingface.co/deepseek-ai/DeepSeek-V4-Pro", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "DeepSeek V4-Pro model card (third-party): gpt-5.4 (per their own canonical config) = 87.5.", "candidates": [] }, { "model_id": "kimi-k2.6", "benchmark_id": "mmlu_pro", "score": 87.1, "reference_url": "https://huggingface.co/deepseek-ai/DeepSeek-V4-Pro", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "DeepSeek V4-Pro model card (third-party): kimi-k2.6 (per their own canonical config) = 87.1.", "candidates": [ { "score": 88.1, "reference_url": "https://research.nvidia.com/labs/nemotron/files/NVIDIA-Nemotron-3-Ultra-Technical-Report.pdf", "source_type": "tech_report", "reported_setting": { "version_metric": "MMLU-Pro / %", "tools": "none", "harness": "NeMo Evaluator", "sampling": "pass@1" }, "notes": "NVIDIA Nemotron 3 Ultra technical report, Final Table 10: Kimi K2.6: mmlu_pro=88.1." } ] }, { "model_id": "glm-5.1", "benchmark_id": "mmlu_pro", "score": 86.0, "reference_url": "https://huggingface.co/deepseek-ai/DeepSeek-V4-Pro", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "DeepSeek V4-Pro model card (third-party): glm-5.1 (per their own canonical config) = 86.0.", "candidates": [ { "score": 85.9, "reference_url": "https://research.nvidia.com/labs/nemotron/files/NVIDIA-Nemotron-3-Ultra-Technical-Report.pdf", "source_type": "tech_report", "reported_setting": { "version_metric": "MMLU-Pro / %", "tools": "none", "harness": "NeMo Evaluator", "sampling": "pass@1" }, "notes": "NVIDIA Nemotron 3 Ultra technical report, Final Table 10: GLM-5.1: mmlu_pro=85.9." } ] }, { "model_id": "claude-opus-4.6", "benchmark_id": "chinese_simpleqa", "score": 76.4, "reference_url": "https://huggingface.co/deepseek-ai/DeepSeek-V4-Pro", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "DeepSeek V4-Pro model card (third-party): claude-opus-4.6 (per their own canonical config) = 76.4.", "candidates": [] }, { "model_id": "gpt-5.4", "benchmark_id": "chinese_simpleqa", "score": 76.8, "reference_url": "https://huggingface.co/deepseek-ai/DeepSeek-V4-Pro", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "DeepSeek V4-Pro model card (third-party): gpt-5.4 (per their own canonical config) = 76.8.", "candidates": [] }, { "model_id": "gemini-3.1-pro", "benchmark_id": "chinese_simpleqa", "score": 85.9, "reference_url": "https://huggingface.co/deepseek-ai/DeepSeek-V4-Pro", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "DeepSeek V4-Pro model card (third-party): gemini-3.1-pro (per their own canonical config) = 85.9.", "candidates": [] }, { "model_id": "kimi-k2.6", "benchmark_id": "chinese_simpleqa", "score": 75.9, "reference_url": "https://huggingface.co/deepseek-ai/DeepSeek-V4-Pro", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "DeepSeek V4-Pro model card (third-party): kimi-k2.6 (per their own canonical config) = 75.9.", "candidates": [] }, { "model_id": "glm-5.1", "benchmark_id": "chinese_simpleqa", "score": 75.0, "reference_url": "https://huggingface.co/deepseek-ai/DeepSeek-V4-Pro", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "DeepSeek V4-Pro model card (third-party): glm-5.1 (per their own canonical config) = 75.0.", "candidates": [] }, { "model_id": "gpt-5.4", "benchmark_id": "codeforces_rating", "score": 3168, "reference_url": "https://huggingface.co/deepseek-ai/DeepSeek-V4-Pro", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "DeepSeek V4-Pro model card (third-party): gpt-5.4 (per their own canonical config) = 3168.", "candidates": [] }, { "model_id": "gemini-3.1-pro", "benchmark_id": "imo_answerbench", "score": 81.0, "reference_url": "https://huggingface.co/deepseek-ai/DeepSeek-V4-Pro", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "DeepSeek V4-Pro model card (third-party): gemini-3.1-pro (per their own canonical config) = 81.0.", "candidates": [ { "score": 91.0, "reference_url": "https://huggingface.co/moonshotai/Kimi-K2.6", "source_type": "model_card", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "1.0" }, "notes": "(merged from imoanswerbench) Kimi K2.6 model card: gemini-3.1-pro=91.0." }, { "score": 90.0, "reference_url": "https://longcat.chat/blog/longcat-2.0/", "source_type": "official_blog", "reported_setting": { "mode": "thinking", "effort": "source does not state", "tools": "none", "sampling": "source does not state", "judge": "IMO-AnswerBench answer autograder", "harness": "Meituan unified in-house harness", "prompt_style": "default", "temperature": "1.0; top_k=-1; top_p=0.95", "context": "source does not state", "notes": "Measured in-house by Meituan unless otherwise noted." }, "notes": "LongCat-2.0 official tech blog: IMO-AnswerBench = 90.0. Measured in-house by Meituan under the reported unified harness." } ] }, { "model_id": "claude-opus-4.6", "benchmark_id": "matharena_apex_2025", "score": 34.5, "reference_url": "https://huggingface.co/deepseek-ai/DeepSeek-V4-Pro", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "DeepSeek V4-Pro model card (third-party): claude-opus-4.6 (per their own canonical config) = 34.5.", "candidates": [] }, { "model_id": "gpt-5.4", "benchmark_id": "matharena_apex_2025", "score": 54.1, "reference_url": "https://huggingface.co/deepseek-ai/DeepSeek-V4-Pro", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "DeepSeek V4-Pro model card (third-party): gpt-5.4 (per their own canonical config) = 54.1.", "candidates": [] }, { "model_id": "kimi-k2.6", "benchmark_id": "matharena_apex_2025", "score": 24.0, "reference_url": "https://huggingface.co/deepseek-ai/DeepSeek-V4-Pro", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "DeepSeek V4-Pro model card (third-party): kimi-k2.6 (per their own canonical config) = 24.0.", "candidates": [] }, { "model_id": "glm-5.1", "benchmark_id": "matharena_apex_2025", "score": 11.5, "reference_url": "https://huggingface.co/deepseek-ai/DeepSeek-V4-Pro", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "DeepSeek V4-Pro model card (third-party): glm-5.1 (per their own canonical config) = 11.5.", "candidates": [] }, { "model_id": "claude-opus-4.6", "benchmark_id": "mrcr_v1", "score": 92.9, "reference_url": "https://huggingface.co/deepseek-ai/DeepSeek-V4-Pro", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "DeepSeek V4-Pro model card (third-party): claude-opus-4.6 (per their own canonical config) = 92.9.", "candidates": [] }, { "model_id": "gemini-3.1-pro", "benchmark_id": "mrcr_v1", "score": 76.3, "reference_url": "https://huggingface.co/deepseek-ai/DeepSeek-V4-Pro", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "DeepSeek V4-Pro model card (third-party): gemini-3.1-pro (per their own canonical config) = 76.3.", "candidates": [] }, { "model_id": "claude-opus-4.6", "benchmark_id": "corpusqa_1m", "score": 71.7, "reference_url": "https://huggingface.co/deepseek-ai/DeepSeek-V4-Pro", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "DeepSeek V4-Pro model card (third-party): claude-opus-4.6 (per their own canonical config) = 71.7.", "candidates": [] }, { "model_id": "gemini-3.1-pro", "benchmark_id": "corpusqa_1m", "score": 53.8, "reference_url": "https://huggingface.co/deepseek-ai/DeepSeek-V4-Pro", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "DeepSeek V4-Pro model card (third-party): gemini-3.1-pro (per their own canonical config) = 53.8.", "candidates": [] }, { "model_id": "glm-5.1", "benchmark_id": "swe_bench_multilingual", "score": 73.3, "reference_url": "https://huggingface.co/deepseek-ai/DeepSeek-V4-Pro", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "agentic (bash + edit)", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "DeepSeek V4-Pro model card (third-party): glm-5.1 (per their own canonical config) = 73.3.", "candidates": [ { "score": 74.8, "reference_url": "https://research.nvidia.com/labs/nemotron/files/NVIDIA-Nemotron-3-Ultra-Technical-Report.pdf", "source_type": "tech_report", "reported_setting": { "version_metric": "Multilingual / % resolved", "tools": "agentic repository editing", "harness": "Harbor / OpenHands / AWS ECS", "sampling": "3-run average" }, "notes": "NVIDIA Nemotron 3 Ultra technical report, Final Table 10: GLM-5.1: swe_bench_multilingual=74.8." } ] }, { "model_id": "gpt-5.4", "benchmark_id": "gdpval_aa_elo", "score": 1674, "reference_url": "https://huggingface.co/deepseek-ai/DeepSeek-V4-Pro", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "DeepSeek V4-Pro model card (third-party): gpt-5.4 (per their own canonical config) = 1674.", "candidates": [] }, { "model_id": "gemini-3.1-pro", "benchmark_id": "gdpval_aa_elo", "score": 1314, "reference_url": "https://huggingface.co/deepseek-ai/DeepSeek-V4-Pro", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "DeepSeek V4-Pro model card (third-party): gemini-3.1-pro (per their own canonical config) = 1314.", "candidates": [ { "score": 1317, "reference_url": "https://deepmind.google/models/gemini/pro/", "source_type": "official_blog", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "notes": "DeepMind /models/gemini/pro/ (model card) Performance table: GDPval-AA Elo 1317 (col 1, Gemini 3.1 Pro Thinking High). Differs from primary 1314 from DeepSeek V4-Pro 3rd-party self-test." }, { "score": 962.3, "reference_url": "https://openai.com/index/gpt-5-6/", "source_type": "official_blog", "reported_setting": { "effort": "n/a", "eval_variant": null, "code_mode": "false" }, "notes": "OpenAI GPT-5.6 release table (table 0): Gemini 3.1 Pro Preview; GDPval-AA v2=962.3. OpenAI row explicitly identifies GDPval-AA v2. Exact effort/variant resolved from the rendered chart point." }, { "score": 1317, "reference_url": "https://deepmind.google/models/gemini/pro/", "source_type": "official_blog", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official Google Gemini evaluation", "prompt_style": "provider official evaluation prompt", "temperature": "source/provider setting", "context": "source/provider setting", "input_modalities": "text", "trials": "source-reported" }, "notes": "Exact provider-official score and reported setting." } ] }, { "model_id": "kimi-k2.6", "benchmark_id": "gdpval_aa_elo", "score": 1482, "reference_url": "https://huggingface.co/deepseek-ai/DeepSeek-V4-Pro", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "DeepSeek V4-Pro model card (third-party): kimi-k2.6 (per their own canonical config) = 1482.", "candidates": [ { "score": 1481.0, "reference_url": "https://static.stepfun.com/blog/step-3.7-flash/", "source_type": "official_blog", "reported_setting": { "quantization_config": "common release evaluation; artifact precision not established", "configuration": "base Step evaluation configuration", "mode": "source does not state", "effort": "source does not state", "tools_search": "official Artificial Analysis Stirrup harness", "shots": "source does not state", "trials_samples": "source does not state", "aggregation_pass_k": "source-reported score", "context": "source does not state", "output_cap": "source does not state", "temperature": "source does not state", "top_p": "source does not state", "judge": "pairwise Elo evaluation anchored to human experts", "harness_agent": "official Artificial Analysis Stirrup leaderboard", "dataset_version_split": "GDPval-AA historical 220-task Stirrup leaderboard quoted 2026-05-29", "multimodal_input": false }, "notes": "Official StepFun launch comparison table." } ] }, { "model_id": "glm-5.1", "benchmark_id": "gdpval_aa_elo", "score": 1535, "reference_url": "https://huggingface.co/deepseek-ai/DeepSeek-V4-Pro", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "DeepSeek V4-Pro model card (third-party): glm-5.1 (per their own canonical config) = 1535.", "candidates": [ { "score": 1535.0, "reference_url": "https://static.stepfun.com/blog/step-3.7-flash/", "source_type": "official_blog", "reported_setting": { "quantization_config": "common release evaluation; artifact precision not established", "configuration": "base Step evaluation configuration", "mode": "source does not state", "effort": "source does not state", "tools_search": "official Artificial Analysis Stirrup harness", "shots": "source does not state", "trials_samples": "source does not state", "aggregation_pass_k": "source-reported score", "context": "source does not state", "output_cap": "source does not state", "temperature": "source does not state", "top_p": "source does not state", "judge": "pairwise Elo evaluation anchored to human experts", "harness_agent": "official Artificial Analysis Stirrup leaderboard", "dataset_version_split": "GDPval-AA historical 220-task Stirrup leaderboard quoted 2026-05-29", "multimodal_input": false }, "notes": "Official StepFun launch comparison table." } ] }, { "model_id": "qwen3-32b", "benchmark_id": "mmlu_redux", "score": 90.9, "reference_url": "https://arxiv.org/abs/2505.09388", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Qwen3 tech report Thinking: MMLU-Redux 90.9", "candidates": [] }, { "model_id": "qwen3-32b", "benchmark_id": "bfcl_v3", "score": 70.3, "reference_url": "https://arxiv.org/abs/2505.09388", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Qwen3 tech report Thinking: BFCL v3 70.3", "candidates": [] }, { "model_id": "qwen3-32b", "benchmark_id": "multi_if", "score": 73.0, "reference_url": "https://arxiv.org/abs/2505.09388", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Qwen3 tech report Thinking: Multi-IF 73.0", "candidates": [] }, { "model_id": "qwen3-32b", "benchmark_id": "mt_aime_2024", "score": 75.0, "reference_url": "https://arxiv.org/abs/2505.09388", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Qwen3 tech report Thinking: MT-AIME2024 75.0", "candidates": [] }, { "model_id": "qwen3-32b", "benchmark_id": "mmmlu", "score": 80.6, "reference_url": "https://arxiv.org/abs/2505.09388", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Qwen3 tech report Thinking: MMMLU 14 langs 80.6", "candidates": [] }, { "model_id": "qwen3-32b", "benchmark_id": "livebench", "score": 74.9, "reference_url": "https://arxiv.org/abs/2505.09388", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Qwen3 tech report Thinking: LiveBench 2024-11-25 74.9", "candidates": [] }, { "model_id": "qwen3-30b-a3b", "benchmark_id": "mmlu_redux", "score": 89.5, "reference_url": "https://arxiv.org/abs/2505.09388", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Qwen3 tech report Thinking: MMLU-Redux 89.5", "candidates": [] }, { "model_id": "qwen3-30b-a3b", "benchmark_id": "bfcl_v3", "score": 69.1, "reference_url": "https://arxiv.org/abs/2505.09388", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Qwen3 tech report Thinking: BFCL v3 69.1", "candidates": [] }, { "model_id": "qwen3-30b-a3b", "benchmark_id": "multi_if", "score": 72.2, "reference_url": "https://arxiv.org/abs/2505.09388", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Qwen3 tech report Thinking: Multi-IF 72.2", "candidates": [] }, { "model_id": "qwen3-30b-a3b", "benchmark_id": "mt_aime_2024", "score": 73.9, "reference_url": "https://arxiv.org/abs/2505.09388", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Qwen3 tech report Thinking: MT-AIME2024 73.9", "candidates": [] }, { "model_id": "qwen3-30b-a3b", "benchmark_id": "mmmlu", "score": 78.4, "reference_url": "https://arxiv.org/abs/2505.09388", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Qwen3 tech report Thinking: MMMLU 14 langs 78.4", "candidates": [] }, { "model_id": "qwen3-30b-a3b", "benchmark_id": "livebench", "score": 74.3, "reference_url": "https://arxiv.org/abs/2505.09388", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Qwen3 tech report Thinking: LiveBench 2024-11-25 74.3", "candidates": [] }, { "model_id": "qwen3-14b", "benchmark_id": "mmlu_redux", "score": 88.6, "reference_url": "https://arxiv.org/abs/2505.09388", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Qwen3 tech report Thinking: MMLU-Redux 88.6", "candidates": [] }, { "model_id": "qwen3-14b", "benchmark_id": "bfcl_v3", "score": 70.4, "reference_url": "https://arxiv.org/abs/2505.09388", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Qwen3 tech report Thinking: BFCL v3 70.4", "candidates": [] }, { "model_id": "qwen3-14b", "benchmark_id": "multi_if", "score": 74.8, "reference_url": "https://arxiv.org/abs/2505.09388", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Qwen3 tech report Thinking: Multi-IF 74.8", "candidates": [] }, { "model_id": "qwen3-14b", "benchmark_id": "mt_aime_2024", "score": 73.3, "reference_url": "https://arxiv.org/abs/2505.09388", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Qwen3 tech report Thinking: MT-AIME2024 73.3", "candidates": [] }, { "model_id": "qwen3-14b", "benchmark_id": "mmmlu", "score": 77.9, "reference_url": "https://arxiv.org/abs/2505.09388", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Qwen3 tech report Thinking: MMMLU 14 langs 77.9", "candidates": [] }, { "model_id": "qwen3-14b", "benchmark_id": "livebench", "score": 71.3, "reference_url": "https://arxiv.org/abs/2505.09388", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Qwen3 tech report Thinking: LiveBench 2024-11-25 71.3", "candidates": [] }, { "model_id": "qwen3-14b", "benchmark_id": "ifeval", "score": 85.4, "reference_url": "https://arxiv.org/abs/2505.09388", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Qwen3 tech report Thinking: IFEval strict 85.4", "candidates": [] }, { "model_id": "qwen3-8b", "benchmark_id": "mmlu_redux", "score": 87.5, "reference_url": "https://arxiv.org/abs/2505.09388", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Qwen3 tech report Thinking: MMLU-Redux 87.5", "candidates": [] }, { "model_id": "qwen3-8b", "benchmark_id": "bfcl_v3", "score": 68.1, "reference_url": "https://arxiv.org/abs/2505.09388", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Qwen3 tech report Thinking: BFCL v3 68.1", "candidates": [] }, { "model_id": "qwen3-8b", "benchmark_id": "multi_if", "score": 71.2, "reference_url": "https://arxiv.org/abs/2505.09388", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Qwen3 tech report Thinking: Multi-IF 71.2", "candidates": [] }, { "model_id": "qwen3-8b", "benchmark_id": "mt_aime_2024", "score": 65.4, "reference_url": "https://arxiv.org/abs/2505.09388", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Qwen3 tech report Thinking: MT-AIME2024 65.4", "candidates": [] }, { "model_id": "qwen3-8b", "benchmark_id": "mmmlu", "score": 74.4, "reference_url": "https://arxiv.org/abs/2505.09388", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Qwen3 tech report Thinking: MMMLU 14 langs 74.4", "candidates": [] }, { "model_id": "qwen3-8b", "benchmark_id": "livebench", "score": 67.1, "reference_url": "https://arxiv.org/abs/2505.09388", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Qwen3 tech report Thinking: LiveBench 2024-11-25 67.1", "candidates": [] }, { "model_id": "qwen3-4b", "benchmark_id": "mmlu_redux", "score": 83.7, "reference_url": "https://arxiv.org/abs/2505.09388", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Qwen3 tech report Thinking: MMLU-Redux 83.7", "candidates": [] }, { "model_id": "qwen3-4b", "benchmark_id": "bfcl_v3", "score": 65.9, "reference_url": "https://arxiv.org/abs/2505.09388", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Qwen3 tech report Thinking: BFCL v3 65.9", "candidates": [] }, { "model_id": "qwen3-4b", "benchmark_id": "multi_if", "score": 66.3, "reference_url": "https://arxiv.org/abs/2505.09388", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Qwen3 tech report Thinking: Multi-IF 66.3", "candidates": [] }, { "model_id": "qwen3-4b", "benchmark_id": "mt_aime_2024", "score": 60.7, "reference_url": "https://arxiv.org/abs/2505.09388", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Qwen3 tech report Thinking: MT-AIME2024 60.7", "candidates": [] }, { "model_id": "qwen3-4b", "benchmark_id": "mmmlu", "score": 69.8, "reference_url": "https://arxiv.org/abs/2505.09388", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Qwen3 tech report Thinking: MMMLU 14 langs 69.8", "candidates": [] }, { "model_id": "qwen3-4b", "benchmark_id": "livebench", "score": 63.6, "reference_url": "https://arxiv.org/abs/2505.09388", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Qwen3 tech report Thinking: LiveBench 2024-11-25 63.6", "candidates": [] }, { "model_id": "qwen3-1.7b", "benchmark_id": "mmlu_redux", "score": 73.9, "reference_url": "https://arxiv.org/abs/2505.09388", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Qwen3 tech report Thinking: MMLU-Redux 73.9", "candidates": [] }, { "model_id": "qwen3-1.7b", "benchmark_id": "bfcl_v3", "score": 56.6, "reference_url": "https://arxiv.org/abs/2505.09388", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Qwen3 tech report Thinking: BFCL v3 56.6", "candidates": [ { "score": 55.41, "reference_url": "https://www.liquid.ai/blog/lfm2-5-1-2b-thinking-on-device-reasoning-under-1gb", "source_type": "official_blog", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1 (mean of 5 runs)", "judge": "rule-based", "harness": "official (LFM eval, third-party)", "prompt_style": "default", "temperature": "default" }, "notes": "Liquid AI LFM2.5-1.2B-Thinking blog: qwen3-1.7b=55.41. (third-party self-test by Liquid AI)" } ] }, { "model_id": "qwen3-1.7b", "benchmark_id": "multi_if", "score": 51.2, "reference_url": "https://arxiv.org/abs/2505.09388", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Qwen3 tech report Thinking: Multi-IF 51.2", "candidates": [ { "score": 60.33, "reference_url": "https://www.liquid.ai/blog/lfm2-5-1-2b-thinking-on-device-reasoning-under-1gb", "source_type": "official_blog", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1 (mean of 5 runs)", "judge": "rule-based", "harness": "official (LFM eval, third-party)", "prompt_style": "default", "temperature": "default" }, "notes": "Liquid AI LFM2.5-1.2B-Thinking blog: qwen3-1.7b=60.33. (third-party self-test by Liquid AI)" } ] }, { "model_id": "qwen3-1.7b", "benchmark_id": "mt_aime_2024", "score": 36.1, "reference_url": "https://arxiv.org/abs/2505.09388", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Qwen3 tech report Thinking: MT-AIME2024 36.1", "candidates": [] }, { "model_id": "qwen3-1.7b", "benchmark_id": "mmmlu", "score": 59.1, "reference_url": "https://arxiv.org/abs/2505.09388", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Qwen3 tech report Thinking: MMMLU 14 langs 59.1", "candidates": [] }, { "model_id": "qwen3-0.6b", "benchmark_id": "mmlu_redux", "score": 55.6, "reference_url": "https://arxiv.org/abs/2505.09388", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Qwen3 tech report Thinking: MMLU-Redux 55.6", "candidates": [] }, { "model_id": "qwen3-0.6b", "benchmark_id": "bfcl_v3", "score": 46.4, "reference_url": "https://arxiv.org/abs/2505.09388", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Qwen3 tech report Thinking: BFCL v3 46.4", "candidates": [] }, { "model_id": "qwen3-0.6b", "benchmark_id": "multi_if", "score": 36.1, "reference_url": "https://arxiv.org/abs/2505.09388", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Qwen3 tech report Thinking: Multi-IF 36.1", "candidates": [] }, { "model_id": "qwen3-0.6b", "benchmark_id": "mt_aime_2024", "score": 7.8, "reference_url": "https://arxiv.org/abs/2505.09388", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Qwen3 tech report Thinking: MT-AIME2024 7.8", "candidates": [] }, { "model_id": "qwen3-0.6b", "benchmark_id": "mmmlu", "score": 43.1, "reference_url": "https://arxiv.org/abs/2505.09388", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Qwen3 tech report Thinking: MMMLU 14 langs 43.1", "candidates": [] }, { "model_id": "claude-sonnet-4.5", "benchmark_id": "hmmt_feb_2025", "score": 79.2, "reference_url": "https://huggingface.co/deepseek-ai/DeepSeek-V3.2/resolve/main/assets/paper.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "DeepSeek V3.2 tech report Table 2: claude-sonnet-4.5=79.2.", "candidates": [ { "score": 74.6, "reference_url": "https://huggingface.co/moonshotai/Kimi-K2-Thinking", "source_type": "model_card", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "1.0" }, "notes": "Kimi K2-Thinking model card: claude-sonnet-4.5=74.6." } ] }, { "model_id": "gpt-5", "benchmark_id": "hmmt_feb_2025", "score": 88.3, "reference_url": "https://huggingface.co/deepseek-ai/DeepSeek-V3.2/resolve/main/assets/paper.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "DeepSeek V3.2 tech report Table 2: gpt-5=88.3.", "candidates": [ { "score": 93.3, "reference_url": "https://huggingface.co/moonshotai/Kimi-K2-Thinking", "source_type": "model_card", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "1.0" }, "notes": "Kimi K2-Thinking model card: gpt-5=93.3." }, { "score": 93.3, "reference_url": "https://openai.com/index/introducing-gpt-5-for-developers/", "source_type": "official_blog", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "notes": "(merged from hmmt_2025) OpenAI GPT-5 dev blog: gpt-5=93.3 (high reasoning effort)." }, { "score": 93.3, "reference_url": "https://openai.com/index/introducing-gpt-5-for-developers/", "source_type": "official_blog", "notes": "Dev blog says HMMT 2025 no tools=93.3% for GPT-5 (high). Primary=88.3 from DeepSeek third-party self-test." } ] }, { "model_id": "gemini-3-pro", "benchmark_id": "hmmt_feb_2025", "score": 97.5, "reference_url": "https://huggingface.co/deepseek-ai/DeepSeek-V3.2/resolve/main/assets/paper.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "DeepSeek V3.2 tech report Table 2: gemini-3-pro=97.5.", "candidates": [ { "score": 97.3, "reference_url": "https://huggingface.co/moonshotai/Kimi-K2.5", "source_type": "model_card", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "1.0" }, "notes": "Kimi K2.5 model card: gemini-3-pro=97.3." }, { "score": 97.3, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "source_type": "model_card", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "notes": "Doubao Seed 2.0 model card Table 3: gemini-3-pro=97.3." }, { "score": 97.3, "reference_url": "https://huggingface.co/Qwen/Qwen3.5-397B-A17B", "source_type": "model_card", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "notes": "(merged from hmmt_2025) Qwen3.5-397B-A17B HF model card cross-model table (Qwen self-test of 3rd-party model, mc=false): hmmt_2025 97.3 (3rd-party Qwen self-test). [R5d: prior unverified value 97.5 from https://matharena.ai/ deleted.]" } ] }, { "model_id": "kimi-k2-thinking", "benchmark_id": "hmmt_feb_2025", "score": 89.4, "reference_url": "https://huggingface.co/deepseek-ai/DeepSeek-V3.2/resolve/main/assets/paper.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "DeepSeek V3.2 tech report Table 2: kimi-k2-thinking=89.4.", "candidates": [] }, { "model_id": "deepseek-v3.2", "benchmark_id": "hmmt_feb_2025", "score": 92.5, "reference_url": "https://huggingface.co/deepseek-ai/DeepSeek-V3.2/resolve/main/assets/paper.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "1.0" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "DeepSeek V3.2 tech report Table 2: deepseek-v3.2=92.5.", "candidates": [ { "score": 83.6, "reference_url": "https://huggingface.co/moonshotai/Kimi-K2-Thinking", "source_type": "model_card", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "1.0" }, "notes": "Kimi K2-Thinking model card: deepseek-v3.2=83.6." }, { "score": 90.2, "reference_url": "https://llm-stats.com/benchmarks/hmmt-2025", "source_type": "third_party_aggregator", "reported_setting": { "mode": "thinking (DeepSeek-Reasoner endpoint)" }, "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=DeepSeek-V3.2 (Thinking), slug=deepseek-reasoner, provider=DeepSeek" }, { "score": 83.6, "reference_url": "https://llm-stats.com/benchmarks/hmmt-2025", "source_type": "third_party_aggregator", "reported_setting": { "variant": "exp" }, "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=DeepSeek-V3.2-Exp, slug=deepseek-v3.2-exp, provider=DeepSeek" } ] }, { "model_id": "claude-sonnet-4.5", "benchmark_id": "hmmt_nov_2025", "score": 81.7, "reference_url": "https://huggingface.co/deepseek-ai/DeepSeek-V3.2/resolve/main/assets/paper.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "DeepSeek V3.2 tech report Table 2: claude-sonnet-4.5=81.7.", "candidates": [] }, { "model_id": "gpt-5", "benchmark_id": "imo_answerbench", "score": 76.0, "reference_url": "https://huggingface.co/deepseek-ai/DeepSeek-V3.2/resolve/main/assets/paper.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "DeepSeek V3.2 tech report Table 2: gpt-5=76.0.", "candidates": [] }, { "model_id": "gemini-3-pro", "benchmark_id": "imo_answerbench", "score": 83.3, "reference_url": "https://huggingface.co/deepseek-ai/DeepSeek-V3.2/resolve/main/assets/paper.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "DeepSeek V3.2 tech report Table 2: gemini-3-pro=83.3.", "candidates": [ { "score": 83.1, "reference_url": "https://huggingface.co/moonshotai/Kimi-K2.5", "source_type": "model_card", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "1.0" }, "notes": "(merged from imoanswerbench) Kimi K2.5 model card: gemini-3-pro=83.1." } ] }, { "model_id": "deepseek-v3.2", "benchmark_id": "imo_answerbench", "score": 78.3, "reference_url": "https://huggingface.co/deepseek-ai/DeepSeek-V3.2/resolve/main/assets/paper.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "1.0" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "DeepSeek V3.2 tech report Table 2: deepseek-v3.2=78.3.", "candidates": [] }, { "model_id": "claude-sonnet-4.5", "benchmark_id": "swe_bench_multilingual", "score": 68.0, "reference_url": "https://huggingface.co/deepseek-ai/DeepSeek-V3.2/resolve/main/assets/paper.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "agentic (bash + edit)", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "DeepSeek V3.2 tech report Table 2: claude-sonnet-4.5=68.0.", "candidates": [ { "score": 64.1, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "source_type": "model_card", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "varies", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "notes": "(merged from swe_multilingual) Doubao Seed 2.0 model card: claude-sonnet-4.5=64.1." } ] }, { "model_id": "gpt-5", "benchmark_id": "swe_bench_multilingual", "score": 55.3, "reference_url": "https://huggingface.co/deepseek-ai/DeepSeek-V3.2/resolve/main/assets/paper.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "agentic (bash + edit)", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "DeepSeek V3.2 tech report Table 2: gpt-5=55.3.", "candidates": [] }, { "model_id": "kimi-k2-thinking", "benchmark_id": "swe_bench_multilingual", "score": 61.1, "reference_url": "https://huggingface.co/deepseek-ai/DeepSeek-V3.2/resolve/main/assets/paper.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "agentic (bash + edit)", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "DeepSeek V3.2 tech report Table 2: kimi-k2-thinking=61.1.", "candidates": [] }, { "model_id": "minimax-m2", "benchmark_id": "swe_bench_multilingual", "score": 56.5, "reference_url": "https://huggingface.co/deepseek-ai/DeepSeek-V3.2/resolve/main/assets/paper.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "agentic (bash + edit)", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "DeepSeek V3.2 tech report Table 2: minimax-m2=56.5.", "candidates": [] }, { "model_id": "deepseek-v3.2", "benchmark_id": "swe_bench_multilingual", "score": 70.2, "reference_url": "https://huggingface.co/deepseek-ai/DeepSeek-V3.2/resolve/main/assets/paper.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "agentic (bash + edit)", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "1.0" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "DeepSeek V3.2 tech report Table 2: deepseek-v3.2=70.2.", "candidates": [ { "score": 57.9, "reference_url": "https://huggingface.co/MiniMaxAI/MiniMax-M2", "source_type": "model_card", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "agentic (bash + edit)", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "notes": "MiniMax M2 model card: deepseek-v3.2=57.9." }, { "score": 70.2, "reference_url": "https://llm-stats.com/benchmarks/swe-bench-multilingual", "source_type": "third_party_aggregator", "reported_setting": { "mode": "thinking (DeepSeek-Reasoner endpoint)" }, "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=DeepSeek-V3.2 (Thinking), slug=deepseek-reasoner, provider=DeepSeek" }, { "score": 57.9, "reference_url": "https://llm-stats.com/benchmarks/swe-bench-multilingual", "source_type": "third_party_aggregator", "reported_setting": { "variant": "exp" }, "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=DeepSeek-V3.2-Exp, slug=deepseek-v3.2-exp, provider=DeepSeek" } ] }, { "model_id": "deepseek-v3.2", "benchmark_id": "tau2_bench_airline", "score": 63.8, "reference_url": "https://huggingface.co/deepseek-ai/DeepSeek-V3.2/resolve/main/assets/paper.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "agentic", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "1.0" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "DeepSeek V3.2 tech report Table 2: deepseek-v3.2=63.8.", "candidates": [] }, { "model_id": "deepseek-v3.2", "benchmark_id": "tau2_bench_retail", "score": 81.1, "reference_url": "https://huggingface.co/deepseek-ai/DeepSeek-V3.2/resolve/main/assets/paper.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "agentic", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "1.0" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "DeepSeek V3.2 tech report Table 2: deepseek-v3.2=81.1.", "candidates": [] }, { "model_id": "deepseek-v3.2", "benchmark_id": "tau2_bench_telecom", "score": 96.2, "reference_url": "https://huggingface.co/deepseek-ai/DeepSeek-V3.2/resolve/main/assets/paper.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "agentic", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "1.0" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "DeepSeek V3.2 tech report Table 2: deepseek-v3.2=96.2.", "candidates": [ { "score": 34, "reference_url": "https://huggingface.co/MiniMaxAI/MiniMax-M2", "source_type": "model_card", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "agentic", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "notes": "MiniMax M2 model card: deepseek-v3.2=34." } ] }, { "model_id": "kimi-k2", "benchmark_id": "mmlu_redux", "score": 92.7, "reference_url": "https://arxiv.org/abs/2507.20534", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none (per-bench override)", "sampling": "pass@1 (avg of 4-64 trials per benchmark)", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Kimi K2 tech report (arxiv:2507.20534) Table 3 Kimi-K2-Instruct column: MMLU-Redux EM 92.7", "candidates": [ { "score": 92.7, "reference_url": "https://llm-stats.com/benchmarks/mmlu-redux", "source_type": "third_party_aggregator", "reported_setting": { "variant": "instruct (BP kimi-k2 is base)" }, "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=Kimi K2 Instruct, slug=kimi-k2-instruct, provider=Moonshot AI" }, { "score": 92.7, "reference_url": "https://llm-stats.com/benchmarks/mmlu-redux", "source_type": "third_party_aggregator", "reported_setting": { "variant": "instruct", "snapshot": "0905" }, "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=Kimi K2-Instruct-0905, slug=kimi-k2-instruct-0905, provider=Moonshot AI" }, { "score": 90.2, "reference_url": "https://huggingface.co/XiaomiMiMo/MiMo-V2.5-Pro/resolve/21d1ecfecd7bd70f31be25ca49d7edd21f003659/README.md", "source_type": "official_model_card_base_checkpoint", "reported_setting": { "mode": "base pretrained checkpoint", "effort": "not applicable", "tools": "none", "sampling": "5-shot; single completion per prompt", "judge": "benchmark-native rule-based evaluator", "harness": "Xiaomi MiMo base-checkpoint evaluation", "prompt_style": "5-shot few-shot continuation", "temperature": "source does not state", "context": "source does not state", "input_modalities": "text", "trials": "one completion per item", "checkpoint": "base" }, "notes": "Base checkpoint observation attached to the canonical post-trained model ID. Base result is attached to the canonical post-trained model ID. The Xiaomi table does not state this comparator Base checkpoint's context length, and no comparator Base config is locked; no Pro context value or conflict is copied onto this observation." } ] }, { "model_id": "kimi-k2", "benchmark_id": "swe_bench_multilingual", "score": 47.3, "reference_url": "https://arxiv.org/abs/2507.20534", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none (per-bench override)", "sampling": "pass@1 (avg of 4-64 trials per benchmark)", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Kimi K2 tech report (arxiv:2507.20534) Table 3 Kimi-K2-Instruct column: SWE-bench Multilingual Pass@1 47.3", "candidates": [ { "score": 55.9, "reference_url": "https://huggingface.co/MiniMaxAI/MiniMax-M2", "source_type": "model_card", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "agentic (bash + edit)", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "notes": "MiniMax M2 model card: kimi-k2=55.9." }, { "score": 47.3, "reference_url": "https://llm-stats.com/benchmarks/swe-bench-multilingual", "source_type": "third_party_aggregator", "reported_setting": { "variant": "instruct", "snapshot": "0905" }, "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=Kimi K2-Instruct-0905, slug=kimi-k2-instruct-0905, provider=Moonshot AI" }, { "score": 47.3, "reference_url": "https://llm-stats.com/benchmarks/swe-bench-multilingual", "source_type": "third_party_aggregator", "reported_setting": { "variant": "instruct (BP kimi-k2 is base)" }, "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=Kimi K2 Instruct, slug=kimi-k2-instruct, provider=Moonshot AI" } ] }, { "model_id": "kimi-k2", "benchmark_id": "tau2_bench_retail", "score": 70.6, "reference_url": "https://arxiv.org/abs/2507.20534", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none (per-bench override)", "sampling": "pass@1 (avg of 4-64 trials per benchmark)", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Kimi K2 tech report (arxiv:2507.20534) Table 3 Kimi-K2-Instruct column: Tau2 retail Avg@4 70.6", "candidates": [] }, { "model_id": "kimi-k2", "benchmark_id": "tau2_bench_airline", "score": 56.5, "reference_url": "https://arxiv.org/abs/2507.20534", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none (per-bench override)", "sampling": "pass@1 (avg of 4-64 trials per benchmark)", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Kimi K2 tech report (arxiv:2507.20534) Table 3 Kimi-K2-Instruct column: Tau2 airline Avg@4 56.5", "candidates": [] }, { "model_id": "kimi-k2", "benchmark_id": "tau2_bench_telecom", "score": 65.8, "reference_url": "https://arxiv.org/abs/2507.20534", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none (per-bench override)", "sampling": "pass@1 (avg of 4-64 trials per benchmark)", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Kimi K2 tech report (arxiv:2507.20534) Table 3 Kimi-K2-Instruct column: Tau2 telecom Avg@4 65.8", "candidates": [ { "score": 73, "reference_url": "https://huggingface.co/MiniMaxAI/MiniMax-M2", "source_type": "model_card", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "agentic", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "notes": "MiniMax M2 model card: kimi-k2=73." } ] }, { "model_id": "kimi-k2", "benchmark_id": "supergpqa", "score": 57.2, "reference_url": "https://arxiv.org/abs/2507.20534", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none (per-bench override)", "sampling": "pass@1 (avg of 4-64 trials per benchmark)", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Kimi K2 tech report (arxiv:2507.20534) Table 3 Kimi-K2-Instruct column: SuperGPQA Acc 57.2", "candidates": [ { "score": 57.2, "reference_url": "https://llm-stats.com/benchmarks/supergpqa", "source_type": "third_party_aggregator", "reported_setting": { "variant": "instruct (BP kimi-k2 is base)" }, "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=Kimi K2 Instruct, slug=kimi-k2-instruct, provider=Moonshot AI" }, { "score": 57.2, "reference_url": "https://llm-stats.com/benchmarks/supergpqa", "source_type": "third_party_aggregator", "reported_setting": { "variant": "instruct", "snapshot": "0905" }, "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=Kimi K2-Instruct-0905, slug=kimi-k2-instruct-0905, provider=Moonshot AI" } ] }, { "model_id": "kimi-k2", "benchmark_id": "livebench", "score": 76.4, "reference_url": "https://arxiv.org/abs/2507.20534", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none (per-bench override)", "sampling": "pass@1 (avg of 4-64 trials per benchmark)", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Kimi K2 tech report (arxiv:2507.20534) Table 3 Kimi-K2-Instruct column: Livebench Pass@1 76.4", "candidates": [ { "score": 76.4, "reference_url": "https://llm-stats.com/benchmarks/livebench", "source_type": "third_party_aggregator", "reported_setting": { "variant": "instruct", "snapshot": "0905" }, "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=Kimi K2-Instruct-0905, slug=kimi-k2-instruct-0905, provider=Moonshot AI" }, { "score": 76.4, "reference_url": "https://llm-stats.com/benchmarks/livebench", "source_type": "third_party_aggregator", "reported_setting": { "variant": "instruct (BP kimi-k2 is base)" }, "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=Kimi K2 Instruct, slug=kimi-k2-instruct, provider=Moonshot AI" } ] }, { "model_id": "kimi-k2-thinking", "benchmark_id": "imo_answerbench", "score": 78.6, "reference_url": "https://moonshotai.github.io/Kimi-K2/thinking.html", "reported_setting": { "mode": "thinking", "effort": "default (96k thinking-token budget for HLE/AIME/HMMT/GPQA; 128k for IMO/LCB/OJ)", "tools": "none (per-bench override)", "sampling": "pass@1 (avg@32 for AIME/HMMT no-tools)", "judge": "rule-based (HLE uses o3-mini)", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "Kimi K2 Thinking blog cross-model table: IMO-AnswerBench no-tools avg@8 78.6", "candidates": [] }, { "model_id": "kimi-k2-thinking", "benchmark_id": "mmlu_redux", "score": 94.4, "reference_url": "https://moonshotai.github.io/Kimi-K2/thinking.html", "reported_setting": { "mode": "thinking", "effort": "default (96k thinking-token budget for HLE/AIME/HMMT/GPQA; 128k for IMO/LCB/OJ)", "tools": "none (per-bench override)", "sampling": "pass@1 (avg@32 for AIME/HMMT no-tools)", "judge": "rule-based (HLE uses o3-mini)", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "Kimi K2 Thinking blog cross-model table: MMLU-Redux no-tools 94.4", "candidates": [] }, { "model_id": "kimi-k2-thinking", "benchmark_id": "multi_swe_bench", "score": 41.9, "reference_url": "https://moonshotai.github.io/Kimi-K2/thinking.html", "reported_setting": { "mode": "thinking", "effort": "default (96k thinking-token budget for HLE/AIME/HMMT/GPQA; 128k for IMO/LCB/OJ)", "tools": "none (per-bench override)", "sampling": "pass@1 (avg@32 for AIME/HMMT no-tools)", "judge": "rule-based (HLE uses o3-mini)", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "Kimi K2 Thinking blog cross-model table: Multi-SWE-bench w/ tools 41.9", "candidates": [] }, { "model_id": "kimi-k2-thinking", "benchmark_id": "scicode", "score": 44.8, "reference_url": "https://moonshotai.github.io/Kimi-K2/thinking.html", "reported_setting": { "mode": "thinking", "effort": "default (96k thinking-token budget for HLE/AIME/HMMT/GPQA; 128k for IMO/LCB/OJ)", "tools": "none (per-bench override)", "sampling": "pass@1 (avg@32 for AIME/HMMT no-tools)", "judge": "rule-based (HLE uses o3-mini)", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "Kimi K2 Thinking blog cross-model table: SciCode no-tools 44.8", "candidates": [] }, { "model_id": "kimi-k2-thinking", "benchmark_id": "ojbench", "score": 48.7, "reference_url": "https://moonshotai.github.io/Kimi-K2/thinking.html", "reported_setting": { "mode": "thinking", "effort": "default (96k thinking-token budget for HLE/AIME/HMMT/GPQA; 128k for IMO/LCB/OJ)", "tools": "none (per-bench override)", "sampling": "pass@1 (avg@32 for AIME/HMMT no-tools)", "judge": "rule-based (HLE uses o3-mini)", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "Kimi K2 Thinking blog cross-model table: OJ-Bench cpp no-tools 48.7", "candidates": [] }, { "model_id": "kimi-k2-thinking", "benchmark_id": "browsecomp_zh", "score": 62.3, "reference_url": "https://moonshotai.github.io/Kimi-K2/thinking.html", "reported_setting": { "mode": "thinking", "effort": "default (96k thinking-token budget for HLE/AIME/HMMT/GPQA; 128k for IMO/LCB/OJ)", "tools": "none (per-bench override)", "sampling": "pass@1 (avg@32 for AIME/HMMT no-tools)", "judge": "rule-based (HLE uses o3-mini)", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "Kimi K2 Thinking blog cross-model table: BrowseComp-ZH w/ tools 62.3", "candidates": [] }, { "model_id": "kimi-k2-thinking", "benchmark_id": "frames", "score": 87.0, "reference_url": "https://moonshotai.github.io/Kimi-K2/thinking.html", "reported_setting": { "mode": "thinking", "effort": "default (96k thinking-token budget for HLE/AIME/HMMT/GPQA; 128k for IMO/LCB/OJ)", "tools": "none (per-bench override)", "sampling": "pass@1 (avg@32 for AIME/HMMT no-tools)", "judge": "rule-based (HLE uses o3-mini)", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "Kimi K2 Thinking blog cross-model table: Frames w/ tools 87.0", "candidates": [] }, { "model_id": "kimi-k2.5", "benchmark_id": "mathvista_mini", "score": 90.1, "reference_url": "https://huggingface.co/moonshotai/Kimi-K2.5", "reported_setting": { "mode": "thinking", "effort": "default (96k for reasoning, 64k for vision)", "tools": "none (per-bench override)", "sampling": "pass@1 (avg@32 AIME/HMMT, avg@8 GPQA, avg@3 vision, avg@4 Seal/WideSearch)", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "1.0" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Kimi K2.5 model card (HF moonshotai/Kimi-K2.5) Evaluation table: MathVista mini 90.1", "candidates": [] }, { "model_id": "kimi-k2.5", "benchmark_id": "ocrbench", "score": 92.3, "reference_url": "https://huggingface.co/moonshotai/Kimi-K2.5", "reported_setting": { "mode": "thinking", "effort": "default (96k for reasoning, 64k for vision)", "tools": "none (per-bench override)", "sampling": "pass@1 (avg@32 AIME/HMMT, avg@8 GPQA, avg@3 vision, avg@4 Seal/WideSearch)", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "1.0" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Kimi K2.5 model card (HF moonshotai/Kimi-K2.5) Evaluation table: OCRBench 92.3", "candidates": [] }, { "model_id": "kimi-k2.5", "benchmark_id": "omnidocbench", "score": 0.112, "reference_url": "https://huggingface.co/moonshotai/Kimi-K2.5", "reported_setting": { "mode": "thinking", "effort": "default (96k for reasoning, 64k for vision)", "tools": "none (per-bench override)", "sampling": "pass@1 (avg@32 AIME/HMMT, avg@8 GPQA, avg@3 vision, avg@4 Seal/WideSearch)", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "1.0" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Kimi K2.5 model card (HF moonshotai/Kimi-K2.5) Evaluation table: OmniDocBench 1.5 (1 - normalized Levenshtein) ×100 = 88.8 [Converted 88.8% (1-LED higher-better) → 0.112 (edit distance lower-better) for cross-source consistency]", "candidates": [] }, { "model_id": "kimi-k2.5", "benchmark_id": "longbench_v2", "score": 61.0, "reference_url": "https://huggingface.co/moonshotai/Kimi-K2.5", "reported_setting": { "mode": "thinking", "effort": "default (96k for reasoning, 64k for vision)", "tools": "none (per-bench override)", "sampling": "pass@1 (avg@32 AIME/HMMT, avg@8 GPQA, avg@3 vision, avg@4 Seal/WideSearch)", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "1.0" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Kimi K2.5 model card (HF moonshotai/Kimi-K2.5) Evaluation table: LongBench v2 61.0", "candidates": [] }, { "model_id": "kimi-k2.5", "benchmark_id": "paperbench", "score": 63.5, "reference_url": "https://huggingface.co/moonshotai/Kimi-K2.5", "reported_setting": { "mode": "thinking", "effort": "default (96k for reasoning, 64k for vision)", "tools": "none (per-bench override)", "sampling": "pass@1 (avg@32 AIME/HMMT, avg@8 GPQA, avg@3 vision, avg@4 Seal/WideSearch)", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "1.0" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Kimi K2.5 model card (HF moonshotai/Kimi-K2.5) Evaluation table: PaperBench 63.5", "candidates": [] }, { "model_id": "kimi-k2.5", "benchmark_id": "cybergym", "score": 41.3, "reference_url": "https://huggingface.co/moonshotai/Kimi-K2.5", "reported_setting": { "mode": "thinking", "effort": "default (96k for reasoning, 64k for vision)", "tools": "none (per-bench override)", "sampling": "pass@1 (avg@32 AIME/HMMT, avg@8 GPQA, avg@3 vision, avg@4 Seal/WideSearch)", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "1.0" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Kimi K2.5 model card (HF moonshotai/Kimi-K2.5) Evaluation table: CyberGym 41.3", "candidates": [] }, { "model_id": "kimi-k2.5", "benchmark_id": "ojbench", "score": 57.4, "reference_url": "https://huggingface.co/moonshotai/Kimi-K2.5", "reported_setting": { "mode": "thinking", "effort": "default (96k for reasoning, 64k for vision)", "tools": "none (per-bench override)", "sampling": "pass@1 (avg@32 AIME/HMMT, avg@8 GPQA, avg@3 vision, avg@4 Seal/WideSearch)", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "1.0" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Kimi K2.5 model card (HF moonshotai/Kimi-K2.5) Evaluation table: OJBench cpp 57.4", "candidates": [ { "score": 54.7, "reference_url": "https://huggingface.co/moonshotai/Kimi-K2.6", "source_type": "model_card", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "1.0" }, "notes": "Kimi K2.6 model card: kimi-k2.5=54.7." } ] }, { "model_id": "kimi-k2.5", "benchmark_id": "video_mme", "score": 87.4, "reference_url": "https://huggingface.co/moonshotai/Kimi-K2.5", "reported_setting": { "mode": "thinking", "effort": "default (96k for reasoning, 64k for vision)", "tools": "none (per-bench override)", "sampling": "pass@1 (avg@32 AIME/HMMT, avg@8 GPQA, avg@3 vision, avg@4 Seal/WideSearch)", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "1.0" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Kimi K2.5 model card (HF moonshotai/Kimi-K2.5) Evaluation table: VideoMME 87.4", "candidates": [] }, { "model_id": "kimi-k2.6", "benchmark_id": "babyvision", "score": 39.8, "reference_url": "https://huggingface.co/moonshotai/Kimi-K2.6", "reported_setting": { "mode": "thinking", "effort": "max (98304 generation tokens)", "tools": "none (per-bench override)", "sampling": "pass@1 (avg@10 coding, avg@3 vision)", "judge": "rule-based", "harness": "official (in-house SWE-agent-derived for SWE-Bench)", "prompt_style": "default", "temperature": "1.0" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Kimi K2.6 model card (HF moonshotai/Kimi-K2.6) Evaluation table: BabyVision 39.8", "candidates": [] }, { "model_id": "kimi-k2.6", "benchmark_id": "claw_eval_pass3", "score": 62.3, "reference_url": "https://huggingface.co/moonshotai/Kimi-K2.6", "reported_setting": { "mode": "thinking", "effort": "max (98304 generation tokens)", "tools": "none (per-bench override)", "sampling": "pass@1 (avg@10 coding, avg@3 vision)", "judge": "rule-based", "harness": "official (in-house SWE-agent-derived for SWE-Bench)", "prompt_style": "default", "temperature": "1.0" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Kimi K2.6 model card (HF moonshotai/Kimi-K2.6) Evaluation table: Claw Eval pass^3 62.3", "candidates": [ { "score": 62.3, "reference_url": "https://raw.githubusercontent.com/claw-eval/claw-eval.github.io/19d378a646d892ca6c8148691aea02da1b184eb8/assets/benchmark-0fIVfCD7.js", "source_type": "official_blog", "reported_setting": { "mode": "source does not state", "effort": "source does not state", "tools": "official Claw-Eval agent environment", "sampling": "Pass^3; all three trajectories must pass", "judge": "full-trajectory benchmark grading", "harness": "official Claw-Eval v1.1 leaderboard", "prompt_style": "official Claw-Eval task prompts", "temperature": "source does not state", "context": "source does not state", "input_modalities": "text", "trials": "3 per task; 597 generations", "task_subset": "non-multimodal 199-task subset", "task_count": 199 }, "notes": "Exact official leaderboard field for the non-multimodal 199-task subset. Model mode/effort is not inferred." }, { "score": 61.5, "reference_url": "https://raw.githubusercontent.com/claw-eval/claw-eval.github.io/19d378a646d892ca6c8148691aea02da1b184eb8/assets/benchmark-0fIVfCD7.js", "source_type": "official_blog", "reported_setting": { "mode": "source does not state", "effort": "source does not state", "tools": "official Claw-Eval agent environment", "sampling": "Pass^3; all three trajectories must pass", "judge": "full-trajectory benchmark grading", "harness": "official Claw-Eval v1.1 leaderboard", "prompt_style": "official Claw-Eval task prompts", "temperature": "source does not state", "context": "source does not state", "input_modalities": "text", "trials": "3 per task; 483 generations", "task_subset": "core-general 161-task subset", "task_count": 161 }, "notes": "Average tokens per trajectory=120816 is a physical efficiency annotation, not a benchmark score or score setting. Exact official leaderboard field for the core-general 161-task subset. Model mode/effort is not inferred." } ] }, { "model_id": "kimi-k2.6", "benchmark_id": "mcpmark", "score": 55.9, "reference_url": "https://huggingface.co/moonshotai/Kimi-K2.6", "reported_setting": { "mode": "thinking", "effort": "max (98304 generation tokens)", "tools": "none (per-bench override)", "sampling": "pass@1 (avg@10 coding, avg@3 vision)", "judge": "rule-based", "harness": "official (in-house SWE-agent-derived for SWE-Bench)", "prompt_style": "default", "temperature": "1.0" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Kimi K2.6 model card (HF moonshotai/Kimi-K2.6) Evaluation table: MCPMark 55.9", "candidates": [] }, { "model_id": "deepseek-v3.2-speciale", "benchmark_id": "imo_answerbench", "score": 84.5, "reference_url": "https://huggingface.co/deepseek-ai/DeepSeek-V3.2/resolve/main/assets/paper.pdf", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "1.0" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "DS V3.2 tech report Table 3: V3.2-Speciale=84.5.", "candidates": [] }, { "model_id": "deepseek-r1", "benchmark_id": "mmlu_redux", "score": 92.9, "reference_url": "https://huggingface.co/deepseek-ai/DeepSeek-R1-0528", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "DeepSeek R1-0528 model card: deepseek-r1=92.9.", "candidates": [] }, { "model_id": "deepseek-r1-0528", "benchmark_id": "mmlu_redux", "score": 93.4, "reference_url": "https://huggingface.co/deepseek-ai/DeepSeek-R1-0528", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "DeepSeek R1-0528 model card: deepseek-r1-0528=93.4.", "candidates": [] }, { "model_id": "deepseek-r1", "benchmark_id": "frames", "score": 82.5, "reference_url": "https://huggingface.co/deepseek-ai/DeepSeek-R1-0528", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "DeepSeek R1-0528 model card: deepseek-r1=82.5.", "candidates": [] }, { "model_id": "deepseek-r1-0528", "benchmark_id": "frames", "score": 83.0, "reference_url": "https://huggingface.co/deepseek-ai/DeepSeek-R1-0528", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "DeepSeek R1-0528 model card: deepseek-r1-0528=83.0.", "candidates": [] }, { "model_id": "deepseek-r1", "benchmark_id": "aider_polyglot_diff", "score": 53.3, "reference_url": "https://huggingface.co/deepseek-ai/DeepSeek-R1-0528", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "agentic (bash + edit)", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "DeepSeek R1-0528 model card: deepseek-r1=53.3.", "candidates": [ { "score": 56.9, "reference_url": "https://arxiv.org/abs/2504.13914", "source_type": "tech_report", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "agentic (bash + edit)", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "notes": "Seed-Thinking-v1.5 paper Table 2: deepseek-r1=56.9." } ] }, { "model_id": "deepseek-r1-0528", "benchmark_id": "aider_polyglot_diff", "score": 71.6, "reference_url": "https://huggingface.co/deepseek-ai/DeepSeek-R1-0528", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "agentic (bash + edit)", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "DeepSeek R1-0528 model card: deepseek-r1-0528=71.6.", "candidates": [] }, { "model_id": "deepseek-r1", "benchmark_id": "cnmo_2024", "score": 78.8, "reference_url": "https://huggingface.co/deepseek-ai/DeepSeek-R1-0528", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "DeepSeek R1-0528 model card: deepseek-r1=78.8.", "candidates": [] }, { "model_id": "deepseek-r1-0528", "benchmark_id": "cnmo_2024", "score": 86.9, "reference_url": "https://huggingface.co/deepseek-ai/DeepSeek-R1-0528", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "DeepSeek R1-0528 model card: deepseek-r1-0528=86.9.", "candidates": [] }, { "model_id": "deepseek-r1-0528", "benchmark_id": "bfcl_v3_multiturn", "score": 37.0, "reference_url": "https://huggingface.co/deepseek-ai/DeepSeek-R1-0528", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "agentic", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "DeepSeek R1-0528 model card: deepseek-r1-0528=37.0.", "candidates": [] }, { "model_id": "deepseek-r1-0528", "benchmark_id": "tau_bench_airline", "score": 53.5, "reference_url": "https://huggingface.co/deepseek-ai/DeepSeek-R1-0528", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "agentic", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "DeepSeek R1-0528 model card: deepseek-r1-0528=53.5.", "candidates": [] }, { "model_id": "deepseek-r1-0528", "benchmark_id": "tau_bench_retail", "score": 63.9, "reference_url": "https://huggingface.co/deepseek-ai/DeepSeek-R1-0528", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "agentic", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "DeepSeek R1-0528 model card: deepseek-r1-0528=63.9.", "candidates": [] }, { "model_id": "gemini-2.5-flash", "benchmark_id": "aider_polyglot_whole", "score": 61.9, "reference_url": "https://storage.googleapis.com/deepmind-media/Model-Cards/Gemini-2-5-Flash-Model-Card.pdf", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "AI Studio API", "prompt_style": "whole", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Gemini 2.5 Flash Model Card (Dec 2025) GA Thinking column: Aider Polyglot whole 61.9 (whole/diff: 61.9/56.7)", "candidates": [] }, { "model_id": "gemini-2.5-flash", "benchmark_id": "facts_grounding", "score": 85.3, "reference_url": "https://storage.googleapis.com/deepmind-media/Model-Cards/Gemini-2-5-Flash-Model-Card.pdf", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "AI Studio API", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Gemini 2.5 Flash Model Card (Dec 2025) GA Thinking column: FACTS Grounding 85.3", "candidates": [] }, { "model_id": "gemini-2.5-flash", "benchmark_id": "vibe_eval", "score": 65.4, "reference_url": "https://storage.googleapis.com/deepmind-media/Model-Cards/Gemini-2-5-Flash-Model-Card.pdf", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "Gemini", "harness": "AI Studio API", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Gemini 2.5 Flash Model Card (Dec 2025) GA Thinking column: Vibe-Eval (Reka) 65.4 (Gemini-as-judge)", "candidates": [] }, { "model_id": "gemini-2.5-flash", "benchmark_id": "mrcr_v1", "score": 74.0, "reference_url": "https://storage.googleapis.com/deepmind-media/Model-Cards/Gemini-2-5-Flash-Model-Card.pdf", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "AI Studio API", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Gemini 2.5 Flash Model Card (Dec 2025) GA Thinking column: MRCR (v1) 128k average 74", "candidates": [] }, { "model_id": "gemini-2.5-flash", "benchmark_id": "global_mmlu_lite", "score": 88.4, "reference_url": "https://storage.googleapis.com/deepmind-media/Model-Cards/Gemini-2-5-Flash-Model-Card.pdf", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "AI Studio API", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Gemini 2.5 Flash Model Card (Dec 2025) GA Thinking column: Global MMLU Lite 88.4", "candidates": [] }, { "model_id": "deepseek-v2-0506", "benchmark_id": "mmlu", "score": 78.2, "reference_url": "https://arxiv.org/abs/2412.19437", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "DeepSeek V3 paper Table 6: deepseek-v2-0506=78.2.", "candidates": [] }, { "model_id": "deepseek-v2.5-0905", "benchmark_id": "mmlu", "score": 80.6, "reference_url": "https://arxiv.org/abs/2412.19437", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "DeepSeek V3 paper Table 6: deepseek-v2.5-0905=80.6.", "candidates": [ { "score": 80.4, "reference_url": "https://llm-stats.com/benchmarks/mmlu", "source_type": "third_party_aggregator", "reported_setting": { "snapshot": "0905" }, "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=DeepSeek-V2.5, slug=deepseek-v2.5, provider=DeepSeek" } ] }, { "model_id": "qwen2.5-72b-instruct", "benchmark_id": "mmlu", "score": 85.3, "reference_url": "https://arxiv.org/abs/2412.19437", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "DeepSeek V3 paper Table 6: qwen2.5-72b-instruct=85.3.", "candidates": [ { "score": 85.5, "reference_url": "https://arxiv.org/abs/2501.00656", "source_type": "tech_report", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "OLMES", "prompt_style": "default", "temperature": "default" }, "notes": "OLMo 2 paper Table 7 Instruct: alt measurement 85.5 (mc=false)." } ] }, { "model_id": "llama-3.1-405b-instruct", "benchmark_id": "mmlu", "score": 88.6, "reference_url": "https://arxiv.org/abs/2412.19437", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "DeepSeek V3 paper Table 6: llama-3.1-405b-instruct=88.6.", "candidates": [] }, { "model_id": "claude-3.5-sonnet", "benchmark_id": "mmlu", "score": 88.3, "reference_url": "https://arxiv.org/abs/2412.19437", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "DeepSeek V3 paper Table 6: claude-3.5-sonnet=88.3.", "candidates": [ { "score": 89.5, "reference_url": "https://cohere.com/research/papers/command-a-technical-report.pdf", "source_type": "tech_report", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "Cohere internal eval", "prompt_style": "default", "temperature": "default" }, "notes": "Command A paper: alt measurement 89.5 (mc=false)." } ] }, { "model_id": "gpt-4o-0513", "benchmark_id": "mmlu", "score": 87.2, "reference_url": "https://arxiv.org/abs/2412.19437", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "DeepSeek V3 paper Table 6: gpt-4o-0513=87.2.", "candidates": [ { "score": 88.1, "reference_url": "https://arxiv.org/abs/2412.08905", "source_type": "tech_report", "reported_setting": { "mode": "unknown", "effort": "unknown", "tools": "unknown", "sampling": "unknown", "judge": "rule-based", "harness": "Microsoft Phi internal eval", "prompt_style": "unknown", "temperature": "unknown" }, "notes": "Phi-4 paper Table 1 (gpt-4o-0513 column): mmlu=88.1 (alt measurement, mc=false)" }, { "score": 88.5, "reference_url": "https://cohere.com/research/papers/command-a-technical-report.pdf", "source_type": "tech_report", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "Cohere internal eval", "prompt_style": "default", "temperature": "default" }, "notes": "Command A paper: alt measurement 88.5 (mc=false)." }, { "score": 89.2, "reference_url": "https://cohere.com/research/papers/command-a-technical-report.pdf", "source_type": "tech_report", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "Cohere internal eval", "prompt_style": "default", "temperature": "default" }, "notes": "Command A paper: alt measurement 89.2 (mc=false)." } ] }, { "model_id": "deepseek-v2-0506", "benchmark_id": "mmlu_redux", "score": 77.9, "reference_url": "https://arxiv.org/abs/2412.19437", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "DeepSeek V3 paper Table 6: deepseek-v2-0506=77.9.", "candidates": [] }, { "model_id": "deepseek-v2.5-0905", "benchmark_id": "mmlu_redux", "score": 80.3, "reference_url": "https://arxiv.org/abs/2412.19437", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "DeepSeek V3 paper Table 6: deepseek-v2.5-0905=80.3.", "candidates": [] }, { "model_id": "qwen2.5-72b-instruct", "benchmark_id": "mmlu_redux", "score": 85.6, "reference_url": "https://arxiv.org/abs/2412.19437", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "DeepSeek V3 paper Table 6: qwen2.5-72b-instruct=85.6.", "candidates": [] }, { "model_id": "llama-3.1-405b-instruct", "benchmark_id": "mmlu_redux", "score": 86.2, "reference_url": "https://arxiv.org/abs/2412.19437", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "DeepSeek V3 paper Table 6: llama-3.1-405b-instruct=86.2.", "candidates": [] }, { "model_id": "claude-3.5-sonnet", "benchmark_id": "mmlu_redux", "score": 88.9, "reference_url": "https://arxiv.org/abs/2412.19437", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "DeepSeek V3 paper Table 6: claude-3.5-sonnet=88.9.", "candidates": [] }, { "model_id": "gpt-4o-0513", "benchmark_id": "mmlu_redux", "score": 88.0, "reference_url": "https://arxiv.org/abs/2412.19437", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "DeepSeek V3 paper Table 6: gpt-4o-0513=88.0.", "candidates": [] }, { "model_id": "deepseek-v3", "benchmark_id": "mmlu_redux", "score": 89.1, "reference_url": "https://arxiv.org/abs/2412.19437", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "DeepSeek V3 paper Table 6: deepseek-v3=89.1.", "candidates": [] }, { "model_id": "deepseek-v2-0506", "benchmark_id": "mmlu_pro", "score": 58.5, "reference_url": "https://arxiv.org/abs/2412.19437", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "DeepSeek V3 paper Table 6: deepseek-v2-0506=58.5.", "candidates": [] }, { "model_id": "deepseek-v2.5-0905", "benchmark_id": "mmlu_pro", "score": 66.2, "reference_url": "https://arxiv.org/abs/2412.19437", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "DeepSeek V3 paper Table 6: deepseek-v2.5-0905=66.2.", "candidates": [] }, { "model_id": "qwen2.5-72b-instruct", "benchmark_id": "mmlu_pro", "score": 71.6, "reference_url": "https://arxiv.org/abs/2412.19437", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "DeepSeek V3 paper Table 6: qwen2.5-72b-instruct=71.6.", "candidates": [ { "score": 69.6, "reference_url": "https://arxiv.org/abs/2412.08905", "source_type": "tech_report", "reported_setting": { "mode": "unknown", "effort": "unknown", "tools": "unknown", "sampling": "unknown", "judge": "rule-based", "harness": "Microsoft Phi internal eval", "prompt_style": "unknown", "temperature": "unknown" }, "notes": "Phi-4 paper Table 1 (qwen2.5-72b-instruct column): mmlu_pro=69.6 (alt measurement, mc=false)" } ] }, { "model_id": "llama-3.1-405b-instruct", "benchmark_id": "mmlu_pro", "score": 73.3, "reference_url": "https://arxiv.org/abs/2412.19437", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "DeepSeek V3 paper Table 6: llama-3.1-405b-instruct=73.3.", "candidates": [ { "score": 73.0, "reference_url": "https://cohere.com/research/papers/command-a-technical-report.pdf", "source_type": "tech_report", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "Cohere internal eval", "prompt_style": "default", "temperature": "default" }, "notes": "Command A paper: alt measurement 73.0 (mc=false)." } ] }, { "model_id": "claude-3.5-sonnet", "benchmark_id": "mmlu_pro", "score": 78.0, "reference_url": "https://arxiv.org/abs/2412.19437", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "DeepSeek V3 paper Table 6: claude-3.5-sonnet=78.0.", "candidates": [] }, { "model_id": "gpt-4o-0513", "benchmark_id": "mmlu_pro", "score": 72.6, "reference_url": "https://arxiv.org/abs/2412.19437", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "DeepSeek V3 paper Table 6: gpt-4o-0513=72.6.", "candidates": [ { "score": 73.0, "reference_url": "https://arxiv.org/abs/2412.08905", "source_type": "tech_report", "reported_setting": { "mode": "unknown", "effort": "unknown", "tools": "unknown", "sampling": "unknown", "judge": "rule-based", "harness": "Microsoft Phi internal eval", "prompt_style": "unknown", "temperature": "unknown" }, "notes": "Phi-4 paper Table 1 (gpt-4o-0513 column): mmlu_pro=73.0 (alt measurement, mc=false)" }, { "score": 73.5, "reference_url": "https://arxiv.org/abs/2504.21318", "source_type": "tech_report", "reported_setting": { "mode": "unknown", "effort": "unknown", "tools": "unknown", "sampling": "unknown", "judge": "rule-based", "harness": "Microsoft Phi internal eval", "prompt_style": "unknown", "temperature": "unknown" }, "notes": "Phi-4-reasoning paper Table 2 (gpt-4o-0513 column): mmlu_pro=73.5 (alt measurement, mc=false)" }, { "score": 77.9, "reference_url": "https://cohere.com/research/papers/command-a-technical-report.pdf", "source_type": "tech_report", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "Cohere internal eval", "prompt_style": "default", "temperature": "default" }, "notes": "Command A paper: alt measurement 77.9 (mc=false)." } ] }, { "model_id": "deepseek-v3", "benchmark_id": "mmlu_pro", "score": 75.9, "reference_url": "https://arxiv.org/abs/2412.19437", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "DeepSeek V3 paper Table 6: deepseek-v3=75.9.", "candidates": [] }, { "model_id": "deepseek-v2-0506", "benchmark_id": "drop", "score": 83.0, "reference_url": "https://arxiv.org/abs/2412.19437", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "DeepSeek V3 paper Table 6: deepseek-v2-0506=83.0.", "candidates": [] }, { "model_id": "deepseek-v2.5-0905", "benchmark_id": "drop", "score": 87.8, "reference_url": "https://arxiv.org/abs/2412.19437", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "DeepSeek V3 paper Table 6: deepseek-v2.5-0905=87.8.", "candidates": [] }, { "model_id": "qwen2.5-72b-instruct", "benchmark_id": "drop", "score": 76.7, "reference_url": "https://arxiv.org/abs/2412.19437", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "DeepSeek V3 paper Table 6: qwen2.5-72b-instruct=76.7.", "candidates": [ { "score": 34.2, "reference_url": "https://arxiv.org/abs/2501.00656", "source_type": "tech_report", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "OLMES", "prompt_style": "default", "temperature": "default" }, "notes": "OLMo 2 paper Table 7 Instruct: alt measurement 34.2 (mc=false)." } ] }, { "model_id": "llama-3.1-405b-instruct", "benchmark_id": "drop", "score": 88.7, "reference_url": "https://arxiv.org/abs/2412.19437", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "DeepSeek V3 paper Table 6: llama-3.1-405b-instruct=88.7.", "candidates": [] }, { "model_id": "claude-3.5-sonnet", "benchmark_id": "drop", "score": 88.3, "reference_url": "https://arxiv.org/abs/2412.19437", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "DeepSeek V3 paper Table 6: claude-3.5-sonnet=88.3.", "candidates": [] }, { "model_id": "gpt-4o-0513", "benchmark_id": "drop", "score": 83.7, "reference_url": "https://arxiv.org/abs/2412.19437", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "DeepSeek V3 paper Table 6: gpt-4o-0513=83.7.", "candidates": [ { "score": 80.9, "reference_url": "https://arxiv.org/abs/2412.08905", "source_type": "tech_report", "reported_setting": { "mode": "unknown", "effort": "unknown", "tools": "unknown", "sampling": "unknown", "judge": "rule-based", "harness": "Microsoft Phi internal eval", "prompt_style": "unknown", "temperature": "unknown" }, "notes": "Phi-4 paper Table 1 (gpt-4o-0513 column): drop=80.9 (alt measurement, mc=false)" } ] }, { "model_id": "deepseek-v3", "benchmark_id": "drop", "score": 91.6, "reference_url": "https://arxiv.org/abs/2412.19437", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "DeepSeek V3 paper Table 6: deepseek-v3=91.6.", "candidates": [] }, { "model_id": "deepseek-v2-0506", "benchmark_id": "ifeval", "score": 57.7, "reference_url": "https://arxiv.org/abs/2412.19437", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "DeepSeek V3 paper Table 6: deepseek-v2-0506=57.7.", "candidates": [] }, { "model_id": "deepseek-v2.5-0905", "benchmark_id": "ifeval", "score": 80.6, "reference_url": "https://arxiv.org/abs/2412.19437", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "DeepSeek V3 paper Table 6: deepseek-v2.5-0905=80.6.", "candidates": [] }, { "model_id": "qwen2.5-72b-instruct", "benchmark_id": "ifeval", "score": 84.1, "reference_url": "https://arxiv.org/abs/2412.19437", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "DeepSeek V3 paper Table 6: qwen2.5-72b-instruct=84.1.", "candidates": [ { "score": 85.0, "reference_url": "https://arxiv.org/abs/2412.08905", "source_type": "tech_report", "reported_setting": { "mode": "unknown", "effort": "unknown", "tools": "unknown", "sampling": "unknown", "judge": "rule-based", "harness": "Microsoft Phi internal eval", "prompt_style": "unknown", "temperature": "unknown" }, "notes": "Phi-4 paper Table 1 (qwen2.5-72b-instruct column): ifeval=85.0 (alt measurement, mc=false)" }, { "score": 87.6, "reference_url": "https://arxiv.org/abs/2501.00656", "source_type": "tech_report", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "OLMES", "prompt_style": "default", "temperature": "default" }, "notes": "OLMo 2 paper Table 7 Instruct: alt measurement 87.6 (mc=false)." } ] }, { "model_id": "llama-3.1-405b-instruct", "benchmark_id": "ifeval", "score": 86.0, "reference_url": "https://arxiv.org/abs/2412.19437", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "DeepSeek V3 paper Table 6: llama-3.1-405b-instruct=86.0.", "candidates": [ { "score": 88.6, "reference_url": "https://cohere.com/research/papers/command-a-technical-report.pdf", "source_type": "tech_report", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "Cohere internal eval", "prompt_style": "default", "temperature": "default" }, "notes": "Command A paper: alt measurement 88.6 (mc=false)." } ] }, { "model_id": "claude-3.5-sonnet", "benchmark_id": "ifeval", "score": 86.5, "reference_url": "https://arxiv.org/abs/2412.19437", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "DeepSeek V3 paper Table 6: claude-3.5-sonnet=86.5.", "candidates": [ { "score": 90.2, "reference_url": "https://cohere.com/research/papers/command-a-technical-report.pdf", "source_type": "tech_report", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "Cohere internal eval", "prompt_style": "default", "temperature": "default" }, "notes": "Command A paper: alt measurement 90.2 (mc=false)." } ] }, { "model_id": "gpt-4o-0513", "benchmark_id": "ifeval", "score": 84.3, "reference_url": "https://arxiv.org/abs/2412.19437", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "DeepSeek V3 paper Table 6: gpt-4o-0513=84.3.", "candidates": [ { "score": 84.8, "reference_url": "https://arxiv.org/abs/2412.08905", "source_type": "tech_report", "reported_setting": { "mode": "unknown", "effort": "unknown", "tools": "unknown", "sampling": "unknown", "judge": "rule-based", "harness": "Microsoft Phi internal eval", "prompt_style": "unknown", "temperature": "unknown" }, "notes": "Phi-4 paper Table 1 (gpt-4o-0513 column): ifeval=84.8 (alt measurement, mc=false)" }, { "score": 81.8, "reference_url": "https://arxiv.org/abs/2504.21318", "source_type": "tech_report", "reported_setting": { "mode": "unknown", "effort": "unknown", "tools": "unknown", "sampling": "unknown", "judge": "rule-based", "harness": "Microsoft Phi internal eval", "prompt_style": "unknown", "temperature": "unknown" }, "notes": "Phi-4-reasoning paper Table 2 (gpt-4o-0513 column): ifeval=81.8 (alt measurement, mc=false)" }, { "score": 86.1, "reference_url": "https://cohere.com/research/papers/command-a-technical-report.pdf", "source_type": "tech_report", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "Cohere internal eval", "prompt_style": "default", "temperature": "default" }, "notes": "Command A paper: alt measurement 86.1 (mc=false)." }, { "score": 83.8, "reference_url": "https://cohere.com/research/papers/command-a-technical-report.pdf", "source_type": "tech_report", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "Cohere internal eval", "prompt_style": "default", "temperature": "default" }, "notes": "Command A paper: alt measurement 83.8 (mc=false)." } ] }, { "model_id": "deepseek-v2-0506", "benchmark_id": "gpqa_diamond", "score": 35.3, "reference_url": "https://arxiv.org/abs/2412.19437", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "DeepSeek V3 paper Table 6: deepseek-v2-0506=35.3.", "candidates": [] }, { "model_id": "deepseek-v2.5-0905", "benchmark_id": "gpqa_diamond", "score": 41.3, "reference_url": "https://arxiv.org/abs/2412.19437", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "DeepSeek V3 paper Table 6: deepseek-v2.5-0905=41.3.", "candidates": [] }, { "model_id": "qwen2.5-72b-instruct", "benchmark_id": "gpqa_diamond", "score": 49.0, "reference_url": "https://arxiv.org/abs/2412.19437", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "DeepSeek V3 paper Table 6: qwen2.5-72b-instruct=49.0.", "candidates": [] }, { "model_id": "llama-3.1-405b-instruct", "benchmark_id": "gpqa_diamond", "score": 51.1, "reference_url": "https://arxiv.org/abs/2412.19437", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "DeepSeek V3 paper Table 6: llama-3.1-405b-instruct=51.1.", "candidates": [] }, { "model_id": "claude-3.5-sonnet", "benchmark_id": "gpqa_diamond", "score": 65.0, "reference_url": "https://arxiv.org/abs/2412.19437", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "DeepSeek V3 paper Table 6: claude-3.5-sonnet=65.0.", "candidates": [] }, { "model_id": "gpt-4o-0513", "benchmark_id": "gpqa_diamond", "score": 49.9, "reference_url": "https://arxiv.org/abs/2412.19437", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "DeepSeek V3 paper Table 6: gpt-4o-0513=49.9.", "candidates": [ { "score": 50.6, "reference_url": "https://arxiv.org/abs/2412.08905", "source_type": "tech_report", "reported_setting": { "mode": "unknown", "effort": "unknown", "tools": "unknown", "sampling": "unknown", "judge": "rule-based", "harness": "Microsoft Phi internal eval", "prompt_style": "unknown", "temperature": "unknown" }, "notes": "Phi-4 paper Table 1 (gpt-4o-0513 column): gpqa_diamond=50.6 (alt measurement, mc=false)" } ] }, { "model_id": "deepseek-v2-0506", "benchmark_id": "simpleqa", "score": 9.0, "reference_url": "https://arxiv.org/abs/2412.19437", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "DeepSeek V3 paper Table 6: deepseek-v2-0506=9.0.", "candidates": [] }, { "model_id": "deepseek-v2.5-0905", "benchmark_id": "simpleqa", "score": 10.2, "reference_url": "https://arxiv.org/abs/2412.19437", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "DeepSeek V3 paper Table 6: deepseek-v2.5-0905=10.2.", "candidates": [] }, { "model_id": "qwen2.5-72b-instruct", "benchmark_id": "simpleqa", "score": 9.1, "reference_url": "https://arxiv.org/abs/2412.19437", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "DeepSeek V3 paper Table 6: qwen2.5-72b-instruct=9.1.", "candidates": [ { "score": 10.2, "reference_url": "https://arxiv.org/abs/2412.08905", "source_type": "tech_report", "reported_setting": { "mode": "unknown", "effort": "unknown", "tools": "unknown", "sampling": "unknown", "judge": "rule-based", "harness": "Microsoft Phi internal eval", "prompt_style": "unknown", "temperature": "unknown" }, "notes": "Phi-4 paper Table 1 (qwen2.5-72b-instruct column): simpleqa=10.2 (alt measurement, mc=false)" } ] }, { "model_id": "llama-3.1-405b-instruct", "benchmark_id": "simpleqa", "score": 17.1, "reference_url": "https://arxiv.org/abs/2412.19437", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "DeepSeek V3 paper Table 6: llama-3.1-405b-instruct=17.1.", "candidates": [] }, { "model_id": "claude-3.5-sonnet", "benchmark_id": "simpleqa", "score": 28.4, "reference_url": "https://arxiv.org/abs/2412.19437", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "DeepSeek V3 paper Table 6: claude-3.5-sonnet=28.4.", "candidates": [] }, { "model_id": "gpt-4o-0513", "benchmark_id": "simpleqa", "score": 38.2, "reference_url": "https://arxiv.org/abs/2412.19437", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "DeepSeek V3 paper Table 6: gpt-4o-0513=38.2.", "candidates": [ { "score": 39.4, "reference_url": "https://arxiv.org/abs/2412.08905", "source_type": "tech_report", "reported_setting": { "mode": "unknown", "effort": "unknown", "tools": "unknown", "sampling": "unknown", "judge": "rule-based", "harness": "Microsoft Phi internal eval", "prompt_style": "unknown", "temperature": "unknown" }, "notes": "Phi-4 paper Table 1 (gpt-4o-0513 column): simpleqa=39.4 (alt measurement, mc=false)" } ] }, { "model_id": "deepseek-v3", "benchmark_id": "simpleqa", "score": 24.9, "reference_url": "https://arxiv.org/abs/2412.19437", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "DeepSeek V3 paper Table 6: deepseek-v3=24.9.", "candidates": [] }, { "model_id": "deepseek-v2-0506", "benchmark_id": "frames", "score": 66.9, "reference_url": "https://arxiv.org/abs/2412.19437", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "DeepSeek V3 paper Table 6: deepseek-v2-0506=66.9.", "candidates": [] }, { "model_id": "deepseek-v2.5-0905", "benchmark_id": "frames", "score": 65.4, "reference_url": "https://arxiv.org/abs/2412.19437", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "DeepSeek V3 paper Table 6: deepseek-v2.5-0905=65.4.", "candidates": [] }, { "model_id": "qwen2.5-72b-instruct", "benchmark_id": "frames", "score": 69.8, "reference_url": "https://arxiv.org/abs/2412.19437", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "DeepSeek V3 paper Table 6: qwen2.5-72b-instruct=69.8.", "candidates": [] }, { "model_id": "llama-3.1-405b-instruct", "benchmark_id": "frames", "score": 70.0, "reference_url": "https://arxiv.org/abs/2412.19437", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "DeepSeek V3 paper Table 6: llama-3.1-405b-instruct=70.0.", "candidates": [] }, { "model_id": "claude-3.5-sonnet", "benchmark_id": "frames", "score": 72.5, "reference_url": "https://arxiv.org/abs/2412.19437", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "DeepSeek V3 paper Table 6: claude-3.5-sonnet=72.5.", "candidates": [] }, { "model_id": "gpt-4o-0513", "benchmark_id": "frames", "score": 80.5, "reference_url": "https://arxiv.org/abs/2412.19437", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "DeepSeek V3 paper Table 6: gpt-4o-0513=80.5.", "candidates": [] }, { "model_id": "deepseek-v3", "benchmark_id": "frames", "score": 73.3, "reference_url": "https://arxiv.org/abs/2412.19437", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "DeepSeek V3 paper Table 6: deepseek-v3=73.3.", "candidates": [] }, { "model_id": "deepseek-v2-0506", "benchmark_id": "longbench_v2", "score": 31.6, "reference_url": "https://arxiv.org/abs/2412.19437", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "DeepSeek V3 paper Table 6: deepseek-v2-0506=31.6.", "candidates": [] }, { "model_id": "deepseek-v2.5-0905", "benchmark_id": "longbench_v2", "score": 35.4, "reference_url": "https://arxiv.org/abs/2412.19437", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "DeepSeek V3 paper Table 6: deepseek-v2.5-0905=35.4.", "candidates": [] }, { "model_id": "qwen2.5-72b-instruct", "benchmark_id": "longbench_v2", "score": 39.4, "reference_url": "https://arxiv.org/abs/2412.19437", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "DeepSeek V3 paper Table 6: qwen2.5-72b-instruct=39.4.", "candidates": [] }, { "model_id": "llama-3.1-405b-instruct", "benchmark_id": "longbench_v2", "score": 36.1, "reference_url": "https://arxiv.org/abs/2412.19437", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "DeepSeek V3 paper Table 6: llama-3.1-405b-instruct=36.1.", "candidates": [] }, { "model_id": "claude-3.5-sonnet", "benchmark_id": "longbench_v2", "score": 41.0, "reference_url": "https://arxiv.org/abs/2412.19437", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "DeepSeek V3 paper Table 6: claude-3.5-sonnet=41.0.", "candidates": [] }, { "model_id": "gpt-4o-0513", "benchmark_id": "longbench_v2", "score": 48.1, "reference_url": "https://arxiv.org/abs/2412.19437", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "DeepSeek V3 paper Table 6: gpt-4o-0513=48.1.", "candidates": [] }, { "model_id": "deepseek-v3", "benchmark_id": "longbench_v2", "score": 48.7, "reference_url": "https://arxiv.org/abs/2412.19437", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "DeepSeek V3 paper Table 6: deepseek-v3=48.7.", "candidates": [] }, { "model_id": "deepseek-v2-0506", "benchmark_id": "humaneval", "score": 69.3, "reference_url": "https://arxiv.org/abs/2412.19437", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "DeepSeek V3 paper Table 6: deepseek-v2-0506=69.3.", "candidates": [] }, { "model_id": "deepseek-v2.5-0905", "benchmark_id": "humaneval", "score": 77.4, "reference_url": "https://arxiv.org/abs/2412.19437", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "DeepSeek V3 paper Table 6: deepseek-v2.5-0905=77.4.", "candidates": [ { "score": 89.0, "reference_url": "https://llm-stats.com/benchmarks/humaneval", "source_type": "third_party_aggregator", "reported_setting": { "snapshot": "0905" }, "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=DeepSeek-V2.5, slug=deepseek-v2.5, provider=DeepSeek" } ] }, { "model_id": "qwen2.5-72b-instruct", "benchmark_id": "humaneval", "score": 77.3, "reference_url": "https://arxiv.org/abs/2412.19437", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "DeepSeek V3 paper Table 6: qwen2.5-72b-instruct=77.3.", "candidates": [ { "score": 80.4, "reference_url": "https://arxiv.org/abs/2412.08905", "source_type": "tech_report", "reported_setting": { "mode": "unknown", "effort": "unknown", "tools": "unknown", "sampling": "unknown", "judge": "rule-based", "harness": "Microsoft Phi internal eval", "prompt_style": "unknown", "temperature": "unknown" }, "notes": "Phi-4 paper Table 1 (qwen2.5-72b-instruct column): humaneval=80.4 (alt measurement, mc=false)" }, { "score": 78.5, "reference_url": "https://cohere.com/research/papers/command-a-technical-report.pdf", "source_type": "tech_report", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "Cohere internal eval", "prompt_style": "default", "temperature": "default" }, "notes": "Command A paper: alt measurement 78.5 (mc=false)." } ] }, { "model_id": "llama-3.1-405b-instruct", "benchmark_id": "humaneval", "score": 77.2, "reference_url": "https://arxiv.org/abs/2412.19437", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "DeepSeek V3 paper Table 6: llama-3.1-405b-instruct=77.2.", "candidates": [ { "score": 76.7, "reference_url": "https://cohere.com/research/papers/command-a-technical-report.pdf", "source_type": "tech_report", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "Cohere internal eval", "prompt_style": "default", "temperature": "default" }, "notes": "Command A paper: alt measurement 76.7 (mc=false)." } ] }, { "model_id": "claude-3.5-sonnet", "benchmark_id": "humaneval", "score": 81.7, "reference_url": "https://arxiv.org/abs/2412.19437", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "DeepSeek V3 paper Table 6: claude-3.5-sonnet=81.7.", "candidates": [] }, { "model_id": "gpt-4o-0513", "benchmark_id": "humaneval", "score": 80.5, "reference_url": "https://arxiv.org/abs/2412.19437", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "DeepSeek V3 paper Table 6: gpt-4o-0513=80.5.", "candidates": [ { "score": 90.6, "reference_url": "https://arxiv.org/abs/2412.08905", "source_type": "tech_report", "reported_setting": { "mode": "unknown", "effort": "unknown", "tools": "unknown", "sampling": "unknown", "judge": "rule-based", "harness": "Microsoft Phi internal eval", "prompt_style": "unknown", "temperature": "unknown" }, "notes": "Phi-4 paper Table 1 (gpt-4o-0513 column): humaneval=90.6 (alt measurement, mc=false)" } ] }, { "model_id": "deepseek-v2-0506", "benchmark_id": "livecodebench", "score": 18.8, "reference_url": "https://arxiv.org/abs/2412.19437", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "DeepSeek V3 paper Table 6: deepseek-v2-0506=18.8.", "candidates": [] }, { "model_id": "deepseek-v2.5-0905", "benchmark_id": "livecodebench", "score": 29.2, "reference_url": "https://arxiv.org/abs/2412.19437", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "DeepSeek V3 paper Table 6: deepseek-v2.5-0905=29.2.", "candidates": [] }, { "model_id": "qwen2.5-72b-instruct", "benchmark_id": "livecodebench", "score": 31.1, "reference_url": "https://arxiv.org/abs/2412.19437", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "DeepSeek V3 paper Table 6: qwen2.5-72b-instruct=31.1.", "candidates": [ { "score": 26.3, "reference_url": "https://cohere.com/research/papers/command-a-technical-report.pdf", "source_type": "tech_report", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "Cohere internal eval", "prompt_style": "default", "temperature": "default" }, "notes": "Command A paper: alt measurement 26.3 (mc=false)." } ] }, { "model_id": "llama-3.1-405b-instruct", "benchmark_id": "livecodebench", "score": 28.4, "reference_url": "https://arxiv.org/abs/2412.19437", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "DeepSeek V3 paper Table 6: llama-3.1-405b-instruct=28.4.", "candidates": [ { "score": 29.3, "reference_url": "https://cohere.com/research/papers/command-a-technical-report.pdf", "source_type": "tech_report", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "Cohere internal eval", "prompt_style": "default", "temperature": "default" }, "notes": "Command A paper: alt measurement 29.3 (mc=false)." } ] }, { "model_id": "claude-3.5-sonnet", "benchmark_id": "livecodebench", "score": 36.3, "reference_url": "https://arxiv.org/abs/2412.19437", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "DeepSeek V3 paper Table 6: claude-3.5-sonnet=36.3.", "candidates": [ { "score": 38.9, "reference_url": "https://arxiv.org/abs/2501.12948", "source_type": "tech_report", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "notes": "DeepSeek R1 paper Table 15: claude-3.5-sonnet=38.9." } ] }, { "model_id": "gpt-4o-0513", "benchmark_id": "livecodebench", "score": 33.4, "reference_url": "https://arxiv.org/abs/2412.19437", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "DeepSeek V3 paper Table 6: gpt-4o-0513=33.4.", "candidates": [ { "score": 32.9, "reference_url": "https://arxiv.org/abs/2501.12948", "source_type": "tech_report", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "notes": "DeepSeek R1 paper Table 15: gpt-4o-0513=32.9." } ] }, { "model_id": "deepseek-v3", "benchmark_id": "livecodebench", "score": 40.5, "reference_url": "https://arxiv.org/abs/2412.19437", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "DeepSeek V3 paper Table 6: deepseek-v3=40.5.", "candidates": [ { "score": 33.5, "reference_url": "https://cohere.com/research/papers/command-a-technical-report.pdf", "source_type": "tech_report", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "Cohere internal eval", "prompt_style": "default", "temperature": "default" }, "notes": "Command A paper: alt measurement 33.5 (mc=false)." } ] }, { "model_id": "deepseek-v2.5-0905", "benchmark_id": "swe_bench_verified", "score": 22.6, "reference_url": "https://arxiv.org/abs/2412.19437", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "agentic (bash + edit)", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "DeepSeek V3 paper Table 6: deepseek-v2.5-0905=22.6.", "candidates": [ { "score": 16.8, "reference_url": "https://llm-stats.com/benchmarks/swe-bench-verified", "source_type": "third_party_aggregator", "reported_setting": { "snapshot": "0905" }, "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=DeepSeek-V2.5, slug=deepseek-v2.5, provider=DeepSeek" } ] }, { "model_id": "qwen2.5-72b-instruct", "benchmark_id": "swe_bench_verified", "score": 23.8, "reference_url": "https://arxiv.org/abs/2412.19437", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "agentic (bash + edit)", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "DeepSeek V3 paper Table 6: qwen2.5-72b-instruct=23.8.", "candidates": [ { "score": 33.0, "reference_url": "https://cohere.com/research/papers/command-a-technical-report.pdf", "source_type": "tech_report", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "Cohere internal eval", "prompt_style": "default", "temperature": "default" }, "notes": "Command A paper: alt measurement 33.0 (mc=false)." } ] }, { "model_id": "llama-3.1-405b-instruct", "benchmark_id": "swe_bench_verified", "score": 24.5, "reference_url": "https://arxiv.org/abs/2412.19437", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "agentic (bash + edit)", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "DeepSeek V3 paper Table 6: llama-3.1-405b-instruct=24.5.", "candidates": [ { "score": 33.4, "reference_url": "https://cohere.com/research/papers/command-a-technical-report.pdf", "source_type": "tech_report", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "Cohere internal eval", "prompt_style": "default", "temperature": "default" }, "notes": "Command A paper: alt measurement 33.4 (mc=false)." } ] }, { "model_id": "claude-3.5-sonnet", "benchmark_id": "swe_bench_verified", "score": 50.8, "reference_url": "https://arxiv.org/abs/2412.19437", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "agentic (bash + edit)", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "DeepSeek V3 paper Table 6: claude-3.5-sonnet=50.8.", "candidates": [ { "score": 49.0, "reference_url": "https://www.anthropic.com/news/claude-3-7-sonnet", "source_type": "official_blog", "reported_setting": { "mode": "non-thinking", "effort": "default", "tools": "agentic (bash+editor)", "sampling": "pass@1", "judge": "rule-based", "harness": "simple bash+editor scaffold", "prompt_style": "default + planning tool", "temperature": "default" }, "notes": "Anthropic 3.7 Sonnet blog cross-model: Claude 3.5 Sonnet (new) SWE-bench Verified=49.0%; differs from arxiv 50.8%" } ] }, { "model_id": "gpt-4o-0513", "benchmark_id": "swe_bench_verified", "score": 38.8, "reference_url": "https://arxiv.org/abs/2412.19437", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "agentic (bash + edit)", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "DeepSeek V3 paper Table 6: gpt-4o-0513=38.8.", "candidates": [] }, { "model_id": "deepseek-v2.5-0905", "benchmark_id": "aider_polyglot_diff", "score": 18.2, "reference_url": "https://arxiv.org/abs/2412.19437", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "agentic (bash + edit)", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "DeepSeek V3 paper Table 6: deepseek-v2.5-0905=18.2.", "candidates": [] }, { "model_id": "qwen2.5-72b-instruct", "benchmark_id": "aider_polyglot_diff", "score": 7.6, "reference_url": "https://arxiv.org/abs/2412.19437", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "agentic (bash + edit)", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "DeepSeek V3 paper Table 6: qwen2.5-72b-instruct=7.6.", "candidates": [] }, { "model_id": "llama-3.1-405b-instruct", "benchmark_id": "aider_polyglot_diff", "score": 5.8, "reference_url": "https://arxiv.org/abs/2412.19437", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "agentic (bash + edit)", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "DeepSeek V3 paper Table 6: llama-3.1-405b-instruct=5.8.", "candidates": [] }, { "model_id": "claude-3.5-sonnet", "benchmark_id": "aider_polyglot_diff", "score": 45.3, "reference_url": "https://arxiv.org/abs/2412.19437", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "agentic (bash + edit)", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "DeepSeek V3 paper Table 6: claude-3.5-sonnet=45.3.", "candidates": [] }, { "model_id": "gpt-4o-0513", "benchmark_id": "aider_polyglot_diff", "score": 16.0, "reference_url": "https://arxiv.org/abs/2412.19437", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "agentic (bash + edit)", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "DeepSeek V3 paper Table 6: gpt-4o-0513=16.0.", "candidates": [] }, { "model_id": "deepseek-v3", "benchmark_id": "aider_polyglot_diff", "score": 49.6, "reference_url": "https://arxiv.org/abs/2412.19437", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "agentic (bash + edit)", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "DeepSeek V3 paper Table 6: deepseek-v3=49.6.", "candidates": [] }, { "model_id": "deepseek-v2-0506", "benchmark_id": "aime_2024", "score": 4.6, "reference_url": "https://arxiv.org/abs/2412.19437", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "DeepSeek V3 paper Table 6: deepseek-v2-0506=4.6.", "candidates": [] }, { "model_id": "deepseek-v2.5-0905", "benchmark_id": "aime_2024", "score": 16.7, "reference_url": "https://arxiv.org/abs/2412.19437", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "DeepSeek V3 paper Table 6: deepseek-v2.5-0905=16.7.", "candidates": [] }, { "model_id": "qwen2.5-72b-instruct", "benchmark_id": "aime_2024", "score": 23.3, "reference_url": "https://arxiv.org/abs/2412.19437", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "DeepSeek V3 paper Table 6: qwen2.5-72b-instruct=23.3.", "candidates": [] }, { "model_id": "llama-3.1-405b-instruct", "benchmark_id": "aime_2024", "score": 23.3, "reference_url": "https://arxiv.org/abs/2412.19437", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "DeepSeek V3 paper Table 6: llama-3.1-405b-instruct=23.3.", "candidates": [] }, { "model_id": "claude-3.5-sonnet", "benchmark_id": "aime_2024", "score": 16.0, "reference_url": "https://arxiv.org/abs/2412.19437", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "DeepSeek V3 paper Table 6: claude-3.5-sonnet=16.0.", "candidates": [] }, { "model_id": "gpt-4o-0513", "benchmark_id": "aime_2024", "score": 9.3, "reference_url": "https://arxiv.org/abs/2412.19437", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "DeepSeek V3 paper Table 6: gpt-4o-0513=9.3.", "candidates": [] }, { "model_id": "deepseek-v3", "benchmark_id": "aime_2024", "score": 39.2, "reference_url": "https://arxiv.org/abs/2412.19437", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "DeepSeek V3 paper Table 6: deepseek-v3=39.2.", "candidates": [] }, { "model_id": "deepseek-v2-0506", "benchmark_id": "math_500", "score": 56.3, "reference_url": "https://arxiv.org/abs/2412.19437", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "DeepSeek V3 paper Table 6: deepseek-v2-0506=56.3.", "candidates": [] }, { "model_id": "deepseek-v2.5-0905", "benchmark_id": "math_500", "score": 74.7, "reference_url": "https://arxiv.org/abs/2412.19437", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "DeepSeek V3 paper Table 6: deepseek-v2.5-0905=74.7.", "candidates": [] }, { "model_id": "qwen2.5-72b-instruct", "benchmark_id": "math_500", "score": 80.0, "reference_url": "https://arxiv.org/abs/2412.19437", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "DeepSeek V3 paper Table 6: qwen2.5-72b-instruct=80.0.", "candidates": [] }, { "model_id": "llama-3.1-405b-instruct", "benchmark_id": "math_500", "score": 73.8, "reference_url": "https://arxiv.org/abs/2412.19437", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "DeepSeek V3 paper Table 6: llama-3.1-405b-instruct=73.8.", "candidates": [] }, { "model_id": "claude-3.5-sonnet", "benchmark_id": "math_500", "score": 78.3, "reference_url": "https://arxiv.org/abs/2412.19437", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "DeepSeek V3 paper Table 6: claude-3.5-sonnet=78.3.", "candidates": [] }, { "model_id": "gpt-4o-0513", "benchmark_id": "math_500", "score": 74.6, "reference_url": "https://arxiv.org/abs/2412.19437", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "DeepSeek V3 paper Table 6: gpt-4o-0513=74.6.", "candidates": [] }, { "model_id": "deepseek-v2-0506", "benchmark_id": "cnmo_2024", "score": 2.8, "reference_url": "https://arxiv.org/abs/2412.19437", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "DeepSeek V3 paper Table 6: deepseek-v2-0506=2.8.", "candidates": [] }, { "model_id": "deepseek-v2.5-0905", "benchmark_id": "cnmo_2024", "score": 10.8, "reference_url": "https://arxiv.org/abs/2412.19437", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "DeepSeek V3 paper Table 6: deepseek-v2.5-0905=10.8.", "candidates": [] }, { "model_id": "qwen2.5-72b-instruct", "benchmark_id": "cnmo_2024", "score": 15.9, "reference_url": "https://arxiv.org/abs/2412.19437", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "DeepSeek V3 paper Table 6: qwen2.5-72b-instruct=15.9.", "candidates": [] }, { "model_id": "llama-3.1-405b-instruct", "benchmark_id": "cnmo_2024", "score": 6.8, "reference_url": "https://arxiv.org/abs/2412.19437", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "DeepSeek V3 paper Table 6: llama-3.1-405b-instruct=6.8.", "candidates": [] }, { "model_id": "claude-3.5-sonnet", "benchmark_id": "cnmo_2024", "score": 13.1, "reference_url": "https://arxiv.org/abs/2412.19437", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "DeepSeek V3 paper Table 6: claude-3.5-sonnet=13.1.", "candidates": [] }, { "model_id": "gpt-4o-0513", "benchmark_id": "cnmo_2024", "score": 10.8, "reference_url": "https://arxiv.org/abs/2412.19437", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "DeepSeek V3 paper Table 6: gpt-4o-0513=10.8.", "candidates": [] }, { "model_id": "deepseek-v3", "benchmark_id": "cnmo_2024", "score": 43.2, "reference_url": "https://arxiv.org/abs/2412.19437", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "DeepSeek V3 paper Table 6: deepseek-v3=43.2.", "candidates": [] }, { "model_id": "deepseek-v2-0506", "benchmark_id": "cluewsc", "score": 89.9, "reference_url": "https://arxiv.org/abs/2412.19437", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "DeepSeek V3 paper Table 6: deepseek-v2-0506=89.9.", "candidates": [] }, { "model_id": "deepseek-v2.5-0905", "benchmark_id": "cluewsc", "score": 90.4, "reference_url": "https://arxiv.org/abs/2412.19437", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "DeepSeek V3 paper Table 6: deepseek-v2.5-0905=90.4.", "candidates": [] }, { "model_id": "qwen2.5-72b-instruct", "benchmark_id": "cluewsc", "score": 91.4, "reference_url": "https://arxiv.org/abs/2412.19437", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "DeepSeek V3 paper Table 6: qwen2.5-72b-instruct=91.4.", "candidates": [] }, { "model_id": "llama-3.1-405b-instruct", "benchmark_id": "cluewsc", "score": 84.7, "reference_url": "https://arxiv.org/abs/2412.19437", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "DeepSeek V3 paper Table 6: llama-3.1-405b-instruct=84.7.", "candidates": [] }, { "model_id": "claude-3.5-sonnet", "benchmark_id": "cluewsc", "score": 85.4, "reference_url": "https://arxiv.org/abs/2412.19437", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "DeepSeek V3 paper Table 6: claude-3.5-sonnet=85.4.", "candidates": [] }, { "model_id": "gpt-4o-0513", "benchmark_id": "cluewsc", "score": 87.9, "reference_url": "https://arxiv.org/abs/2412.19437", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "DeepSeek V3 paper Table 6: gpt-4o-0513=87.9.", "candidates": [] }, { "model_id": "deepseek-v3", "benchmark_id": "cluewsc", "score": 90.9, "reference_url": "https://arxiv.org/abs/2412.19437", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "DeepSeek V3 paper Table 6: deepseek-v3=90.9.", "candidates": [] }, { "model_id": "deepseek-v2-0506", "benchmark_id": "c_eval", "score": 78.6, "reference_url": "https://arxiv.org/abs/2412.19437", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "DeepSeek V3 paper Table 6: deepseek-v2-0506=78.6.", "candidates": [] }, { "model_id": "deepseek-v2.5-0905", "benchmark_id": "c_eval", "score": 79.5, "reference_url": "https://arxiv.org/abs/2412.19437", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "DeepSeek V3 paper Table 6: deepseek-v2.5-0905=79.5.", "candidates": [] }, { "model_id": "qwen2.5-72b-instruct", "benchmark_id": "c_eval", "score": 86.1, "reference_url": "https://arxiv.org/abs/2412.19437", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "DeepSeek V3 paper Table 6: qwen2.5-72b-instruct=86.1.", "candidates": [] }, { "model_id": "llama-3.1-405b-instruct", "benchmark_id": "c_eval", "score": 61.5, "reference_url": "https://arxiv.org/abs/2412.19437", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "DeepSeek V3 paper Table 6: llama-3.1-405b-instruct=61.5.", "candidates": [] }, { "model_id": "claude-3.5-sonnet", "benchmark_id": "c_eval", "score": 76.7, "reference_url": "https://arxiv.org/abs/2412.19437", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "DeepSeek V3 paper Table 6: claude-3.5-sonnet=76.7.", "candidates": [] }, { "model_id": "gpt-4o-0513", "benchmark_id": "c_eval", "score": 76.0, "reference_url": "https://arxiv.org/abs/2412.19437", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "DeepSeek V3 paper Table 6: gpt-4o-0513=76.0.", "candidates": [] }, { "model_id": "deepseek-v3", "benchmark_id": "c_eval", "score": 86.5, "reference_url": "https://arxiv.org/abs/2412.19437", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "DeepSeek V3 paper Table 6: deepseek-v3=86.5.", "candidates": [] }, { "model_id": "deepseek-v2-0506", "benchmark_id": "chinese_simpleqa", "score": 48.5, "reference_url": "https://arxiv.org/abs/2412.19437", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "DeepSeek V3 paper Table 6: deepseek-v2-0506=48.5.", "candidates": [] }, { "model_id": "deepseek-v2.5-0905", "benchmark_id": "chinese_simpleqa", "score": 54.1, "reference_url": "https://arxiv.org/abs/2412.19437", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "DeepSeek V3 paper Table 6: deepseek-v2.5-0905=54.1.", "candidates": [] }, { "model_id": "qwen2.5-72b-instruct", "benchmark_id": "chinese_simpleqa", "score": 48.4, "reference_url": "https://arxiv.org/abs/2412.19437", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "DeepSeek V3 paper Table 6: qwen2.5-72b-instruct=48.4.", "candidates": [] }, { "model_id": "llama-3.1-405b-instruct", "benchmark_id": "chinese_simpleqa", "score": 50.4, "reference_url": "https://arxiv.org/abs/2412.19437", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "DeepSeek V3 paper Table 6: llama-3.1-405b-instruct=50.4.", "candidates": [] }, { "model_id": "claude-3.5-sonnet", "benchmark_id": "chinese_simpleqa", "score": 51.3, "reference_url": "https://arxiv.org/abs/2412.19437", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "DeepSeek V3 paper Table 6: claude-3.5-sonnet=51.3.", "candidates": [] }, { "model_id": "gpt-4o-0513", "benchmark_id": "chinese_simpleqa", "score": 59.3, "reference_url": "https://arxiv.org/abs/2412.19437", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "DeepSeek V3 paper Table 6: gpt-4o-0513=59.3.", "candidates": [] }, { "model_id": "deepseek-v3", "benchmark_id": "chinese_simpleqa", "score": 64.8, "reference_url": "https://arxiv.org/abs/2412.19437", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "DeepSeek V3 paper Table 6: deepseek-v3=64.8.", "candidates": [] }, { "model_id": "deepseek-r1", "benchmark_id": "drop", "score": 92.2, "reference_url": "https://arxiv.org/abs/2501.12948", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "DeepSeek R1 paper Table 3 (page 9, R1 final): 92.2.", "candidates": [] }, { "model_id": "deepseek-r1", "benchmark_id": "alpacaeval_2", "score": 87.6, "reference_url": "https://arxiv.org/abs/2501.12948", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "DeepSeek R1 paper Table 3 (page 9, R1 final): 87.6.", "candidates": [] }, { "model_id": "deepseek-r1", "benchmark_id": "cluewsc", "score": 92.8, "reference_url": "https://arxiv.org/abs/2501.12948", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "DeepSeek R1 paper Table 3 (page 9, R1 final): 92.8.", "candidates": [] }, { "model_id": "deepseek-r1", "benchmark_id": "c_eval", "score": 91.8, "reference_url": "https://arxiv.org/abs/2501.12948", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "DeepSeek R1 paper Table 3 (page 9, R1 final): 91.8.", "candidates": [] }, { "model_id": "deepseek-r1", "benchmark_id": "chinese_simpleqa", "score": 63.7, "reference_url": "https://arxiv.org/abs/2501.12948", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "DeepSeek R1 paper Table 3 (page 9, R1 final): 63.7.", "candidates": [] }, { "model_id": "gpt-4o-0513", "benchmark_id": "codeforces_rating", "score": 759, "reference_url": "https://arxiv.org/abs/2501.12948", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "DeepSeek R1 paper Table 15: gpt-4o-0513=759.", "candidates": [] }, { "model_id": "claude-3.5-sonnet", "benchmark_id": "codeforces_rating", "score": 717, "reference_url": "https://arxiv.org/abs/2501.12948", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "DeepSeek R1 paper Table 15: claude-3.5-sonnet=717.", "candidates": [] }, { "model_id": "qwq-32b", "benchmark_id": "livebench", "score": 72.0, "reference_url": "https://arxiv.org/abs/2505.09388", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "QwQ-32B blog (qwenlm.github.io/blog/qwq-32b/) + Qwen3 tech report Table 13 baseline column: LiveBench 2024-11-25 72.0", "candidates": [] }, { "model_id": "qwq-32b", "benchmark_id": "bfcl_v3", "score": 66.4, "reference_url": "https://arxiv.org/abs/2505.09388", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "QwQ-32B blog (qwenlm.github.io/blog/qwq-32b/) + Qwen3 tech report Table 13 baseline column: BFCL v3 66.4", "candidates": [] }, { "model_id": "qwq-32b", "benchmark_id": "multi_if", "score": 68.3, "reference_url": "https://arxiv.org/abs/2505.09388", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "QwQ-32B blog (qwenlm.github.io/blog/qwq-32b/) + Qwen3 tech report Table 13 baseline column: Multi-IF 68.3", "candidates": [] }, { "model_id": "qwq-32b", "benchmark_id": "mmmlu", "score": 80.9, "reference_url": "https://arxiv.org/abs/2505.09388", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "QwQ-32B blog (qwenlm.github.io/blog/qwq-32b/) + Qwen3 tech report Table 13 baseline column: MMMLU 14 langs 80.9", "candidates": [] }, { "model_id": "qwq-32b", "benchmark_id": "mt_aime_2024", "score": 68.0, "reference_url": "https://arxiv.org/abs/2505.09388", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "QwQ-32B blog (qwenlm.github.io/blog/qwq-32b/) + Qwen3 tech report Table 13 baseline column: MT-AIME2024 68.0", "candidates": [] }, { "model_id": "qwq-32b", "benchmark_id": "mmlu_redux", "score": 90.0, "reference_url": "https://arxiv.org/abs/2505.09388", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "QwQ-32B blog (qwenlm.github.io/blog/qwq-32b/) + Qwen3 tech report Table 13 baseline column: MMLU-Redux 90.0", "candidates": [] }, { "model_id": "deepseek-r1-distill-llama-70b", "benchmark_id": "ifeval", "score": 79.3, "reference_url": "https://arxiv.org/abs/2505.09388", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Qwen3 tech report (arxiv:2505.09388) Table 13 baseline column for DeepSeek-R1-Distill-Llama-70B (third-party Qwen self-test, matches_canonical=false): IFEval strict prompt 79.3", "candidates": [] }, { "model_id": "deepseek-r1-distill-llama-70b", "benchmark_id": "livebench", "score": 54.5, "reference_url": "https://arxiv.org/abs/2505.09388", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Qwen3 tech report (arxiv:2505.09388) Table 13 baseline column for DeepSeek-R1-Distill-Llama-70B (third-party Qwen self-test, matches_canonical=false): LiveBench 2024-11-25 54.5", "candidates": [] }, { "model_id": "deepseek-r1-distill-llama-70b", "benchmark_id": "arena_hard", "score": 60.6, "reference_url": "https://arxiv.org/abs/2505.09388", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Qwen3 tech report (arxiv:2505.09388) Table 13 baseline column for DeepSeek-R1-Distill-Llama-70B (third-party Qwen self-test, matches_canonical=false): Arena-Hard 60.6", "candidates": [] }, { "model_id": "deepseek-r1-distill-llama-70b", "benchmark_id": "mmlu_redux", "score": 89.3, "reference_url": "https://arxiv.org/abs/2505.09388", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Qwen3 tech report (arxiv:2505.09388) Table 13 baseline column for DeepSeek-R1-Distill-Llama-70B (third-party Qwen self-test, matches_canonical=false): MMLU-Redux 89.3", "candidates": [] }, { "model_id": "deepseek-r1-distill-llama-70b", "benchmark_id": "bfcl_v3", "score": 49.3, "reference_url": "https://arxiv.org/abs/2505.09388", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Qwen3 tech report (arxiv:2505.09388) Table 13 baseline column for DeepSeek-R1-Distill-Llama-70B (third-party Qwen self-test, matches_canonical=false): BFCL v3 49.3", "candidates": [] }, { "model_id": "deepseek-r1-distill-llama-70b", "benchmark_id": "multi_if", "score": 57.6, "reference_url": "https://arxiv.org/abs/2505.09388", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Qwen3 tech report (arxiv:2505.09388) Table 13 baseline column for DeepSeek-R1-Distill-Llama-70B (third-party Qwen self-test, matches_canonical=false): Multi-IF 57.6", "candidates": [] }, { "model_id": "deepseek-r1-distill-llama-70b", "benchmark_id": "mmmlu", "score": 69.6, "reference_url": "https://arxiv.org/abs/2505.09388", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Qwen3 tech report (arxiv:2505.09388) Table 13 baseline column for DeepSeek-R1-Distill-Llama-70B (third-party Qwen self-test, matches_canonical=false): MMMLU 14 langs 69.6", "candidates": [] }, { "model_id": "deepseek-r1-distill-llama-70b", "benchmark_id": "mt_aime_2024", "score": 29.3, "reference_url": "https://arxiv.org/abs/2505.09388", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Qwen3 tech report (arxiv:2505.09388) Table 13 baseline column for DeepSeek-R1-Distill-Llama-70B (third-party Qwen self-test, matches_canonical=false): MT-AIME2024 29.3", "candidates": [] }, { "model_id": "glm-5", "benchmark_id": "hle", "score": 30.5, "reference_url": "https://z.ai/blog/glm-5.1", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "GLM-5.1 blog: glm-5=30.5.", "candidates": [] }, { "model_id": "minimax-m2.7", "benchmark_id": "hle", "score": 28.0, "reference_url": "https://z.ai/blog/glm-5.1", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "GLM-5.1 blog: minimax-m2.7=28.0.", "candidates": [] }, { "model_id": "minimax-m2.7", "benchmark_id": "aime_2026", "score": 89.8, "reference_url": "https://z.ai/blog/glm-5.1", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "GLM-5.1 blog: minimax-m2.7=89.8.", "candidates": [] }, { "model_id": "claude-opus-4.6", "benchmark_id": "aime_2026", "score": 95.6, "reference_url": "https://z.ai/blog/glm-5.1", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "GLM-5.1 blog: claude-opus-4.6=95.6.", "candidates": [ { "score": 96.7, "reference_url": "https://huggingface.co/moonshotai/Kimi-K2.6", "source_type": "model_card", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "1.0" }, "notes": "Kimi K2.6 model card: claude-opus-4.6=96.7." }, { "score": 81.2, "reference_url": "https://ernie.baidu.com/blog/posts/ernie-5.1-0508-release/benchmark.png", "source_type": "official_blog", "reported_setting": { "tools": "tools enabled (unspecified)", "sampling": "avg@32", "metric": "score" }, "notes": "With-tool avg@32 value is noncanonical against pass@1/no-tools. Research observation obs-003." }, { "score": 96.7, "reference_url": "https://ernie.baidu.com/blog/posts/ernie-5.1-0508-release/benchmark.png", "source_type": "official_blog", "reported_setting": { "tools": "none implied by contrast with w/tools row", "sampling": "avg@32" }, "notes": "No-tool avg@32 value is noncanonical against pass@1. Research observation obs-019. Literal marker ▲ has no source legend. Marker has no legend in image or article." } ] }, { "model_id": "gpt-5.4", "benchmark_id": "aime_2026", "score": 98.7, "reference_url": "https://z.ai/blog/glm-5.1", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "GLM-5.1 blog: gpt-5.4=98.7.", "candidates": [ { "score": 99.2, "reference_url": "https://huggingface.co/moonshotai/Kimi-K2.6", "source_type": "model_card", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "1.0" }, "notes": "Kimi K2.6 model card: gpt-5.4=99.2." } ] }, { "model_id": "glm-5.1", "benchmark_id": "hmmt_nov_2025", "score": 94.0, "reference_url": "https://z.ai/blog/glm-5.1", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "GLM-5.1 blog: glm-5.1=94.0.", "candidates": [] }, { "model_id": "glm-5", "benchmark_id": "hmmt_nov_2025", "score": 96.9, "reference_url": "https://z.ai/blog/glm-5.1", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "GLM-5.1 blog: glm-5=96.9.", "candidates": [] }, { "model_id": "qwen3.6-plus", "benchmark_id": "hmmt_nov_2025", "score": 94.6, "reference_url": "https://z.ai/blog/glm-5.1", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "GLM-5.1 blog: qwen3.6-plus=94.6.", "candidates": [] }, { "model_id": "minimax-m2.7", "benchmark_id": "hmmt_nov_2025", "score": 81.0, "reference_url": "https://z.ai/blog/glm-5.1", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "GLM-5.1 blog: minimax-m2.7=81.0.", "candidates": [] }, { "model_id": "claude-opus-4.6", "benchmark_id": "hmmt_nov_2025", "score": 96.3, "reference_url": "https://z.ai/blog/glm-5.1", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "GLM-5.1 blog: claude-opus-4.6=96.3.", "candidates": [] }, { "model_id": "gemini-3.1-pro", "benchmark_id": "hmmt_nov_2025", "score": 94.8, "reference_url": "https://z.ai/blog/glm-5.1", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "GLM-5.1 blog: gemini-3.1-pro=94.8.", "candidates": [] }, { "model_id": "gpt-5.4", "benchmark_id": "hmmt_nov_2025", "score": 95.8, "reference_url": "https://z.ai/blog/glm-5.1", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "GLM-5.1 blog: gpt-5.4=95.8.", "candidates": [] }, { "model_id": "qwen3.6-plus", "benchmark_id": "hmmt_feb_2026", "score": 87.8, "reference_url": "https://z.ai/blog/glm-5.1", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "GLM-5.1 blog: qwen3.6-plus=87.8.", "candidates": [] }, { "model_id": "minimax-m2.7", "benchmark_id": "hmmt_feb_2026", "score": 72.7, "reference_url": "https://z.ai/blog/glm-5.1", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "GLM-5.1 blog: minimax-m2.7=72.7.", "candidates": [] }, { "model_id": "glm-5", "benchmark_id": "imo_answerbench", "score": 82.5, "reference_url": "https://z.ai/blog/glm-5.1", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "GLM-5.1 blog: glm-5=82.5.", "candidates": [] }, { "model_id": "qwen3.6-plus", "benchmark_id": "imo_answerbench", "score": 83.8, "reference_url": "https://z.ai/blog/glm-5.1", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "GLM-5.1 blog: qwen3.6-plus=83.8.", "candidates": [] }, { "model_id": "minimax-m2.7", "benchmark_id": "imo_answerbench", "score": 66.3, "reference_url": "https://z.ai/blog/glm-5.1", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "GLM-5.1 blog: minimax-m2.7=66.3.", "candidates": [ { "score": 68.3, "reference_url": "https://research.nvidia.com/labs/nemotron/files/NVIDIA-Nemotron-3-Ultra-Technical-Report.pdf", "source_type": "tech_report", "reported_setting": { "version_metric": "no tools / %", "tools": "none", "harness": "NeMo Evaluator", "sampling": "5-run average", "variant": "no tools" }, "notes": "NVIDIA Nemotron 3 Ultra technical report, Final Table 10: MiniMax M2.7: imo_answerbench=68.3, variant=no tools." }, { "score": 75.1, "reference_url": "https://research.nvidia.com/labs/nemotron/files/NVIDIA-Nemotron-3-Ultra-Technical-Report.pdf", "source_type": "tech_report", "reported_setting": { "version_metric": "with tools / %", "tools": "benchmark tool environment", "harness": "NeMo Evaluator", "sampling": "pass@1", "variant": "with tools" }, "notes": "NVIDIA Nemotron 3 Ultra technical report, Final Table 10: MiniMax M2.7: imo_answerbench=75.1, variant=with tools." } ] }, { "model_id": "glm-5", "benchmark_id": "gpqa_diamond", "score": 86.0, "reference_url": "https://z.ai/blog/glm-5.1", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "GLM-5.1 blog: glm-5=86.0.", "candidates": [] }, { "model_id": "minimax-m2.7", "benchmark_id": "gpqa_diamond", "score": 87.0, "reference_url": "https://z.ai/blog/glm-5.1", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "GLM-5.1 blog: minimax-m2.7=87.0.", "candidates": [ { "score": 86.6, "reference_url": "https://research.nvidia.com/labs/nemotron/files/NVIDIA-Nemotron-3-Ultra-Technical-Report.pdf", "source_type": "tech_report", "reported_setting": { "version_metric": "no tools / %", "tools": "none", "harness": "NeMo Evaluator", "sampling": "8-run average" }, "notes": "NVIDIA Nemotron 3 Ultra technical report, Final Table 10: MiniMax M2.7: gpqa_diamond=86.6." } ] }, { "model_id": "glm-5", "benchmark_id": "swe_bench_pro", "score": 55.1, "reference_url": "https://z.ai/blog/glm-5.1", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "agentic (bash + edit)", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "GLM-5.1 blog: glm-5=55.1.", "candidates": [] }, { "model_id": "minimax-m2.7", "benchmark_id": "swe_bench_pro", "score": 56.2, "reference_url": "https://z.ai/blog/glm-5.1", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "agentic (bash + edit)", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "GLM-5.1 blog: minimax-m2.7=56.2.", "candidates": [] }, { "model_id": "glm-5.1", "benchmark_id": "nl2repo_bench", "score": 42.7, "reference_url": "https://z.ai/blog/glm-5.1", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "agentic (bash + edit)", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "GLM-5.1 blog: glm-5.1=42.7.", "candidates": [] }, { "model_id": "glm-5", "benchmark_id": "nl2repo_bench", "score": 35.9, "reference_url": "https://z.ai/blog/glm-5.1", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "agentic (bash + edit)", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "GLM-5.1 blog: glm-5=35.9.", "candidates": [] }, { "model_id": "qwen3.6-plus", "benchmark_id": "nl2repo_bench", "score": 37.9, "reference_url": "https://z.ai/blog/glm-5.1", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "agentic (bash + edit)", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "GLM-5.1 blog: qwen3.6-plus=37.9.", "candidates": [] }, { "model_id": "minimax-m2.7", "benchmark_id": "nl2repo_bench", "score": 39.8, "reference_url": "https://z.ai/blog/glm-5.1", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "agentic (bash + edit)", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "GLM-5.1 blog: minimax-m2.7=39.8.", "candidates": [] }, { "model_id": "kimi-k2.5", "benchmark_id": "nl2repo_bench", "score": 32.0, "reference_url": "https://z.ai/blog/glm-5.1", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "agentic (bash + edit)", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "GLM-5.1 blog: kimi-k2.5=32.0.", "candidates": [] }, { "model_id": "claude-opus-4.6", "benchmark_id": "nl2repo_bench", "score": 49.8, "reference_url": "https://z.ai/blog/glm-5.1", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "agentic (bash + edit)", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "GLM-5.1 blog: claude-opus-4.6=49.8.", "candidates": [] }, { "model_id": "gemini-3.1-pro", "benchmark_id": "nl2repo_bench", "score": 33.4, "reference_url": "https://z.ai/blog/glm-5.1", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "agentic (bash + edit)", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "GLM-5.1 blog: gemini-3.1-pro=33.4.", "candidates": [ { "score": 33.4, "reference_url": "https://seed.bytedance.com/en/blog/seed2-1-officially-released-advancing-ai-productivity", "source_type": "official_blog", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "Trae or benchmark-specific coding-agent environment", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official", "prompt_style": "official default", "temperature": "unknown" }, "notes": "Release blog EN: Coding chart (https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/user-upload/4xfa4mqq4wlbt.png)." } ] }, { "model_id": "gpt-5.4", "benchmark_id": "nl2repo_bench", "score": 41.3, "reference_url": "https://z.ai/blog/glm-5.1", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "agentic (bash + edit)", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "GLM-5.1 blog: gpt-5.4=41.3.", "candidates": [] }, { "model_id": "glm-5", "benchmark_id": "terminal_bench", "score": 56.2, "reference_url": "https://z.ai/blog/glm-5.1", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "agentic", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "GLM-5.1 blog: glm-5=56.2.", "candidates": [ { "score": 59.6, "reference_url": "https://cursor.com/resources/Composer2.pdf", "source_type": "tech_report", "reported_setting": { "effort": "source does not state", "tools": "Cursor coding agent", "sampling": "pass@1", "harness": "Cursor harness" }, "notes": "Composer 2 technical report Table 1 / Terminal-Bench 2.0 / GLM-5 / value 1: For third-party public benchmarks, slash-separated values are Cursor/official-harness then self-reported. SWE prompts prepend 'please solve this github issue'; Terminal prompts add solution-format instructions." } ] }, { "model_id": "glm-5.1", "benchmark_id": "cybergym", "score": 68.7, "reference_url": "https://z.ai/blog/glm-5.1", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "agentic", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "GLM-5.1 blog: glm-5.1=68.7.", "candidates": [] }, { "model_id": "glm-5", "benchmark_id": "cybergym", "score": 48.3, "reference_url": "https://z.ai/blog/glm-5.1", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "agentic", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "GLM-5.1 blog: glm-5=48.3.", "candidates": [ { "score": 43.2, "reference_url": "https://z.ai/blog/glm-5", "source_type": "official_blog", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "agentic", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "notes": "GLM-5 blog: glm-5=43.2." } ] }, { "model_id": "deepseek-v3.2", "benchmark_id": "cybergym", "score": 17.3, "reference_url": "https://z.ai/blog/glm-5.1", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "agentic", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "GLM-5.1 blog: deepseek-v3.2=17.3.", "candidates": [] }, { "model_id": "gemini-3.1-pro", "benchmark_id": "cybergym", "score": 38.8, "reference_url": "https://z.ai/blog/glm-5.1", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "agentic", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "GLM-5.1 blog: gemini-3.1-pro=38.8.", "candidates": [] }, { "model_id": "glm-5", "benchmark_id": "browsecomp", "score": 62.0, "reference_url": "https://z.ai/blog/glm-5.1", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "agentic", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "GLM-5.1 blog: glm-5=62.0.", "candidates": [] }, { "model_id": "glm-5.1", "benchmark_id": "tau3_bench", "score": 70.6, "reference_url": "https://z.ai/blog/glm-5.1", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "agentic", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "GLM-5.1 blog: glm-5.1=70.6.", "candidates": [ { "score": 69.7, "reference_url": "https://research.nvidia.com/labs/nemotron/files/NVIDIA-Nemotron-3-Ultra-Technical-Report.pdf", "source_type": "tech_report", "reported_setting": { "version_metric": "four-domain / 8-trial average", "tools": "domain customer-service tools", "harness": "NeMo Gym / tau3", "sampling": "375 tasks x 8 trials" }, "notes": "NVIDIA Nemotron 3 Ultra technical report, Final Table 10: GLM-5.1: tau3_bench=69.7." }, { "score": 70.6, "reference_url": "https://z.ai/blog/assets/glm-5.1-B3CLmgrT.js", "source_type": "official_blog", "reported_setting": { "mode": "thinking", "effort": "official reported setting", "tools": "source-reported agentic tools", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official Z.ai GLM-5.1 evaluation", "prompt_style": "provider official evaluation prompt", "temperature": "source/provider setting", "context": "source/provider setting", "input_modalities": "text", "trials": "source-reported" }, "notes": "Exact provider-official score and reported setting." } ] }, { "model_id": "glm-5", "benchmark_id": "tau3_bench", "score": 69.2, "reference_url": "https://z.ai/blog/glm-5.1", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "agentic", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "GLM-5.1 blog: glm-5=69.2.", "candidates": [] }, { "model_id": "qwen3.6-plus", "benchmark_id": "tau3_bench", "score": 70.7, "reference_url": "https://z.ai/blog/glm-5.1", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "agentic", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "GLM-5.1 blog: qwen3.6-plus=70.7.", "candidates": [] }, { "model_id": "minimax-m2.7", "benchmark_id": "tau3_bench", "score": 67.6, "reference_url": "https://z.ai/blog/glm-5.1", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "agentic", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "GLM-5.1 blog: minimax-m2.7=67.6.", "candidates": [ { "score": 66.1, "reference_url": "https://research.nvidia.com/labs/nemotron/files/NVIDIA-Nemotron-3-Ultra-Technical-Report.pdf", "source_type": "tech_report", "reported_setting": { "version_metric": "four-domain / 8-trial average", "tools": "domain customer-service tools", "harness": "NeMo Gym / tau3", "sampling": "375 tasks x 8 trials" }, "notes": "NVIDIA Nemotron 3 Ultra technical report, Final Table 10: MiniMax M2.7: tau3_bench=66.1." } ] }, { "model_id": "deepseek-v3.2", "benchmark_id": "tau3_bench", "score": 69.2, "reference_url": "https://z.ai/blog/glm-5.1", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "agentic", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "GLM-5.1 blog: deepseek-v3.2=69.2.", "candidates": [] }, { "model_id": "kimi-k2.5", "benchmark_id": "tau3_bench", "score": 66.0, "reference_url": "https://z.ai/blog/glm-5.1", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "agentic", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "GLM-5.1 blog: kimi-k2.5=66.0.", "candidates": [] }, { "model_id": "claude-opus-4.6", "benchmark_id": "tau3_bench", "score": 72.4, "reference_url": "https://z.ai/blog/glm-5.1", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "agentic", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "GLM-5.1 blog: claude-opus-4.6=72.4.", "candidates": [ { "score": 72.4, "reference_url": "https://ernie.baidu.com/blog/posts/ernie-5.1-0508-release/benchmark.png", "source_type": "official_blog", "reported_setting": { "tools": "agentic", "sampling": "avg@4" }, "notes": "The source's avg@4 agentic setting matches the campaign row. Research observation obs-007. Literal marker ◆ has no source legend. Marker has no legend in image or article." } ] }, { "model_id": "gemini-3.1-pro", "benchmark_id": "tau3_bench", "score": 67.1, "reference_url": "https://z.ai/blog/glm-5.1", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "agentic", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "GLM-5.1 blog: gemini-3.1-pro=67.1.", "candidates": [ { "score": 67.1, "reference_url": "https://ernie.baidu.com/blog/posts/ernie-5.1-0508-release/benchmark.png", "source_type": "official_blog", "reported_setting": { "tools": "agentic", "sampling": "avg@4" }, "notes": "The source's avg@4 agentic setting matches the campaign row. Research observation obs-008. Literal marker ◆ has no source legend. Marker has no legend in image or article." } ] }, { "model_id": "gpt-5.4", "benchmark_id": "tau3_bench", "score": 72.9, "reference_url": "https://z.ai/blog/glm-5.1", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "agentic", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "GLM-5.1 blog: gpt-5.4=72.9.", "candidates": [] }, { "model_id": "glm-5", "benchmark_id": "mcpatlas", "score": 69.2, "reference_url": "https://z.ai/blog/glm-5.1", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "agentic", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "GLM-5.1 blog: glm-5=69.2.", "candidates": [ { "score": 67.8, "reference_url": "https://z.ai/blog/glm-5", "source_type": "official_blog", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "agentic", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "notes": "GLM-5 blog: glm-5=67.8." } ] }, { "model_id": "qwen3.6-plus", "benchmark_id": "mcpatlas", "score": 74.1, "reference_url": "https://z.ai/blog/glm-5.1", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "agentic", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "GLM-5.1 blog: qwen3.6-plus=74.1.", "candidates": [] }, { "model_id": "minimax-m2.7", "benchmark_id": "mcpatlas", "score": 48.8, "reference_url": "https://z.ai/blog/glm-5.1", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "agentic", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "GLM-5.1 blog: minimax-m2.7=48.8.", "candidates": [] }, { "model_id": "deepseek-v3.2", "benchmark_id": "mcpatlas", "score": 62.2, "reference_url": "https://z.ai/blog/glm-5.1", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "agentic", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "GLM-5.1 blog: deepseek-v3.2=62.2.", "candidates": [] }, { "model_id": "kimi-k2.5", "benchmark_id": "mcpatlas", "score": 63.8, "reference_url": "https://z.ai/blog/glm-5.1", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "agentic", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "GLM-5.1 blog: kimi-k2.5=63.8.", "candidates": [] }, { "model_id": "glm-5", "benchmark_id": "toolathlon", "score": 38.0, "reference_url": "https://z.ai/blog/glm-5.1", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "agentic", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "GLM-5.1 blog: glm-5=38.0.", "candidates": [ { "score": 39.2, "reference_url": "https://z.ai/blog/glm-5", "source_type": "official_blog", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "agentic", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "notes": "GLM-5 blog: glm-5=39.2." } ] }, { "model_id": "qwen3.6-plus", "benchmark_id": "toolathlon", "score": 39.8, "reference_url": "https://z.ai/blog/glm-5.1", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "agentic", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "GLM-5.1 blog: qwen3.6-plus=39.8.", "candidates": [] }, { "model_id": "minimax-m2.7", "benchmark_id": "toolathlon", "score": 46.3, "reference_url": "https://z.ai/blog/glm-5.1", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "agentic", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "GLM-5.1 blog: minimax-m2.7=46.3.", "candidates": [] }, { "model_id": "deepseek-v3.2", "benchmark_id": "toolathlon", "score": 35.2, "reference_url": "https://z.ai/blog/glm-5.1", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "agentic", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "GLM-5.1 blog: deepseek-v3.2=35.2.", "candidates": [ { "score": 35.2, "reference_url": "https://llm-stats.com/benchmarks/toolathlon", "source_type": "third_party_aggregator", "reported_setting": { "mode": "thinking (DeepSeek-Reasoner endpoint)" }, "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=DeepSeek-V3.2 (Thinking), slug=deepseek-reasoner, provider=DeepSeek" } ] }, { "model_id": "glm-5.1", "benchmark_id": "vending_bench_2", "score": 5634.41, "reference_url": "https://z.ai/blog/glm-5.1", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "agentic", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "GLM-5.1 blog: glm-5.1=5634.41.", "candidates": [] }, { "model_id": "glm-5", "benchmark_id": "vending_bench_2", "score": 4432.12, "reference_url": "https://z.ai/blog/glm-5.1", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "agentic", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "GLM-5.1 blog: glm-5=4432.12.", "candidates": [] }, { "model_id": "qwen3.6-plus", "benchmark_id": "vending_bench_2", "score": 5114.87, "reference_url": "https://z.ai/blog/glm-5.1", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "agentic", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "GLM-5.1 blog: qwen3.6-plus=5114.87.", "candidates": [] }, { "model_id": "deepseek-v3.2", "benchmark_id": "vending_bench_2", "score": 1034.0, "reference_url": "https://z.ai/blog/glm-5.1", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "agentic", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "GLM-5.1 blog: deepseek-v3.2=1034.0.", "candidates": [] }, { "model_id": "kimi-k2.5", "benchmark_id": "vending_bench_2", "score": 1198.46, "reference_url": "https://z.ai/blog/glm-5.1", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "agentic", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "GLM-5.1 blog: kimi-k2.5=1198.46.", "candidates": [] }, { "model_id": "claude-opus-4.6", "benchmark_id": "vending_bench_2", "score": 8017.59, "reference_url": "https://z.ai/blog/glm-5.1", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "agentic", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "GLM-5.1 blog: claude-opus-4.6=8017.59.", "candidates": [] }, { "model_id": "gemini-3.1-pro", "benchmark_id": "vending_bench_2", "score": 911.21, "reference_url": "https://z.ai/blog/glm-5.1", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "agentic", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "GLM-5.1 blog: gemini-3.1-pro=911.21.", "candidates": [] }, { "model_id": "gpt-5.4", "benchmark_id": "vending_bench_2", "score": 6144.18, "reference_url": "https://z.ai/blog/glm-5.1", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "agentic", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "GLM-5.1 blog: gpt-5.4=6144.18.", "candidates": [] }, { "model_id": "glm-4.7", "benchmark_id": "hmmt_nov_2025", "score": 93.5, "reference_url": "https://z.ai/blog/glm-5", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "GLM-5 blog: glm-4.7=93.5.", "candidates": [] }, { "model_id": "claude-opus-4.5", "benchmark_id": "hmmt_nov_2025", "score": 91.7, "reference_url": "https://z.ai/blog/glm-5", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "GLM-5 blog: claude-opus-4.5=91.7.", "candidates": [ { "score": 93.3, "reference_url": "https://huggingface.co/Qwen/Qwen3.5-397B-A17B", "source_type": "model_card", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "notes": "Qwen3.5-397B-A17B HF model card cross-model table (Qwen self-test of 3rd-party model, mc=false): alternate measurement 93.3 (3rd-party Qwen self-test)." }, { "score": 93.3, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "source_type": "model_card", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "notes": "Doubao Seed 2.0 model card Table 3: claude-opus-4.5=93.3." } ] }, { "model_id": "glm-4.7", "benchmark_id": "imo_answerbench", "score": 82.0, "reference_url": "https://z.ai/blog/glm-5", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "GLM-5 blog: glm-4.7=82.0.", "candidates": [] }, { "model_id": "claude-opus-4.5", "benchmark_id": "imo_answerbench", "score": 78.5, "reference_url": "https://z.ai/blog/glm-5", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "GLM-5 blog: claude-opus-4.5=78.5.", "candidates": [ { "score": 84.0, "reference_url": "https://huggingface.co/Qwen/Qwen3.5-397B-A17B", "source_type": "model_card", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "notes": "Qwen3.5-397B-A17B HF model card cross-model table (Qwen self-test of 3rd-party model, mc=false): alternate measurement 84.0 (3rd-party Qwen self-test)." } ] }, { "model_id": "gpt-5.2", "benchmark_id": "imo_answerbench", "score": 86.3, "reference_url": "https://z.ai/blog/glm-5", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "GLM-5 blog: gpt-5.2=86.3.", "candidates": [] }, { "model_id": "glm-5", "benchmark_id": "swe_bench_verified", "score": 77.8, "reference_url": "https://z.ai/blog/glm-5", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "agentic (bash + edit)", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "GLM-5 blog: glm-5=77.8.", "candidates": [] }, { "model_id": "glm-5", "benchmark_id": "swe_bench_multilingual", "score": 73.3, "reference_url": "https://z.ai/blog/glm-5", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "agentic (bash + edit)", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "GLM-5 blog: glm-5=73.3.", "candidates": [ { "score": 66.9, "reference_url": "https://cursor.com/resources/Composer2.pdf", "source_type": "tech_report", "reported_setting": { "effort": "source does not state", "tools": "Cursor coding agent", "sampling": "pass@1", "harness": "Cursor harness" }, "notes": "Composer 2 technical report Table 1 / SWE-Bench Multilingual / GLM-5 / value 1: For third-party public benchmarks, slash-separated values are Cursor/official-harness then self-reported. SWE prompts prepend 'please solve this github issue'; Terminal prompts add solution-format instructions." } ] }, { "model_id": "glm-4.7", "benchmark_id": "swe_bench_multilingual", "score": 66.7, "reference_url": "https://z.ai/blog/glm-5", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "agentic (bash + edit)", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "GLM-5 blog: glm-4.7=66.7.", "candidates": [] }, { "model_id": "claude-opus-4.5", "benchmark_id": "swe_bench_multilingual", "score": 77.5, "reference_url": "https://z.ai/blog/glm-5", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "agentic (bash + edit)", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "GLM-5 blog: claude-opus-4.5=77.5.", "candidates": [ { "score": 74.0, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "source_type": "model_card", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "varies", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "notes": "(merged from swe_multilingual) Doubao Seed 2.0 model card: claude-opus-4.5=74.0." }, { "score": 73.8, "reference_url": "https://cursor.com/resources/Composer2.pdf", "source_type": "tech_report", "reported_setting": { "effort": "high", "tools": "Cursor coding agent", "sampling": "pass@1", "harness": "Cursor harness" }, "notes": "Composer 2 technical report Table 1 / SWE-Bench Multilingual / Opus 4.5 High / value 1: For third-party public benchmarks, slash-separated values are Cursor/official-harness then self-reported. SWE prompts prepend 'please solve this github issue'; Terminal prompts add solution-format instructions." }, { "score": 76.2, "reference_url": "https://cursor.com/resources/Composer2.pdf", "source_type": "tech_report", "reported_setting": { "effort": "high", "tools": "Cursor coding agent", "sampling": "pass@1", "harness": "self-reported" }, "notes": "Composer 2 technical report Table 1 / SWE-Bench Multilingual / Opus 4.5 High / value 2: For third-party public benchmarks, slash-separated values are Cursor/official-harness then self-reported. SWE prompts prepend 'please solve this github issue'; Terminal prompts add solution-format instructions." } ] }, { "model_id": "gemini-3-pro", "benchmark_id": "swe_bench_multilingual", "score": 65.0, "reference_url": "https://z.ai/blog/glm-5", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "agentic (bash + edit)", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "GLM-5 blog: gemini-3-pro=65.0.", "candidates": [ { "score": 72.7, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "source_type": "model_card", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "varies", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "notes": "(merged from swe_multilingual) Doubao Seed 2.0 model card: gemini-3-pro=72.7." } ] }, { "model_id": "gpt-5.2", "benchmark_id": "swe_bench_multilingual", "score": 72.0, "reference_url": "https://z.ai/blog/glm-5", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "agentic (bash + edit)", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "GLM-5 blog: gpt-5.2=72.0.", "candidates": [ { "score": 68.8, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "source_type": "model_card", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "varies", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "notes": "(merged from swe_multilingual) Doubao Seed 2.0 model card: gpt-5.2=68.8." }, { "score": 68.3, "reference_url": "https://cursor.com/resources/Composer2.pdf", "source_type": "tech_report", "reported_setting": { "effort": "high", "tools": "Cursor coding agent", "sampling": "pass@1", "harness": "Cursor harness" }, "notes": "Composer 2 technical report Table 1 / SWE-Bench Multilingual / GPT-5.2 / value 1: For third-party public benchmarks, slash-separated values are Cursor/official-harness then self-reported. SWE prompts prepend 'please solve this github issue'; Terminal prompts add solution-format instructions." } ] }, { "model_id": "glm-4.7", "benchmark_id": "cybergym", "score": 23.5, "reference_url": "https://z.ai/blog/glm-5", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "agentic", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "GLM-5 blog: glm-4.7=23.5.", "candidates": [] }, { "model_id": "claude-opus-4.5", "benchmark_id": "cybergym", "score": 50.6, "reference_url": "https://z.ai/blog/glm-5", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "agentic", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "GLM-5 blog: claude-opus-4.5=50.6.", "candidates": [] }, { "model_id": "gemini-3-pro", "benchmark_id": "cybergym", "score": 39.9, "reference_url": "https://z.ai/blog/glm-5", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "agentic", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "GLM-5 blog: gemini-3-pro=39.9.", "candidates": [] }, { "model_id": "glm-5", "benchmark_id": "browsecomp_zh", "score": 72.7, "reference_url": "https://z.ai/blog/glm-5", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "agentic", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "GLM-5 blog: glm-5=72.7.", "candidates": [] }, { "model_id": "glm-4.7", "benchmark_id": "browsecomp_zh", "score": 66.6, "reference_url": "https://z.ai/blog/glm-5", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "agentic", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "GLM-5 blog: glm-4.7=66.6.", "candidates": [] }, { "model_id": "deepseek-v3.2", "benchmark_id": "browsecomp_zh", "score": 65.0, "reference_url": "https://z.ai/blog/glm-5", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "agentic", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "GLM-5 blog: deepseek-v3.2=65.0.", "candidates": [ { "score": 47.9, "reference_url": "https://huggingface.co/MiniMaxAI/MiniMax-M2", "source_type": "model_card", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "agentic", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "notes": "MiniMax M2 model card: deepseek-v3.2=47.9." }, { "score": 65.0, "reference_url": "https://llm-stats.com/benchmarks/browsecomp-zh", "source_type": "third_party_aggregator", "reported_setting": { "mode": "thinking (DeepSeek-Reasoner endpoint)" }, "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=DeepSeek-V3.2 (Thinking), slug=deepseek-reasoner, provider=DeepSeek" }, { "score": 47.9, "reference_url": "https://llm-stats.com/benchmarks/browsecomp-zh", "source_type": "third_party_aggregator", "reported_setting": { "variant": "exp" }, "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=DeepSeek-V3.2-Exp, slug=deepseek-v3.2-exp, provider=DeepSeek" } ] }, { "model_id": "kimi-k2.5", "benchmark_id": "browsecomp_zh", "score": 62.3, "reference_url": "https://z.ai/blog/glm-5", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "agentic", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "GLM-5 blog: kimi-k2.5=62.3.", "candidates": [] }, { "model_id": "claude-opus-4.5", "benchmark_id": "browsecomp_zh", "score": 62.4, "reference_url": "https://z.ai/blog/glm-5", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "agentic", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "GLM-5 blog: claude-opus-4.5=62.4.", "candidates": [] }, { "model_id": "gemini-3-pro", "benchmark_id": "browsecomp_zh", "score": 66.8, "reference_url": "https://z.ai/blog/glm-5", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "agentic", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "GLM-5 blog: gemini-3-pro=66.8.", "candidates": [] }, { "model_id": "gpt-5.2", "benchmark_id": "browsecomp_zh", "score": 76.1, "reference_url": "https://z.ai/blog/glm-5", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "agentic", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "GLM-5 blog: gpt-5.2=76.1.", "candidates": [] }, { "model_id": "glm-4.7", "benchmark_id": "mcpatlas", "score": 52.0, "reference_url": "https://z.ai/blog/glm-5", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "agentic", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "GLM-5 blog: glm-4.7=52.0.", "candidates": [] }, { "model_id": "glm-4.7", "benchmark_id": "toolathlon", "score": 23.8, "reference_url": "https://z.ai/blog/glm-5", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "agentic", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "GLM-5 blog: glm-4.7=23.8.", "candidates": [] }, { "model_id": "claude-opus-4.5", "benchmark_id": "toolathlon", "score": 43.5, "reference_url": "https://z.ai/blog/glm-5", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "agentic", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "GLM-5 blog: claude-opus-4.5=43.5.", "candidates": [] }, { "model_id": "glm-4.7", "benchmark_id": "vending_bench_2", "score": 2376.82, "reference_url": "https://z.ai/blog/glm-5", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "agentic", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "GLM-5 blog: glm-4.7=2376.82.", "candidates": [] }, { "model_id": "claude-opus-4.5", "benchmark_id": "vending_bench_2", "score": 4967.06, "reference_url": "https://z.ai/blog/glm-5", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "agentic", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "GLM-5 blog: claude-opus-4.5=4967.06.", "candidates": [] }, { "model_id": "glm-4.7", "benchmark_id": "hmmt_feb_2025", "score": 97.1, "reference_url": "https://z.ai/blog/glm-4.7", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "GLM-4.7 blog: glm-4.7=97.1.", "candidates": [] }, { "model_id": "glm-4.6", "benchmark_id": "hmmt_feb_2025", "score": 89.2, "reference_url": "https://z.ai/blog/glm-4.7", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "GLM-4.7 blog: glm-4.6=89.2.", "candidates": [] }, { "model_id": "glm-4.6", "benchmark_id": "hmmt_nov_2025", "score": 87.7, "reference_url": "https://z.ai/blog/glm-4.7", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "GLM-4.7 blog: glm-4.6=87.7.", "candidates": [] }, { "model_id": "glm-4.6", "benchmark_id": "imo_answerbench", "score": 73.5, "reference_url": "https://z.ai/blog/glm-4.7", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "GLM-4.7 blog: glm-4.6=73.5.", "candidates": [] }, { "model_id": "claude-sonnet-4.5", "benchmark_id": "imo_answerbench", "score": 65.8, "reference_url": "https://z.ai/blog/glm-4.7", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "GLM-4.7 blog: claude-sonnet-4.5=65.8.", "candidates": [ { "score": 65.9, "reference_url": "https://huggingface.co/moonshotai/Kimi-K2-Thinking", "source_type": "model_card", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "1.0" }, "notes": "(merged from imoanswerbench) Kimi K2-Thinking model card: claude-sonnet-4.5=65.9." } ] }, { "model_id": "glm-4.7", "benchmark_id": "livecodebench_v6", "score": 84.9, "reference_url": "https://z.ai/blog/glm-4.7", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "GLM-4.7 blog: glm-4.7=84.9.", "candidates": [] }, { "model_id": "glm-4.6", "benchmark_id": "livecodebench_v6", "score": 82.8, "reference_url": "https://z.ai/blog/glm-4.7", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "GLM-4.7 blog: glm-4.6=82.8.", "candidates": [] }, { "model_id": "kimi-k2-thinking", "benchmark_id": "livecodebench_v6", "score": 83.1, "reference_url": "https://z.ai/blog/glm-4.7", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "GLM-4.7 blog: kimi-k2-thinking=83.1.", "candidates": [] }, { "model_id": "deepseek-v3.2", "benchmark_id": "livecodebench_v6", "score": 83.3, "reference_url": "https://z.ai/blog/glm-4.7", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "GLM-4.7 blog: deepseek-v3.2=83.3.", "candidates": [ { "score": 74.1, "reference_url": "https://huggingface.co/moonshotai/Kimi-K2-Thinking", "source_type": "model_card", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "1.0" }, "notes": "Kimi K2-Thinking model card: deepseek-v3.2=74.1." }, { "score": 83.3, "reference_url": "https://microsoft.ai/wp-content/uploads/2026/06/main_20260602_2.pdf", "source_type": "official_paper", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none (per-bench override)", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "1.0", "context": "128K", "notes": "Per DeepSeek V3.2 tech report: temperature=1.0, context=128K, thinking mode for tool-use." }, "notes": "Displayed exactly as 83.3. Official Microsoft-reported result. Comparator value is quoted from that provider's official model card/release; use the campaign canonical setting." } ] }, { "model_id": "gemini-3-pro", "benchmark_id": "livecodebench_v6", "score": 90.7, "reference_url": "https://z.ai/blog/glm-4.7", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "GLM-4.7 blog: gemini-3-pro=90.7.", "candidates": [ { "score": 87.4, "reference_url": "https://huggingface.co/moonshotai/Kimi-K2.5", "source_type": "model_card", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "1.0" }, "notes": "Kimi K2.5 model card: gemini-3-pro=87.4." } ] }, { "model_id": "claude-sonnet-4.5", "benchmark_id": "livecodebench_v6", "score": 64.0, "reference_url": "https://z.ai/blog/glm-4.7", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "GLM-4.7 blog: claude-sonnet-4.5=64.0.", "candidates": [] }, { "model_id": "gpt-5", "benchmark_id": "livecodebench_v6", "score": 87.0, "reference_url": "https://z.ai/blog/glm-4.7", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "GLM-4.7 blog: gpt-5=87.0.", "candidates": [] }, { "model_id": "gpt-5.1", "benchmark_id": "livecodebench_v6", "score": 87.0, "reference_url": "https://z.ai/blog/glm-4.7", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "GLM-4.7 blog: gpt-5.1=87.0.", "candidates": [] }, { "model_id": "glm-4.6", "benchmark_id": "swe_bench_multilingual", "score": 53.8, "reference_url": "https://z.ai/blog/glm-4.7", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "agentic (bash + edit)", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "GLM-4.7 blog: glm-4.6=53.8.", "candidates": [] }, { "model_id": "glm-4.7", "benchmark_id": "terminal_bench_hard", "score": 33.3, "reference_url": "https://z.ai/blog/glm-4.7", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "agentic", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "GLM-4.7 blog: glm-4.7=33.3.", "candidates": [] }, { "model_id": "glm-4.6", "benchmark_id": "terminal_bench_hard", "score": 23.6, "reference_url": "https://z.ai/blog/glm-4.7", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "agentic", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "GLM-4.7 blog: glm-4.6=23.6.", "candidates": [ { "score": 23, "reference_url": "https://huggingface.co/MiniMaxAI/MiniMax-M2", "source_type": "model_card", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "agentic", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "notes": "MiniMax M2 model card: glm-4.6=23." } ] }, { "model_id": "kimi-k2-thinking", "benchmark_id": "terminal_bench_hard", "score": 30.6, "reference_url": "https://z.ai/blog/glm-4.7", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "agentic", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "GLM-4.7 blog: kimi-k2-thinking=30.6.", "candidates": [] }, { "model_id": "deepseek-v3.2", "benchmark_id": "terminal_bench_hard", "score": 35.4, "reference_url": "https://z.ai/blog/glm-4.7", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "agentic", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "GLM-4.7 blog: deepseek-v3.2=35.4.", "candidates": [ { "score": 29, "reference_url": "https://huggingface.co/MiniMaxAI/MiniMax-M2", "source_type": "model_card", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "agentic", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "notes": "MiniMax M2 model card: deepseek-v3.2=29." } ] }, { "model_id": "gemini-3-pro", "benchmark_id": "terminal_bench_hard", "score": 39.0, "reference_url": "https://z.ai/blog/glm-4.7", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "agentic", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "GLM-4.7 blog: gemini-3-pro=39.0.", "candidates": [] }, { "model_id": "claude-sonnet-4.5", "benchmark_id": "terminal_bench_hard", "score": 33.3, "reference_url": "https://z.ai/blog/glm-4.7", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "agentic", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "GLM-4.7 blog: claude-sonnet-4.5=33.3.", "candidates": [ { "score": 33, "reference_url": "https://huggingface.co/MiniMaxAI/MiniMax-M2", "source_type": "model_card", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "agentic", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "notes": "MiniMax M2 model card: claude-sonnet-4.5=33." } ] }, { "model_id": "gpt-5", "benchmark_id": "terminal_bench_hard", "score": 30.5, "reference_url": "https://z.ai/blog/glm-4.7", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "agentic", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "GLM-4.7 blog: gpt-5=30.5.", "candidates": [ { "score": 31, "reference_url": "https://huggingface.co/MiniMaxAI/MiniMax-M2", "source_type": "model_card", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "agentic", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "notes": "MiniMax M2 model card: gpt-5=31." } ] }, { "model_id": "gpt-5.1", "benchmark_id": "terminal_bench_hard", "score": 43.0, "reference_url": "https://z.ai/blog/glm-4.7", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "agentic", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "GLM-4.7 blog: gpt-5.1=43.0.", "candidates": [] }, { "model_id": "glm-4.6", "benchmark_id": "browsecomp", "score": 45.1, "reference_url": "https://z.ai/blog/glm-4.7", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "agentic", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "GLM-4.7 blog: glm-4.6=45.1.", "candidates": [] }, { "model_id": "glm-4.6", "benchmark_id": "browsecomp_zh", "score": 49.5, "reference_url": "https://z.ai/blog/glm-4.7", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "agentic", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "GLM-4.7 blog: glm-4.6=49.5.", "candidates": [] }, { "model_id": "claude-sonnet-4.5", "benchmark_id": "browsecomp_zh", "score": 42.4, "reference_url": "https://z.ai/blog/glm-4.7", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "agentic", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "GLM-4.7 blog: claude-sonnet-4.5=42.4.", "candidates": [ { "score": 40.8, "reference_url": "https://huggingface.co/MiniMaxAI/MiniMax-M2", "source_type": "model_card", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "agentic", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "notes": "MiniMax M2 model card: claude-sonnet-4.5=40.8." } ] }, { "model_id": "gpt-5", "benchmark_id": "browsecomp_zh", "score": 63.0, "reference_url": "https://z.ai/blog/glm-4.7", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "agentic", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "GLM-4.7 blog: gpt-5=63.0.", "candidates": [ { "score": 65, "reference_url": "https://huggingface.co/MiniMaxAI/MiniMax-M2", "source_type": "model_card", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "agentic", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "notes": "MiniMax M2 model card: gpt-5=65." } ] }, { "model_id": "kimi-k2.6", "benchmark_id": "deepsearchqa_f1", "score": 92.5, "reference_url": "https://huggingface.co/moonshotai/Kimi-K2.6", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "agentic", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "1.0" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Kimi K2.6 model card: kimi-k2.6=92.5.", "candidates": [] }, { "model_id": "gpt-5.4", "benchmark_id": "deepsearchqa_f1", "score": 78.6, "reference_url": "https://huggingface.co/moonshotai/Kimi-K2.6", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "agentic", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "1.0" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Kimi K2.6 model card: gpt-5.4=78.6.", "candidates": [] }, { "model_id": "claude-opus-4.6", "benchmark_id": "deepsearchqa_f1", "score": 91.3, "reference_url": "https://huggingface.co/moonshotai/Kimi-K2.6", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "agentic", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "1.0" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Kimi K2.6 model card: claude-opus-4.6=91.3.", "candidates": [ { "score": 82.0, "reference_url": "https://ernie.baidu.com/blog/posts/ernie-5.1-0508-release/benchmark.png", "source_type": "official_blog", "reported_setting": { "tools": "search/agentic", "metric": "F1" }, "notes": "F1 search-agent row matches the campaign benchmark identity. Research observation obs-011. Literal marker ● has no source legend. Marker has no legend in image or article." } ] }, { "model_id": "gemini-3.1-pro", "benchmark_id": "deepsearchqa_f1", "score": 81.9, "reference_url": "https://huggingface.co/moonshotai/Kimi-K2.6", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "agentic", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "1.0" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Kimi K2.6 model card: gemini-3.1-pro=81.9.", "candidates": [ { "score": 90.4, "reference_url": "https://huggingface.co/tencent/Hy3", "source_type": "model_card", "reported_setting": { "mode": "thinking", "effort": "high / highest available", "tools": "web search/browser", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "Tencent internal search harness; BrowseComp uses self-summary context management", "prompt_style": "default", "temperature": "0.9 for Hy3 family; otherwise source/provider setting", "context": "256000 for Hy3 family; otherwise source/provider setting", "notes": "Tencent own test" }, "notes": "Hy3 official model-card appendix: DeepSearchQA = 90.4*. Tencent own testing." }, { "score": 71.3, "reference_url": "https://ai.meta.com/static-resource/muse-spark-1.1-evaluation-report", "source_type": "tech_report", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "search/browser", "sampling": "pass@1 unless source row states otherwise", "harness": "Meta MSL general-capability evaluation", "temperature": "1.0 for Muse; provider/source setting for comparisons", "context": "benchmark-specific", "configuration": "default" }, "notes": "Figure 44 · DeepSearchQA F1; metric=headline_metric. Exact printed value in the general-capability summary." }, { "score": 79.3, "reference_url": "https://ernie.baidu.com/blog/posts/ernie-5.1-0508-release/benchmark.png", "source_type": "official_blog", "reported_setting": { "tools": "search/agentic", "metric": "F1" }, "notes": "F1 search-agent row matches the campaign benchmark identity. Research observation obs-012." } ] }, { "model_id": "kimi-k2.5", "benchmark_id": "deepsearchqa_f1", "score": 89.0, "reference_url": "https://huggingface.co/moonshotai/Kimi-K2.6", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "agentic", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "1.0" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Kimi K2.6 model card: kimi-k2.5=89.0.", "candidates": [] }, { "model_id": "kimi-k2.6", "benchmark_id": "deepsearchqa_acc", "score": 83.0, "reference_url": "https://huggingface.co/moonshotai/Kimi-K2.6", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "agentic", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "1.0" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Kimi K2.6 model card: kimi-k2.6=83.0.", "candidates": [] }, { "model_id": "gpt-5.4", "benchmark_id": "deepsearchqa_acc", "score": 63.7, "reference_url": "https://huggingface.co/moonshotai/Kimi-K2.6", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "agentic", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "1.0" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Kimi K2.6 model card: gpt-5.4=63.7.", "candidates": [] }, { "model_id": "claude-opus-4.6", "benchmark_id": "deepsearchqa_acc", "score": 80.6, "reference_url": "https://huggingface.co/moonshotai/Kimi-K2.6", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "agentic", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "1.0" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Kimi K2.6 model card: claude-opus-4.6=80.6.", "candidates": [] }, { "model_id": "gemini-3.1-pro", "benchmark_id": "deepsearchqa_acc", "score": 60.2, "reference_url": "https://huggingface.co/moonshotai/Kimi-K2.6", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "agentic", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "1.0" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Kimi K2.6 model card: gemini-3.1-pro=60.2.", "candidates": [] }, { "model_id": "kimi-k2.5", "benchmark_id": "deepsearchqa_acc", "score": 77.1, "reference_url": "https://huggingface.co/moonshotai/Kimi-K2.6", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "agentic", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "1.0" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Kimi K2.6 model card: kimi-k2.5=77.1.", "candidates": [] }, { "model_id": "kimi-k2.6", "benchmark_id": "widesearch", "score": 80.8, "reference_url": "https://huggingface.co/moonshotai/Kimi-K2.6", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "agentic", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "1.0" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Kimi K2.6 model card: kimi-k2.6=80.8.", "candidates": [] }, { "model_id": "kimi-k2.5", "benchmark_id": "widesearch", "score": 72.7, "reference_url": "https://huggingface.co/moonshotai/Kimi-K2.6", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "agentic", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "1.0" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Kimi K2.6 model card: kimi-k2.5=72.7.", "candidates": [] }, { "model_id": "gpt-5.4", "benchmark_id": "mcpmark", "score": 62.5, "reference_url": "https://huggingface.co/moonshotai/Kimi-K2.6", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "agentic", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "1.0" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Kimi K2.6 model card: gpt-5.4=62.5.", "candidates": [] }, { "model_id": "claude-opus-4.6", "benchmark_id": "mcpmark", "score": 56.7, "reference_url": "https://huggingface.co/moonshotai/Kimi-K2.6", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "agentic", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "1.0" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Kimi K2.6 model card: claude-opus-4.6=56.7.", "candidates": [] }, { "model_id": "gemini-3.1-pro", "benchmark_id": "mcpmark", "score": 55.9, "reference_url": "https://huggingface.co/moonshotai/Kimi-K2.6", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "agentic", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "1.0" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Kimi K2.6 model card: gemini-3.1-pro=55.9.", "candidates": [] }, { "model_id": "kimi-k2.5", "benchmark_id": "mcpmark", "score": 29.5, "reference_url": "https://huggingface.co/moonshotai/Kimi-K2.6", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "agentic", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "1.0" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Kimi K2.6 model card: kimi-k2.5=29.5.", "candidates": [] }, { "model_id": "kimi-k2.6", "benchmark_id": "apex_agents", "score": 27.9, "reference_url": "https://huggingface.co/moonshotai/Kimi-K2.6", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "agentic", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "1.0" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Kimi K2.6 model card: kimi-k2.6=27.9.", "candidates": [] }, { "model_id": "gpt-5.4", "benchmark_id": "apex_agents", "score": 33.3, "reference_url": "https://huggingface.co/moonshotai/Kimi-K2.6", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "agentic", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "1.0" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Kimi K2.6 model card: gpt-5.4=33.3.", "candidates": [] }, { "model_id": "kimi-k2.5", "benchmark_id": "apex_agents", "score": 11.5, "reference_url": "https://huggingface.co/moonshotai/Kimi-K2.6", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "agentic", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "1.0" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Kimi K2.6 model card: kimi-k2.5=11.5.", "candidates": [] }, { "model_id": "kimi-k2.6", "benchmark_id": "osworld", "score": 73.1, "reference_url": "https://huggingface.co/moonshotai/Kimi-K2.6", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "agentic", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "1.0" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Kimi K2.6 model card: kimi-k2.6=73.1.", "candidates": [] }, { "model_id": "gemini-3.1-pro", "benchmark_id": "swe_bench_multilingual", "score": 76.9, "reference_url": "https://huggingface.co/moonshotai/Kimi-K2.6", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "agentic (bash + edit)", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "1.0" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Kimi K2.6 model card: gemini-3.1-pro=76.9.", "candidates": [ { "score": 44.0, "reference_url": "https://huggingface.co/internlm/Intern-S2-Preview-397B", "source_type": "model_card", "reported_setting": { "mode": "source does not state", "effort": "source does not state", "tools": "agentic code execution", "sampling": "source does not state", "judge": "benchmark-specific; exact version not stated", "harness": "Mini-SWE-Agent", "prompt_style": "source does not state", "temperature": "source does not state", "context": "source does not state", "notes": "Shanghai AI Lab evaluated all model columns using OpenCompass, VLMEvalKit, and AgentCompass. Per-model mode/effort/sampling are not reported; only Intern-S2 maximum evaluation lengths are disclosed." }, "notes": "Intern-S2-Preview-397B official model card: SWE-Bench-Multilingual = 44.0. This is a Shanghai AI Lab cross-model evaluation; the source does not disclose this model's exact mode, effort, or sampling." } ] }, { "model_id": "gpt-5.4", "benchmark_id": "scicode", "score": 56.6, "reference_url": "https://huggingface.co/moonshotai/Kimi-K2.6", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "1.0" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Kimi K2.6 model card: gpt-5.4=56.6.", "candidates": [] }, { "model_id": "kimi-k2.6", "benchmark_id": "ojbench", "score": 60.6, "reference_url": "https://huggingface.co/moonshotai/Kimi-K2.6", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "1.0" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Kimi K2.6 model card: kimi-k2.6=60.6.", "candidates": [] }, { "model_id": "claude-opus-4.6", "benchmark_id": "ojbench", "score": 60.3, "reference_url": "https://huggingface.co/moonshotai/Kimi-K2.6", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "1.0" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Kimi K2.6 model card: claude-opus-4.6=60.3.", "candidates": [] }, { "model_id": "gemini-3.1-pro", "benchmark_id": "ojbench", "score": 70.7, "reference_url": "https://huggingface.co/moonshotai/Kimi-K2.6", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "1.0" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Kimi K2.6 model card: gemini-3.1-pro=70.7.", "candidates": [] }, { "model_id": "kimi-k2.6", "benchmark_id": "livecodebench_v6", "score": 89.6, "reference_url": "https://huggingface.co/moonshotai/Kimi-K2.6", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "1.0" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Kimi K2.6 model card: kimi-k2.6=89.6.", "candidates": [ { "score": 90.2, "reference_url": "https://research.nvidia.com/labs/nemotron/files/NVIDIA-Nemotron-3-Ultra-Technical-Report.pdf", "source_type": "tech_report", "reported_setting": { "version_metric": "v6 / pass@1", "tools": "code execution", "harness": "NeMo Gym", "sampling": "pass@1" }, "notes": "NVIDIA Nemotron 3 Ultra technical report, Final Table 10: Kimi K2.6: livecodebench_v6=90.2." }, { "score": 89.6, "reference_url": "https://microsoft.ai/wp-content/uploads/2026/06/main_20260602_2.pdf", "source_type": "official_paper", "reported_setting": { "mode": "thinking", "effort": "max (98304 generation tokens)", "tools": "none (per-bench override)", "sampling": "pass@1 (avg@10 coding, avg@3 vision)", "judge": "rule-based", "harness": "official (in-house SWE-agent-derived for SWE-Bench)", "prompt_style": "default", "temperature": "1.0", "context": "262144 (256K)", "notes": "Per Kimi K2.6 model card (HF moonshotai/Kimi-K2.6). Thinking mode, max effort, t=1.0 top_p=1.0." }, "notes": "Displayed exactly as 89.6. Official Microsoft-reported result. Comparator value is quoted from that provider's official model card/release; use the campaign canonical setting." } ] }, { "model_id": "claude-opus-4.6", "benchmark_id": "livecodebench_v6", "score": 88.8, "reference_url": "https://huggingface.co/moonshotai/Kimi-K2.6", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "1.0" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Kimi K2.6 model card: claude-opus-4.6=88.8.", "candidates": [] }, { "model_id": "gemini-3.1-pro", "benchmark_id": "livecodebench_v6", "score": 91.7, "reference_url": "https://huggingface.co/moonshotai/Kimi-K2.6", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "1.0" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Kimi K2.6 model card: gemini-3.1-pro=91.7.", "candidates": [] }, { "model_id": "kimi-k2.5", "benchmark_id": "livecodebench_v6", "score": 85.0, "reference_url": "https://huggingface.co/moonshotai/Kimi-K2.6", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "1.0" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Kimi K2.6 model card: kimi-k2.5=85.0.", "candidates": [] }, { "model_id": "gpt-5.4", "benchmark_id": "charxiv_reasoning", "score": 82.8, "reference_url": "https://huggingface.co/moonshotai/Kimi-K2.6", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "1.0" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Kimi K2.6 model card: gpt-5.4=82.8.", "candidates": [] }, { "model_id": "gemini-3.1-pro", "benchmark_id": "charxiv_reasoning", "score": 80.2, "reference_url": "https://huggingface.co/moonshotai/Kimi-K2.6", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "1.0" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Kimi K2.6 model card: gemini-3.1-pro=80.2.", "candidates": [ { "score": 81.6, "reference_url": "https://ai.meta.com/static-resource/muse-spark-1.1-evaluation-report", "source_type": "tech_report", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "Python", "sampling": "pass@1 unless source row states otherwise", "harness": "Meta MSL general-capability evaluation", "temperature": "1.0 for Muse; provider/source setting for comparisons", "context": "benchmark-specific", "configuration": "default" }, "notes": "Figure 44 · CharXiv Reasoning w/code execution; metric=headline_metric. Exact printed value in the general-capability summary." }, { "score": 83.3, "reference_url": "https://deepmind.google/models/model-cards/gemini-3-5-flash/", "source_type": "official_model_card", "reported_setting": { "mode": "thinking/reasoning", "effort": "source does not state", "tools": "none", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "Google self-computed for Gemini/GPT; provider for Claude", "temperature": "default", "snapshot": "May 2026" }, "notes": "Google Gemini 3.5 Flash May 2026 table: CharXiv Reasoning; no tools. Settings and provenance are preserved per cell." }, { "score": 83.2, "reference_url": "https://deepmind.google/models/model-cards/gemini-3-6-flash/", "source_type": "official_model_card", "reported_setting": { "mode": "thinking/reasoning", "effort": "high/default", "tools": "search + code execution", "sampling": "pass@1", "judge": "official CharXiv binary answer judge", "harness": "Google self-computed; search/code details unpublished", "temperature": "default", "snapshot": "July 2026" }, "notes": "Same CharXiv reasoning set with search and code execution." }, { "score": 83.5, "reference_url": "https://seed.bytedance.com/en/blog/seed2-1-officially-released-advancing-ai-productivity", "source_type": "official_blog", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "none", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official", "prompt_style": "official default", "temperature": "unknown" }, "notes": "Release blog EN: Vision capability chart (https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/user-upload/4xfa4mqq4yer3.png)." } ] }, { "model_id": "gpt-5.4", "benchmark_id": "mathvision", "score": 92.0, "reference_url": "https://huggingface.co/moonshotai/Kimi-K2.6", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "1.0" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Kimi K2.6 model card: gpt-5.4=92.0.", "candidates": [] }, { "model_id": "claude-opus-4.6", "benchmark_id": "mathvision", "score": 71.2, "reference_url": "https://huggingface.co/moonshotai/Kimi-K2.6", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "1.0" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Kimi K2.6 model card: claude-opus-4.6=71.2.", "candidates": [] }, { "model_id": "gpt-5.4", "benchmark_id": "babyvision", "score": 49.7, "reference_url": "https://huggingface.co/moonshotai/Kimi-K2.6", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "1.0" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Kimi K2.6 model card: gpt-5.4=49.7.", "candidates": [] }, { "model_id": "claude-opus-4.6", "benchmark_id": "babyvision", "score": 14.8, "reference_url": "https://huggingface.co/moonshotai/Kimi-K2.6", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "1.0" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Kimi K2.6 model card: claude-opus-4.6=14.8.", "candidates": [] }, { "model_id": "gemini-3.1-pro", "benchmark_id": "babyvision", "score": 51.6, "reference_url": "https://huggingface.co/moonshotai/Kimi-K2.6", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "1.0" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Kimi K2.6 model card: gemini-3.1-pro=51.6.", "candidates": [ { "score": 51.5, "reference_url": "https://ai.meta.com/static-resource/muse-spark-1.1-evaluation-report", "source_type": "tech_report", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "Python", "sampling": "pass@1 unless source row states otherwise", "harness": "Meta MSL general-capability evaluation", "temperature": "1.0 for Muse; provider/source setting for comparisons", "context": "benchmark-specific", "configuration": "default" }, "notes": "Figure 44 · BabyVision w/code execution; metric=headline_metric. Exact printed value in the general-capability summary." }, { "score": 54.4, "reference_url": "https://seed.bytedance.com/en/blog/seed2-1-officially-released-advancing-ai-productivity", "source_type": "official_blog", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "none", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official", "prompt_style": "official default", "temperature": "unknown" }, "notes": "Release blog EN: Vision capability chart (https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/user-upload/4xfa4mqq4yer3.png)." } ] }, { "model_id": "kimi-k2.5", "benchmark_id": "babyvision", "score": 36.5, "reference_url": "https://huggingface.co/moonshotai/Kimi-K2.6", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "1.0" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Kimi K2.6 model card: kimi-k2.5=36.5.", "candidates": [] }, { "model_id": "qwen3.5-397b", "benchmark_id": "supergpqa", "score": 70.4, "reference_url": "https://huggingface.co/Qwen/Qwen3.5-397B-A17B", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Qwen3.5-397B-A17B HF model card cross-model table: SuperGPQA 70.4", "candidates": [] }, { "model_id": "qwen3.5-397b", "benchmark_id": "mmlu_redux", "score": 94.9, "reference_url": "https://huggingface.co/Qwen/Qwen3.5-397B-A17B", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Qwen3.5-397B-A17B HF model card cross-model table: MMLU-Redux 94.9", "candidates": [] }, { "model_id": "qwen3.5-397b", "benchmark_id": "multichallenge", "score": 67.6, "reference_url": "https://huggingface.co/Qwen/Qwen3.5-397B-A17B", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Qwen3.5-397B-A17B HF model card cross-model table: MultiChallenge 67.6", "candidates": [ { "score": 63.9, "reference_url": "https://research.nvidia.com/labs/nemotron/files/NVIDIA-Nemotron-3-Ultra-Technical-Report.pdf", "source_type": "tech_report", "reported_setting": { "version_metric": "multi-turn score / %", "tools": "none", "harness": "NeMo Evaluator", "sampling": "8-run average" }, "notes": "NVIDIA Nemotron 3 Ultra technical report, Final Table 10: Qwen3.5-397B-A17B: multichallenge=63.9." } ] }, { "model_id": "qwen3.5-397b", "benchmark_id": "longbench_v2", "score": 63.2, "reference_url": "https://huggingface.co/Qwen/Qwen3.5-397B-A17B", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Qwen3.5-397B-A17B HF model card cross-model table: LongBench v2 63.2", "candidates": [ { "score": 68.9, "reference_url": "https://research.nvidia.com/labs/nemotron/files/NVIDIA-Nemotron-3-Ultra-Technical-Report.pdf", "source_type": "tech_report", "reported_setting": { "version_metric": "<=1M / %", "tools": "none", "harness": "NeMo Evaluator", "sampling": "pass@1" }, "notes": "NVIDIA Nemotron 3 Ultra technical report, Final Table 10: Qwen3.5-397B-A17B: longbench_v2=68.9." } ] }, { "model_id": "qwen3.5-397b", "benchmark_id": "mcpmark", "score": 46.1, "reference_url": "https://huggingface.co/Qwen/Qwen3.5-397B-A17B", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Qwen3.5-397B-A17B HF model card cross-model table: MCP-Mark 46.1", "candidates": [] }, { "model_id": "qwen3.5-397b", "benchmark_id": "toolathlon", "score": 38.3, "reference_url": "https://huggingface.co/Qwen/Qwen3.5-397B-A17B", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Qwen3.5-397B-A17B HF model card cross-model table: Tool Decathlon 38.3", "candidates": [] }, { "model_id": "qwen3.5-397b", "benchmark_id": "browsecomp_zh", "score": 70.3, "reference_url": "https://huggingface.co/Qwen/Qwen3.5-397B-A17B", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Qwen3.5-397B-A17B HF model card cross-model table: BrowseComp-zh 70.3", "candidates": [] }, { "model_id": "qwen3.5-397b", "benchmark_id": "global_piqa", "score": 89.8, "reference_url": "https://huggingface.co/Qwen/Qwen3.5-397B-A17B", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Qwen3.5-397B-A17B HF model card cross-model table: Global PIQA 89.8", "candidates": [] }, { "model_id": "qwen3.5-397b", "benchmark_id": "mmmu_pro", "score": 79.0, "reference_url": "https://huggingface.co/Qwen/Qwen3.5-397B-A17B", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Qwen3.5-397B-A17B HF model card cross-model table: MMMU-Pro 79.0", "candidates": [ { "score": 80.29, "reference_url": "https://huggingface.co/internlm/Intern-S2-Preview-397B", "source_type": "model_card", "reported_setting": { "mode": "source does not state", "effort": "source does not state", "tools": "none", "sampling": "source does not state", "judge": "benchmark-specific; exact version not stated", "harness": "VLMEvalKit or AgentCompass; exact adapter not stated", "prompt_style": "source does not state", "temperature": "source does not state", "context": "source does not state", "notes": "Shanghai AI Lab evaluated all model columns using OpenCompass, VLMEvalKit, and AgentCompass. Per-model mode/effort/sampling are not reported; only Intern-S2 maximum evaluation lengths are disclosed." }, "notes": "Intern-S2-Preview-397B official model card: MMMU Pro = 80.29. This is a Shanghai AI Lab cross-model evaluation; the source does not disclose this model's exact mode, effort, or sampling." } ] }, { "model_id": "qwen3.5-397b", "benchmark_id": "mathvista_mini", "score": 90.3, "reference_url": "https://huggingface.co/Qwen/Qwen3.5-397B-A17B", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Qwen3.5-397B-A17B HF model card cross-model table: MathVista (mini) 90.3", "candidates": [] }, { "model_id": "qwen3.5-397b", "benchmark_id": "charxiv_reasoning", "score": 80.8, "reference_url": "https://huggingface.co/Qwen/Qwen3.5-397B-A17B", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Qwen3.5-397B-A17B HF model card cross-model table: CharXiv RQ 80.8", "candidates": [] }, { "model_id": "qwen3.5-397b", "benchmark_id": "ocrbench", "score": 93.1, "reference_url": "https://huggingface.co/Qwen/Qwen3.5-397B-A17B", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Qwen3.5-397B-A17B HF model card cross-model table: OCRBench 93.1", "candidates": [] }, { "model_id": "qwen3.5-397b", "benchmark_id": "screenspot_pro", "score": 65.6, "reference_url": "https://huggingface.co/Qwen/Qwen3.5-397B-A17B", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Qwen3.5-397B-A17B HF model card cross-model table: ScreenSpot Pro 65.6", "candidates": [] }, { "model_id": "qwen3.5-397b", "benchmark_id": "video_mme", "score": 87.5, "reference_url": "https://huggingface.co/Qwen/Qwen3.5-397B-A17B", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Qwen3.5-397B-A17B HF model card cross-model table: Video-MME with subtitle 87.5", "candidates": [] }, { "model_id": "qwen3.5-397b", "benchmark_id": "video_mmmu", "score": 84.7, "reference_url": "https://huggingface.co/Qwen/Qwen3.5-397B-A17B", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Qwen3.5-397B-A17B HF model card cross-model table: Video-MMU 84.7", "candidates": [] }, { "model_id": "kimi-k2.5", "benchmark_id": "hmmt_feb_2025", "score": 95.4, "reference_url": "https://huggingface.co/moonshotai/Kimi-K2.5", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "1.0" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Kimi K2.5 model card: kimi-k2.5=95.4.", "candidates": [] }, { "model_id": "claude-opus-4.5", "benchmark_id": "hmmt_feb_2025", "score": 92.9, "reference_url": "https://huggingface.co/moonshotai/Kimi-K2.5", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "1.0" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Kimi K2.5 model card: claude-opus-4.5=92.9.", "candidates": [] }, { "model_id": "claude-opus-4.5", "benchmark_id": "charxiv_reasoning", "score": 67.2, "reference_url": "https://huggingface.co/moonshotai/Kimi-K2.5", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "1.0" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Kimi K2.5 model card: claude-opus-4.5=67.2.", "candidates": [ { "score": 68.5, "reference_url": "https://huggingface.co/Qwen/Qwen3.5-397B-A17B", "source_type": "model_card", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "notes": "Qwen3.5-397B-A17B HF model card cross-model table (Qwen self-test of 3rd-party model, mc=false): alternate measurement 68.5 (3rd-party Qwen self-test)." }, { "score": 65.5, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "source_type": "model_card", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "notes": "(merged from charxiv_rq) Doubao Seed 2.0 model card: claude-opus-4.5=65.5." } ] }, { "model_id": "gpt-5.2", "benchmark_id": "mathvision", "score": 83.0, "reference_url": "https://huggingface.co/moonshotai/Kimi-K2.5", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "1.0" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Kimi K2.5 model card: gpt-5.2=83.0.", "candidates": [ { "score": 86.8, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "source_type": "model_card", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "notes": "Doubao Seed 2.0 model card: gpt-5.2=86.8." } ] }, { "model_id": "claude-opus-4.5", "benchmark_id": "mathvision", "score": 77.1, "reference_url": "https://huggingface.co/moonshotai/Kimi-K2.5", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "1.0" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Kimi K2.5 model card: claude-opus-4.5=77.1.", "candidates": [ { "score": 74.3, "reference_url": "https://huggingface.co/Qwen/Qwen3.5-397B-A17B", "source_type": "model_card", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "notes": "Qwen3.5-397B-A17B HF model card cross-model table (Qwen self-test of 3rd-party model, mc=false): alternate measurement 74.3 (3rd-party Qwen self-test)." }, { "score": 74.3, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "source_type": "model_card", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "notes": "Doubao Seed 2.0 model card: claude-opus-4.5=74.3." } ] }, { "model_id": "gemini-3-pro", "benchmark_id": "mathvision", "score": 86.1, "reference_url": "https://huggingface.co/moonshotai/Kimi-K2.5", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "1.0" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Kimi K2.5 model card: gemini-3-pro=86.1.", "candidates": [ { "score": 86.6, "reference_url": "https://huggingface.co/Qwen/Qwen3.5-397B-A17B", "source_type": "model_card", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "notes": "Qwen3.5-397B-A17B HF model card cross-model table (Qwen self-test of 3rd-party model, mc=false): alternate measurement 86.6 (3rd-party Qwen self-test)." } ] }, { "model_id": "gpt-5.2", "benchmark_id": "mathvista_mini", "score": 82.8, "reference_url": "https://huggingface.co/moonshotai/Kimi-K2.5", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "1.0" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Kimi K2.5 model card: gpt-5.2=82.8.", "candidates": [ { "score": 83.1, "reference_url": "https://huggingface.co/Qwen/Qwen3.5-397B-A17B", "source_type": "model_card", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "notes": "Qwen3.5-397B-A17B HF model card cross-model table (Qwen self-test of 3rd-party model, mc=false): alternate measurement 83.1 (3rd-party Qwen self-test)." } ] }, { "model_id": "claude-opus-4.5", "benchmark_id": "mathvista_mini", "score": 80.2, "reference_url": "https://huggingface.co/moonshotai/Kimi-K2.5", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "1.0" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Kimi K2.5 model card: claude-opus-4.5=80.2.", "candidates": [ { "score": 80.0, "reference_url": "https://huggingface.co/Qwen/Qwen3.5-397B-A17B", "source_type": "model_card", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "notes": "Qwen3.5-397B-A17B HF model card cross-model table (Qwen self-test of 3rd-party model, mc=false): alternate measurement 80.0 (3rd-party Qwen self-test)." } ] }, { "model_id": "gemini-3-pro", "benchmark_id": "mathvista_mini", "score": 89.8, "reference_url": "https://huggingface.co/moonshotai/Kimi-K2.5", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "1.0" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Kimi K2.5 model card: gemini-3-pro=89.8.", "candidates": [ { "score": 87.9, "reference_url": "https://huggingface.co/Qwen/Qwen3.5-397B-A17B", "source_type": "model_card", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "notes": "Qwen3.5-397B-A17B HF model card cross-model table (Qwen self-test of 3rd-party model, mc=false): alternate measurement 87.9 (3rd-party Qwen self-test)." } ] }, { "model_id": "kimi-k2.5", "benchmark_id": "zerobench_main", "score": 9, "reference_url": "https://huggingface.co/moonshotai/Kimi-K2.5", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "1.0" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Kimi K2.5 model card: kimi-k2.5=9.", "candidates": [] }, { "model_id": "gpt-5.2", "benchmark_id": "ocrbench", "score": 80.7, "reference_url": "https://huggingface.co/moonshotai/Kimi-K2.5", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "1.0" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Kimi K2.5 model card: gpt-5.2=80.7.", "candidates": [] }, { "model_id": "claude-opus-4.5", "benchmark_id": "ocrbench", "score": 86.5, "reference_url": "https://huggingface.co/moonshotai/Kimi-K2.5", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "1.0" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Kimi K2.5 model card: claude-opus-4.5=86.5.", "candidates": [ { "score": 85.8, "reference_url": "https://huggingface.co/Qwen/Qwen3.5-397B-A17B", "source_type": "model_card", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "notes": "Qwen3.5-397B-A17B HF model card cross-model table (Qwen self-test of 3rd-party model, mc=false): alternate measurement 85.8 (3rd-party Qwen self-test)." } ] }, { "model_id": "gemini-3-pro", "benchmark_id": "ocrbench", "score": 90.3, "reference_url": "https://huggingface.co/moonshotai/Kimi-K2.5", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "1.0" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Kimi K2.5 model card: gemini-3-pro=90.3.", "candidates": [ { "score": 90.4, "reference_url": "https://huggingface.co/Qwen/Qwen3.5-397B-A17B", "source_type": "model_card", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "notes": "Qwen3.5-397B-A17B HF model card cross-model table (Qwen self-test of 3rd-party model, mc=false): alternate measurement 90.4 (3rd-party Qwen self-test)." } ] }, { "model_id": "kimi-k2.5", "benchmark_id": "omnidocbench_1.5", "score": 0.112, "reference_url": "https://huggingface.co/moonshotai/Kimi-K2.5", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "1.0" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Kimi K2.5 model card: kimi-k2.5=88.8. [Converted 88.8% (1-LED higher-better) → 0.112 (edit distance lower-better) for cross-source consistency]", "candidates": [] }, { "model_id": "gpt-5.2", "benchmark_id": "omnidocbench_1.5", "score": 0.143, "reference_url": "https://huggingface.co/moonshotai/Kimi-K2.5", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "1.0" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Kimi K2.5 model card: gpt-5.2=85.7. [Converted 85.7% (1-LED higher-better) → 0.143 (edit distance lower-better) for cross-source consistency]", "candidates": [] }, { "model_id": "claude-opus-4.5", "benchmark_id": "omnidocbench_1.5", "score": 0.123, "reference_url": "https://huggingface.co/moonshotai/Kimi-K2.5", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "1.0" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Kimi K2.5 model card: claude-opus-4.5=87.7. [Converted 87.7% (1-LED higher-better) → 0.123 (edit distance lower-better) for cross-source consistency]", "candidates": [ { "score": 0.153, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "source_type": "model_card", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "notes": "(merged from omnidocbench_1.5_lower) Doubao Seed 2.0 model card: claude-opus-4.5=0.153." } ] }, { "model_id": "gemini-3-pro", "benchmark_id": "omnidocbench_1.5", "score": 0.115, "reference_url": "https://huggingface.co/moonshotai/Kimi-K2.5", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "1.0" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Kimi K2.5 model card: gemini-3-pro=88.5. [Converted 88.5% (1-LED higher-better) → 0.115 (edit distance lower-better) for cross-source consistency]", "candidates": [] }, { "model_id": "kimi-k2.5", "benchmark_id": "infovqa", "score": 92.6, "reference_url": "https://huggingface.co/moonshotai/Kimi-K2.5", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "1.0" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Kimi K2.5 model card: kimi-k2.5=92.6.", "candidates": [] }, { "model_id": "gpt-5.2", "benchmark_id": "infovqa", "score": 84, "reference_url": "https://huggingface.co/moonshotai/Kimi-K2.5", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "1.0" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Kimi K2.5 model card: gpt-5.2=84.", "candidates": [] }, { "model_id": "claude-opus-4.5", "benchmark_id": "infovqa", "score": 76.9, "reference_url": "https://huggingface.co/moonshotai/Kimi-K2.5", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "1.0" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Kimi K2.5 model card: claude-opus-4.5=76.9.", "candidates": [] }, { "model_id": "gemini-3-pro", "benchmark_id": "infovqa", "score": 57.2, "reference_url": "https://huggingface.co/moonshotai/Kimi-K2.5", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "1.0" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Kimi K2.5 model card: gemini-3-pro=57.2.", "candidates": [] }, { "model_id": "kimi-k2.5", "benchmark_id": "simplevqa", "score": 71.2, "reference_url": "https://huggingface.co/moonshotai/Kimi-K2.5", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "1.0" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Kimi K2.5 model card: kimi-k2.5=71.2.", "candidates": [] }, { "model_id": "gpt-5.2", "benchmark_id": "simplevqa", "score": 55.8, "reference_url": "https://huggingface.co/moonshotai/Kimi-K2.5", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "1.0" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Kimi K2.5 model card: gpt-5.2=55.8.", "candidates": [ { "score": 54.1, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "source_type": "model_card", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "notes": "Doubao Seed 2.0 model card: gpt-5.2=54.1." } ] }, { "model_id": "claude-opus-4.5", "benchmark_id": "simplevqa", "score": 69.7, "reference_url": "https://huggingface.co/moonshotai/Kimi-K2.5", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "1.0" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Kimi K2.5 model card: claude-opus-4.5=69.7.", "candidates": [ { "score": 57.9, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "source_type": "model_card", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "notes": "Doubao Seed 2.0 model card: claude-opus-4.5=57.9." } ] }, { "model_id": "gemini-3-pro", "benchmark_id": "simplevqa", "score": 69.7, "reference_url": "https://huggingface.co/moonshotai/Kimi-K2.5", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "1.0" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Kimi K2.5 model card: gemini-3-pro=69.7.", "candidates": [] }, { "model_id": "kimi-k2.5", "benchmark_id": "worldvqa", "score": 46.3, "reference_url": "https://huggingface.co/moonshotai/Kimi-K2.5", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "1.0" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Kimi K2.5 model card: kimi-k2.5=46.3.", "candidates": [] }, { "model_id": "gpt-5.2", "benchmark_id": "worldvqa", "score": 28.0, "reference_url": "https://huggingface.co/moonshotai/Kimi-K2.5", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "1.0" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Kimi K2.5 model card: gpt-5.2=28.0.", "candidates": [ { "score": 26.3, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "source_type": "model_card", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "notes": "Doubao Seed 2.0 model card: gpt-5.2=26.3." } ] }, { "model_id": "claude-opus-4.5", "benchmark_id": "worldvqa", "score": 36.8, "reference_url": "https://huggingface.co/moonshotai/Kimi-K2.5", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "1.0" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Kimi K2.5 model card: claude-opus-4.5=36.8.", "candidates": [ { "score": 36.6, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "source_type": "model_card", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "notes": "Doubao Seed 2.0 model card: claude-opus-4.5=36.6." } ] }, { "model_id": "gemini-3-pro", "benchmark_id": "worldvqa", "score": 47.4, "reference_url": "https://huggingface.co/moonshotai/Kimi-K2.5", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "1.0" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Kimi K2.5 model card: gemini-3-pro=47.4.", "candidates": [ { "score": 47.5, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "source_type": "model_card", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "notes": "Doubao Seed 2.0 model card: gemini-3-pro=47.5." } ] }, { "model_id": "kimi-k2.5", "benchmark_id": "mmvu", "score": 80.4, "reference_url": "https://huggingface.co/moonshotai/Kimi-K2.5", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "1.0" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Kimi K2.5 model card: kimi-k2.5=80.4.", "candidates": [] }, { "model_id": "gpt-5.2", "benchmark_id": "mmvu", "score": 80.8, "reference_url": "https://huggingface.co/moonshotai/Kimi-K2.5", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "1.0" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Kimi K2.5 model card: gpt-5.2=80.8.", "candidates": [] }, { "model_id": "claude-opus-4.5", "benchmark_id": "mmvu", "score": 77.3, "reference_url": "https://huggingface.co/moonshotai/Kimi-K2.5", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "1.0" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Kimi K2.5 model card: claude-opus-4.5=77.3.", "candidates": [] }, { "model_id": "gemini-3-pro", "benchmark_id": "mmvu", "score": 77.5, "reference_url": "https://huggingface.co/moonshotai/Kimi-K2.5", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "1.0" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Kimi K2.5 model card: gemini-3-pro=77.5.", "candidates": [ { "score": 76.3, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "source_type": "model_card", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "notes": "Doubao Seed 2.0 model card: gemini-3-pro=76.3." } ] }, { "model_id": "kimi-k2.5", "benchmark_id": "motionbench", "score": 70.4, "reference_url": "https://huggingface.co/moonshotai/Kimi-K2.5", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "1.0" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Kimi K2.5 model card: kimi-k2.5=70.4.", "candidates": [] }, { "model_id": "gpt-5.2", "benchmark_id": "motionbench", "score": 64.8, "reference_url": "https://huggingface.co/moonshotai/Kimi-K2.5", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "1.0" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Kimi K2.5 model card: gpt-5.2=64.8.", "candidates": [] }, { "model_id": "claude-opus-4.5", "benchmark_id": "motionbench", "score": 60.3, "reference_url": "https://huggingface.co/moonshotai/Kimi-K2.5", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "1.0" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Kimi K2.5 model card: claude-opus-4.5=60.3.", "candidates": [] }, { "model_id": "gemini-3-pro", "benchmark_id": "motionbench", "score": 70.3, "reference_url": "https://huggingface.co/moonshotai/Kimi-K2.5", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "1.0" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Kimi K2.5 model card: gemini-3-pro=70.3.", "candidates": [] }, { "model_id": "gpt-5.2", "benchmark_id": "video_mme", "score": 86.0, "reference_url": "https://huggingface.co/moonshotai/Kimi-K2.5", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "1.0" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Kimi K2.5 model card: gpt-5.2=86.0.", "candidates": [] }, { "model_id": "gemini-3-pro", "benchmark_id": "video_mme", "score": 88.4, "reference_url": "https://huggingface.co/moonshotai/Kimi-K2.5", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "1.0" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Kimi K2.5 model card: gemini-3-pro=88.4.", "candidates": [] }, { "model_id": "kimi-k2.5", "benchmark_id": "longvideobench", "score": 79.8, "reference_url": "https://huggingface.co/moonshotai/Kimi-K2.5", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "1.0" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Kimi K2.5 model card: kimi-k2.5=79.8.", "candidates": [] }, { "model_id": "gpt-5.2", "benchmark_id": "longvideobench", "score": 76.5, "reference_url": "https://huggingface.co/moonshotai/Kimi-K2.5", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "1.0" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Kimi K2.5 model card: gpt-5.2=76.5.", "candidates": [] }, { "model_id": "claude-opus-4.5", "benchmark_id": "longvideobench", "score": 67.2, "reference_url": "https://huggingface.co/moonshotai/Kimi-K2.5", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "1.0" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Kimi K2.5 model card: claude-opus-4.5=67.2.", "candidates": [] }, { "model_id": "gemini-3-pro", "benchmark_id": "longvideobench", "score": 77.7, "reference_url": "https://huggingface.co/moonshotai/Kimi-K2.5", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "1.0" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Kimi K2.5 model card: gemini-3-pro=77.7.", "candidates": [ { "score": 76.7, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "source_type": "model_card", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "notes": "Doubao Seed 2.0 model card: gemini-3-pro=76.7." } ] }, { "model_id": "kimi-k2.5", "benchmark_id": "lvbench", "score": 75.9, "reference_url": "https://huggingface.co/moonshotai/Kimi-K2.5", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "1.0" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Kimi K2.5 model card: kimi-k2.5=75.9.", "candidates": [] }, { "model_id": "gemini-3-pro", "benchmark_id": "lvbench", "score": 73.5, "reference_url": "https://huggingface.co/moonshotai/Kimi-K2.5", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "1.0" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Kimi K2.5 model card: gemini-3-pro=73.5.", "candidates": [] }, { "model_id": "gpt-5.2", "benchmark_id": "paperbench", "score": 63.7, "reference_url": "https://huggingface.co/moonshotai/Kimi-K2.5", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "agentic", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "1.0" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Kimi K2.5 model card: gpt-5.2=63.7.", "candidates": [] }, { "model_id": "claude-opus-4.5", "benchmark_id": "paperbench", "score": 72.9, "reference_url": "https://huggingface.co/moonshotai/Kimi-K2.5", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "agentic", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "1.0" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Kimi K2.5 model card: claude-opus-4.5=72.9.", "candidates": [] }, { "model_id": "deepseek-v3.2", "benchmark_id": "paperbench", "score": 47.1, "reference_url": "https://huggingface.co/moonshotai/Kimi-K2.5", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "agentic", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "1.0" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Kimi K2.5 model card: deepseek-v3.2=47.1.", "candidates": [] }, { "model_id": "claude-opus-4.5", "benchmark_id": "scicode", "score": 49.5, "reference_url": "https://huggingface.co/moonshotai/Kimi-K2.5", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "1.0" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Kimi K2.5 model card: claude-opus-4.5=49.5.", "candidates": [ { "score": 50.0, "reference_url": "https://huggingface.co/MiniMaxAI/MiniMax-M2.5", "source_type": "model_card", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "notes": "MiniMax M2.5 model card: claude-opus-4.5=50.0." }, { "score": 52.8, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "source_type": "model_card", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "varies", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "notes": "Doubao Seed 2.0 model card: claude-opus-4.5=52.8." } ] }, { "model_id": "deepseek-v3.2", "benchmark_id": "scicode", "score": 38.9, "reference_url": "https://huggingface.co/moonshotai/Kimi-K2.5", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "1.0" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Kimi K2.5 model card: deepseek-v3.2=38.9.", "candidates": [ { "score": 37.7, "reference_url": "https://huggingface.co/moonshotai/Kimi-K2-Thinking", "source_type": "model_card", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "1.0" }, "notes": "Kimi K2-Thinking model card: deepseek-v3.2=37.7." }, { "score": 38, "reference_url": "https://huggingface.co/MiniMaxAI/MiniMax-M2", "source_type": "model_card", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "notes": "MiniMax M2 model card: deepseek-v3.2=38." } ] }, { "model_id": "claude-opus-4.5", "benchmark_id": "ojbench", "score": 54.6, "reference_url": "https://huggingface.co/moonshotai/Kimi-K2.5", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "1.0" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Kimi K2.5 model card: claude-opus-4.5=54.6.", "candidates": [] }, { "model_id": "gemini-3-pro", "benchmark_id": "ojbench", "score": 68.5, "reference_url": "https://huggingface.co/moonshotai/Kimi-K2.5", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "1.0" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Kimi K2.5 model card: gemini-3-pro=68.5.", "candidates": [] }, { "model_id": "deepseek-v3.2", "benchmark_id": "ojbench", "score": 54.7, "reference_url": "https://huggingface.co/moonshotai/Kimi-K2.5", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "1.0" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Kimi K2.5 model card: deepseek-v3.2=54.7.", "candidates": [ { "score": 38.2, "reference_url": "https://huggingface.co/moonshotai/Kimi-K2-Thinking", "source_type": "model_card", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "1.0" }, "notes": "Kimi K2-Thinking model card: deepseek-v3.2=38.2." } ] }, { "model_id": "claude-opus-4.5", "benchmark_id": "livecodebench_v6", "score": 82.2, "reference_url": "https://huggingface.co/moonshotai/Kimi-K2.5", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "1.0" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Kimi K2.5 model card: claude-opus-4.5=82.2.", "candidates": [ { "score": 84.8, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "source_type": "model_card", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "notes": "Doubao Seed 2.0 model card Table 3: claude-opus-4.5=84.8." } ] }, { "model_id": "gpt-5.2", "benchmark_id": "longbench_v2", "score": 54.5, "reference_url": "https://huggingface.co/moonshotai/Kimi-K2.5", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "1.0" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Kimi K2.5 model card: gpt-5.2=54.5.", "candidates": [ { "score": 63.2, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "source_type": "model_card", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "notes": "Doubao Seed 2.0 model card Table 3: gpt-5.2=63.2." } ] }, { "model_id": "claude-opus-4.5", "benchmark_id": "longbench_v2", "score": 64.4, "reference_url": "https://huggingface.co/moonshotai/Kimi-K2.5", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "1.0" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Kimi K2.5 model card: claude-opus-4.5=64.4.", "candidates": [ { "score": 65.0, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "source_type": "model_card", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "notes": "Doubao Seed 2.0 model card Table 3: claude-opus-4.5=65.0." } ] }, { "model_id": "gemini-3-pro", "benchmark_id": "longbench_v2", "score": 68.2, "reference_url": "https://huggingface.co/moonshotai/Kimi-K2.5", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "1.0" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Kimi K2.5 model card: gemini-3-pro=68.2.", "candidates": [ { "score": 67.4, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "source_type": "model_card", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "notes": "Doubao Seed 2.0 model card Table 3: gemini-3-pro=67.4." } ] }, { "model_id": "deepseek-v3.2", "benchmark_id": "longbench_v2", "score": 59.8, "reference_url": "https://huggingface.co/moonshotai/Kimi-K2.5", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "1.0" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Kimi K2.5 model card: deepseek-v3.2=59.8.", "candidates": [] }, { "model_id": "gemini-3-pro", "benchmark_id": "aa_lcr", "score": 65.3, "reference_url": "https://huggingface.co/moonshotai/Kimi-K2.5", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "1.0" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Kimi K2.5 model card: gemini-3-pro=65.3.", "candidates": [ { "score": 70.7, "reference_url": "https://huggingface.co/Qwen/Qwen3.5-397B-A17B", "source_type": "model_card", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "notes": "Qwen3.5-397B-A17B HF model card cross-model table (Qwen self-test of 3rd-party model, mc=false): alternate measurement 70.7 (3rd-party Qwen self-test)." }, { "score": 71.0, "reference_url": "https://huggingface.co/MiniMaxAI/MiniMax-M2.5", "source_type": "model_card", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "notes": "MiniMax M2.5 model card: gemini-3-pro=71.0." } ] }, { "model_id": "deepseek-v3.2", "benchmark_id": "aa_lcr", "score": 64.3, "reference_url": "https://huggingface.co/moonshotai/Kimi-K2.5", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "1.0" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Kimi K2.5 model card: deepseek-v3.2=64.3.", "candidates": [ { "score": 69, "reference_url": "https://huggingface.co/MiniMaxAI/MiniMax-M2", "source_type": "model_card", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "notes": "MiniMax M2 model card: deepseek-v3.2=69." } ] }, { "model_id": "claude-opus-4.5", "benchmark_id": "widesearch", "score": 76.2, "reference_url": "https://huggingface.co/moonshotai/Kimi-K2.5", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "agentic", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "1.0" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Kimi K2.5 model card: claude-opus-4.5=76.2.", "candidates": [] }, { "model_id": "gemini-3-pro", "benchmark_id": "widesearch", "score": 57.0, "reference_url": "https://huggingface.co/moonshotai/Kimi-K2.5", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "agentic", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "1.0" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Kimi K2.5 model card: gemini-3-pro=57.0.", "candidates": [ { "score": 67.3, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "source_type": "model_card", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "varies", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "notes": "Doubao Seed 2.0 model card: gemini-3-pro=67.3." } ] }, { "model_id": "gpt-5.2", "benchmark_id": "deepsearchqa_acc", "score": 71.3, "reference_url": "https://huggingface.co/moonshotai/Kimi-K2.5", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "agentic", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "1.0" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Kimi K2.5 model card: gpt-5.2=71.3.", "candidates": [] }, { "model_id": "claude-opus-4.5", "benchmark_id": "deepsearchqa_acc", "score": 76.1, "reference_url": "https://huggingface.co/moonshotai/Kimi-K2.5", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "agentic", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "1.0" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Kimi K2.5 model card: claude-opus-4.5=76.1.", "candidates": [] }, { "model_id": "gemini-3-pro", "benchmark_id": "deepsearchqa_acc", "score": 63.2, "reference_url": "https://huggingface.co/moonshotai/Kimi-K2.5", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "agentic", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "1.0" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Kimi K2.5 model card: gemini-3-pro=63.2.", "candidates": [ { "score": 63.9, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "source_type": "model_card", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "varies", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "notes": "Doubao Seed 2.0 model card: gemini-3-pro=63.9." } ] }, { "model_id": "deepseek-v3.2", "benchmark_id": "deepsearchqa_acc", "score": 60.9, "reference_url": "https://huggingface.co/moonshotai/Kimi-K2.5", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "agentic", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "1.0" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Kimi K2.5 model card: deepseek-v3.2=60.9.", "candidates": [] }, { "model_id": "kimi-k2.5", "benchmark_id": "finsearchcompt23", "score": 67.8, "reference_url": "https://huggingface.co/moonshotai/Kimi-K2.5", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "agentic", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "1.0" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Kimi K2.5 model card: kimi-k2.5=67.8.", "candidates": [] }, { "model_id": "claude-opus-4.5", "benchmark_id": "finsearchcompt23", "score": 66.2, "reference_url": "https://huggingface.co/moonshotai/Kimi-K2.5", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "agentic", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "1.0" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Kimi K2.5 model card: claude-opus-4.5=66.2.", "candidates": [] }, { "model_id": "gemini-3-pro", "benchmark_id": "finsearchcompt23", "score": 49.9, "reference_url": "https://huggingface.co/moonshotai/Kimi-K2.5", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "agentic", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "1.0" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Kimi K2.5 model card: gemini-3-pro=49.9.", "candidates": [] }, { "model_id": "deepseek-v3.2", "benchmark_id": "finsearchcompt23", "score": 59.1, "reference_url": "https://huggingface.co/moonshotai/Kimi-K2.5", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "agentic", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "1.0" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Kimi K2.5 model card: deepseek-v3.2=59.1.", "candidates": [] }, { "model_id": "kimi-k2.5", "benchmark_id": "seal_0", "score": 57.4, "reference_url": "https://huggingface.co/moonshotai/Kimi-K2.5", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "agentic", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "1.0" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Kimi K2.5 model card: kimi-k2.5=57.4.", "candidates": [] }, { "model_id": "gpt-5.2", "benchmark_id": "seal_0", "score": 45.0, "reference_url": "https://huggingface.co/moonshotai/Kimi-K2.5", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "agentic", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "1.0" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Kimi K2.5 model card: gpt-5.2=45.0.", "candidates": [ { "score": 51.4, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "source_type": "model_card", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "varies", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "notes": "Doubao Seed 2.0 model card: gpt-5.2=51.4." } ] }, { "model_id": "claude-opus-4.5", "benchmark_id": "seal_0", "score": 47.7, "reference_url": "https://huggingface.co/moonshotai/Kimi-K2.5", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "agentic", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "1.0" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Kimi K2.5 model card: claude-opus-4.5=47.7.", "candidates": [] }, { "model_id": "gemini-3-pro", "benchmark_id": "seal_0", "score": 45.5, "reference_url": "https://huggingface.co/moonshotai/Kimi-K2.5", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "agentic", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "1.0" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Kimi K2.5 model card: gemini-3-pro=45.5.", "candidates": [] }, { "model_id": "deepseek-v3.2", "benchmark_id": "seal_0", "score": 49.5, "reference_url": "https://huggingface.co/moonshotai/Kimi-K2.5", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "agentic", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "1.0" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Kimi K2.5 model card: deepseek-v3.2=49.5.", "candidates": [] }, { "model_id": "gpt-5.2", "benchmark_id": "mmlu_redux", "score": 95.0, "reference_url": "https://huggingface.co/Qwen/Qwen3.5-397B-A17B", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Qwen3.5-397B-A17B HF model card cross-model table (Qwen self-test of 3rd-party model, mc=false): mmlu_redux 95.0 (3rd-party Qwen self-test).", "candidates": [] }, { "model_id": "claude-opus-4.5", "benchmark_id": "mmlu_redux", "score": 95.6, "reference_url": "https://huggingface.co/Qwen/Qwen3.5-397B-A17B", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Qwen3.5-397B-A17B HF model card cross-model table (Qwen self-test of 3rd-party model, mc=false): mmlu_redux 95.6 (3rd-party Qwen self-test).", "candidates": [] }, { "model_id": "gemini-3-pro", "benchmark_id": "mmlu_redux", "score": 95.9, "reference_url": "https://huggingface.co/Qwen/Qwen3.5-397B-A17B", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Qwen3.5-397B-A17B HF model card cross-model table (Qwen self-test of 3rd-party model, mc=false): mmlu_redux 95.9 (3rd-party Qwen self-test).", "candidates": [] }, { "model_id": "kimi-k2.5", "benchmark_id": "mmlu_redux", "score": 94.5, "reference_url": "https://huggingface.co/Qwen/Qwen3.5-397B-A17B", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Qwen3.5-397B-A17B HF model card cross-model table (Qwen self-test of 3rd-party model, mc=false): mmlu_redux 94.5 (3rd-party Qwen self-test).", "candidates": [] }, { "model_id": "gpt-5.2", "benchmark_id": "supergpqa", "score": 67.9, "reference_url": "https://huggingface.co/Qwen/Qwen3.5-397B-A17B", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Qwen3.5-397B-A17B HF model card cross-model table (Qwen self-test of 3rd-party model, mc=false): supergpqa 67.9 (3rd-party Qwen self-test).", "candidates": [] }, { "model_id": "claude-opus-4.5", "benchmark_id": "supergpqa", "score": 70.6, "reference_url": "https://huggingface.co/Qwen/Qwen3.5-397B-A17B", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Qwen3.5-397B-A17B HF model card cross-model table (Qwen self-test of 3rd-party model, mc=false): supergpqa 70.6 (3rd-party Qwen self-test).", "candidates": [] }, { "model_id": "gemini-3-pro", "benchmark_id": "supergpqa", "score": 74.0, "reference_url": "https://huggingface.co/Qwen/Qwen3.5-397B-A17B", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Qwen3.5-397B-A17B HF model card cross-model table (Qwen self-test of 3rd-party model, mc=false): supergpqa 74.0 (3rd-party Qwen self-test).", "candidates": [ { "score": 73.8, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "source_type": "model_card", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "notes": "Doubao Seed 2.0 model card Table 3: gemini-3-pro=73.8." } ] }, { "model_id": "kimi-k2.5", "benchmark_id": "supergpqa", "score": 69.2, "reference_url": "https://huggingface.co/Qwen/Qwen3.5-397B-A17B", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Qwen3.5-397B-A17B HF model card cross-model table (Qwen self-test of 3rd-party model, mc=false): supergpqa 69.2 (3rd-party Qwen self-test).", "candidates": [] }, { "model_id": "gpt-5.2", "benchmark_id": "ifbench", "score": 75.4, "reference_url": "https://huggingface.co/Qwen/Qwen3.5-397B-A17B", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Qwen3.5-397B-A17B HF model card cross-model table (Qwen self-test of 3rd-party model, mc=false): ifbench 75.4 (3rd-party Qwen self-test).", "candidates": [ { "score": 75.0, "reference_url": "https://huggingface.co/MiniMaxAI/MiniMax-M2.5", "source_type": "model_card", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "notes": "MiniMax M2.5 model card: gpt-5.2=75.0." } ] }, { "model_id": "claude-opus-4.5", "benchmark_id": "ifbench", "score": 58.0, "reference_url": "https://huggingface.co/Qwen/Qwen3.5-397B-A17B", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Qwen3.5-397B-A17B HF model card cross-model table (Qwen self-test of 3rd-party model, mc=false): ifbench 58.0 (3rd-party Qwen self-test).", "candidates": [] }, { "model_id": "gemini-3-pro", "benchmark_id": "ifbench", "score": 70.4, "reference_url": "https://huggingface.co/Qwen/Qwen3.5-397B-A17B", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Qwen3.5-397B-A17B HF model card cross-model table (Qwen self-test of 3rd-party model, mc=false): ifbench 70.4 (3rd-party Qwen self-test).", "candidates": [ { "score": 70.0, "reference_url": "https://huggingface.co/MiniMaxAI/MiniMax-M2.5", "source_type": "model_card", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "notes": "MiniMax M2.5 model card: gemini-3-pro=70.0." } ] }, { "model_id": "kimi-k2.5", "benchmark_id": "ifbench", "score": 70.2, "reference_url": "https://huggingface.co/Qwen/Qwen3.5-397B-A17B", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Qwen3.5-397B-A17B HF model card cross-model table (Qwen self-test of 3rd-party model, mc=false): ifbench 70.2 (3rd-party Qwen self-test).", "candidates": [ { "score": 70.1, "reference_url": "https://huggingface.co/mistralai/Mistral-Medium-3.5-128B/resolve/22b2b868a15677cfa6061277ed2f653d1349a9ab/images/image1.png?download=true", "source_type": "official_model_card", "reported_setting": { "mode": "thinking", "effort": "default (maximum reasoning setting)", "tools": "none", "sampling": "pass@1; repeat count undisclosed", "judge": "benchmark-specified", "harness": "Mistral official maximum-reasoning comparison", "prompt_style": "default", "temperature": "1.0", "context": "256K" }, "notes": "Displayed exactly: '70.1'. Research observation: medium-image1.png:2:4:score:reported. Chart footnotes: *All model evaluations were run with maximum reasoning settings. | **Sonnet 4.6 encountered higher rates of reasoning truncation due to external API restrictions than other models, which affected its performance." } ] }, { "model_id": "gpt-5.2", "benchmark_id": "multichallenge", "score": 57.9, "reference_url": "https://huggingface.co/Qwen/Qwen3.5-397B-A17B", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Qwen3.5-397B-A17B HF model card cross-model table (Qwen self-test of 3rd-party model, mc=false): multichallenge 57.9 (3rd-party Qwen self-test).", "candidates": [ { "score": 59.5, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "source_type": "model_card", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "notes": "Doubao Seed 2.0 model card Table 3: gpt-5.2=59.5." } ] }, { "model_id": "claude-opus-4.5", "benchmark_id": "multichallenge", "score": 54.2, "reference_url": "https://huggingface.co/Qwen/Qwen3.5-397B-A17B", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Qwen3.5-397B-A17B HF model card cross-model table (Qwen self-test of 3rd-party model, mc=false): multichallenge 54.2 (3rd-party Qwen self-test).", "candidates": [ { "score": 59.0, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "source_type": "model_card", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "notes": "Doubao Seed 2.0 model card Table 3: claude-opus-4.5=59.0." } ] }, { "model_id": "gemini-3-pro", "benchmark_id": "multichallenge", "score": 64.2, "reference_url": "https://huggingface.co/Qwen/Qwen3.5-397B-A17B", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Qwen3.5-397B-A17B HF model card cross-model table (Qwen self-test of 3rd-party model, mc=false): multichallenge 64.2 (3rd-party Qwen self-test).", "candidates": [ { "score": 68.7, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "source_type": "model_card", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "notes": "Doubao Seed 2.0 model card Table 3: gemini-3-pro=68.7." } ] }, { "model_id": "kimi-k2.5", "benchmark_id": "multichallenge", "score": 62.7, "reference_url": "https://huggingface.co/Qwen/Qwen3.5-397B-A17B", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Qwen3.5-397B-A17B HF model card cross-model table (Qwen self-test of 3rd-party model, mc=false): multichallenge 62.7 (3rd-party Qwen self-test).", "candidates": [] }, { "model_id": "gpt-5.2", "benchmark_id": "aime_2026", "score": 96.7, "reference_url": "https://huggingface.co/Qwen/Qwen3.5-397B-A17B", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Qwen3.5-397B-A17B HF model card cross-model table (Qwen self-test of 3rd-party model, mc=false): aime_2026 96.7 (3rd-party Qwen self-test).", "candidates": [ { "score": 97.5, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "source_type": "model_card", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "notes": "Doubao Seed 2.0 model card Table 3: gpt-5.2=97.5." } ] }, { "model_id": "claude-opus-4.5", "benchmark_id": "aime_2026", "score": 93.3, "reference_url": "https://huggingface.co/Qwen/Qwen3.5-397B-A17B", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Qwen3.5-397B-A17B HF model card cross-model table (Qwen self-test of 3rd-party model, mc=false): aime_2026 93.3 (3rd-party Qwen self-test).", "candidates": [ { "score": 92.5, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "source_type": "model_card", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "notes": "Doubao Seed 2.0 model card Table 3: claude-opus-4.5=92.5." } ] }, { "model_id": "gemini-3-pro", "benchmark_id": "aime_2026", "score": 90.6, "reference_url": "https://huggingface.co/Qwen/Qwen3.5-397B-A17B", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Qwen3.5-397B-A17B HF model card cross-model table (Qwen self-test of 3rd-party model, mc=false): aime_2026 90.6 (3rd-party Qwen self-test).", "candidates": [ { "score": 93.3, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "source_type": "model_card", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "notes": "Doubao Seed 2.0 model card Table 3: gemini-3-pro=93.3." } ] }, { "model_id": "kimi-k2.5", "benchmark_id": "mmmlu", "score": 86.0, "reference_url": "https://huggingface.co/Qwen/Qwen3.5-397B-A17B", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Qwen3.5-397B-A17B HF model card cross-model table (Qwen self-test of 3rd-party model, mc=false): mmmlu 86.0 (3rd-party Qwen self-test).", "candidates": [] }, { "model_id": "claude-opus-4.5", "benchmark_id": "global_piqa", "score": 91.6, "reference_url": "https://huggingface.co/Qwen/Qwen3.5-397B-A17B", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Qwen3.5-397B-A17B HF model card cross-model table (Qwen self-test of 3rd-party model, mc=false): global_piqa 91.6 (3rd-party Qwen self-test).", "candidates": [ { "score": 93.9, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "source_type": "model_card", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "notes": "Doubao Seed 2.0 model card Table 3: claude-opus-4.5=93.9." } ] }, { "model_id": "kimi-k2.5", "benchmark_id": "global_piqa", "score": 89.3, "reference_url": "https://huggingface.co/Qwen/Qwen3.5-397B-A17B", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Qwen3.5-397B-A17B HF model card cross-model table (Qwen self-test of 3rd-party model, mc=false): global_piqa 89.3 (3rd-party Qwen self-test).", "candidates": [] }, { "model_id": "gpt-5.2", "benchmark_id": "mcpmark", "score": 57.5, "reference_url": "https://huggingface.co/Qwen/Qwen3.5-397B-A17B", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Qwen3.5-397B-A17B HF model card cross-model table (Qwen self-test of 3rd-party model, mc=false): mcpmark 57.5 (3rd-party Qwen self-test).", "candidates": [] }, { "model_id": "claude-opus-4.5", "benchmark_id": "mcpmark", "score": 42.3, "reference_url": "https://huggingface.co/Qwen/Qwen3.5-397B-A17B", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Qwen3.5-397B-A17B HF model card cross-model table (Qwen self-test of 3rd-party model, mc=false): mcpmark 42.3 (3rd-party Qwen self-test).", "candidates": [] }, { "model_id": "gemini-3-pro", "benchmark_id": "mcpmark", "score": 53.9, "reference_url": "https://huggingface.co/Qwen/Qwen3.5-397B-A17B", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Qwen3.5-397B-A17B HF model card cross-model table (Qwen self-test of 3rd-party model, mc=false): mcpmark 53.9 (3rd-party Qwen self-test).", "candidates": [ { "score": 43.1, "reference_url": "https://huggingface.co/deepseek-ai/DeepSeek-V3.2/resolve/main/assets/paper.pdf", "source_type": "tech_report", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "agentic", "sampling": "pass@1", "judge": "rule-based", "harness": "internal", "prompt_style": "default", "temperature": "1.0", "context": "128K" }, "notes": "DS V3.2 tech report Table 2: 43.1. DS uses internal MCP environment (non-official). Primary (53.9) from Qwen3.5 model card; large gap likely harness difference." } ] }, { "model_id": "claude-opus-4.5", "benchmark_id": "screenspot_pro", "score": 45.7, "reference_url": "https://huggingface.co/Qwen/Qwen3.5-397B-A17B", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Qwen3.5-397B-A17B HF model card cross-model table (Qwen self-test of 3rd-party model, mc=false): screenspot_pro 45.7 (3rd-party Qwen self-test).", "candidates": [] }, { "model_id": "claude-opus-4.5", "benchmark_id": "video_mme", "score": 77.6, "reference_url": "https://huggingface.co/Qwen/Qwen3.5-397B-A17B", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Qwen3.5-397B-A17B HF model card cross-model table (Qwen self-test of 3rd-party model, mc=false): video_mme 77.6 (3rd-party Qwen self-test).", "candidates": [] }, { "model_id": "kimi-k2", "benchmark_id": "hmmt_feb_2025", "score": 38.8, "reference_url": "https://huggingface.co/moonshotai/Kimi-K2-Thinking", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "1.0" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Kimi K2-Thinking model card: kimi-k2=38.8.", "candidates": [ { "score": 38.8, "reference_url": "https://llm-stats.com/benchmarks/hmmt-2025", "source_type": "third_party_aggregator", "reported_setting": { "variant": "instruct", "snapshot": "0905" }, "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=Kimi K2-Instruct-0905, slug=kimi-k2-instruct-0905, provider=Moonshot AI" }, { "score": 38.8, "reference_url": "https://llm-stats.com/benchmarks/hmmt-2025", "source_type": "third_party_aggregator", "reported_setting": { "variant": "instruct (BP kimi-k2 is base)" }, "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=Kimi K2 Instruct, slug=kimi-k2-instruct, provider=Moonshot AI" } ] }, { "model_id": "grok-4", "benchmark_id": "hmmt_feb_2025", "score": 90.0, "reference_url": "https://huggingface.co/moonshotai/Kimi-K2-Thinking", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "1.0" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Kimi K2-Thinking model card: grok-4=90.0.", "candidates": [] }, { "model_id": "kimi-k2", "benchmark_id": "imo_answerbench", "score": 45.8, "reference_url": "https://huggingface.co/moonshotai/Kimi-K2-Thinking", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "1.0" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Kimi K2-Thinking model card: kimi-k2=45.8.", "candidates": [] }, { "model_id": "grok-4", "benchmark_id": "imo_answerbench", "score": 73.1, "reference_url": "https://huggingface.co/moonshotai/Kimi-K2-Thinking", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "1.0" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Kimi K2-Thinking model card: grok-4=73.1.", "candidates": [] }, { "model_id": "gpt-5", "benchmark_id": "mmlu_redux", "score": 95.3, "reference_url": "https://huggingface.co/moonshotai/Kimi-K2-Thinking", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "1.0" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Kimi K2-Thinking model card: gpt-5=95.3.", "candidates": [] }, { "model_id": "claude-sonnet-4.5", "benchmark_id": "mmlu_redux", "score": 95.6, "reference_url": "https://huggingface.co/moonshotai/Kimi-K2-Thinking", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "1.0" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Kimi K2-Thinking model card: claude-sonnet-4.5=95.6.", "candidates": [] }, { "model_id": "deepseek-v3.2", "benchmark_id": "mmlu_redux", "score": 93.7, "reference_url": "https://huggingface.co/moonshotai/Kimi-K2-Thinking", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "1.0" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Kimi K2-Thinking model card: deepseek-v3.2=93.7.", "candidates": [] }, { "model_id": "kimi-k2-thinking", "benchmark_id": "longform_writing", "score": 73.8, "reference_url": "https://huggingface.co/moonshotai/Kimi-K2-Thinking", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "1.0" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Kimi K2-Thinking model card: kimi-k2-thinking=73.8.", "candidates": [] }, { "model_id": "gpt-5", "benchmark_id": "longform_writing", "score": 71.4, "reference_url": "https://huggingface.co/moonshotai/Kimi-K2-Thinking", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "1.0" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Kimi K2-Thinking model card: gpt-5=71.4.", "candidates": [] }, { "model_id": "claude-sonnet-4.5", "benchmark_id": "longform_writing", "score": 79.8, "reference_url": "https://huggingface.co/moonshotai/Kimi-K2-Thinking", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "1.0" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Kimi K2-Thinking model card: claude-sonnet-4.5=79.8.", "candidates": [] }, { "model_id": "kimi-k2", "benchmark_id": "longform_writing", "score": 62.8, "reference_url": "https://huggingface.co/moonshotai/Kimi-K2-Thinking", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "1.0" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Kimi K2-Thinking model card: kimi-k2=62.8.", "candidates": [] }, { "model_id": "deepseek-v3.2", "benchmark_id": "longform_writing", "score": 72.5, "reference_url": "https://huggingface.co/moonshotai/Kimi-K2-Thinking", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "1.0" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Kimi K2-Thinking model card: deepseek-v3.2=72.5.", "candidates": [] }, { "model_id": "kimi-k2-thinking", "benchmark_id": "healthbench", "score": 58.0, "reference_url": "https://huggingface.co/moonshotai/Kimi-K2-Thinking", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "1.0" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Kimi K2-Thinking model card: kimi-k2-thinking=58.0.", "candidates": [] }, { "model_id": "gpt-5", "benchmark_id": "healthbench", "score": 67.2, "reference_url": "https://huggingface.co/moonshotai/Kimi-K2-Thinking", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "1.0" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Kimi K2-Thinking model card: gpt-5=67.2.", "candidates": [ { "score": 63.1, "reference_url": "https://deploymentsafety.openai.com/gpt-5-6", "source_type": "model_card", "reported_setting": { "effort": "source does not state", "tools": "none", "harness": "OpenAI official HealthBench scoring" }, "notes": "OpenAI GPT-5.6 System Card table (HealthBench length-adjusted scores): gpt-5; HealthBench unadjusted=63.1. System Card Table 6. Adjusted and unadjusted metrics are kept as distinct benchmark identities." } ] }, { "model_id": "claude-sonnet-4.5", "benchmark_id": "healthbench", "score": 44.2, "reference_url": "https://huggingface.co/moonshotai/Kimi-K2-Thinking", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "1.0" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Kimi K2-Thinking model card: claude-sonnet-4.5=44.2.", "candidates": [ { "score": 28.7, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "source_type": "model_card", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "notes": "Doubao Seed 2.0 model card Table 3: claude-sonnet-4.5=28.7." } ] }, { "model_id": "kimi-k2", "benchmark_id": "healthbench", "score": 43.8, "reference_url": "https://huggingface.co/moonshotai/Kimi-K2-Thinking", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "1.0" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Kimi K2-Thinking model card: kimi-k2=43.8.", "candidates": [] }, { "model_id": "deepseek-v3.2", "benchmark_id": "healthbench", "score": 46.9, "reference_url": "https://huggingface.co/moonshotai/Kimi-K2-Thinking", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "1.0" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Kimi K2-Thinking model card: deepseek-v3.2=46.9.", "candidates": [] }, { "model_id": "gpt-5", "benchmark_id": "scicode", "score": 42.9, "reference_url": "https://huggingface.co/moonshotai/Kimi-K2-Thinking", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "1.0" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Kimi K2-Thinking model card: gpt-5=42.9.", "candidates": [ { "score": 43, "reference_url": "https://huggingface.co/MiniMaxAI/MiniMax-M2", "source_type": "model_card", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "notes": "MiniMax M2 model card: gpt-5=43." } ] }, { "model_id": "claude-sonnet-4.5", "benchmark_id": "scicode", "score": 44.7, "reference_url": "https://huggingface.co/moonshotai/Kimi-K2-Thinking", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "1.0" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Kimi K2-Thinking model card: claude-sonnet-4.5=44.7.", "candidates": [ { "score": 45.0, "reference_url": "https://huggingface.co/MiniMaxAI/MiniMax-M2.5", "source_type": "model_card", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "notes": "MiniMax M2.5 model card: claude-sonnet-4.5=45.0." }, { "score": 45, "reference_url": "https://huggingface.co/MiniMaxAI/MiniMax-M2", "source_type": "model_card", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "notes": "MiniMax M2 model card: claude-sonnet-4.5=45." }, { "score": 47.9, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "source_type": "model_card", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "varies", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "notes": "Doubao Seed 2.0 model card: claude-sonnet-4.5=47.9." } ] }, { "model_id": "kimi-k2", "benchmark_id": "scicode", "score": 30.7, "reference_url": "https://huggingface.co/moonshotai/Kimi-K2-Thinking", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "1.0" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Kimi K2-Thinking model card: kimi-k2=30.7.", "candidates": [ { "score": 31, "reference_url": "https://huggingface.co/MiniMaxAI/MiniMax-M2", "source_type": "model_card", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "notes": "MiniMax M2 model card: kimi-k2=31." } ] }, { "model_id": "kimi-k2", "benchmark_id": "livecodebench_v6", "score": 56.1, "reference_url": "https://huggingface.co/moonshotai/Kimi-K2-Thinking", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "1.0" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Kimi K2-Thinking model card: kimi-k2=56.1.", "candidates": [ { "score": 53.7, "reference_url": "https://huggingface.co/moonshotai/Kimi-K2-Instruct", "source_type": "model_card", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "notes": "Kimi K2-Instruct model card: 53.7." }, { "score": 53.7, "reference_url": "https://llm-stats.com/benchmarks/livecodebench-v6", "source_type": "third_party_aggregator", "reported_setting": { "variant": "instruct (BP kimi-k2 is base)" }, "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=Kimi K2 Instruct, slug=kimi-k2-instruct, provider=Moonshot AI" }, { "score": 26.3, "reference_url": "https://huggingface.co/XiaomiMiMo/MiMo-V2.5-Pro/resolve/21d1ecfecd7bd70f31be25ca49d7edd21f003659/README.md", "source_type": "official_model_card_base_checkpoint", "reported_setting": { "mode": "base pretrained checkpoint", "effort": "not applicable", "tools": "none", "sampling": "1-shot; single completion per prompt", "judge": "benchmark-native rule-based evaluator", "harness": "Xiaomi MiMo base-checkpoint evaluation", "prompt_style": "1-shot few-shot continuation", "temperature": "source does not state", "context": "source does not state", "input_modalities": "text", "trials": "one completion per item", "checkpoint": "base" }, "notes": "Base checkpoint observation attached to the canonical post-trained model ID. Base result is attached to the canonical post-trained model ID. The Xiaomi table does not state this comparator Base checkpoint's context length, and no comparator Base config is locked; no Pro context value or conflict is copied onto this observation." } ] }, { "model_id": "gpt-5", "benchmark_id": "ojbench", "score": 56.2, "reference_url": "https://huggingface.co/moonshotai/Kimi-K2-Thinking", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "1.0" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Kimi K2-Thinking model card: gpt-5=56.2.", "candidates": [] }, { "model_id": "claude-sonnet-4.5", "benchmark_id": "ojbench", "score": 30.4, "reference_url": "https://huggingface.co/moonshotai/Kimi-K2-Thinking", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "1.0" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Kimi K2-Thinking model card: claude-sonnet-4.5=30.4.", "candidates": [] }, { "model_id": "kimi-k2", "benchmark_id": "ojbench", "score": 25.5, "reference_url": "https://huggingface.co/moonshotai/Kimi-K2-Thinking", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "1.0" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Kimi K2-Thinking model card: kimi-k2=25.5.", "candidates": [ { "score": 27.1, "reference_url": "https://huggingface.co/moonshotai/Kimi-K2-Instruct", "source_type": "model_card", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "notes": "Kimi K2-Instruct model card: 27.1." }, { "score": 27.1, "reference_url": "https://llm-stats.com/benchmarks/ojbench", "source_type": "third_party_aggregator", "reported_setting": { "variant": "instruct", "snapshot": "0905" }, "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=Kimi K2-Instruct-0905, slug=kimi-k2-instruct-0905, provider=Moonshot AI" }, { "score": 27.1, "reference_url": "https://llm-stats.com/benchmarks/ojbench", "source_type": "third_party_aggregator", "reported_setting": { "variant": "instruct (BP kimi-k2 is base)" }, "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=Kimi K2 Instruct, slug=kimi-k2-instruct, provider=Moonshot AI" } ] }, { "model_id": "gpt-oss-120b", "benchmark_id": "aider_polyglot_diff", "score": 44.4, "reference_url": "https://arxiv.org/abs/2508.10925", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "with tools", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "harmony", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "gpt-oss model card Table 3 (high effort): Aider Polyglot 44.4 (high)", "candidates": [] }, { "model_id": "gpt-oss-120b", "benchmark_id": "mmmlu", "score": 81.3, "reference_url": "https://arxiv.org/abs/2508.10925", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "with tools", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "harmony", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "gpt-oss model card Table 3 (high effort): MMMLU Average 81.3", "candidates": [ { "score": 83.8, "reference_url": "https://llm-stats.com/benchmarks/mmmlu", "source_type": "third_party_aggregator", "reported_setting": { "effort": "high" }, "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=GPT OSS 120B High, slug=gpt-oss-120b-high, provider=OpenAI" } ] }, { "model_id": "gpt-oss-120b", "benchmark_id": "healthbench", "score": 57.6, "reference_url": "https://arxiv.org/abs/2508.10925", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "with tools", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "harmony", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "gpt-oss model card Table 3 (high effort): HealthBench 57.6", "candidates": [] }, { "model_id": "gpt-oss-120b", "benchmark_id": "healthbench_hard", "score": 30.0, "reference_url": "https://arxiv.org/abs/2508.10925", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "with tools", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "harmony", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "gpt-oss model card Table 3 (high effort): HealthBench Hard 30.0", "candidates": [] }, { "model_id": "gpt-oss-120b", "benchmark_id": "healthbench_consensus", "score": 89.9, "reference_url": "https://arxiv.org/abs/2508.10925", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "with tools", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "harmony", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "gpt-oss model card Table 3 (high effort): HealthBench Consensus 89.9", "candidates": [] }, { "model_id": "gpt-oss-120b", "benchmark_id": "hle", "score": 19.0, "reference_url": "https://arxiv.org/abs/2508.10925", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "with tools", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "harmony", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "gpt-oss model card Table 3 (high effort): HLE with-tools 19.0", "candidates": [] }, { "model_id": "gpt-oss-120b", "benchmark_id": "tau_bench_airline", "score": 49.2, "reference_url": "https://arxiv.org/abs/2508.10925", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "with tools", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "harmony", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "gpt-oss model card Table 3 (high effort): Tau-Bench Airline with-tools 49.2", "candidates": [] }, { "model_id": "gpt-oss-20b", "benchmark_id": "aider_polyglot_diff", "score": 34.2, "reference_url": "https://arxiv.org/abs/2508.10925", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "with tools", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "harmony", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "gpt-oss model card Table 3 (high effort): Aider Polyglot 34.2 (high)", "candidates": [] }, { "model_id": "gpt-oss-20b", "benchmark_id": "mmmlu", "score": 75.7, "reference_url": "https://arxiv.org/abs/2508.10925", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "with tools", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "harmony", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "gpt-oss model card Table 3 (high effort): MMMLU Average 75.7", "candidates": [] }, { "model_id": "gpt-oss-20b", "benchmark_id": "healthbench", "score": 42.5, "reference_url": "https://arxiv.org/abs/2508.10925", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "with tools", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "harmony", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "gpt-oss model card Table 3 (high effort): HealthBench 42.5", "candidates": [] }, { "model_id": "gpt-oss-20b", "benchmark_id": "healthbench_hard", "score": 10.8, "reference_url": "https://arxiv.org/abs/2508.10925", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "with tools", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "harmony", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "gpt-oss model card Table 3 (high effort): HealthBench Hard 10.8", "candidates": [] }, { "model_id": "gpt-oss-20b", "benchmark_id": "healthbench_consensus", "score": 82.6, "reference_url": "https://arxiv.org/abs/2508.10925", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "with tools", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "harmony", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "gpt-oss model card Table 3 (high effort): HealthBench Consensus 82.6", "candidates": [] }, { "model_id": "gpt-oss-20b", "benchmark_id": "hle", "score": 17.3, "reference_url": "https://arxiv.org/abs/2508.10925", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "with tools", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "harmony", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "gpt-oss model card Table 3 (high effort): HLE with-tools 17.3", "candidates": [] }, { "model_id": "gpt-oss-20b", "benchmark_id": "tau_bench_retail", "score": 54.8, "reference_url": "https://arxiv.org/abs/2508.10925", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "with tools", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "harmony", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "gpt-oss model card Table 3 (high effort): Tau-Bench Retail with-tools 54.8", "candidates": [] }, { "model_id": "gpt-oss-20b", "benchmark_id": "tau_bench_airline", "score": 38.0, "reference_url": "https://arxiv.org/abs/2508.10925", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "with tools", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "harmony", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "gpt-oss model card Table 3 (high effort): Tau-Bench Airline with-tools 38.0", "candidates": [] }, { "model_id": "kimi-k2", "benchmark_id": "aider_polyglot_diff", "score": 60.0, "reference_url": "https://huggingface.co/moonshotai/Kimi-K2-Instruct", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "agentic (bash + edit)", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Kimi K2-Instruct model card: 60.0.", "candidates": [ { "score": 60.0, "reference_url": "https://arxiv.org/abs/2507.20534", "source_type": "tech_report", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none (per-bench override)", "sampling": "pass@1 (avg of 4-64 trials per benchmark)", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "notes": "(merged from aider_polyglot) Kimi K2 tech report (arxiv:2507.20534) Table 3 Kimi-K2-Instruct column: Aider-Polyglot Acc 60.0" }, { "score": 60.0, "reference_url": "https://llm-stats.com/benchmarks/aider-polyglot", "source_type": "third_party_aggregator", "reported_setting": { "variant": "instruct", "snapshot": "0905" }, "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=Kimi K2-Instruct-0905, slug=kimi-k2-instruct-0905, provider=Moonshot AI" }, { "score": 60.0, "reference_url": "https://llm-stats.com/benchmarks/aider-polyglot", "source_type": "third_party_aggregator", "reported_setting": { "variant": "instruct (BP kimi-k2 is base)" }, "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=Kimi K2 Instruct, slug=kimi-k2-instruct, provider=Moonshot AI" } ] }, { "model_id": "kimi-k2", "benchmark_id": "cnmo_2024", "score": 74.3, "reference_url": "https://huggingface.co/moonshotai/Kimi-K2-Instruct", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Kimi K2-Instruct model card: 74.3.", "candidates": [ { "score": 74.3, "reference_url": "https://llm-stats.com/benchmarks/cnmo-2024", "source_type": "third_party_aggregator", "reported_setting": { "variant": "instruct (BP kimi-k2 is base)" }, "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=Kimi K2 Instruct, slug=kimi-k2-instruct, provider=Moonshot AI" }, { "score": 74.3, "reference_url": "https://llm-stats.com/benchmarks/cnmo-2024", "source_type": "third_party_aggregator", "reported_setting": { "variant": "instruct", "snapshot": "0905" }, "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=Kimi K2-Instruct-0905, slug=kimi-k2-instruct-0905, provider=Moonshot AI" } ] }, { "model_id": "gemma-3-27b", "benchmark_id": "bird_sql", "score": 54.4, "reference_url": "https://arxiv.org/abs/2503.19786", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Gemma 3 tech report Table 6 IT 27B: Bird-SQL Dev 54.4", "candidates": [] }, { "model_id": "gemma-3-27b", "benchmark_id": "facts_grounding", "score": 74.9, "reference_url": "https://arxiv.org/abs/2503.19786", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Gemma 3 tech report Table 6 IT 27B: FACTS Grounding 74.9", "candidates": [] }, { "model_id": "gemma-3-27b", "benchmark_id": "global_mmlu_lite", "score": 75.1, "reference_url": "https://arxiv.org/abs/2503.19786", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Gemma 3 tech report Table 6 IT 27B: Global MMLU-Lite 75.1", "candidates": [] }, { "model_id": "gemma-3-27b", "benchmark_id": "math", "score": 89.0, "reference_url": "https://arxiv.org/abs/2503.19786", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Gemma 3 tech report Table 6 IT 27B: MATH 89.0", "candidates": [ { "score": 76.2, "reference_url": "https://arxiv.org/abs/2501.00656", "source_type": "tech_report", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "OLMES", "prompt_style": "default", "temperature": "default" }, "notes": "OLMo 2 paper Table 7 Instruct: alt measurement 76.2 (mc=false)." } ] }, { "model_id": "gemma-3-27b", "benchmark_id": "hiddenmath", "score": 60.3, "reference_url": "https://arxiv.org/abs/2503.19786", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Gemma 3 tech report Table 6 IT 27B: HiddenMath 60.3", "candidates": [] }, { "model_id": "gemma-3-27b", "benchmark_id": "mmmu", "score": 64.9, "reference_url": "https://arxiv.org/abs/2503.19786", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Gemma 3 tech report Table 6 IT 27B: MMMU 64.9", "candidates": [] }, { "model_id": "gemma-4-31b", "benchmark_id": "bigbench_extra_hard", "score": 74.4, "reference_url": "https://arxiv.org/abs/2607.02770", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "provider-reported", "judge": "benchmark-specified", "harness": "official Google evaluation", "prompt_style": "official", "temperature": "not disclosed", "context": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Gemma 4 Technical Report Table 5; Gemma 4 thinking unless explicitly stated, Gemma 3 27B non-thinking." }, { "model_id": "gemma-4-26b-a4b", "benchmark_id": "bigbench_extra_hard", "score": 64.8, "reference_url": "https://arxiv.org/abs/2607.02770", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "provider-reported", "judge": "benchmark-specified", "harness": "official Google evaluation", "prompt_style": "official", "temperature": "not disclosed", "context": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Gemma 4 Technical Report Table 5; Gemma 4 thinking unless explicitly stated, Gemma 3 27B non-thinking.", "candidates": [ { "score": 56.2, "reference_url": "https://arxiv.org/abs/2608.00146", "source_type": "tech_report", "reported_setting": { "mode": "non-thinking", "decoding": "AR MTP", "tools": "none", "sampling": "replicate count N undisclosed", "judge": "benchmark-specified", "harness": "official Google cross-model evaluation", "prompt_style": "official provider chat template", "context": "default" }, "notes": "DiffusionGemma Technical Report Table 3; replicate count N is undisclosed." } ] }, { "model_id": "gemma-4-e4b", "benchmark_id": "bigbench_extra_hard", "score": 33.1, "reference_url": "https://arxiv.org/abs/2607.02770", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "provider-reported", "judge": "benchmark-specified", "harness": "official Google evaluation", "prompt_style": "official", "temperature": "not disclosed", "context": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Gemma 4 Technical Report Table 5; Gemma 4 thinking unless explicitly stated, Gemma 3 27B non-thinking." }, { "model_id": "gemma-4-e2b", "benchmark_id": "bigbench_extra_hard", "score": 21.9, "reference_url": "https://arxiv.org/abs/2607.02770", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "provider-reported", "judge": "benchmark-specified", "harness": "official Google evaluation", "prompt_style": "official", "temperature": "not disclosed", "context": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Gemma 4 Technical Report Table 5; Gemma 4 thinking unless explicitly stated, Gemma 3 27B non-thinking." }, { "model_id": "gemma-4-31b", "benchmark_id": "medxpertqa_mm", "score": 61.3, "reference_url": "https://arxiv.org/abs/2607.02770", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "provider-reported", "judge": "benchmark-specified", "harness": "official Google evaluation", "prompt_style": "official", "temperature": "not disclosed", "context": "default", "visual_tokens": 1120 }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Gemma 4 Technical Report Table 6; maximum supported resolution, 1,120 visual tokens.", "candidates": [ { "score": 60.7, "reference_url": "https://arxiv.org/abs/2607.02770", "source_type": "tech_report", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "provider-reported", "judge": "benchmark-specified", "harness": "official Google evaluation", "prompt_style": "official", "temperature": "not disclosed", "context": "default", "visual_tokens": 280 }, "notes": "Gemma 4 Technical Report Appendix Table 12; thinking with 280 visual tokens." } ] }, { "model_id": "gemma-4-26b-a4b", "benchmark_id": "medxpertqa_mm", "score": 58.1, "reference_url": "https://arxiv.org/abs/2607.02770", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "provider-reported", "judge": "benchmark-specified", "harness": "official Google evaluation", "prompt_style": "official", "temperature": "not disclosed", "context": "default", "visual_tokens": 1120 }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Gemma 4 Technical Report Table 6; maximum supported resolution, 1,120 visual tokens.", "candidates": [ { "score": 55.7, "reference_url": "https://arxiv.org/abs/2607.02770", "source_type": "tech_report", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "provider-reported", "judge": "benchmark-specified", "harness": "official Google evaluation", "prompt_style": "official", "temperature": "not disclosed", "context": "default", "visual_tokens": 280 }, "notes": "Gemma 4 Technical Report Appendix Table 12; thinking with 280 visual tokens." } ] }, { "model_id": "gemma-4-e4b", "benchmark_id": "medxpertqa_mm", "score": 28.7, "reference_url": "https://arxiv.org/abs/2607.02770", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "provider-reported", "judge": "benchmark-specified", "harness": "official Google evaluation", "prompt_style": "official", "temperature": "not disclosed", "context": "default", "visual_tokens": 1120 }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Gemma 4 Technical Report Table 6; maximum supported resolution, 1,120 visual tokens.", "candidates": [ { "score": 28.7, "reference_url": "https://arxiv.org/abs/2607.02770", "source_type": "tech_report", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "provider-reported", "judge": "benchmark-specified", "harness": "official Google evaluation", "prompt_style": "official", "temperature": "not disclosed", "context": "default", "visual_tokens": 280 }, "notes": "Gemma 4 Technical Report Appendix Table 12; thinking with 280 visual tokens." } ] }, { "model_id": "gemma-4-e2b", "benchmark_id": "medxpertqa_mm", "score": 23.5, "reference_url": "https://arxiv.org/abs/2607.02770", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "provider-reported", "judge": "benchmark-specified", "harness": "official Google evaluation", "prompt_style": "official", "temperature": "not disclosed", "context": "default", "visual_tokens": 1120 }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Gemma 4 Technical Report Table 6; maximum supported resolution, 1,120 visual tokens.", "candidates": [ { "score": 22.5, "reference_url": "https://arxiv.org/abs/2607.02770", "source_type": "tech_report", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "provider-reported", "judge": "benchmark-specified", "harness": "official Google evaluation", "prompt_style": "official", "temperature": "not disclosed", "context": "default", "visual_tokens": 280 }, "notes": "Gemma 4 Technical Report Appendix Table 12; thinking with 280 visual tokens." } ] }, { "model_id": "gemma-4-e4b", "benchmark_id": "fleurs", "score": 0.075, "reference_url": "https://arxiv.org/abs/2607.02770", "reported_setting": { "mode": "unknown", "effort": "n/a", "tools": "audio input; no external tools", "sampling": "provider-reported", "judge": "WER/CER", "harness": "official Google audio evaluation", "prompt_style": "official", "temperature": "not disclosed", "context": "default" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Gemma 4 Technical Report Table 7; WER except CER for Korean, Japanese and Chinese.", "candidates": [ { "score": 0.08, "reference_url": "https://huggingface.co/blog/gemma4", "source_type": "official_blog", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "notes": "Gemma 4 HF blog benchmark table: fleurs 0.08" } ] }, { "model_id": "gemma-4-e2b", "benchmark_id": "fleurs", "score": 0.09, "reference_url": "https://arxiv.org/abs/2607.02770", "reported_setting": { "mode": "unknown", "effort": "n/a", "tools": "audio input; no external tools", "sampling": "provider-reported", "judge": "WER/CER", "harness": "official Google audio evaluation", "prompt_style": "official", "temperature": "not disclosed", "context": "default" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Gemma 4 Technical Report Table 7; WER except CER for Korean, Japanese and Chinese." }, { "model_id": "minimax-m2.5", "benchmark_id": "aime_2025", "score": 86.3, "reference_url": "https://huggingface.co/MiniMaxAI/MiniMax-M2.5", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "MiniMax M2.5 model card: minimax-m2.5=86.3.", "candidates": [] }, { "model_id": "minimax-m2.1", "benchmark_id": "aime_2025", "score": 83.0, "reference_url": "https://huggingface.co/MiniMaxAI/MiniMax-M2.5", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "MiniMax M2.5 model card: minimax-m2.1=83.0.", "candidates": [] }, { "model_id": "minimax-m2.5", "benchmark_id": "gpqa_diamond", "score": 85.2, "reference_url": "https://huggingface.co/MiniMaxAI/MiniMax-M2.5", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "MiniMax M2.5 model card: minimax-m2.5=85.2.", "candidates": [] }, { "model_id": "minimax-m2.1", "benchmark_id": "gpqa_diamond", "score": 83.0, "reference_url": "https://huggingface.co/MiniMaxAI/MiniMax-M2.5", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "MiniMax M2.5 model card: minimax-m2.1=83.0.", "candidates": [] }, { "model_id": "minimax-m2.5", "benchmark_id": "hle", "score": 19.4, "reference_url": "https://huggingface.co/MiniMaxAI/MiniMax-M2.5", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "MiniMax M2.5 model card: minimax-m2.5=19.4.", "candidates": [] }, { "model_id": "minimax-m2.1", "benchmark_id": "hle", "score": 22.2, "reference_url": "https://huggingface.co/MiniMaxAI/MiniMax-M2.5", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "MiniMax M2.5 model card: minimax-m2.1=22.2.", "candidates": [] }, { "model_id": "minimax-m2.5", "benchmark_id": "scicode", "score": 44.4, "reference_url": "https://huggingface.co/MiniMaxAI/MiniMax-M2.5", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "MiniMax M2.5 model card: minimax-m2.5=44.4.", "candidates": [] }, { "model_id": "minimax-m2.1", "benchmark_id": "scicode", "score": 41.0, "reference_url": "https://huggingface.co/MiniMaxAI/MiniMax-M2.5", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "MiniMax M2.5 model card: minimax-m2.1=41.0.", "candidates": [] }, { "model_id": "minimax-m2.5", "benchmark_id": "ifbench", "score": 70.0, "reference_url": "https://huggingface.co/MiniMaxAI/MiniMax-M2.5", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "MiniMax M2.5 model card: minimax-m2.5=70.0.", "candidates": [] }, { "model_id": "minimax-m2.1", "benchmark_id": "ifbench", "score": 70.0, "reference_url": "https://huggingface.co/MiniMaxAI/MiniMax-M2.5", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "MiniMax M2.5 model card: minimax-m2.1=70.0.", "candidates": [] }, { "model_id": "claude-sonnet-4.5", "benchmark_id": "ifbench", "score": 57.0, "reference_url": "https://huggingface.co/MiniMaxAI/MiniMax-M2.5", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "MiniMax M2.5 model card: claude-sonnet-4.5=57.0.", "candidates": [ { "score": 55.4, "reference_url": "https://huggingface.co/mistralai/Mistral-Medium-3.5-128B/resolve/22b2b868a15677cfa6061277ed2f653d1349a9ab/images/image1.png?download=true", "source_type": "official_model_card", "reported_setting": { "mode": "thinking", "effort": "high (maximum)", "tools": "none", "sampling": "pass@1; repeat count undisclosed", "judge": "benchmark-specified", "harness": "Mistral official maximum-reasoning comparison", "prompt_style": "interleaved scratchpads", "temperature": "default (top_p, temperature)", "context": "200K (default)" }, "notes": "Displayed exactly: '55.4'. Research observation: medium-image1.png:2:2:score:reported. Chart footnotes: *All model evaluations were run with maximum reasoning settings. | **Sonnet 4.6 encountered higher rates of reasoning truncation due to external API restrictions than other models, which affected its performance." } ] }, { "model_id": "claude-opus-4.6", "benchmark_id": "ifbench", "score": 53.0, "reference_url": "https://huggingface.co/MiniMaxAI/MiniMax-M2.5", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "MiniMax M2.5 model card: claude-opus-4.6=53.0.", "candidates": [] }, { "model_id": "minimax-m2.5", "benchmark_id": "aa_lcr", "score": 69.5, "reference_url": "https://huggingface.co/MiniMaxAI/MiniMax-M2.5", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "MiniMax M2.5 model card: minimax-m2.5=69.5.", "candidates": [] }, { "model_id": "minimax-m2.1", "benchmark_id": "aa_lcr", "score": 62.0, "reference_url": "https://huggingface.co/MiniMaxAI/MiniMax-M2.5", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "MiniMax M2.5 model card: minimax-m2.1=62.0.", "candidates": [] }, { "model_id": "claude-sonnet-4.5", "benchmark_id": "aa_lcr", "score": 66.0, "reference_url": "https://huggingface.co/MiniMaxAI/MiniMax-M2.5", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "MiniMax M2.5 model card: claude-sonnet-4.5=66.0.", "candidates": [] }, { "model_id": "claude-opus-4.6", "benchmark_id": "aa_lcr", "score": 71.0, "reference_url": "https://huggingface.co/MiniMaxAI/MiniMax-M2.5", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "MiniMax M2.5 model card: claude-opus-4.6=71.0.", "candidates": [] }, { "model_id": "minimax-m2", "benchmark_id": "multi_swe_bench", "score": 36.2, "reference_url": "https://huggingface.co/MiniMaxAI/MiniMax-M2", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "agentic (bash + edit)", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "MiniMax M2 model card: minimax-m2=36.2.", "candidates": [] }, { "model_id": "claude-sonnet-4", "benchmark_id": "multi_swe_bench", "score": 35.7, "reference_url": "https://huggingface.co/MiniMaxAI/MiniMax-M2", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "agentic (bash + edit)", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "MiniMax M2 model card: claude-sonnet-4=35.7.", "candidates": [] }, { "model_id": "claude-sonnet-4.5", "benchmark_id": "multi_swe_bench", "score": 44.3, "reference_url": "https://huggingface.co/MiniMaxAI/MiniMax-M2", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "agentic (bash + edit)", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "MiniMax M2 model card: claude-sonnet-4.5=44.3.", "candidates": [ { "score": 47.7, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "source_type": "model_card", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "varies", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "notes": "Doubao Seed 2.0 model card: claude-sonnet-4.5=47.7." } ] }, { "model_id": "glm-4.6", "benchmark_id": "multi_swe_bench", "score": 30, "reference_url": "https://huggingface.co/MiniMaxAI/MiniMax-M2", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "agentic (bash + edit)", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "MiniMax M2 model card: glm-4.6=30.", "candidates": [] }, { "model_id": "kimi-k2", "benchmark_id": "multi_swe_bench", "score": 33.5, "reference_url": "https://huggingface.co/MiniMaxAI/MiniMax-M2", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "agentic (bash + edit)", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "MiniMax M2 model card: kimi-k2=33.5.", "candidates": [] }, { "model_id": "deepseek-v3.2", "benchmark_id": "multi_swe_bench", "score": 30.6, "reference_url": "https://huggingface.co/MiniMaxAI/MiniMax-M2", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "agentic (bash + edit)", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "MiniMax M2 model card: deepseek-v3.2=30.6.", "candidates": [] }, { "model_id": "claude-sonnet-4", "benchmark_id": "swe_bench_multilingual", "score": 56.9, "reference_url": "https://huggingface.co/MiniMaxAI/MiniMax-M2", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "agentic (bash + edit)", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "MiniMax M2 model card: claude-sonnet-4=56.9.", "candidates": [] }, { "model_id": "minimax-m2", "benchmark_id": "artifactsbench", "score": 66.8, "reference_url": "https://huggingface.co/MiniMaxAI/MiniMax-M2", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "agentic", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "MiniMax M2 model card: minimax-m2=66.8.", "candidates": [] }, { "model_id": "claude-sonnet-4", "benchmark_id": "artifactsbench", "score": 57.3, "reference_url": "https://huggingface.co/MiniMaxAI/MiniMax-M2", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "agentic", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "MiniMax M2 model card: claude-sonnet-4=57.3.", "candidates": [] }, { "model_id": "claude-sonnet-4.5", "benchmark_id": "artifactsbench", "score": 61.5, "reference_url": "https://huggingface.co/MiniMaxAI/MiniMax-M2", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "agentic", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "MiniMax M2 model card: claude-sonnet-4.5=61.5.", "candidates": [ { "score": 59.1, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "source_type": "model_card", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "varies", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "notes": "Doubao Seed 2.0 model card: claude-sonnet-4.5=59.1." } ] }, { "model_id": "gemini-2.5-pro", "benchmark_id": "artifactsbench", "score": 57.7, "reference_url": "https://huggingface.co/MiniMaxAI/MiniMax-M2", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "agentic", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "MiniMax M2 model card: gemini-2.5-pro=57.7.", "candidates": [] }, { "model_id": "gpt-5", "benchmark_id": "artifactsbench", "score": 73, "reference_url": "https://huggingface.co/MiniMaxAI/MiniMax-M2", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "agentic", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "MiniMax M2 model card: gpt-5=73.", "candidates": [] }, { "model_id": "glm-4.6", "benchmark_id": "artifactsbench", "score": 59.8, "reference_url": "https://huggingface.co/MiniMaxAI/MiniMax-M2", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "agentic", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "MiniMax M2 model card: glm-4.6=59.8.", "candidates": [] }, { "model_id": "kimi-k2", "benchmark_id": "artifactsbench", "score": 54.2, "reference_url": "https://huggingface.co/MiniMaxAI/MiniMax-M2", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "agentic", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "MiniMax M2 model card: kimi-k2=54.2.", "candidates": [] }, { "model_id": "deepseek-v3.2", "benchmark_id": "artifactsbench", "score": 55.8, "reference_url": "https://huggingface.co/MiniMaxAI/MiniMax-M2", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "agentic", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "MiniMax M2 model card: deepseek-v3.2=55.8.", "candidates": [] }, { "model_id": "claude-sonnet-4", "benchmark_id": "browsecomp", "score": 12.2, "reference_url": "https://huggingface.co/MiniMaxAI/MiniMax-M2", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "agentic", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "MiniMax M2 model card: claude-sonnet-4=12.2.", "candidates": [] }, { "model_id": "gemini-2.5-pro", "benchmark_id": "browsecomp", "score": 9.9, "reference_url": "https://huggingface.co/MiniMaxAI/MiniMax-M2", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "agentic", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "MiniMax M2 model card: gemini-2.5-pro=9.9.", "candidates": [] }, { "model_id": "kimi-k2", "benchmark_id": "browsecomp", "score": 14.1, "reference_url": "https://huggingface.co/MiniMaxAI/MiniMax-M2", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "agentic", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "MiniMax M2 model card: kimi-k2=14.1.", "candidates": [] }, { "model_id": "minimax-m2", "benchmark_id": "browsecomp_zh", "score": 48.5, "reference_url": "https://huggingface.co/MiniMaxAI/MiniMax-M2", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "agentic", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "MiniMax M2 model card: minimax-m2=48.5.", "candidates": [] }, { "model_id": "claude-sonnet-4", "benchmark_id": "browsecomp_zh", "score": 29.1, "reference_url": "https://huggingface.co/MiniMaxAI/MiniMax-M2", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "agentic", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "MiniMax M2 model card: claude-sonnet-4=29.1.", "candidates": [] }, { "model_id": "gemini-2.5-pro", "benchmark_id": "browsecomp_zh", "score": 32.2, "reference_url": "https://huggingface.co/MiniMaxAI/MiniMax-M2", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "agentic", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "MiniMax M2 model card: gemini-2.5-pro=32.2.", "candidates": [] }, { "model_id": "kimi-k2", "benchmark_id": "browsecomp_zh", "score": 28.8, "reference_url": "https://huggingface.co/MiniMaxAI/MiniMax-M2", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "agentic", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "MiniMax M2 model card: kimi-k2=28.8.", "candidates": [] }, { "model_id": "minimax-m2", "benchmark_id": "gaia", "score": 75.7, "reference_url": "https://huggingface.co/MiniMaxAI/MiniMax-M2", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "agentic", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "MiniMax M2 model card: minimax-m2=75.7.", "candidates": [] }, { "model_id": "claude-sonnet-4", "benchmark_id": "gaia", "score": 68.3, "reference_url": "https://huggingface.co/MiniMaxAI/MiniMax-M2", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "agentic", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "MiniMax M2 model card: claude-sonnet-4=68.3.", "candidates": [] }, { "model_id": "claude-sonnet-4.5", "benchmark_id": "gaia", "score": 71.2, "reference_url": "https://huggingface.co/MiniMaxAI/MiniMax-M2", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "agentic", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "MiniMax M2 model card: claude-sonnet-4.5=71.2.", "candidates": [] }, { "model_id": "gemini-2.5-pro", "benchmark_id": "gaia", "score": 60.2, "reference_url": "https://huggingface.co/MiniMaxAI/MiniMax-M2", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "agentic", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "MiniMax M2 model card: gemini-2.5-pro=60.2.", "candidates": [] }, { "model_id": "gpt-5", "benchmark_id": "gaia", "score": 76.4, "reference_url": "https://huggingface.co/MiniMaxAI/MiniMax-M2", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "agentic", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "MiniMax M2 model card: gpt-5=76.4.", "candidates": [] }, { "model_id": "glm-4.6", "benchmark_id": "gaia", "score": 71.9, "reference_url": "https://huggingface.co/MiniMaxAI/MiniMax-M2", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "agentic", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "MiniMax M2 model card: glm-4.6=71.9.", "candidates": [] }, { "model_id": "kimi-k2", "benchmark_id": "gaia", "score": 60.2, "reference_url": "https://huggingface.co/MiniMaxAI/MiniMax-M2", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "agentic", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "MiniMax M2 model card: kimi-k2=60.2.", "candidates": [] }, { "model_id": "deepseek-v3.2", "benchmark_id": "gaia", "score": 63.5, "reference_url": "https://huggingface.co/MiniMaxAI/MiniMax-M2", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "agentic", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "MiniMax M2 model card: deepseek-v3.2=63.5.", "candidates": [] }, { "model_id": "minimax-m2", "benchmark_id": "xbench_deepsearch", "score": 72, "reference_url": "https://huggingface.co/MiniMaxAI/MiniMax-M2", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "agentic", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "MiniMax M2 model card: minimax-m2=72.", "candidates": [] }, { "model_id": "claude-sonnet-4", "benchmark_id": "xbench_deepsearch", "score": 64.6, "reference_url": "https://huggingface.co/MiniMaxAI/MiniMax-M2", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "agentic", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "MiniMax M2 model card: claude-sonnet-4=64.6.", "candidates": [] }, { "model_id": "claude-sonnet-4.5", "benchmark_id": "xbench_deepsearch", "score": 66, "reference_url": "https://huggingface.co/MiniMaxAI/MiniMax-M2", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "agentic", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "MiniMax M2 model card: claude-sonnet-4.5=66.", "candidates": [] }, { "model_id": "gemini-2.5-pro", "benchmark_id": "xbench_deepsearch", "score": 56, "reference_url": "https://huggingface.co/MiniMaxAI/MiniMax-M2", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "agentic", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "MiniMax M2 model card: gemini-2.5-pro=56.", "candidates": [] }, { "model_id": "gpt-5", "benchmark_id": "xbench_deepsearch", "score": 77.8, "reference_url": "https://huggingface.co/MiniMaxAI/MiniMax-M2", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "agentic", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "MiniMax M2 model card: gpt-5=77.8.", "candidates": [] }, { "model_id": "glm-4.6", "benchmark_id": "xbench_deepsearch", "score": 70, "reference_url": "https://huggingface.co/MiniMaxAI/MiniMax-M2", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "agentic", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "MiniMax M2 model card: glm-4.6=70.", "candidates": [] }, { "model_id": "kimi-k2", "benchmark_id": "xbench_deepsearch", "score": 61, "reference_url": "https://huggingface.co/MiniMaxAI/MiniMax-M2", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "agentic", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "MiniMax M2 model card: kimi-k2=61.", "candidates": [] }, { "model_id": "deepseek-v3.2", "benchmark_id": "xbench_deepsearch", "score": 71, "reference_url": "https://huggingface.co/MiniMaxAI/MiniMax-M2", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "agentic", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "MiniMax M2 model card: deepseek-v3.2=71.", "candidates": [] }, { "model_id": "minimax-m2", "benchmark_id": "tau2_bench_avg", "score": 77.2, "reference_url": "https://huggingface.co/MiniMaxAI/MiniMax-M2", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "agentic", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "MiniMax M2 model card: minimax-m2=77.2.", "candidates": [ { "score": 76.9, "reference_url": "https://huggingface.co/deepseek-ai/DeepSeek-V3.2/resolve/main/assets/paper.pdf", "source_type": "tech_report", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "agentic", "sampling": "pass@1", "judge": "rule-based", "harness": "internal", "prompt_style": "default", "temperature": "1.0", "context": "128K" }, "notes": "DS V3.2 tech report Table 2: 76.9. DS uses test model itself as user agent (non-standard harness). Primary (77.2) from MiniMax M2 official card." } ] }, { "model_id": "claude-sonnet-4", "benchmark_id": "tau2_bench_avg", "score": 65.5, "reference_url": "https://huggingface.co/MiniMaxAI/MiniMax-M2", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "agentic", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "MiniMax M2 model card: claude-sonnet-4=65.5.", "candidates": [] }, { "model_id": "claude-sonnet-4.5", "benchmark_id": "tau2_bench_avg", "score": 84.7, "reference_url": "https://huggingface.co/MiniMaxAI/MiniMax-M2", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "agentic", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "MiniMax M2 model card: claude-sonnet-4.5=84.7.", "candidates": [ { "score": 87.2, "reference_url": "https://deepmind.google/models/gemini/flash/", "source_type": "official_blog", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "notes": "DeepMind /models/gemini/flash/ Performance table: τ2-bench avg 87.2% for Claude Sonnet 4.5 Thinking (third-party self-test by Google DeepMind). Primary 84.7 from MiniMax M2 model card." } ] }, { "model_id": "gemini-2.5-pro", "benchmark_id": "tau2_bench_avg", "score": 59.2, "reference_url": "https://huggingface.co/MiniMaxAI/MiniMax-M2", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "agentic", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "MiniMax M2 model card: gemini-2.5-pro=59.2.", "candidates": [ { "score": 77.8, "reference_url": "https://deepmind.google/models/gemini/flash/", "source_type": "official_blog", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "notes": "DeepMind /models/gemini/flash/ Performance table: τ2-bench avg 77.8% for Gemini 2.5 Pro Thinking. Primary 59.2 from MiniMax M2 model card (third-party); official Flash page value is strongly preferred — see review notes." } ] }, { "model_id": "gpt-5", "benchmark_id": "tau2_bench_avg", "score": 80.1, "reference_url": "https://huggingface.co/MiniMaxAI/MiniMax-M2", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "agentic", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "MiniMax M2 model card: gpt-5=80.1.", "candidates": [ { "score": 80.2, "reference_url": "https://huggingface.co/deepseek-ai/DeepSeek-V3.2/resolve/main/assets/paper.pdf", "source_type": "tech_report", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "agentic", "sampling": "pass@1", "judge": "rule-based", "harness": "internal", "prompt_style": "default", "temperature": "1.0", "context": "128K" }, "notes": "DS V3.2 tech report Table 2: 80.2. DS uses test model itself as user agent (non-standard harness). Primary (80.1) from MiniMax M2 card uses simulated user." } ] }, { "model_id": "glm-4.6", "benchmark_id": "tau2_bench_avg", "score": 75.9, "reference_url": "https://huggingface.co/MiniMaxAI/MiniMax-M2", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "agentic", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "MiniMax M2 model card: glm-4.6=75.9.", "candidates": [ { "score": 75.2, "reference_url": "https://z.ai/blog/glm-4.7", "source_type": "official_blog", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "web+API", "sampling": "pass@1", "judge": "rule-based", "harness": "official" }, "notes": "GLM-4.7 blog (same provider Zhipu): GLM-4.6 τ²-Bench=75.2. Primary is 75.9 from MiniMax M2 card; discrepancy may be different eval dates or run conditions." } ] }, { "model_id": "kimi-k2", "benchmark_id": "tau2_bench_avg", "score": 70.3, "reference_url": "https://huggingface.co/MiniMaxAI/MiniMax-M2", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "agentic", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "MiniMax M2 model card: kimi-k2=70.3.", "candidates": [] }, { "model_id": "deepseek-v3.2", "benchmark_id": "tau2_bench_avg", "score": 66.7, "reference_url": "https://huggingface.co/MiniMaxAI/MiniMax-M2", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "agentic", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "MiniMax M2 model card: deepseek-v3.2=66.7.", "candidates": [ { "score": 80.3, "reference_url": "https://huggingface.co/deepseek-ai/DeepSeek-V3.2/resolve/main/assets/paper.pdf", "source_type": "tech_report", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "agentic", "sampling": "pass@1", "judge": "rule-based", "harness": "internal", "prompt_style": "default", "temperature": "1.0", "context": "128K" }, "notes": "DS V3.2 tech report Table 2: 80.3. DS uses the model itself as user agent for tau2-bench (airline=63.8, retail=81.1, telecom=96.2); non-standard harness. Primary (66.7) from MiniMax M2 card using official simulated-user harness." } ] }, { "model_id": "minimax-m2", "benchmark_id": "finsearchcomp_global", "score": 65.5, "reference_url": "https://huggingface.co/MiniMaxAI/MiniMax-M2", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "agentic", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "MiniMax M2 model card: minimax-m2=65.5.", "candidates": [] }, { "model_id": "claude-sonnet-4", "benchmark_id": "finsearchcomp_global", "score": 42, "reference_url": "https://huggingface.co/MiniMaxAI/MiniMax-M2", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "agentic", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "MiniMax M2 model card: claude-sonnet-4=42.", "candidates": [] }, { "model_id": "claude-sonnet-4.5", "benchmark_id": "finsearchcomp_global", "score": 60.8, "reference_url": "https://huggingface.co/MiniMaxAI/MiniMax-M2", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "agentic", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "MiniMax M2 model card: claude-sonnet-4.5=60.8.", "candidates": [] }, { "model_id": "gemini-2.5-pro", "benchmark_id": "finsearchcomp_global", "score": 42.6, "reference_url": "https://huggingface.co/MiniMaxAI/MiniMax-M2", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "agentic", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "MiniMax M2 model card: gemini-2.5-pro=42.6.", "candidates": [] }, { "model_id": "gpt-5", "benchmark_id": "finsearchcomp_global", "score": 63.9, "reference_url": "https://huggingface.co/MiniMaxAI/MiniMax-M2", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "agentic", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "MiniMax M2 model card: gpt-5=63.9.", "candidates": [] }, { "model_id": "glm-4.6", "benchmark_id": "finsearchcomp_global", "score": 29.2, "reference_url": "https://huggingface.co/MiniMaxAI/MiniMax-M2", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "agentic", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "MiniMax M2 model card: glm-4.6=29.2.", "candidates": [] }, { "model_id": "kimi-k2", "benchmark_id": "finsearchcomp_global", "score": 29.5, "reference_url": "https://huggingface.co/MiniMaxAI/MiniMax-M2", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "agentic", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "MiniMax M2 model card: kimi-k2=29.5.", "candidates": [] }, { "model_id": "deepseek-v3.2", "benchmark_id": "finsearchcomp_global", "score": 26.2, "reference_url": "https://huggingface.co/MiniMaxAI/MiniMax-M2", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "agentic", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "MiniMax M2 model card: deepseek-v3.2=26.2.", "candidates": [] }, { "model_id": "minimax-m2", "benchmark_id": "agentcompany", "score": 36, "reference_url": "https://huggingface.co/MiniMaxAI/MiniMax-M2", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "agentic", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "MiniMax M2 model card: minimax-m2=36.", "candidates": [] }, { "model_id": "claude-sonnet-4", "benchmark_id": "agentcompany", "score": 37, "reference_url": "https://huggingface.co/MiniMaxAI/MiniMax-M2", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "agentic", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "MiniMax M2 model card: claude-sonnet-4=37.", "candidates": [] }, { "model_id": "claude-sonnet-4.5", "benchmark_id": "agentcompany", "score": 41, "reference_url": "https://huggingface.co/MiniMaxAI/MiniMax-M2", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "agentic", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "MiniMax M2 model card: claude-sonnet-4.5=41.", "candidates": [] }, { "model_id": "gemini-2.5-pro", "benchmark_id": "agentcompany", "score": 39.3, "reference_url": "https://huggingface.co/MiniMaxAI/MiniMax-M2", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "agentic", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "MiniMax M2 model card: gemini-2.5-pro=39.3.", "candidates": [] }, { "model_id": "glm-4.6", "benchmark_id": "agentcompany", "score": 35, "reference_url": "https://huggingface.co/MiniMaxAI/MiniMax-M2", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "agentic", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "MiniMax M2 model card: glm-4.6=35.", "candidates": [] }, { "model_id": "kimi-k2", "benchmark_id": "agentcompany", "score": 30, "reference_url": "https://huggingface.co/MiniMaxAI/MiniMax-M2", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "agentic", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "MiniMax M2 model card: kimi-k2=30.", "candidates": [] }, { "model_id": "deepseek-v3.2", "benchmark_id": "agentcompany", "score": 34, "reference_url": "https://huggingface.co/MiniMaxAI/MiniMax-M2", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "agentic", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "MiniMax M2 model card: deepseek-v3.2=34.", "candidates": [] }, { "model_id": "minimax-m2", "benchmark_id": "scicode", "score": 36, "reference_url": "https://huggingface.co/MiniMaxAI/MiniMax-M2", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "MiniMax M2 model card: minimax-m2=36.", "candidates": [] }, { "model_id": "claude-sonnet-4", "benchmark_id": "scicode", "score": 40, "reference_url": "https://huggingface.co/MiniMaxAI/MiniMax-M2", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "MiniMax M2 model card: claude-sonnet-4=40.", "candidates": [] }, { "model_id": "gemini-2.5-pro", "benchmark_id": "scicode", "score": 43, "reference_url": "https://huggingface.co/MiniMaxAI/MiniMax-M2", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "MiniMax M2 model card: gemini-2.5-pro=43.", "candidates": [] }, { "model_id": "glm-4.6", "benchmark_id": "scicode", "score": 38, "reference_url": "https://huggingface.co/MiniMaxAI/MiniMax-M2", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "MiniMax M2 model card: glm-4.6=38.", "candidates": [] }, { "model_id": "minimax-m2", "benchmark_id": "ifbench", "score": 72, "reference_url": "https://huggingface.co/MiniMaxAI/MiniMax-M2", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "MiniMax M2 model card: minimax-m2=72.", "candidates": [] }, { "model_id": "gpt-5", "benchmark_id": "ifbench", "score": 73, "reference_url": "https://huggingface.co/MiniMaxAI/MiniMax-M2", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "MiniMax M2 model card: gpt-5=73.", "candidates": [] }, { "model_id": "glm-4.6", "benchmark_id": "ifbench", "score": 43, "reference_url": "https://huggingface.co/MiniMaxAI/MiniMax-M2", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "MiniMax M2 model card: glm-4.6=43.", "candidates": [] }, { "model_id": "kimi-k2", "benchmark_id": "ifbench", "score": 42, "reference_url": "https://huggingface.co/MiniMaxAI/MiniMax-M2", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "MiniMax M2 model card: kimi-k2=42.", "candidates": [] }, { "model_id": "deepseek-v3.2", "benchmark_id": "ifbench", "score": 54, "reference_url": "https://huggingface.co/MiniMaxAI/MiniMax-M2", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "MiniMax M2 model card: deepseek-v3.2=54.", "candidates": [] }, { "model_id": "minimax-m2", "benchmark_id": "aa_lcr", "score": 61, "reference_url": "https://huggingface.co/MiniMaxAI/MiniMax-M2", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "MiniMax M2 model card: minimax-m2=61.", "candidates": [] }, { "model_id": "claude-sonnet-4", "benchmark_id": "aa_lcr", "score": 65, "reference_url": "https://huggingface.co/MiniMaxAI/MiniMax-M2", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "MiniMax M2 model card: claude-sonnet-4=65.", "candidates": [] }, { "model_id": "gemini-2.5-pro", "benchmark_id": "aa_lcr", "score": 66, "reference_url": "https://huggingface.co/MiniMaxAI/MiniMax-M2", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "MiniMax M2 model card: gemini-2.5-pro=66.", "candidates": [] }, { "model_id": "glm-4.6", "benchmark_id": "aa_lcr", "score": 54, "reference_url": "https://huggingface.co/MiniMaxAI/MiniMax-M2", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "MiniMax M2 model card: glm-4.6=54.", "candidates": [] }, { "model_id": "kimi-k2", "benchmark_id": "aa_lcr", "score": 52, "reference_url": "https://huggingface.co/MiniMaxAI/MiniMax-M2", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "MiniMax M2 model card: kimi-k2=52.", "candidates": [] }, { "model_id": "minimax-m2", "benchmark_id": "tau2_bench_telecom", "score": 87, "reference_url": "https://huggingface.co/MiniMaxAI/MiniMax-M2", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "agentic", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "MiniMax M2 model card: minimax-m2=87.", "candidates": [] }, { "model_id": "gemini-2.5-pro", "benchmark_id": "tau2_bench_telecom", "score": 54, "reference_url": "https://huggingface.co/MiniMaxAI/MiniMax-M2", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "agentic", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "MiniMax M2 model card: gemini-2.5-pro=54.", "candidates": [] }, { "model_id": "glm-4.6", "benchmark_id": "tau2_bench_telecom", "score": 71, "reference_url": "https://huggingface.co/MiniMaxAI/MiniMax-M2", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "agentic", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "MiniMax M2 model card: glm-4.6=71.", "candidates": [] }, { "model_id": "minimax-m2", "benchmark_id": "terminal_bench_hard", "score": 24, "reference_url": "https://huggingface.co/MiniMaxAI/MiniMax-M2", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "agentic", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "MiniMax M2 model card: minimax-m2=24.", "candidates": [] }, { "model_id": "claude-sonnet-4", "benchmark_id": "terminal_bench_hard", "score": 30, "reference_url": "https://huggingface.co/MiniMaxAI/MiniMax-M2", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "agentic", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "MiniMax M2 model card: claude-sonnet-4=30.", "candidates": [] }, { "model_id": "gemini-2.5-pro", "benchmark_id": "terminal_bench_hard", "score": 25, "reference_url": "https://huggingface.co/MiniMaxAI/MiniMax-M2", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "agentic", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "MiniMax M2 model card: gemini-2.5-pro=25.", "candidates": [] }, { "model_id": "kimi-k2", "benchmark_id": "terminal_bench_hard", "score": 23, "reference_url": "https://huggingface.co/MiniMaxAI/MiniMax-M2", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "agentic", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "MiniMax M2 model card: kimi-k2=23.", "candidates": [] }, { "model_id": "claude-sonnet-4", "benchmark_id": "aa_intelligence_index", "score": 57, "reference_url": "https://huggingface.co/MiniMaxAI/MiniMax-M2", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "MiniMax M2 model card: claude-sonnet-4=57.", "candidates": [] }, { "model_id": "claude-sonnet-4.5", "benchmark_id": "aa_intelligence_index", "score": 63, "reference_url": "https://huggingface.co/MiniMaxAI/MiniMax-M2", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "MiniMax M2 model card: claude-sonnet-4.5=63.", "candidates": [] }, { "model_id": "glm-4.6", "benchmark_id": "aa_intelligence_index", "score": 56, "reference_url": "https://huggingface.co/MiniMaxAI/MiniMax-M2", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "MiniMax M2 model card: glm-4.6=56.", "candidates": [] }, { "model_id": "kimi-k2", "benchmark_id": "aa_intelligence_index", "score": 50, "reference_url": "https://huggingface.co/MiniMaxAI/MiniMax-M2", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "MiniMax M2 model card: kimi-k2=50.", "candidates": [] }, { "model_id": "deepseek-v3.2", "benchmark_id": "aa_intelligence_index", "score": 57, "reference_url": "https://huggingface.co/MiniMaxAI/MiniMax-M2", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "MiniMax M2 model card: deepseek-v3.2=57.", "candidates": [] }, { "model_id": "gemini-2.0-pro", "benchmark_id": "mmlu_pro", "score": 79.1, "reference_url": "https://arxiv.org/abs/2503.19786", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Gemma 3 tech report (arxiv:2503.19786) Table 6 IT: mmlu_pro 79.1 (gemini-2.0-pro column).", "candidates": [] }, { "model_id": "gemini-2.0-pro", "benchmark_id": "livecodebench", "score": 36.0, "reference_url": "https://arxiv.org/abs/2503.19786", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Gemma 3 tech report (arxiv:2503.19786) Table 6 IT: livecodebench 36.0 (gemini-2.0-pro column).", "candidates": [] }, { "model_id": "gemini-2.0-pro", "benchmark_id": "bird_sql", "score": 59.3, "reference_url": "https://arxiv.org/abs/2503.19786", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Gemma 3 tech report (arxiv:2503.19786) Table 6 IT: bird_sql 59.3 (gemini-2.0-pro column).", "candidates": [] }, { "model_id": "gemini-2.0-pro", "benchmark_id": "gpqa_diamond", "score": 64.7, "reference_url": "https://arxiv.org/abs/2503.19786", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Gemma 3 tech report (arxiv:2503.19786) Table 6 IT: gpqa_diamond 64.7 (gemini-2.0-pro column).", "candidates": [] }, { "model_id": "gemini-2.0-pro", "benchmark_id": "simpleqa", "score": 44.3, "reference_url": "https://arxiv.org/abs/2503.19786", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Gemma 3 tech report (arxiv:2503.19786) Table 6 IT: simpleqa 44.3 (gemini-2.0-pro column).", "candidates": [] }, { "model_id": "gemini-2.0-pro", "benchmark_id": "facts_grounding", "score": 82.8, "reference_url": "https://arxiv.org/abs/2503.19786", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Gemma 3 tech report (arxiv:2503.19786) Table 6 IT: facts_grounding 82.8 (gemini-2.0-pro column).", "candidates": [] }, { "model_id": "gemini-2.0-pro", "benchmark_id": "global_mmlu_lite", "score": 86.5, "reference_url": "https://arxiv.org/abs/2503.19786", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Gemma 3 tech report (arxiv:2503.19786) Table 6 IT: global_mmlu_lite 86.5 (gemini-2.0-pro column).", "candidates": [] }, { "model_id": "gemini-2.0-pro", "benchmark_id": "math", "score": 91.8, "reference_url": "https://arxiv.org/abs/2503.19786", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Gemma 3 tech report (arxiv:2503.19786) Table 6 IT: math 91.8 (gemini-2.0-pro column).", "candidates": [] }, { "model_id": "gemini-2.0-pro", "benchmark_id": "hiddenmath", "score": 65.2, "reference_url": "https://arxiv.org/abs/2503.19786", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Gemma 3 tech report (arxiv:2503.19786) Table 6 IT: hiddenmath 65.2 (gemini-2.0-pro column).", "candidates": [] }, { "model_id": "gemini-2.0-pro", "benchmark_id": "mmmu", "score": 72.7, "reference_url": "https://arxiv.org/abs/2503.19786", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Gemma 3 tech report (arxiv:2503.19786) Table 6 IT: mmmu 72.7 (gemini-2.0-pro column).", "candidates": [] }, { "model_id": "gemma-2-2b", "benchmark_id": "mmlu_pro", "score": 15.6, "reference_url": "https://arxiv.org/abs/2503.19786", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Gemma 3 tech report (arxiv:2503.19786) Table 6 IT: mmlu_pro 15.6 (gemma-2-2b column).", "candidates": [] }, { "model_id": "gemma-2-2b", "benchmark_id": "livecodebench", "score": 1.2, "reference_url": "https://arxiv.org/abs/2503.19786", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Gemma 3 tech report (arxiv:2503.19786) Table 6 IT: livecodebench 1.2 (gemma-2-2b column).", "candidates": [] }, { "model_id": "gemma-2-2b", "benchmark_id": "bird_sql", "score": 12.2, "reference_url": "https://arxiv.org/abs/2503.19786", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Gemma 3 tech report (arxiv:2503.19786) Table 6 IT: bird_sql 12.2 (gemma-2-2b column).", "candidates": [] }, { "model_id": "gemma-2-2b", "benchmark_id": "gpqa_diamond", "score": 24.7, "reference_url": "https://arxiv.org/abs/2503.19786", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Gemma 3 tech report (arxiv:2503.19786) Table 6 IT: gpqa_diamond 24.7 (gemma-2-2b column).", "candidates": [] }, { "model_id": "gemma-2-2b", "benchmark_id": "simpleqa", "score": 2.8, "reference_url": "https://arxiv.org/abs/2503.19786", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Gemma 3 tech report (arxiv:2503.19786) Table 6 IT: simpleqa 2.8 (gemma-2-2b column).", "candidates": [] }, { "model_id": "gemma-2-2b", "benchmark_id": "facts_grounding", "score": 43.8, "reference_url": "https://arxiv.org/abs/2503.19786", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Gemma 3 tech report (arxiv:2503.19786) Table 6 IT: facts_grounding 43.8 (gemma-2-2b column).", "candidates": [] }, { "model_id": "gemma-2-2b", "benchmark_id": "global_mmlu_lite", "score": 41.9, "reference_url": "https://arxiv.org/abs/2503.19786", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Gemma 3 tech report (arxiv:2503.19786) Table 6 IT: global_mmlu_lite 41.9 (gemma-2-2b column).", "candidates": [] }, { "model_id": "gemma-2-2b", "benchmark_id": "math", "score": 27.2, "reference_url": "https://arxiv.org/abs/2503.19786", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Gemma 3 tech report (arxiv:2503.19786) Table 6 IT: math 27.2 (gemma-2-2b column).", "candidates": [] }, { "model_id": "gemma-2-2b", "benchmark_id": "hiddenmath", "score": 1.8, "reference_url": "https://arxiv.org/abs/2503.19786", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Gemma 3 tech report (arxiv:2503.19786) Table 6 IT: hiddenmath 1.8 (gemma-2-2b column).", "candidates": [] }, { "model_id": "gemma-2-9b", "benchmark_id": "mmlu_pro", "score": 46.8, "reference_url": "https://arxiv.org/abs/2503.19786", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Gemma 3 tech report (arxiv:2503.19786) Table 6 IT: mmlu_pro 46.8 (gemma-2-9b column).", "candidates": [ { "score": 50.6, "reference_url": "https://cohere.com/research/papers/command-a-technical-report.pdf", "source_type": "tech_report", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "Cohere internal eval", "prompt_style": "default", "temperature": "default" }, "notes": "Command A paper: alt measurement 50.6 (mc=false)." } ] }, { "model_id": "gemma-2-9b", "benchmark_id": "livecodebench", "score": 10.8, "reference_url": "https://arxiv.org/abs/2503.19786", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Gemma 3 tech report (arxiv:2503.19786) Table 6 IT: livecodebench 10.8 (gemma-2-9b column).", "candidates": [] }, { "model_id": "gemma-2-9b", "benchmark_id": "bird_sql", "score": 33.8, "reference_url": "https://arxiv.org/abs/2503.19786", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Gemma 3 tech report (arxiv:2503.19786) Table 6 IT: bird_sql 33.8 (gemma-2-9b column).", "candidates": [] }, { "model_id": "gemma-2-9b", "benchmark_id": "gpqa_diamond", "score": 28.8, "reference_url": "https://arxiv.org/abs/2503.19786", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Gemma 3 tech report (arxiv:2503.19786) Table 6 IT: gpqa_diamond 28.8 (gemma-2-9b column).", "candidates": [] }, { "model_id": "gemma-2-9b", "benchmark_id": "simpleqa", "score": 5.3, "reference_url": "https://arxiv.org/abs/2503.19786", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Gemma 3 tech report (arxiv:2503.19786) Table 6 IT: simpleqa 5.3 (gemma-2-9b column).", "candidates": [] }, { "model_id": "gemma-2-9b", "benchmark_id": "facts_grounding", "score": 62.0, "reference_url": "https://arxiv.org/abs/2503.19786", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Gemma 3 tech report (arxiv:2503.19786) Table 6 IT: facts_grounding 62.0 (gemma-2-9b column).", "candidates": [] }, { "model_id": "gemma-2-9b", "benchmark_id": "global_mmlu_lite", "score": 64.8, "reference_url": "https://arxiv.org/abs/2503.19786", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Gemma 3 tech report (arxiv:2503.19786) Table 6 IT: global_mmlu_lite 64.8 (gemma-2-9b column).", "candidates": [] }, { "model_id": "gemma-2-9b", "benchmark_id": "math", "score": 49.4, "reference_url": "https://arxiv.org/abs/2503.19786", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Gemma 3 tech report (arxiv:2503.19786) Table 6 IT: math 49.4 (gemma-2-9b column).", "candidates": [ { "score": 29.8, "reference_url": "https://arxiv.org/abs/2501.00656", "source_type": "tech_report", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "OLMES", "prompt_style": "default", "temperature": "default" }, "notes": "OLMo 2 paper Table 7 Instruct: alt measurement 29.8 (mc=false)." } ] }, { "model_id": "gemma-2-9b", "benchmark_id": "hiddenmath", "score": 10.4, "reference_url": "https://arxiv.org/abs/2503.19786", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Gemma 3 tech report (arxiv:2503.19786) Table 6 IT: hiddenmath 10.4 (gemma-2-9b column).", "candidates": [] }, { "model_id": "gemma-2-27b", "benchmark_id": "mmlu_pro", "score": 56.9, "reference_url": "https://arxiv.org/abs/2503.19786", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Gemma 3 tech report (arxiv:2503.19786) Table 6 IT: mmlu_pro 56.9 (gemma-2-27b column).", "candidates": [] }, { "model_id": "gemma-2-27b", "benchmark_id": "livecodebench", "score": 20.4, "reference_url": "https://arxiv.org/abs/2503.19786", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Gemma 3 tech report (arxiv:2503.19786) Table 6 IT: livecodebench 20.4 (gemma-2-27b column).", "candidates": [] }, { "model_id": "gemma-2-27b", "benchmark_id": "bird_sql", "score": 46.7, "reference_url": "https://arxiv.org/abs/2503.19786", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Gemma 3 tech report (arxiv:2503.19786) Table 6 IT: bird_sql 46.7 (gemma-2-27b column).", "candidates": [] }, { "model_id": "gemma-2-27b", "benchmark_id": "gpqa_diamond", "score": 34.3, "reference_url": "https://arxiv.org/abs/2503.19786", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Gemma 3 tech report (arxiv:2503.19786) Table 6 IT: gpqa_diamond 34.3 (gemma-2-27b column).", "candidates": [] }, { "model_id": "gemma-2-27b", "benchmark_id": "simpleqa", "score": 9.2, "reference_url": "https://arxiv.org/abs/2503.19786", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Gemma 3 tech report (arxiv:2503.19786) Table 6 IT: simpleqa 9.2 (gemma-2-27b column).", "candidates": [] }, { "model_id": "gemma-2-27b", "benchmark_id": "facts_grounding", "score": 62.4, "reference_url": "https://arxiv.org/abs/2503.19786", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Gemma 3 tech report (arxiv:2503.19786) Table 6 IT: facts_grounding 62.4 (gemma-2-27b column).", "candidates": [] }, { "model_id": "gemma-2-27b", "benchmark_id": "global_mmlu_lite", "score": 68.6, "reference_url": "https://arxiv.org/abs/2503.19786", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Gemma 3 tech report (arxiv:2503.19786) Table 6 IT: global_mmlu_lite 68.6 (gemma-2-27b column).", "candidates": [] }, { "model_id": "gemma-2-27b", "benchmark_id": "math", "score": 55.6, "reference_url": "https://arxiv.org/abs/2503.19786", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Gemma 3 tech report (arxiv:2503.19786) Table 6 IT: math 55.6 (gemma-2-27b column).", "candidates": [ { "score": 35.1, "reference_url": "https://arxiv.org/abs/2501.00656", "source_type": "tech_report", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "OLMES", "prompt_style": "default", "temperature": "default" }, "notes": "OLMo 2 paper Table 7 Instruct: alt measurement 35.1 (mc=false)." } ] }, { "model_id": "gemma-2-27b", "benchmark_id": "hiddenmath", "score": 14.8, "reference_url": "https://arxiv.org/abs/2503.19786", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Gemma 3 tech report (arxiv:2503.19786) Table 6 IT: hiddenmath 14.8 (gemma-2-27b column).", "candidates": [] }, { "model_id": "gemma-3-1b", "benchmark_id": "mmlu_pro", "score": 14.7, "reference_url": "https://arxiv.org/abs/2503.19786", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Gemma 3 tech report (arxiv:2503.19786) Table 6 IT: mmlu_pro 14.7 (gemma-3-1b column).", "candidates": [] }, { "model_id": "gemma-3-1b", "benchmark_id": "livecodebench", "score": 1.9, "reference_url": "https://arxiv.org/abs/2503.19786", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Gemma 3 tech report (arxiv:2503.19786) Table 6 IT: livecodebench 1.9 (gemma-3-1b column).", "candidates": [] }, { "model_id": "gemma-3-1b", "benchmark_id": "bird_sql", "score": 6.4, "reference_url": "https://arxiv.org/abs/2503.19786", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Gemma 3 tech report (arxiv:2503.19786) Table 6 IT: bird_sql 6.4 (gemma-3-1b column).", "candidates": [] }, { "model_id": "gemma-3-1b", "benchmark_id": "gpqa_diamond", "score": 19.2, "reference_url": "https://arxiv.org/abs/2503.19786", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Gemma 3 tech report (arxiv:2503.19786) Table 6 IT: gpqa_diamond 19.2 (gemma-3-1b column).", "candidates": [] }, { "model_id": "gemma-3-1b", "benchmark_id": "simpleqa", "score": 2.2, "reference_url": "https://arxiv.org/abs/2503.19786", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Gemma 3 tech report (arxiv:2503.19786) Table 6 IT: simpleqa 2.2 (gemma-3-1b column).", "candidates": [] }, { "model_id": "gemma-3-1b", "benchmark_id": "facts_grounding", "score": 36.4, "reference_url": "https://arxiv.org/abs/2503.19786", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Gemma 3 tech report (arxiv:2503.19786) Table 6 IT: facts_grounding 36.4 (gemma-3-1b column).", "candidates": [] }, { "model_id": "gemma-3-1b", "benchmark_id": "global_mmlu_lite", "score": 34.2, "reference_url": "https://arxiv.org/abs/2503.19786", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Gemma 3 tech report (arxiv:2503.19786) Table 6 IT: global_mmlu_lite 34.2 (gemma-3-1b column).", "candidates": [] }, { "model_id": "gemma-3-1b", "benchmark_id": "math", "score": 48.0, "reference_url": "https://arxiv.org/abs/2503.19786", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Gemma 3 tech report (arxiv:2503.19786) Table 6 IT: math 48.0 (gemma-3-1b column).", "candidates": [ { "score": 40.3, "reference_url": "https://arxiv.org/abs/2501.00656", "source_type": "tech_report", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "OLMES", "prompt_style": "default", "temperature": "default" }, "notes": "OLMo 2 paper Table 7 Instruct: alt measurement 40.3 (mc=false)." } ] }, { "model_id": "gemma-3-1b", "benchmark_id": "hiddenmath", "score": 15.8, "reference_url": "https://arxiv.org/abs/2503.19786", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Gemma 3 tech report (arxiv:2503.19786) Table 6 IT: hiddenmath 15.8 (gemma-3-1b column).", "candidates": [] }, { "model_id": "gemma-3-4b", "benchmark_id": "mmlu_pro", "score": 43.6, "reference_url": "https://arxiv.org/abs/2503.19786", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Gemma 3 tech report (arxiv:2503.19786) Table 6 IT: mmlu_pro 43.6 (gemma-3-4b column).", "candidates": [] }, { "model_id": "gemma-3-4b", "benchmark_id": "livecodebench", "score": 12.6, "reference_url": "https://arxiv.org/abs/2503.19786", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Gemma 3 tech report (arxiv:2503.19786) Table 6 IT: livecodebench 12.6 (gemma-3-4b column).", "candidates": [] }, { "model_id": "gemma-3-4b", "benchmark_id": "bird_sql", "score": 36.3, "reference_url": "https://arxiv.org/abs/2503.19786", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Gemma 3 tech report (arxiv:2503.19786) Table 6 IT: bird_sql 36.3 (gemma-3-4b column).", "candidates": [] }, { "model_id": "gemma-3-4b", "benchmark_id": "gpqa_diamond", "score": 30.8, "reference_url": "https://arxiv.org/abs/2503.19786", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Gemma 3 tech report (arxiv:2503.19786) Table 6 IT: gpqa_diamond 30.8 (gemma-3-4b column).", "candidates": [] }, { "model_id": "gemma-3-4b", "benchmark_id": "simpleqa", "score": 4.0, "reference_url": "https://arxiv.org/abs/2503.19786", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Gemma 3 tech report (arxiv:2503.19786) Table 6 IT: simpleqa 4.0 (gemma-3-4b column).", "candidates": [] }, { "model_id": "gemma-3-4b", "benchmark_id": "facts_grounding", "score": 70.1, "reference_url": "https://arxiv.org/abs/2503.19786", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Gemma 3 tech report (arxiv:2503.19786) Table 6 IT: facts_grounding 70.1 (gemma-3-4b column).", "candidates": [] }, { "model_id": "gemma-3-4b", "benchmark_id": "global_mmlu_lite", "score": 54.5, "reference_url": "https://arxiv.org/abs/2503.19786", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Gemma 3 tech report (arxiv:2503.19786) Table 6 IT: global_mmlu_lite 54.5 (gemma-3-4b column).", "candidates": [] }, { "model_id": "gemma-3-4b", "benchmark_id": "math", "score": 75.6, "reference_url": "https://arxiv.org/abs/2503.19786", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Gemma 3 tech report (arxiv:2503.19786) Table 6 IT: math 75.6 (gemma-3-4b column).", "candidates": [] }, { "model_id": "gemma-3-4b", "benchmark_id": "hiddenmath", "score": 43.0, "reference_url": "https://arxiv.org/abs/2503.19786", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Gemma 3 tech report (arxiv:2503.19786) Table 6 IT: hiddenmath 43.0 (gemma-3-4b column).", "candidates": [] }, { "model_id": "gemma-3-4b", "benchmark_id": "mmmu", "score": 48.8, "reference_url": "https://arxiv.org/abs/2503.19786", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Gemma 3 tech report (arxiv:2503.19786) Table 6 IT: mmmu 48.8 (gemma-3-4b column).", "candidates": [] }, { "model_id": "gemma-3-12b", "benchmark_id": "mmlu_pro", "score": 60.6, "reference_url": "https://arxiv.org/abs/2503.19786", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Gemma 3 tech report (arxiv:2503.19786) Table 6 IT: mmlu_pro 60.6 (gemma-3-12b column).", "candidates": [] }, { "model_id": "gemma-3-12b", "benchmark_id": "livecodebench", "score": 24.6, "reference_url": "https://arxiv.org/abs/2503.19786", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Gemma 3 tech report (arxiv:2503.19786) Table 6 IT: livecodebench 24.6 (gemma-3-12b column).", "candidates": [] }, { "model_id": "gemma-3-12b", "benchmark_id": "bird_sql", "score": 47.9, "reference_url": "https://arxiv.org/abs/2503.19786", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Gemma 3 tech report (arxiv:2503.19786) Table 6 IT: bird_sql 47.9 (gemma-3-12b column).", "candidates": [] }, { "model_id": "gemma-3-12b", "benchmark_id": "gpqa_diamond", "score": 40.9, "reference_url": "https://arxiv.org/abs/2503.19786", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Gemma 3 tech report (arxiv:2503.19786) Table 6 IT: gpqa_diamond 40.9 (gemma-3-12b column).", "candidates": [] }, { "model_id": "gemma-3-12b", "benchmark_id": "simpleqa", "score": 6.3, "reference_url": "https://arxiv.org/abs/2503.19786", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Gemma 3 tech report (arxiv:2503.19786) Table 6 IT: simpleqa 6.3 (gemma-3-12b column).", "candidates": [] }, { "model_id": "gemma-3-12b", "benchmark_id": "facts_grounding", "score": 75.8, "reference_url": "https://arxiv.org/abs/2503.19786", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Gemma 3 tech report (arxiv:2503.19786) Table 6 IT: facts_grounding 75.8 (gemma-3-12b column).", "candidates": [] }, { "model_id": "gemma-3-12b", "benchmark_id": "global_mmlu_lite", "score": 69.5, "reference_url": "https://arxiv.org/abs/2503.19786", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Gemma 3 tech report (arxiv:2503.19786) Table 6 IT: global_mmlu_lite 69.5 (gemma-3-12b column).", "candidates": [] }, { "model_id": "gemma-3-12b", "benchmark_id": "math", "score": 83.8, "reference_url": "https://arxiv.org/abs/2503.19786", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Gemma 3 tech report (arxiv:2503.19786) Table 6 IT: math 83.8 (gemma-3-12b column).", "candidates": [] }, { "model_id": "gemma-3-12b", "benchmark_id": "hiddenmath", "score": 54.5, "reference_url": "https://arxiv.org/abs/2503.19786", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Gemma 3 tech report (arxiv:2503.19786) Table 6 IT: hiddenmath 54.5 (gemma-3-12b column).", "candidates": [] }, { "model_id": "gemma-3-12b", "benchmark_id": "mmmu", "score": 59.6, "reference_url": "https://arxiv.org/abs/2503.19786", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Gemma 3 tech report (arxiv:2503.19786) Table 6 IT: mmmu 59.6 (gemma-3-12b column).", "candidates": [] }, { "model_id": "seed-thinking-v1.5", "benchmark_id": "beyond_aime", "score": 48.0, "reference_url": "https://arxiv.org/abs/2504.13914", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Seed-Thinking-v1.5 paper Table 2: seed-thinking-v1.5=48.0.", "candidates": [] }, { "model_id": "deepseek-r1", "benchmark_id": "beyond_aime", "score": 42.4, "reference_url": "https://arxiv.org/abs/2504.13914", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Seed-Thinking-v1.5 paper Table 2: deepseek-r1=42.4.", "candidates": [] }, { "model_id": "o3-mini-high", "benchmark_id": "beyond_aime", "score": 63.6, "reference_url": "https://arxiv.org/abs/2504.13914", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Seed-Thinking-v1.5 paper Table 2: o3-mini-high=63.6.", "candidates": [] }, { "model_id": "gemini-2.5-pro", "benchmark_id": "beyond_aime", "score": 58.8, "reference_url": "https://arxiv.org/abs/2504.13914", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Seed-Thinking-v1.5 paper Table 2: gemini-2.5-pro=58.8.", "candidates": [] }, { "model_id": "seed-thinking-v1.5", "benchmark_id": "supergpqa", "score": 62.1, "reference_url": "https://arxiv.org/abs/2504.13914", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Seed-Thinking-v1.5 paper Table 2: seed-thinking-v1.5=62.1.", "candidates": [] }, { "model_id": "deepseek-r1", "benchmark_id": "supergpqa", "score": 60.5, "reference_url": "https://arxiv.org/abs/2504.13914", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Seed-Thinking-v1.5 paper Table 2: deepseek-r1=60.5.", "candidates": [] }, { "model_id": "o3-mini-high", "benchmark_id": "supergpqa", "score": 52.2, "reference_url": "https://arxiv.org/abs/2504.13914", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Seed-Thinking-v1.5 paper Table 2: o3-mini-high=52.2.", "candidates": [] }, { "model_id": "grok-3-beta", "benchmark_id": "supergpqa", "score": 62.8, "reference_url": "https://arxiv.org/abs/2504.13914", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Seed-Thinking-v1.5 paper Table 2: grok-3-beta=62.8.", "candidates": [] }, { "model_id": "gemini-2.5-pro", "benchmark_id": "supergpqa", "score": 65.3, "reference_url": "https://arxiv.org/abs/2504.13914", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Seed-Thinking-v1.5 paper Table 2: gemini-2.5-pro=65.3.", "candidates": [] }, { "model_id": "seed-thinking-v1.5", "benchmark_id": "codeforces_pass8", "score": 55.0, "reference_url": "https://arxiv.org/abs/2504.13914", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Seed-Thinking-v1.5 paper Table 2: seed-thinking-v1.5=55.0.", "candidates": [] }, { "model_id": "deepseek-r1", "benchmark_id": "codeforces_pass8", "score": 45.0, "reference_url": "https://arxiv.org/abs/2504.13914", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Seed-Thinking-v1.5 paper Table 2: deepseek-r1=45.0.", "candidates": [] }, { "model_id": "o3-mini-high", "benchmark_id": "codeforces_pass8", "score": 67.5, "reference_url": "https://arxiv.org/abs/2504.13914", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Seed-Thinking-v1.5 paper Table 2: o3-mini-high=67.5.", "candidates": [] }, { "model_id": "gemini-2.5-pro", "benchmark_id": "codeforces_pass8", "score": 56.3, "reference_url": "https://arxiv.org/abs/2504.13914", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Seed-Thinking-v1.5 paper Table 2: gemini-2.5-pro=56.3.", "candidates": [] }, { "model_id": "seed-thinking-v1.5", "benchmark_id": "codeforces_avg8", "score": 36.3, "reference_url": "https://arxiv.org/abs/2504.13914", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Seed-Thinking-v1.5 paper Table 2: seed-thinking-v1.5=36.3.", "candidates": [] }, { "model_id": "deepseek-r1", "benchmark_id": "codeforces_avg8", "score": 32.0, "reference_url": "https://arxiv.org/abs/2504.13914", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Seed-Thinking-v1.5 paper Table 2: deepseek-r1=32.0.", "candidates": [] }, { "model_id": "o3-mini-high", "benchmark_id": "codeforces_avg8", "score": 50.9, "reference_url": "https://arxiv.org/abs/2504.13914", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Seed-Thinking-v1.5 paper Table 2: o3-mini-high=50.9.", "candidates": [] }, { "model_id": "gemini-2.5-pro", "benchmark_id": "codeforces_avg8", "score": 40.3, "reference_url": "https://arxiv.org/abs/2504.13914", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Seed-Thinking-v1.5 paper Table 2: gemini-2.5-pro=40.3.", "candidates": [] }, { "model_id": "seed-thinking-v1.5", "benchmark_id": "livecodebench_v5", "score": 64.9, "reference_url": "https://arxiv.org/abs/2504.13914", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Seed-Thinking-v1.5 paper Table 2: seed-thinking-v1.5=64.9.", "candidates": [] }, { "model_id": "deepseek-r1", "benchmark_id": "livecodebench_v5", "score": 64.3, "reference_url": "https://arxiv.org/abs/2504.13914", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Seed-Thinking-v1.5 paper Table 2: deepseek-r1=64.3.", "candidates": [] }, { "model_id": "o3-mini-high", "benchmark_id": "livecodebench_v5", "score": 74.1, "reference_url": "https://arxiv.org/abs/2504.13914", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Seed-Thinking-v1.5 paper Table 2: o3-mini-high=74.1.", "candidates": [] }, { "model_id": "grok-3-beta", "benchmark_id": "livecodebench_v5", "score": 70.6, "reference_url": "https://arxiv.org/abs/2504.13914", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Seed-Thinking-v1.5 paper Table 2: grok-3-beta=70.6.", "candidates": [] }, { "model_id": "gemini-2.5-pro", "benchmark_id": "livecodebench_v5", "score": 70.4, "reference_url": "https://arxiv.org/abs/2504.13914", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Seed-Thinking-v1.5 paper Table 2: gemini-2.5-pro=70.4.", "candidates": [] }, { "model_id": "seed-thinking-v1.5", "benchmark_id": "aider_polyglot_diff", "score": 54.2, "reference_url": "https://arxiv.org/abs/2504.13914", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "agentic (bash + edit)", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Seed-Thinking-v1.5 paper Table 2: seed-thinking-v1.5=54.2.", "candidates": [] }, { "model_id": "seed-thinking-v1.5", "benchmark_id": "arc_agi_1", "score": 39.9, "reference_url": "https://arxiv.org/abs/2504.13914", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Seed-Thinking-v1.5 paper Table 2: seed-thinking-v1.5=39.9.", "candidates": [] }, { "model_id": "seed-thinking-v1.5", "benchmark_id": "collie", "score": 73.1, "reference_url": "https://arxiv.org/abs/2504.13914", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Seed-Thinking-v1.5 paper Table 2: seed-thinking-v1.5=73.1.", "candidates": [] }, { "model_id": "deepseek-r1", "benchmark_id": "collie", "score": 34.2, "reference_url": "https://arxiv.org/abs/2504.13914", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Seed-Thinking-v1.5 paper Table 2: deepseek-r1=34.2.", "candidates": [] }, { "model_id": "grok-3-beta", "benchmark_id": "collie", "score": 33.6, "reference_url": "https://arxiv.org/abs/2504.13914", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Seed-Thinking-v1.5 paper Table 2: grok-3-beta=33.6.", "candidates": [] }, { "model_id": "gemini-2.5-pro", "benchmark_id": "collie", "score": 62.5, "reference_url": "https://arxiv.org/abs/2504.13914", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Seed-Thinking-v1.5 paper Table 2: gemini-2.5-pro=62.5.", "candidates": [] }, { "model_id": "claude-opus-4.5", "benchmark_id": "simpleqa_verified", "score": 48.6, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card Table 3: claude-opus-4.5=48.6.", "candidates": [] }, { "model_id": "doubao-seed-2.0-pro", "benchmark_id": "simpleqa_verified", "score": 36.0, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card Table 3: doubao-seed-2.0-pro=36.0.", "candidates": [] }, { "model_id": "gpt-5.2", "benchmark_id": "healthbench", "score": 63.3, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card Table 3: gpt-5.2=63.3.", "candidates": [ { "score": 60.7, "reference_url": "https://deploymentsafety.openai.com/gpt-5-6", "source_type": "model_card", "reported_setting": { "effort": "source does not state", "tools": "none", "harness": "OpenAI official HealthBench scoring" }, "notes": "OpenAI GPT-5.6 System Card table (HealthBench length-adjusted scores): gpt-5.2; HealthBench unadjusted=60.7. System Card Table 6. Adjusted and unadjusted metrics are kept as distinct benchmark identities." } ] }, { "model_id": "claude-opus-4.5", "benchmark_id": "healthbench", "score": 36.3, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card Table 3: claude-opus-4.5=36.3.", "candidates": [] }, { "model_id": "gemini-3-pro", "benchmark_id": "healthbench", "score": 37.9, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card Table 3: gemini-3-pro=37.9.", "candidates": [] }, { "model_id": "doubao-seed-2.0-pro", "benchmark_id": "healthbench", "score": 57.7, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card Table 3: doubao-seed-2.0-pro=57.7.", "candidates": [] }, { "model_id": "gpt-5.2", "benchmark_id": "healthbench_hard", "score": 42.0, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card Table 3: gpt-5.2=42.0.", "candidates": [ { "score": 38.9, "reference_url": "https://deploymentsafety.openai.com/gpt-5-6", "source_type": "model_card", "reported_setting": { "effort": "source does not state", "tools": "none", "harness": "OpenAI official HealthBench scoring" }, "notes": "OpenAI GPT-5.6 System Card table (HealthBench length-adjusted scores): gpt-5.2; HealthBench Hard unadjusted=38.9. System Card Table 6. Adjusted and unadjusted metrics are kept as distinct benchmark identities." } ] }, { "model_id": "claude-sonnet-4.5", "benchmark_id": "healthbench_hard", "score": 10.9, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card Table 3: claude-sonnet-4.5=10.9.", "candidates": [] }, { "model_id": "claude-opus-4.5", "benchmark_id": "healthbench_hard", "score": 11.0, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card Table 3: claude-opus-4.5=11.0.", "candidates": [] }, { "model_id": "gemini-3-pro", "benchmark_id": "healthbench_hard", "score": 15.0, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card Table 3: gemini-3-pro=15.0.", "candidates": [] }, { "model_id": "doubao-seed-2.0-pro", "benchmark_id": "healthbench_hard", "score": 29.1, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card Table 3: doubao-seed-2.0-pro=29.1.", "candidates": [] }, { "model_id": "claude-sonnet-4.5", "benchmark_id": "supergpqa", "score": 65.5, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card Table 3: claude-sonnet-4.5=65.5.", "candidates": [] }, { "model_id": "doubao-seed-2.0-pro", "benchmark_id": "supergpqa", "score": 68.7, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card Table 3: doubao-seed-2.0-pro=68.7.", "candidates": [] }, { "model_id": "gpt-5.2", "benchmark_id": "lpfqa", "score": 54.4, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card Table 3: gpt-5.2=54.4.", "candidates": [] }, { "model_id": "claude-sonnet-4.5", "benchmark_id": "lpfqa", "score": 54.9, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card Table 3: claude-sonnet-4.5=54.9.", "candidates": [] }, { "model_id": "claude-opus-4.5", "benchmark_id": "lpfqa", "score": 52.6, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card Table 3: claude-opus-4.5=52.6.", "candidates": [] }, { "model_id": "gemini-3-pro", "benchmark_id": "lpfqa", "score": 51.2, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card Table 3: gemini-3-pro=51.2.", "candidates": [] }, { "model_id": "doubao-seed-2.0-pro", "benchmark_id": "lpfqa", "score": 52.6, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card Table 3: doubao-seed-2.0-pro=52.6.", "candidates": [] }, { "model_id": "gpt-5.2", "benchmark_id": "encyclo_k", "score": 61.0, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card Table 3: gpt-5.2=61.0.", "candidates": [] }, { "model_id": "claude-sonnet-4.5", "benchmark_id": "encyclo_k", "score": 58.0, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card Table 3: claude-sonnet-4.5=58.0.", "candidates": [] }, { "model_id": "claude-opus-4.5", "benchmark_id": "encyclo_k", "score": 63.3, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card Table 3: claude-opus-4.5=63.3.", "candidates": [] }, { "model_id": "gemini-3-pro", "benchmark_id": "encyclo_k", "score": 64.9, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card Table 3: gemini-3-pro=64.9.", "candidates": [] }, { "model_id": "doubao-seed-2.0-pro", "benchmark_id": "encyclo_k", "score": 65.7, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card Table 3: doubao-seed-2.0-pro=65.7.", "candidates": [] }, { "model_id": "claude-sonnet-4.5", "benchmark_id": "aime_2026", "score": 82.5, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card Table 3: claude-sonnet-4.5=82.5.", "candidates": [] }, { "model_id": "doubao-seed-2.0-pro", "benchmark_id": "aime_2026", "score": 94.2, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card Table 3: doubao-seed-2.0-pro=94.2.", "candidates": [] }, { "model_id": "doubao-seed-2.0-pro", "benchmark_id": "hmmt_feb_2025", "score": 97.3, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card Table 3: doubao-seed-2.0-pro=97.3.", "candidates": [] }, { "model_id": "doubao-seed-2.0-pro", "benchmark_id": "hmmt_nov_2025", "score": 93.3, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card Table 3: doubao-seed-2.0-pro=93.3.", "candidates": [] }, { "model_id": "claude-opus-4.5", "benchmark_id": "matharena_apex_2025", "score": 1.6, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card Table 3: claude-opus-4.5=1.6.", "candidates": [] }, { "model_id": "doubao-seed-2.0-pro", "benchmark_id": "matharena_apex_2025", "score": 20.3, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card Table 3: doubao-seed-2.0-pro=20.3.", "candidates": [] }, { "model_id": "gpt-5.2", "benchmark_id": "apex_shortlist", "score": 80.1, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card Table 3: gpt-5.2=80.1.", "candidates": [] }, { "model_id": "claude-sonnet-4.5", "benchmark_id": "apex_shortlist", "score": 26.0, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card Table 3: claude-sonnet-4.5=26.0.", "candidates": [] }, { "model_id": "claude-opus-4.5", "benchmark_id": "apex_shortlist", "score": 47.4, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card Table 3: claude-opus-4.5=47.4.", "candidates": [] }, { "model_id": "gemini-3-pro", "benchmark_id": "apex_shortlist", "score": 71.4, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card Table 3: gemini-3-pro=71.4.", "candidates": [] }, { "model_id": "doubao-seed-2.0-pro", "benchmark_id": "apex_shortlist", "score": 82.1, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card Table 3: doubao-seed-2.0-pro=82.1.", "candidates": [] }, { "model_id": "gpt-5.2", "benchmark_id": "beyond_aime", "score": 86.0, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card Table 3: gpt-5.2=86.0.", "candidates": [] }, { "model_id": "claude-sonnet-4.5", "benchmark_id": "beyond_aime", "score": 57.0, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card Table 3: claude-sonnet-4.5=57.0.", "candidates": [ { "score": 59.8, "reference_url": "https://huggingface.co/mistralai/Mistral-Medium-3.5-128B/resolve/22b2b868a15677cfa6061277ed2f653d1349a9ab/images/image1.png?download=true", "source_type": "official_model_card", "reported_setting": { "mode": "thinking", "effort": "high (maximum)", "tools": "none", "sampling": "samples=16 (reported as avg@16)", "judge": "benchmark-specified", "harness": "Mistral official maximum-reasoning comparison", "prompt_style": "interleaved scratchpads", "temperature": "default (top_p, temperature)", "context": "200K (default)" }, "notes": "Displayed exactly: '59.8'. Research observation: medium-image1.png:4:2:score:reported. Chart footnotes: *All model evaluations were run with maximum reasoning settings. | **Sonnet 4.6 encountered higher rates of reasoning truncation due to external API restrictions than other models, which affected its performance." } ] }, { "model_id": "claude-opus-4.5", "benchmark_id": "beyond_aime", "score": 69.0, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card Table 3: claude-opus-4.5=69.0.", "candidates": [] }, { "model_id": "gemini-3-pro", "benchmark_id": "beyond_aime", "score": 83.0, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card Table 3: gemini-3-pro=83.0.", "candidates": [] }, { "model_id": "doubao-seed-2.0-pro", "benchmark_id": "beyond_aime", "score": 86.5, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card Table 3: doubao-seed-2.0-pro=86.5.", "candidates": [] }, { "model_id": "doubao-seed-2.0-pro", "benchmark_id": "imo_answerbench", "score": 89.3, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card Table 3: doubao-seed-2.0-pro=89.3.", "candidates": [] }, { "model_id": "gpt-5.2", "benchmark_id": "aethercode", "score": 73.8, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card Table 3: gpt-5.2=73.8.", "candidates": [] }, { "model_id": "claude-sonnet-4.5", "benchmark_id": "aethercode", "score": 16.4, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card Table 3: claude-sonnet-4.5=16.4.", "candidates": [] }, { "model_id": "claude-opus-4.5", "benchmark_id": "aethercode", "score": 31.6, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card Table 3: claude-opus-4.5=31.6.", "candidates": [] }, { "model_id": "gemini-3-pro", "benchmark_id": "aethercode", "score": 57.8, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card Table 3: gemini-3-pro=57.8.", "candidates": [] }, { "model_id": "doubao-seed-2.0-pro", "benchmark_id": "aethercode", "score": 60.6, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card Table 3: doubao-seed-2.0-pro=60.6.", "candidates": [] }, { "model_id": "gpt-5.2", "benchmark_id": "livecodebench_v6", "score": 87.7, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card Table 3: gpt-5.2=87.7.", "candidates": [] }, { "model_id": "doubao-seed-2.0-pro", "benchmark_id": "livecodebench_v6", "score": 87.8, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card Table 3: doubao-seed-2.0-pro=87.8.", "candidates": [] }, { "model_id": "gpt-5.2", "benchmark_id": "superchem", "score": 58.0, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card Table 3: gpt-5.2=58.0.", "candidates": [] }, { "model_id": "claude-sonnet-4.5", "benchmark_id": "superchem", "score": 32.4, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card Table 3: claude-sonnet-4.5=32.4.", "candidates": [] }, { "model_id": "claude-opus-4.5", "benchmark_id": "superchem", "score": 43.2, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card Table 3: claude-opus-4.5=43.2.", "candidates": [] }, { "model_id": "gemini-3-pro", "benchmark_id": "superchem", "score": 63.2, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card Table 3: gemini-3-pro=63.2.", "candidates": [] }, { "model_id": "doubao-seed-2.0-pro", "benchmark_id": "superchem", "score": 51.6, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card Table 3: doubao-seed-2.0-pro=51.6.", "candidates": [] }, { "model_id": "gpt-5.2", "benchmark_id": "babe", "score": 58.1, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card Table 3: gpt-5.2=58.1.", "candidates": [] }, { "model_id": "claude-sonnet-4.5", "benchmark_id": "babe", "score": 44.7, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card Table 3: claude-sonnet-4.5=44.7.", "candidates": [] }, { "model_id": "claude-opus-4.5", "benchmark_id": "babe", "score": 49.3, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card Table 3: claude-opus-4.5=49.3.", "candidates": [] }, { "model_id": "gemini-3-pro", "benchmark_id": "babe", "score": 51.3, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card Table 3: gemini-3-pro=51.3.", "candidates": [] }, { "model_id": "doubao-seed-2.0-pro", "benchmark_id": "babe", "score": 50.0, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card Table 3: doubao-seed-2.0-pro=50.0.", "candidates": [] }, { "model_id": "gpt-5.2", "benchmark_id": "phybench", "score": 74.0, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card Table 3: gpt-5.2=74.0.", "candidates": [] }, { "model_id": "claude-sonnet-4.5", "benchmark_id": "phybench", "score": 48.0, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card Table 3: claude-sonnet-4.5=48.0.", "candidates": [] }, { "model_id": "claude-opus-4.5", "benchmark_id": "phybench", "score": 69.0, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card Table 3: claude-opus-4.5=69.0.", "candidates": [] }, { "model_id": "gemini-3-pro", "benchmark_id": "phybench", "score": 80.0, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card Table 3: gemini-3-pro=80.0.", "candidates": [] }, { "model_id": "doubao-seed-2.0-pro", "benchmark_id": "phybench", "score": 74.0, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card Table 3: doubao-seed-2.0-pro=74.0.", "candidates": [] }, { "model_id": "gpt-5.2", "benchmark_id": "frontiersci_research", "score": 25.0, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card Table 3: gpt-5.2=25.0.", "candidates": [] }, { "model_id": "claude-sonnet-4.5", "benchmark_id": "frontiersci_research", "score": 16.7, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card Table 3: claude-sonnet-4.5=16.7.", "candidates": [] }, { "model_id": "claude-opus-4.5", "benchmark_id": "frontiersci_research", "score": 21.7, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card Table 3: claude-opus-4.5=21.7.", "candidates": [] }, { "model_id": "gemini-3-pro", "benchmark_id": "frontiersci_research", "score": 15.0, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card Table 3: gemini-3-pro=15.0.", "candidates": [] }, { "model_id": "doubao-seed-2.0-pro", "benchmark_id": "frontiersci_research", "score": 25.0, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card Table 3: doubao-seed-2.0-pro=25.0.", "candidates": [] }, { "model_id": "gpt-5.2", "benchmark_id": "frontiersci_olympiad", "score": 75.0, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card Table 3: gpt-5.2=75.0.", "candidates": [] }, { "model_id": "claude-sonnet-4.5", "benchmark_id": "frontiersci_olympiad", "score": 60.0, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card Table 3: claude-sonnet-4.5=60.0.", "candidates": [] }, { "model_id": "claude-opus-4.5", "benchmark_id": "frontiersci_olympiad", "score": 71.0, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card Table 3: claude-opus-4.5=71.0.", "candidates": [] }, { "model_id": "gemini-3-pro", "benchmark_id": "frontiersci_olympiad", "score": 73.0, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card Table 3: gemini-3-pro=73.0.", "candidates": [] }, { "model_id": "doubao-seed-2.0-pro", "benchmark_id": "frontiersci_olympiad", "score": 74.0, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card Table 3: doubao-seed-2.0-pro=74.0.", "candidates": [] }, { "model_id": "doubao-seed-2.0-pro", "benchmark_id": "arc_agi_1", "score": 85.4, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card Table 3: doubao-seed-2.0-pro=85.4.", "candidates": [] }, { "model_id": "doubao-seed-2.0-pro", "benchmark_id": "arc_agi_2", "score": 37.5, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card Table 3: doubao-seed-2.0-pro=37.5.", "candidates": [] }, { "model_id": "gpt-5.2", "benchmark_id": "korbench", "score": 79.2, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card Table 3: gpt-5.2=79.2.", "candidates": [] }, { "model_id": "claude-sonnet-4.5", "benchmark_id": "korbench", "score": 73.0, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card Table 3: claude-sonnet-4.5=73.0.", "candidates": [] }, { "model_id": "claude-opus-4.5", "benchmark_id": "korbench", "score": 77.4, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card Table 3: claude-opus-4.5=77.4.", "candidates": [] }, { "model_id": "gemini-3-pro", "benchmark_id": "korbench", "score": 73.9, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card Table 3: gemini-3-pro=73.9.", "candidates": [] }, { "model_id": "doubao-seed-2.0-pro", "benchmark_id": "korbench", "score": 77.5, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card Table 3: doubao-seed-2.0-pro=77.5.", "candidates": [] }, { "model_id": "gpt-5.2", "benchmark_id": "procbench", "score": 95.0, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card Table 3: gpt-5.2=95.0.", "candidates": [] }, { "model_id": "claude-sonnet-4.5", "benchmark_id": "procbench", "score": 87.5, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card Table 3: claude-sonnet-4.5=87.5.", "candidates": [] }, { "model_id": "claude-opus-4.5", "benchmark_id": "procbench", "score": 92.5, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card Table 3: claude-opus-4.5=92.5.", "candidates": [] }, { "model_id": "gemini-3-pro", "benchmark_id": "procbench", "score": 90.0, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card Table 3: gemini-3-pro=90.0.", "candidates": [] }, { "model_id": "doubao-seed-2.0-pro", "benchmark_id": "procbench", "score": 96.6, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card Table 3: doubao-seed-2.0-pro=96.6.", "candidates": [] }, { "model_id": "gpt-5.2", "benchmark_id": "mrcr_v2_8needle", "score": 89.4, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card Table 3: gpt-5.2=89.4.", "candidates": [] }, { "model_id": "claude-sonnet-4.5", "benchmark_id": "mrcr_v2_8needle", "score": 47.1, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card Table 3: claude-sonnet-4.5=47.1.", "candidates": [] }, { "model_id": "claude-opus-4.5", "benchmark_id": "mrcr_v2_8needle", "score": 56.2, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card Table 3: claude-opus-4.5=56.2.", "candidates": [] }, { "model_id": "gemini-3-pro", "benchmark_id": "mrcr_v2_8needle", "score": 79.7, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card Table 3: gemini-3-pro=79.7.", "candidates": [] }, { "model_id": "doubao-seed-2.0-pro", "benchmark_id": "mrcr_v2_8needle", "score": 54.0, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card Table 3: doubao-seed-2.0-pro=54.0.", "candidates": [] }, { "model_id": "claude-sonnet-4.5", "benchmark_id": "graphwalks_bfs_0k_128k", "score": 80.5, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card Table 3: claude-sonnet-4.5=80.5.", "candidates": [] }, { "model_id": "claude-opus-4.5", "benchmark_id": "graphwalks_bfs_0k_128k", "score": 92.0, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card Table 3: claude-opus-4.5=92.0.", "candidates": [] }, { "model_id": "gemini-3-pro", "benchmark_id": "graphwalks_bfs_0k_128k", "score": 79.9, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card Table 3: gemini-3-pro=79.9.", "candidates": [] }, { "model_id": "doubao-seed-2.0-pro", "benchmark_id": "graphwalks_bfs_0k_128k", "score": 68.9, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card Table 3: doubao-seed-2.0-pro=68.9.", "candidates": [] }, { "model_id": "claude-sonnet-4.5", "benchmark_id": "graphwalks_parents_0k_128k", "score": 99.0, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card Table 3: claude-sonnet-4.5=99.0.", "candidates": [] }, { "model_id": "claude-opus-4.5", "benchmark_id": "graphwalks_parents_0k_128k", "score": 96.2, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card Table 3: claude-opus-4.5=96.2.", "candidates": [] }, { "model_id": "gemini-3-pro", "benchmark_id": "graphwalks_parents_0k_128k", "score": 99.7, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card Table 3: gemini-3-pro=99.7.", "candidates": [] }, { "model_id": "doubao-seed-2.0-pro", "benchmark_id": "graphwalks_parents_0k_128k", "score": 97.6, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card Table 3: doubao-seed-2.0-pro=97.6.", "candidates": [] }, { "model_id": "claude-sonnet-4.5", "benchmark_id": "longbench_v2", "score": 62.0, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card Table 3: claude-sonnet-4.5=62.0.", "candidates": [] }, { "model_id": "doubao-seed-2.0-pro", "benchmark_id": "longbench_v2", "score": 63.8, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card Table 3: doubao-seed-2.0-pro=63.8.", "candidates": [] }, { "model_id": "gpt-5.2", "benchmark_id": "frames", "score": 84.0, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card Table 3: gpt-5.2=84.0.", "candidates": [] }, { "model_id": "claude-sonnet-4.5", "benchmark_id": "frames", "score": 78.7, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card Table 3: claude-sonnet-4.5=78.7.", "candidates": [] }, { "model_id": "claude-opus-4.5", "benchmark_id": "frames", "score": 84.7, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card Table 3: claude-opus-4.5=84.7.", "candidates": [] }, { "model_id": "gemini-3-pro", "benchmark_id": "frames", "score": 81.9, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card Table 3: gemini-3-pro=81.9.", "candidates": [] }, { "model_id": "doubao-seed-2.0-pro", "benchmark_id": "frames", "score": 84.5, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card Table 3: doubao-seed-2.0-pro=84.5.", "candidates": [] }, { "model_id": "gpt-5.2", "benchmark_id": "der2_bench", "score": 69.0, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card Table 3: gpt-5.2=69.0.", "candidates": [] }, { "model_id": "claude-sonnet-4.5", "benchmark_id": "der2_bench", "score": 58.9, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card Table 3: claude-sonnet-4.5=58.9.", "candidates": [] }, { "model_id": "claude-opus-4.5", "benchmark_id": "der2_bench", "score": 60.4, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card Table 3: claude-opus-4.5=60.4.", "candidates": [] }, { "model_id": "gemini-3-pro", "benchmark_id": "der2_bench", "score": 66.1, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card Table 3: gemini-3-pro=66.1.", "candidates": [] }, { "model_id": "doubao-seed-2.0-pro", "benchmark_id": "der2_bench", "score": 58.2, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card Table 3: doubao-seed-2.0-pro=58.2.", "candidates": [] }, { "model_id": "gpt-5.2", "benchmark_id": "cl_bench", "score": 23.9, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card Table 3: gpt-5.2=23.9.", "candidates": [] }, { "model_id": "claude-sonnet-4.5", "benchmark_id": "cl_bench", "score": 18.1, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card Table 3: claude-sonnet-4.5=18.1.", "candidates": [] }, { "model_id": "claude-opus-4.5", "benchmark_id": "cl_bench", "score": 22.6, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card Table 3: claude-opus-4.5=22.6.", "candidates": [] }, { "model_id": "gemini-3-pro", "benchmark_id": "cl_bench", "score": 15.6, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card Table 3: gemini-3-pro=15.6.", "candidates": [] }, { "model_id": "doubao-seed-2.0-pro", "benchmark_id": "cl_bench", "score": 20.8, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card Table 3: doubao-seed-2.0-pro=20.8.", "candidates": [] }, { "model_id": "doubao-seed-2.0-pro", "benchmark_id": "global_piqa", "score": 92.3, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card Table 3: doubao-seed-2.0-pro=92.3.", "candidates": [] }, { "model_id": "doubao-seed-2.0-pro", "benchmark_id": "mmmlu", "score": 88.1, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card Table 3: doubao-seed-2.0-pro=88.1.", "candidates": [] }, { "model_id": "gpt-5.2", "benchmark_id": "disco_x", "score": 76.3, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card Table 3: gpt-5.2=76.3.", "candidates": [] }, { "model_id": "claude-sonnet-4.5", "benchmark_id": "disco_x", "score": 70.3, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card Table 3: claude-sonnet-4.5=70.3.", "candidates": [] }, { "model_id": "claude-opus-4.5", "benchmark_id": "disco_x", "score": 78.6, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card Table 3: claude-opus-4.5=78.6.", "candidates": [] }, { "model_id": "gemini-3-pro", "benchmark_id": "disco_x", "score": 76.8, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card Table 3: gemini-3-pro=76.8.", "candidates": [] }, { "model_id": "doubao-seed-2.0-pro", "benchmark_id": "disco_x", "score": 82.0, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card Table 3: doubao-seed-2.0-pro=82.0.", "candidates": [] }, { "model_id": "claude-sonnet-4.5", "benchmark_id": "multichallenge", "score": 57.3, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card Table 3: claude-sonnet-4.5=57.3.", "candidates": [] }, { "model_id": "doubao-seed-2.0-pro", "benchmark_id": "multichallenge", "score": 68.3, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card Table 3: doubao-seed-2.0-pro=68.3.", "candidates": [] }, { "model_id": "gpt-5.2", "benchmark_id": "collie", "score": 96.9, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card Table 3: gpt-5.2=96.9.", "candidates": [] }, { "model_id": "claude-sonnet-4.5", "benchmark_id": "collie", "score": 77.3, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card Table 3: claude-sonnet-4.5=77.3.", "candidates": [ { "score": 90.5, "reference_url": "https://huggingface.co/mistralai/Mistral-Medium-3.5-128B/resolve/22b2b868a15677cfa6061277ed2f653d1349a9ab/images/image1.png?download=true", "source_type": "official_model_card", "reported_setting": { "mode": "thinking", "effort": "high (maximum)", "tools": "none", "sampling": "pass@1; repeat count undisclosed", "judge": "benchmark-specified", "harness": "Mistral official maximum-reasoning comparison", "prompt_style": "interleaved scratchpads", "temperature": "default (top_p, temperature)", "context": "200K (default)" }, "notes": "Displayed exactly: '90.5'. Research observation: medium-image1.png:3:2:score:reported. Chart footnotes: *All model evaluations were run with maximum reasoning settings. | **Sonnet 4.6 encountered higher rates of reasoning truncation due to external API restrictions than other models, which affected its performance." } ] }, { "model_id": "claude-opus-4.5", "benchmark_id": "collie", "score": 79.8, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card Table 3: claude-opus-4.5=79.8.", "candidates": [] }, { "model_id": "gemini-3-pro", "benchmark_id": "collie", "score": 95.0, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card Table 3: gemini-3-pro=95.0.", "candidates": [] }, { "model_id": "doubao-seed-2.0-pro", "benchmark_id": "collie", "score": 93.9, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card Table 3: doubao-seed-2.0-pro=93.9.", "candidates": [] }, { "model_id": "gpt-5.2", "benchmark_id": "mars_bench", "score": 87.9, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card Table 3: gpt-5.2=87.9.", "candidates": [] }, { "model_id": "claude-sonnet-4.5", "benchmark_id": "mars_bench", "score": 72.9, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card Table 3: claude-sonnet-4.5=72.9.", "candidates": [] }, { "model_id": "claude-opus-4.5", "benchmark_id": "mars_bench", "score": 87.7, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card Table 3: claude-opus-4.5=87.7.", "candidates": [] }, { "model_id": "gemini-3-pro", "benchmark_id": "mars_bench", "score": 85.6, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card Table 3: gemini-3-pro=85.6.", "candidates": [] }, { "model_id": "doubao-seed-2.0-pro", "benchmark_id": "mars_bench", "score": 85.6, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card Table 3: doubao-seed-2.0-pro=85.6.", "candidates": [] }, { "model_id": "gpt-5.2", "benchmark_id": "inverse_ifeval", "score": 72.3, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card Table 3: gpt-5.2=72.3.", "candidates": [] }, { "model_id": "claude-sonnet-4.5", "benchmark_id": "inverse_ifeval", "score": 69.3, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card Table 3: claude-sonnet-4.5=69.3.", "candidates": [] }, { "model_id": "claude-opus-4.5", "benchmark_id": "inverse_ifeval", "score": 72.4, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card Table 3: claude-opus-4.5=72.4.", "candidates": [] }, { "model_id": "gemini-3-pro", "benchmark_id": "inverse_ifeval", "score": 79.6, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card Table 3: gemini-3-pro=79.6.", "candidates": [] }, { "model_id": "doubao-seed-2.0-pro", "benchmark_id": "inverse_ifeval", "score": 78.9, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card Table 3: doubao-seed-2.0-pro=78.9.", "candidates": [] }, { "model_id": "gpt-5.2", "benchmark_id": "longfact_objects", "score": 99.2, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card Table 3: gpt-5.2=99.2.", "candidates": [] }, { "model_id": "claude-sonnet-4.5", "benchmark_id": "longfact_objects", "score": 98.8, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card Table 3: claude-sonnet-4.5=98.8.", "candidates": [] }, { "model_id": "claude-opus-4.5", "benchmark_id": "longfact_objects", "score": 99.0, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card Table 3: claude-opus-4.5=99.0.", "candidates": [] }, { "model_id": "gemini-3-pro", "benchmark_id": "longfact_objects", "score": 98.1, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card Table 3: gemini-3-pro=98.1.", "candidates": [] }, { "model_id": "doubao-seed-2.0-pro", "benchmark_id": "longfact_objects", "score": 92.9, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card Table 3: doubao-seed-2.0-pro=92.9.", "candidates": [] }, { "model_id": "gpt-5.2", "benchmark_id": "longfact_concepts", "score": 99.7, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card Table 3: gpt-5.2=99.7.", "candidates": [] }, { "model_id": "claude-sonnet-4.5", "benchmark_id": "longfact_concepts", "score": 98.5, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card Table 3: claude-sonnet-4.5=98.5.", "candidates": [] }, { "model_id": "claude-opus-4.5", "benchmark_id": "longfact_concepts", "score": 98.8, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card Table 3: claude-opus-4.5=98.8.", "candidates": [] }, { "model_id": "gemini-3-pro", "benchmark_id": "longfact_concepts", "score": 98.5, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card Table 3: gemini-3-pro=98.5.", "candidates": [] }, { "model_id": "doubao-seed-2.0-pro", "benchmark_id": "longfact_concepts", "score": 92.8, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card Table 3: doubao-seed-2.0-pro=92.8.", "candidates": [] }, { "model_id": "gpt-5.2", "benchmark_id": "factscore", "score": 91.9, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card Table 3: gpt-5.2=91.9.", "candidates": [] }, { "model_id": "claude-sonnet-4.5", "benchmark_id": "factscore", "score": 90.6, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card Table 3: claude-sonnet-4.5=90.6.", "candidates": [] }, { "model_id": "claude-opus-4.5", "benchmark_id": "factscore", "score": 91.1, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card Table 3: claude-opus-4.5=91.1.", "candidates": [] }, { "model_id": "gemini-3-pro", "benchmark_id": "factscore", "score": 92.6, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card Table 3: gemini-3-pro=92.6.", "candidates": [] }, { "model_id": "doubao-seed-2.0-pro", "benchmark_id": "factscore", "score": 71.2, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card Table 3: doubao-seed-2.0-pro=71.2.", "candidates": [] }, { "model_id": "claude-opus-4.5", "benchmark_id": "mathvista", "score": 80.6, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card: claude-opus-4.5=80.6.", "candidates": [] }, { "model_id": "gpt-5.2", "benchmark_id": "mathvista", "score": 83.1, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card: gpt-5.2=83.1.", "candidates": [] }, { "model_id": "gemini-3-pro", "benchmark_id": "mathvista", "score": 89.8, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card: gemini-3-pro=89.8.", "candidates": [] }, { "model_id": "doubao-seed-2.0-pro", "benchmark_id": "mathvista", "score": 89.8, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card: doubao-seed-2.0-pro=89.8.", "candidates": [] }, { "model_id": "claude-opus-4.5", "benchmark_id": "dynamath", "score": 52.5, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card: claude-opus-4.5=52.5.", "candidates": [] }, { "model_id": "gpt-5.2", "benchmark_id": "dynamath", "score": 70.1, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card: gpt-5.2=70.1.", "candidates": [] }, { "model_id": "gemini-3-pro", "benchmark_id": "dynamath", "score": 63.3, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card: gemini-3-pro=63.3.", "candidates": [] }, { "model_id": "doubao-seed-2.0-pro", "benchmark_id": "dynamath", "score": 68.9, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card: doubao-seed-2.0-pro=68.9.", "candidates": [] }, { "model_id": "claude-opus-4.5", "benchmark_id": "mathkangaroo", "score": 69.6, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card: claude-opus-4.5=69.6.", "candidates": [] }, { "model_id": "gpt-5.2", "benchmark_id": "mathkangaroo", "score": 86.9, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card: gpt-5.2=86.9.", "candidates": [] }, { "model_id": "gemini-3-pro", "benchmark_id": "mathkangaroo", "score": 84.4, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card: gemini-3-pro=84.4.", "candidates": [] }, { "model_id": "doubao-seed-2.0-pro", "benchmark_id": "mathkangaroo", "score": 90.5, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card: doubao-seed-2.0-pro=90.5.", "candidates": [] }, { "model_id": "claude-opus-4.5", "benchmark_id": "mathcanvas", "score": 52.9, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card: claude-opus-4.5=52.9.", "candidates": [] }, { "model_id": "gpt-5.2", "benchmark_id": "mathcanvas", "score": 55.3, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card: gpt-5.2=55.3.", "candidates": [] }, { "model_id": "gemini-3-pro", "benchmark_id": "mathcanvas", "score": 58.8, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card: gemini-3-pro=58.8.", "candidates": [] }, { "model_id": "doubao-seed-2.0-pro", "benchmark_id": "mathcanvas", "score": 61.9, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card: doubao-seed-2.0-pro=61.9.", "candidates": [] }, { "model_id": "doubao-seed-2.0-pro", "benchmark_id": "mmmu_pro", "score": 78.2, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card: doubao-seed-2.0-pro=78.2.", "candidates": [] }, { "model_id": "claude-opus-4.5", "benchmark_id": "emma", "score": 60.4, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card: claude-opus-4.5=60.4.", "candidates": [] }, { "model_id": "gpt-5.2", "benchmark_id": "emma", "score": 69.4, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card: gpt-5.2=69.4.", "candidates": [] }, { "model_id": "gemini-3-pro", "benchmark_id": "emma", "score": 66.5, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card: gemini-3-pro=66.5.", "candidates": [] }, { "model_id": "doubao-seed-2.0-pro", "benchmark_id": "emma", "score": 72.0, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card: doubao-seed-2.0-pro=72.0.", "candidates": [] }, { "model_id": "claude-opus-4.5", "benchmark_id": "sfe", "score": 55.8, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card: claude-opus-4.5=55.8.", "candidates": [] }, { "model_id": "gpt-5.2", "benchmark_id": "sfe", "score": 50.1, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card: gpt-5.2=50.1.", "candidates": [] }, { "model_id": "gemini-3-pro", "benchmark_id": "sfe", "score": 61.9, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card: gemini-3-pro=61.9.", "candidates": [] }, { "model_id": "doubao-seed-2.0-pro", "benchmark_id": "sfe", "score": 55.6, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card: doubao-seed-2.0-pro=55.6.", "candidates": [] }, { "model_id": "claude-opus-4.5", "benchmark_id": "hipho", "score": 81.8, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card: claude-opus-4.5=81.8.", "candidates": [] }, { "model_id": "gpt-5.2", "benchmark_id": "hipho", "score": 77.7, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card: gpt-5.2=77.7.", "candidates": [] }, { "model_id": "gemini-3-pro", "benchmark_id": "hipho", "score": 79.1, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card: gemini-3-pro=79.1.", "candidates": [] }, { "model_id": "doubao-seed-2.0-pro", "benchmark_id": "hipho", "score": 74.1, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card: doubao-seed-2.0-pro=74.1.", "candidates": [] }, { "model_id": "claude-opus-4.5", "benchmark_id": "xlrs_macro", "score": 50.4, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card: claude-opus-4.5=50.4.", "candidates": [] }, { "model_id": "gpt-5.2", "benchmark_id": "xlrs_macro", "score": 49.9, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card: gpt-5.2=49.9.", "candidates": [] }, { "model_id": "gemini-3-pro", "benchmark_id": "xlrs_macro", "score": 51.7, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card: gemini-3-pro=51.7.", "candidates": [] }, { "model_id": "doubao-seed-2.0-pro", "benchmark_id": "xlrs_macro", "score": 54.6, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card: doubao-seed-2.0-pro=54.6.", "candidates": [] }, { "model_id": "claude-opus-4.5", "benchmark_id": "phyx_openended", "score": 61.3, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card: claude-opus-4.5=61.3.", "candidates": [] }, { "model_id": "gpt-5.2", "benchmark_id": "phyx_openended", "score": 71.5, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card: gpt-5.2=71.5.", "candidates": [] }, { "model_id": "gemini-3-pro", "benchmark_id": "phyx_openended", "score": 71.0, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card: gemini-3-pro=71.0.", "candidates": [] }, { "model_id": "doubao-seed-2.0-pro", "benchmark_id": "phyx_openended", "score": 72.1, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card: doubao-seed-2.0-pro=72.1.", "candidates": [] }, { "model_id": "claude-opus-4.5", "benchmark_id": "logicvista", "score": 68.9, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card: claude-opus-4.5=68.9.", "candidates": [] }, { "model_id": "gpt-5.2", "benchmark_id": "logicvista", "score": 81.0, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card: gpt-5.2=81.0.", "candidates": [] }, { "model_id": "gemini-3-pro", "benchmark_id": "logicvista", "score": 80.8, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card: gemini-3-pro=80.8.", "candidates": [] }, { "model_id": "doubao-seed-2.0-pro", "benchmark_id": "logicvista", "score": 81.9, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card: doubao-seed-2.0-pro=81.9.", "candidates": [] }, { "model_id": "claude-opus-4.5", "benchmark_id": "vpct", "score": 29.0, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card: claude-opus-4.5=29.0.", "candidates": [] }, { "model_id": "gpt-5.2", "benchmark_id": "vpct", "score": 56.0, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card: gpt-5.2=56.0.", "candidates": [] }, { "model_id": "gemini-3-pro", "benchmark_id": "vpct", "score": 90.0, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card: gemini-3-pro=90.0.", "candidates": [] }, { "model_id": "doubao-seed-2.0-pro", "benchmark_id": "vpct", "score": 76.0, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card: doubao-seed-2.0-pro=76.0.", "candidates": [] }, { "model_id": "claude-opus-4.5", "benchmark_id": "zerobench_main", "score": 4.0, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card: claude-opus-4.5=4.0.", "candidates": [ { "score": 3, "reference_url": "https://huggingface.co/moonshotai/Kimi-K2.5", "source_type": "model_card", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "1.0" }, "notes": "(merged from zerobench) Kimi K2.5 model card: claude-opus-4.5=3." } ] }, { "model_id": "gpt-5.2", "benchmark_id": "zerobench_main", "score": 11.0, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card: gpt-5.2=11.0.", "candidates": [ { "score": 9, "reference_url": "https://huggingface.co/moonshotai/Kimi-K2.5", "source_type": "model_card", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "1.0" }, "notes": "(merged from zerobench) Kimi K2.5 model card: gpt-5.2=9." } ] }, { "model_id": "gemini-3-pro", "benchmark_id": "zerobench_main", "score": 10.0, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card: gemini-3-pro=10.0.", "candidates": [ { "score": 8, "reference_url": "https://huggingface.co/moonshotai/Kimi-K2.5", "source_type": "model_card", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "1.0" }, "notes": "(merged from zerobench) Kimi K2.5 model card: gemini-3-pro=8." } ] }, { "model_id": "doubao-seed-2.0-pro", "benchmark_id": "zerobench_main", "score": 12.0, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card: doubao-seed-2.0-pro=12.0.", "candidates": [] }, { "model_id": "claude-opus-4.5", "benchmark_id": "zerobench_sub", "score": 30.8, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card: claude-opus-4.5=30.8.", "candidates": [] }, { "model_id": "gpt-5.2", "benchmark_id": "zerobench_sub", "score": 38.9, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card: gpt-5.2=38.9.", "candidates": [] }, { "model_id": "gemini-3-pro", "benchmark_id": "zerobench_sub", "score": 42.2, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card: gemini-3-pro=42.2.", "candidates": [] }, { "model_id": "doubao-seed-2.0-pro", "benchmark_id": "zerobench_sub", "score": 47.6, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card: doubao-seed-2.0-pro=47.6.", "candidates": [] }, { "model_id": "claude-opus-4.5", "benchmark_id": "arcagi1_image", "score": 75.8, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card: claude-opus-4.5=75.8.", "candidates": [] }, { "model_id": "gpt-5.2", "benchmark_id": "arcagi1_image", "score": 93.1, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card: gpt-5.2=93.1.", "candidates": [] }, { "model_id": "gemini-3-pro", "benchmark_id": "arcagi1_image", "score": 69.4, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card: gemini-3-pro=69.4.", "candidates": [] }, { "model_id": "doubao-seed-2.0-pro", "benchmark_id": "arcagi1_image", "score": 88.8, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card: doubao-seed-2.0-pro=88.8.", "candidates": [] }, { "model_id": "claude-opus-4.5", "benchmark_id": "arcagi2_image", "score": 26.1, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card: claude-opus-4.5=26.1.", "candidates": [] }, { "model_id": "gpt-5.2", "benchmark_id": "arcagi2_image", "score": 54.4, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card: gpt-5.2=54.4.", "candidates": [] }, { "model_id": "gemini-3-pro", "benchmark_id": "arcagi2_image", "score": 21.5, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card: gemini-3-pro=21.5.", "candidates": [] }, { "model_id": "doubao-seed-2.0-pro", "benchmark_id": "arcagi2_image", "score": 43.3, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card: doubao-seed-2.0-pro=43.3.", "candidates": [] }, { "model_id": "claude-opus-4.5", "benchmark_id": "visulogic", "score": 27.6, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card: claude-opus-4.5=27.6.", "candidates": [] }, { "model_id": "gpt-5.2", "benchmark_id": "visulogic", "score": 37.0, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card: gpt-5.2=37.0.", "candidates": [] }, { "model_id": "gemini-3-pro", "benchmark_id": "visulogic", "score": 39.0, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card: gemini-3-pro=39.0.", "candidates": [] }, { "model_id": "doubao-seed-2.0-pro", "benchmark_id": "visulogic", "score": 47.4, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card: doubao-seed-2.0-pro=47.4.", "candidates": [] }, { "model_id": "claude-opus-4.5", "benchmark_id": "vlms_are_biased", "score": 21.4, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card: claude-opus-4.5=21.4.", "candidates": [] }, { "model_id": "gpt-5.2", "benchmark_id": "vlms_are_biased", "score": 28.0, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card: gpt-5.2=28.0.", "candidates": [] }, { "model_id": "gemini-3-pro", "benchmark_id": "vlms_are_biased", "score": 50.6, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card: gemini-3-pro=50.6.", "candidates": [] }, { "model_id": "doubao-seed-2.0-pro", "benchmark_id": "vlms_are_biased", "score": 77.4, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card: doubao-seed-2.0-pro=77.4.", "candidates": [] }, { "model_id": "claude-opus-4.5", "benchmark_id": "vlms_are_blind", "score": 77.2, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card: claude-opus-4.5=77.2.", "candidates": [] }, { "model_id": "gpt-5.2", "benchmark_id": "vlms_are_blind", "score": 84.2, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card: gpt-5.2=84.2.", "candidates": [] }, { "model_id": "gemini-3-pro", "benchmark_id": "vlms_are_blind", "score": 97.5, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card: gemini-3-pro=97.5.", "candidates": [] }, { "model_id": "doubao-seed-2.0-pro", "benchmark_id": "vlms_are_blind", "score": 98.6, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card: doubao-seed-2.0-pro=98.6.", "candidates": [] }, { "model_id": "claude-opus-4.5", "benchmark_id": "visfactor", "score": 24.5, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card: claude-opus-4.5=24.5.", "candidates": [] }, { "model_id": "gpt-5.2", "benchmark_id": "visfactor", "score": 33.6, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card: gpt-5.2=33.6.", "candidates": [] }, { "model_id": "gemini-3-pro", "benchmark_id": "visfactor", "score": 45.8, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card: gemini-3-pro=45.8.", "candidates": [] }, { "model_id": "doubao-seed-2.0-pro", "benchmark_id": "visfactor", "score": 36.8, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card: doubao-seed-2.0-pro=36.8.", "candidates": [] }, { "model_id": "claude-opus-4.5", "benchmark_id": "realworldqa", "score": 75.9, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card: claude-opus-4.5=75.9.", "candidates": [] }, { "model_id": "gpt-5.2", "benchmark_id": "realworldqa", "score": 82.1, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card: gpt-5.2=82.1.", "candidates": [] }, { "model_id": "gemini-3-pro", "benchmark_id": "realworldqa", "score": 84.7, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card: gemini-3-pro=84.7.", "candidates": [] }, { "model_id": "doubao-seed-2.0-pro", "benchmark_id": "realworldqa", "score": 86.0, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card: doubao-seed-2.0-pro=86.0.", "candidates": [] }, { "model_id": "claude-opus-4.5", "benchmark_id": "babyvision", "score": 16.2, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card: claude-opus-4.5=16.2.", "candidates": [] }, { "model_id": "gpt-5.2", "benchmark_id": "babyvision", "score": 37.4, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card: gpt-5.2=37.4.", "candidates": [] }, { "model_id": "gemini-3-pro", "benchmark_id": "babyvision", "score": 49.7, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card: gemini-3-pro=49.7.", "candidates": [] }, { "model_id": "doubao-seed-2.0-pro", "benchmark_id": "babyvision", "score": 60.6, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card: doubao-seed-2.0-pro=60.6.", "candidates": [] }, { "model_id": "doubao-seed-2.0-pro", "benchmark_id": "simplevqa", "score": 71.4, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card: doubao-seed-2.0-pro=71.4.", "candidates": [] }, { "model_id": "claude-opus-4.5", "benchmark_id": "hallusionbench", "score": 65.3, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card: claude-opus-4.5=65.3.", "candidates": [] }, { "model_id": "gpt-5.2", "benchmark_id": "hallusionbench", "score": 67.7, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card: gpt-5.2=67.7.", "candidates": [] }, { "model_id": "gemini-3-pro", "benchmark_id": "hallusionbench", "score": 69.9, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card: gemini-3-pro=69.9.", "candidates": [] }, { "model_id": "doubao-seed-2.0-pro", "benchmark_id": "hallusionbench", "score": 68.0, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card: doubao-seed-2.0-pro=68.0.", "candidates": [] }, { "model_id": "claude-opus-4.5", "benchmark_id": "mme_cc", "score": 25.2, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card: claude-opus-4.5=25.2.", "candidates": [] }, { "model_id": "gpt-5.2", "benchmark_id": "mme_cc", "score": 44.4, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card: gpt-5.2=44.4.", "candidates": [] }, { "model_id": "gemini-3-pro", "benchmark_id": "mme_cc", "score": 56.9, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card: gemini-3-pro=56.9.", "candidates": [] }, { "model_id": "doubao-seed-2.0-pro", "benchmark_id": "mme_cc", "score": 57.0, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card: doubao-seed-2.0-pro=57.0.", "candidates": [] }, { "model_id": "claude-opus-4.5", "benchmark_id": "mmstar", "score": 73.9, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card: claude-opus-4.5=73.9.", "candidates": [] }, { "model_id": "gpt-5.2", "benchmark_id": "mmstar", "score": 78.2, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card: gpt-5.2=78.2.", "candidates": [] }, { "model_id": "gemini-3-pro", "benchmark_id": "mmstar", "score": 83.1, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card: gemini-3-pro=83.1.", "candidates": [] }, { "model_id": "doubao-seed-2.0-pro", "benchmark_id": "mmstar", "score": 83.0, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card: doubao-seed-2.0-pro=83.0.", "candidates": [] }, { "model_id": "claude-opus-4.5", "benchmark_id": "muirbench", "score": 78.9, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card: claude-opus-4.5=78.9.", "candidates": [] }, { "model_id": "gpt-5.2", "benchmark_id": "muirbench", "score": 77.4, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card: gpt-5.2=77.4.", "candidates": [] }, { "model_id": "gemini-3-pro", "benchmark_id": "muirbench", "score": 78.2, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card: gemini-3-pro=78.2.", "candidates": [] }, { "model_id": "doubao-seed-2.0-pro", "benchmark_id": "muirbench", "score": 81.8, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card: doubao-seed-2.0-pro=81.8.", "candidates": [] }, { "model_id": "claude-opus-4.5", "benchmark_id": "mtvqa", "score": 53.1, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card: claude-opus-4.5=53.1.", "candidates": [] }, { "model_id": "gpt-5.2", "benchmark_id": "mtvqa", "score": 48.5, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card: gpt-5.2=48.5.", "candidates": [] }, { "model_id": "gemini-3-pro", "benchmark_id": "mtvqa", "score": 50.8, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card: gemini-3-pro=50.8.", "candidates": [] }, { "model_id": "doubao-seed-2.0-pro", "benchmark_id": "mtvqa", "score": 51.1, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card: doubao-seed-2.0-pro=51.1.", "candidates": [] }, { "model_id": "doubao-seed-2.0-pro", "benchmark_id": "worldvqa", "score": 49.9, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card: doubao-seed-2.0-pro=49.9.", "candidates": [] }, { "model_id": "claude-opus-4.5", "benchmark_id": "vibe_eval", "score": 70.3, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card: claude-opus-4.5=70.3.", "candidates": [] }, { "model_id": "gpt-5.2", "benchmark_id": "vibe_eval", "score": 73.1, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card: gpt-5.2=73.1.", "candidates": [] }, { "model_id": "gemini-3-pro", "benchmark_id": "vibe_eval", "score": 77.7, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card: gemini-3-pro=77.7.", "candidates": [] }, { "model_id": "doubao-seed-2.0-pro", "benchmark_id": "vibe_eval", "score": 81.4, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card: doubao-seed-2.0-pro=81.4.", "candidates": [] }, { "model_id": "claude-opus-4.5", "benchmark_id": "viverbench", "score": 72.4, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card: claude-opus-4.5=72.4.", "candidates": [] }, { "model_id": "gpt-5.2", "benchmark_id": "viverbench", "score": 74.8, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card: gpt-5.2=74.8.", "candidates": [] }, { "model_id": "gemini-3-pro", "benchmark_id": "viverbench", "score": 75.9, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card: gemini-3-pro=75.9.", "candidates": [] }, { "model_id": "doubao-seed-2.0-pro", "benchmark_id": "viverbench", "score": 75.9, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card: doubao-seed-2.0-pro=75.9.", "candidates": [] }, { "model_id": "claude-opus-4.5", "benchmark_id": "countbench", "score": 90.3, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card: claude-opus-4.5=90.3.", "candidates": [] }, { "model_id": "gpt-5.2", "benchmark_id": "countbench", "score": 91.2, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card: gpt-5.2=91.2.", "candidates": [] }, { "model_id": "gemini-3-pro", "benchmark_id": "countbench", "score": 97.3, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card: gemini-3-pro=97.3.", "candidates": [] }, { "model_id": "doubao-seed-2.0-pro", "benchmark_id": "countbench", "score": 95.5, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card: doubao-seed-2.0-pro=95.5.", "candidates": [] }, { "model_id": "claude-opus-4.5", "benchmark_id": "fsc_147", "score": 20.9, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card: claude-opus-4.5=20.9.", "candidates": [] }, { "model_id": "gpt-5.2", "benchmark_id": "fsc_147", "score": 21.1, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card: gpt-5.2=21.1.", "candidates": [] }, { "model_id": "gemini-3-pro", "benchmark_id": "fsc_147", "score": 12.1, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card: gemini-3-pro=12.1.", "candidates": [] }, { "model_id": "doubao-seed-2.0-pro", "benchmark_id": "fsc_147", "score": 11.3, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card: doubao-seed-2.0-pro=11.3.", "candidates": [] }, { "model_id": "gemini-3-pro", "benchmark_id": "point_bench", "score": 85.5, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card: gemini-3-pro=85.5.", "candidates": [] }, { "model_id": "doubao-seed-2.0-pro", "benchmark_id": "point_bench", "score": 81.4, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card: doubao-seed-2.0-pro=81.4.", "candidates": [] }, { "model_id": "claude-opus-4.5", "benchmark_id": "blink", "score": 68.1, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card: claude-opus-4.5=68.1.", "candidates": [] }, { "model_id": "gpt-5.2", "benchmark_id": "blink", "score": 70.3, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card: gpt-5.2=70.3.", "candidates": [] }, { "model_id": "gemini-3-pro", "benchmark_id": "blink", "score": 77.1, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card: gemini-3-pro=77.1.", "candidates": [] }, { "model_id": "doubao-seed-2.0-pro", "benchmark_id": "blink", "score": 79.5, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card: doubao-seed-2.0-pro=79.5.", "candidates": [] }, { "model_id": "claude-opus-4.5", "benchmark_id": "mmsibench_circular", "score": 20.2, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card: claude-opus-4.5=20.2.", "candidates": [] }, { "model_id": "gpt-5.2", "benchmark_id": "mmsibench_circular", "score": 26.1, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card: gpt-5.2=26.1.", "candidates": [] }, { "model_id": "gemini-3-pro", "benchmark_id": "mmsibench_circular", "score": 25.4, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card: gemini-3-pro=25.4.", "candidates": [] }, { "model_id": "doubao-seed-2.0-pro", "benchmark_id": "mmsibench_circular", "score": 32.5, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card: doubao-seed-2.0-pro=32.5.", "candidates": [] }, { "model_id": "claude-opus-4.5", "benchmark_id": "treebench", "score": 53.6, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card: claude-opus-4.5=53.6.", "candidates": [] }, { "model_id": "gpt-5.2", "benchmark_id": "treebench", "score": 58.8, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card: gpt-5.2=58.8.", "candidates": [] }, { "model_id": "gemini-3-pro", "benchmark_id": "treebench", "score": 62.7, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card: gemini-3-pro=62.7.", "candidates": [] }, { "model_id": "doubao-seed-2.0-pro", "benchmark_id": "treebench", "score": 64.7, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card: doubao-seed-2.0-pro=64.7.", "candidates": [] }, { "model_id": "gpt-5.2", "benchmark_id": "refspatialbench", "score": 25.5, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card: gpt-5.2=25.5.", "candidates": [] }, { "model_id": "gemini-3-pro", "benchmark_id": "refspatialbench", "score": 65.5, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card: gemini-3-pro=65.5.", "candidates": [] }, { "model_id": "doubao-seed-2.0-pro", "benchmark_id": "refspatialbench", "score": 72.6, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card: doubao-seed-2.0-pro=72.6.", "candidates": [] }, { "model_id": "claude-opus-4.5", "benchmark_id": "da_2k", "score": 70.3, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card: claude-opus-4.5=70.3.", "candidates": [] }, { "model_id": "gpt-5.2", "benchmark_id": "da_2k", "score": 78.9, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card: gpt-5.2=78.9.", "candidates": [] }, { "model_id": "gemini-3-pro", "benchmark_id": "da_2k", "score": 82.1, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card: gemini-3-pro=82.1.", "candidates": [] }, { "model_id": "doubao-seed-2.0-pro", "benchmark_id": "da_2k", "score": 92.3, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card: doubao-seed-2.0-pro=92.3.", "candidates": [] }, { "model_id": "claude-opus-4.5", "benchmark_id": "all_angles", "score": 63.1, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card: claude-opus-4.5=63.1.", "candidates": [] }, { "model_id": "gpt-5.2", "benchmark_id": "all_angles", "score": 71.5, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card: gpt-5.2=71.5.", "candidates": [] }, { "model_id": "gemini-3-pro", "benchmark_id": "all_angles", "score": 73.5, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card: gemini-3-pro=73.5.", "candidates": [] }, { "model_id": "doubao-seed-2.0-pro", "benchmark_id": "all_angles", "score": 72.1, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card: doubao-seed-2.0-pro=72.1.", "candidates": [] }, { "model_id": "claude-opus-4.5", "benchmark_id": "erqa", "score": 48.3, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card: claude-opus-4.5=48.3.", "candidates": [] }, { "model_id": "gpt-5.2", "benchmark_id": "erqa", "score": 59.8, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card: gpt-5.2=59.8.", "candidates": [] }, { "model_id": "gemini-3-pro", "benchmark_id": "erqa", "score": 70.5, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card: gemini-3-pro=70.5.", "candidates": [] }, { "model_id": "doubao-seed-2.0-pro", "benchmark_id": "erqa", "score": 68.5, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card: doubao-seed-2.0-pro=68.5.", "candidates": [] }, { "model_id": "gpt-5.2", "benchmark_id": "chartqapro", "score": 67.6, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card: gpt-5.2=67.6.", "candidates": [] }, { "model_id": "gemini-3-pro", "benchmark_id": "chartqapro", "score": 69.0, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card: gemini-3-pro=69.0.", "candidates": [] }, { "model_id": "doubao-seed-2.0-pro", "benchmark_id": "chartqapro", "score": 71.2, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card: doubao-seed-2.0-pro=71.2.", "candidates": [] }, { "model_id": "claude-opus-4.5", "benchmark_id": "ocrbench_v2", "score": 55.5, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card: claude-opus-4.5=55.5.", "candidates": [] }, { "model_id": "gpt-5.2", "benchmark_id": "ocrbench_v2", "score": 55.6, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card: gpt-5.2=55.6.", "candidates": [] }, { "model_id": "gemini-3-pro", "benchmark_id": "ocrbench_v2", "score": 63.3, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card: gemini-3-pro=63.3.", "candidates": [] }, { "model_id": "doubao-seed-2.0-pro", "benchmark_id": "ocrbench_v2", "score": 62.5, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card: doubao-seed-2.0-pro=62.5.", "candidates": [] }, { "model_id": "doubao-seed-2.0-pro", "benchmark_id": "omnidocbench_1.5", "score": 0.099, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card: doubao-seed-2.0-pro=0.099.", "candidates": [] }, { "model_id": "claude-opus-4.5", "benchmark_id": "charxiv_descriptive", "score": 92.7, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card: claude-opus-4.5=92.7.", "candidates": [] }, { "model_id": "gpt-5.2", "benchmark_id": "charxiv_descriptive", "score": 93.8, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card: gpt-5.2=93.8.", "candidates": [] }, { "model_id": "gemini-3-pro", "benchmark_id": "charxiv_descriptive", "score": 94.4, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card: gemini-3-pro=94.4.", "candidates": [] }, { "model_id": "doubao-seed-2.0-pro", "benchmark_id": "charxiv_descriptive", "score": 93.5, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card: doubao-seed-2.0-pro=93.5.", "candidates": [] }, { "model_id": "doubao-seed-2.0-pro", "benchmark_id": "charxiv_reasoning", "score": 80.5, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card: doubao-seed-2.0-pro=80.5.", "candidates": [] }, { "model_id": "claude-opus-4.5", "benchmark_id": "dude", "score": 55.6, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card: claude-opus-4.5=55.6.", "candidates": [] }, { "model_id": "gpt-5.2", "benchmark_id": "dude", "score": 68.2, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card: gpt-5.2=68.2.", "candidates": [] }, { "model_id": "gemini-3-pro", "benchmark_id": "dude", "score": 70.1, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card: gemini-3-pro=70.1.", "candidates": [] }, { "model_id": "doubao-seed-2.0-pro", "benchmark_id": "dude", "score": 72.4, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card: doubao-seed-2.0-pro=72.4.", "candidates": [] }, { "model_id": "gemini-3-pro", "benchmark_id": "mmlongbench", "score": 73.6, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card: gemini-3-pro=73.6.", "candidates": [] }, { "model_id": "doubao-seed-2.0-pro", "benchmark_id": "mmlongbench", "score": 74.8, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card: doubao-seed-2.0-pro=74.8.", "candidates": [] }, { "model_id": "gemini-3-pro", "benchmark_id": "longdocurl", "score": 72.0, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card: gemini-3-pro=72.0.", "candidates": [] }, { "model_id": "doubao-seed-2.0-pro", "benchmark_id": "longdocurl", "score": 74.7, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card: doubao-seed-2.0-pro=74.7.", "candidates": [] }, { "model_id": "gemini-3-pro", "benchmark_id": "mmlongbench_doc", "score": 59.5, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card: gemini-3-pro=59.5.", "candidates": [] }, { "model_id": "doubao-seed-2.0-pro", "benchmark_id": "mmlongbench_doc", "score": 61.4, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card: doubao-seed-2.0-pro=61.4.", "candidates": [] }, { "model_id": "doubao-seed-2.0-pro", "benchmark_id": "mmvu", "score": 78.2, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card: doubao-seed-2.0-pro=78.2.", "candidates": [] }, { "model_id": "gemini-3-pro", "benchmark_id": "videosimpleqa", "score": 71.9, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card: gemini-3-pro=71.9.", "candidates": [] }, { "model_id": "doubao-seed-2.0-pro", "benchmark_id": "videosimpleqa", "score": 71.9, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card: doubao-seed-2.0-pro=71.9.", "candidates": [] }, { "model_id": "gemini-3-pro", "benchmark_id": "videoreasonbench", "score": 59.5, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card: gemini-3-pro=59.5.", "candidates": [] }, { "model_id": "doubao-seed-2.0-pro", "benchmark_id": "videoreasonbench", "score": 77.8, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card: doubao-seed-2.0-pro=77.8.", "candidates": [] }, { "model_id": "gemini-3-pro", "benchmark_id": "morse_500", "score": 33.0, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card: gemini-3-pro=33.0.", "candidates": [] }, { "model_id": "doubao-seed-2.0-pro", "benchmark_id": "morse_500", "score": 37.4, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card: doubao-seed-2.0-pro=37.4.", "candidates": [] }, { "model_id": "gemini-3-pro", "benchmark_id": "videoholmes", "score": 64.2, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card: gemini-3-pro=64.2.", "candidates": [] }, { "model_id": "doubao-seed-2.0-pro", "benchmark_id": "videoholmes", "score": 67.4, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card: doubao-seed-2.0-pro=67.4.", "candidates": [] }, { "model_id": "gemini-3-pro", "benchmark_id": "minerva", "score": 65.0, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card: gemini-3-pro=65.0.", "candidates": [] }, { "model_id": "doubao-seed-2.0-pro", "benchmark_id": "minerva", "score": 66.5, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card: doubao-seed-2.0-pro=66.5.", "candidates": [] }, { "model_id": "gemini-3-pro", "benchmark_id": "tvbench", "score": 71.1, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card: gemini-3-pro=71.1.", "candidates": [] }, { "model_id": "doubao-seed-2.0-pro", "benchmark_id": "tvbench", "score": 75.0, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card: doubao-seed-2.0-pro=75.0.", "candidates": [] }, { "model_id": "gemini-3-pro", "benchmark_id": "contphy", "score": 58.0, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card: gemini-3-pro=58.0.", "candidates": [] }, { "model_id": "doubao-seed-2.0-pro", "benchmark_id": "contphy", "score": 67.4, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card: doubao-seed-2.0-pro=67.4.", "candidates": [] }, { "model_id": "gemini-3-pro", "benchmark_id": "tempcompass", "score": 88.0, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card: gemini-3-pro=88.0.", "candidates": [] }, { "model_id": "doubao-seed-2.0-pro", "benchmark_id": "tempcompass", "score": 89.6, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card: doubao-seed-2.0-pro=89.6.", "candidates": [] }, { "model_id": "gemini-3-pro", "benchmark_id": "egotempo", "score": 65.4, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card: gemini-3-pro=65.4.", "candidates": [] }, { "model_id": "doubao-seed-2.0-pro", "benchmark_id": "egotempo", "score": 71.8, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card: doubao-seed-2.0-pro=71.8.", "candidates": [] }, { "model_id": "doubao-seed-2.0-pro", "benchmark_id": "motionbench", "score": 75.2, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card: doubao-seed-2.0-pro=75.2.", "candidates": [] }, { "model_id": "gemini-3-pro", "benchmark_id": "tomato", "score": 59.6, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card: gemini-3-pro=59.6.", "candidates": [] }, { "model_id": "doubao-seed-2.0-pro", "benchmark_id": "tomato", "score": 59.9, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card: doubao-seed-2.0-pro=59.9.", "candidates": [] }, { "model_id": "doubao-seed-2.0-pro", "benchmark_id": "video_mme", "score": 89.5, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card: doubao-seed-2.0-pro=89.5.", "candidates": [] }, { "model_id": "gemini-3-pro", "benchmark_id": "cgbench", "score": 65.5, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card: gemini-3-pro=65.5.", "candidates": [] }, { "model_id": "doubao-seed-2.0-pro", "benchmark_id": "cgbench", "score": 65.0, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card: doubao-seed-2.0-pro=65.0.", "candidates": [] }, { "model_id": "doubao-seed-2.0-pro", "benchmark_id": "longvideobench", "score": 80.3, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card: doubao-seed-2.0-pro=80.3.", "candidates": [] }, { "model_id": "gemini-3-pro", "benchmark_id": "videoeval_pro", "score": 52.7, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card: gemini-3-pro=52.7.", "candidates": [] }, { "model_id": "doubao-seed-2.0-pro", "benchmark_id": "videoeval_pro", "score": 48.0, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card: doubao-seed-2.0-pro=48.0.", "candidates": [] }, { "model_id": "doubao-seed-2.0-pro", "benchmark_id": "lvbench", "score": 76.4, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card: doubao-seed-2.0-pro=76.4.", "candidates": [] }, { "model_id": "gemini-3-pro", "benchmark_id": "crossvid", "score": 53.0, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card: gemini-3-pro=53.0.", "candidates": [] }, { "model_id": "doubao-seed-2.0-pro", "benchmark_id": "crossvid", "score": 60.3, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card: doubao-seed-2.0-pro=60.3.", "candidates": [] }, { "model_id": "gemini-3-pro", "benchmark_id": "ovbench", "score": 62.7, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card: gemini-3-pro=62.7.", "candidates": [] }, { "model_id": "doubao-seed-2.0-pro", "benchmark_id": "ovbench", "score": 69.2, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card: doubao-seed-2.0-pro=69.2.", "candidates": [] }, { "model_id": "gemini-3-pro", "benchmark_id": "livesports_3k", "score": 74.5, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card: gemini-3-pro=74.5.", "candidates": [] }, { "model_id": "doubao-seed-2.0-pro", "benchmark_id": "livesports_3k", "score": 78.0, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card: doubao-seed-2.0-pro=78.0.", "candidates": [] }, { "model_id": "gemini-3-pro", "benchmark_id": "ovobench", "score": 70.1, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card: gemini-3-pro=70.1.", "candidates": [] }, { "model_id": "doubao-seed-2.0-pro", "benchmark_id": "ovobench", "score": 77.0, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card: doubao-seed-2.0-pro=77.0.", "candidates": [] }, { "model_id": "gemini-3-pro", "benchmark_id": "odvbench", "score": 63.6, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card: gemini-3-pro=63.6.", "candidates": [] }, { "model_id": "doubao-seed-2.0-pro", "benchmark_id": "odvbench", "score": 72.5, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card: doubao-seed-2.0-pro=72.5.", "candidates": [] }, { "model_id": "gemini-3-pro", "benchmark_id": "vispeak", "score": 89.0, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card: gemini-3-pro=89.0.", "candidates": [] }, { "model_id": "doubao-seed-2.0-pro", "benchmark_id": "vispeak", "score": 78.5, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card: doubao-seed-2.0-pro=78.5.", "candidates": [] }, { "model_id": "mistral-small-3.1", "benchmark_id": "math", "score": 69.3, "reference_url": "https://huggingface.co/mistralai/Mistral-Small-3.1-24B-Instruct-2503", "reported_setting": { "prompt_style": "default", "judge": "rule-based", "harness": "mistral-eval", "sampling": "pass@1", "temperature": "0.0" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "HF MC Table 1: MATH = 69.30% (4-shot CoT, official Mistral eval)." }, { "model_id": "mistral-small-3.1", "benchmark_id": "mmmu", "score": 64.0, "reference_url": "https://huggingface.co/mistralai/Mistral-Small-3.1-24B-Instruct-2503", "reported_setting": { "prompt_style": "default", "judge": "rule-based", "harness": "mistral-eval", "sampling": "pass@1", "temperature": "0.0" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "HF MC Table 2: MMMU = 64.00%." }, { "model_id": "mistral-small-3.1", "benchmark_id": "mmmu_pro", "score": 49.25, "reference_url": "https://huggingface.co/mistralai/Mistral-Small-3.1-24B-Instruct-2503", "reported_setting": { "prompt_style": "default", "judge": "rule-based", "harness": "mistral-eval", "sampling": "pass@1", "temperature": "0.0" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "HF MC Table 2: MMMU PRO = 49.25%." }, { "model_id": "mistral-small-3.1", "benchmark_id": "mathvista", "score": 68.91, "reference_url": "https://huggingface.co/mistralai/Mistral-Small-3.1-24B-Instruct-2503", "reported_setting": { "prompt_style": "default", "judge": "rule-based", "harness": "mistral-eval", "sampling": "pass@1", "temperature": "0.0" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "HF MC Table 2: Mathvista = 68.91%." }, { "model_id": "mistral-small-3.1", "benchmark_id": "longbench_v2", "score": 37.18, "reference_url": "https://huggingface.co/mistralai/Mistral-Small-3.1-24B-Instruct-2503", "reported_setting": { "prompt_style": "default", "judge": "rule-based", "harness": "mistral-eval", "sampling": "pass@1", "temperature": "0.0" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "HF MC Table 4: LongBench v2 = 37.18%." }, { "model_id": "gpt-4o-mini", "benchmark_id": "math_500", "score": 73.0, "reference_url": "https://arxiv.org/abs/2412.08905", "reported_setting": { "mode": "unknown", "effort": "unknown", "tools": "unknown", "sampling": "unknown", "judge": "rule-based", "harness": "Microsoft Phi internal eval", "prompt_style": "unknown", "temperature": "unknown" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Phi-4 paper Table 1 (gpt-4o-mini column): math_500=73.0", "candidates": [] }, { "model_id": "gpt-4o-mini", "benchmark_id": "humaneval", "score": 86.2, "reference_url": "https://arxiv.org/abs/2412.08905", "reported_setting": { "mode": "unknown", "effort": "unknown", "tools": "unknown", "sampling": "unknown", "judge": "rule-based", "harness": "Microsoft Phi internal eval", "prompt_style": "unknown", "temperature": "unknown" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Phi-4 paper Table 1 (gpt-4o-mini column): humaneval=86.2", "candidates": [] }, { "model_id": "phi-4", "benchmark_id": "mgsm", "score": 80.6, "reference_url": "https://arxiv.org/abs/2412.08905", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "simple-evals", "prompt_style": "default", "temperature": "0.5" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Phi-4 paper Table 1 (phi-4 column): mgsm=80.6", "candidates": [] }, { "model_id": "gpt-4o-mini", "benchmark_id": "mgsm", "score": 86.5, "reference_url": "https://arxiv.org/abs/2412.08905", "reported_setting": { "mode": "unknown", "effort": "unknown", "tools": "unknown", "sampling": "unknown", "judge": "rule-based", "harness": "Microsoft Phi internal eval", "prompt_style": "unknown", "temperature": "unknown" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Phi-4 paper Table 1 (gpt-4o-mini column): mgsm=86.5", "candidates": [] }, { "model_id": "qwen2.5-72b-instruct", "benchmark_id": "mgsm", "score": 87.3, "reference_url": "https://arxiv.org/abs/2412.08905", "reported_setting": { "mode": "unknown", "effort": "unknown", "tools": "unknown", "sampling": "unknown", "judge": "rule-based", "harness": "Microsoft Phi internal eval", "prompt_style": "unknown", "temperature": "unknown" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Phi-4 paper Table 1 (qwen2.5-72b-instruct column): mgsm=87.3", "candidates": [] }, { "model_id": "gpt-4o-0513", "benchmark_id": "mgsm", "score": 90.4, "reference_url": "https://arxiv.org/abs/2412.08905", "reported_setting": { "mode": "unknown", "effort": "unknown", "tools": "unknown", "sampling": "unknown", "judge": "rule-based", "harness": "Microsoft Phi internal eval", "prompt_style": "unknown", "temperature": "unknown" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Phi-4 paper Table 1 (gpt-4o-0513 column): mgsm=90.4", "candidates": [] }, { "model_id": "gpt-4o-mini", "benchmark_id": "simpleqa", "score": 9.9, "reference_url": "https://arxiv.org/abs/2412.08905", "reported_setting": { "mode": "unknown", "effort": "unknown", "tools": "unknown", "sampling": "unknown", "judge": "rule-based", "harness": "Microsoft Phi internal eval", "prompt_style": "unknown", "temperature": "unknown" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Phi-4 paper Table 1 (gpt-4o-mini column): simpleqa=9.9", "candidates": [] }, { "model_id": "phi-4", "benchmark_id": "drop", "score": 75.5, "reference_url": "https://arxiv.org/abs/2412.08905", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "simple-evals", "prompt_style": "default", "temperature": "0.5" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Phi-4 paper Table 1 (phi-4 column): drop=75.5", "candidates": [] }, { "model_id": "gpt-4o-mini", "benchmark_id": "drop", "score": 79.3, "reference_url": "https://arxiv.org/abs/2412.08905", "reported_setting": { "mode": "unknown", "effort": "unknown", "tools": "unknown", "sampling": "unknown", "judge": "rule-based", "harness": "Microsoft Phi internal eval", "prompt_style": "unknown", "temperature": "unknown" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Phi-4 paper Table 1 (gpt-4o-mini column): drop=79.3", "candidates": [ { "score": 36.3, "reference_url": "https://arxiv.org/abs/2501.00656", "source_type": "tech_report", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "OLMES", "prompt_style": "default", "temperature": "default" }, "notes": "OLMo 2 paper Table 7 Instruct: alt measurement 36.3 (mc=false)." } ] }, { "model_id": "gpt-4o-mini", "benchmark_id": "mmlu_pro", "score": 63.4, "reference_url": "https://arxiv.org/abs/2412.08905", "reported_setting": { "mode": "unknown", "effort": "unknown", "tools": "unknown", "sampling": "unknown", "judge": "rule-based", "harness": "Microsoft Phi internal eval", "prompt_style": "unknown", "temperature": "unknown" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Phi-4 paper Table 1 (gpt-4o-mini column): mmlu_pro=63.4", "candidates": [] }, { "model_id": "phi-4", "benchmark_id": "humaneval_plus", "score": 82.8, "reference_url": "https://arxiv.org/abs/2412.08905", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "simple-evals", "prompt_style": "default", "temperature": "0.5" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Phi-4 paper Table 1 (phi-4 column): humaneval_plus=82.8", "candidates": [ { "score": 83.5, "reference_url": "https://arxiv.org/abs/2504.21318", "source_type": "tech_report", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "simple-evals", "prompt_style": "default", "temperature": "0.5" }, "notes": "Phi-4-reasoning paper Table 2 (phi-4 column): humaneval_plus=83.5 (alt measurement, mc=true)" } ] }, { "model_id": "gpt-4o-mini", "benchmark_id": "humaneval_plus", "score": 82.0, "reference_url": "https://arxiv.org/abs/2412.08905", "reported_setting": { "mode": "unknown", "effort": "unknown", "tools": "unknown", "sampling": "unknown", "judge": "rule-based", "harness": "Microsoft Phi internal eval", "prompt_style": "unknown", "temperature": "unknown" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Phi-4 paper Table 1 (gpt-4o-mini column): humaneval_plus=82.0", "candidates": [] }, { "model_id": "qwen2.5-72b-instruct", "benchmark_id": "humaneval_plus", "score": 78.4, "reference_url": "https://arxiv.org/abs/2412.08905", "reported_setting": { "mode": "unknown", "effort": "unknown", "tools": "unknown", "sampling": "unknown", "judge": "rule-based", "harness": "Microsoft Phi internal eval", "prompt_style": "unknown", "temperature": "unknown" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Phi-4 paper Table 1 (qwen2.5-72b-instruct column): humaneval_plus=78.4", "candidates": [] }, { "model_id": "gpt-4o-0513", "benchmark_id": "humaneval_plus", "score": 88.0, "reference_url": "https://arxiv.org/abs/2412.08905", "reported_setting": { "mode": "unknown", "effort": "unknown", "tools": "unknown", "sampling": "unknown", "judge": "rule-based", "harness": "Microsoft Phi internal eval", "prompt_style": "unknown", "temperature": "unknown" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Phi-4 paper Table 1 (gpt-4o-0513 column): humaneval_plus=88.0", "candidates": [ { "score": 84.9, "reference_url": "https://arxiv.org/abs/2504.21318", "source_type": "tech_report", "reported_setting": { "mode": "unknown", "effort": "unknown", "tools": "unknown", "sampling": "unknown", "judge": "rule-based", "harness": "Microsoft Phi internal eval", "prompt_style": "unknown", "temperature": "unknown" }, "notes": "Phi-4-reasoning paper Table 2 (gpt-4o-0513 column): humaneval_plus=84.9 (alt measurement, mc=false)" } ] }, { "model_id": "gpt-4o-mini", "benchmark_id": "arena_hard", "score": 76.2, "reference_url": "https://arxiv.org/abs/2412.08905", "reported_setting": { "mode": "unknown", "effort": "unknown", "tools": "unknown", "sampling": "unknown", "judge": "rule-based", "harness": "Microsoft Phi internal eval", "prompt_style": "unknown", "temperature": "unknown" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Phi-4 paper Table 1 (gpt-4o-mini column): arena_hard=76.2", "candidates": [] }, { "model_id": "qwen2.5-72b-instruct", "benchmark_id": "arena_hard", "score": 78.4, "reference_url": "https://arxiv.org/abs/2412.08905", "reported_setting": { "mode": "unknown", "effort": "unknown", "tools": "unknown", "sampling": "unknown", "judge": "rule-based", "harness": "Microsoft Phi internal eval", "prompt_style": "unknown", "temperature": "unknown" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Phi-4 paper Table 1 (qwen2.5-72b-instruct column): arena_hard=78.4", "candidates": [] }, { "model_id": "gpt-4o-0513", "benchmark_id": "arena_hard", "score": 75.6, "reference_url": "https://arxiv.org/abs/2412.08905", "reported_setting": { "mode": "unknown", "effort": "unknown", "tools": "unknown", "sampling": "unknown", "judge": "rule-based", "harness": "Microsoft Phi internal eval", "prompt_style": "unknown", "temperature": "unknown" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Phi-4 paper Table 1 (gpt-4o-0513 column): arena_hard=75.6", "candidates": [ { "score": 69.0, "reference_url": "https://arxiv.org/abs/2504.21318", "source_type": "tech_report", "reported_setting": { "mode": "unknown", "effort": "unknown", "tools": "unknown", "sampling": "unknown", "judge": "rule-based", "harness": "Microsoft Phi internal eval", "prompt_style": "unknown", "temperature": "unknown" }, "notes": "Phi-4-reasoning paper Table 2 (gpt-4o-0513 column): arena_hard=69.0 (alt measurement, mc=false)" } ] }, { "model_id": "phi-4", "benchmark_id": "livebench", "score": 47.6, "reference_url": "https://arxiv.org/abs/2412.08905", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "simple-evals", "prompt_style": "default", "temperature": "0.5" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Phi-4 paper Table 1 (phi-4 column): livebench=47.6", "candidates": [] }, { "model_id": "gpt-4o-mini", "benchmark_id": "livebench", "score": 48.1, "reference_url": "https://arxiv.org/abs/2412.08905", "reported_setting": { "mode": "unknown", "effort": "unknown", "tools": "unknown", "sampling": "unknown", "judge": "rule-based", "harness": "Microsoft Phi internal eval", "prompt_style": "unknown", "temperature": "unknown" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Phi-4 paper Table 1 (gpt-4o-mini column): livebench=48.1", "candidates": [] }, { "model_id": "qwen2.5-72b-instruct", "benchmark_id": "livebench", "score": 55.3, "reference_url": "https://arxiv.org/abs/2412.08905", "reported_setting": { "mode": "unknown", "effort": "unknown", "tools": "unknown", "sampling": "unknown", "judge": "rule-based", "harness": "Microsoft Phi internal eval", "prompt_style": "unknown", "temperature": "unknown" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Phi-4 paper Table 1 (qwen2.5-72b-instruct column): livebench=55.3", "candidates": [] }, { "model_id": "gpt-4o-0513", "benchmark_id": "livebench", "score": 57.6, "reference_url": "https://arxiv.org/abs/2412.08905", "reported_setting": { "mode": "unknown", "effort": "unknown", "tools": "unknown", "sampling": "unknown", "judge": "rule-based", "harness": "Microsoft Phi internal eval", "prompt_style": "unknown", "temperature": "unknown" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Phi-4 paper Table 1 (gpt-4o-0513 column): livebench=57.6", "candidates": [] }, { "model_id": "o1-high", "benchmark_id": "aime_2025", "score": 71.4, "reference_url": "https://arxiv.org/abs/2504.21318", "reported_setting": { "mode": "unknown", "effort": "unknown", "tools": "unknown", "sampling": "unknown", "judge": "rule-based", "harness": "Microsoft Phi internal eval", "prompt_style": "unknown", "temperature": "unknown" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Phi-4-reasoning paper Table 1 (o1-high column): aime_2025=71.4", "candidates": [] }, { "model_id": "o1-high", "benchmark_id": "hmmt_feb_2025", "score": 48.3, "reference_url": "https://arxiv.org/abs/2504.21318", "reported_setting": { "mode": "unknown", "effort": "unknown", "tools": "unknown", "sampling": "unknown", "judge": "rule-based", "harness": "Microsoft Phi internal eval", "prompt_style": "unknown", "temperature": "unknown" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Phi-4-reasoning paper Table 1 (o1-high column): hmmt_2025=48.3", "candidates": [] }, { "model_id": "claude-3.7-sonnet", "benchmark_id": "hmmt_feb_2025", "score": 31.7, "reference_url": "https://arxiv.org/abs/2504.21318", "reported_setting": { "mode": "unknown", "effort": "unknown", "tools": "unknown", "sampling": "unknown", "judge": "rule-based", "harness": "Microsoft Phi internal eval", "prompt_style": "unknown", "temperature": "unknown" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Phi-4-reasoning paper Table 1 (claude-3.7-sonnet column): hmmt_2025=31.7", "candidates": [] }, { "model_id": "phi-4-reasoning", "benchmark_id": "omnimath", "score": 76.6, "reference_url": "https://arxiv.org/abs/2504.21318", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "0.8" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Phi-4-reasoning paper Table 1 (phi-4-reasoning column): omnimath=76.6", "candidates": [] }, { "model_id": "phi-4-reasoning-plus", "benchmark_id": "omnimath", "score": 81.9, "reference_url": "https://arxiv.org/abs/2504.21318", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "0.8" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Phi-4-reasoning paper Table 1 (phi-4-reasoning-plus column): omnimath=81.9", "candidates": [] }, { "model_id": "deepseek-r1-distill-llama-70b", "benchmark_id": "omnimath", "score": 63.4, "reference_url": "https://arxiv.org/abs/2504.21318", "reported_setting": { "mode": "unknown", "effort": "unknown", "tools": "unknown", "sampling": "unknown", "judge": "rule-based", "harness": "Microsoft Phi internal eval", "prompt_style": "unknown", "temperature": "unknown" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Phi-4-reasoning paper Table 1 (deepseek-r1-distill-llama-70b column): omnimath=63.4", "candidates": [] }, { "model_id": "deepseek-r1", "benchmark_id": "omnimath", "score": 85.0, "reference_url": "https://arxiv.org/abs/2504.21318", "reported_setting": { "mode": "unknown", "effort": "unknown", "tools": "unknown", "sampling": "unknown", "judge": "rule-based", "harness": "Microsoft Phi internal eval", "prompt_style": "unknown", "temperature": "unknown" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Phi-4-reasoning paper Table 1 (deepseek-r1 column): omnimath=85.0", "candidates": [] }, { "model_id": "o1-high", "benchmark_id": "omnimath", "score": 67.5, "reference_url": "https://arxiv.org/abs/2504.21318", "reported_setting": { "mode": "unknown", "effort": "unknown", "tools": "unknown", "sampling": "unknown", "judge": "rule-based", "harness": "Microsoft Phi internal eval", "prompt_style": "unknown", "temperature": "unknown" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Phi-4-reasoning paper Table 1 (o1-high column): omnimath=67.5", "candidates": [] }, { "model_id": "o3-mini-high", "benchmark_id": "omnimath", "score": 74.6, "reference_url": "https://arxiv.org/abs/2504.21318", "reported_setting": { "mode": "unknown", "effort": "unknown", "tools": "unknown", "sampling": "unknown", "judge": "rule-based", "harness": "Microsoft Phi internal eval", "prompt_style": "unknown", "temperature": "unknown" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Phi-4-reasoning paper Table 1 (o3-mini-high column): omnimath=74.6", "candidates": [] }, { "model_id": "claude-3.7-sonnet", "benchmark_id": "omnimath", "score": 54.6, "reference_url": "https://arxiv.org/abs/2504.21318", "reported_setting": { "mode": "unknown", "effort": "unknown", "tools": "unknown", "sampling": "unknown", "judge": "rule-based", "harness": "Microsoft Phi internal eval", "prompt_style": "unknown", "temperature": "unknown" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Phi-4-reasoning paper Table 1 (claude-3.7-sonnet column): omnimath=54.6", "candidates": [] }, { "model_id": "o1-high", "benchmark_id": "livecodebench", "score": 63.4, "reference_url": "https://arxiv.org/abs/2504.21318", "reported_setting": { "mode": "unknown", "effort": "unknown", "tools": "unknown", "sampling": "unknown", "judge": "rule-based", "harness": "Microsoft Phi internal eval", "prompt_style": "unknown", "temperature": "unknown" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Phi-4-reasoning paper Table 1 (o1-high column): livecodebench=63.4", "candidates": [] }, { "model_id": "o1-high", "benchmark_id": "codeforces_rating", "score": 1891, "reference_url": "https://arxiv.org/abs/2504.21318", "reported_setting": { "mode": "unknown", "effort": "unknown", "tools": "unknown", "sampling": "unknown", "judge": "rule-based", "harness": "Microsoft Phi internal eval", "prompt_style": "unknown", "temperature": "unknown" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Phi-4-reasoning paper Table 1 (o1-high column): codeforces_rating=1891", "candidates": [] }, { "model_id": "phi-4", "benchmark_id": "flenqa_3k", "score": 82.0, "reference_url": "https://arxiv.org/abs/2504.21318", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "simple-evals", "prompt_style": "default", "temperature": "0.5" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Phi-4-reasoning paper Table 2 (phi-4 column): flenqa_3k=82.0", "candidates": [] }, { "model_id": "phi-4-reasoning", "benchmark_id": "flenqa_3k", "score": 97.7, "reference_url": "https://arxiv.org/abs/2504.21318", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "0.8" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Phi-4-reasoning paper Table 2 (phi-4-reasoning column): flenqa_3k=97.7", "candidates": [] }, { "model_id": "phi-4-reasoning-plus", "benchmark_id": "flenqa_3k", "score": 97.9, "reference_url": "https://arxiv.org/abs/2504.21318", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "0.8" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Phi-4-reasoning paper Table 2 (phi-4-reasoning-plus column): flenqa_3k=97.9", "candidates": [] }, { "model_id": "o3-mini-high", "benchmark_id": "flenqa_3k", "score": 96.8, "reference_url": "https://arxiv.org/abs/2504.21318", "reported_setting": { "mode": "unknown", "effort": "unknown", "tools": "unknown", "sampling": "unknown", "judge": "rule-based", "harness": "Microsoft Phi internal eval", "prompt_style": "unknown", "temperature": "unknown" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Phi-4-reasoning paper Table 2 (o3-mini-high column): flenqa_3k=96.8", "candidates": [] }, { "model_id": "gpt-4o-0513", "benchmark_id": "flenqa_3k", "score": 90.8, "reference_url": "https://arxiv.org/abs/2504.21318", "reported_setting": { "mode": "unknown", "effort": "unknown", "tools": "unknown", "sampling": "unknown", "judge": "rule-based", "harness": "Microsoft Phi internal eval", "prompt_style": "unknown", "temperature": "unknown" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Phi-4-reasoning paper Table 2 (gpt-4o-0513 column): flenqa_3k=90.8", "candidates": [] }, { "model_id": "phi-4-reasoning", "benchmark_id": "humaneval_plus", "score": 92.9, "reference_url": "https://arxiv.org/abs/2504.21318", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "0.8" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Phi-4-reasoning paper Table 2 (phi-4-reasoning column): humaneval_plus=92.9", "candidates": [] }, { "model_id": "phi-4-reasoning-plus", "benchmark_id": "humaneval_plus", "score": 92.3, "reference_url": "https://arxiv.org/abs/2504.21318", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "0.8" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Phi-4-reasoning paper Table 2 (phi-4-reasoning-plus column): humaneval_plus=92.3", "candidates": [] }, { "model_id": "o3-mini-high", "benchmark_id": "humaneval_plus", "score": 94.0, "reference_url": "https://arxiv.org/abs/2504.21318", "reported_setting": { "mode": "unknown", "effort": "unknown", "tools": "unknown", "sampling": "unknown", "judge": "rule-based", "harness": "Microsoft Phi internal eval", "prompt_style": "unknown", "temperature": "unknown" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Phi-4-reasoning paper Table 2 (o3-mini-high column): humaneval_plus=94.0", "candidates": [] }, { "model_id": "phi-4", "benchmark_id": "phibench", "score": 58.2, "reference_url": "https://arxiv.org/abs/2504.21318", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "simple-evals", "prompt_style": "default", "temperature": "0.5" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Phi-4-reasoning paper Table 2 (phi-4 column): phibench=58.2", "candidates": [] }, { "model_id": "phi-4-reasoning", "benchmark_id": "phibench", "score": 70.6, "reference_url": "https://arxiv.org/abs/2504.21318", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "0.8" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Phi-4-reasoning paper Table 2 (phi-4-reasoning column): phibench=70.6", "candidates": [] }, { "model_id": "phi-4-reasoning-plus", "benchmark_id": "phibench", "score": 74.2, "reference_url": "https://arxiv.org/abs/2504.21318", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "0.8" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Phi-4-reasoning paper Table 2 (phi-4-reasoning-plus column): phibench=74.2", "candidates": [] }, { "model_id": "o3-mini-high", "benchmark_id": "phibench", "score": 78.0, "reference_url": "https://arxiv.org/abs/2504.21318", "reported_setting": { "mode": "unknown", "effort": "unknown", "tools": "unknown", "sampling": "unknown", "judge": "rule-based", "harness": "Microsoft Phi internal eval", "prompt_style": "unknown", "temperature": "unknown" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Phi-4-reasoning paper Table 2 (o3-mini-high column): phibench=78.0", "candidates": [] }, { "model_id": "gpt-4o-0513", "benchmark_id": "phibench", "score": 73.1, "reference_url": "https://arxiv.org/abs/2504.21318", "reported_setting": { "mode": "unknown", "effort": "unknown", "tools": "unknown", "sampling": "unknown", "judge": "rule-based", "harness": "Microsoft Phi internal eval", "prompt_style": "unknown", "temperature": "unknown" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Phi-4-reasoning paper Table 2 (gpt-4o-0513 column): phibench=73.1", "candidates": [] }, { "model_id": "phi-4-mini", "benchmark_id": "bigbench_hard", "score": 70.4, "reference_url": "https://huggingface.co/microsoft/Phi-4-mini-instruct", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "phi-4-mini HF model card: bigbench_hard=70.4", "candidates": [] }, { "model_id": "phi-4-mini", "benchmark_id": "arc_challenge", "score": 83.7, "reference_url": "https://huggingface.co/microsoft/Phi-4-mini-instruct", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "phi-4-mini HF model card: arc_challenge=83.7", "candidates": [] }, { "model_id": "phi-4-mini", "benchmark_id": "gpqa_main", "score": 25.2, "reference_url": "https://huggingface.co/microsoft/Phi-4-mini-instruct", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "phi-4-mini HF model card: gpqa=25.2", "candidates": [] }, { "model_id": "phi-4-mini", "benchmark_id": "math", "score": 64.0, "reference_url": "https://huggingface.co/microsoft/Phi-4-mini-instruct", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "phi-4-mini HF model card: math=64.0", "candidates": [] }, { "model_id": "phi-4-mini", "benchmark_id": "multilingual_mmlu", "score": 49.3, "reference_url": "https://huggingface.co/microsoft/Phi-4-mini-instruct", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "phi-4-mini HF model card: multilingual_mmlu=49.3", "candidates": [] }, { "model_id": "phi-4-mini", "benchmark_id": "mgsm", "score": 63.9, "reference_url": "https://huggingface.co/microsoft/Phi-4-mini-instruct", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "phi-4-mini HF model card: mgsm=63.9", "candidates": [] }, { "model_id": "claude-sonnet-4.5", "benchmark_id": "swelancer", "score": 45.7, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "varies", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card: claude-sonnet-4.5=45.7.", "candidates": [] }, { "model_id": "claude-opus-4.5", "benchmark_id": "swelancer", "score": 56.1, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "varies", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card: claude-opus-4.5=56.1.", "candidates": [] }, { "model_id": "gemini-3-pro", "benchmark_id": "swelancer", "score": 44.3, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "varies", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card: gemini-3-pro=44.3.", "candidates": [] }, { "model_id": "doubao-seed-2.0-pro", "benchmark_id": "swelancer", "score": 49.4, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "varies", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card: doubao-seed-2.0-pro=49.4.", "candidates": [] }, { "model_id": "gpt-5.2", "benchmark_id": "multi_swe_bench", "score": 47.7, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "varies", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card: gpt-5.2=47.7.", "candidates": [] }, { "model_id": "claude-opus-4.5", "benchmark_id": "multi_swe_bench", "score": 52.8, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "varies", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card: claude-opus-4.5=52.8.", "candidates": [] }, { "model_id": "gemini-3-pro", "benchmark_id": "multi_swe_bench", "score": 50.2, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "varies", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card: gemini-3-pro=50.2.", "candidates": [] }, { "model_id": "doubao-seed-2.0-pro", "benchmark_id": "multi_swe_bench", "score": 45.2, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "varies", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card: doubao-seed-2.0-pro=45.2.", "candidates": [] }, { "model_id": "doubao-seed-2.0-pro", "benchmark_id": "swe_bench_pro", "score": 46.9, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "varies", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card: doubao-seed-2.0-pro=46.9.", "candidates": [] }, { "model_id": "doubao-seed-2.0-pro", "benchmark_id": "swe_bench_multilingual", "score": 71.7, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "varies", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card: doubao-seed-2.0-pro=71.7.", "candidates": [] }, { "model_id": "doubao-seed-2.0-pro", "benchmark_id": "scicode", "score": 48.5, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "varies", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card: doubao-seed-2.0-pro=48.5.", "candidates": [] }, { "model_id": "gpt-5.2", "benchmark_id": "swe_evo", "score": 12.5, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "varies", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card: gpt-5.2=12.5.", "candidates": [] }, { "model_id": "claude-sonnet-4.5", "benchmark_id": "swe_evo", "score": 16.7, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "varies", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card: claude-sonnet-4.5=16.7.", "candidates": [] }, { "model_id": "claude-opus-4.5", "benchmark_id": "swe_evo", "score": 27.1, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "varies", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card: claude-opus-4.5=27.1.", "candidates": [] }, { "model_id": "gemini-3-pro", "benchmark_id": "swe_evo", "score": 8.9, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "varies", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card: gemini-3-pro=8.9.", "candidates": [] }, { "model_id": "doubao-seed-2.0-pro", "benchmark_id": "swe_evo", "score": 8.5, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "varies", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card: doubao-seed-2.0-pro=8.5.", "candidates": [] }, { "model_id": "gpt-5.2", "benchmark_id": "aider_polyglot_diff", "score": 91.1, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "varies", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card: gpt-5.2=91.1.", "candidates": [] }, { "model_id": "claude-sonnet-4.5", "benchmark_id": "aider_polyglot_diff", "score": 82.2, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "varies", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card: claude-sonnet-4.5=82.2.", "candidates": [] }, { "model_id": "claude-opus-4.5", "benchmark_id": "aider_polyglot_diff", "score": 92.4, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "varies", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card: claude-opus-4.5=92.4.", "candidates": [] }, { "model_id": "gemini-3-pro", "benchmark_id": "aider_polyglot_diff", "score": 94.2, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "varies", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card: gemini-3-pro=94.2.", "candidates": [] }, { "model_id": "doubao-seed-2.0-pro", "benchmark_id": "aider_polyglot_diff", "score": 80.0, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "varies", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card: doubao-seed-2.0-pro=80.0.", "candidates": [] }, { "model_id": "gpt-5.2", "benchmark_id": "artifactsbench", "score": 71.1, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "varies", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card: gpt-5.2=71.1.", "candidates": [] }, { "model_id": "claude-opus-4.5", "benchmark_id": "artifactsbench", "score": 68.5, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "varies", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card: claude-opus-4.5=68.5.", "candidates": [] }, { "model_id": "gemini-3-pro", "benchmark_id": "artifactsbench", "score": 58.4, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "varies", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card: gemini-3-pro=58.4.", "candidates": [] }, { "model_id": "doubao-seed-2.0-pro", "benchmark_id": "artifactsbench", "score": 66.6, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "varies", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card: doubao-seed-2.0-pro=66.6.", "candidates": [] }, { "model_id": "gpt-5.2", "benchmark_id": "codesimpleqa", "score": 62.3, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "varies", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card: gpt-5.2=62.3.", "candidates": [] }, { "model_id": "claude-sonnet-4.5", "benchmark_id": "codesimpleqa", "score": 59.6, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "varies", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card: claude-sonnet-4.5=59.6.", "candidates": [] }, { "model_id": "claude-opus-4.5", "benchmark_id": "codesimpleqa", "score": 63.0, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "varies", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card: claude-opus-4.5=63.0.", "candidates": [] }, { "model_id": "gemini-3-pro", "benchmark_id": "codesimpleqa", "score": 54.7, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "varies", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card: gemini-3-pro=54.7.", "candidates": [] }, { "model_id": "doubao-seed-2.0-pro", "benchmark_id": "codesimpleqa", "score": 58.0, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "varies", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card: doubao-seed-2.0-pro=58.0.", "candidates": [] }, { "model_id": "gpt-5.2", "benchmark_id": "spreadsheetbench_verified", "score": 69.9, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "varies", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card: gpt-5.2=69.9.", "candidates": [] }, { "model_id": "claude-sonnet-4.5", "benchmark_id": "spreadsheetbench_verified", "score": 75.9, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "varies", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card: claude-sonnet-4.5=75.9.", "candidates": [] }, { "model_id": "claude-opus-4.5", "benchmark_id": "spreadsheetbench_verified", "score": 78.6, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "varies", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card: claude-opus-4.5=78.6.", "candidates": [] }, { "model_id": "gemini-3-pro", "benchmark_id": "spreadsheetbench_verified", "score": 70.8, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "varies", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card: gemini-3-pro=70.8.", "candidates": [] }, { "model_id": "doubao-seed-2.0-pro", "benchmark_id": "spreadsheetbench_verified", "score": 79.1, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "varies", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card: doubao-seed-2.0-pro=79.1.", "candidates": [] }, { "model_id": "doubao-seed-2.0-pro", "benchmark_id": "browsecomp_zh", "score": 82.4, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "varies", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card: doubao-seed-2.0-pro=82.4.", "candidates": [] }, { "model_id": "gpt-5.2", "benchmark_id": "hle_text", "score": 45.5, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "varies", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card: gpt-5.2=45.5.", "candidates": [ { "score": 36.6, "reference_url": "https://llm-stats.com/benchmarks/humanity's-last-exam", "source_type": "third_party_aggregator", "reported_setting": { "effort": "pro" }, "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=GPT-5.2 Pro, slug=gpt-5.2-pro-2025-12-11, provider=OpenAI" } ] }, { "model_id": "claude-sonnet-4.5", "benchmark_id": "hle_text", "score": 32.0, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "varies", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card: claude-sonnet-4.5=32.0.", "candidates": [] }, { "model_id": "claude-opus-4.5", "benchmark_id": "hle_text", "score": 43.2, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "varies", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card: claude-opus-4.5=43.2.", "candidates": [] }, { "model_id": "gemini-3-pro", "benchmark_id": "hle_text", "score": 46.9, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "varies", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card: gemini-3-pro=46.9.", "candidates": [] }, { "model_id": "doubao-seed-2.0-pro", "benchmark_id": "hle_text", "score": 54.2, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "varies", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card: doubao-seed-2.0-pro=54.2.", "candidates": [] }, { "model_id": "gpt-5.2", "benchmark_id": "hle_verified", "score": 68.5, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "varies", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card: gpt-5.2=68.5.", "candidates": [] }, { "model_id": "claude-sonnet-4.5", "benchmark_id": "hle_verified", "score": 37.6, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "varies", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card: claude-sonnet-4.5=37.6.", "candidates": [] }, { "model_id": "claude-opus-4.5", "benchmark_id": "hle_verified", "score": 56.6, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "varies", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card: claude-opus-4.5=56.6.", "candidates": [] }, { "model_id": "gemini-3-pro", "benchmark_id": "hle_verified", "score": 67.5, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "varies", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card: gemini-3-pro=67.5.", "candidates": [] }, { "model_id": "doubao-seed-2.0-pro", "benchmark_id": "hle_verified", "score": 73.6, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "varies", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card: doubao-seed-2.0-pro=73.6.", "candidates": [] }, { "model_id": "gpt-5.2", "benchmark_id": "widesearch", "score": 76.8, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "varies", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card: gpt-5.2=76.8.", "candidates": [] }, { "model_id": "claude-sonnet-4.5", "benchmark_id": "widesearch", "score": 65.1, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "varies", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card: claude-sonnet-4.5=65.1.", "candidates": [] }, { "model_id": "doubao-seed-2.0-pro", "benchmark_id": "widesearch", "score": 74.7, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "varies", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card: doubao-seed-2.0-pro=74.7.", "candidates": [] }, { "model_id": "gpt-5.2", "benchmark_id": "finsearchcomp", "score": 73.8, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "varies", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card: gpt-5.2=73.8.", "candidates": [] }, { "model_id": "claude-sonnet-4.5", "benchmark_id": "finsearchcomp", "score": 58.6, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "varies", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card: claude-sonnet-4.5=58.6.", "candidates": [] }, { "model_id": "claude-opus-4.5", "benchmark_id": "finsearchcomp", "score": 66.2, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "varies", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card: claude-opus-4.5=66.2.", "candidates": [] }, { "model_id": "gemini-3-pro", "benchmark_id": "finsearchcomp", "score": 52.7, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "varies", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card: gemini-3-pro=52.7.", "candidates": [] }, { "model_id": "doubao-seed-2.0-pro", "benchmark_id": "finsearchcomp", "score": 70.2, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "varies", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card: doubao-seed-2.0-pro=70.2.", "candidates": [] }, { "model_id": "claude-sonnet-4.5", "benchmark_id": "deepsearchqa_acc", "score": 36.3, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "varies", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card: claude-sonnet-4.5=36.3.", "candidates": [] }, { "model_id": "doubao-seed-2.0-pro", "benchmark_id": "deepsearchqa_acc", "score": 77.4, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "varies", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card: doubao-seed-2.0-pro=77.4.", "candidates": [] }, { "model_id": "claude-sonnet-4.5", "benchmark_id": "seal_0", "score": 53.4, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "varies", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card: claude-sonnet-4.5=53.4.", "candidates": [] }, { "model_id": "doubao-seed-2.0-pro", "benchmark_id": "seal_0", "score": 49.5, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "varies", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card: doubao-seed-2.0-pro=49.5.", "candidates": [] }, { "model_id": "doubao-seed-2.0-pro", "benchmark_id": "tau2_bench_retail", "score": 90.4, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "varies", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card: doubao-seed-2.0-pro=90.4.", "candidates": [] }, { "model_id": "doubao-seed-2.0-pro", "benchmark_id": "tau2_bench_telecom", "score": 94.2, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "varies", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card: doubao-seed-2.0-pro=94.2.", "candidates": [] }, { "model_id": "claude-sonnet-4.5", "benchmark_id": "mcpmark", "score": 32.1, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "varies", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card: claude-sonnet-4.5=32.1.", "candidates": [ { "score": 33.3, "reference_url": "https://huggingface.co/deepseek-ai/DeepSeek-V3.2/resolve/main/assets/paper.pdf", "source_type": "tech_report", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "agentic", "sampling": "pass@1", "judge": "rule-based", "harness": "internal", "prompt_style": "default", "temperature": "1.0", "context": "128K" }, "notes": "DS V3.2 tech report Table 2: 33.3. DS uses internal MCP env (non-official). Primary (32.1) from Doubao Seed2 blog." } ] }, { "model_id": "doubao-seed-2.0-pro", "benchmark_id": "mcpmark", "score": 54.7, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "varies", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card: doubao-seed-2.0-pro=54.7.", "candidates": [] }, { "model_id": "gpt-5.2", "benchmark_id": "bfcl_v4", "score": 65.9, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "varies", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card: gpt-5.2=65.9.", "candidates": [] }, { "model_id": "claude-sonnet-4.5", "benchmark_id": "bfcl_v4", "score": 72.9, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "varies", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card: claude-sonnet-4.5=72.9.", "candidates": [] }, { "model_id": "claude-opus-4.5", "benchmark_id": "bfcl_v4", "score": 76.5, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "varies", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card: claude-opus-4.5=76.5.", "candidates": [] }, { "model_id": "gemini-3-pro", "benchmark_id": "bfcl_v4", "score": 71.0, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "varies", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card: gemini-3-pro=71.0.", "candidates": [] }, { "model_id": "doubao-seed-2.0-pro", "benchmark_id": "bfcl_v4", "score": 73.4, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "varies", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card: doubao-seed-2.0-pro=73.4.", "candidates": [] }, { "model_id": "gpt-5.2", "benchmark_id": "vitabench", "score": 41.8, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "varies", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card: gpt-5.2=41.8.", "candidates": [] }, { "model_id": "claude-sonnet-4.5", "benchmark_id": "vitabench", "score": 40.8, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "varies", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card: claude-sonnet-4.5=40.8.", "candidates": [] }, { "model_id": "claude-opus-4.5", "benchmark_id": "vitabench", "score": 55.3, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "varies", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card: claude-opus-4.5=55.3.", "candidates": [] }, { "model_id": "gemini-3-pro", "benchmark_id": "vitabench", "score": 48.8, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "varies", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card: gemini-3-pro=48.8.", "candidates": [] }, { "model_id": "doubao-seed-2.0-pro", "benchmark_id": "vitabench", "score": 47.0, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "varies", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card: doubao-seed-2.0-pro=47.0.", "candidates": [] }, { "model_id": "gpt-5.2", "benchmark_id": "deepconsult", "score": 54.3, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "varies", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card: gpt-5.2=54.3.", "candidates": [] }, { "model_id": "claude-sonnet-4.5", "benchmark_id": "deepconsult", "score": 55.8, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "varies", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card: claude-sonnet-4.5=55.8.", "candidates": [] }, { "model_id": "claude-opus-4.5", "benchmark_id": "deepconsult", "score": 61.0, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "varies", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card: claude-opus-4.5=61.0.", "candidates": [] }, { "model_id": "gemini-3-pro", "benchmark_id": "deepconsult", "score": 48.0, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "varies", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card: gemini-3-pro=48.0.", "candidates": [] }, { "model_id": "doubao-seed-2.0-pro", "benchmark_id": "deepconsult", "score": 61.1, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "varies", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card: doubao-seed-2.0-pro=61.1.", "candidates": [] }, { "model_id": "gpt-5.2", "benchmark_id": "deepresearchbench", "score": 52.2, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "varies", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card: gpt-5.2=52.2.", "candidates": [] }, { "model_id": "claude-sonnet-4.5", "benchmark_id": "deepresearchbench", "score": 47.2, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "varies", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card: claude-sonnet-4.5=47.2.", "candidates": [] }, { "model_id": "claude-opus-4.5", "benchmark_id": "deepresearchbench", "score": 50.6, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "varies", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card: claude-opus-4.5=50.6.", "candidates": [] }, { "model_id": "gemini-3-pro", "benchmark_id": "deepresearchbench", "score": 49.6, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "varies", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card: gemini-3-pro=49.6.", "candidates": [] }, { "model_id": "doubao-seed-2.0-pro", "benchmark_id": "deepresearchbench", "score": 53.3, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "varies", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card: doubao-seed-2.0-pro=53.3.", "candidates": [] }, { "model_id": "gpt-5.2", "benchmark_id": "researchrubrics", "score": 42.3, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "varies", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card: gpt-5.2=42.3.", "candidates": [] }, { "model_id": "claude-sonnet-4.5", "benchmark_id": "researchrubrics", "score": 38.6, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "varies", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card: claude-sonnet-4.5=38.6.", "candidates": [] }, { "model_id": "claude-opus-4.5", "benchmark_id": "researchrubrics", "score": 45.0, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "varies", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card: claude-opus-4.5=45.0.", "candidates": [] }, { "model_id": "gemini-3-pro", "benchmark_id": "researchrubrics", "score": 37.7, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "varies", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card: gemini-3-pro=37.7.", "candidates": [] }, { "model_id": "doubao-seed-2.0-pro", "benchmark_id": "researchrubrics", "score": 50.7, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "varies", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card: doubao-seed-2.0-pro=50.7.", "candidates": [] }, { "model_id": "gpt-5.2", "benchmark_id": "minedojo_verified", "score": 18.3, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "varies", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card: gpt-5.2=18.3.", "candidates": [] }, { "model_id": "gemini-3-pro", "benchmark_id": "minedojo_verified", "score": 23.3, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "varies", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card: gemini-3-pro=23.3.", "candidates": [] }, { "model_id": "doubao-seed-2.0-pro", "benchmark_id": "minedojo_verified", "score": 49.0, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "varies", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card: doubao-seed-2.0-pro=49.0.", "candidates": [] }, { "model_id": "gpt-5.2", "benchmark_id": "mm_browsecomp", "score": 26.3, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "varies", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card: gpt-5.2=26.3.", "candidates": [] }, { "model_id": "gemini-3-pro", "benchmark_id": "mm_browsecomp", "score": 25.0, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "varies", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card: gemini-3-pro=25.0.", "candidates": [] }, { "model_id": "doubao-seed-2.0-pro", "benchmark_id": "mm_browsecomp", "score": 48.8, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "varies", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card: doubao-seed-2.0-pro=48.8.", "candidates": [] }, { "model_id": "gpt-5.2", "benchmark_id": "hle_vl", "score": 31.0, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "varies", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card: gpt-5.2=31.0.", "candidates": [] }, { "model_id": "gemini-3-pro", "benchmark_id": "hle_vl", "score": 36.0, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "varies", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card: gemini-3-pro=36.0.", "candidates": [] }, { "model_id": "doubao-seed-2.0-pro", "benchmark_id": "hle_vl", "score": 39.2, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "varies", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card: doubao-seed-2.0-pro=39.2.", "candidates": [] }, { "model_id": "gpt-5.2", "benchmark_id": "biobench", "score": 58.1, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "varies", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card: gpt-5.2=58.1.", "candidates": [] }, { "model_id": "claude-sonnet-4.5", "benchmark_id": "biobench", "score": 44.7, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "varies", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card: claude-sonnet-4.5=44.7.", "candidates": [] }, { "model_id": "claude-opus-4.5", "benchmark_id": "biobench", "score": 49.3, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "varies", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card: claude-opus-4.5=49.3.", "candidates": [] }, { "model_id": "gemini-3-pro", "benchmark_id": "biobench", "score": 51.3, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "varies", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card: gemini-3-pro=51.3.", "candidates": [] }, { "model_id": "doubao-seed-2.0-pro", "benchmark_id": "biobench", "score": 53.5, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "varies", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card: doubao-seed-2.0-pro=53.5.", "candidates": [] }, { "model_id": "gpt-5.2", "benchmark_id": "ainstein_bench", "score": 41.3, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "varies", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card: gpt-5.2=41.3.", "candidates": [] }, { "model_id": "claude-sonnet-4.5", "benchmark_id": "ainstein_bench", "score": 33.7, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "varies", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card: claude-sonnet-4.5=33.7.", "candidates": [] }, { "model_id": "claude-opus-4.5", "benchmark_id": "ainstein_bench", "score": 44.0, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "varies", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card: claude-opus-4.5=44.0.", "candidates": [] }, { "model_id": "gemini-3-pro", "benchmark_id": "ainstein_bench", "score": 42.8, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "varies", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card: gemini-3-pro=42.8.", "candidates": [] }, { "model_id": "doubao-seed-2.0-pro", "benchmark_id": "ainstein_bench", "score": 47.7, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "varies", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card: doubao-seed-2.0-pro=47.7.", "candidates": [] }, { "model_id": "gpt-5.2", "benchmark_id": "nl2repo_bench", "score": 49.3, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "varies", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card: gpt-5.2=49.3.", "candidates": [] }, { "model_id": "claude-sonnet-4.5", "benchmark_id": "nl2repo_bench", "score": 39.9, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "varies", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card: claude-sonnet-4.5=39.9.", "candidates": [] }, { "model_id": "claude-opus-4.5", "benchmark_id": "nl2repo_bench", "score": 43.2, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "varies", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card: claude-opus-4.5=43.2.", "candidates": [] }, { "model_id": "gemini-3-pro", "benchmark_id": "nl2repo_bench", "score": 34.2, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "varies", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card: gemini-3-pro=34.2.", "candidates": [] }, { "model_id": "doubao-seed-2.0-pro", "benchmark_id": "nl2repo_bench", "score": 27.9, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "varies", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card: doubao-seed-2.0-pro=27.9.", "candidates": [] }, { "model_id": "gpt-5.2", "benchmark_id": "nl2repo_pass1", "score": 8.0, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "varies", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card: gpt-5.2=8.0.", "candidates": [] }, { "model_id": "claude-sonnet-4.5", "benchmark_id": "nl2repo_pass1", "score": 3.0, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "varies", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card: claude-sonnet-4.5=3.0.", "candidates": [] }, { "model_id": "claude-opus-4.5", "benchmark_id": "nl2repo_pass1", "score": 3.0, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "varies", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card: claude-opus-4.5=3.0.", "candidates": [] }, { "model_id": "gemini-3-pro", "benchmark_id": "nl2repo_pass1", "score": 4.0, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "varies", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card: gemini-3-pro=4.0.", "candidates": [] }, { "model_id": "doubao-seed-2.0-pro", "benchmark_id": "nl2repo_pass1", "score": 3.0, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "varies", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card: doubao-seed-2.0-pro=3.0.", "candidates": [] }, { "model_id": "gpt-5.2", "benchmark_id": "tob_complex_workflows", "score": 45.0, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "varies", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card: gpt-5.2=45.0.", "candidates": [] }, { "model_id": "claude-sonnet-4.5", "benchmark_id": "tob_complex_workflows", "score": 61.0, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "varies", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card: claude-sonnet-4.5=61.0.", "candidates": [] }, { "model_id": "claude-opus-4.5", "benchmark_id": "tob_complex_workflows", "score": 64.8, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "varies", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card: claude-opus-4.5=64.8.", "candidates": [] }, { "model_id": "gemini-3-pro", "benchmark_id": "tob_complex_workflows", "score": 69.2, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "varies", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card: gemini-3-pro=69.2.", "candidates": [] }, { "model_id": "doubao-seed-2.0-pro", "benchmark_id": "tob_complex_workflows", "score": 64.7, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "varies", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card: doubao-seed-2.0-pro=64.7.", "candidates": [] }, { "model_id": "gpt-5.2", "benchmark_id": "tob_referenceqa", "score": 63.6, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "varies", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card: gpt-5.2=63.6.", "candidates": [] }, { "model_id": "claude-sonnet-4.5", "benchmark_id": "tob_referenceqa", "score": 58.9, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "varies", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card: claude-sonnet-4.5=58.9.", "candidates": [] }, { "model_id": "claude-opus-4.5", "benchmark_id": "tob_referenceqa", "score": 67.9, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "varies", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card: claude-opus-4.5=67.9.", "candidates": [] }, { "model_id": "gemini-3-pro", "benchmark_id": "tob_referenceqa", "score": 68.3, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "varies", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card: gemini-3-pro=68.3.", "candidates": [] }, { "model_id": "doubao-seed-2.0-pro", "benchmark_id": "tob_referenceqa", "score": 72.4, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "varies", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card: doubao-seed-2.0-pro=72.4.", "candidates": [] }, { "model_id": "gpt-5.2", "benchmark_id": "gdpval_diamond", "score": 26.9, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "varies", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card: gpt-5.2=26.9.", "candidates": [] }, { "model_id": "claude-sonnet-4.5", "benchmark_id": "gdpval_diamond", "score": 15.2, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "varies", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card: claude-sonnet-4.5=15.2.", "candidates": [] }, { "model_id": "claude-opus-4.5", "benchmark_id": "gdpval_diamond", "score": 20.7, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "varies", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card: claude-opus-4.5=20.7.", "candidates": [] }, { "model_id": "gemini-3-pro", "benchmark_id": "gdpval_diamond", "score": 19.4, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "varies", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card: gemini-3-pro=19.4.", "candidates": [] }, { "model_id": "doubao-seed-2.0-pro", "benchmark_id": "gdpval_diamond", "score": 21.3, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "varies", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card: doubao-seed-2.0-pro=21.3.", "candidates": [] }, { "model_id": "gpt-5.2", "benchmark_id": "xpert_bench", "score": 53.3, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "varies", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card: gpt-5.2=53.3.", "candidates": [] }, { "model_id": "claude-sonnet-4.5", "benchmark_id": "xpert_bench", "score": 44.7, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "varies", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card: claude-sonnet-4.5=44.7.", "candidates": [] }, { "model_id": "claude-opus-4.5", "benchmark_id": "xpert_bench", "score": 50.5, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "varies", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card: claude-opus-4.5=50.5.", "candidates": [] }, { "model_id": "gemini-3-pro", "benchmark_id": "xpert_bench", "score": 53.1, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "varies", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card: gemini-3-pro=53.1.", "candidates": [] }, { "model_id": "doubao-seed-2.0-pro", "benchmark_id": "xpert_bench", "score": 64.5, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "varies", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card: doubao-seed-2.0-pro=64.5.", "candidates": [] }, { "model_id": "gpt-5.2", "benchmark_id": "tob_k12_education", "score": 61.6, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "varies", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card: gpt-5.2=61.6.", "candidates": [] }, { "model_id": "claude-sonnet-4.5", "benchmark_id": "tob_k12_education", "score": 50.1, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "varies", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card: claude-sonnet-4.5=50.1.", "candidates": [] }, { "model_id": "claude-opus-4.5", "benchmark_id": "tob_k12_education", "score": 56.2, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "varies", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card: claude-opus-4.5=56.2.", "candidates": [] }, { "model_id": "gemini-3-pro", "benchmark_id": "tob_k12_education", "score": 59.4, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "varies", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card: gemini-3-pro=59.4.", "candidates": [] }, { "model_id": "doubao-seed-2.0-pro", "benchmark_id": "tob_k12_education", "score": 62.8, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "varies", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card: doubao-seed-2.0-pro=62.8.", "candidates": [] }, { "model_id": "gpt-5.2", "benchmark_id": "tob_compositional_tasks", "score": 51.5, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "varies", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card: gpt-5.2=51.5.", "candidates": [] }, { "model_id": "claude-sonnet-4.5", "benchmark_id": "tob_compositional_tasks", "score": 57.3, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "varies", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card: claude-sonnet-4.5=57.3.", "candidates": [] }, { "model_id": "claude-opus-4.5", "benchmark_id": "tob_compositional_tasks", "score": 63.6, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "varies", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card: claude-opus-4.5=63.6.", "candidates": [] }, { "model_id": "gemini-3-pro", "benchmark_id": "tob_compositional_tasks", "score": 64.8, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "varies", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card: gemini-3-pro=64.8.", "candidates": [] }, { "model_id": "doubao-seed-2.0-pro", "benchmark_id": "tob_compositional_tasks", "score": 59.1, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "varies", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card: doubao-seed-2.0-pro=59.1.", "candidates": [] }, { "model_id": "gpt-5.2", "benchmark_id": "tob_text_classification", "score": 62.1, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "varies", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card: gpt-5.2=62.1.", "candidates": [] }, { "model_id": "claude-sonnet-4.5", "benchmark_id": "tob_text_classification", "score": 64.5, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "varies", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card: claude-sonnet-4.5=64.5.", "candidates": [] }, { "model_id": "claude-opus-4.5", "benchmark_id": "tob_text_classification", "score": 63.9, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "varies", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card: claude-opus-4.5=63.9.", "candidates": [] }, { "model_id": "gemini-3-pro", "benchmark_id": "tob_text_classification", "score": 67.5, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "varies", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card: gemini-3-pro=67.5.", "candidates": [] }, { "model_id": "doubao-seed-2.0-pro", "benchmark_id": "tob_text_classification", "score": 69.0, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "varies", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card: doubao-seed-2.0-pro=69.0.", "candidates": [] }, { "model_id": "gpt-5.2", "benchmark_id": "tob_information_extraction", "score": 44.7, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "varies", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card: gpt-5.2=44.7.", "candidates": [] }, { "model_id": "claude-sonnet-4.5", "benchmark_id": "tob_information_extraction", "score": 48.3, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "varies", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card: claude-sonnet-4.5=48.3.", "candidates": [] }, { "model_id": "claude-opus-4.5", "benchmark_id": "tob_information_extraction", "score": 50.1, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "varies", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card: claude-opus-4.5=50.1.", "candidates": [] }, { "model_id": "gemini-3-pro", "benchmark_id": "tob_information_extraction", "score": 49.0, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "varies", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card: gemini-3-pro=49.0.", "candidates": [] }, { "model_id": "doubao-seed-2.0-pro", "benchmark_id": "tob_information_extraction", "score": 52.0, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "varies", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card: doubao-seed-2.0-pro=52.0.", "candidates": [] }, { "model_id": "gpt-5.2", "benchmark_id": "world_travel_vlm", "score": 19.33, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "varies", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card: gpt-5.2=19.33.", "candidates": [] }, { "model_id": "claude-sonnet-4.5", "benchmark_id": "world_travel_vlm", "score": 2.67, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "varies", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card: claude-sonnet-4.5=2.67.", "candidates": [] }, { "model_id": "claude-opus-4.5", "benchmark_id": "world_travel_vlm", "score": 14.0, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "varies", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card: claude-opus-4.5=14.0.", "candidates": [] }, { "model_id": "gemini-3-pro", "benchmark_id": "world_travel_vlm", "score": 8.0, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "varies", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card: gemini-3-pro=8.0.", "candidates": [] }, { "model_id": "doubao-seed-2.0-pro", "benchmark_id": "world_travel_vlm", "score": 12.0, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "varies", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card: doubao-seed-2.0-pro=12.0.", "candidates": [] }, { "model_id": "gpt-5.2", "benchmark_id": "world_travel_text", "score": 32.67, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "varies", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card: gpt-5.2=32.67.", "candidates": [] }, { "model_id": "claude-sonnet-4.5", "benchmark_id": "world_travel_text", "score": 10.0, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "varies", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card: claude-sonnet-4.5=10.0.", "candidates": [] }, { "model_id": "claude-opus-4.5", "benchmark_id": "world_travel_text", "score": 21.3, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "varies", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card: claude-opus-4.5=21.3.", "candidates": [] }, { "model_id": "gemini-3-pro", "benchmark_id": "world_travel_text", "score": 14.7, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "varies", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card: gemini-3-pro=14.7.", "candidates": [] }, { "model_id": "doubao-seed-2.0-pro", "benchmark_id": "world_travel_text", "score": 23.3, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "varies", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Doubao Seed 2.0 model card: doubao-seed-2.0-pro=23.3.", "candidates": [] }, { "model_id": "phi-3-14b", "benchmark_id": "mmlu", "score": 77.9, "reference_url": "https://arxiv.org/abs/2412.08905", "reported_setting": { "mode": "unknown", "effort": "unknown", "tools": "unknown", "sampling": "unknown", "judge": "rule-based", "harness": "Microsoft Phi internal eval", "prompt_style": "unknown", "temperature": "unknown" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Phi-4 paper Table 1 baseline (phi-3-14b column): mmlu=77.9. Third-party Microsoft self-test, mc=false.", "candidates": [] }, { "model_id": "qwen-2.5-14b-instruct", "benchmark_id": "mmlu", "score": 79.9, "reference_url": "https://arxiv.org/abs/2412.08905", "reported_setting": { "mode": "unknown", "effort": "unknown", "tools": "unknown", "sampling": "unknown", "judge": "rule-based", "harness": "Microsoft Phi internal eval", "prompt_style": "unknown", "temperature": "unknown" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Phi-4 paper Table 1 baseline (qwen-2.5-14b-instruct column): mmlu=79.9. Third-party Microsoft self-test, mc=false.", "candidates": [] }, { "model_id": "llama-3.3-70b-instruct", "benchmark_id": "mmlu", "score": 86.3, "reference_url": "https://arxiv.org/abs/2412.08905", "reported_setting": { "mode": "unknown", "effort": "unknown", "tools": "unknown", "sampling": "unknown", "judge": "rule-based", "harness": "Microsoft Phi internal eval", "prompt_style": "unknown", "temperature": "unknown" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Phi-4 paper Table 1 baseline (llama-3.3-70b-instruct column): mmlu=86.3. Third-party Microsoft self-test, mc=false.", "candidates": [ { "score": 85.9, "reference_url": "https://arxiv.org/abs/2501.00656", "source_type": "tech_report", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "OLMES", "prompt_style": "default", "temperature": "default" }, "notes": "OLMo 2 paper Table 7 Instruct: alt measurement 85.9 (mc=false)." }, { "score": 86.0, "reference_url": "https://cohere.com/research/papers/command-a-technical-report.pdf", "source_type": "tech_report", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "Cohere internal eval", "prompt_style": "default", "temperature": "default" }, "notes": "Command A paper: alt measurement 86.0 (mc=false)." }, { "score": 86.0, "reference_url": "https://cohere.com/research/papers/command-a-technical-report.pdf", "source_type": "tech_report", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "Cohere internal eval", "prompt_style": "default", "temperature": "default" }, "notes": "Command A paper: alt measurement 86.0 (mc=false)." } ] }, { "model_id": "phi-3-14b", "benchmark_id": "gpqa_diamond", "score": 31.2, "reference_url": "https://arxiv.org/abs/2412.08905", "reported_setting": { "mode": "unknown", "effort": "unknown", "tools": "unknown", "sampling": "unknown", "judge": "rule-based", "harness": "Microsoft Phi internal eval", "prompt_style": "unknown", "temperature": "unknown" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Phi-4 paper Table 1 baseline (phi-3-14b column): gpqa_diamond=31.2. Third-party Microsoft self-test, mc=false.", "candidates": [] }, { "model_id": "qwen-2.5-14b-instruct", "benchmark_id": "gpqa_diamond", "score": 42.9, "reference_url": "https://arxiv.org/abs/2412.08905", "reported_setting": { "mode": "unknown", "effort": "unknown", "tools": "unknown", "sampling": "unknown", "judge": "rule-based", "harness": "Microsoft Phi internal eval", "prompt_style": "unknown", "temperature": "unknown" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Phi-4 paper Table 1 baseline (qwen-2.5-14b-instruct column): gpqa_diamond=42.9. Third-party Microsoft self-test, mc=false.", "candidates": [] }, { "model_id": "llama-3.3-70b-instruct", "benchmark_id": "gpqa_diamond", "score": 49.1, "reference_url": "https://arxiv.org/abs/2412.08905", "reported_setting": { "mode": "unknown", "effort": "unknown", "tools": "unknown", "sampling": "unknown", "judge": "rule-based", "harness": "Microsoft Phi internal eval", "prompt_style": "unknown", "temperature": "unknown" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Phi-4 paper Table 1 baseline (llama-3.3-70b-instruct column): gpqa_diamond=49.1. Third-party Microsoft self-test, mc=false.", "candidates": [] }, { "model_id": "phi-3-14b", "benchmark_id": "math_500", "score": 44.6, "reference_url": "https://arxiv.org/abs/2412.08905", "reported_setting": { "mode": "unknown", "effort": "unknown", "tools": "unknown", "sampling": "unknown", "judge": "rule-based", "harness": "Microsoft Phi internal eval", "prompt_style": "unknown", "temperature": "unknown" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Phi-4 paper Table 1 baseline (phi-3-14b column): math_500=44.6. Third-party Microsoft self-test, mc=false.", "candidates": [] }, { "model_id": "qwen-2.5-14b-instruct", "benchmark_id": "math_500", "score": 75.6, "reference_url": "https://arxiv.org/abs/2412.08905", "reported_setting": { "mode": "unknown", "effort": "unknown", "tools": "unknown", "sampling": "unknown", "judge": "rule-based", "harness": "Microsoft Phi internal eval", "prompt_style": "unknown", "temperature": "unknown" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Phi-4 paper Table 1 baseline (qwen-2.5-14b-instruct column): math_500=75.6. Third-party Microsoft self-test, mc=false.", "candidates": [] }, { "model_id": "llama-3.3-70b-instruct", "benchmark_id": "math_500", "score": 66.31, "reference_url": "https://arxiv.org/abs/2412.08905", "reported_setting": { "mode": "unknown", "effort": "unknown", "tools": "unknown", "sampling": "unknown", "judge": "rule-based", "harness": "Microsoft Phi internal eval", "prompt_style": "unknown", "temperature": "unknown" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Phi-4 paper Table 1 baseline (llama-3.3-70b-instruct column): math_500=66.31. Third-party Microsoft self-test, mc=false.", "candidates": [] }, { "model_id": "phi-3-14b", "benchmark_id": "humaneval", "score": 67.8, "reference_url": "https://arxiv.org/abs/2412.08905", "reported_setting": { "mode": "unknown", "effort": "unknown", "tools": "unknown", "sampling": "unknown", "judge": "rule-based", "harness": "Microsoft Phi internal eval", "prompt_style": "unknown", "temperature": "unknown" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Phi-4 paper Table 1 baseline (phi-3-14b column): humaneval=67.8. Third-party Microsoft self-test, mc=false.", "candidates": [] }, { "model_id": "qwen-2.5-14b-instruct", "benchmark_id": "humaneval", "score": 72.1, "reference_url": "https://arxiv.org/abs/2412.08905", "reported_setting": { "mode": "unknown", "effort": "unknown", "tools": "unknown", "sampling": "unknown", "judge": "rule-based", "harness": "Microsoft Phi internal eval", "prompt_style": "unknown", "temperature": "unknown" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Phi-4 paper Table 1 baseline (qwen-2.5-14b-instruct column): humaneval=72.1. Third-party Microsoft self-test, mc=false.", "candidates": [] }, { "model_id": "llama-3.3-70b-instruct", "benchmark_id": "humaneval", "score": 78.91, "reference_url": "https://arxiv.org/abs/2412.08905", "reported_setting": { "mode": "unknown", "effort": "unknown", "tools": "unknown", "sampling": "unknown", "judge": "rule-based", "harness": "Microsoft Phi internal eval", "prompt_style": "unknown", "temperature": "unknown" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Phi-4 paper Table 1 baseline (llama-3.3-70b-instruct column): humaneval=78.91. Third-party Microsoft self-test, mc=false.", "candidates": [ { "score": 75.5, "reference_url": "https://cohere.com/research/papers/command-a-technical-report.pdf", "source_type": "tech_report", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "Cohere internal eval", "prompt_style": "default", "temperature": "default" }, "notes": "Command A paper: alt measurement 75.5 (mc=false)." } ] }, { "model_id": "phi-3-14b", "benchmark_id": "mgsm", "score": 53.5, "reference_url": "https://arxiv.org/abs/2412.08905", "reported_setting": { "mode": "unknown", "effort": "unknown", "tools": "unknown", "sampling": "unknown", "judge": "rule-based", "harness": "Microsoft Phi internal eval", "prompt_style": "unknown", "temperature": "unknown" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Phi-4 paper Table 1 baseline (phi-3-14b column): mgsm=53.5. Third-party Microsoft self-test, mc=false.", "candidates": [] }, { "model_id": "qwen-2.5-14b-instruct", "benchmark_id": "mgsm", "score": 79.6, "reference_url": "https://arxiv.org/abs/2412.08905", "reported_setting": { "mode": "unknown", "effort": "unknown", "tools": "unknown", "sampling": "unknown", "judge": "rule-based", "harness": "Microsoft Phi internal eval", "prompt_style": "unknown", "temperature": "unknown" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Phi-4 paper Table 1 baseline (qwen-2.5-14b-instruct column): mgsm=79.6. Third-party Microsoft self-test, mc=false.", "candidates": [] }, { "model_id": "llama-3.3-70b-instruct", "benchmark_id": "mgsm", "score": 89.1, "reference_url": "https://arxiv.org/abs/2412.08905", "reported_setting": { "mode": "unknown", "effort": "unknown", "tools": "unknown", "sampling": "unknown", "judge": "rule-based", "harness": "Microsoft Phi internal eval", "prompt_style": "unknown", "temperature": "unknown" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Phi-4 paper Table 1 baseline (llama-3.3-70b-instruct column): mgsm=89.1. Third-party Microsoft self-test, mc=false.", "candidates": [] }, { "model_id": "phi-3-14b", "benchmark_id": "simpleqa", "score": 7.6, "reference_url": "https://arxiv.org/abs/2412.08905", "reported_setting": { "mode": "unknown", "effort": "unknown", "tools": "unknown", "sampling": "unknown", "judge": "rule-based", "harness": "Microsoft Phi internal eval", "prompt_style": "unknown", "temperature": "unknown" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Phi-4 paper Table 1 baseline (phi-3-14b column): simpleqa=7.6. Third-party Microsoft self-test, mc=false.", "candidates": [] }, { "model_id": "qwen-2.5-14b-instruct", "benchmark_id": "simpleqa", "score": 5.4, "reference_url": "https://arxiv.org/abs/2412.08905", "reported_setting": { "mode": "unknown", "effort": "unknown", "tools": "unknown", "sampling": "unknown", "judge": "rule-based", "harness": "Microsoft Phi internal eval", "prompt_style": "unknown", "temperature": "unknown" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Phi-4 paper Table 1 baseline (qwen-2.5-14b-instruct column): simpleqa=5.4. Third-party Microsoft self-test, mc=false.", "candidates": [] }, { "model_id": "llama-3.3-70b-instruct", "benchmark_id": "simpleqa", "score": 20.9, "reference_url": "https://arxiv.org/abs/2412.08905", "reported_setting": { "mode": "unknown", "effort": "unknown", "tools": "unknown", "sampling": "unknown", "judge": "rule-based", "harness": "Microsoft Phi internal eval", "prompt_style": "unknown", "temperature": "unknown" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Phi-4 paper Table 1 baseline (llama-3.3-70b-instruct column): simpleqa=20.9. Third-party Microsoft self-test, mc=false.", "candidates": [] }, { "model_id": "phi-3-14b", "benchmark_id": "drop", "score": 68.3, "reference_url": "https://arxiv.org/abs/2412.08905", "reported_setting": { "mode": "unknown", "effort": "unknown", "tools": "unknown", "sampling": "unknown", "judge": "rule-based", "harness": "Microsoft Phi internal eval", "prompt_style": "unknown", "temperature": "unknown" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Phi-4 paper Table 1 baseline (phi-3-14b column): drop=68.3. Third-party Microsoft self-test, mc=false.", "candidates": [] }, { "model_id": "qwen-2.5-14b-instruct", "benchmark_id": "drop", "score": 85.5, "reference_url": "https://arxiv.org/abs/2412.08905", "reported_setting": { "mode": "unknown", "effort": "unknown", "tools": "unknown", "sampling": "unknown", "judge": "rule-based", "harness": "Microsoft Phi internal eval", "prompt_style": "unknown", "temperature": "unknown" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Phi-4 paper Table 1 baseline (qwen-2.5-14b-instruct column): drop=85.5. Third-party Microsoft self-test, mc=false.", "candidates": [] }, { "model_id": "llama-3.3-70b-instruct", "benchmark_id": "drop", "score": 90.2, "reference_url": "https://arxiv.org/abs/2412.08905", "reported_setting": { "mode": "unknown", "effort": "unknown", "tools": "unknown", "sampling": "unknown", "judge": "rule-based", "harness": "Microsoft Phi internal eval", "prompt_style": "unknown", "temperature": "unknown" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Phi-4 paper Table 1 baseline (llama-3.3-70b-instruct column): drop=90.2. Third-party Microsoft self-test, mc=false.", "candidates": [ { "score": 78.0, "reference_url": "https://arxiv.org/abs/2501.00656", "source_type": "tech_report", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "OLMES", "prompt_style": "default", "temperature": "default" }, "notes": "OLMo 2 paper Table 7 Instruct: alt measurement 78.0 (mc=false)." } ] }, { "model_id": "phi-3-14b", "benchmark_id": "mmlu_pro", "score": 51.3, "reference_url": "https://arxiv.org/abs/2412.08905", "reported_setting": { "mode": "unknown", "effort": "unknown", "tools": "unknown", "sampling": "unknown", "judge": "rule-based", "harness": "Microsoft Phi internal eval", "prompt_style": "unknown", "temperature": "unknown" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Phi-4 paper Table 1 baseline (phi-3-14b column): mmlu_pro=51.3. Third-party Microsoft self-test, mc=false.", "candidates": [] }, { "model_id": "qwen-2.5-14b-instruct", "benchmark_id": "mmlu_pro", "score": 63.2, "reference_url": "https://arxiv.org/abs/2412.08905", "reported_setting": { "mode": "unknown", "effort": "unknown", "tools": "unknown", "sampling": "unknown", "judge": "rule-based", "harness": "Microsoft Phi internal eval", "prompt_style": "unknown", "temperature": "unknown" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Phi-4 paper Table 1 baseline (qwen-2.5-14b-instruct column): mmlu_pro=63.2. Third-party Microsoft self-test, mc=false.", "candidates": [] }, { "model_id": "llama-3.3-70b-instruct", "benchmark_id": "mmlu_pro", "score": 64.4, "reference_url": "https://arxiv.org/abs/2412.08905", "reported_setting": { "mode": "unknown", "effort": "unknown", "tools": "unknown", "sampling": "unknown", "judge": "rule-based", "harness": "Microsoft Phi internal eval", "prompt_style": "unknown", "temperature": "unknown" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Phi-4 paper Table 1 baseline (llama-3.3-70b-instruct column): mmlu_pro=64.4. Third-party Microsoft self-test, mc=false.", "candidates": [ { "score": 66.0, "reference_url": "https://cohere.com/research/papers/command-a-technical-report.pdf", "source_type": "tech_report", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "Cohere internal eval", "prompt_style": "default", "temperature": "default" }, "notes": "Command A paper: alt measurement 66.0 (mc=false)." } ] }, { "model_id": "phi-3-14b", "benchmark_id": "humaneval_plus", "score": 69.2, "reference_url": "https://arxiv.org/abs/2412.08905", "reported_setting": { "mode": "unknown", "effort": "unknown", "tools": "unknown", "sampling": "unknown", "judge": "rule-based", "harness": "Microsoft Phi internal eval", "prompt_style": "unknown", "temperature": "unknown" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Phi-4 paper Table 1 baseline (phi-3-14b column): humaneval_plus=69.2. Third-party Microsoft self-test, mc=false.", "candidates": [] }, { "model_id": "qwen-2.5-14b-instruct", "benchmark_id": "humaneval_plus", "score": 79.1, "reference_url": "https://arxiv.org/abs/2412.08905", "reported_setting": { "mode": "unknown", "effort": "unknown", "tools": "unknown", "sampling": "unknown", "judge": "rule-based", "harness": "Microsoft Phi internal eval", "prompt_style": "unknown", "temperature": "unknown" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Phi-4 paper Table 1 baseline (qwen-2.5-14b-instruct column): humaneval_plus=79.1. Third-party Microsoft self-test, mc=false.", "candidates": [] }, { "model_id": "llama-3.3-70b-instruct", "benchmark_id": "humaneval_plus", "score": 77.9, "reference_url": "https://arxiv.org/abs/2412.08905", "reported_setting": { "mode": "unknown", "effort": "unknown", "tools": "unknown", "sampling": "unknown", "judge": "rule-based", "harness": "Microsoft Phi internal eval", "prompt_style": "unknown", "temperature": "unknown" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Phi-4 paper Table 1 baseline (llama-3.3-70b-instruct column): humaneval_plus=77.9. Third-party Microsoft self-test, mc=false.", "candidates": [] }, { "model_id": "phi-3-14b", "benchmark_id": "arena_hard", "score": 45.8, "reference_url": "https://arxiv.org/abs/2412.08905", "reported_setting": { "mode": "unknown", "effort": "unknown", "tools": "unknown", "sampling": "unknown", "judge": "rule-based", "harness": "Microsoft Phi internal eval", "prompt_style": "unknown", "temperature": "unknown" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Phi-4 paper Table 1 baseline (phi-3-14b column): arena_hard=45.8. Third-party Microsoft self-test, mc=false.", "candidates": [] }, { "model_id": "qwen-2.5-14b-instruct", "benchmark_id": "arena_hard", "score": 70.2, "reference_url": "https://arxiv.org/abs/2412.08905", "reported_setting": { "mode": "unknown", "effort": "unknown", "tools": "unknown", "sampling": "unknown", "judge": "rule-based", "harness": "Microsoft Phi internal eval", "prompt_style": "unknown", "temperature": "unknown" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Phi-4 paper Table 1 baseline (qwen-2.5-14b-instruct column): arena_hard=70.2. Third-party Microsoft self-test, mc=false.", "candidates": [] }, { "model_id": "llama-3.3-70b-instruct", "benchmark_id": "arena_hard", "score": 65.5, "reference_url": "https://arxiv.org/abs/2412.08905", "reported_setting": { "mode": "unknown", "effort": "unknown", "tools": "unknown", "sampling": "unknown", "judge": "rule-based", "harness": "Microsoft Phi internal eval", "prompt_style": "unknown", "temperature": "unknown" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Phi-4 paper Table 1 baseline (llama-3.3-70b-instruct column): arena_hard=65.5. Third-party Microsoft self-test, mc=false.", "candidates": [] }, { "model_id": "phi-3-14b", "benchmark_id": "livebench", "score": 28.1, "reference_url": "https://arxiv.org/abs/2412.08905", "reported_setting": { "mode": "unknown", "effort": "unknown", "tools": "unknown", "sampling": "unknown", "judge": "rule-based", "harness": "Microsoft Phi internal eval", "prompt_style": "unknown", "temperature": "unknown" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Phi-4 paper Table 1 baseline (phi-3-14b column): livebench=28.1. Third-party Microsoft self-test, mc=false.", "candidates": [] }, { "model_id": "qwen-2.5-14b-instruct", "benchmark_id": "livebench", "score": 46.6, "reference_url": "https://arxiv.org/abs/2412.08905", "reported_setting": { "mode": "unknown", "effort": "unknown", "tools": "unknown", "sampling": "unknown", "judge": "rule-based", "harness": "Microsoft Phi internal eval", "prompt_style": "unknown", "temperature": "unknown" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Phi-4 paper Table 1 baseline (qwen-2.5-14b-instruct column): livebench=46.6. Third-party Microsoft self-test, mc=false.", "candidates": [] }, { "model_id": "llama-3.3-70b-instruct", "benchmark_id": "livebench", "score": 57.6, "reference_url": "https://arxiv.org/abs/2412.08905", "reported_setting": { "mode": "unknown", "effort": "unknown", "tools": "unknown", "sampling": "unknown", "judge": "rule-based", "harness": "Microsoft Phi internal eval", "prompt_style": "unknown", "temperature": "unknown" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Phi-4 paper Table 1 baseline (llama-3.3-70b-instruct column): livebench=57.6. Third-party Microsoft self-test, mc=false.", "candidates": [] }, { "model_id": "phi-3-14b", "benchmark_id": "ifeval", "score": 57.9, "reference_url": "https://arxiv.org/abs/2412.08905", "reported_setting": { "mode": "unknown", "effort": "unknown", "tools": "unknown", "sampling": "unknown", "judge": "rule-based", "harness": "Microsoft Phi internal eval", "prompt_style": "unknown", "temperature": "unknown" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Phi-4 paper Table 1 baseline (phi-3-14b column): ifeval=57.9. Third-party Microsoft self-test, mc=false.", "candidates": [] }, { "model_id": "qwen-2.5-14b-instruct", "benchmark_id": "ifeval", "score": 78.7, "reference_url": "https://arxiv.org/abs/2412.08905", "reported_setting": { "mode": "unknown", "effort": "unknown", "tools": "unknown", "sampling": "unknown", "judge": "rule-based", "harness": "Microsoft Phi internal eval", "prompt_style": "unknown", "temperature": "unknown" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Phi-4 paper Table 1 baseline (qwen-2.5-14b-instruct column): ifeval=78.7. Third-party Microsoft self-test, mc=false.", "candidates": [] }, { "model_id": "llama-3.3-70b-instruct", "benchmark_id": "ifeval", "score": 89.3, "reference_url": "https://arxiv.org/abs/2412.08905", "reported_setting": { "mode": "unknown", "effort": "unknown", "tools": "unknown", "sampling": "unknown", "judge": "rule-based", "harness": "Microsoft Phi internal eval", "prompt_style": "unknown", "temperature": "unknown" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Phi-4 paper Table 1 baseline (llama-3.3-70b-instruct column): ifeval=89.3. Third-party Microsoft self-test, mc=false.", "candidates": [ { "score": 90.8, "reference_url": "https://arxiv.org/abs/2501.00656", "source_type": "tech_report", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "OLMES", "prompt_style": "default", "temperature": "default" }, "notes": "OLMo 2 paper Table 7 Instruct: alt measurement 90.8 (mc=false)." }, { "score": 92.1, "reference_url": "https://cohere.com/research/papers/command-a-technical-report.pdf", "source_type": "tech_report", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "Cohere internal eval", "prompt_style": "default", "temperature": "default" }, "notes": "Command A paper: alt measurement 92.1 (mc=false)." }, { "score": 92.1, "reference_url": "https://cohere.com/research/papers/command-a-technical-report.pdf", "source_type": "tech_report", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "Cohere internal eval", "prompt_style": "default", "temperature": "default" }, "notes": "Command A paper: alt measurement 92.1 (mc=false)." } ] }, { "model_id": "openthinker2-32b", "benchmark_id": "aime_2024", "score": 58.0, "reference_url": "https://arxiv.org/abs/2504.21318", "reported_setting": { "mode": "unknown", "effort": "unknown", "tools": "unknown", "sampling": "unknown", "judge": "rule-based", "harness": "Microsoft Phi internal eval", "prompt_style": "unknown", "temperature": "unknown" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Phi-4-reasoning paper Table 1 baseline (openthinker2-32b column): aime_2024=58.0. Third-party Microsoft self-test, mc=false.", "candidates": [] }, { "model_id": "exaone-deep-32b", "benchmark_id": "aime_2024", "score": 72.1, "reference_url": "https://arxiv.org/abs/2504.21318", "reported_setting": { "mode": "unknown", "effort": "unknown", "tools": "unknown", "sampling": "unknown", "judge": "rule-based", "harness": "Microsoft Phi internal eval", "prompt_style": "unknown", "temperature": "unknown" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Phi-4-reasoning paper Table 1 baseline (exaone-deep-32b column): aime_2024=72.1. Third-party Microsoft self-test, mc=false.", "candidates": [] }, { "model_id": "o1-mini", "benchmark_id": "aime_2024", "score": 63.6, "reference_url": "https://arxiv.org/abs/2504.21318", "reported_setting": { "mode": "unknown", "effort": "unknown", "tools": "unknown", "sampling": "unknown", "judge": "rule-based", "harness": "Microsoft Phi internal eval", "prompt_style": "unknown", "temperature": "unknown" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Phi-4-reasoning paper Table 1 baseline (o1-mini column): aime_2024=63.6. Third-party Microsoft self-test, mc=false.", "candidates": [] }, { "model_id": "openthinker2-32b", "benchmark_id": "aime_2025", "score": 58.0, "reference_url": "https://arxiv.org/abs/2504.21318", "reported_setting": { "mode": "unknown", "effort": "unknown", "tools": "unknown", "sampling": "unknown", "judge": "rule-based", "harness": "Microsoft Phi internal eval", "prompt_style": "unknown", "temperature": "unknown" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Phi-4-reasoning paper Table 1 baseline (openthinker2-32b column): aime_2025=58.0. Third-party Microsoft self-test, mc=false.", "candidates": [] }, { "model_id": "exaone-deep-32b", "benchmark_id": "aime_2025", "score": 65.8, "reference_url": "https://arxiv.org/abs/2504.21318", "reported_setting": { "mode": "unknown", "effort": "unknown", "tools": "unknown", "sampling": "unknown", "judge": "rule-based", "harness": "Microsoft Phi internal eval", "prompt_style": "unknown", "temperature": "unknown" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Phi-4-reasoning paper Table 1 baseline (exaone-deep-32b column): aime_2025=65.8. Third-party Microsoft self-test, mc=false.", "candidates": [] }, { "model_id": "o1-mini", "benchmark_id": "aime_2025", "score": 54.8, "reference_url": "https://arxiv.org/abs/2504.21318", "reported_setting": { "mode": "unknown", "effort": "unknown", "tools": "unknown", "sampling": "unknown", "judge": "rule-based", "harness": "Microsoft Phi internal eval", "prompt_style": "unknown", "temperature": "unknown" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Phi-4-reasoning paper Table 1 baseline (o1-mini column): aime_2025=54.8. Third-party Microsoft self-test, mc=false.", "candidates": [] }, { "model_id": "o1-mini", "benchmark_id": "hmmt_feb_2025", "score": 38.0, "reference_url": "https://arxiv.org/abs/2504.21318", "reported_setting": { "mode": "unknown", "effort": "unknown", "tools": "unknown", "sampling": "unknown", "judge": "rule-based", "harness": "Microsoft Phi internal eval", "prompt_style": "unknown", "temperature": "unknown" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Phi-4-reasoning paper Table 1 baseline (o1-mini column): hmmt_2025=38.0. Third-party Microsoft self-test, mc=false.", "candidates": [] }, { "model_id": "o1-mini", "benchmark_id": "omnimath", "score": 60.5, "reference_url": "https://arxiv.org/abs/2504.21318", "reported_setting": { "mode": "unknown", "effort": "unknown", "tools": "unknown", "sampling": "unknown", "judge": "rule-based", "harness": "Microsoft Phi internal eval", "prompt_style": "unknown", "temperature": "unknown" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Phi-4-reasoning paper Table 1 baseline (o1-mini column): omnimath=60.5. Third-party Microsoft self-test, mc=false.", "candidates": [] }, { "model_id": "openthinker2-32b", "benchmark_id": "gpqa_diamond", "score": 64.1, "reference_url": "https://arxiv.org/abs/2504.21318", "reported_setting": { "mode": "unknown", "effort": "unknown", "tools": "unknown", "sampling": "unknown", "judge": "rule-based", "harness": "Microsoft Phi internal eval", "prompt_style": "unknown", "temperature": "unknown" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Phi-4-reasoning paper Table 1 baseline (openthinker2-32b column): gpqa_diamond=64.1. Third-party Microsoft self-test, mc=false.", "candidates": [] }, { "model_id": "exaone-deep-32b", "benchmark_id": "gpqa_diamond", "score": 66.1, "reference_url": "https://arxiv.org/abs/2504.21318", "reported_setting": { "mode": "unknown", "effort": "unknown", "tools": "unknown", "sampling": "unknown", "judge": "rule-based", "harness": "Microsoft Phi internal eval", "prompt_style": "unknown", "temperature": "unknown" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Phi-4-reasoning paper Table 1 baseline (exaone-deep-32b column): gpqa_diamond=66.1. Third-party Microsoft self-test, mc=false.", "candidates": [] }, { "model_id": "o1-mini", "benchmark_id": "gpqa_diamond", "score": 60.0, "reference_url": "https://arxiv.org/abs/2504.21318", "reported_setting": { "mode": "unknown", "effort": "unknown", "tools": "unknown", "sampling": "unknown", "judge": "rule-based", "harness": "Microsoft Phi internal eval", "prompt_style": "unknown", "temperature": "unknown" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Phi-4-reasoning paper Table 1 baseline (o1-mini column): gpqa_diamond=60.0. Third-party Microsoft self-test, mc=false.", "candidates": [] }, { "model_id": "exaone-deep-32b", "benchmark_id": "livecodebench", "score": 59.5, "reference_url": "https://arxiv.org/abs/2504.21318", "reported_setting": { "mode": "unknown", "effort": "unknown", "tools": "unknown", "sampling": "unknown", "judge": "rule-based", "harness": "Microsoft Phi internal eval", "prompt_style": "unknown", "temperature": "unknown" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Phi-4-reasoning paper Table 1 baseline (exaone-deep-32b column): livecodebench=59.5. Third-party Microsoft self-test, mc=false.", "candidates": [] }, { "model_id": "o1-mini", "benchmark_id": "livecodebench", "score": 53.8, "reference_url": "https://arxiv.org/abs/2504.21318", "reported_setting": { "mode": "unknown", "effort": "unknown", "tools": "unknown", "sampling": "unknown", "judge": "rule-based", "harness": "Microsoft Phi internal eval", "prompt_style": "unknown", "temperature": "unknown" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Phi-4-reasoning paper Table 1 baseline (o1-mini column): livecodebench=53.8. Third-party Microsoft self-test, mc=false.", "candidates": [] }, { "model_id": "o1-mini", "benchmark_id": "codeforces_rating", "score": 1650, "reference_url": "https://arxiv.org/abs/2504.21318", "reported_setting": { "mode": "unknown", "effort": "unknown", "tools": "unknown", "sampling": "unknown", "judge": "rule-based", "harness": "Microsoft Phi internal eval", "prompt_style": "unknown", "temperature": "unknown" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Phi-4-reasoning paper Table 1 baseline (o1-mini column): codeforces_rating=1650. Third-party Microsoft self-test, mc=false.", "candidates": [] }, { "model_id": "gpt-3.5-turbo", "benchmark_id": "alpacaeval_2", "score": 38.7, "reference_url": "https://arxiv.org/abs/2501.00656", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "OLMES", "prompt_style": "default", "temperature": "default" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "OLMo 2 paper Table 7 Instruct (gpt-3.5-turbo column): alpacaeval_2=38.7. Third-party Allen AI self-test, mc=false.", "candidates": [] }, { "model_id": "gpt-3.5-turbo", "benchmark_id": "bigbench_hard", "score": 66.6, "reference_url": "https://arxiv.org/abs/2501.00656", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "OLMES", "prompt_style": "default", "temperature": "default" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "OLMo 2 paper Table 7 Instruct (gpt-3.5-turbo column): bigbench_hard=66.6. Third-party Allen AI self-test, mc=false.", "candidates": [] }, { "model_id": "gpt-3.5-turbo", "benchmark_id": "drop", "score": 70.2, "reference_url": "https://arxiv.org/abs/2501.00656", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "OLMES", "prompt_style": "default", "temperature": "default" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "OLMo 2 paper Table 7 Instruct (gpt-3.5-turbo column): drop=70.2. Third-party Allen AI self-test, mc=false.", "candidates": [] }, { "model_id": "gpt-3.5-turbo", "benchmark_id": "gsm8k", "score": 74.3, "reference_url": "https://arxiv.org/abs/2501.00656", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "OLMES", "prompt_style": "default", "temperature": "default" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "OLMo 2 paper Table 7 Instruct (gpt-3.5-turbo column): gsm8k=74.3. Third-party Allen AI self-test, mc=false.", "candidates": [] }, { "model_id": "gpt-3.5-turbo", "benchmark_id": "ifeval", "score": 66.9, "reference_url": "https://arxiv.org/abs/2501.00656", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "OLMES", "prompt_style": "default", "temperature": "default" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "OLMo 2 paper Table 7 Instruct (gpt-3.5-turbo column): ifeval=66.9. Third-party Allen AI self-test, mc=false.", "candidates": [] }, { "model_id": "gpt-3.5-turbo", "benchmark_id": "math", "score": 41.2, "reference_url": "https://arxiv.org/abs/2501.00656", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "OLMES", "prompt_style": "default", "temperature": "default" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "OLMo 2 paper Table 7 Instruct (gpt-3.5-turbo column): math=41.2. Third-party Allen AI self-test, mc=false.", "candidates": [] }, { "model_id": "gpt-3.5-turbo", "benchmark_id": "mmlu", "score": 70.2, "reference_url": "https://arxiv.org/abs/2501.00656", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "OLMES", "prompt_style": "default", "temperature": "default" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "OLMo 2 paper Table 7 Instruct (gpt-3.5-turbo column): mmlu=70.2. Third-party Allen AI self-test, mc=false.", "candidates": [] }, { "model_id": "gpt-3.5-turbo", "benchmark_id": "safety", "score": 69.1, "reference_url": "https://arxiv.org/abs/2501.00656", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "OLMES", "prompt_style": "default", "temperature": "default" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "OLMo 2 paper Table 7 Instruct (gpt-3.5-turbo column): safety=69.1. Third-party Allen AI self-test, mc=false.", "candidates": [] }, { "model_id": "gpt-3.5-turbo", "benchmark_id": "popqa", "score": 45.0, "reference_url": "https://arxiv.org/abs/2501.00656", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "OLMES", "prompt_style": "default", "temperature": "default" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "OLMo 2 paper Table 7 Instruct (gpt-3.5-turbo column): popqa=45.0. Third-party Allen AI self-test, mc=false.", "candidates": [] }, { "model_id": "gpt-3.5-turbo", "benchmark_id": "truthfulqa", "score": 62.9, "reference_url": "https://arxiv.org/abs/2501.00656", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "OLMES", "prompt_style": "default", "temperature": "default" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "OLMo 2 paper Table 7 Instruct (gpt-3.5-turbo column): truthfulqa=62.9. Third-party Allen AI self-test, mc=false.", "candidates": [] }, { "model_id": "gpt-4o-mini", "benchmark_id": "alpacaeval_2", "score": 49.7, "reference_url": "https://arxiv.org/abs/2501.00656", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "OLMES", "prompt_style": "default", "temperature": "default" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "OLMo 2 paper Table 7 Instruct (gpt-4o-mini column): alpacaeval_2=49.7. Third-party Allen AI self-test, mc=false.", "candidates": [] }, { "model_id": "gpt-4o-mini", "benchmark_id": "bigbench_hard", "score": 65.9, "reference_url": "https://arxiv.org/abs/2501.00656", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "OLMES", "prompt_style": "default", "temperature": "default" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "OLMo 2 paper Table 7 Instruct (gpt-4o-mini column): bigbench_hard=65.9. Third-party Allen AI self-test, mc=false.", "candidates": [] }, { "model_id": "gpt-4o-mini", "benchmark_id": "gsm8k", "score": 83.0, "reference_url": "https://arxiv.org/abs/2501.00656", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "OLMES", "prompt_style": "default", "temperature": "default" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "OLMo 2 paper Table 7 Instruct (gpt-4o-mini column): gsm8k=83.0. Third-party Allen AI self-test, mc=false.", "candidates": [] }, { "model_id": "gpt-4o-mini", "benchmark_id": "math", "score": 67.9, "reference_url": "https://arxiv.org/abs/2501.00656", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "OLMES", "prompt_style": "default", "temperature": "default" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "OLMo 2 paper Table 7 Instruct (gpt-4o-mini column): math=67.9. Third-party Allen AI self-test, mc=false.", "candidates": [] }, { "model_id": "gpt-4o-mini", "benchmark_id": "safety", "score": 84.9, "reference_url": "https://arxiv.org/abs/2501.00656", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "OLMES", "prompt_style": "default", "temperature": "default" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "OLMo 2 paper Table 7 Instruct (gpt-4o-mini column): safety=84.9. Third-party Allen AI self-test, mc=false.", "candidates": [] }, { "model_id": "gpt-4o-mini", "benchmark_id": "popqa", "score": 39.0, "reference_url": "https://arxiv.org/abs/2501.00656", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "OLMES", "prompt_style": "default", "temperature": "default" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "OLMo 2 paper Table 7 Instruct (gpt-4o-mini column): popqa=39.0. Third-party Allen AI self-test, mc=false.", "candidates": [] }, { "model_id": "gpt-4o-mini", "benchmark_id": "truthfulqa", "score": 64.8, "reference_url": "https://arxiv.org/abs/2501.00656", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "OLMES", "prompt_style": "default", "temperature": "default" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "OLMo 2 paper Table 7 Instruct (gpt-4o-mini column): truthfulqa=64.8. Third-party Allen AI self-test, mc=false.", "candidates": [] }, { "model_id": "gemma-3-1b", "benchmark_id": "alpacaeval_2", "score": 20.4, "reference_url": "https://arxiv.org/abs/2501.00656", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "OLMES", "prompt_style": "default", "temperature": "default" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "OLMo 2 paper Table 7 Instruct (gemma-3-1b column): alpacaeval_2=20.4. Third-party Allen AI self-test, mc=false.", "candidates": [] }, { "model_id": "gemma-3-1b", "benchmark_id": "bigbench_hard", "score": 39.4, "reference_url": "https://arxiv.org/abs/2501.00656", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "OLMES", "prompt_style": "default", "temperature": "default" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "OLMo 2 paper Table 7 Instruct (gemma-3-1b column): bigbench_hard=39.4. Third-party Allen AI self-test, mc=false.", "candidates": [] }, { "model_id": "gemma-3-1b", "benchmark_id": "drop", "score": 25.1, "reference_url": "https://arxiv.org/abs/2501.00656", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "OLMES", "prompt_style": "default", "temperature": "default" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "OLMo 2 paper Table 7 Instruct (gemma-3-1b column): drop=25.1. Third-party Allen AI self-test, mc=false.", "candidates": [] }, { "model_id": "gemma-3-1b", "benchmark_id": "gsm8k", "score": 35.0, "reference_url": "https://arxiv.org/abs/2501.00656", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "OLMES", "prompt_style": "default", "temperature": "default" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "OLMo 2 paper Table 7 Instruct (gemma-3-1b column): gsm8k=35.0. Third-party Allen AI self-test, mc=false.", "candidates": [] }, { "model_id": "gemma-3-1b", "benchmark_id": "ifeval", "score": 60.6, "reference_url": "https://arxiv.org/abs/2501.00656", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "OLMES", "prompt_style": "default", "temperature": "default" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "OLMo 2 paper Table 7 Instruct (gemma-3-1b column): ifeval=60.6. Third-party Allen AI self-test, mc=false.", "candidates": [] }, { "model_id": "gemma-3-1b", "benchmark_id": "mmlu", "score": 38.9, "reference_url": "https://arxiv.org/abs/2501.00656", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "OLMES", "prompt_style": "default", "temperature": "default" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "OLMo 2 paper Table 7 Instruct (gemma-3-1b column): mmlu=38.9. Third-party Allen AI self-test, mc=false.", "candidates": [] }, { "model_id": "gemma-3-1b", "benchmark_id": "safety", "score": 70.2, "reference_url": "https://arxiv.org/abs/2501.00656", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "OLMES", "prompt_style": "default", "temperature": "default" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "OLMo 2 paper Table 7 Instruct (gemma-3-1b column): safety=70.2. Third-party Allen AI self-test, mc=false.", "candidates": [] }, { "model_id": "gemma-3-1b", "benchmark_id": "popqa", "score": 9.6, "reference_url": "https://arxiv.org/abs/2501.00656", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "OLMES", "prompt_style": "default", "temperature": "default" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "OLMo 2 paper Table 7 Instruct (gemma-3-1b column): popqa=9.6. Third-party Allen AI self-test, mc=false.", "candidates": [] }, { "model_id": "gemma-3-1b", "benchmark_id": "truthfulqa", "score": 43.8, "reference_url": "https://arxiv.org/abs/2501.00656", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "OLMES", "prompt_style": "default", "temperature": "default" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "OLMo 2 paper Table 7 Instruct (gemma-3-1b column): truthfulqa=43.8. Third-party Allen AI self-test, mc=false.", "candidates": [] }, { "model_id": "llama-3.2-1b", "benchmark_id": "alpacaeval_2", "score": 10.1, "reference_url": "https://arxiv.org/abs/2501.00656", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "OLMES", "prompt_style": "default", "temperature": "default" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "OLMo 2 paper Table 7 Instruct (llama-3.2-1b column): alpacaeval_2=10.1. Third-party Allen AI self-test, mc=false.", "candidates": [] }, { "model_id": "llama-3.2-1b", "benchmark_id": "bigbench_hard", "score": 40.2, "reference_url": "https://arxiv.org/abs/2501.00656", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "OLMES", "prompt_style": "default", "temperature": "default" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "OLMo 2 paper Table 7 Instruct (llama-3.2-1b column): bigbench_hard=40.2. Third-party Allen AI self-test, mc=false.", "candidates": [] }, { "model_id": "llama-3.2-1b", "benchmark_id": "drop", "score": 32.2, "reference_url": "https://arxiv.org/abs/2501.00656", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "OLMES", "prompt_style": "default", "temperature": "default" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "OLMo 2 paper Table 7 Instruct (llama-3.2-1b column): drop=32.2. Third-party Allen AI self-test, mc=false.", "candidates": [] }, { "model_id": "llama-3.2-1b", "benchmark_id": "gsm8k", "score": 45.4, "reference_url": "https://arxiv.org/abs/2501.00656", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "OLMES", "prompt_style": "default", "temperature": "default" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "OLMo 2 paper Table 7 Instruct (llama-3.2-1b column): gsm8k=45.4. Third-party Allen AI self-test, mc=false.", "candidates": [] }, { "model_id": "llama-3.2-1b", "benchmark_id": "ifeval", "score": 54.0, "reference_url": "https://arxiv.org/abs/2501.00656", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "OLMES", "prompt_style": "default", "temperature": "default" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "OLMo 2 paper Table 7 Instruct (llama-3.2-1b column): ifeval=54.0. Third-party Allen AI self-test, mc=false.", "candidates": [] }, { "model_id": "llama-3.2-1b", "benchmark_id": "math", "score": 21.6, "reference_url": "https://arxiv.org/abs/2501.00656", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "OLMES", "prompt_style": "default", "temperature": "default" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "OLMo 2 paper Table 7 Instruct (llama-3.2-1b column): math=21.6. Third-party Allen AI self-test, mc=false.", "candidates": [] }, { "model_id": "llama-3.2-1b", "benchmark_id": "mmlu", "score": 46.7, "reference_url": "https://arxiv.org/abs/2501.00656", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "OLMES", "prompt_style": "default", "temperature": "default" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "OLMo 2 paper Table 7 Instruct (llama-3.2-1b column): mmlu=46.7. Third-party Allen AI self-test, mc=false.", "candidates": [] }, { "model_id": "llama-3.2-1b", "benchmark_id": "safety", "score": 87.2, "reference_url": "https://arxiv.org/abs/2501.00656", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "OLMES", "prompt_style": "default", "temperature": "default" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "OLMo 2 paper Table 7 Instruct (llama-3.2-1b column): safety=87.2. Third-party Allen AI self-test, mc=false.", "candidates": [] }, { "model_id": "llama-3.2-1b", "benchmark_id": "popqa", "score": 13.8, "reference_url": "https://arxiv.org/abs/2501.00656", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "OLMES", "prompt_style": "default", "temperature": "default" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "OLMo 2 paper Table 7 Instruct (llama-3.2-1b column): popqa=13.8. Third-party Allen AI self-test, mc=false.", "candidates": [] }, { "model_id": "llama-3.2-1b", "benchmark_id": "truthfulqa", "score": 41.5, "reference_url": "https://arxiv.org/abs/2501.00656", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "OLMES", "prompt_style": "default", "temperature": "default" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "OLMo 2 paper Table 7 Instruct (llama-3.2-1b column): truthfulqa=41.5. Third-party Allen AI self-test, mc=false.", "candidates": [] }, { "model_id": "qwen-2.5-1.5b", "benchmark_id": "alpacaeval_2", "score": 7.4, "reference_url": "https://arxiv.org/abs/2501.00656", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "OLMES", "prompt_style": "default", "temperature": "default" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "OLMo 2 paper Table 7 Instruct (qwen-2.5-1.5b column): alpacaeval_2=7.4. Third-party Allen AI self-test, mc=false.", "candidates": [] }, { "model_id": "qwen-2.5-1.5b", "benchmark_id": "bigbench_hard", "score": 45.8, "reference_url": "https://arxiv.org/abs/2501.00656", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "OLMES", "prompt_style": "default", "temperature": "default" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "OLMo 2 paper Table 7 Instruct (qwen-2.5-1.5b column): bigbench_hard=45.8. Third-party Allen AI self-test, mc=false.", "candidates": [] }, { "model_id": "qwen-2.5-1.5b", "benchmark_id": "drop", "score": 13.4, "reference_url": "https://arxiv.org/abs/2501.00656", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "OLMES", "prompt_style": "default", "temperature": "default" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "OLMo 2 paper Table 7 Instruct (qwen-2.5-1.5b column): drop=13.4. Third-party Allen AI self-test, mc=false.", "candidates": [] }, { "model_id": "qwen-2.5-1.5b", "benchmark_id": "gsm8k", "score": 66.2, "reference_url": "https://arxiv.org/abs/2501.00656", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "OLMES", "prompt_style": "default", "temperature": "default" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "OLMo 2 paper Table 7 Instruct (qwen-2.5-1.5b column): gsm8k=66.2. Third-party Allen AI self-test, mc=false.", "candidates": [] }, { "model_id": "qwen-2.5-1.5b", "benchmark_id": "ifeval", "score": 44.2, "reference_url": "https://arxiv.org/abs/2501.00656", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "OLMES", "prompt_style": "default", "temperature": "default" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "OLMo 2 paper Table 7 Instruct (qwen-2.5-1.5b column): ifeval=44.2. Third-party Allen AI self-test, mc=false.", "candidates": [] }, { "model_id": "qwen-2.5-1.5b", "benchmark_id": "math", "score": 40.6, "reference_url": "https://arxiv.org/abs/2501.00656", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "OLMES", "prompt_style": "default", "temperature": "default" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "OLMo 2 paper Table 7 Instruct (qwen-2.5-1.5b column): math=40.6. Third-party Allen AI self-test, mc=false.", "candidates": [] }, { "model_id": "qwen-2.5-1.5b", "benchmark_id": "mmlu", "score": 59.7, "reference_url": "https://arxiv.org/abs/2501.00656", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "OLMES", "prompt_style": "default", "temperature": "default" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "OLMo 2 paper Table 7 Instruct (qwen-2.5-1.5b column): mmlu=59.7. Third-party Allen AI self-test, mc=false.", "candidates": [] }, { "model_id": "qwen-2.5-1.5b", "benchmark_id": "safety", "score": 77.6, "reference_url": "https://arxiv.org/abs/2501.00656", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "OLMES", "prompt_style": "default", "temperature": "default" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "OLMo 2 paper Table 7 Instruct (qwen-2.5-1.5b column): safety=77.6. Third-party Allen AI self-test, mc=false.", "candidates": [] }, { "model_id": "qwen-2.5-1.5b", "benchmark_id": "popqa", "score": 15.5, "reference_url": "https://arxiv.org/abs/2501.00656", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "OLMES", "prompt_style": "default", "temperature": "default" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "OLMo 2 paper Table 7 Instruct (qwen-2.5-1.5b column): popqa=15.5. Third-party Allen AI self-test, mc=false.", "candidates": [] }, { "model_id": "qwen-2.5-1.5b", "benchmark_id": "truthfulqa", "score": 46.5, "reference_url": "https://arxiv.org/abs/2501.00656", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "OLMES", "prompt_style": "default", "temperature": "default" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "OLMo 2 paper Table 7 Instruct (qwen-2.5-1.5b column): truthfulqa=46.5. Third-party Allen AI self-test, mc=false.", "candidates": [] }, { "model_id": "ministral-8b", "benchmark_id": "alpacaeval_2", "score": 31.4, "reference_url": "https://arxiv.org/abs/2501.00656", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "OLMES", "prompt_style": "default", "temperature": "default" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "OLMo 2 paper Table 7 Instruct (ministral-8b column): alpacaeval_2=31.4. Third-party Allen AI self-test, mc=false.", "candidates": [] }, { "model_id": "ministral-8b", "benchmark_id": "bigbench_hard", "score": 70.8, "reference_url": "https://arxiv.org/abs/2501.00656", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "OLMES", "prompt_style": "default", "temperature": "default" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "OLMo 2 paper Table 7 Instruct (ministral-8b column): bigbench_hard=70.8. Third-party Allen AI self-test, mc=false.", "candidates": [] }, { "model_id": "ministral-8b", "benchmark_id": "drop", "score": 56.2, "reference_url": "https://arxiv.org/abs/2501.00656", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "OLMES", "prompt_style": "default", "temperature": "default" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "OLMo 2 paper Table 7 Instruct (ministral-8b column): drop=56.2. Third-party Allen AI self-test, mc=false.", "candidates": [] }, { "model_id": "ministral-8b", "benchmark_id": "gsm8k", "score": 80.0, "reference_url": "https://arxiv.org/abs/2501.00656", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "OLMES", "prompt_style": "default", "temperature": "default" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "OLMo 2 paper Table 7 Instruct (ministral-8b column): gsm8k=80.0. Third-party Allen AI self-test, mc=false.", "candidates": [] }, { "model_id": "ministral-8b", "benchmark_id": "ifeval", "score": 56.4, "reference_url": "https://arxiv.org/abs/2501.00656", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "OLMES", "prompt_style": "default", "temperature": "default" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "OLMo 2 paper Table 7 Instruct (ministral-8b column): ifeval=56.4. Third-party Allen AI self-test, mc=false.", "candidates": [ { "score": 59.0, "reference_url": "https://cohere.com/research/papers/command-a-technical-report.pdf", "source_type": "tech_report", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "Cohere internal eval", "prompt_style": "default", "temperature": "default" }, "notes": "Command A paper: alt measurement 59.0 (mc=false)." }, { "score": 59.0, "reference_url": "https://cohere.com/research/papers/command-a-technical-report.pdf", "source_type": "tech_report", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "Cohere internal eval", "prompt_style": "default", "temperature": "default" }, "notes": "Command A paper: alt measurement 59.0 (mc=false)." } ] }, { "model_id": "ministral-8b", "benchmark_id": "math", "score": 40.0, "reference_url": "https://arxiv.org/abs/2501.00656", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "OLMES", "prompt_style": "default", "temperature": "default" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "OLMo 2 paper Table 7 Instruct (ministral-8b column): math=40.0. Third-party Allen AI self-test, mc=false.", "candidates": [ { "score": 54.5, "reference_url": "https://cohere.com/research/papers/command-a-technical-report.pdf", "source_type": "tech_report", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "Cohere internal eval", "prompt_style": "default", "temperature": "default" }, "notes": "Command A paper: alt measurement 54.5 (mc=false)." } ] }, { "model_id": "ministral-8b", "benchmark_id": "mmlu", "score": 68.5, "reference_url": "https://arxiv.org/abs/2501.00656", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "OLMES", "prompt_style": "default", "temperature": "default" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "OLMo 2 paper Table 7 Instruct (ministral-8b column): mmlu=68.5. Third-party Allen AI self-test, mc=false.", "candidates": [ { "score": 71.1, "reference_url": "https://cohere.com/research/papers/command-a-technical-report.pdf", "source_type": "tech_report", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "Cohere internal eval", "prompt_style": "default", "temperature": "default" }, "notes": "Command A paper: alt measurement 71.1 (mc=false)." }, { "score": 71.1, "reference_url": "https://cohere.com/research/papers/command-a-technical-report.pdf", "source_type": "tech_report", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "Cohere internal eval", "prompt_style": "default", "temperature": "default" }, "notes": "Command A paper: alt measurement 71.1 (mc=false)." } ] }, { "model_id": "ministral-8b", "benchmark_id": "safety", "score": 56.2, "reference_url": "https://arxiv.org/abs/2501.00656", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "OLMES", "prompt_style": "default", "temperature": "default" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "OLMo 2 paper Table 7 Instruct (ministral-8b column): safety=56.2. Third-party Allen AI self-test, mc=false.", "candidates": [] }, { "model_id": "ministral-8b", "benchmark_id": "popqa", "score": 20.2, "reference_url": "https://arxiv.org/abs/2501.00656", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "OLMES", "prompt_style": "default", "temperature": "default" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "OLMo 2 paper Table 7 Instruct (ministral-8b column): popqa=20.2. Third-party Allen AI self-test, mc=false.", "candidates": [] }, { "model_id": "ministral-8b", "benchmark_id": "truthfulqa", "score": 55.5, "reference_url": "https://arxiv.org/abs/2501.00656", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "OLMES", "prompt_style": "default", "temperature": "default" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "OLMo 2 paper Table 7 Instruct (ministral-8b column): truthfulqa=55.5. Third-party Allen AI self-test, mc=false.", "candidates": [] }, { "model_id": "llama-3.1-8b", "benchmark_id": "alpacaeval_2", "score": 25.8, "reference_url": "https://arxiv.org/abs/2501.00656", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "OLMES", "prompt_style": "default", "temperature": "default" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "OLMo 2 paper Table 7 Instruct (llama-3.1-8b column): alpacaeval_2=25.8. Third-party Allen AI self-test, mc=false.", "candidates": [] }, { "model_id": "llama-3.1-8b", "benchmark_id": "bigbench_hard", "score": 71.9, "reference_url": "https://arxiv.org/abs/2501.00656", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "OLMES", "prompt_style": "default", "temperature": "default" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "OLMo 2 paper Table 7 Instruct (llama-3.1-8b column): bigbench_hard=71.9. Third-party Allen AI self-test, mc=false.", "candidates": [] }, { "model_id": "llama-3.1-8b", "benchmark_id": "drop", "score": 61.7, "reference_url": "https://arxiv.org/abs/2501.00656", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "OLMES", "prompt_style": "default", "temperature": "default" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "OLMo 2 paper Table 7 Instruct (llama-3.1-8b column): drop=61.7. Third-party Allen AI self-test, mc=false.", "candidates": [] }, { "model_id": "llama-3.1-8b", "benchmark_id": "gsm8k", "score": 83.4, "reference_url": "https://arxiv.org/abs/2501.00656", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "OLMES", "prompt_style": "default", "temperature": "default" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "OLMo 2 paper Table 7 Instruct (llama-3.1-8b column): gsm8k=83.4. Third-party Allen AI self-test, mc=false.", "candidates": [] }, { "model_id": "llama-3.1-8b", "benchmark_id": "ifeval", "score": 80.6, "reference_url": "https://arxiv.org/abs/2501.00656", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "OLMES", "prompt_style": "default", "temperature": "default" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "OLMo 2 paper Table 7 Instruct (llama-3.1-8b column): ifeval=80.6. Third-party Allen AI self-test, mc=false.", "candidates": [ { "score": 78.6, "reference_url": "https://cohere.com/research/papers/command-a-technical-report.pdf", "source_type": "tech_report", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "Cohere internal eval", "prompt_style": "default", "temperature": "default" }, "notes": "Command A paper: alt measurement 78.6 (mc=false)." }, { "score": 78.6, "reference_url": "https://cohere.com/research/papers/command-a-technical-report.pdf", "source_type": "tech_report", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "Cohere internal eval", "prompt_style": "default", "temperature": "default" }, "notes": "Command A paper: alt measurement 78.6 (mc=false)." } ] }, { "model_id": "llama-3.1-8b", "benchmark_id": "math", "score": 42.5, "reference_url": "https://arxiv.org/abs/2501.00656", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "OLMES", "prompt_style": "default", "temperature": "default" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "OLMo 2 paper Table 7 Instruct (llama-3.1-8b column): math=42.5. Third-party Allen AI self-test, mc=false.", "candidates": [ { "score": 51.9, "reference_url": "https://cohere.com/research/papers/command-a-technical-report.pdf", "source_type": "tech_report", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "Cohere internal eval", "prompt_style": "default", "temperature": "default" }, "notes": "Command A paper: alt measurement 51.9 (mc=false)." } ] }, { "model_id": "llama-3.1-8b", "benchmark_id": "mmlu", "score": 71.3, "reference_url": "https://arxiv.org/abs/2501.00656", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "OLMES", "prompt_style": "default", "temperature": "default" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "OLMo 2 paper Table 7 Instruct (llama-3.1-8b column): mmlu=71.3. Third-party Allen AI self-test, mc=false.", "candidates": [ { "score": 71.1, "reference_url": "https://cohere.com/research/papers/command-a-technical-report.pdf", "source_type": "tech_report", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "Cohere internal eval", "prompt_style": "default", "temperature": "default" }, "notes": "Command A paper: alt measurement 71.1 (mc=false)." }, { "score": 71.1, "reference_url": "https://cohere.com/research/papers/command-a-technical-report.pdf", "source_type": "tech_report", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "Cohere internal eval", "prompt_style": "default", "temperature": "default" }, "notes": "Command A paper: alt measurement 71.1 (mc=false)." } ] }, { "model_id": "llama-3.1-8b", "benchmark_id": "safety", "score": 70.2, "reference_url": "https://arxiv.org/abs/2501.00656", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "OLMES", "prompt_style": "default", "temperature": "default" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "OLMo 2 paper Table 7 Instruct (llama-3.1-8b column): safety=70.2. Third-party Allen AI self-test, mc=false.", "candidates": [] }, { "model_id": "llama-3.1-8b", "benchmark_id": "popqa", "score": 28.4, "reference_url": "https://arxiv.org/abs/2501.00656", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "OLMES", "prompt_style": "default", "temperature": "default" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "OLMo 2 paper Table 7 Instruct (llama-3.1-8b column): popqa=28.4. Third-party Allen AI self-test, mc=false.", "candidates": [] }, { "model_id": "llama-3.1-8b", "benchmark_id": "truthfulqa", "score": 55.1, "reference_url": "https://arxiv.org/abs/2501.00656", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "OLMES", "prompt_style": "default", "temperature": "default" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "OLMo 2 paper Table 7 Instruct (llama-3.1-8b column): truthfulqa=55.1. Third-party Allen AI self-test, mc=false.", "candidates": [] }, { "model_id": "tulu-3-8b", "benchmark_id": "alpacaeval_2", "score": 34.0, "reference_url": "https://arxiv.org/abs/2501.00656", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "OLMES", "prompt_style": "default", "temperature": "default" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "OLMo 2 paper Table 7 Instruct (tulu-3-8b column): alpacaeval_2=34.0. Third-party Allen AI self-test, mc=false.", "candidates": [] }, { "model_id": "tulu-3-8b", "benchmark_id": "bigbench_hard", "score": 69.0, "reference_url": "https://arxiv.org/abs/2501.00656", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "OLMES", "prompt_style": "default", "temperature": "default" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "OLMo 2 paper Table 7 Instruct (tulu-3-8b column): bigbench_hard=69.0. Third-party Allen AI self-test, mc=false.", "candidates": [] }, { "model_id": "tulu-3-8b", "benchmark_id": "drop", "score": 62.6, "reference_url": "https://arxiv.org/abs/2501.00656", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "OLMES", "prompt_style": "default", "temperature": "default" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "OLMo 2 paper Table 7 Instruct (tulu-3-8b column): drop=62.6. Third-party Allen AI self-test, mc=false.", "candidates": [] }, { "model_id": "tulu-3-8b", "benchmark_id": "gsm8k", "score": 87.6, "reference_url": "https://arxiv.org/abs/2501.00656", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "OLMES", "prompt_style": "default", "temperature": "default" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "OLMo 2 paper Table 7 Instruct (tulu-3-8b column): gsm8k=87.6. Third-party Allen AI self-test, mc=false.", "candidates": [] }, { "model_id": "tulu-3-8b", "benchmark_id": "ifeval", "score": 82.4, "reference_url": "https://arxiv.org/abs/2501.00656", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "OLMES", "prompt_style": "default", "temperature": "default" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "OLMo 2 paper Table 7 Instruct (tulu-3-8b column): ifeval=82.4. Third-party Allen AI self-test, mc=false.", "candidates": [] }, { "model_id": "tulu-3-8b", "benchmark_id": "math", "score": 43.7, "reference_url": "https://arxiv.org/abs/2501.00656", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "OLMES", "prompt_style": "default", "temperature": "default" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "OLMo 2 paper Table 7 Instruct (tulu-3-8b column): math=43.7. Third-party Allen AI self-test, mc=false.", "candidates": [] }, { "model_id": "tulu-3-8b", "benchmark_id": "mmlu", "score": 68.2, "reference_url": "https://arxiv.org/abs/2501.00656", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "OLMES", "prompt_style": "default", "temperature": "default" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "OLMo 2 paper Table 7 Instruct (tulu-3-8b column): mmlu=68.2. Third-party Allen AI self-test, mc=false.", "candidates": [] }, { "model_id": "tulu-3-8b", "benchmark_id": "safety", "score": 75.4, "reference_url": "https://arxiv.org/abs/2501.00656", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "OLMES", "prompt_style": "default", "temperature": "default" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "OLMo 2 paper Table 7 Instruct (tulu-3-8b column): safety=75.4. Third-party Allen AI self-test, mc=false.", "candidates": [] }, { "model_id": "tulu-3-8b", "benchmark_id": "popqa", "score": 29.1, "reference_url": "https://arxiv.org/abs/2501.00656", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "OLMES", "prompt_style": "default", "temperature": "default" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "OLMo 2 paper Table 7 Instruct (tulu-3-8b column): popqa=29.1. Third-party Allen AI self-test, mc=false.", "candidates": [] }, { "model_id": "tulu-3-8b", "benchmark_id": "truthfulqa", "score": 55.0, "reference_url": "https://arxiv.org/abs/2501.00656", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "OLMES", "prompt_style": "default", "temperature": "default" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "OLMo 2 paper Table 7 Instruct (tulu-3-8b column): truthfulqa=55.0. Third-party Allen AI self-test, mc=false.", "candidates": [] }, { "model_id": "qwen-2.5-7b", "benchmark_id": "alpacaeval_2", "score": 29.7, "reference_url": "https://arxiv.org/abs/2501.00656", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "OLMES", "prompt_style": "default", "temperature": "default" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "OLMo 2 paper Table 7 Instruct (qwen-2.5-7b column): alpacaeval_2=29.7. Third-party Allen AI self-test, mc=false.", "candidates": [] }, { "model_id": "qwen-2.5-7b", "benchmark_id": "bigbench_hard", "score": 70.2, "reference_url": "https://arxiv.org/abs/2501.00656", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "OLMES", "prompt_style": "default", "temperature": "default" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "OLMo 2 paper Table 7 Instruct (qwen-2.5-7b column): bigbench_hard=70.2. Third-party Allen AI self-test, mc=false.", "candidates": [] }, { "model_id": "qwen-2.5-7b", "benchmark_id": "drop", "score": 54.4, "reference_url": "https://arxiv.org/abs/2501.00656", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "OLMES", "prompt_style": "default", "temperature": "default" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "OLMo 2 paper Table 7 Instruct (qwen-2.5-7b column): drop=54.4. Third-party Allen AI self-test, mc=false.", "candidates": [] }, { "model_id": "qwen-2.5-7b", "benchmark_id": "gsm8k", "score": 83.8, "reference_url": "https://arxiv.org/abs/2501.00656", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "OLMES", "prompt_style": "default", "temperature": "default" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "OLMo 2 paper Table 7 Instruct (qwen-2.5-7b column): gsm8k=83.8. Third-party Allen AI self-test, mc=false.", "candidates": [] }, { "model_id": "qwen-2.5-7b", "benchmark_id": "ifeval", "score": 74.7, "reference_url": "https://arxiv.org/abs/2501.00656", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "OLMES", "prompt_style": "default", "temperature": "default" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "OLMo 2 paper Table 7 Instruct (qwen-2.5-7b column): ifeval=74.7. Third-party Allen AI self-test, mc=false.", "candidates": [] }, { "model_id": "qwen-2.5-7b", "benchmark_id": "math", "score": 69.9, "reference_url": "https://arxiv.org/abs/2501.00656", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "OLMES", "prompt_style": "default", "temperature": "default" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "OLMo 2 paper Table 7 Instruct (qwen-2.5-7b column): math=69.9. Third-party Allen AI self-test, mc=false.", "candidates": [] }, { "model_id": "qwen-2.5-7b", "benchmark_id": "mmlu", "score": 76.6, "reference_url": "https://arxiv.org/abs/2501.00656", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "OLMES", "prompt_style": "default", "temperature": "default" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "OLMo 2 paper Table 7 Instruct (qwen-2.5-7b column): mmlu=76.6. Third-party Allen AI self-test, mc=false.", "candidates": [] }, { "model_id": "qwen-2.5-7b", "benchmark_id": "safety", "score": 75.0, "reference_url": "https://arxiv.org/abs/2501.00656", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "OLMES", "prompt_style": "default", "temperature": "default" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "OLMo 2 paper Table 7 Instruct (qwen-2.5-7b column): safety=75.0. Third-party Allen AI self-test, mc=false.", "candidates": [] }, { "model_id": "qwen-2.5-7b", "benchmark_id": "popqa", "score": 18.1, "reference_url": "https://arxiv.org/abs/2501.00656", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "OLMES", "prompt_style": "default", "temperature": "default" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "OLMo 2 paper Table 7 Instruct (qwen-2.5-7b column): popqa=18.1. Third-party Allen AI self-test, mc=false.", "candidates": [] }, { "model_id": "qwen-2.5-7b", "benchmark_id": "truthfulqa", "score": 63.1, "reference_url": "https://arxiv.org/abs/2501.00656", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "OLMES", "prompt_style": "default", "temperature": "default" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "OLMo 2 paper Table 7 Instruct (qwen-2.5-7b column): truthfulqa=63.1. Third-party Allen AI self-test, mc=false.", "candidates": [] }, { "model_id": "gemma-2-9b", "benchmark_id": "alpacaeval_2", "score": 43.7, "reference_url": "https://arxiv.org/abs/2501.00656", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "OLMES", "prompt_style": "default", "temperature": "default" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "OLMo 2 paper Table 7 Instruct (gemma-2-9b column): alpacaeval_2=43.7. Third-party Allen AI self-test, mc=false.", "candidates": [] }, { "model_id": "gemma-2-9b", "benchmark_id": "bigbench_hard", "score": 64.9, "reference_url": "https://arxiv.org/abs/2501.00656", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "OLMES", "prompt_style": "default", "temperature": "default" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "OLMo 2 paper Table 7 Instruct (gemma-2-9b column): bigbench_hard=64.9. Third-party Allen AI self-test, mc=false.", "candidates": [] }, { "model_id": "gemma-2-9b", "benchmark_id": "drop", "score": 58.8, "reference_url": "https://arxiv.org/abs/2501.00656", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "OLMES", "prompt_style": "default", "temperature": "default" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "OLMo 2 paper Table 7 Instruct (gemma-2-9b column): drop=58.8. Third-party Allen AI self-test, mc=false.", "candidates": [] }, { "model_id": "gemma-2-9b", "benchmark_id": "gsm8k", "score": 79.7, "reference_url": "https://arxiv.org/abs/2501.00656", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "OLMES", "prompt_style": "default", "temperature": "default" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "OLMo 2 paper Table 7 Instruct (gemma-2-9b column): gsm8k=79.7. Third-party Allen AI self-test, mc=false.", "candidates": [] }, { "model_id": "gemma-2-9b", "benchmark_id": "ifeval", "score": 69.9, "reference_url": "https://arxiv.org/abs/2501.00656", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "OLMES", "prompt_style": "default", "temperature": "default" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "OLMo 2 paper Table 7 Instruct (gemma-2-9b column): ifeval=69.9. Third-party Allen AI self-test, mc=false.", "candidates": [ { "score": 74.4, "reference_url": "https://cohere.com/research/papers/command-a-technical-report.pdf", "source_type": "tech_report", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "Cohere internal eval", "prompt_style": "default", "temperature": "default" }, "notes": "Command A paper: alt measurement 74.4 (mc=false)." } ] }, { "model_id": "gemma-2-9b", "benchmark_id": "mmlu", "score": 69.1, "reference_url": "https://arxiv.org/abs/2501.00656", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "OLMES", "prompt_style": "default", "temperature": "default" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "OLMo 2 paper Table 7 Instruct (gemma-2-9b column): mmlu=69.1. Third-party Allen AI self-test, mc=false.", "candidates": [ { "score": 73.5, "reference_url": "https://cohere.com/research/papers/command-a-technical-report.pdf", "source_type": "tech_report", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "Cohere internal eval", "prompt_style": "default", "temperature": "default" }, "notes": "Command A paper: alt measurement 73.5 (mc=false)." } ] }, { "model_id": "gemma-2-9b", "benchmark_id": "safety", "score": 75.5, "reference_url": "https://arxiv.org/abs/2501.00656", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "OLMES", "prompt_style": "default", "temperature": "default" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "OLMo 2 paper Table 7 Instruct (gemma-2-9b column): safety=75.5. Third-party Allen AI self-test, mc=false.", "candidates": [] }, { "model_id": "gemma-2-9b", "benchmark_id": "popqa", "score": 28.3, "reference_url": "https://arxiv.org/abs/2501.00656", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "OLMES", "prompt_style": "default", "temperature": "default" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "OLMo 2 paper Table 7 Instruct (gemma-2-9b column): popqa=28.3. Third-party Allen AI self-test, mc=false.", "candidates": [] }, { "model_id": "gemma-2-9b", "benchmark_id": "truthfulqa", "score": 61.4, "reference_url": "https://arxiv.org/abs/2501.00656", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "OLMES", "prompt_style": "default", "temperature": "default" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "OLMo 2 paper Table 7 Instruct (gemma-2-9b column): truthfulqa=61.4. Third-party Allen AI self-test, mc=false.", "candidates": [] }, { "model_id": "qwen-2.5-14b", "benchmark_id": "alpacaeval_2", "score": 34.6, "reference_url": "https://arxiv.org/abs/2501.00656", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "OLMES", "prompt_style": "default", "temperature": "default" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "OLMo 2 paper Table 7 Instruct (qwen-2.5-14b column): alpacaeval_2=34.6. Third-party Allen AI self-test, mc=false.", "candidates": [] }, { "model_id": "qwen-2.5-14b", "benchmark_id": "bigbench_hard", "score": 78.4, "reference_url": "https://arxiv.org/abs/2501.00656", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "OLMES", "prompt_style": "default", "temperature": "default" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "OLMo 2 paper Table 7 Instruct (qwen-2.5-14b column): bigbench_hard=78.4. Third-party Allen AI self-test, mc=false.", "candidates": [] }, { "model_id": "qwen-2.5-14b", "benchmark_id": "drop", "score": 50.5, "reference_url": "https://arxiv.org/abs/2501.00656", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "OLMES", "prompt_style": "default", "temperature": "default" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "OLMo 2 paper Table 7 Instruct (qwen-2.5-14b column): drop=50.5. Third-party Allen AI self-test, mc=false.", "candidates": [] }, { "model_id": "qwen-2.5-14b", "benchmark_id": "gsm8k", "score": 83.9, "reference_url": "https://arxiv.org/abs/2501.00656", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "OLMES", "prompt_style": "default", "temperature": "default" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "OLMo 2 paper Table 7 Instruct (qwen-2.5-14b column): gsm8k=83.9. Third-party Allen AI self-test, mc=false.", "candidates": [] }, { "model_id": "qwen-2.5-14b", "benchmark_id": "ifeval", "score": 82.4, "reference_url": "https://arxiv.org/abs/2501.00656", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "OLMES", "prompt_style": "default", "temperature": "default" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "OLMo 2 paper Table 7 Instruct (qwen-2.5-14b column): ifeval=82.4. Third-party Allen AI self-test, mc=false.", "candidates": [] }, { "model_id": "qwen-2.5-14b", "benchmark_id": "math", "score": 70.6, "reference_url": "https://arxiv.org/abs/2501.00656", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "OLMES", "prompt_style": "default", "temperature": "default" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "OLMo 2 paper Table 7 Instruct (qwen-2.5-14b column): math=70.6. Third-party Allen AI self-test, mc=false.", "candidates": [] }, { "model_id": "qwen-2.5-14b", "benchmark_id": "mmlu", "score": 81.1, "reference_url": "https://arxiv.org/abs/2501.00656", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "OLMES", "prompt_style": "default", "temperature": "default" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "OLMo 2 paper Table 7 Instruct (qwen-2.5-14b column): mmlu=81.1. Third-party Allen AI self-test, mc=false.", "candidates": [] }, { "model_id": "qwen-2.5-14b", "benchmark_id": "safety", "score": 79.3, "reference_url": "https://arxiv.org/abs/2501.00656", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "OLMES", "prompt_style": "default", "temperature": "default" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "OLMo 2 paper Table 7 Instruct (qwen-2.5-14b column): safety=79.3. Third-party Allen AI self-test, mc=false.", "candidates": [] }, { "model_id": "qwen-2.5-14b", "benchmark_id": "popqa", "score": 21.1, "reference_url": "https://arxiv.org/abs/2501.00656", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "OLMES", "prompt_style": "default", "temperature": "default" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "OLMo 2 paper Table 7 Instruct (qwen-2.5-14b column): popqa=21.1. Third-party Allen AI self-test, mc=false.", "candidates": [] }, { "model_id": "qwen-2.5-14b", "benchmark_id": "truthfulqa", "score": 70.8, "reference_url": "https://arxiv.org/abs/2501.00656", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "OLMES", "prompt_style": "default", "temperature": "default" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "OLMo 2 paper Table 7 Instruct (qwen-2.5-14b column): truthfulqa=70.8. Third-party Allen AI self-test, mc=false.", "candidates": [] }, { "model_id": "gemma-2-27b", "benchmark_id": "alpacaeval_2", "score": 49.0, "reference_url": "https://arxiv.org/abs/2501.00656", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "OLMES", "prompt_style": "default", "temperature": "default" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "OLMo 2 paper Table 7 Instruct (gemma-2-27b column): alpacaeval_2=49.0. Third-party Allen AI self-test, mc=false.", "candidates": [] }, { "model_id": "gemma-2-27b", "benchmark_id": "bigbench_hard", "score": 72.7, "reference_url": "https://arxiv.org/abs/2501.00656", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "OLMES", "prompt_style": "default", "temperature": "default" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "OLMo 2 paper Table 7 Instruct (gemma-2-27b column): bigbench_hard=72.7. Third-party Allen AI self-test, mc=false.", "candidates": [] }, { "model_id": "gemma-2-27b", "benchmark_id": "drop", "score": 67.5, "reference_url": "https://arxiv.org/abs/2501.00656", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "OLMES", "prompt_style": "default", "temperature": "default" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "OLMo 2 paper Table 7 Instruct (gemma-2-27b column): drop=67.5. Third-party Allen AI self-test, mc=false.", "candidates": [] }, { "model_id": "gemma-2-27b", "benchmark_id": "gsm8k", "score": 80.7, "reference_url": "https://arxiv.org/abs/2501.00656", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "OLMES", "prompt_style": "default", "temperature": "default" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "OLMo 2 paper Table 7 Instruct (gemma-2-27b column): gsm8k=80.7. Third-party Allen AI self-test, mc=false.", "candidates": [] }, { "model_id": "gemma-2-27b", "benchmark_id": "ifeval", "score": 63.2, "reference_url": "https://arxiv.org/abs/2501.00656", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "OLMES", "prompt_style": "default", "temperature": "default" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "OLMo 2 paper Table 7 Instruct (gemma-2-27b column): ifeval=63.2. Third-party Allen AI self-test, mc=false.", "candidates": [] }, { "model_id": "gemma-2-27b", "benchmark_id": "mmlu", "score": 70.7, "reference_url": "https://arxiv.org/abs/2501.00656", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "OLMES", "prompt_style": "default", "temperature": "default" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "OLMo 2 paper Table 7 Instruct (gemma-2-27b column): mmlu=70.7. Third-party Allen AI self-test, mc=false.", "candidates": [] }, { "model_id": "gemma-2-27b", "benchmark_id": "safety", "score": 75.9, "reference_url": "https://arxiv.org/abs/2501.00656", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "OLMES", "prompt_style": "default", "temperature": "default" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "OLMo 2 paper Table 7 Instruct (gemma-2-27b column): safety=75.9. Third-party Allen AI self-test, mc=false.", "candidates": [] }, { "model_id": "gemma-2-27b", "benchmark_id": "popqa", "score": 33.9, "reference_url": "https://arxiv.org/abs/2501.00656", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "OLMES", "prompt_style": "default", "temperature": "default" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "OLMo 2 paper Table 7 Instruct (gemma-2-27b column): popqa=33.9. Third-party Allen AI self-test, mc=false.", "candidates": [] }, { "model_id": "gemma-2-27b", "benchmark_id": "truthfulqa", "score": 64.6, "reference_url": "https://arxiv.org/abs/2501.00656", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "OLMES", "prompt_style": "default", "temperature": "default" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "OLMo 2 paper Table 7 Instruct (gemma-2-27b column): truthfulqa=64.6. Third-party Allen AI self-test, mc=false.", "candidates": [] }, { "model_id": "qwen-2.5-32b", "benchmark_id": "alpacaeval_2", "score": 39.1, "reference_url": "https://arxiv.org/abs/2501.00656", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "OLMES", "prompt_style": "default", "temperature": "default" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "OLMo 2 paper Table 7 Instruct (qwen-2.5-32b column): alpacaeval_2=39.1. Third-party Allen AI self-test, mc=false.", "candidates": [] }, { "model_id": "qwen-2.5-32b", "benchmark_id": "bigbench_hard", "score": 82.3, "reference_url": "https://arxiv.org/abs/2501.00656", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "OLMES", "prompt_style": "default", "temperature": "default" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "OLMo 2 paper Table 7 Instruct (qwen-2.5-32b column): bigbench_hard=82.3. Third-party Allen AI self-test, mc=false.", "candidates": [] }, { "model_id": "qwen-2.5-32b", "benchmark_id": "drop", "score": 48.3, "reference_url": "https://arxiv.org/abs/2501.00656", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "OLMES", "prompt_style": "default", "temperature": "default" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "OLMo 2 paper Table 7 Instruct (qwen-2.5-32b column): drop=48.3. Third-party Allen AI self-test, mc=false.", "candidates": [] }, { "model_id": "qwen-2.5-32b", "benchmark_id": "gsm8k", "score": 87.5, "reference_url": "https://arxiv.org/abs/2501.00656", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "OLMES", "prompt_style": "default", "temperature": "default" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "OLMo 2 paper Table 7 Instruct (qwen-2.5-32b column): gsm8k=87.5. Third-party Allen AI self-test, mc=false.", "candidates": [] }, { "model_id": "qwen-2.5-32b", "benchmark_id": "ifeval", "score": 82.4, "reference_url": "https://arxiv.org/abs/2501.00656", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "OLMES", "prompt_style": "default", "temperature": "default" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "OLMo 2 paper Table 7 Instruct (qwen-2.5-32b column): ifeval=82.4. Third-party Allen AI self-test, mc=false.", "candidates": [] }, { "model_id": "qwen-2.5-32b", "benchmark_id": "math", "score": 77.9, "reference_url": "https://arxiv.org/abs/2501.00656", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "OLMES", "prompt_style": "default", "temperature": "default" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "OLMo 2 paper Table 7 Instruct (qwen-2.5-32b column): math=77.9. Third-party Allen AI self-test, mc=false.", "candidates": [] }, { "model_id": "qwen-2.5-32b", "benchmark_id": "mmlu", "score": 84.7, "reference_url": "https://arxiv.org/abs/2501.00656", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "OLMES", "prompt_style": "default", "temperature": "default" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "OLMo 2 paper Table 7 Instruct (qwen-2.5-32b column): mmlu=84.7. Third-party Allen AI self-test, mc=false.", "candidates": [] }, { "model_id": "qwen-2.5-32b", "benchmark_id": "safety", "score": 82.4, "reference_url": "https://arxiv.org/abs/2501.00656", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "OLMES", "prompt_style": "default", "temperature": "default" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "OLMo 2 paper Table 7 Instruct (qwen-2.5-32b column): safety=82.4. Third-party Allen AI self-test, mc=false.", "candidates": [] }, { "model_id": "qwen-2.5-32b", "benchmark_id": "popqa", "score": 26.1, "reference_url": "https://arxiv.org/abs/2501.00656", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "OLMES", "prompt_style": "default", "temperature": "default" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "OLMo 2 paper Table 7 Instruct (qwen-2.5-32b column): popqa=26.1. Third-party Allen AI self-test, mc=false.", "candidates": [] }, { "model_id": "qwen-2.5-32b", "benchmark_id": "truthfulqa", "score": 70.6, "reference_url": "https://arxiv.org/abs/2501.00656", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "OLMES", "prompt_style": "default", "temperature": "default" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "OLMo 2 paper Table 7 Instruct (qwen-2.5-32b column): truthfulqa=70.6. Third-party Allen AI self-test, mc=false.", "candidates": [] }, { "model_id": "mistral-small-24b", "benchmark_id": "alpacaeval_2", "score": 43.2, "reference_url": "https://arxiv.org/abs/2501.00656", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "OLMES", "prompt_style": "default", "temperature": "default" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "OLMo 2 paper Table 7 Instruct (mistral-small-24b column): alpacaeval_2=43.2. Third-party Allen AI self-test, mc=false.", "candidates": [] }, { "model_id": "mistral-small-24b", "benchmark_id": "bigbench_hard", "score": 80.1, "reference_url": "https://arxiv.org/abs/2501.00656", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "OLMES", "prompt_style": "default", "temperature": "default" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "OLMo 2 paper Table 7 Instruct (mistral-small-24b column): bigbench_hard=80.1. Third-party Allen AI self-test, mc=false.", "candidates": [] }, { "model_id": "mistral-small-24b", "benchmark_id": "drop", "score": 78.5, "reference_url": "https://arxiv.org/abs/2501.00656", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "OLMES", "prompt_style": "default", "temperature": "default" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "OLMo 2 paper Table 7 Instruct (mistral-small-24b column): drop=78.5. Third-party Allen AI self-test, mc=false.", "candidates": [] }, { "model_id": "mistral-small-24b", "benchmark_id": "gsm8k", "score": 87.2, "reference_url": "https://arxiv.org/abs/2501.00656", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "OLMES", "prompt_style": "default", "temperature": "default" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "OLMo 2 paper Table 7 Instruct (mistral-small-24b column): gsm8k=87.2. Third-party Allen AI self-test, mc=false.", "candidates": [] }, { "model_id": "mistral-small-24b", "benchmark_id": "ifeval", "score": 77.3, "reference_url": "https://arxiv.org/abs/2501.00656", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "OLMES", "prompt_style": "default", "temperature": "default" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "OLMo 2 paper Table 7 Instruct (mistral-small-24b column): ifeval=77.3. Third-party Allen AI self-test, mc=false.", "candidates": [] }, { "model_id": "mistral-small-24b", "benchmark_id": "math", "score": 65.9, "reference_url": "https://arxiv.org/abs/2501.00656", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "OLMES", "prompt_style": "default", "temperature": "default" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "OLMo 2 paper Table 7 Instruct (mistral-small-24b column): math=65.9. Third-party Allen AI self-test, mc=false.", "candidates": [] }, { "model_id": "mistral-small-24b", "benchmark_id": "mmlu", "score": 83.7, "reference_url": "https://arxiv.org/abs/2501.00656", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "OLMES", "prompt_style": "default", "temperature": "default" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "OLMo 2 paper Table 7 Instruct (mistral-small-24b column): mmlu=83.7. Third-party Allen AI self-test, mc=false.", "candidates": [] }, { "model_id": "mistral-small-24b", "benchmark_id": "safety", "score": 66.5, "reference_url": "https://arxiv.org/abs/2501.00656", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "OLMES", "prompt_style": "default", "temperature": "default" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "OLMo 2 paper Table 7 Instruct (mistral-small-24b column): safety=66.5. Third-party Allen AI self-test, mc=false.", "candidates": [] }, { "model_id": "mistral-small-24b", "benchmark_id": "popqa", "score": 24.4, "reference_url": "https://arxiv.org/abs/2501.00656", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "OLMES", "prompt_style": "default", "temperature": "default" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "OLMo 2 paper Table 7 Instruct (mistral-small-24b column): popqa=24.4. Third-party Allen AI self-test, mc=false.", "candidates": [] }, { "model_id": "mistral-small-24b", "benchmark_id": "truthfulqa", "score": 68.1, "reference_url": "https://arxiv.org/abs/2501.00656", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "OLMES", "prompt_style": "default", "temperature": "default" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "OLMo 2 paper Table 7 Instruct (mistral-small-24b column): truthfulqa=68.1. Third-party Allen AI self-test, mc=false.", "candidates": [] }, { "model_id": "qwq-32b", "benchmark_id": "alpacaeval_2", "score": 82.4, "reference_url": "https://arxiv.org/abs/2501.00656", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "OLMES", "prompt_style": "default", "temperature": "default" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "OLMo 2 paper Table 7 Instruct (qwq-32b column): alpacaeval_2=82.4. Third-party Allen AI self-test, mc=false.", "candidates": [] }, { "model_id": "qwq-32b", "benchmark_id": "bigbench_hard", "score": 89.6, "reference_url": "https://arxiv.org/abs/2501.00656", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "OLMES", "prompt_style": "default", "temperature": "default" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "OLMo 2 paper Table 7 Instruct (qwq-32b column): bigbench_hard=89.6. Third-party Allen AI self-test, mc=false.", "candidates": [] }, { "model_id": "qwq-32b", "benchmark_id": "drop", "score": 54.7, "reference_url": "https://arxiv.org/abs/2501.00656", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "OLMES", "prompt_style": "default", "temperature": "default" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "OLMo 2 paper Table 7 Instruct (qwq-32b column): drop=54.7. Third-party Allen AI self-test, mc=false.", "candidates": [] }, { "model_id": "qwq-32b", "benchmark_id": "gsm8k", "score": 95.5, "reference_url": "https://arxiv.org/abs/2501.00656", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "OLMES", "prompt_style": "default", "temperature": "default" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "OLMo 2 paper Table 7 Instruct (qwq-32b column): gsm8k=95.5. Third-party Allen AI self-test, mc=false.", "candidates": [] }, { "model_id": "qwq-32b", "benchmark_id": "math", "score": 98.1, "reference_url": "https://arxiv.org/abs/2501.00656", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "OLMES", "prompt_style": "default", "temperature": "default" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "OLMo 2 paper Table 7 Instruct (qwq-32b column): math=98.1. Third-party Allen AI self-test, mc=false.", "candidates": [] }, { "model_id": "qwq-32b", "benchmark_id": "safety", "score": 69.9, "reference_url": "https://arxiv.org/abs/2501.00656", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "OLMES", "prompt_style": "default", "temperature": "default" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "OLMo 2 paper Table 7 Instruct (qwq-32b column): safety=69.9. Third-party Allen AI self-test, mc=false.", "candidates": [] }, { "model_id": "gemma-3-27b", "benchmark_id": "alpacaeval_2", "score": 63.4, "reference_url": "https://arxiv.org/abs/2501.00656", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "OLMES", "prompt_style": "default", "temperature": "default" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "OLMo 2 paper Table 7 Instruct (gemma-3-27b column): alpacaeval_2=63.4. Third-party Allen AI self-test, mc=false.", "candidates": [] }, { "model_id": "gemma-3-27b", "benchmark_id": "bigbench_hard", "score": 83.7, "reference_url": "https://arxiv.org/abs/2501.00656", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "OLMES", "prompt_style": "default", "temperature": "default" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "OLMo 2 paper Table 7 Instruct (gemma-3-27b column): bigbench_hard=83.7. Third-party Allen AI self-test, mc=false.", "candidates": [] }, { "model_id": "gemma-3-27b", "benchmark_id": "drop", "score": 69.2, "reference_url": "https://arxiv.org/abs/2501.00656", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "OLMES", "prompt_style": "default", "temperature": "default" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "OLMo 2 paper Table 7 Instruct (gemma-3-27b column): drop=69.2. Third-party Allen AI self-test, mc=false.", "candidates": [] }, { "model_id": "gemma-3-27b", "benchmark_id": "safety", "score": 69.1, "reference_url": "https://arxiv.org/abs/2501.00656", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "OLMES", "prompt_style": "default", "temperature": "default" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "OLMo 2 paper Table 7 Instruct (gemma-3-27b column): safety=69.1. Third-party Allen AI self-test, mc=false.", "candidates": [] }, { "model_id": "gemma-3-27b", "benchmark_id": "popqa", "score": 30.9, "reference_url": "https://arxiv.org/abs/2501.00656", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "OLMES", "prompt_style": "default", "temperature": "default" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "OLMo 2 paper Table 7 Instruct (gemma-3-27b column): popqa=30.9. Third-party Allen AI self-test, mc=false.", "candidates": [] }, { "model_id": "gemma-3-27b", "benchmark_id": "truthfulqa", "score": 63.9, "reference_url": "https://arxiv.org/abs/2501.00656", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "OLMES", "prompt_style": "default", "temperature": "default" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "OLMo 2 paper Table 7 Instruct (gemma-3-27b column): truthfulqa=63.9. Third-party Allen AI self-test, mc=false.", "candidates": [] }, { "model_id": "qwen2.5-72b-instruct", "benchmark_id": "alpacaeval_2", "score": 47.7, "reference_url": "https://arxiv.org/abs/2501.00656", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "OLMES", "prompt_style": "default", "temperature": "default" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "OLMo 2 paper Table 7 Instruct (qwen2.5-72b-instruct column): alpacaeval_2=47.7. Third-party Allen AI self-test, mc=false.", "candidates": [] }, { "model_id": "qwen2.5-72b-instruct", "benchmark_id": "bigbench_hard", "score": 80.4, "reference_url": "https://arxiv.org/abs/2501.00656", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "OLMES", "prompt_style": "default", "temperature": "default" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "OLMo 2 paper Table 7 Instruct (qwen2.5-72b-instruct column): bigbench_hard=80.4. Third-party Allen AI self-test, mc=false.", "candidates": [] }, { "model_id": "qwen2.5-72b-instruct", "benchmark_id": "gsm8k", "score": 89.5, "reference_url": "https://arxiv.org/abs/2501.00656", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "OLMES", "prompt_style": "default", "temperature": "default" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "OLMo 2 paper Table 7 Instruct (qwen2.5-72b-instruct column): gsm8k=89.5. Third-party Allen AI self-test, mc=false.", "candidates": [] }, { "model_id": "qwen2.5-72b-instruct", "benchmark_id": "math", "score": 75.9, "reference_url": "https://arxiv.org/abs/2501.00656", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "OLMES", "prompt_style": "default", "temperature": "default" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "OLMo 2 paper Table 7 Instruct (qwen2.5-72b-instruct column): math=75.9. Third-party Allen AI self-test, mc=false.", "candidates": [] }, { "model_id": "qwen2.5-72b-instruct", "benchmark_id": "safety", "score": 87.0, "reference_url": "https://arxiv.org/abs/2501.00656", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "OLMES", "prompt_style": "default", "temperature": "default" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "OLMo 2 paper Table 7 Instruct (qwen2.5-72b-instruct column): safety=87.0. Third-party Allen AI self-test, mc=false.", "candidates": [] }, { "model_id": "qwen2.5-72b-instruct", "benchmark_id": "popqa", "score": 30.6, "reference_url": "https://arxiv.org/abs/2501.00656", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "OLMES", "prompt_style": "default", "temperature": "default" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "OLMo 2 paper Table 7 Instruct (qwen2.5-72b-instruct column): popqa=30.6. Third-party Allen AI self-test, mc=false.", "candidates": [] }, { "model_id": "qwen2.5-72b-instruct", "benchmark_id": "truthfulqa", "score": 69.9, "reference_url": "https://arxiv.org/abs/2501.00656", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "OLMES", "prompt_style": "default", "temperature": "default" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "OLMo 2 paper Table 7 Instruct (qwen2.5-72b-instruct column): truthfulqa=69.9. Third-party Allen AI self-test, mc=false.", "candidates": [] }, { "model_id": "llama-3.1-70b", "benchmark_id": "alpacaeval_2", "score": 32.9, "reference_url": "https://arxiv.org/abs/2501.00656", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "OLMES", "prompt_style": "default", "temperature": "default" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "OLMo 2 paper Table 7 Instruct (llama-3.1-70b column): alpacaeval_2=32.9. Third-party Allen AI self-test, mc=false.", "candidates": [] }, { "model_id": "llama-3.1-70b", "benchmark_id": "bigbench_hard", "score": 83.0, "reference_url": "https://arxiv.org/abs/2501.00656", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "OLMES", "prompt_style": "default", "temperature": "default" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "OLMo 2 paper Table 7 Instruct (llama-3.1-70b column): bigbench_hard=83.0. Third-party Allen AI self-test, mc=false.", "candidates": [] }, { "model_id": "llama-3.1-70b", "benchmark_id": "drop", "score": 77.0, "reference_url": "https://arxiv.org/abs/2501.00656", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "OLMES", "prompt_style": "default", "temperature": "default" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "OLMo 2 paper Table 7 Instruct (llama-3.1-70b column): drop=77.0. Third-party Allen AI self-test, mc=false.", "candidates": [] }, { "model_id": "llama-3.1-70b", "benchmark_id": "gsm8k", "score": 94.5, "reference_url": "https://arxiv.org/abs/2501.00656", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "OLMES", "prompt_style": "default", "temperature": "default" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "OLMo 2 paper Table 7 Instruct (llama-3.1-70b column): gsm8k=94.5. Third-party Allen AI self-test, mc=false.", "candidates": [] }, { "model_id": "llama-3.1-70b", "benchmark_id": "ifeval", "score": 88.0, "reference_url": "https://arxiv.org/abs/2501.00656", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "OLMES", "prompt_style": "default", "temperature": "default" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "OLMo 2 paper Table 7 Instruct (llama-3.1-70b column): ifeval=88.0. Third-party Allen AI self-test, mc=false.", "candidates": [] }, { "model_id": "llama-3.1-70b", "benchmark_id": "math", "score": 56.2, "reference_url": "https://arxiv.org/abs/2501.00656", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "OLMES", "prompt_style": "default", "temperature": "default" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "OLMo 2 paper Table 7 Instruct (llama-3.1-70b column): math=56.2. Third-party Allen AI self-test, mc=false.", "candidates": [] }, { "model_id": "llama-3.1-70b", "benchmark_id": "mmlu", "score": 85.2, "reference_url": "https://arxiv.org/abs/2501.00656", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "OLMES", "prompt_style": "default", "temperature": "default" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "OLMo 2 paper Table 7 Instruct (llama-3.1-70b column): mmlu=85.2. Third-party Allen AI self-test, mc=false.", "candidates": [] }, { "model_id": "llama-3.1-70b", "benchmark_id": "safety", "score": 76.4, "reference_url": "https://arxiv.org/abs/2501.00656", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "OLMES", "prompt_style": "default", "temperature": "default" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "OLMo 2 paper Table 7 Instruct (llama-3.1-70b column): safety=76.4. Third-party Allen AI self-test, mc=false.", "candidates": [] }, { "model_id": "llama-3.1-70b", "benchmark_id": "popqa", "score": 46.5, "reference_url": "https://arxiv.org/abs/2501.00656", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "OLMES", "prompt_style": "default", "temperature": "default" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "OLMo 2 paper Table 7 Instruct (llama-3.1-70b column): popqa=46.5. Third-party Allen AI self-test, mc=false.", "candidates": [] }, { "model_id": "llama-3.1-70b", "benchmark_id": "truthfulqa", "score": 66.8, "reference_url": "https://arxiv.org/abs/2501.00656", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "OLMES", "prompt_style": "default", "temperature": "default" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "OLMo 2 paper Table 7 Instruct (llama-3.1-70b column): truthfulqa=66.8. Third-party Allen AI self-test, mc=false.", "candidates": [] }, { "model_id": "llama-3.3-70b-instruct", "benchmark_id": "alpacaeval_2", "score": 36.5, "reference_url": "https://arxiv.org/abs/2501.00656", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "OLMES", "prompt_style": "default", "temperature": "default" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "OLMo 2 paper Table 7 Instruct (llama-3.3-70b-instruct column): alpacaeval_2=36.5. Third-party Allen AI self-test, mc=false.", "candidates": [] }, { "model_id": "llama-3.3-70b-instruct", "benchmark_id": "bigbench_hard", "score": 85.8, "reference_url": "https://arxiv.org/abs/2501.00656", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "OLMES", "prompt_style": "default", "temperature": "default" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "OLMo 2 paper Table 7 Instruct (llama-3.3-70b-instruct column): bigbench_hard=85.8. Third-party Allen AI self-test, mc=false.", "candidates": [] }, { "model_id": "llama-3.3-70b-instruct", "benchmark_id": "gsm8k", "score": 93.6, "reference_url": "https://arxiv.org/abs/2501.00656", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "OLMES", "prompt_style": "default", "temperature": "default" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "OLMo 2 paper Table 7 Instruct (llama-3.3-70b-instruct column): gsm8k=93.6. Third-party Allen AI self-test, mc=false.", "candidates": [] }, { "model_id": "llama-3.3-70b-instruct", "benchmark_id": "math", "score": 71.8, "reference_url": "https://arxiv.org/abs/2501.00656", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "OLMES", "prompt_style": "default", "temperature": "default" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "OLMo 2 paper Table 7 Instruct (llama-3.3-70b-instruct column): math=71.8. Third-party Allen AI self-test, mc=false.", "candidates": [ { "score": 77.0, "reference_url": "https://cohere.com/research/papers/command-a-technical-report.pdf", "source_type": "tech_report", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "Cohere internal eval", "prompt_style": "default", "temperature": "default" }, "notes": "Command A paper: alt measurement 77.0 (mc=false)." } ] }, { "model_id": "llama-3.3-70b-instruct", "benchmark_id": "safety", "score": 70.4, "reference_url": "https://arxiv.org/abs/2501.00656", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "OLMES", "prompt_style": "default", "temperature": "default" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "OLMo 2 paper Table 7 Instruct (llama-3.3-70b-instruct column): safety=70.4. Third-party Allen AI self-test, mc=false.", "candidates": [] }, { "model_id": "llama-3.3-70b-instruct", "benchmark_id": "popqa", "score": 48.2, "reference_url": "https://arxiv.org/abs/2501.00656", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "OLMES", "prompt_style": "default", "temperature": "default" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "OLMo 2 paper Table 7 Instruct (llama-3.3-70b-instruct column): popqa=48.2. Third-party Allen AI self-test, mc=false.", "candidates": [] }, { "model_id": "llama-3.3-70b-instruct", "benchmark_id": "truthfulqa", "score": 66.1, "reference_url": "https://arxiv.org/abs/2501.00656", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "OLMES", "prompt_style": "default", "temperature": "default" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "OLMo 2 paper Table 7 Instruct (llama-3.3-70b-instruct column): truthfulqa=66.1. Third-party Allen AI self-test, mc=false.", "candidates": [] }, { "model_id": "olmo-1b", "benchmark_id": "alpacaeval_2", "score": 2.4, "reference_url": "https://arxiv.org/abs/2501.00656", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "OLMES", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "OLMo 2 paper Table 7 Instruct (olmo-1b column): alpacaeval_2=2.4.", "candidates": [] }, { "model_id": "olmo-1b", "benchmark_id": "bigbench_hard", "score": 29.9, "reference_url": "https://arxiv.org/abs/2501.00656", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "OLMES", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "OLMo 2 paper Table 7 Instruct (olmo-1b column): bigbench_hard=29.9.", "candidates": [] }, { "model_id": "olmo-1b", "benchmark_id": "drop", "score": 27.9, "reference_url": "https://arxiv.org/abs/2501.00656", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "OLMES", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "OLMo 2 paper Table 7 Instruct (olmo-1b column): drop=27.9.", "candidates": [] }, { "model_id": "olmo-1b", "benchmark_id": "gsm8k", "score": 10.8, "reference_url": "https://arxiv.org/abs/2501.00656", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "OLMES", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "OLMo 2 paper Table 7 Instruct (olmo-1b column): gsm8k=10.8.", "candidates": [] }, { "model_id": "olmo-1b", "benchmark_id": "ifeval", "score": 25.3, "reference_url": "https://arxiv.org/abs/2501.00656", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "OLMES", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "OLMo 2 paper Table 7 Instruct (olmo-1b column): ifeval=25.3.", "candidates": [] }, { "model_id": "olmo-1b", "benchmark_id": "math", "score": 2.2, "reference_url": "https://arxiv.org/abs/2501.00656", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "OLMES", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "OLMo 2 paper Table 7 Instruct (olmo-1b column): math=2.2.", "candidates": [] }, { "model_id": "olmo-1b", "benchmark_id": "mmlu", "score": 36.6, "reference_url": "https://arxiv.org/abs/2501.00656", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "OLMES", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "OLMo 2 paper Table 7 Instruct (olmo-1b column): mmlu=36.6.", "candidates": [] }, { "model_id": "olmo-1b", "benchmark_id": "safety", "score": 52.0, "reference_url": "https://arxiv.org/abs/2501.00656", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "OLMES", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "OLMo 2 paper Table 7 Instruct (olmo-1b column): safety=52.0.", "candidates": [] }, { "model_id": "olmo-1b", "benchmark_id": "popqa", "score": 12.1, "reference_url": "https://arxiv.org/abs/2501.00656", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "OLMES", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "OLMo 2 paper Table 7 Instruct (olmo-1b column): popqa=12.1.", "candidates": [] }, { "model_id": "olmo-1b", "benchmark_id": "truthfulqa", "score": 44.3, "reference_url": "https://arxiv.org/abs/2501.00656", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "OLMES", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "OLMo 2 paper Table 7 Instruct (olmo-1b column): truthfulqa=44.3.", "candidates": [] }, { "model_id": "smollm2-1.7b", "benchmark_id": "alpacaeval_2", "score": 5.8, "reference_url": "https://arxiv.org/abs/2501.00656", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "OLMES", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "OLMo 2 paper Table 7 Instruct (smollm2-1.7b column): alpacaeval_2=5.8.", "candidates": [] }, { "model_id": "smollm2-1.7b", "benchmark_id": "bigbench_hard", "score": 39.8, "reference_url": "https://arxiv.org/abs/2501.00656", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "OLMES", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "OLMo 2 paper Table 7 Instruct (smollm2-1.7b column): bigbench_hard=39.8.", "candidates": [] }, { "model_id": "smollm2-1.7b", "benchmark_id": "drop", "score": 30.9, "reference_url": "https://arxiv.org/abs/2501.00656", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "OLMES", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "OLMo 2 paper Table 7 Instruct (smollm2-1.7b column): drop=30.9.", "candidates": [] }, { "model_id": "smollm2-1.7b", "benchmark_id": "gsm8k", "score": 45.3, "reference_url": "https://arxiv.org/abs/2501.00656", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "OLMES", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "OLMo 2 paper Table 7 Instruct (smollm2-1.7b column): gsm8k=45.3.", "candidates": [] }, { "model_id": "smollm2-1.7b", "benchmark_id": "ifeval", "score": 51.6, "reference_url": "https://arxiv.org/abs/2501.00656", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "OLMES", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "OLMo 2 paper Table 7 Instruct (smollm2-1.7b column): ifeval=51.6.", "candidates": [] }, { "model_id": "smollm2-1.7b", "benchmark_id": "math", "score": 20.3, "reference_url": "https://arxiv.org/abs/2501.00656", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "OLMES", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "OLMo 2 paper Table 7 Instruct (smollm2-1.7b column): math=20.3.", "candidates": [] }, { "model_id": "smollm2-1.7b", "benchmark_id": "mmlu", "score": 34.3, "reference_url": "https://arxiv.org/abs/2501.00656", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "OLMES", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "OLMo 2 paper Table 7 Instruct (smollm2-1.7b column): mmlu=34.3.", "candidates": [] }, { "model_id": "smollm2-1.7b", "benchmark_id": "safety", "score": 52.4, "reference_url": "https://arxiv.org/abs/2501.00656", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "OLMES", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "OLMo 2 paper Table 7 Instruct (smollm2-1.7b column): safety=52.4.", "candidates": [] }, { "model_id": "smollm2-1.7b", "benchmark_id": "popqa", "score": 16.4, "reference_url": "https://arxiv.org/abs/2501.00656", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "OLMES", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "OLMo 2 paper Table 7 Instruct (smollm2-1.7b column): popqa=16.4.", "candidates": [] }, { "model_id": "smollm2-1.7b", "benchmark_id": "truthfulqa", "score": 45.3, "reference_url": "https://arxiv.org/abs/2501.00656", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "OLMES", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "OLMo 2 paper Table 7 Instruct (smollm2-1.7b column): truthfulqa=45.3.", "candidates": [] }, { "model_id": "olmo-7b-0424", "benchmark_id": "alpacaeval_2", "score": 8.5, "reference_url": "https://arxiv.org/abs/2501.00656", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "OLMES", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "OLMo 2 paper Table 7 Instruct (olmo-7b-0424 column): alpacaeval_2=8.5.", "candidates": [] }, { "model_id": "olmo-7b-0424", "benchmark_id": "bigbench_hard", "score": 34.4, "reference_url": "https://arxiv.org/abs/2501.00656", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "OLMES", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "OLMo 2 paper Table 7 Instruct (olmo-7b-0424 column): bigbench_hard=34.4.", "candidates": [] }, { "model_id": "olmo-7b-0424", "benchmark_id": "drop", "score": 47.9, "reference_url": "https://arxiv.org/abs/2501.00656", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "OLMES", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "OLMo 2 paper Table 7 Instruct (olmo-7b-0424 column): drop=47.9.", "candidates": [] }, { "model_id": "olmo-7b-0424", "benchmark_id": "gsm8k", "score": 23.2, "reference_url": "https://arxiv.org/abs/2501.00656", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "OLMES", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "OLMo 2 paper Table 7 Instruct (olmo-7b-0424 column): gsm8k=23.2.", "candidates": [] }, { "model_id": "olmo-7b-0424", "benchmark_id": "ifeval", "score": 39.2, "reference_url": "https://arxiv.org/abs/2501.00656", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "OLMES", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "OLMo 2 paper Table 7 Instruct (olmo-7b-0424 column): ifeval=39.2.", "candidates": [] }, { "model_id": "olmo-7b-0424", "benchmark_id": "math", "score": 5.2, "reference_url": "https://arxiv.org/abs/2501.00656", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "OLMES", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "OLMo 2 paper Table 7 Instruct (olmo-7b-0424 column): math=5.2.", "candidates": [] }, { "model_id": "olmo-7b-0424", "benchmark_id": "mmlu", "score": 48.9, "reference_url": "https://arxiv.org/abs/2501.00656", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "OLMES", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "OLMo 2 paper Table 7 Instruct (olmo-7b-0424 column): mmlu=48.9.", "candidates": [] }, { "model_id": "olmo-7b-0424", "benchmark_id": "safety", "score": 49.3, "reference_url": "https://arxiv.org/abs/2501.00656", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "OLMES", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "OLMo 2 paper Table 7 Instruct (olmo-7b-0424 column): safety=49.3.", "candidates": [] }, { "model_id": "olmo-7b-0424", "benchmark_id": "popqa", "score": 18.9, "reference_url": "https://arxiv.org/abs/2501.00656", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "OLMES", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "OLMo 2 paper Table 7 Instruct (olmo-7b-0424 column): popqa=18.9.", "candidates": [] }, { "model_id": "olmo-7b-0424", "benchmark_id": "truthfulqa", "score": 55.2, "reference_url": "https://arxiv.org/abs/2501.00656", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "OLMES", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "OLMo 2 paper Table 7 Instruct (olmo-7b-0424 column): truthfulqa=55.2.", "candidates": [] }, { "model_id": "olmo-2-1b", "benchmark_id": "alpacaeval_2", "score": 9.1, "reference_url": "https://arxiv.org/abs/2501.00656", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "OLMES", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "OLMo 2 paper Table 7 Instruct (olmo-2-1b column): alpacaeval_2=9.1.", "candidates": [] }, { "model_id": "olmo-2-1b", "benchmark_id": "bigbench_hard", "score": 35.0, "reference_url": "https://arxiv.org/abs/2501.00656", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "OLMES", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "OLMo 2 paper Table 7 Instruct (olmo-2-1b column): bigbench_hard=35.0.", "candidates": [] }, { "model_id": "olmo-2-1b", "benchmark_id": "drop", "score": 34.6, "reference_url": "https://arxiv.org/abs/2501.00656", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "OLMES", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "OLMo 2 paper Table 7 Instruct (olmo-2-1b column): drop=34.6.", "candidates": [] }, { "model_id": "olmo-2-1b", "benchmark_id": "gsm8k", "score": 68.3, "reference_url": "https://arxiv.org/abs/2501.00656", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "OLMES", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "OLMo 2 paper Table 7 Instruct (olmo-2-1b column): gsm8k=68.3.", "candidates": [] }, { "model_id": "olmo-2-1b", "benchmark_id": "ifeval", "score": 70.1, "reference_url": "https://arxiv.org/abs/2501.00656", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "OLMES", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "OLMo 2 paper Table 7 Instruct (olmo-2-1b column): ifeval=70.1.", "candidates": [] }, { "model_id": "olmo-2-1b", "benchmark_id": "math", "score": 20.7, "reference_url": "https://arxiv.org/abs/2501.00656", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "OLMES", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "OLMo 2 paper Table 7 Instruct (olmo-2-1b column): math=20.7.", "candidates": [] }, { "model_id": "olmo-2-1b", "benchmark_id": "mmlu", "score": 40.0, "reference_url": "https://arxiv.org/abs/2501.00656", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "OLMES", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "OLMo 2 paper Table 7 Instruct (olmo-2-1b column): mmlu=40.0.", "candidates": [] }, { "model_id": "olmo-2-1b", "benchmark_id": "safety", "score": 87.6, "reference_url": "https://arxiv.org/abs/2501.00656", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "OLMES", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "OLMo 2 paper Table 7 Instruct (olmo-2-1b column): safety=87.6.", "candidates": [] }, { "model_id": "olmo-2-1b", "benchmark_id": "popqa", "score": 12.9, "reference_url": "https://arxiv.org/abs/2501.00656", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "OLMES", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "OLMo 2 paper Table 7 Instruct (olmo-2-1b column): popqa=12.9.", "candidates": [] }, { "model_id": "olmo-2-1b", "benchmark_id": "truthfulqa", "score": 48.7, "reference_url": "https://arxiv.org/abs/2501.00656", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "OLMES", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "OLMo 2 paper Table 7 Instruct (olmo-2-1b column): truthfulqa=48.7.", "candidates": [] }, { "model_id": "olmo-2-7b", "benchmark_id": "alpacaeval_2", "score": 29.1, "reference_url": "https://arxiv.org/abs/2501.00656", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "OLMES", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "OLMo 2 paper Table 7 Instruct (olmo-2-7b column): alpacaeval_2=29.1.", "candidates": [] }, { "model_id": "olmo-2-7b", "benchmark_id": "bigbench_hard", "score": 51.4, "reference_url": "https://arxiv.org/abs/2501.00656", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "OLMES", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "OLMo 2 paper Table 7 Instruct (olmo-2-7b column): bigbench_hard=51.4.", "candidates": [] }, { "model_id": "olmo-2-7b", "benchmark_id": "drop", "score": 60.5, "reference_url": "https://arxiv.org/abs/2501.00656", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "OLMES", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "OLMo 2 paper Table 7 Instruct (olmo-2-7b column): drop=60.5.", "candidates": [] }, { "model_id": "olmo-2-7b", "benchmark_id": "gsm8k", "score": 85.1, "reference_url": "https://arxiv.org/abs/2501.00656", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "OLMES", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "OLMo 2 paper Table 7 Instruct (olmo-2-7b column): gsm8k=85.1.", "candidates": [] }, { "model_id": "olmo-2-7b", "benchmark_id": "ifeval", "score": 72.3, "reference_url": "https://arxiv.org/abs/2501.00656", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "OLMES", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "OLMo 2 paper Table 7 Instruct (olmo-2-7b column): ifeval=72.3.", "candidates": [] }, { "model_id": "olmo-2-7b", "benchmark_id": "math", "score": 32.5, "reference_url": "https://arxiv.org/abs/2501.00656", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "OLMES", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "OLMo 2 paper Table 7 Instruct (olmo-2-7b column): math=32.5.", "candidates": [] }, { "model_id": "olmo-2-7b", "benchmark_id": "mmlu", "score": 61.3, "reference_url": "https://arxiv.org/abs/2501.00656", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "OLMES", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "OLMo 2 paper Table 7 Instruct (olmo-2-7b column): mmlu=61.3.", "candidates": [] }, { "model_id": "olmo-2-7b", "benchmark_id": "safety", "score": 93.3, "reference_url": "https://arxiv.org/abs/2501.00656", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "OLMES", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "OLMo 2 paper Table 7 Instruct (olmo-2-7b column): safety=93.3.", "candidates": [] }, { "model_id": "olmo-2-7b", "benchmark_id": "popqa", "score": 23.2, "reference_url": "https://arxiv.org/abs/2501.00656", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "OLMES", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "OLMo 2 paper Table 7 Instruct (olmo-2-7b column): popqa=23.2.", "candidates": [] }, { "model_id": "olmo-2-7b", "benchmark_id": "truthfulqa", "score": 56.5, "reference_url": "https://arxiv.org/abs/2501.00656", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "OLMES", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "OLMo 2 paper Table 7 Instruct (olmo-2-7b column): truthfulqa=56.5.", "candidates": [] }, { "model_id": "olmo-2-13b", "benchmark_id": "alpacaeval_2", "score": 39.5, "reference_url": "https://arxiv.org/abs/2501.00656", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "OLMES", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "OLMo 2 paper Table 7 Instruct (olmo-2-13b column): alpacaeval_2=39.5.", "candidates": [] }, { "model_id": "olmo-2-13b", "benchmark_id": "bigbench_hard", "score": 63.0, "reference_url": "https://arxiv.org/abs/2501.00656", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "OLMES", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "OLMo 2 paper Table 7 Instruct (olmo-2-13b column): bigbench_hard=63.0.", "candidates": [] }, { "model_id": "olmo-2-13b", "benchmark_id": "drop", "score": 71.5, "reference_url": "https://arxiv.org/abs/2501.00656", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "OLMES", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "OLMo 2 paper Table 7 Instruct (olmo-2-13b column): drop=71.5.", "candidates": [] }, { "model_id": "olmo-2-13b", "benchmark_id": "gsm8k", "score": 87.4, "reference_url": "https://arxiv.org/abs/2501.00656", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "OLMES", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "OLMo 2 paper Table 7 Instruct (olmo-2-13b column): gsm8k=87.4.", "candidates": [] }, { "model_id": "olmo-2-13b", "benchmark_id": "math", "score": 39.2, "reference_url": "https://arxiv.org/abs/2501.00656", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "OLMES", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "OLMo 2 paper Table 7 Instruct (olmo-2-13b column): math=39.2.", "candidates": [] }, { "model_id": "olmo-2-13b", "benchmark_id": "safety", "score": 89.7, "reference_url": "https://arxiv.org/abs/2501.00656", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "OLMES", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "OLMo 2 paper Table 7 Instruct (olmo-2-13b column): safety=89.7.", "candidates": [] }, { "model_id": "olmo-2-13b", "benchmark_id": "popqa", "score": 28.8, "reference_url": "https://arxiv.org/abs/2501.00656", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "OLMES", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "OLMo 2 paper Table 7 Instruct (olmo-2-13b column): popqa=28.8.", "candidates": [] }, { "model_id": "olmo-2-13b", "benchmark_id": "truthfulqa", "score": 64.3, "reference_url": "https://arxiv.org/abs/2501.00656", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "OLMES", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "OLMo 2 paper Table 7 Instruct (olmo-2-13b column): truthfulqa=64.3.", "candidates": [] }, { "model_id": "olmo-2-32b", "benchmark_id": "alpacaeval_2", "score": 42.8, "reference_url": "https://arxiv.org/abs/2501.00656", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "OLMES", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "OLMo 2 paper Table 7 Instruct (olmo-2-32b column): alpacaeval_2=42.8.", "candidates": [] }, { "model_id": "olmo-2-32b", "benchmark_id": "bigbench_hard", "score": 70.6, "reference_url": "https://arxiv.org/abs/2501.00656", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "OLMES", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "OLMo 2 paper Table 7 Instruct (olmo-2-32b column): bigbench_hard=70.6.", "candidates": [] }, { "model_id": "olmo-2-32b", "benchmark_id": "drop", "score": 78.0, "reference_url": "https://arxiv.org/abs/2501.00656", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "OLMES", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "OLMo 2 paper Table 7 Instruct (olmo-2-32b column): drop=78.0.", "candidates": [] }, { "model_id": "olmo-2-32b", "benchmark_id": "gsm8k", "score": 87.6, "reference_url": "https://arxiv.org/abs/2501.00656", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "OLMES", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "OLMo 2 paper Table 7 Instruct (olmo-2-32b column): gsm8k=87.6.", "candidates": [] }, { "model_id": "olmo-2-32b", "benchmark_id": "ifeval", "score": 85.6, "reference_url": "https://arxiv.org/abs/2501.00656", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "OLMES", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "OLMo 2 paper Table 7 Instruct (olmo-2-32b column): ifeval=85.6.", "candidates": [] }, { "model_id": "olmo-2-32b", "benchmark_id": "math", "score": 49.7, "reference_url": "https://arxiv.org/abs/2501.00656", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "OLMES", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "OLMo 2 paper Table 7 Instruct (olmo-2-32b column): math=49.7.", "candidates": [] }, { "model_id": "olmo-2-32b", "benchmark_id": "mmlu", "score": 77.3, "reference_url": "https://arxiv.org/abs/2501.00656", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "OLMES", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "OLMo 2 paper Table 7 Instruct (olmo-2-32b column): mmlu=77.3.", "candidates": [] }, { "model_id": "olmo-2-32b", "benchmark_id": "safety", "score": 85.9, "reference_url": "https://arxiv.org/abs/2501.00656", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "OLMES", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "OLMo 2 paper Table 7 Instruct (olmo-2-32b column): safety=85.9.", "candidates": [] }, { "model_id": "olmo-2-32b", "benchmark_id": "popqa", "score": 37.5, "reference_url": "https://arxiv.org/abs/2501.00656", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "OLMES", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "OLMo 2 paper Table 7 Instruct (olmo-2-32b column): popqa=37.5.", "candidates": [] }, { "model_id": "olmo-2-32b", "benchmark_id": "truthfulqa", "score": 73.2, "reference_url": "https://arxiv.org/abs/2501.00656", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "OLMES", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "OLMo 2 paper Table 7 Instruct (olmo-2-32b column): truthfulqa=73.2.", "candidates": [] }, { "model_id": "lfm2.5-1.2b-thinking", "benchmark_id": "ifbench", "score": 44.85, "reference_url": "https://www.liquid.ai/blog/lfm2-5-1-2b-thinking-on-device-reasoning-under-1gb", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1 (mean of 5 runs)", "judge": "rule-based", "harness": "official (LFM eval)", "prompt_style": "default", "temperature": 0.05 }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "Liquid AI LFM2.5-1.2B-Thinking blog: lfm2.5-1.2b-thinking=44.85.", "candidates": [] }, { "model_id": "lfm2.5-1.2b-thinking", "benchmark_id": "multi_if", "score": 69.33, "reference_url": "https://www.liquid.ai/blog/lfm2-5-1-2b-thinking-on-device-reasoning-under-1gb", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1 (mean of 5 runs)", "judge": "rule-based", "harness": "official (LFM eval)", "prompt_style": "default", "temperature": 0.05 }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "Liquid AI LFM2.5-1.2B-Thinking blog: lfm2.5-1.2b-thinking=69.33.", "candidates": [] }, { "model_id": "lfm2.5-1.2b-thinking", "benchmark_id": "gsm8k", "score": 85.6, "reference_url": "https://www.liquid.ai/blog/lfm2-5-1-2b-thinking-on-device-reasoning-under-1gb", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1 (mean of 5 runs)", "judge": "rule-based", "harness": "official (LFM eval)", "prompt_style": "default", "temperature": 0.05 }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "Liquid AI LFM2.5-1.2B-Thinking blog: lfm2.5-1.2b-thinking=85.6.", "candidates": [] }, { "model_id": "lfm2.5-1.2b-thinking", "benchmark_id": "bfcl_v3", "score": 56.97, "reference_url": "https://www.liquid.ai/blog/lfm2-5-1-2b-thinking-on-device-reasoning-under-1gb", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1 (mean of 5 runs)", "judge": "rule-based", "harness": "official (LFM eval)", "prompt_style": "default", "temperature": 0.05 }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "Liquid AI LFM2.5-1.2B-Thinking blog: lfm2.5-1.2b-thinking=56.97.", "candidates": [] }, { "model_id": "qwen3-1.7b", "benchmark_id": "ifbench", "score": 25.88, "reference_url": "https://www.liquid.ai/blog/lfm2-5-1-2b-thinking-on-device-reasoning-under-1gb", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1 (mean of 5 runs)", "judge": "rule-based", "harness": "official (LFM eval, third-party)", "prompt_style": "default", "temperature": "default" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Liquid AI LFM2.5-1.2B-Thinking blog: qwen3-1.7b=25.88. (third-party self-test by Liquid AI)", "candidates": [] }, { "model_id": "lfm2.5-1.2b-instruct", "benchmark_id": "gpqa_diamond", "score": 38.89, "reference_url": "https://www.liquid.ai/blog/lfm2-5-1-2b-thinking-on-device-reasoning-under-1gb", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1 (mean of 5 runs)", "judge": "rule-based", "harness": "official (LFM eval, third-party)", "prompt_style": "default", "temperature": "default" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Liquid AI LFM2.5-1.2B-Thinking blog: lfm2.5-1.2b-instruct=38.89. (third-party self-test by Liquid AI)", "candidates": [] }, { "model_id": "lfm2.5-1.2b-instruct", "benchmark_id": "mmlu_pro", "score": 44.35, "reference_url": "https://www.liquid.ai/blog/lfm2-5-1-2b-thinking-on-device-reasoning-under-1gb", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1 (mean of 5 runs)", "judge": "rule-based", "harness": "official (LFM eval, third-party)", "prompt_style": "default", "temperature": "default" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Liquid AI LFM2.5-1.2B-Thinking blog: lfm2.5-1.2b-instruct=44.35. (third-party self-test by Liquid AI)", "candidates": [] }, { "model_id": "lfm2.5-1.2b-instruct", "benchmark_id": "ifeval", "score": 86.23, "reference_url": "https://www.liquid.ai/blog/lfm2-5-1-2b-thinking-on-device-reasoning-under-1gb", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1 (mean of 5 runs)", "judge": "rule-based", "harness": "official (LFM eval, third-party)", "prompt_style": "default", "temperature": "default" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Liquid AI LFM2.5-1.2B-Thinking blog: lfm2.5-1.2b-instruct=86.23. (third-party self-test by Liquid AI)", "candidates": [] }, { "model_id": "lfm2.5-1.2b-instruct", "benchmark_id": "ifbench", "score": 47.33, "reference_url": "https://www.liquid.ai/blog/lfm2-5-1-2b-thinking-on-device-reasoning-under-1gb", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1 (mean of 5 runs)", "judge": "rule-based", "harness": "official (LFM eval, third-party)", "prompt_style": "default", "temperature": "default" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Liquid AI LFM2.5-1.2B-Thinking blog: lfm2.5-1.2b-instruct=47.33. (third-party self-test by Liquid AI)", "candidates": [] }, { "model_id": "lfm2.5-1.2b-instruct", "benchmark_id": "multi_if", "score": 60.98, "reference_url": "https://www.liquid.ai/blog/lfm2-5-1-2b-thinking-on-device-reasoning-under-1gb", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1 (mean of 5 runs)", "judge": "rule-based", "harness": "official (LFM eval, third-party)", "prompt_style": "default", "temperature": "default" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Liquid AI LFM2.5-1.2B-Thinking blog: lfm2.5-1.2b-instruct=60.98. (third-party self-test by Liquid AI)", "candidates": [] }, { "model_id": "lfm2.5-1.2b-instruct", "benchmark_id": "gsm8k", "score": 64.52, "reference_url": "https://www.liquid.ai/blog/lfm2-5-1-2b-thinking-on-device-reasoning-under-1gb", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1 (mean of 5 runs)", "judge": "rule-based", "harness": "official (LFM eval, third-party)", "prompt_style": "default", "temperature": "default" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Liquid AI LFM2.5-1.2B-Thinking blog: lfm2.5-1.2b-instruct=64.52. (third-party self-test by Liquid AI)", "candidates": [] }, { "model_id": "lfm2.5-1.2b-instruct", "benchmark_id": "math_500", "score": 63.2, "reference_url": "https://www.liquid.ai/blog/lfm2-5-1-2b-thinking-on-device-reasoning-under-1gb", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1 (mean of 5 runs)", "judge": "rule-based", "harness": "official (LFM eval, third-party)", "prompt_style": "default", "temperature": "default" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Liquid AI LFM2.5-1.2B-Thinking blog: lfm2.5-1.2b-instruct=63.2. (third-party self-test by Liquid AI)", "candidates": [] }, { "model_id": "lfm2.5-1.2b-instruct", "benchmark_id": "aime_2025", "score": 14.0, "reference_url": "https://www.liquid.ai/blog/lfm2-5-1-2b-thinking-on-device-reasoning-under-1gb", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1 (mean of 5 runs)", "judge": "rule-based", "harness": "official (LFM eval, third-party)", "prompt_style": "default", "temperature": "default" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Liquid AI LFM2.5-1.2B-Thinking blog: lfm2.5-1.2b-instruct=14.0. (third-party self-test by Liquid AI)", "candidates": [] }, { "model_id": "lfm2.5-1.2b-instruct", "benchmark_id": "bfcl_v3", "score": 49.12, "reference_url": "https://www.liquid.ai/blog/lfm2-5-1-2b-thinking-on-device-reasoning-under-1gb", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1 (mean of 5 runs)", "judge": "rule-based", "harness": "official (LFM eval, third-party)", "prompt_style": "default", "temperature": "default" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Liquid AI LFM2.5-1.2B-Thinking blog: lfm2.5-1.2b-instruct=49.12. (third-party self-test by Liquid AI)", "candidates": [] }, { "model_id": "granite-4.0-h-1b", "benchmark_id": "gpqa_diamond", "score": 24.34, "reference_url": "https://www.liquid.ai/blog/lfm2-5-1-2b-thinking-on-device-reasoning-under-1gb", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1 (mean of 5 runs)", "judge": "rule-based", "harness": "official (LFM eval, third-party)", "prompt_style": "default", "temperature": "default" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Liquid AI LFM2.5-1.2B-Thinking blog: granite-4.0-h-1b=24.34. (third-party self-test by Liquid AI)", "candidates": [] }, { "model_id": "granite-4.0-h-1b", "benchmark_id": "mmlu_pro", "score": 27.64, "reference_url": "https://www.liquid.ai/blog/lfm2-5-1-2b-thinking-on-device-reasoning-under-1gb", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1 (mean of 5 runs)", "judge": "rule-based", "harness": "official (LFM eval, third-party)", "prompt_style": "default", "temperature": "default" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Liquid AI LFM2.5-1.2B-Thinking blog: granite-4.0-h-1b=27.64. (third-party self-test by Liquid AI)", "candidates": [] }, { "model_id": "granite-4.0-h-1b", "benchmark_id": "ifeval", "score": 80.08, "reference_url": "https://www.liquid.ai/blog/lfm2-5-1-2b-thinking-on-device-reasoning-under-1gb", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1 (mean of 5 runs)", "judge": "rule-based", "harness": "official (LFM eval, third-party)", "prompt_style": "default", "temperature": "default" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Liquid AI LFM2.5-1.2B-Thinking blog: granite-4.0-h-1b=80.08. (third-party self-test by Liquid AI)", "candidates": [] }, { "model_id": "granite-4.0-h-1b", "benchmark_id": "ifbench", "score": 24.93, "reference_url": "https://www.liquid.ai/blog/lfm2-5-1-2b-thinking-on-device-reasoning-under-1gb", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1 (mean of 5 runs)", "judge": "rule-based", "harness": "official (LFM eval, third-party)", "prompt_style": "default", "temperature": "default" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Liquid AI LFM2.5-1.2B-Thinking blog: granite-4.0-h-1b=24.93. (third-party self-test by Liquid AI)", "candidates": [] }, { "model_id": "granite-4.0-h-1b", "benchmark_id": "multi_if", "score": 47.56, "reference_url": "https://www.liquid.ai/blog/lfm2-5-1-2b-thinking-on-device-reasoning-under-1gb", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1 (mean of 5 runs)", "judge": "rule-based", "harness": "official (LFM eval, third-party)", "prompt_style": "default", "temperature": "default" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Liquid AI LFM2.5-1.2B-Thinking blog: granite-4.0-h-1b=47.56. (third-party self-test by Liquid AI)", "candidates": [] }, { "model_id": "granite-4.0-h-1b", "benchmark_id": "gsm8k", "score": 69.6, "reference_url": "https://www.liquid.ai/blog/lfm2-5-1-2b-thinking-on-device-reasoning-under-1gb", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1 (mean of 5 runs)", "judge": "rule-based", "harness": "official (LFM eval, third-party)", "prompt_style": "default", "temperature": "default" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Liquid AI LFM2.5-1.2B-Thinking blog: granite-4.0-h-1b=69.6. (third-party self-test by Liquid AI)", "candidates": [] }, { "model_id": "granite-4.0-h-1b", "benchmark_id": "math_500", "score": 47.2, "reference_url": "https://www.liquid.ai/blog/lfm2-5-1-2b-thinking-on-device-reasoning-under-1gb", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1 (mean of 5 runs)", "judge": "rule-based", "harness": "official (LFM eval, third-party)", "prompt_style": "default", "temperature": "default" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Liquid AI LFM2.5-1.2B-Thinking blog: granite-4.0-h-1b=47.2. (third-party self-test by Liquid AI)", "candidates": [] }, { "model_id": "granite-4.0-h-1b", "benchmark_id": "aime_2025", "score": 1.0, "reference_url": "https://www.liquid.ai/blog/lfm2-5-1-2b-thinking-on-device-reasoning-under-1gb", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1 (mean of 5 runs)", "judge": "rule-based", "harness": "official (LFM eval, third-party)", "prompt_style": "default", "temperature": "default" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Liquid AI LFM2.5-1.2B-Thinking blog: granite-4.0-h-1b=1.0. (third-party self-test by Liquid AI)", "candidates": [] }, { "model_id": "granite-4.0-h-1b", "benchmark_id": "bfcl_v3", "score": 50.69, "reference_url": "https://www.liquid.ai/blog/lfm2-5-1-2b-thinking-on-device-reasoning-under-1gb", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1 (mean of 5 runs)", "judge": "rule-based", "harness": "official (LFM eval, third-party)", "prompt_style": "default", "temperature": "default" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Liquid AI LFM2.5-1.2B-Thinking blog: granite-4.0-h-1b=50.69. (third-party self-test by Liquid AI)", "candidates": [] }, { "model_id": "granite-4.0-1b", "benchmark_id": "gpqa_diamond", "score": 24.24, "reference_url": "https://www.liquid.ai/blog/lfm2-5-1-2b-thinking-on-device-reasoning-under-1gb", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1 (mean of 5 runs)", "judge": "rule-based", "harness": "official (LFM eval, third-party)", "prompt_style": "default", "temperature": "default" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Liquid AI LFM2.5-1.2B-Thinking blog: granite-4.0-1b=24.24. (third-party self-test by Liquid AI)", "candidates": [] }, { "model_id": "granite-4.0-1b", "benchmark_id": "mmlu_pro", "score": 33.53, "reference_url": "https://www.liquid.ai/blog/lfm2-5-1-2b-thinking-on-device-reasoning-under-1gb", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1 (mean of 5 runs)", "judge": "rule-based", "harness": "official (LFM eval, third-party)", "prompt_style": "default", "temperature": "default" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Liquid AI LFM2.5-1.2B-Thinking blog: granite-4.0-1b=33.53. (third-party self-test by Liquid AI)", "candidates": [] }, { "model_id": "granite-4.0-1b", "benchmark_id": "ifeval", "score": 79.61, "reference_url": "https://www.liquid.ai/blog/lfm2-5-1-2b-thinking-on-device-reasoning-under-1gb", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1 (mean of 5 runs)", "judge": "rule-based", "harness": "official (LFM eval, third-party)", "prompt_style": "default", "temperature": "default" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Liquid AI LFM2.5-1.2B-Thinking blog: granite-4.0-1b=79.61. (third-party self-test by Liquid AI)", "candidates": [] }, { "model_id": "granite-4.0-1b", "benchmark_id": "ifbench", "score": 21.0, "reference_url": "https://www.liquid.ai/blog/lfm2-5-1-2b-thinking-on-device-reasoning-under-1gb", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1 (mean of 5 runs)", "judge": "rule-based", "harness": "official (LFM eval, third-party)", "prompt_style": "default", "temperature": "default" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Liquid AI LFM2.5-1.2B-Thinking blog: granite-4.0-1b=21.0. (third-party self-test by Liquid AI)", "candidates": [] }, { "model_id": "granite-4.0-1b", "benchmark_id": "multi_if", "score": 43.65, "reference_url": "https://www.liquid.ai/blog/lfm2-5-1-2b-thinking-on-device-reasoning-under-1gb", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1 (mean of 5 runs)", "judge": "rule-based", "harness": "official (LFM eval, third-party)", "prompt_style": "default", "temperature": "default" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Liquid AI LFM2.5-1.2B-Thinking blog: granite-4.0-1b=43.65. (third-party self-test by Liquid AI)", "candidates": [] }, { "model_id": "granite-4.0-1b", "benchmark_id": "gsm8k", "score": 73.42, "reference_url": "https://www.liquid.ai/blog/lfm2-5-1-2b-thinking-on-device-reasoning-under-1gb", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1 (mean of 5 runs)", "judge": "rule-based", "harness": "official (LFM eval, third-party)", "prompt_style": "default", "temperature": "default" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Liquid AI LFM2.5-1.2B-Thinking blog: granite-4.0-1b=73.42. (third-party self-test by Liquid AI)", "candidates": [] }, { "model_id": "granite-4.0-1b", "benchmark_id": "math_500", "score": 44.8, "reference_url": "https://www.liquid.ai/blog/lfm2-5-1-2b-thinking-on-device-reasoning-under-1gb", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1 (mean of 5 runs)", "judge": "rule-based", "harness": "official (LFM eval, third-party)", "prompt_style": "default", "temperature": "default" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Liquid AI LFM2.5-1.2B-Thinking blog: granite-4.0-1b=44.8. (third-party self-test by Liquid AI)", "candidates": [] }, { "model_id": "granite-4.0-1b", "benchmark_id": "aime_2025", "score": 3.33, "reference_url": "https://www.liquid.ai/blog/lfm2-5-1-2b-thinking-on-device-reasoning-under-1gb", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1 (mean of 5 runs)", "judge": "rule-based", "harness": "official (LFM eval, third-party)", "prompt_style": "default", "temperature": "default" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Liquid AI LFM2.5-1.2B-Thinking blog: granite-4.0-1b=3.33. (third-party self-test by Liquid AI)", "candidates": [] }, { "model_id": "granite-4.0-1b", "benchmark_id": "bfcl_v3", "score": 52.43, "reference_url": "https://www.liquid.ai/blog/lfm2-5-1-2b-thinking-on-device-reasoning-under-1gb", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1 (mean of 5 runs)", "judge": "rule-based", "harness": "official (LFM eval, third-party)", "prompt_style": "default", "temperature": "default" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Liquid AI LFM2.5-1.2B-Thinking blog: granite-4.0-1b=52.43. (third-party self-test by Liquid AI)", "candidates": [] }, { "model_id": "gemma-3-1b-it", "benchmark_id": "gpqa_diamond", "score": 24.24, "reference_url": "https://www.liquid.ai/blog/lfm2-5-1-2b-thinking-on-device-reasoning-under-1gb", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1 (mean of 5 runs)", "judge": "rule-based", "harness": "official (LFM eval, third-party)", "prompt_style": "default", "temperature": "default" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Liquid AI LFM2.5-1.2B-Thinking blog: gemma-3-1b-it=24.24. (third-party self-test by Liquid AI)", "candidates": [ { "score": 23.89, "reference_url": "https://www.liquid.ai/blog/lfm2-5-230m", "source_type": "official_blog", "reported_setting": { "mode": "non-thinking", "effort": "source does not state", "tools": "none", "sampling": "source does not state trial count", "judge": "benchmark-specified", "harness": "Liquid AI official comparative evaluation", "prompt_style": "source does not state", "temperature": "source does not state evaluation parameters", "context": "source does not state evaluation context" }, "notes": "Displayed exactly as 23.89. Official Liquid AI reported result." } ] }, { "model_id": "gemma-3-1b-it", "benchmark_id": "mmlu_pro", "score": 14.04, "reference_url": "https://www.liquid.ai/blog/lfm2-5-1-2b-thinking-on-device-reasoning-under-1gb", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1 (mean of 5 runs)", "judge": "rule-based", "harness": "official (LFM eval, third-party)", "prompt_style": "default", "temperature": "default" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Liquid AI LFM2.5-1.2B-Thinking blog: gemma-3-1b-it=14.04. (third-party self-test by Liquid AI)", "candidates": [ { "score": 14.04, "reference_url": "https://www.liquid.ai/blog/lfm2-5-230m", "source_type": "official_blog", "reported_setting": { "mode": "non-thinking", "effort": "source does not state", "tools": "none", "sampling": "source does not state trial count", "judge": "benchmark-specified", "harness": "Liquid AI official comparative evaluation", "prompt_style": "source does not state", "temperature": "source does not state evaluation parameters", "context": "source does not state evaluation context" }, "notes": "Displayed exactly as 14.04. Official Liquid AI reported result." } ] }, { "model_id": "gemma-3-1b-it", "benchmark_id": "ifeval", "score": 63.25, "reference_url": "https://www.liquid.ai/blog/lfm2-5-1-2b-thinking-on-device-reasoning-under-1gb", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1 (mean of 5 runs)", "judge": "rule-based", "harness": "official (LFM eval, third-party)", "prompt_style": "default", "temperature": "default" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Liquid AI LFM2.5-1.2B-Thinking blog: gemma-3-1b-it=63.25. (third-party self-test by Liquid AI)", "candidates": [ { "score": 63.49, "reference_url": "https://www.liquid.ai/blog/lfm2-5-230m", "source_type": "official_blog", "reported_setting": { "mode": "non-thinking", "effort": "source does not state", "tools": "none", "sampling": "source does not state trial count", "judge": "benchmark-specified", "harness": "Liquid AI official comparative evaluation", "prompt_style": "source does not state", "temperature": "source does not state evaluation parameters", "context": "source does not state evaluation context" }, "notes": "Displayed exactly as 63.49. Official Liquid AI reported result." } ] }, { "model_id": "gemma-3-1b-it", "benchmark_id": "ifbench", "score": 20.47, "reference_url": "https://www.liquid.ai/blog/lfm2-5-1-2b-thinking-on-device-reasoning-under-1gb", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1 (mean of 5 runs)", "judge": "rule-based", "harness": "official (LFM eval, third-party)", "prompt_style": "default", "temperature": "default" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Liquid AI LFM2.5-1.2B-Thinking blog: gemma-3-1b-it=20.47. (third-party self-test by Liquid AI)", "candidates": [ { "score": 20.33, "reference_url": "https://www.liquid.ai/blog/lfm2-5-230m", "source_type": "official_blog", "reported_setting": { "mode": "non-thinking", "effort": "source does not state", "tools": "none", "sampling": "source does not state trial count", "judge": "benchmark-specified", "harness": "Liquid AI official comparative evaluation", "prompt_style": "source does not state", "temperature": "source does not state evaluation parameters", "context": "source does not state evaluation context" }, "notes": "Displayed exactly as 20.33. Official Liquid AI reported result." } ] }, { "model_id": "gemma-3-1b-it", "benchmark_id": "multi_if", "score": 44.31, "reference_url": "https://www.liquid.ai/blog/lfm2-5-1-2b-thinking-on-device-reasoning-under-1gb", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1 (mean of 5 runs)", "judge": "rule-based", "harness": "official (LFM eval, third-party)", "prompt_style": "default", "temperature": "default" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Liquid AI LFM2.5-1.2B-Thinking blog: gemma-3-1b-it=44.31. (third-party self-test by Liquid AI)", "candidates": [ { "score": 44.25, "reference_url": "https://www.liquid.ai/blog/lfm2-5-230m", "source_type": "official_blog", "reported_setting": { "mode": "non-thinking", "effort": "source does not state", "tools": "none", "sampling": "source does not state trial count", "judge": "benchmark-specified", "harness": "Liquid AI official comparative evaluation", "prompt_style": "source does not state", "temperature": "source does not state evaluation parameters", "context": "source does not state evaluation context" }, "notes": "Displayed exactly as 44.25. Official Liquid AI reported result." } ] }, { "model_id": "gemma-3-1b-it", "benchmark_id": "gsm8k", "score": 42.15, "reference_url": "https://www.liquid.ai/blog/lfm2-5-1-2b-thinking-on-device-reasoning-under-1gb", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1 (mean of 5 runs)", "judge": "rule-based", "harness": "official (LFM eval, third-party)", "prompt_style": "default", "temperature": "default" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Liquid AI LFM2.5-1.2B-Thinking blog: gemma-3-1b-it=42.15. (third-party self-test by Liquid AI)", "candidates": [] }, { "model_id": "gemma-3-1b-it", "benchmark_id": "math_500", "score": 45.2, "reference_url": "https://www.liquid.ai/blog/lfm2-5-1-2b-thinking-on-device-reasoning-under-1gb", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1 (mean of 5 runs)", "judge": "rule-based", "harness": "official (LFM eval, third-party)", "prompt_style": "default", "temperature": "default" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Liquid AI LFM2.5-1.2B-Thinking blog: gemma-3-1b-it=45.2. (third-party self-test by Liquid AI)", "candidates": [] }, { "model_id": "gemma-3-1b-it", "benchmark_id": "aime_2025", "score": 1.0, "reference_url": "https://www.liquid.ai/blog/lfm2-5-1-2b-thinking-on-device-reasoning-under-1gb", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1 (mean of 5 runs)", "judge": "rule-based", "harness": "official (LFM eval, third-party)", "prompt_style": "default", "temperature": "default" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Liquid AI LFM2.5-1.2B-Thinking blog: gemma-3-1b-it=1.0. (third-party self-test by Liquid AI)", "candidates": [] }, { "model_id": "gemma-3-1b-it", "benchmark_id": "bfcl_v3", "score": 16.64, "reference_url": "https://www.liquid.ai/blog/lfm2-5-1-2b-thinking-on-device-reasoning-under-1gb", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1 (mean of 5 runs)", "judge": "rule-based", "harness": "official (LFM eval, third-party)", "prompt_style": "default", "temperature": "default" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Liquid AI LFM2.5-1.2B-Thinking blog: gemma-3-1b-it=16.64. (third-party self-test by Liquid AI)", "candidates": [ { "score": 16.61, "reference_url": "https://www.liquid.ai/blog/lfm2-5-230m", "source_type": "official_blog", "reported_setting": { "mode": "non-thinking", "effort": "source does not state", "tools": "benchmark-provided tools", "sampling": "source does not state trial count", "judge": "benchmark-specified", "harness": "Liquid AI official comparative evaluation", "prompt_style": "source does not state", "temperature": "source does not state evaluation parameters", "context": "source does not state evaluation context" }, "notes": "Displayed exactly as 16.61. Official Liquid AI reported result." } ] }, { "model_id": "llama-3.2-1b-instruct", "benchmark_id": "gpqa_diamond", "score": 16.57, "reference_url": "https://www.liquid.ai/blog/lfm2-5-1-2b-thinking-on-device-reasoning-under-1gb", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1 (mean of 5 runs)", "judge": "rule-based", "harness": "official (LFM eval, third-party)", "prompt_style": "default", "temperature": "default" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Liquid AI LFM2.5-1.2B-Thinking blog: llama-3.2-1b-instruct=16.57. (third-party self-test by Liquid AI)", "candidates": [] }, { "model_id": "llama-3.2-1b-instruct", "benchmark_id": "mmlu_pro", "score": 20.8, "reference_url": "https://www.liquid.ai/blog/lfm2-5-1-2b-thinking-on-device-reasoning-under-1gb", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1 (mean of 5 runs)", "judge": "rule-based", "harness": "official (LFM eval, third-party)", "prompt_style": "default", "temperature": "default" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Liquid AI LFM2.5-1.2B-Thinking blog: llama-3.2-1b-instruct=20.8. (third-party self-test by Liquid AI)", "candidates": [] }, { "model_id": "llama-3.2-1b-instruct", "benchmark_id": "ifeval", "score": 52.37, "reference_url": "https://www.liquid.ai/blog/lfm2-5-1-2b-thinking-on-device-reasoning-under-1gb", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1 (mean of 5 runs)", "judge": "rule-based", "harness": "official (LFM eval, third-party)", "prompt_style": "default", "temperature": "default" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Liquid AI LFM2.5-1.2B-Thinking blog: llama-3.2-1b-instruct=52.37. (third-party self-test by Liquid AI)", "candidates": [] }, { "model_id": "llama-3.2-1b-instruct", "benchmark_id": "ifbench", "score": 15.93, "reference_url": "https://www.liquid.ai/blog/lfm2-5-1-2b-thinking-on-device-reasoning-under-1gb", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1 (mean of 5 runs)", "judge": "rule-based", "harness": "official (LFM eval, third-party)", "prompt_style": "default", "temperature": "default" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Liquid AI LFM2.5-1.2B-Thinking blog: llama-3.2-1b-instruct=15.93. (third-party self-test by Liquid AI)", "candidates": [] }, { "model_id": "llama-3.2-1b-instruct", "benchmark_id": "multi_if", "score": 30.16, "reference_url": "https://www.liquid.ai/blog/lfm2-5-1-2b-thinking-on-device-reasoning-under-1gb", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1 (mean of 5 runs)", "judge": "rule-based", "harness": "official (LFM eval, third-party)", "prompt_style": "default", "temperature": "default" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Liquid AI LFM2.5-1.2B-Thinking blog: llama-3.2-1b-instruct=30.16. (third-party self-test by Liquid AI)", "candidates": [] }, { "model_id": "llama-3.2-1b-instruct", "benchmark_id": "gsm8k", "score": 39.04, "reference_url": "https://www.liquid.ai/blog/lfm2-5-1-2b-thinking-on-device-reasoning-under-1gb", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1 (mean of 5 runs)", "judge": "rule-based", "harness": "official (LFM eval, third-party)", "prompt_style": "default", "temperature": "default" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Liquid AI LFM2.5-1.2B-Thinking blog: llama-3.2-1b-instruct=39.04. (third-party self-test by Liquid AI)", "candidates": [] }, { "model_id": "llama-3.2-1b-instruct", "benchmark_id": "math_500", "score": 23.4, "reference_url": "https://www.liquid.ai/blog/lfm2-5-1-2b-thinking-on-device-reasoning-under-1gb", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1 (mean of 5 runs)", "judge": "rule-based", "harness": "official (LFM eval, third-party)", "prompt_style": "default", "temperature": "default" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Liquid AI LFM2.5-1.2B-Thinking blog: llama-3.2-1b-instruct=23.4. (third-party self-test by Liquid AI)", "candidates": [] }, { "model_id": "llama-3.2-1b-instruct", "benchmark_id": "aime_2025", "score": 0.33, "reference_url": "https://www.liquid.ai/blog/lfm2-5-1-2b-thinking-on-device-reasoning-under-1gb", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1 (mean of 5 runs)", "judge": "rule-based", "harness": "official (LFM eval, third-party)", "prompt_style": "default", "temperature": "default" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Liquid AI LFM2.5-1.2B-Thinking blog: llama-3.2-1b-instruct=0.33. (third-party self-test by Liquid AI)", "candidates": [] }, { "model_id": "llama-3.2-1b-instruct", "benchmark_id": "bfcl_v3", "score": 21.44, "reference_url": "https://www.liquid.ai/blog/lfm2-5-1-2b-thinking-on-device-reasoning-under-1gb", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1 (mean of 5 runs)", "judge": "rule-based", "harness": "official (LFM eval, third-party)", "prompt_style": "default", "temperature": "default" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Liquid AI LFM2.5-1.2B-Thinking blog: llama-3.2-1b-instruct=21.44. (third-party self-test by Liquid AI)", "candidates": [] }, { "model_id": "mistral-medium-3", "benchmark_id": "livecodebench_v6", "score": 30.3, "reference_url": "https://mistral.ai/news/mistral-medium-3", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "mistral-eval", "prompt_style": "0-shot", "temperature": "0.0" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "Mistral Medium 3 blog table: 30.3." }, { "model_id": "mistral-medium-3", "benchmark_id": "multipl_e_avg", "score": 81.4, "reference_url": "https://mistral.ai/news/mistral-medium-3", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "mistral-eval", "prompt_style": "0-shot", "temperature": "0.0" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "Mistral Medium 3 blog table: 81.4." }, { "model_id": "mistral-medium-3", "benchmark_id": "ruler_32k", "score": 96.0, "reference_url": "https://mistral.ai/news/mistral-medium-3", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "mistral-eval", "prompt_style": "0-shot", "temperature": "0.0" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "Mistral Medium 3 blog table: 96.0." }, { "model_id": "mistral-medium-3", "benchmark_id": "ruler_128k", "score": 90.2, "reference_url": "https://mistral.ai/news/mistral-medium-3", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "mistral-eval", "prompt_style": "0-shot", "temperature": "0.0" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "Mistral Medium 3 blog table: 90.2." }, { "model_id": "mistral-medium-3", "benchmark_id": "mmmu", "score": 66.1, "reference_url": "https://mistral.ai/news/mistral-medium-3", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "mistral-eval", "prompt_style": "0-shot", "temperature": "0.0" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "Mistral Medium 3 blog table: 66.1." }, { "model_id": "mistral-medium-3", "benchmark_id": "docvqa", "score": 95.3, "reference_url": "https://mistral.ai/news/mistral-medium-3", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "mistral-eval", "prompt_style": "0-shot", "temperature": "0.0" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "Mistral Medium 3 blog table: 95.3." }, { "model_id": "mistral-medium-3", "benchmark_id": "ai2d", "score": 93.7, "reference_url": "https://mistral.ai/news/mistral-medium-3", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "mistral-eval", "prompt_style": "0-shot", "temperature": "0.0" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "Mistral Medium 3 blog table: 93.7." }, { "model_id": "mistral-medium-3", "benchmark_id": "chartqa", "score": 82.6, "reference_url": "https://mistral.ai/news/mistral-medium-3", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "mistral-eval", "prompt_style": "0-shot", "temperature": "0.0" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "Mistral Medium 3 blog table: 82.6." }, { "model_id": "llama-4-maverick", "benchmark_id": "livecodebench_v6", "score": 28.7, "reference_url": "https://mistral.ai/news/mistral-medium-3", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "mistral-eval", "prompt_style": "0-shot", "temperature": "0.0" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Mistral Medium 3 blog table: 28.7. Third-party self-test by Mistral (their internal eval pipeline)." }, { "model_id": "llama-4-maverick", "benchmark_id": "multipl_e_avg", "score": 76.4, "reference_url": "https://mistral.ai/news/mistral-medium-3", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "mistral-eval", "prompt_style": "0-shot", "temperature": "0.0" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Mistral Medium 3 blog table: 76.4. Third-party self-test by Mistral (their internal eval pipeline)." }, { "model_id": "llama-4-maverick", "benchmark_id": "ruler_32k", "score": 94.8, "reference_url": "https://mistral.ai/news/mistral-medium-3", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "mistral-eval", "prompt_style": "0-shot", "temperature": "0.0" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Mistral Medium 3 blog table: 94.8. Third-party self-test by Mistral (their internal eval pipeline)." }, { "model_id": "llama-4-maverick", "benchmark_id": "ruler_128k", "score": 86.7, "reference_url": "https://mistral.ai/news/mistral-medium-3", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "mistral-eval", "prompt_style": "0-shot", "temperature": "0.0" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Mistral Medium 3 blog table: 86.7. Third-party self-test by Mistral (their internal eval pipeline)." }, { "model_id": "llama-4-maverick", "benchmark_id": "docvqa", "score": 94.1, "reference_url": "https://mistral.ai/news/mistral-medium-3", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "mistral-eval", "prompt_style": "0-shot", "temperature": "0.0" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Mistral Medium 3 blog table: 94.1. Third-party self-test by Mistral (their internal eval pipeline)." }, { "model_id": "llama-4-maverick", "benchmark_id": "ai2d", "score": 84.4, "reference_url": "https://mistral.ai/news/mistral-medium-3", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "mistral-eval", "prompt_style": "0-shot", "temperature": "0.0" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Mistral Medium 3 blog table: 84.4. Third-party self-test by Mistral (their internal eval pipeline)." }, { "model_id": "llama-4-maverick", "benchmark_id": "chartqa", "score": 90.4, "reference_url": "https://mistral.ai/news/mistral-medium-3", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "mistral-eval", "prompt_style": "0-shot", "temperature": "0.0" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Mistral Medium 3 blog table: 90.4. Third-party self-test by Mistral (their internal eval pipeline)." }, { "model_id": "gpt-4o", "benchmark_id": "humaneval", "score": 91.5, "reference_url": "https://mistral.ai/news/mistral-medium-3", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "mistral-eval", "prompt_style": "0-shot", "temperature": "0.0" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Mistral Medium 3 blog table: 91.5. Third-party self-test by Mistral (their internal eval pipeline)." }, { "model_id": "gpt-4o", "benchmark_id": "livecodebench_v6", "score": 31.4, "reference_url": "https://mistral.ai/news/mistral-medium-3", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "mistral-eval", "prompt_style": "0-shot", "temperature": "0.0" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Mistral Medium 3 blog table: 31.4. Third-party self-test by Mistral (their internal eval pipeline)." }, { "model_id": "gpt-4o", "benchmark_id": "multipl_e_avg", "score": 79.8, "reference_url": "https://mistral.ai/news/mistral-medium-3", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "mistral-eval", "prompt_style": "0-shot", "temperature": "0.0" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Mistral Medium 3 blog table: 79.8. Third-party self-test by Mistral (their internal eval pipeline)." }, { "model_id": "gpt-4o", "benchmark_id": "arena_hard", "score": 95.4, "reference_url": "https://mistral.ai/news/mistral-medium-3", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "mistral-eval", "prompt_style": "0-shot", "temperature": "0.0" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Mistral Medium 3 blog table: 95.4. Third-party self-test by Mistral (their internal eval pipeline)." }, { "model_id": "gpt-4o", "benchmark_id": "math_500", "score": 76.4, "reference_url": "https://mistral.ai/news/mistral-medium-3", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "mistral-eval", "prompt_style": "0-shot", "temperature": "0.0" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Mistral Medium 3 blog table: 76.4. Third-party self-test by Mistral (their internal eval pipeline)." }, { "model_id": "gpt-4o", "benchmark_id": "mmlu_pro", "score": 75.8, "reference_url": "https://mistral.ai/news/mistral-medium-3", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "mistral-eval", "prompt_style": "5-shot CoT", "temperature": "0.0" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Mistral Medium 3 blog table: 75.8. Third-party self-test by Mistral (their internal eval pipeline)." }, { "model_id": "gpt-4o", "benchmark_id": "ruler_32k", "score": 96.0, "reference_url": "https://mistral.ai/news/mistral-medium-3", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "mistral-eval", "prompt_style": "0-shot", "temperature": "0.0" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Mistral Medium 3 blog table: 96.0. Third-party self-test by Mistral (their internal eval pipeline)." }, { "model_id": "gpt-4o", "benchmark_id": "ruler_128k", "score": 88.9, "reference_url": "https://mistral.ai/news/mistral-medium-3", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "mistral-eval", "prompt_style": "0-shot", "temperature": "0.0" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Mistral Medium 3 blog table: 88.9. Third-party self-test by Mistral (their internal eval pipeline)." }, { "model_id": "gpt-4o", "benchmark_id": "docvqa", "score": 85.9, "reference_url": "https://mistral.ai/news/mistral-medium-3", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "mistral-eval", "prompt_style": "0-shot", "temperature": "0.0" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Mistral Medium 3 blog table: 85.9. Third-party self-test by Mistral (their internal eval pipeline)." }, { "model_id": "gpt-4o", "benchmark_id": "ai2d", "score": 93.3, "reference_url": "https://mistral.ai/news/mistral-medium-3", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "mistral-eval", "prompt_style": "0-shot", "temperature": "0.0" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Mistral Medium 3 blog table: 93.3. Third-party self-test by Mistral (their internal eval pipeline)." }, { "model_id": "gpt-4o", "benchmark_id": "chartqa", "score": 86.0, "reference_url": "https://mistral.ai/news/mistral-medium-3", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "mistral-eval", "prompt_style": "0-shot", "temperature": "0.0" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Mistral Medium 3 blog table: 86.0. Third-party self-test by Mistral (their internal eval pipeline)." }, { "model_id": "claude-sonnet-3.7", "benchmark_id": "humaneval", "score": 92.1, "reference_url": "https://mistral.ai/news/mistral-medium-3", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "mistral-eval", "prompt_style": "0-shot", "temperature": "0.0" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Mistral Medium 3 blog table: 92.1. Third-party self-test by Mistral (their internal eval pipeline)." }, { "model_id": "claude-sonnet-3.7", "benchmark_id": "livecodebench_v6", "score": 36.0, "reference_url": "https://mistral.ai/news/mistral-medium-3", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "mistral-eval", "prompt_style": "0-shot", "temperature": "0.0" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Mistral Medium 3 blog table: 36.0. Third-party self-test by Mistral (their internal eval pipeline)." }, { "model_id": "claude-sonnet-3.7", "benchmark_id": "multipl_e_avg", "score": 83.4, "reference_url": "https://mistral.ai/news/mistral-medium-3", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "mistral-eval", "prompt_style": "0-shot", "temperature": "0.0" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Mistral Medium 3 blog table: 83.4. Third-party self-test by Mistral (their internal eval pipeline)." }, { "model_id": "claude-sonnet-3.7", "benchmark_id": "arena_hard", "score": 93.2, "reference_url": "https://mistral.ai/news/mistral-medium-3", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "mistral-eval", "prompt_style": "0-shot", "temperature": "0.0" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Mistral Medium 3 blog table: 93.2. Third-party self-test by Mistral (their internal eval pipeline)." }, { "model_id": "claude-sonnet-3.7", "benchmark_id": "ifeval", "score": 91.8, "reference_url": "https://mistral.ai/news/mistral-medium-3", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "mistral-eval", "prompt_style": "0-shot", "temperature": "0.0" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Mistral Medium 3 blog table: 91.8. Third-party self-test by Mistral (their internal eval pipeline)." }, { "model_id": "claude-sonnet-3.7", "benchmark_id": "math_500", "score": 83.0, "reference_url": "https://mistral.ai/news/mistral-medium-3", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "mistral-eval", "prompt_style": "0-shot", "temperature": "0.0" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Mistral Medium 3 blog table: 83.0. Third-party self-test by Mistral (their internal eval pipeline)." }, { "model_id": "claude-sonnet-3.7", "benchmark_id": "gpqa_diamond", "score": 69.7, "reference_url": "https://mistral.ai/news/mistral-medium-3", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "mistral-eval", "prompt_style": "5-shot CoT", "temperature": "0.0" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Mistral Medium 3 blog table: 69.7. Third-party self-test by Mistral (their internal eval pipeline)." }, { "model_id": "claude-sonnet-3.7", "benchmark_id": "mmlu_pro", "score": 80.0, "reference_url": "https://mistral.ai/news/mistral-medium-3", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "mistral-eval", "prompt_style": "5-shot CoT", "temperature": "0.0" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Mistral Medium 3 blog table: 80.0. Third-party self-test by Mistral (their internal eval pipeline)." }, { "model_id": "claude-sonnet-3.7", "benchmark_id": "ruler_32k", "score": 95.7, "reference_url": "https://mistral.ai/news/mistral-medium-3", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "mistral-eval", "prompt_style": "0-shot", "temperature": "0.0" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Mistral Medium 3 blog table: 95.7. Third-party self-test by Mistral (their internal eval pipeline)." }, { "model_id": "claude-sonnet-3.7", "benchmark_id": "ruler_128k", "score": 93.8, "reference_url": "https://mistral.ai/news/mistral-medium-3", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "mistral-eval", "prompt_style": "0-shot", "temperature": "0.0" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Mistral Medium 3 blog table: 93.8. Third-party self-test by Mistral (their internal eval pipeline)." }, { "model_id": "claude-sonnet-3.7", "benchmark_id": "mmmu", "score": 71.3, "reference_url": "https://mistral.ai/news/mistral-medium-3", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "mistral-eval", "prompt_style": "0-shot", "temperature": "0.0" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Mistral Medium 3 blog table: 71.3. Third-party self-test by Mistral (their internal eval pipeline)." }, { "model_id": "claude-sonnet-3.7", "benchmark_id": "docvqa", "score": 84.3, "reference_url": "https://mistral.ai/news/mistral-medium-3", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "mistral-eval", "prompt_style": "0-shot", "temperature": "0.0" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Mistral Medium 3 blog table: 84.3. Third-party self-test by Mistral (their internal eval pipeline)." }, { "model_id": "claude-sonnet-3.7", "benchmark_id": "ai2d", "score": 78.8, "reference_url": "https://mistral.ai/news/mistral-medium-3", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "mistral-eval", "prompt_style": "0-shot", "temperature": "0.0" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Mistral Medium 3 blog table: 78.8. Third-party self-test by Mistral (their internal eval pipeline)." }, { "model_id": "claude-sonnet-3.7", "benchmark_id": "chartqa", "score": 76.3, "reference_url": "https://mistral.ai/news/mistral-medium-3", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "mistral-eval", "prompt_style": "0-shot", "temperature": "0.0" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Mistral Medium 3 blog table: 76.3. Third-party self-test by Mistral (their internal eval pipeline)." }, { "model_id": "command-a", "benchmark_id": "livecodebench_v6", "score": 26.3, "reference_url": "https://mistral.ai/news/mistral-medium-3", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "mistral-eval", "prompt_style": "0-shot", "temperature": "0.0" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Mistral Medium 3 blog table: 26.3. Third-party self-test by Mistral (their internal eval pipeline)." }, { "model_id": "command-a", "benchmark_id": "multipl_e_avg", "score": 73.1, "reference_url": "https://mistral.ai/news/mistral-medium-3", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "mistral-eval", "prompt_style": "0-shot", "temperature": "0.0" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Mistral Medium 3 blog table: 73.1. Third-party self-test by Mistral (their internal eval pipeline)." }, { "model_id": "command-a", "benchmark_id": "ruler_32k", "score": 95.6, "reference_url": "https://mistral.ai/news/mistral-medium-3", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "mistral-eval", "prompt_style": "0-shot", "temperature": "0.0" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Mistral Medium 3 blog table: 95.6. Third-party self-test by Mistral (their internal eval pipeline)." }, { "model_id": "command-a", "benchmark_id": "ruler_128k", "score": 91.2, "reference_url": "https://mistral.ai/news/mistral-medium-3", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "mistral-eval", "prompt_style": "0-shot", "temperature": "0.0" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Mistral Medium 3 blog table: 91.2. Third-party self-test by Mistral (their internal eval pipeline)." }, { "model_id": "deepseek-3.1", "benchmark_id": "humaneval", "score": 93.3, "reference_url": "https://mistral.ai/news/mistral-medium-3", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "mistral-eval", "prompt_style": "0-shot", "temperature": "0.0" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Mistral Medium 3 blog table: 93.3. Third-party self-test by Mistral (their internal eval pipeline)." }, { "model_id": "deepseek-3.1", "benchmark_id": "livecodebench_v6", "score": 42.9, "reference_url": "https://mistral.ai/news/mistral-medium-3", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "mistral-eval", "prompt_style": "0-shot", "temperature": "0.0" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Mistral Medium 3 blog table: 42.9. Third-party self-test by Mistral (their internal eval pipeline)." }, { "model_id": "deepseek-3.1", "benchmark_id": "multipl_e_avg", "score": 84.9, "reference_url": "https://mistral.ai/news/mistral-medium-3", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "mistral-eval", "prompt_style": "0-shot", "temperature": "0.0" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Mistral Medium 3 blog table: 84.9. Third-party self-test by Mistral (their internal eval pipeline)." }, { "model_id": "deepseek-3.1", "benchmark_id": "arena_hard", "score": 97.3, "reference_url": "https://mistral.ai/news/mistral-medium-3", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "mistral-eval", "prompt_style": "0-shot", "temperature": "0.0" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Mistral Medium 3 blog table: 97.3. Third-party self-test by Mistral (their internal eval pipeline)." }, { "model_id": "deepseek-3.1", "benchmark_id": "ifeval", "score": 89.1, "reference_url": "https://mistral.ai/news/mistral-medium-3", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "mistral-eval", "prompt_style": "0-shot", "temperature": "0.0" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Mistral Medium 3 blog table: 89.1. Third-party self-test by Mistral (their internal eval pipeline)." }, { "model_id": "deepseek-3.1", "benchmark_id": "math_500", "score": 93.8, "reference_url": "https://mistral.ai/news/mistral-medium-3", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "mistral-eval", "prompt_style": "0-shot", "temperature": "0.0" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Mistral Medium 3 blog table: 93.8. Third-party self-test by Mistral (their internal eval pipeline)." }, { "model_id": "deepseek-3.1", "benchmark_id": "gpqa_diamond", "score": 61.1, "reference_url": "https://mistral.ai/news/mistral-medium-3", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "mistral-eval", "prompt_style": "5-shot CoT", "temperature": "0.0" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Mistral Medium 3 blog table: 61.1. Third-party self-test by Mistral (their internal eval pipeline)." }, { "model_id": "deepseek-3.1", "benchmark_id": "mmlu_pro", "score": 81.1, "reference_url": "https://mistral.ai/news/mistral-medium-3", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "mistral-eval", "prompt_style": "5-shot CoT", "temperature": "0.0" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Mistral Medium 3 blog table: 81.1. Third-party self-test by Mistral (their internal eval pipeline)." }, { "model_id": "deepseek-3.1", "benchmark_id": "ruler_32k", "score": 95.8, "reference_url": "https://mistral.ai/news/mistral-medium-3", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "mistral-eval", "prompt_style": "0-shot", "temperature": "0.0" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Mistral Medium 3 blog table: 95.8. Third-party self-test by Mistral (their internal eval pipeline)." }, { "model_id": "deepseek-3.1", "benchmark_id": "ruler_128k", "score": 91.9, "reference_url": "https://mistral.ai/news/mistral-medium-3", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "mistral-eval", "prompt_style": "0-shot", "temperature": "0.0" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Mistral Medium 3 blog table: 91.9. Third-party self-test by Mistral (their internal eval pipeline)." }, { "model_id": "command-r7b", "benchmark_id": "mmlu", "score": 65.2, "reference_url": "https://cohere.com/research/papers/command-a-technical-report.pdf", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "Cohere internal eval", "prompt_style": "default", "temperature": "default" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Command A paper (command-r7b column): mmlu=65.2. Third-party Cohere reproduction, mc=false.", "candidates": [] }, { "model_id": "command-a", "benchmark_id": "math", "score": 80.0, "reference_url": "https://cohere.com/research/papers/command-a-technical-report.pdf", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "Cohere internal eval", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Command A paper (command-a column): math=80.0.", "candidates": [] }, { "model_id": "gpt-4o-0513", "benchmark_id": "math", "score": 70.2, "reference_url": "https://cohere.com/research/papers/command-a-technical-report.pdf", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "Cohere internal eval", "prompt_style": "default", "temperature": "default" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Command A paper (gpt-4o-0513 column): math=70.2. Third-party Cohere reproduction, mc=false.", "candidates": [] }, { "model_id": "deepseek-v3", "benchmark_id": "math", "score": 68.5, "reference_url": "https://cohere.com/research/papers/command-a-technical-report.pdf", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "Cohere internal eval", "prompt_style": "default", "temperature": "default" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Command A paper (deepseek-v3 column): math=68.5. Third-party Cohere reproduction, mc=false.", "candidates": [] }, { "model_id": "command-r7b", "benchmark_id": "math", "score": 59.1, "reference_url": "https://cohere.com/research/papers/command-a-technical-report.pdf", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "Cohere internal eval", "prompt_style": "default", "temperature": "default" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Command A paper (command-r7b column): math=59.1. Third-party Cohere reproduction, mc=false.", "candidates": [] }, { "model_id": "command-r7b", "benchmark_id": "ifeval", "score": 77.9, "reference_url": "https://cohere.com/research/papers/command-a-technical-report.pdf", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "Cohere internal eval", "prompt_style": "default", "temperature": "default" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Command A paper (command-r7b column): ifeval=77.9. Third-party Cohere reproduction, mc=false.", "candidates": [] }, { "model_id": "command-a", "benchmark_id": "gpqa_main", "score": 50.8, "reference_url": "https://cohere.com/research/papers/command-a-technical-report.pdf", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "Cohere internal eval", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Command A paper (command-a column): gpqa=50.8.", "candidates": [] }, { "model_id": "gpt-4o-0513", "benchmark_id": "gpqa_main", "score": 59.1, "reference_url": "https://cohere.com/research/papers/command-a-technical-report.pdf", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "Cohere internal eval", "prompt_style": "default", "temperature": "default" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Command A paper (gpt-4o-0513 column): gpqa=59.1. Third-party Cohere reproduction, mc=false.", "candidates": [ { "score": 53.6, "reference_url": "https://cohere.com/research/papers/command-a-technical-report.pdf", "source_type": "tech_report", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "Cohere internal eval", "prompt_style": "default", "temperature": "default" }, "notes": "Command A paper: alt measurement 53.6 (mc=false)." } ] }, { "model_id": "deepseek-v3", "benchmark_id": "gpqa_main", "score": 53.6, "reference_url": "https://cohere.com/research/papers/command-a-technical-report.pdf", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "Cohere internal eval", "prompt_style": "default", "temperature": "default" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Command A paper (deepseek-v3 column): gpqa=53.6. Third-party Cohere reproduction, mc=false.", "candidates": [ { "score": 59.1, "reference_url": "https://cohere.com/research/papers/command-a-technical-report.pdf", "source_type": "tech_report", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "Cohere internal eval", "prompt_style": "default", "temperature": "default" }, "notes": "Command A paper: alt measurement 59.1 (mc=false)." } ] }, { "model_id": "llama-3.3-70b-instruct", "benchmark_id": "gpqa_main", "score": 50.5, "reference_url": "https://cohere.com/research/papers/command-a-technical-report.pdf", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "Cohere internal eval", "prompt_style": "default", "temperature": "default" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Command A paper (llama-3.3-70b-instruct column): gpqa=50.5. Third-party Cohere reproduction, mc=false.", "candidates": [] }, { "model_id": "command-r7b", "benchmark_id": "gpqa_main", "score": 26.3, "reference_url": "https://cohere.com/research/papers/command-a-technical-report.pdf", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "Cohere internal eval", "prompt_style": "default", "temperature": "default" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Command A paper (command-r7b column): gpqa=26.3. Third-party Cohere reproduction, mc=false.", "candidates": [] }, { "model_id": "llama-3.1-8b", "benchmark_id": "gpqa_main", "score": 23.4, "reference_url": "https://cohere.com/research/papers/command-a-technical-report.pdf", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "Cohere internal eval", "prompt_style": "default", "temperature": "default" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Command A paper (llama-3.1-8b column): gpqa=23.4. Third-party Cohere reproduction, mc=false.", "candidates": [] }, { "model_id": "ministral-8b", "benchmark_id": "gpqa_main", "score": 23.4, "reference_url": "https://cohere.com/research/papers/command-a-technical-report.pdf", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "Cohere internal eval", "prompt_style": "default", "temperature": "default" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Command A paper (ministral-8b column): gpqa=23.4. Third-party Cohere reproduction, mc=false.", "candidates": [] }, { "model_id": "command-a", "benchmark_id": "mbpp_plus", "score": 86.2, "reference_url": "https://cohere.com/research/papers/command-a-technical-report.pdf", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "Cohere internal eval", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Command A paper (command-a column): mbpp_plus=86.2.", "candidates": [] }, { "model_id": "gpt-4o-0513", "benchmark_id": "mbpp_plus", "score": 89.9, "reference_url": "https://cohere.com/research/papers/command-a-technical-report.pdf", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "Cohere internal eval", "prompt_style": "default", "temperature": "default" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Command A paper (gpt-4o-0513 column): mbpp_plus=89.9. Third-party Cohere reproduction, mc=false.", "candidates": [] }, { "model_id": "deepseek-v3", "benchmark_id": "mbpp_plus", "score": 86.5, "reference_url": "https://cohere.com/research/papers/command-a-technical-report.pdf", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "Cohere internal eval", "prompt_style": "default", "temperature": "default" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Command A paper (deepseek-v3 column): mbpp_plus=86.5. Third-party Cohere reproduction, mc=false.", "candidates": [ { "score": 90.0, "reference_url": "https://cohere.com/research/papers/command-a-technical-report.pdf", "source_type": "tech_report", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "Cohere internal eval", "prompt_style": "default", "temperature": "default" }, "notes": "Command A paper: alt measurement 90.0 (mc=false)." } ] }, { "model_id": "llama-3.3-70b-instruct", "benchmark_id": "mbpp_plus", "score": 84.4, "reference_url": "https://cohere.com/research/papers/command-a-technical-report.pdf", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "Cohere internal eval", "prompt_style": "default", "temperature": "default" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Command A paper (llama-3.3-70b-instruct column): mbpp_plus=84.4. Third-party Cohere reproduction, mc=false.", "candidates": [ { "score": 86.0, "reference_url": "https://cohere.com/research/papers/command-a-technical-report.pdf", "source_type": "tech_report", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "Cohere internal eval", "prompt_style": "default", "temperature": "default" }, "notes": "Command A paper: alt measurement 86.0 (mc=false)." } ] }, { "model_id": "command-r7b", "benchmark_id": "mbpp_plus", "score": 72.0, "reference_url": "https://cohere.com/research/papers/command-a-technical-report.pdf", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "Cohere internal eval", "prompt_style": "default", "temperature": "default" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Command A paper (command-r7b column): mbpp_plus=72.0. Third-party Cohere reproduction, mc=false.", "candidates": [] }, { "model_id": "llama-3.1-8b", "benchmark_id": "mbpp_plus", "score": 72.8, "reference_url": "https://cohere.com/research/papers/command-a-technical-report.pdf", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "Cohere internal eval", "prompt_style": "default", "temperature": "default" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Command A paper (llama-3.1-8b column): mbpp_plus=72.8. Third-party Cohere reproduction, mc=false.", "candidates": [] }, { "model_id": "ministral-8b", "benchmark_id": "mbpp_plus", "score": 61.1, "reference_url": "https://cohere.com/research/papers/command-a-technical-report.pdf", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "Cohere internal eval", "prompt_style": "default", "temperature": "default" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Command A paper (ministral-8b column): mbpp_plus=61.1. Third-party Cohere reproduction, mc=false.", "candidates": [] }, { "model_id": "command-a", "benchmark_id": "bird_sql", "score": 59.5, "reference_url": "https://cohere.com/research/papers/command-a-technical-report.pdf", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "Cohere internal eval", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Command A paper (command-a column): bird_sql=59.5.", "candidates": [] }, { "model_id": "gpt-4o-0513", "benchmark_id": "bird_sql", "score": 53.1, "reference_url": "https://cohere.com/research/papers/command-a-technical-report.pdf", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "Cohere internal eval", "prompt_style": "default", "temperature": "default" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Command A paper (gpt-4o-0513 column): bird_sql=53.1. Third-party Cohere reproduction, mc=false.", "candidates": [] }, { "model_id": "deepseek-v3", "benchmark_id": "bird_sql", "score": 50.5, "reference_url": "https://cohere.com/research/papers/command-a-technical-report.pdf", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "Cohere internal eval", "prompt_style": "default", "temperature": "default" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Command A paper (deepseek-v3 column): bird_sql=50.5. Third-party Cohere reproduction, mc=false.", "candidates": [ { "score": 53.1, "reference_url": "https://cohere.com/research/papers/command-a-technical-report.pdf", "source_type": "tech_report", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "Cohere internal eval", "prompt_style": "default", "temperature": "default" }, "notes": "Command A paper: alt measurement 53.1 (mc=false)." } ] }, { "model_id": "llama-3.3-70b-instruct", "benchmark_id": "bird_sql", "score": 58.0, "reference_url": "https://cohere.com/research/papers/command-a-technical-report.pdf", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "Cohere internal eval", "prompt_style": "default", "temperature": "default" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Command A paper (llama-3.3-70b-instruct column): bird_sql=58.0. Third-party Cohere reproduction, mc=false.", "candidates": [] }, { "model_id": "command-r7b", "benchmark_id": "bird_sql", "score": 42.2, "reference_url": "https://cohere.com/research/papers/command-a-technical-report.pdf", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "Cohere internal eval", "prompt_style": "default", "temperature": "default" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Command A paper (command-r7b column): bird_sql=42.2. Third-party Cohere reproduction, mc=false.", "candidates": [] }, { "model_id": "llama-3.1-8b", "benchmark_id": "bird_sql", "score": 41.9, "reference_url": "https://cohere.com/research/papers/command-a-technical-report.pdf", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "Cohere internal eval", "prompt_style": "default", "temperature": "default" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Command A paper (llama-3.1-8b column): bird_sql=41.9. Third-party Cohere reproduction, mc=false.", "candidates": [] }, { "model_id": "ministral-8b", "benchmark_id": "bird_sql", "score": 33.2, "reference_url": "https://cohere.com/research/papers/command-a-technical-report.pdf", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "Cohere internal eval", "prompt_style": "default", "temperature": "default" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Command A paper (ministral-8b column): bird_sql=33.2. Third-party Cohere reproduction, mc=false.", "candidates": [] }, { "model_id": "command-a", "benchmark_id": "repoqa", "score": 92.6, "reference_url": "https://cohere.com/research/papers/command-a-technical-report.pdf", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "Cohere internal eval", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Command A paper (command-a column): repoqa=92.6.", "candidates": [] }, { "model_id": "gpt-4o-0513", "benchmark_id": "repoqa", "score": 92.2, "reference_url": "https://cohere.com/research/papers/command-a-technical-report.pdf", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "Cohere internal eval", "prompt_style": "default", "temperature": "default" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Command A paper (gpt-4o-0513 column): repoqa=92.2. Third-party Cohere reproduction, mc=false.", "candidates": [] }, { "model_id": "deepseek-v3", "benchmark_id": "repoqa", "score": 91.2, "reference_url": "https://cohere.com/research/papers/command-a-technical-report.pdf", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "Cohere internal eval", "prompt_style": "default", "temperature": "default" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Command A paper (deepseek-v3 column): repoqa=91.2. Third-party Cohere reproduction, mc=false.", "candidates": [ { "score": 92.2, "reference_url": "https://cohere.com/research/papers/command-a-technical-report.pdf", "source_type": "tech_report", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "Cohere internal eval", "prompt_style": "default", "temperature": "default" }, "notes": "Command A paper: alt measurement 92.2 (mc=false)." } ] }, { "model_id": "llama-3.3-70b-instruct", "benchmark_id": "repoqa", "score": 85.6, "reference_url": "https://cohere.com/research/papers/command-a-technical-report.pdf", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "Cohere internal eval", "prompt_style": "default", "temperature": "default" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Command A paper (llama-3.3-70b-instruct column): repoqa=85.6. Third-party Cohere reproduction, mc=false.", "candidates": [] }, { "model_id": "command-r7b", "benchmark_id": "repoqa", "score": 69.6, "reference_url": "https://cohere.com/research/papers/command-a-technical-report.pdf", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "Cohere internal eval", "prompt_style": "default", "temperature": "default" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Command A paper (command-r7b column): repoqa=69.6. Third-party Cohere reproduction, mc=false.", "candidates": [] }, { "model_id": "llama-3.1-8b", "benchmark_id": "repoqa", "score": 73.6, "reference_url": "https://cohere.com/research/papers/command-a-technical-report.pdf", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "Cohere internal eval", "prompt_style": "default", "temperature": "default" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Command A paper (llama-3.1-8b column): repoqa=73.6. Third-party Cohere reproduction, mc=false.", "candidates": [] }, { "model_id": "ministral-8b", "benchmark_id": "repoqa", "score": 62.0, "reference_url": "https://cohere.com/research/papers/command-a-technical-report.pdf", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "Cohere internal eval", "prompt_style": "default", "temperature": "default" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Command A paper (ministral-8b column): repoqa=62.0. Third-party Cohere reproduction, mc=false.", "candidates": [] }, { "model_id": "command-a", "benchmark_id": "ntrex", "score": 68.8, "reference_url": "https://cohere.com/research/papers/command-a-technical-report.pdf", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "Cohere internal eval", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Command A paper (command-a column): ntrex=68.8.", "candidates": [] }, { "model_id": "gpt-4o-0513", "benchmark_id": "ntrex", "score": 69.8, "reference_url": "https://cohere.com/research/papers/command-a-technical-report.pdf", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "Cohere internal eval", "prompt_style": "default", "temperature": "default" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Command A paper (gpt-4o-0513 column): ntrex=69.8. Third-party Cohere reproduction, mc=false.", "candidates": [] }, { "model_id": "deepseek-v3", "benchmark_id": "ntrex", "score": 71.0, "reference_url": "https://cohere.com/research/papers/command-a-technical-report.pdf", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "Cohere internal eval", "prompt_style": "default", "temperature": "default" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Command A paper (deepseek-v3 column): ntrex=71.0. Third-party Cohere reproduction, mc=false.", "candidates": [] }, { "model_id": "llama-3.3-70b-instruct", "benchmark_id": "ntrex", "score": 62.5, "reference_url": "https://cohere.com/research/papers/command-a-technical-report.pdf", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "Cohere internal eval", "prompt_style": "default", "temperature": "default" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Command A paper (llama-3.3-70b-instruct column): ntrex=62.5. Third-party Cohere reproduction, mc=false.", "candidates": [] }, { "model_id": "command-r7b", "benchmark_id": "ntrex", "score": 48.1, "reference_url": "https://cohere.com/research/papers/command-a-technical-report.pdf", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "Cohere internal eval", "prompt_style": "default", "temperature": "default" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Command A paper (command-r7b column): ntrex=48.1. Third-party Cohere reproduction, mc=false.", "candidates": [] }, { "model_id": "llama-3.1-8b", "benchmark_id": "ntrex", "score": 49.2, "reference_url": "https://cohere.com/research/papers/command-a-technical-report.pdf", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "Cohere internal eval", "prompt_style": "default", "temperature": "default" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Command A paper (llama-3.1-8b column): ntrex=49.2. Third-party Cohere reproduction, mc=false.", "candidates": [] }, { "model_id": "ministral-8b", "benchmark_id": "ntrex", "score": 36.8, "reference_url": "https://cohere.com/research/papers/command-a-technical-report.pdf", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "Cohere internal eval", "prompt_style": "default", "temperature": "default" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Command A paper (ministral-8b column): ntrex=36.8. Third-party Cohere reproduction, mc=false.", "candidates": [] }, { "model_id": "command-a", "benchmark_id": "bfcl", "score": 63.8, "reference_url": "https://cohere.com/research/papers/command-a-technical-report.pdf", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "Cohere internal eval", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Command A paper (command-a column): bfcl=63.8.", "candidates": [] }, { "model_id": "gpt-4o-0513", "benchmark_id": "bfcl", "score": 58.6, "reference_url": "https://cohere.com/research/papers/command-a-technical-report.pdf", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "Cohere internal eval", "prompt_style": "default", "temperature": "default" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Command A paper (gpt-4o-0513 column): bfcl=58.6. Third-party Cohere reproduction, mc=false.", "candidates": [] }, { "model_id": "deepseek-v3", "benchmark_id": "bfcl", "score": 72.1, "reference_url": "https://cohere.com/research/papers/command-a-technical-report.pdf", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "Cohere internal eval", "prompt_style": "default", "temperature": "default" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Command A paper (deepseek-v3 column): bfcl=72.1. Third-party Cohere reproduction, mc=false.", "candidates": [] }, { "model_id": "llama-3.3-70b-instruct", "benchmark_id": "bfcl", "score": 51.4, "reference_url": "https://cohere.com/research/papers/command-a-technical-report.pdf", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "Cohere internal eval", "prompt_style": "default", "temperature": "default" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Command A paper (llama-3.3-70b-instruct column): bfcl=51.4. Third-party Cohere reproduction, mc=false.", "candidates": [] }, { "model_id": "command-r7b", "benchmark_id": "bfcl", "score": 52.2, "reference_url": "https://cohere.com/research/papers/command-a-technical-report.pdf", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "Cohere internal eval", "prompt_style": "default", "temperature": "default" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Command A paper (command-r7b column): bfcl=52.2. Third-party Cohere reproduction, mc=false.", "candidates": [] }, { "model_id": "llama-3.1-8b", "benchmark_id": "bfcl", "score": 50.9, "reference_url": "https://cohere.com/research/papers/command-a-technical-report.pdf", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "Cohere internal eval", "prompt_style": "default", "temperature": "default" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Command A paper (llama-3.1-8b column): bfcl=50.9. Third-party Cohere reproduction, mc=false.", "candidates": [] }, { "model_id": "ministral-8b", "benchmark_id": "bfcl", "score": 51.8, "reference_url": "https://cohere.com/research/papers/command-a-technical-report.pdf", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "Cohere internal eval", "prompt_style": "default", "temperature": "default" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Command A paper (ministral-8b column): bfcl=51.8. Third-party Cohere reproduction, mc=false.", "candidates": [] }, { "model_id": "command-a", "benchmark_id": "tau1_bench_avg", "score": 51.7, "reference_url": "https://cohere.com/research/papers/command-a-technical-report.pdf", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "Cohere internal eval", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Command A paper (command-a column): taubench_avg=51.7.", "candidates": [] }, { "model_id": "gpt-4o-0513", "benchmark_id": "tau1_bench_avg", "score": 39.1, "reference_url": "https://cohere.com/research/papers/command-a-technical-report.pdf", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "Cohere internal eval", "prompt_style": "default", "temperature": "default" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Command A paper (gpt-4o-0513 column): taubench_avg=39.1. Third-party Cohere reproduction, mc=false.", "candidates": [] }, { "model_id": "deepseek-v3", "benchmark_id": "tau1_bench_avg", "score": 51.2, "reference_url": "https://cohere.com/research/papers/command-a-technical-report.pdf", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "Cohere internal eval", "prompt_style": "default", "temperature": "default" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Command A paper (deepseek-v3 column): taubench_avg=51.2. Third-party Cohere reproduction, mc=false.", "candidates": [] }, { "model_id": "llama-3.3-70b-instruct", "benchmark_id": "tau1_bench_avg", "score": 21.0, "reference_url": "https://cohere.com/research/papers/command-a-technical-report.pdf", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "Cohere internal eval", "prompt_style": "default", "temperature": "default" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Command A paper (llama-3.3-70b-instruct column): taubench_avg=21.0. Third-party Cohere reproduction, mc=false.", "candidates": [] }, { "model_id": "mistral-large-2", "benchmark_id": "mmlu", "score": 85.2, "reference_url": "https://cohere.com/research/papers/command-a-technical-report.pdf", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "Cohere internal eval", "prompt_style": "default", "temperature": "default" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Command A paper (mistral-large-2 column): mmlu=85.2. Third-party Cohere reproduction, mc=false.", "candidates": [] }, { "model_id": "gemini-2.0-pro", "benchmark_id": "mmlu", "score": 89.3, "reference_url": "https://cohere.com/research/papers/command-a-technical-report.pdf", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "Cohere internal eval", "prompt_style": "default", "temperature": "default" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Command A paper (gemini-2.0-pro column): mmlu=89.3. Third-party Cohere reproduction, mc=false.", "candidates": [] }, { "model_id": "gemini-1.5-flash-8b", "benchmark_id": "mmlu", "score": 74.8, "reference_url": "https://cohere.com/research/papers/command-a-technical-report.pdf", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "Cohere internal eval", "prompt_style": "default", "temperature": "default" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Command A paper (gemini-1.5-flash-8b column): mmlu=74.8. Third-party Cohere reproduction, mc=false.", "candidates": [] }, { "model_id": "mistral-large-2", "benchmark_id": "mmlu_pro", "score": 67.9, "reference_url": "https://cohere.com/research/papers/command-a-technical-report.pdf", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "Cohere internal eval", "prompt_style": "default", "temperature": "default" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Command A paper (mistral-large-2 column): mmlu_pro=67.9. Third-party Cohere reproduction, mc=false.", "candidates": [] }, { "model_id": "command-r7b", "benchmark_id": "mmlu_pro", "score": 42.4, "reference_url": "https://cohere.com/research/papers/command-a-technical-report.pdf", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "Cohere internal eval", "prompt_style": "default", "temperature": "default" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Command A paper (command-r7b column): mmlu_pro=42.4. Third-party Cohere reproduction, mc=false.", "candidates": [] }, { "model_id": "llama-3.1-8b", "benchmark_id": "mmlu_pro", "score": 46.5, "reference_url": "https://cohere.com/research/papers/command-a-technical-report.pdf", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "Cohere internal eval", "prompt_style": "default", "temperature": "default" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Command A paper (llama-3.1-8b column): mmlu_pro=46.5. Third-party Cohere reproduction, mc=false.", "candidates": [] }, { "model_id": "ministral-8b", "benchmark_id": "mmlu_pro", "score": 43.0, "reference_url": "https://cohere.com/research/papers/command-a-technical-report.pdf", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "Cohere internal eval", "prompt_style": "default", "temperature": "default" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Command A paper (ministral-8b column): mmlu_pro=43.0. Third-party Cohere reproduction, mc=false.", "candidates": [] }, { "model_id": "gemini-1.5-flash-8b", "benchmark_id": "mmlu_pro", "score": 48.4, "reference_url": "https://cohere.com/research/papers/command-a-technical-report.pdf", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "Cohere internal eval", "prompt_style": "default", "temperature": "default" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Command A paper (gemini-1.5-flash-8b column): mmlu_pro=48.4. Third-party Cohere reproduction, mc=false.", "candidates": [] }, { "model_id": "llama-3.1-405b-instruct", "benchmark_id": "gpqa_main", "score": 49.0, "reference_url": "https://cohere.com/research/papers/command-a-technical-report.pdf", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "Cohere internal eval", "prompt_style": "default", "temperature": "default" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Command A paper (llama-3.1-405b-instruct column): gpqa=49.0. Third-party Cohere reproduction, mc=false.", "candidates": [] }, { "model_id": "mistral-large-2", "benchmark_id": "gpqa_main", "score": 48.6, "reference_url": "https://cohere.com/research/papers/command-a-technical-report.pdf", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "Cohere internal eval", "prompt_style": "default", "temperature": "default" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Command A paper (mistral-large-2 column): gpqa=48.6. Third-party Cohere reproduction, mc=false.", "candidates": [] }, { "model_id": "claude-3.5-sonnet", "benchmark_id": "gpqa_main", "score": 65.0, "reference_url": "https://cohere.com/research/papers/command-a-technical-report.pdf", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "Cohere internal eval", "prompt_style": "default", "temperature": "default" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Command A paper (claude-3.5-sonnet column): gpqa=65.0. Third-party Cohere reproduction, mc=false.", "candidates": [] }, { "model_id": "gemini-2.0-pro", "benchmark_id": "gpqa_main", "score": 64.7, "reference_url": "https://cohere.com/research/papers/command-a-technical-report.pdf", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "Cohere internal eval", "prompt_style": "default", "temperature": "default" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Command A paper (gemini-2.0-pro column): gpqa=64.7. Third-party Cohere reproduction, mc=false.", "candidates": [] }, { "model_id": "gemma-2-9b", "benchmark_id": "gpqa_main", "score": 31.3, "reference_url": "https://cohere.com/research/papers/command-a-technical-report.pdf", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "Cohere internal eval", "prompt_style": "default", "temperature": "default" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Command A paper (gemma-2-9b column): gpqa=31.3. Third-party Cohere reproduction, mc=false.", "candidates": [] }, { "model_id": "gemini-1.5-flash-8b", "benchmark_id": "gpqa_main", "score": 31.6, "reference_url": "https://cohere.com/research/papers/command-a-technical-report.pdf", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "Cohere internal eval", "prompt_style": "default", "temperature": "default" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Command A paper (gemini-1.5-flash-8b column): gpqa=31.6. Third-party Cohere reproduction, mc=false.", "candidates": [] }, { "model_id": "mistral-large-2", "benchmark_id": "ifeval", "score": 83.8, "reference_url": "https://cohere.com/research/papers/command-a-technical-report.pdf", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "Cohere internal eval", "prompt_style": "default", "temperature": "default" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Command A paper (mistral-large-2 column): ifeval=83.8. Third-party Cohere reproduction, mc=false.", "candidates": [] }, { "model_id": "gemini-2.0-pro", "benchmark_id": "ifeval", "score": 87.3, "reference_url": "https://cohere.com/research/papers/command-a-technical-report.pdf", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "Cohere internal eval", "prompt_style": "default", "temperature": "default" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Command A paper (gemini-2.0-pro column): ifeval=87.3. Third-party Cohere reproduction, mc=false.", "candidates": [] }, { "model_id": "gemini-1.5-flash-8b", "benchmark_id": "ifeval", "score": 88.0, "reference_url": "https://cohere.com/research/papers/command-a-technical-report.pdf", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "Cohere internal eval", "prompt_style": "default", "temperature": "default" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Command A paper (gemini-1.5-flash-8b column): ifeval=88.0. Third-party Cohere reproduction, mc=false.", "candidates": [] }, { "model_id": "command-a", "benchmark_id": "infobench", "score": 94.9, "reference_url": "https://cohere.com/research/papers/command-a-technical-report.pdf", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "Cohere internal eval", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Command A paper (command-a column): infobench=94.9.", "candidates": [] }, { "model_id": "gpt-4o-0513", "benchmark_id": "infobench", "score": 94.0, "reference_url": "https://cohere.com/research/papers/command-a-technical-report.pdf", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "Cohere internal eval", "prompt_style": "default", "temperature": "default" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Command A paper (gpt-4o-0513 column): infobench=94.0. Third-party Cohere reproduction, mc=false.", "candidates": [] }, { "model_id": "deepseek-v3", "benchmark_id": "infobench", "score": 94.3, "reference_url": "https://cohere.com/research/papers/command-a-technical-report.pdf", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "Cohere internal eval", "prompt_style": "default", "temperature": "default" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Command A paper (deepseek-v3 column): infobench=94.3. Third-party Cohere reproduction, mc=false.", "candidates": [] }, { "model_id": "llama-3.3-70b-instruct", "benchmark_id": "infobench", "score": 92.8, "reference_url": "https://cohere.com/research/papers/command-a-technical-report.pdf", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "Cohere internal eval", "prompt_style": "default", "temperature": "default" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Command A paper (llama-3.3-70b-instruct column): infobench=92.8. Third-party Cohere reproduction, mc=false.", "candidates": [] }, { "model_id": "llama-3.1-405b-instruct", "benchmark_id": "infobench", "score": 93.9, "reference_url": "https://cohere.com/research/papers/command-a-technical-report.pdf", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "Cohere internal eval", "prompt_style": "default", "temperature": "default" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Command A paper (llama-3.1-405b-instruct column): infobench=93.9. Third-party Cohere reproduction, mc=false.", "candidates": [] }, { "model_id": "mistral-large-2", "benchmark_id": "infobench", "score": 93.3, "reference_url": "https://cohere.com/research/papers/command-a-technical-report.pdf", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "Cohere internal eval", "prompt_style": "default", "temperature": "default" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Command A paper (mistral-large-2 column): infobench=93.3. Third-party Cohere reproduction, mc=false.", "candidates": [] }, { "model_id": "claude-3.5-sonnet", "benchmark_id": "infobench", "score": 93.9, "reference_url": "https://cohere.com/research/papers/command-a-technical-report.pdf", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "Cohere internal eval", "prompt_style": "default", "temperature": "default" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Command A paper (claude-3.5-sonnet column): infobench=93.9. Third-party Cohere reproduction, mc=false.", "candidates": [] }, { "model_id": "gemini-2.0-pro", "benchmark_id": "infobench", "score": 92.2, "reference_url": "https://cohere.com/research/papers/command-a-technical-report.pdf", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "Cohere internal eval", "prompt_style": "default", "temperature": "default" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Command A paper (gemini-2.0-pro column): infobench=92.2. Third-party Cohere reproduction, mc=false.", "candidates": [] }, { "model_id": "command-r7b", "benchmark_id": "infobench", "score": 85.6, "reference_url": "https://cohere.com/research/papers/command-a-technical-report.pdf", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "Cohere internal eval", "prompt_style": "default", "temperature": "default" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Command A paper (command-r7b column): infobench=85.6. Third-party Cohere reproduction, mc=false.", "candidates": [] }, { "model_id": "llama-3.1-8b", "benchmark_id": "infobench", "score": 90.1, "reference_url": "https://cohere.com/research/papers/command-a-technical-report.pdf", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "Cohere internal eval", "prompt_style": "default", "temperature": "default" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Command A paper (llama-3.1-8b column): infobench=90.1. Third-party Cohere reproduction, mc=false.", "candidates": [] }, { "model_id": "ministral-8b", "benchmark_id": "infobench", "score": 88.3, "reference_url": "https://cohere.com/research/papers/command-a-technical-report.pdf", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "Cohere internal eval", "prompt_style": "default", "temperature": "default" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Command A paper (ministral-8b column): infobench=88.3. Third-party Cohere reproduction, mc=false.", "candidates": [] }, { "model_id": "gemma-2-9b", "benchmark_id": "infobench", "score": 87.2, "reference_url": "https://cohere.com/research/papers/command-a-technical-report.pdf", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "Cohere internal eval", "prompt_style": "default", "temperature": "default" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Command A paper (gemma-2-9b column): infobench=87.2. Third-party Cohere reproduction, mc=false.", "candidates": [] }, { "model_id": "gemini-1.5-flash-8b", "benchmark_id": "infobench", "score": 88.3, "reference_url": "https://cohere.com/research/papers/command-a-technical-report.pdf", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "Cohere internal eval", "prompt_style": "default", "temperature": "default" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Command A paper (gemini-1.5-flash-8b column): infobench=88.3. Third-party Cohere reproduction, mc=false.", "candidates": [] }, { "model_id": "mistral-large-2", "benchmark_id": "mbpp_plus", "score": 84.7, "reference_url": "https://cohere.com/research/papers/command-a-technical-report.pdf", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "Cohere internal eval", "prompt_style": "default", "temperature": "default" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Command A paper (mistral-large-2 column): mbpp_plus=84.7. Third-party Cohere reproduction, mc=false.", "candidates": [] }, { "model_id": "qwen2.5-72b-instruct", "benchmark_id": "mbpp_plus", "score": 88.6, "reference_url": "https://cohere.com/research/papers/command-a-technical-report.pdf", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "Cohere internal eval", "prompt_style": "default", "temperature": "default" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Command A paper (qwen2.5-72b-instruct column): mbpp_plus=88.6. Third-party Cohere reproduction, mc=false.", "candidates": [] }, { "model_id": "llama-3.1-405b-instruct", "benchmark_id": "mbpp_plus", "score": 88.6, "reference_url": "https://cohere.com/research/papers/command-a-technical-report.pdf", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "Cohere internal eval", "prompt_style": "default", "temperature": "default" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Command A paper (llama-3.1-405b-instruct column): mbpp_plus=88.6. Third-party Cohere reproduction, mc=false.", "candidates": [] }, { "model_id": "command-r7b", "benchmark_id": "livecodebench", "score": 9.0, "reference_url": "https://cohere.com/research/papers/command-a-technical-report.pdf", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "Cohere internal eval", "prompt_style": "default", "temperature": "default" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Command A paper (command-r7b column): livecodebench=9.0. Third-party Cohere reproduction, mc=false.", "candidates": [] }, { "model_id": "llama-3.3-70b-instruct", "benchmark_id": "livecodebench", "score": 32.9, "reference_url": "https://cohere.com/research/papers/command-a-technical-report.pdf", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "Cohere internal eval", "prompt_style": "default", "temperature": "default" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Command A paper (llama-3.3-70b-instruct column): livecodebench=32.9. Third-party Cohere reproduction, mc=false.", "candidates": [] }, { "model_id": "mistral-large-2", "benchmark_id": "livecodebench", "score": 26.7, "reference_url": "https://cohere.com/research/papers/command-a-technical-report.pdf", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "Cohere internal eval", "prompt_style": "default", "temperature": "default" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Command A paper (mistral-large-2 column): livecodebench=26.7. Third-party Cohere reproduction, mc=false.", "candidates": [] }, { "model_id": "command-r7b", "benchmark_id": "bigcodebench", "score": 30.9, "reference_url": "https://cohere.com/research/papers/command-a-technical-report.pdf", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "Cohere internal eval", "prompt_style": "default", "temperature": "default" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Command A paper (command-r7b column): bigcodebench=30.9. Third-party Cohere reproduction, mc=false.", "candidates": [] }, { "model_id": "llama-3.3-70b-instruct", "benchmark_id": "bigcodebench", "score": 46.9, "reference_url": "https://cohere.com/research/papers/command-a-technical-report.pdf", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "Cohere internal eval", "prompt_style": "default", "temperature": "default" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Command A paper (llama-3.3-70b-instruct column): bigcodebench=46.9. Third-party Cohere reproduction, mc=false.", "candidates": [] }, { "model_id": "mistral-large-2", "benchmark_id": "bigcodebench", "score": 44.7, "reference_url": "https://cohere.com/research/papers/command-a-technical-report.pdf", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "Cohere internal eval", "prompt_style": "default", "temperature": "default" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Command A paper (mistral-large-2 column): bigcodebench=44.7. Third-party Cohere reproduction, mc=false.", "candidates": [] }, { "model_id": "qwen2.5-72b-instruct", "benchmark_id": "bigcodebench", "score": 45.8, "reference_url": "https://cohere.com/research/papers/command-a-technical-report.pdf", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "Cohere internal eval", "prompt_style": "default", "temperature": "default" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Command A paper (qwen2.5-72b-instruct column): bigcodebench=45.8. Third-party Cohere reproduction, mc=false.", "candidates": [] }, { "model_id": "llama-3.1-405b-instruct", "benchmark_id": "bigcodebench", "score": 46.2, "reference_url": "https://cohere.com/research/papers/command-a-technical-report.pdf", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "Cohere internal eval", "prompt_style": "default", "temperature": "default" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Command A paper (llama-3.1-405b-instruct column): bigcodebench=46.2. Third-party Cohere reproduction, mc=false.", "candidates": [] }, { "model_id": "command-r7b", "benchmark_id": "humaneval", "score": 50.7, "reference_url": "https://cohere.com/research/papers/command-a-technical-report.pdf", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "Cohere internal eval", "prompt_style": "default", "temperature": "default" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Command A paper (command-r7b column): humaneval=50.7. Third-party Cohere reproduction, mc=false.", "candidates": [] }, { "model_id": "mistral-large-2", "benchmark_id": "humaneval", "score": 82.9, "reference_url": "https://cohere.com/research/papers/command-a-technical-report.pdf", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "Cohere internal eval", "prompt_style": "default", "temperature": "default" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Command A paper (mistral-large-2 column): humaneval=82.9. Third-party Cohere reproduction, mc=false.", "candidates": [] }, { "model_id": "mistral-large-2", "benchmark_id": "repoqa", "score": 88.0, "reference_url": "https://cohere.com/research/papers/command-a-technical-report.pdf", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "Cohere internal eval", "prompt_style": "default", "temperature": "default" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Command A paper (mistral-large-2 column): repoqa=88.0. Third-party Cohere reproduction, mc=false.", "candidates": [] }, { "model_id": "qwen2.5-72b-instruct", "benchmark_id": "repoqa", "score": 83.2, "reference_url": "https://cohere.com/research/papers/command-a-technical-report.pdf", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "Cohere internal eval", "prompt_style": "default", "temperature": "default" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Command A paper (qwen2.5-72b-instruct column): repoqa=83.2. Third-party Cohere reproduction, mc=false.", "candidates": [] }, { "model_id": "llama-3.1-405b-instruct", "benchmark_id": "repoqa", "score": 90.4, "reference_url": "https://cohere.com/research/papers/command-a-technical-report.pdf", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "Cohere internal eval", "prompt_style": "default", "temperature": "default" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Command A paper (llama-3.1-405b-instruct column): repoqa=90.4. Third-party Cohere reproduction, mc=false.", "candidates": [] }, { "model_id": "command-r7b", "benchmark_id": "swe_bench_verified", "score": 3.6, "reference_url": "https://cohere.com/research/papers/command-a-technical-report.pdf", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "Cohere internal eval", "prompt_style": "default", "temperature": "default" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Command A paper (command-r7b column): swe_bench_verified=3.6. Third-party Cohere reproduction, mc=false.", "candidates": [] }, { "model_id": "llama-3.3-70b-instruct", "benchmark_id": "swe_bench_verified", "score": 29.4, "reference_url": "https://cohere.com/research/papers/command-a-technical-report.pdf", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "Cohere internal eval", "prompt_style": "default", "temperature": "default" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Command A paper (llama-3.3-70b-instruct column): swe_bench_verified=29.4. Third-party Cohere reproduction, mc=false.", "candidates": [] }, { "model_id": "mistral-large-2", "benchmark_id": "swe_bench_verified", "score": 30.0, "reference_url": "https://cohere.com/research/papers/command-a-technical-report.pdf", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "Cohere internal eval", "prompt_style": "default", "temperature": "default" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Command A paper (mistral-large-2 column): swe_bench_verified=30.0. Third-party Cohere reproduction, mc=false.", "candidates": [] }, { "model_id": "command-a", "benchmark_id": "aider_polyglot", "score": 14.7, "reference_url": "https://cohere.com/research/papers/command-a-technical-report.pdf", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "Cohere internal eval", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Command A paper (command-a column): aider_polyglot=14.7.", "candidates": [] }, { "model_id": "command-r7b", "benchmark_id": "aider_polyglot", "score": 2.7, "reference_url": "https://cohere.com/research/papers/command-a-technical-report.pdf", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "Cohere internal eval", "prompt_style": "default", "temperature": "default" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Command A paper (command-r7b column): aider_polyglot=2.7. Third-party Cohere reproduction, mc=false.", "candidates": [] }, { "model_id": "llama-3.3-70b-instruct", "benchmark_id": "aider_polyglot", "score": 8.4, "reference_url": "https://cohere.com/research/papers/command-a-technical-report.pdf", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "Cohere internal eval", "prompt_style": "default", "temperature": "default" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Command A paper (llama-3.3-70b-instruct column): aider_polyglot=8.4. Third-party Cohere reproduction, mc=false.", "candidates": [] }, { "model_id": "mistral-large-2", "benchmark_id": "aider_polyglot", "score": 16.0, "reference_url": "https://cohere.com/research/papers/command-a-technical-report.pdf", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "Cohere internal eval", "prompt_style": "default", "temperature": "default" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Command A paper (mistral-large-2 column): aider_polyglot=16.0. Third-party Cohere reproduction, mc=false.", "candidates": [] }, { "model_id": "qwen2.5-72b-instruct", "benchmark_id": "aider_polyglot", "score": 8.0, "reference_url": "https://cohere.com/research/papers/command-a-technical-report.pdf", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "Cohere internal eval", "prompt_style": "default", "temperature": "default" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Command A paper (qwen2.5-72b-instruct column): aider_polyglot=8.0. Third-party Cohere reproduction, mc=false.", "candidates": [] }, { "model_id": "llama-3.1-405b-instruct", "benchmark_id": "aider_polyglot", "score": 13.8, "reference_url": "https://cohere.com/research/papers/command-a-technical-report.pdf", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "Cohere internal eval", "prompt_style": "default", "temperature": "default" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Command A paper (llama-3.1-405b-instruct column): aider_polyglot=13.8. Third-party Cohere reproduction, mc=false.", "candidates": [] }, { "model_id": "deepseek-v3", "benchmark_id": "aider_polyglot", "score": 49.6, "reference_url": "https://cohere.com/research/papers/command-a-technical-report.pdf", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "Cohere internal eval", "prompt_style": "default", "temperature": "default" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Command A paper (deepseek-v3 column): aider_polyglot=49.6. Third-party Cohere reproduction, mc=false.", "candidates": [] }, { "model_id": "mistral-large-2", "benchmark_id": "bird_sql", "score": 50.0, "reference_url": "https://cohere.com/research/papers/command-a-technical-report.pdf", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "Cohere internal eval", "prompt_style": "default", "temperature": "default" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Command A paper (mistral-large-2 column): bird_sql=50.0. Third-party Cohere reproduction, mc=false.", "candidates": [] }, { "model_id": "qwen2.5-72b-instruct", "benchmark_id": "bird_sql", "score": 50.1, "reference_url": "https://cohere.com/research/papers/command-a-technical-report.pdf", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "Cohere internal eval", "prompt_style": "default", "temperature": "default" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Command A paper (qwen2.5-72b-instruct column): bird_sql=50.1. Third-party Cohere reproduction, mc=false.", "candidates": [] }, { "model_id": "llama-3.1-405b-instruct", "benchmark_id": "bird_sql", "score": 59.4, "reference_url": "https://cohere.com/research/papers/command-a-technical-report.pdf", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "Cohere internal eval", "prompt_style": "default", "temperature": "default" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Command A paper (llama-3.1-405b-instruct column): bird_sql=59.4. Third-party Cohere reproduction, mc=false.", "candidates": [] }, { "model_id": "internlm3-8b", "benchmark_id": "cmmlu", "score": 83.1, "reference_url": "https://github.com/InternLM/InternLM", "source_type": "model_card", "audit_status": "verified", "matches_canonical": true, "reported_setting": { "mode": "non-thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "OpenCompass", "prompt_style": "0-shot" }, "notes": "CMMLU 0-shot = 83.1. No * (non-thinking). GitHub README table." }, { "model_id": "internlm3-8b", "benchmark_id": "drop", "score": 83.1, "reference_url": "https://github.com/InternLM/InternLM", "source_type": "model_card", "audit_status": "verified", "matches_canonical": true, "reported_setting": { "mode": "non-thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "OpenCompass", "prompt_style": "0-shot" }, "notes": "DROP 0-shot = 83.1. No * (non-thinking). GitHub README table." }, { "model_id": "internlm3-8b", "benchmark_id": "hellaswag", "score": 91.2, "reference_url": "https://github.com/InternLM/InternLM", "source_type": "model_card", "audit_status": "verified", "matches_canonical": true, "reported_setting": { "mode": "non-thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "OpenCompass", "prompt_style": "0-shot" }, "notes": "HellaSwag 10-shot = 91.2. No * (non-thinking). GitHub README table." }, { "model_id": "internlm3-8b", "benchmark_id": "korbench", "score": 56.4, "reference_url": "https://github.com/InternLM/InternLM", "source_type": "model_card", "audit_status": "verified", "matches_canonical": true, "reported_setting": { "mode": "non-thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "OpenCompass", "prompt_style": "0-shot" }, "notes": "KOR-Bench 0-shot = 56.4. No * (non-thinking). GitHub README table. Benchmark labeled KOR-Bench in source, maps to korbench in BP." }, { "model_id": "internlm3-8b", "benchmark_id": "alpacaeval_2", "score": 51.1, "reference_url": "https://github.com/InternLM/InternLM", "source_type": "model_card", "audit_status": "verified", "matches_canonical": true, "reported_setting": { "mode": "non-thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "OpenCompass", "prompt_style": "0-shot" }, "notes": "AlpacaEval 2.0 LC WinRate = 51.1. No * (non-thinking). GitHub README table." }, { "model_id": "internlm3-8b", "benchmark_id": "wildbench", "score": 33.1, "reference_url": "https://github.com/InternLM/InternLM", "source_type": "model_card", "audit_status": "verified", "matches_canonical": true, "reported_setting": { "mode": "non-thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "LLM-as-judge", "harness": "OpenCompass", "prompt_style": "0-shot" }, "notes": "WildBench Raw Score = 33.1. No * (non-thinking). GitHub README." }, { "model_id": "internlm3-8b", "benchmark_id": "mt_bench_101", "score": 8.59, "reference_url": "https://github.com/InternLM/InternLM", "source_type": "model_card", "audit_status": "verified", "matches_canonical": true, "reported_setting": { "mode": "non-thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "LLM-as-judge", "harness": "OpenCompass", "prompt_style": "0-shot" }, "notes": "MT-Bench-101 Score 1-10 = 8.59. No * (non-thinking). GitHub README." }, { "model_id": "exaone-4.0-32b", "benchmark_id": "mmlu_redux", "score": 92.3, "reference_url": "https://arxiv.org/html/2507.11407v1", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "LG AI Research internal", "prompt_style": "default", "temperature": "0.6, top_p=0.95, presence_penalty=1.5" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Reasoning mode Table 3: MMLU-Redux=92.3. World Knowledge benchmark.", "candidates": [] }, { "model_id": "exaone-4.0-32b", "benchmark_id": "livecodebench_v5", "score": 72.6, "reference_url": "https://arxiv.org/html/2507.11407v1", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "avg@4", "judge": "rule-based", "harness": "LG AI Research internal", "prompt_style": "default", "temperature": "0.6, top_p=0.95, presence_penalty=1.5" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Reasoning mode Table 3: LiveCodeBench v5=72.6. (moved from generic livecodebench cell)", "candidates": [] }, { "model_id": "exaone-4.0-32b", "benchmark_id": "livecodebench_v6", "score": 66.7, "reference_url": "https://arxiv.org/html/2507.11407v1", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "avg@4", "judge": "rule-based", "harness": "LG AI Research internal", "prompt_style": "default", "temperature": "0.6, top_p=0.95, presence_penalty=1.5" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Reasoning mode Table 3: LiveCodeBench v6=66.7. Also Table 7 (64K budget confirms 66.7).", "candidates": [] }, { "model_id": "exaone-4.0-32b", "benchmark_id": "multi_if", "score": 73.5, "reference_url": "https://arxiv.org/html/2507.11407v1", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "LG AI Research internal", "prompt_style": "default", "temperature": "0.6, top_p=0.95, presence_penalty=1.5" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Reasoning mode Table 3: Multi-IF (EN)=73.5. Instruction following.", "candidates": [] }, { "model_id": "exaone-4.0-32b", "benchmark_id": "bfcl_v3", "score": 63.9, "reference_url": "https://arxiv.org/html/2507.11407v1", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "LG AI Research internal", "prompt_style": "default", "temperature": "0.6, top_p=0.95, presence_penalty=1.5" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Reasoning mode Table 3: BFCL-v3=63.9. Agentic Tool Use.", "candidates": [] }, { "model_id": "exaone-4.0-32b", "benchmark_id": "tau_bench_airline", "score": 51.5, "reference_url": "https://arxiv.org/html/2507.11407v1", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "avg@4", "judge": "rule-based", "harness": "LG AI Research internal", "prompt_style": "default", "temperature": "0.6, top_p=0.95, presence_penalty=1.5" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Reasoning mode Table 3: Tau-Bench (Airline)=51.5. Agentic Tool Use.", "candidates": [] }, { "model_id": "exaone-4.0-32b", "benchmark_id": "tau_bench_retail", "score": 62.8, "reference_url": "https://arxiv.org/html/2507.11407v1", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "avg@4", "judge": "rule-based", "harness": "LG AI Research internal", "prompt_style": "default", "temperature": "0.6, top_p=0.95, presence_penalty=1.5" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Reasoning mode Table 3: Tau-Bench (Retail)=62.8. Agentic Tool Use.", "candidates": [] }, { "model_id": "claude-3.5-sonnet", "benchmark_id": "tau_bench_retail", "score": 71.5, "reference_url": "https://www.anthropic.com/news/claude-3-7-sonnet", "reported_setting": { "mode": "non-thinking", "effort": "default", "tools": "agentic (tau-bench official)", "sampling": "pass@1", "judge": "official harness", "harness": "official", "prompt_style": "default + planning tool addendum (Airline)", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "Anthropic 3.7 Sonnet blog cross-model comparison table: Claude 3.5 Sonnet (new) TAU-bench Retail=71.5%" }, { "model_id": "claude-3.5-sonnet", "benchmark_id": "tau_bench_airline", "score": 48.8, "reference_url": "https://www.anthropic.com/news/claude-3-7-sonnet", "reported_setting": { "mode": "non-thinking", "effort": "default", "tools": "agentic (tau-bench official)", "sampling": "pass@1", "judge": "official harness", "harness": "official", "prompt_style": "default + planning tool addendum (Airline)", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "Anthropic 3.7 Sonnet blog cross-model comparison table: Claude 3.5 Sonnet (new) TAU-bench Airline=48.8%" }, { "model_id": "claude-3.5-sonnet", "benchmark_id": "mmmlu", "score": 82.1, "reference_url": "https://www.anthropic.com/news/claude-3-7-sonnet", "reported_setting": { "mode": "non-thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "Anthropic 3.7 Sonnet blog cross-model: Claude 3.5 Sonnet (new) MMMLU=82.1%" }, { "model_id": "claude-3.5-sonnet", "benchmark_id": "mmmu", "score": 70.4, "reference_url": "https://www.anthropic.com/news/claude-3-7-sonnet", "reported_setting": { "mode": "non-thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "zeroshot-cot", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "Anthropic 3.7 Sonnet blog cross-model: Claude 3.5 Sonnet (new) MMMU=70.4%" }, { "model_id": "kimi-k2.5", "benchmark_id": "hle_tools", "score": 50.2, "reference_url": "https://huggingface.co/moonshotai/Kimi-K2.5", "reported_setting": "thinking mode, tool-augmented, pass@1", "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "First-party, directly from Kimi K2.5 HF model card evaluation table." }, { "model_id": "gpt-5.2", "benchmark_id": "hle_tools", "score": 45.5, "reference_url": "https://huggingface.co/moonshotai/Kimi-K2.5", "reported_setting": "with tools, pass@1", "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Reported in Kimi K2.5 HF model card (no asterisk = Moonshot used publicly available official score). Reference is Kimi card, not GPT primary source." }, { "model_id": "claude-opus-4.5", "benchmark_id": "hle_tools", "score": 43.2, "reference_url": "https://huggingface.co/moonshotai/Kimi-K2.5", "reported_setting": "with tools, pass@1", "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Reported in Kimi K2.5 HF model card (no asterisk = publicly available official score). Reference is Kimi card." }, { "model_id": "gemini-3-pro", "benchmark_id": "hle_tools", "score": 45.8, "reference_url": "https://huggingface.co/moonshotai/Kimi-K2.5", "reported_setting": "with tools, pass@1", "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Reported in Kimi K2.5 HF model card (no asterisk = publicly available official score). Reference is Kimi card." }, { "model_id": "deepseek-v3.2", "benchmark_id": "hle_tools", "score": 40.8, "reference_url": "https://huggingface.co/moonshotai/Kimi-K2.5", "reported_setting": "text-only subset, with tools, pass@1", "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "†: DeepSeek V3.2 HLE (w/tools) score corresponds to text-only subset (vision not supported). Reported in Kimi K2.5 HF model card." }, { "model_id": "kimi-k2.5", "benchmark_id": "zerobench_tools", "score": 11.0, "reference_url": "https://huggingface.co/moonshotai/Kimi-K2.5", "reported_setting": "thinking mode, tool-augmented, pass@1", "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "First-party, directly from Kimi K2.5 HF model card." }, { "model_id": "gpt-5.2", "benchmark_id": "zerobench_tools", "score": 7.0, "reference_url": "https://huggingface.co/moonshotai/Kimi-K2.5", "reported_setting": "tool-augmented, pass@1; re-evaluated by Moonshot", "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "*: Score re-evaluated by Moonshot AI under Kimi K2.5 evaluation conditions (no publicly available score)." }, { "model_id": "claude-opus-4.5", "benchmark_id": "zerobench_tools", "score": 9.0, "reference_url": "https://huggingface.co/moonshotai/Kimi-K2.5", "reported_setting": "tool-augmented, pass@1; re-evaluated by Moonshot", "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "*: Score re-evaluated by Moonshot AI under Kimi K2.5 evaluation conditions." }, { "model_id": "gemini-3-pro", "benchmark_id": "zerobench_tools", "score": 12.0, "reference_url": "https://huggingface.co/moonshotai/Kimi-K2.5", "reported_setting": "tool-augmented, pass@1; re-evaluated by Moonshot", "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "*: Score re-evaluated by Moonshot AI under Kimi K2.5 evaluation conditions." }, { "model_id": "deepseek-v3.2", "benchmark_id": "widesearch", "score": 32.5, "reference_url": "https://huggingface.co/moonshotai/Kimi-K2.5", "reported_setting": "agentic search, pass@1; re-evaluated by Moonshot", "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "*: Score re-evaluated by Moonshot AI under Kimi K2.5 evaluation conditions (WideSearch benchmark). DeepSeek's own report does not include this benchmark." }, { "model_id": "gpt-5", "benchmark_id": "toolathlon", "score": 29.0, "reference_url": "https://huggingface.co/deepseek-ai/DeepSeek-V3.2/resolve/main/assets/paper.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "agentic", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "1.0", "context": "128K" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "DS V3.2 tech report Table 2: GPT-5-High=29.0. Third-party self-test by DeepSeek.", "candidates": [] }, { "model_id": "kimi-k2-thinking", "benchmark_id": "toolathlon", "score": 17.6, "reference_url": "https://huggingface.co/deepseek-ai/DeepSeek-V3.2/resolve/main/assets/paper.pdf", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "agentic", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "1.0", "context": "128K" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "DS V3.2 tech report Table 2: Kimi-K2-Thinking=17.6. Third-party self-test by DeepSeek.", "candidates": [] }, { "model_id": "minimax-m2", "benchmark_id": "toolathlon", "score": 16.0, "reference_url": "https://huggingface.co/deepseek-ai/DeepSeek-V3.2/resolve/main/assets/paper.pdf", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "agentic", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "1.0", "context": "128K" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "DS V3.2 tech report Table 2: MiniMax-M2=16.0. Third-party self-test by DeepSeek.", "candidates": [] }, { "model_id": "gemini-3-pro", "benchmark_id": "tau2_bench_avg", "score": 85.4, "reference_url": "https://huggingface.co/deepseek-ai/DeepSeek-V3.2/resolve/main/assets/paper.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "agentic", "sampling": "pass@1", "judge": "rule-based", "harness": "internal", "prompt_style": "default", "temperature": "1.0", "context": "128K" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "DS V3.2 tech report Table 2: Gemini-3.0-Pro=85.4. DS uses the test model itself as user agent (non-standard harness), not the official simulated-user harness. Third-party self-test.", "candidates": [ { "score": 90.7, "reference_url": "https://deepmind.google/models/gemini/flash/", "source_type": "official_blog", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "notes": "DeepMind /models/gemini/flash/ Performance table: τ2-bench avg 90.7% for Gemini 3 Pro Thinking. Differs from primary 85.4 from DeepSeek V3.2 tech report (matches_canonical=false — Flash page does not specify effort=high)." } ] }, { "model_id": "kimi-k2-thinking", "benchmark_id": "tau2_bench_avg", "score": 74.3, "reference_url": "https://huggingface.co/deepseek-ai/DeepSeek-V3.2/resolve/main/assets/paper.pdf", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "agentic", "sampling": "pass@1", "judge": "rule-based", "harness": "internal", "prompt_style": "default", "temperature": "1.0", "context": "128K" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "DS V3.2 tech report Table 2: Kimi-K2-Thinking=74.3. DS uses the test model itself as user agent (non-standard harness). Third-party self-test.", "candidates": [] }, { "model_id": "gpt-5", "benchmark_id": "mcpmark", "score": 50.9, "reference_url": "https://huggingface.co/deepseek-ai/DeepSeek-V3.2/resolve/main/assets/paper.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "agentic", "sampling": "pass@1", "judge": "rule-based", "harness": "internal", "prompt_style": "default", "temperature": "1.0", "context": "128K" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "DS V3.2 tech report Table 2: GPT-5-High=50.9. DS uses internal MCP environment (not official setting). Third-party self-test.", "candidates": [] }, { "model_id": "kimi-k2-thinking", "benchmark_id": "mcpmark", "score": 20.4, "reference_url": "https://huggingface.co/deepseek-ai/DeepSeek-V3.2/resolve/main/assets/paper.pdf", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "agentic", "sampling": "pass@1", "judge": "rule-based", "harness": "internal", "prompt_style": "default", "temperature": "1.0", "context": "128K" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "DS V3.2 tech report Table 2: Kimi-K2-Thinking=20.4. DS uses internal MCP environment. Third-party self-test.", "candidates": [] }, { "model_id": "minimax-m2", "benchmark_id": "mcpmark", "score": 24.4, "reference_url": "https://huggingface.co/deepseek-ai/DeepSeek-V3.2/resolve/main/assets/paper.pdf", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "agentic", "sampling": "pass@1", "judge": "rule-based", "harness": "internal", "prompt_style": "default", "temperature": "1.0", "context": "128K" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "DS V3.2 tech report Table 2: MiniMax-M2=24.4. DS uses internal MCP environment. Third-party self-test.", "candidates": [] }, { "model_id": "deepseek-v3.2", "benchmark_id": "mcpmark", "score": 38.0, "reference_url": "https://huggingface.co/deepseek-ai/DeepSeek-V3.2/resolve/main/assets/paper.pdf", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "agentic", "sampling": "pass@1", "judge": "rule-based", "harness": "internal", "prompt_style": "default", "temperature": "1.0", "context": "128K" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "DS V3.2 tech report Table 2: DeepSeek-V3.2-Thinking=38.0. DS uses internal MCP environment (not official MCP-Mark setting).", "candidates": [] }, { "model_id": "gemini-3-flash", "benchmark_id": "omnidocbench_1.5", "score": 0.121, "audit_status": "verified", "matches_canonical": true, "source_url": "https://deepmind.google/models/gemini/flash/", "source_type": "official_blog", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "notes": "DeepMind /models/gemini/flash/ Performance table: OmniDocBench 1.5 overall edit distance 0.121 (lower=better)", "candidates": [] }, { "model_id": "gemini-2.5-flash", "benchmark_id": "omnidocbench_1.5", "score": 0.154, "audit_status": "verified", "matches_canonical": true, "source_url": "https://deepmind.google/models/gemini/flash/", "source_type": "official_blog", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "notes": "DeepMind /models/gemini/flash/ Performance table: OmniDocBench 1.5 overall edit distance 0.154", "candidates": [] }, { "model_id": "gemini-2.5-pro", "benchmark_id": "omnidocbench_1.5", "score": 0.145, "audit_status": "verified", "matches_canonical": true, "source_url": "https://deepmind.google/models/gemini/flash/", "source_type": "official_blog", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "notes": "DeepMind /models/gemini/flash/ Performance table: OmniDocBench 1.5 overall edit distance 0.145", "candidates": [] }, { "model_id": "claude-sonnet-4.5", "benchmark_id": "omnidocbench_1.5", "score": 0.145, "audit_status": "verified", "matches_canonical": false, "source_url": "https://deepmind.google/models/gemini/flash/", "source_type": "official_blog", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "notes": "DeepMind /models/gemini/flash/ Performance table: OmniDocBench 1.5 0.145 (third-party self-test by Google DeepMind)", "candidates": [] }, { "model_id": "gemini-3-flash", "benchmark_id": "video_mmmu", "score": 86.9, "audit_status": "verified", "matches_canonical": true, "source_url": "https://deepmind.google/models/gemini/flash/", "source_type": "official_blog", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "notes": "DeepMind /models/gemini/flash/ Performance table: Video-MMMU (Knowledge acquisition from videos) 86.9%", "candidates": [] }, { "model_id": "gemini-3-pro", "benchmark_id": "video_mmmu", "score": 87.6, "audit_status": "verified", "matches_canonical": false, "source_url": "https://deepmind.google/models/gemini/flash/", "source_type": "official_blog", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "notes": "DeepMind /models/gemini/flash/ Performance table: Video-MMMU 87.6% (cross-model comparison; Flash page column label \"Gemini 3 Pro Thinking\" does not specify \"High\" effort — matches_canonical=false)", "candidates": [] }, { "model_id": "gemini-2.5-flash", "benchmark_id": "video_mmmu", "score": 79.2, "audit_status": "verified", "matches_canonical": true, "source_url": "https://deepmind.google/models/gemini/flash/", "source_type": "official_blog", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "notes": "DeepMind /models/gemini/flash/ Performance table: Video-MMMU 79.2%", "candidates": [] }, { "model_id": "gemini-2.5-pro", "benchmark_id": "video_mmmu", "score": 83.6, "audit_status": "verified", "matches_canonical": true, "source_url": "https://deepmind.google/models/gemini/flash/", "source_type": "official_blog", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "notes": "DeepMind /models/gemini/flash/ Performance table: Video-MMMU 83.6%", "candidates": [] }, { "model_id": "claude-sonnet-4.5", "benchmark_id": "video_mmmu", "score": 77.8, "audit_status": "verified", "matches_canonical": false, "source_url": "https://deepmind.google/models/gemini/flash/", "source_type": "official_blog", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "notes": "DeepMind /models/gemini/flash/ Performance table: Video-MMMU 77.8% (third-party self-test by Google DeepMind)", "candidates": [] }, { "model_id": "gpt-5.2", "benchmark_id": "video_mmmu", "score": 85.9, "audit_status": "verified", "matches_canonical": false, "source_url": "https://deepmind.google/models/gemini/flash/", "source_type": "official_blog", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "notes": "DeepMind /models/gemini/flash/ Performance table: Video-MMMU 85.9% (third-party self-test by Google DeepMind)", "candidates": [] }, { "model_id": "gemini-3-flash", "benchmark_id": "tau2_bench_avg", "score": 90.2, "audit_status": "verified", "matches_canonical": true, "source_url": "https://deepmind.google/models/gemini/flash/", "source_type": "official_blog", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "notes": "DeepMind /models/gemini/flash/ Performance table: τ2-bench avg (agentic tool use) 90.2%", "candidates": [] }, { "model_id": "gemini-2.5-flash", "benchmark_id": "tau2_bench_avg", "score": 79.5, "audit_status": "verified", "matches_canonical": true, "source_url": "https://deepmind.google/models/gemini/flash/", "source_type": "official_blog", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "notes": "DeepMind /models/gemini/flash/ Performance table: τ2-bench avg 79.5%", "candidates": [] }, { "model_id": "grok-4.1", "benchmark_id": "vending_bench_2", "score": 1107, "audit_status": "verified", "matches_canonical": true, "source_url": "https://deepmind.google/models/gemini/flash/", "source_type": "official_blog", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "notes": "DeepMind /models/gemini/flash/ Performance table: Vending-Bench 2 net worth (mean) $1,107 (Grok 4.1 Fast Reasoning column)", "candidates": [] }, { "model_id": "gpt-4o", "benchmark_id": "arc_agi_1", "score": 4.5, "reference_url": "https://arcprize.org/leaderboard", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1 (single-pass)", "judge": "rule-based", "harness": "official (arcprize.org)", "system_type": "Base LLM" }, "matches_canonical": true, "source_type": "leaderboard", "audit_status": "verified", "notes": "arcprize.org leaderboard audit R5b: GPT-4o Base LLM on leaderboard" }, { "model_id": "gpt-4o", "benchmark_id": "arc_agi_2", "score": 0.0, "reference_url": "https://arcprize.org/leaderboard", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1 (single-pass)", "judge": "rule-based", "harness": "official (arcprize.org)", "system_type": "Base LLM" }, "matches_canonical": true, "source_type": "leaderboard", "audit_status": "verified", "notes": "arcprize.org leaderboard audit R5b: GPT-4o Base LLM on leaderboard" }, { "model_id": "gpt-4o-mini", "benchmark_id": "arc_agi_2", "score": 0.0, "reference_url": "https://arcprize.org/leaderboard", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1 (single-pass)", "judge": "rule-based", "harness": "official (arcprize.org)", "system_type": "Base LLM" }, "matches_canonical": true, "source_type": "leaderboard", "audit_status": "verified", "notes": "arcprize.org leaderboard audit R5b: GPT-4o-mini Base LLM on leaderboard (arc_agi_1 N/A)" }, { "model_id": "o1-mini", "benchmark_id": "arc_agi_1", "score": 14.0, "reference_url": "https://arcprize.org/leaderboard", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1 (single-pass)", "judge": "rule-based", "harness": "official (arcprize.org)", "system_type": "CoT" }, "matches_canonical": false, "source_type": "leaderboard", "audit_status": "verified", "notes": "arcprize.org leaderboard audit R5b: o1-mini on leaderboard" }, { "model_id": "o1-mini", "benchmark_id": "arc_agi_2", "score": 0.8, "reference_url": "https://arcprize.org/leaderboard", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1 (single-pass)", "judge": "rule-based", "harness": "official (arcprize.org)", "system_type": "CoT" }, "matches_canonical": false, "source_type": "leaderboard", "audit_status": "verified", "notes": "arcprize.org leaderboard audit R5b: o1-mini on leaderboard" }, { "model_id": "gemini-1.5-pro", "benchmark_id": "arc_agi_2", "score": 0.8, "reference_url": "https://arcprize.org/leaderboard", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1 (single-pass)", "judge": "rule-based", "harness": "official (arcprize.org)", "system_type": "Base LLM" }, "matches_canonical": true, "source_type": "leaderboard", "audit_status": "verified", "notes": "arcprize.org leaderboard audit R5b: Gemini 1.5 Pro on leaderboard (arc_agi_1 N/A)" }, { "model_id": "grok-4.20", "benchmark_id": "terminal_bench", "score": 57.3, "reference_url": "https://www.tbench.ai/leaderboard/terminal-bench/2.0", "audit_status": "verified", "source_type": "leaderboard", "rule_ids": [ "R5b" ], "notes": "grok-cli third-party agent, score=57.3% ± N/A. Only available entry on tbench 2.0." }, { "model_id": "minimax-m2.1", "benchmark_id": "terminal_bench", "score": 29.2, "reference_url": "https://www.tbench.ai/leaderboard/terminal-bench/2.0", "audit_status": "verified", "source_type": "leaderboard", "rule_ids": [ "R5b" ], "notes": "Confirmed 29.2% on tbench.ai Terminal-Bench 2.0 (Terminus 2 scaffold, 2025-12-23)." }, { "model_id": "minimax-m2.5", "benchmark_id": "terminal_bench", "score": 42.2, "reference_url": "https://www.tbench.ai/leaderboard/terminal-bench/2.0", "audit_status": "verified", "source_type": "leaderboard", "rule_ids": [ "R5b" ], "notes": "Confirmed 42.2% on tbench.ai Terminal-Bench 2.0 (Terminus 2 scaffold, 2026-02-23). minimax-m2.5 listed as \"Minimax m2.5\"." }, { "model_id": "minimax-m2", "benchmark_id": "hle_tools", "score": 31.8, "reference_url": "https://huggingface.co/MiniMaxAI/MiniMax-M2", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "search + python (text-only HLE subset)", "sampling": "pass@1", "judge": "rule-based", "harness": "WebExplorer agent framework", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "rule_ids": [ "R5b" ], "notes": "MiniMax M2 model card Table 1: HLE (w/ tools)=31.8. Uses search tools + Python tool (Jupyter), text-only HLE subset.", "candidates": [] }, { "model_id": "minimax-m2.7", "benchmark_id": "multi_swe_bench", "score": 52.7, "reference_url": "https://huggingface.co/MiniMaxAI/MiniMax-M2.7", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "agentic (bash + edit)", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "rule_ids": [ "R5b" ], "notes": "MiniMax M2.7 model card: Multi-SWE-Bench=52.7.", "candidates": [] }, { "model_id": "minimax-m2.7", "benchmark_id": "swe_bench_multilingual", "score": 76.5, "reference_url": "https://huggingface.co/MiniMaxAI/MiniMax-M2.7", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "agentic (bash + edit)", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "rule_ids": [ "R5b" ], "notes": "MiniMax M2.7 model card: SWE-bench Multilingual=76.5.", "candidates": [ { "score": 71.8, "reference_url": "https://research.nvidia.com/labs/nemotron/files/NVIDIA-Nemotron-3-Ultra-Technical-Report.pdf", "source_type": "tech_report", "reported_setting": { "version_metric": "Multilingual / % resolved", "tools": "agentic repository editing", "harness": "Harbor / OpenHands / AWS ECS", "sampling": "3-run average" }, "notes": "NVIDIA Nemotron 3 Ultra technical report, Final Table 10: MiniMax M2.7: swe_bench_multilingual=71.8." } ] }, { "model_id": "minimax-m2.7", "benchmark_id": "terminal_bench", "score": 57.0, "reference_url": "https://huggingface.co/MiniMaxAI/MiniMax-M2.7", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "agentic", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "rule_ids": [ "R5b" ], "notes": "MiniMax M2.7 model card: Terminal-Bench 2=57.0.", "candidates": [] }, { "model_id": "glm-4.7", "benchmark_id": "hle_tools", "score": 42.8, "reference_url": "https://z.ai/blog/glm-4.7", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "search+code+web", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "GLM-4.7 blog (first-party): GLM-4.7=42.8. Also confirmed in z.ai/blog/glm-5 (third-party: 42.8)." }, { "model_id": "glm-4.6", "benchmark_id": "hle_tools", "score": 30.4, "reference_url": "https://z.ai/blog/glm-4.7", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "search+code+web", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "GLM-4.7 blog (same provider Zhipu, third-party for GLM-4.6): GLM-4.6=30.4 in comparison table." }, { "model_id": "glm-5", "benchmark_id": "hle_tools", "score": 50.4, "reference_url": "https://z.ai/blog/glm-5", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "search+code+web", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "GLM-5 blog (first-party): GLM-5=50.4. Also confirmed in z.ai/blog/glm-5.1 (50.4)." }, { "model_id": "glm-4.7", "benchmark_id": "tau2_bench_avg", "score": 87.4, "reference_url": "https://z.ai/blog/glm-4.7", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "web+API", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "GLM-4.7 blog (first-party): τ²-Bench=87.4. Confirmed by z.ai/blog/glm-5 (third-party: 87.4)." }, { "model_id": "glm-5", "benchmark_id": "tau2_bench_avg", "score": 89.7, "reference_url": "https://z.ai/blog/glm-5", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "web+API", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "GLM-5 blog (first-party): τ²-Bench=89.7. Confirmed by z.ai/blog/glm-5.1 (same value 89.7, via comparison)." }, { "model_id": "glm-4.6", "benchmark_id": "browsecomp_cm", "score": 57.5, "reference_url": "https://z.ai/blog/glm-4.7", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "agentic", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "context_management": "discard-all", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "GLM-4.7 blog (same provider Zhipu): GLM-4.6=57.5 in BrowseComp (w/ Context Manage) table." }, { "model_id": "glm-4.7", "benchmark_id": "browsecomp_cm", "score": 67.5, "reference_url": "https://z.ai/blog/glm-4.7", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "agentic", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "context_management": "discard-all", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "GLM-4.7 blog (first-party): GLM-4.7=67.5. Confirmed in z.ai/blog/glm-5 (third-party: 67.5)." }, { "model_id": "glm-5", "benchmark_id": "browsecomp_cm", "score": 75.9, "reference_url": "https://z.ai/blog/glm-5", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "agentic", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "context_management": "discard-all", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "GLM-5 blog (first-party): GLM-5=75.9. Confirmed in z.ai/blog/glm-5.1 (75.9)." }, { "model_id": "glm-5.1", "benchmark_id": "browsecomp_cm", "score": 79.3, "reference_url": "https://z.ai/blog/glm-5.1", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "agentic", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "context_management": "discard-all", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "GLM-5.1 blog (first-party): GLM-5.1=79.3. NOTE: BP browsecomp primary=79.3 (DeepSeek V4 Pro card) may be mislabeled as plain BrowseComp; GLM-5.1 blog gives plain=68, CM=79.3." }, { "model_id": "claude-opus-4", "benchmark_id": "cybench", "score": 56.4, "reference_url": "https://www-cdn.anthropic.com/07b2a3f9902ee19fe39a36ca638e5ae987bc64dd.pdf", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "agentic (Kali environment with code editor + Terminal Tool; pwntools/metasploit/ghidra)", "sampling": "pass@30", "judge": "rule-based (CTF flag check)", "harness": "Anthropic internal (Kali-based VM)", "prompt_style": "default", "temperature": "default", "notes": "Claude 4 System Card §7.4.7: claude-opus-4 scored 22/39 on Cybench (1 challenge skipped due to infra/timing; benchmark has 40 total). Thinking mode not explicitly specified but consistent with canonical." }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "From Claude 4 System Card (May 2025), §7.4.7 Cybench: 22/39 = 56.4%." }, { "model_id": "claude-sonnet-4", "benchmark_id": "cybench", "score": 56.4, "reference_url": "https://www-cdn.anthropic.com/07b2a3f9902ee19fe39a36ca638e5ae987bc64dd.pdf", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "agentic (Kali environment with code editor + Terminal Tool; pwntools/metasploit/ghidra)", "sampling": "pass@30", "judge": "rule-based (CTF flag check)", "harness": "Anthropic internal (Kali-based VM)", "prompt_style": "default", "temperature": "default", "notes": "Claude 4 System Card §7.4.7: claude-sonnet-4 scored 22/39 on Cybench (1 challenge skipped due to infra/timing; benchmark has 40 total). Thinking mode not explicitly specified but consistent with canonical." }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "From Claude 4 System Card (May 2025), §7.4.7 Cybench: 22/39 = 56.4%." }, { "model_id": "gpt-5-mini", "benchmark_id": "aime_2025", "score": 91.1, "reference_url": "https://openai.com/index/introducing-gpt-5-for-developers/", "source_type": "official_blog", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official" }, "matches_canonical": true, "audit_status": "verified", "notes": "AIME 2025 no tools, per GPT-5 dev blog table 0", "candidates": [] }, { "model_id": "gpt-5-mini", "benchmark_id": "frontiermath", "score": 22.1, "reference_url": "https://openai.com/index/introducing-gpt-5-for-developers/", "source_type": "official_blog", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "python", "sampling": "pass@1", "judge": "rule-based", "harness": "official" }, "matches_canonical": true, "audit_status": "verified", "notes": "FrontierMath with Python tool, per GPT-5 dev blog table 0", "candidates": [] }, { "model_id": "gpt-5-mini", "benchmark_id": "gpqa_diamond", "score": 82.3, "reference_url": "https://openai.com/index/introducing-gpt-5-for-developers/", "source_type": "official_blog", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official" }, "matches_canonical": true, "audit_status": "verified", "notes": "GPQA Diamond no tools, per GPT-5 dev blog table 0", "candidates": [] }, { "model_id": "gpt-5-mini", "benchmark_id": "hle", "score": 16.7, "reference_url": "https://openai.com/index/introducing-gpt-5-for-developers/", "source_type": "official_blog", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official" }, "matches_canonical": true, "audit_status": "verified", "notes": "HLE no tools, per GPT-5 dev blog table 0", "candidates": [] }, { "model_id": "gpt-5-mini", "benchmark_id": "hmmt_feb_2025", "score": 87.8, "reference_url": "https://openai.com/index/introducing-gpt-5-for-developers/", "source_type": "official_blog", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official" }, "matches_canonical": true, "audit_status": "verified", "notes": "HMMT 2025 no tools, per GPT-5 dev blog table 0", "candidates": [] }, { "model_id": "gpt-5-mini", "benchmark_id": "mmmu", "score": 81.6, "reference_url": "https://openai.com/index/introducing-gpt-5-for-developers/", "source_type": "official_blog", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official" }, "matches_canonical": true, "audit_status": "verified", "notes": "MMMU, per GPT-5 dev blog table 1", "candidates": [] }, { "model_id": "gpt-5-mini", "benchmark_id": "mmmu_pro", "score": 74.1, "reference_url": "https://openai.com/index/introducing-gpt-5-for-developers/", "source_type": "official_blog", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official" }, "matches_canonical": true, "audit_status": "verified", "notes": "MMMU-Pro avg, per GPT-5 dev blog table 1", "candidates": [] }, { "model_id": "gpt-5-mini", "benchmark_id": "charxiv_reasoning", "score": 75.5, "reference_url": "https://openai.com/index/introducing-gpt-5-for-developers/", "source_type": "official_blog", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "python", "sampling": "pass@1", "judge": "rule-based", "harness": "official" }, "matches_canonical": true, "audit_status": "verified", "notes": "CharXiv reasoning python-enabled, per GPT-5 dev blog table 1", "candidates": [] }, { "model_id": "gpt-5-mini", "benchmark_id": "video_mmmu", "score": 82.5, "reference_url": "https://openai.com/index/introducing-gpt-5-for-developers/", "source_type": "official_blog", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official" }, "matches_canonical": true, "audit_status": "verified", "notes": "VideoMMMU max-frame 256, per GPT-5 dev blog table 1", "candidates": [] }, { "model_id": "gpt-5-mini", "benchmark_id": "erqa", "score": 62.9, "reference_url": "https://openai.com/index/introducing-gpt-5-for-developers/", "source_type": "official_blog", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official" }, "matches_canonical": true, "audit_status": "verified", "notes": "ERQA, per GPT-5 dev blog table 1", "candidates": [] }, { "model_id": "gpt-5-mini", "benchmark_id": "swelancer_freelance_dollars", "score": 75000, "reference_url": "https://openai.com/index/introducing-gpt-5-for-developers/", "source_type": "official_blog", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official" }, "matches_canonical": true, "audit_status": "verified", "notes": "SWE-Lancer IC SWE Diamond Freelance $75K, per GPT-5 dev blog table 2", "candidates": [] }, { "model_id": "gpt-5-mini", "benchmark_id": "swe_bench_verified", "score": 71.0, "reference_url": "https://openai.com/index/introducing-gpt-5-for-developers/", "source_type": "official_blog", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official" }, "matches_canonical": true, "audit_status": "verified", "notes": "SWE-bench Verified, per GPT-5 dev blog table 2", "candidates": [] }, { "model_id": "gpt-5-mini", "benchmark_id": "aider_polyglot_diff", "score": 71.6, "reference_url": "https://openai.com/index/introducing-gpt-5-for-developers/", "source_type": "official_blog", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official" }, "matches_canonical": true, "audit_status": "verified", "notes": "Aider polyglot diff, per GPT-5 dev blog table 2", "candidates": [] }, { "model_id": "gpt-5-mini", "benchmark_id": "multichallenge_o3mini_grader", "score": 62.3, "reference_url": "https://openai.com/index/introducing-gpt-5-for-developers/", "source_type": "official_blog", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official" }, "matches_canonical": true, "audit_status": "verified", "notes": "Scale Multichallenge o3-mini grader, per GPT-5 dev blog table 3", "candidates": [] }, { "model_id": "gpt-5-mini", "benchmark_id": "internal_api_if_hard", "score": 65.8, "reference_url": "https://openai.com/index/introducing-gpt-5-for-developers/", "source_type": "official_blog", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official" }, "matches_canonical": true, "audit_status": "verified", "notes": "Internal API IF hard, per GPT-5 dev blog table 3", "candidates": [] }, { "model_id": "gpt-5-mini", "benchmark_id": "collie", "score": 98.5, "reference_url": "https://openai.com/index/introducing-gpt-5-for-developers/", "source_type": "official_blog", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official" }, "matches_canonical": true, "audit_status": "verified", "notes": "COLLIE, per GPT-5 dev blog table 3", "candidates": [] }, { "model_id": "gpt-5-mini", "benchmark_id": "tau2_bench_airline", "score": 60.0, "reference_url": "https://openai.com/index/introducing-gpt-5-for-developers/", "source_type": "official_blog", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official" }, "matches_canonical": true, "audit_status": "verified", "notes": "Tau2-bench airline, per GPT-5 dev blog table 4", "candidates": [] }, { "model_id": "gpt-5-mini", "benchmark_id": "tau2_bench_retail", "score": 78.3, "reference_url": "https://openai.com/index/introducing-gpt-5-for-developers/", "source_type": "official_blog", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official" }, "matches_canonical": true, "audit_status": "verified", "notes": "Tau2-bench retail, per GPT-5 dev blog table 4", "candidates": [] }, { "model_id": "gpt-5-mini", "benchmark_id": "tau2_bench_telecom", "score": 74.1, "reference_url": "https://openai.com/index/introducing-gpt-5-for-developers/", "source_type": "official_blog", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official" }, "matches_canonical": true, "audit_status": "verified", "notes": "Tau2-bench telecom, per GPT-5 dev blog table 4", "candidates": [] }, { "model_id": "gpt-5-mini", "benchmark_id": "mrcr_v2_2needle_128k", "score": 84.3, "reference_url": "https://openai.com/index/introducing-gpt-5-for-developers/", "source_type": "official_blog", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official" }, "matches_canonical": true, "audit_status": "verified", "notes": "OpenAI-MRCR 2-needle 128k, per GPT-5 dev blog table 5", "candidates": [] }, { "model_id": "gpt-5-mini", "benchmark_id": "mrcr_v2_2needle_256k", "score": 58.8, "reference_url": "https://openai.com/index/introducing-gpt-5-for-developers/", "source_type": "official_blog", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official" }, "matches_canonical": true, "audit_status": "verified", "notes": "OpenAI-MRCR 2-needle 256k, per GPT-5 dev blog table 5", "candidates": [] }, { "model_id": "gpt-5-mini", "benchmark_id": "graphwalks_bfs_0k_128k", "score": 73.4, "reference_url": "https://openai.com/index/introducing-gpt-5-for-developers/", "source_type": "official_blog", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official" }, "matches_canonical": true, "audit_status": "verified", "notes": "Graphwalks BFS <128k, per GPT-5 dev blog table 5", "candidates": [] }, { "model_id": "gpt-5-mini", "benchmark_id": "graphwalks_parents_0k_128k", "score": 64.3, "reference_url": "https://openai.com/index/introducing-gpt-5-for-developers/", "source_type": "official_blog", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official" }, "matches_canonical": true, "audit_status": "verified", "notes": "Graphwalks parents <128k, per GPT-5 dev blog table 5", "candidates": [] }, { "model_id": "gpt-5-mini", "benchmark_id": "browsecomp_long_context_128k", "score": 89.4, "reference_url": "https://openai.com/index/introducing-gpt-5-for-developers/", "source_type": "official_blog", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official" }, "matches_canonical": true, "audit_status": "verified", "notes": "BrowseComp Long Context 128k, per GPT-5 dev blog table 5", "candidates": [] }, { "model_id": "gpt-5-mini", "benchmark_id": "browsecomp_long_context_256k", "score": 86.0, "reference_url": "https://openai.com/index/introducing-gpt-5-for-developers/", "source_type": "official_blog", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official" }, "matches_canonical": true, "audit_status": "verified", "notes": "BrowseComp Long Context 256k, per GPT-5 dev blog table 5", "candidates": [] }, { "model_id": "gpt-5-mini", "benchmark_id": "video_mme", "score": 78.5, "reference_url": "https://openai.com/index/introducing-gpt-5-for-developers/", "source_type": "official_blog", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official" }, "matches_canonical": true, "audit_status": "verified", "notes": "VideoMME long with subtitle, per GPT-5 dev blog table 5", "candidates": [] }, { "model_id": "gpt-5-mini", "benchmark_id": "longfact_concepts", "score": 0.7, "reference_url": "https://openai.com/index/introducing-gpt-5-for-developers/", "source_type": "official_blog", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official" }, "matches_canonical": true, "audit_status": "verified", "notes": "LongFact-Concepts hallucination rate, per GPT-5 dev blog table 6", "candidates": [] }, { "model_id": "gpt-5-mini", "benchmark_id": "longfact_objects", "score": 1.3, "reference_url": "https://openai.com/index/introducing-gpt-5-for-developers/", "source_type": "official_blog", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official" }, "matches_canonical": true, "audit_status": "verified", "notes": "LongFact-Objects hallucination rate, per GPT-5 dev blog table 6", "candidates": [] }, { "model_id": "gpt-5-mini", "benchmark_id": "factscore", "score": 3.5, "reference_url": "https://openai.com/index/introducing-gpt-5-for-developers/", "source_type": "official_blog", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official" }, "matches_canonical": true, "audit_status": "verified", "notes": "FActScore hallucination rate, per GPT-5 dev blog table 6", "candidates": [] }, { "model_id": "gpt-5-nano", "benchmark_id": "aime_2025", "score": 85.2, "reference_url": "https://openai.com/index/introducing-gpt-5-for-developers/", "source_type": "official_blog", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official" }, "matches_canonical": true, "audit_status": "verified", "notes": "AIME 2025 no tools, per GPT-5 dev blog table 0", "candidates": [] }, { "model_id": "gpt-5-nano", "benchmark_id": "frontiermath", "score": 9.6, "reference_url": "https://openai.com/index/introducing-gpt-5-for-developers/", "source_type": "official_blog", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "python", "sampling": "pass@1", "judge": "rule-based", "harness": "official" }, "matches_canonical": true, "audit_status": "verified", "notes": "FrontierMath with Python tool, per GPT-5 dev blog table 0", "candidates": [] }, { "model_id": "gpt-5-nano", "benchmark_id": "gpqa_diamond", "score": 71.2, "reference_url": "https://openai.com/index/introducing-gpt-5-for-developers/", "source_type": "official_blog", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official" }, "matches_canonical": true, "audit_status": "verified", "notes": "GPQA Diamond no tools, per GPT-5 dev blog table 0", "candidates": [] }, { "model_id": "gpt-5-nano", "benchmark_id": "hle", "score": 8.7, "reference_url": "https://openai.com/index/introducing-gpt-5-for-developers/", "source_type": "official_blog", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official" }, "matches_canonical": true, "audit_status": "verified", "notes": "HLE no tools, per GPT-5 dev blog table 0", "candidates": [] }, { "model_id": "gpt-5-nano", "benchmark_id": "hmmt_feb_2025", "score": 75.6, "reference_url": "https://openai.com/index/introducing-gpt-5-for-developers/", "source_type": "official_blog", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official" }, "matches_canonical": true, "audit_status": "verified", "notes": "HMMT 2025 no tools, per GPT-5 dev blog table 0", "candidates": [] }, { "model_id": "gpt-5-nano", "benchmark_id": "mmmu", "score": 75.6, "reference_url": "https://openai.com/index/introducing-gpt-5-for-developers/", "source_type": "official_blog", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official" }, "matches_canonical": true, "audit_status": "verified", "notes": "MMMU, per GPT-5 dev blog table 1", "candidates": [] }, { "model_id": "gpt-5-nano", "benchmark_id": "mmmu_pro", "score": 62.6, "reference_url": "https://openai.com/index/introducing-gpt-5-for-developers/", "source_type": "official_blog", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official" }, "matches_canonical": true, "audit_status": "verified", "notes": "MMMU-Pro avg, per GPT-5 dev blog table 1", "candidates": [] }, { "model_id": "gpt-5-nano", "benchmark_id": "charxiv_reasoning", "score": 62.7, "reference_url": "https://openai.com/index/introducing-gpt-5-for-developers/", "source_type": "official_blog", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "python", "sampling": "pass@1", "judge": "rule-based", "harness": "official" }, "matches_canonical": true, "audit_status": "verified", "notes": "CharXiv reasoning python-enabled, per GPT-5 dev blog table 1", "candidates": [] }, { "model_id": "gpt-5-nano", "benchmark_id": "video_mmmu", "score": 66.8, "reference_url": "https://openai.com/index/introducing-gpt-5-for-developers/", "source_type": "official_blog", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official" }, "matches_canonical": true, "audit_status": "verified", "notes": "VideoMMMU max-frame 256, per GPT-5 dev blog table 1", "candidates": [] }, { "model_id": "gpt-5-nano", "benchmark_id": "erqa", "score": 50.1, "reference_url": "https://openai.com/index/introducing-gpt-5-for-developers/", "source_type": "official_blog", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official" }, "matches_canonical": true, "audit_status": "verified", "notes": "ERQA, per GPT-5 dev blog table 1", "candidates": [] }, { "model_id": "gpt-5-nano", "benchmark_id": "swelancer_freelance_dollars", "score": 49000, "reference_url": "https://openai.com/index/introducing-gpt-5-for-developers/", "source_type": "official_blog", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official" }, "matches_canonical": true, "audit_status": "verified", "notes": "SWE-Lancer IC SWE Diamond Freelance $49K, per GPT-5 dev blog table 2", "candidates": [] }, { "model_id": "gpt-5-nano", "benchmark_id": "swe_bench_verified", "score": 54.7, "reference_url": "https://openai.com/index/introducing-gpt-5-for-developers/", "source_type": "official_blog", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official" }, "matches_canonical": true, "audit_status": "verified", "notes": "SWE-bench Verified, per GPT-5 dev blog table 2", "candidates": [] }, { "model_id": "gpt-5-nano", "benchmark_id": "aider_polyglot_diff", "score": 48.4, "reference_url": "https://openai.com/index/introducing-gpt-5-for-developers/", "source_type": "official_blog", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official" }, "matches_canonical": true, "audit_status": "verified", "notes": "Aider polyglot diff, per GPT-5 dev blog table 2", "candidates": [] }, { "model_id": "gpt-5-nano", "benchmark_id": "multichallenge_o3mini_grader", "score": 54.9, "reference_url": "https://openai.com/index/introducing-gpt-5-for-developers/", "source_type": "official_blog", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official" }, "matches_canonical": true, "audit_status": "verified", "notes": "Scale Multichallenge o3-mini grader, per GPT-5 dev blog table 3", "candidates": [] }, { "model_id": "gpt-5-nano", "benchmark_id": "internal_api_if_hard", "score": 56.1, "reference_url": "https://openai.com/index/introducing-gpt-5-for-developers/", "source_type": "official_blog", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official" }, "matches_canonical": true, "audit_status": "verified", "notes": "Internal API IF hard, per GPT-5 dev blog table 3", "candidates": [] }, { "model_id": "gpt-5-nano", "benchmark_id": "collie", "score": 96.9, "reference_url": "https://openai.com/index/introducing-gpt-5-for-developers/", "source_type": "official_blog", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official" }, "matches_canonical": true, "audit_status": "verified", "notes": "COLLIE, per GPT-5 dev blog table 3", "candidates": [] }, { "model_id": "gpt-5-nano", "benchmark_id": "tau2_bench_airline", "score": 41.0, "reference_url": "https://openai.com/index/introducing-gpt-5-for-developers/", "source_type": "official_blog", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official" }, "matches_canonical": true, "audit_status": "verified", "notes": "Tau2-bench airline, per GPT-5 dev blog table 4", "candidates": [] }, { "model_id": "gpt-5-nano", "benchmark_id": "tau2_bench_retail", "score": 62.3, "reference_url": "https://openai.com/index/introducing-gpt-5-for-developers/", "source_type": "official_blog", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official" }, "matches_canonical": true, "audit_status": "verified", "notes": "Tau2-bench retail, per GPT-5 dev blog table 4", "candidates": [] }, { "model_id": "gpt-5-nano", "benchmark_id": "tau2_bench_telecom", "score": 35.5, "reference_url": "https://openai.com/index/introducing-gpt-5-for-developers/", "source_type": "official_blog", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official" }, "matches_canonical": true, "audit_status": "verified", "notes": "Tau2-bench telecom, per GPT-5 dev blog table 4", "candidates": [] }, { "model_id": "gpt-5-nano", "benchmark_id": "mrcr_v2_2needle_128k", "score": 43.2, "reference_url": "https://openai.com/index/introducing-gpt-5-for-developers/", "source_type": "official_blog", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official" }, "matches_canonical": true, "audit_status": "verified", "notes": "OpenAI-MRCR 2-needle 128k, per GPT-5 dev blog table 5", "candidates": [] }, { "model_id": "gpt-5-nano", "benchmark_id": "mrcr_v2_2needle_256k", "score": 34.9, "reference_url": "https://openai.com/index/introducing-gpt-5-for-developers/", "source_type": "official_blog", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official" }, "matches_canonical": true, "audit_status": "verified", "notes": "OpenAI-MRCR 2-needle 256k, per GPT-5 dev blog table 5", "candidates": [] }, { "model_id": "gpt-5-nano", "benchmark_id": "graphwalks_bfs_0k_128k", "score": 64.0, "reference_url": "https://openai.com/index/introducing-gpt-5-for-developers/", "source_type": "official_blog", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official" }, "matches_canonical": true, "audit_status": "verified", "notes": "Graphwalks BFS <128k, per GPT-5 dev blog table 5", "candidates": [] }, { "model_id": "gpt-5-nano", "benchmark_id": "graphwalks_parents_0k_128k", "score": 43.8, "reference_url": "https://openai.com/index/introducing-gpt-5-for-developers/", "source_type": "official_blog", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official" }, "matches_canonical": true, "audit_status": "verified", "notes": "Graphwalks parents <128k, per GPT-5 dev blog table 5", "candidates": [] }, { "model_id": "gpt-5-nano", "benchmark_id": "browsecomp_long_context_128k", "score": 80.4, "reference_url": "https://openai.com/index/introducing-gpt-5-for-developers/", "source_type": "official_blog", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official" }, "matches_canonical": true, "audit_status": "verified", "notes": "BrowseComp Long Context 128k, per GPT-5 dev blog table 5", "candidates": [] }, { "model_id": "gpt-5-nano", "benchmark_id": "browsecomp_long_context_256k", "score": 68.4, "reference_url": "https://openai.com/index/introducing-gpt-5-for-developers/", "source_type": "official_blog", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official" }, "matches_canonical": true, "audit_status": "verified", "notes": "BrowseComp Long Context 256k, per GPT-5 dev blog table 5", "candidates": [] }, { "model_id": "gpt-5-nano", "benchmark_id": "video_mme", "score": 65.7, "reference_url": "https://openai.com/index/introducing-gpt-5-for-developers/", "source_type": "official_blog", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official" }, "matches_canonical": true, "audit_status": "verified", "notes": "VideoMME long with subtitle, per GPT-5 dev blog table 5", "candidates": [] }, { "model_id": "gpt-5-nano", "benchmark_id": "longfact_concepts", "score": 1.0, "reference_url": "https://openai.com/index/introducing-gpt-5-for-developers/", "source_type": "official_blog", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official" }, "matches_canonical": true, "audit_status": "verified", "notes": "LongFact-Concepts hallucination rate, per GPT-5 dev blog table 6", "candidates": [] }, { "model_id": "gpt-5-nano", "benchmark_id": "longfact_objects", "score": 2.8, "reference_url": "https://openai.com/index/introducing-gpt-5-for-developers/", "source_type": "official_blog", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official" }, "matches_canonical": true, "audit_status": "verified", "notes": "LongFact-Objects hallucination rate, per GPT-5 dev blog table 6", "candidates": [] }, { "model_id": "gpt-5-nano", "benchmark_id": "factscore", "score": 7.3, "reference_url": "https://openai.com/index/introducing-gpt-5-for-developers/", "source_type": "official_blog", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official" }, "matches_canonical": true, "audit_status": "verified", "notes": "FActScore hallucination rate, per GPT-5 dev blog table 6", "candidates": [] }, { "model_id": "gpt-5.2-codex", "benchmark_id": "swe_bench_pro", "score": 56.4, "reference_url": "https://openai.com/index/introducing-gpt-5-3-codex/", "source_type": "official_blog", "reported_setting": { "mode": "thinking", "effort": "xhigh", "tools": "none (per-bench override)", "sampling": "pass@1", "judge": "rule-based", "harness": "official" }, "matches_canonical": true, "audit_status": "verified", "notes": "SWE-Bench Pro (Public), per GPT-5.3-Codex blog table 0", "candidates": [] }, { "model_id": "gpt-5.2-codex", "benchmark_id": "terminal_bench", "score": 64.0, "reference_url": "https://openai.com/index/introducing-gpt-5-3-codex/", "source_type": "official_blog", "reported_setting": { "mode": "thinking", "effort": "xhigh", "tools": "none (per-bench override)", "sampling": "pass@1", "judge": "rule-based", "harness": "official" }, "matches_canonical": true, "audit_status": "verified", "notes": "Terminal-Bench 2.0, per GPT-5.3-Codex blog table 0", "candidates": [] }, { "model_id": "gpt-5.2-codex", "benchmark_id": "osworld", "score": 38.2, "reference_url": "https://openai.com/index/introducing-gpt-5-3-codex/", "source_type": "official_blog", "reported_setting": { "mode": "thinking", "effort": "xhigh", "tools": "none (per-bench override)", "sampling": "pass@1", "judge": "rule-based", "harness": "official" }, "matches_canonical": true, "audit_status": "verified", "notes": "OSWorld-Verified, per GPT-5.3-Codex blog table 0", "candidates": [] }, { "model_id": "gpt-5.2-codex", "benchmark_id": "cybersecurity_ctf", "score": 67.4, "reference_url": "https://openai.com/index/introducing-gpt-5-3-codex/", "source_type": "official_blog", "reported_setting": { "mode": "thinking", "effort": "xhigh", "tools": "none (per-bench override)", "sampling": "pass@1", "judge": "rule-based", "harness": "official" }, "matches_canonical": true, "audit_status": "verified", "notes": "Cybersecurity CTF, per GPT-5.3-Codex blog table 0", "candidates": [] }, { "model_id": "gpt-5.2-codex", "benchmark_id": "swelancer", "score": 76.0, "reference_url": "https://openai.com/index/introducing-gpt-5-3-codex/", "source_type": "official_blog", "reported_setting": { "mode": "thinking", "effort": "xhigh", "tools": "none (per-bench override)", "sampling": "pass@1", "judge": "rule-based", "harness": "official" }, "matches_canonical": true, "audit_status": "verified", "notes": "SWE-Lancer IC Diamond, per GPT-5.3-Codex blog table 0", "candidates": [] }, { "model_id": "claude-opus-4.7", "benchmark_id": "frontiermath_tier4", "score": 22.9, "reference_url": "https://openai.com/index/introducing-gpt-5-5/", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "OpenAI internal", "prompt_style": "default", "temperature": "default" }, "matches_canonical": false, "source_type": "third_party", "audit_status": "verified", "notes": "GPT-5.5 blog table: FrontierMath Tier 4=22.9%; third-party self-test by OpenAI" }, { "model_id": "claude-opus-4.7", "benchmark_id": "gdpval_oai_woe", "score": 80.3, "reference_url": "https://openai.com/index/introducing-gpt-5-5/", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "OpenAI internal", "prompt_style": "default", "temperature": "default" }, "matches_canonical": false, "source_type": "third_party", "audit_status": "verified", "notes": "GPT-5.5 blog table: GDPval(wins or ties)=80.3%; third-party self-test by OpenAI" }, { "model_id": "claude-opus-4.7", "benchmark_id": "officeqa_pro", "score": 43.6, "reference_url": "https://openai.com/index/introducing-gpt-5-5/", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "OpenAI internal", "prompt_style": "default", "temperature": "default" }, "matches_canonical": false, "source_type": "third_party", "audit_status": "verified", "notes": "GPT-5.5 blog table: OfficeQA Pro=43.6%; third-party self-test by OpenAI", "candidates": [ { "score": 65.0, "reference_url": "https://www.anthropic.com/claude-sonnet-5-system-card", "source_type": "model_card", "reported_setting": { "effort": "source does not state", "tools": "agentic benchmark harness", "sampling": "pass@1", "harness": "office" }, "notes": "Anthropic Claude Sonnet 5 System Card Figure 8.11.1.A, page 131: Claude Opus 4.7; OfficeQA / OfficeQA Pro [OfficeQA Pro]=65.0. Source setting: effort=source does not state; tools=agentic benchmark harness; sampling=pass@1; harness=office." }, { "score": 76.5, "reference_url": "https://seed.bytedance.com/en/blog/seed2-1-officially-released-advancing-ai-productivity", "source_type": "official_blog", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "document and rendered-image analysis", "sampling": "pass@1", "judge": "OfficeQA Pro evaluator", "harness": "official", "prompt_style": "official default", "temperature": "unknown" }, "notes": "Release blog EN: General-agent chart (https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/user-upload/4xfa4mqq4u89c.png). The source says its internal implementations and adaptations can fluctuate relative to the official leaderboard." }, { "score": 76.5, "reference_url": "https://seed.bytedance.com/en/blog/seed2-1-officially-released-advancing-ai-productivity", "source_type": "official_blog", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "high-resolution multimodal agent harness", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official", "prompt_style": "official default", "temperature": "unknown" }, "notes": "Release blog EN: Multimodal-agent chart (https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/user-upload/4xfa4mqq4v1ol.png)." } ] }, { "model_id": "muse-spark", "benchmark_id": "charxiv_reasoning", "score": 86.4, "reference_url": "https://ai.meta.com/blog/introducing-muse-spark-msl/", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1 (single-pass)", "judge": "rule-based", "harness": "internal (Meta MSL)", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "rule_ids": [ "R5b" ], "notes": "Meta MSL blog main table: CharXiv Reasoning 86.4, Thinking mode.", "candidates": [ { "score": 88.9, "reference_url": "https://ai.meta.com/static-resource/muse-spark-1.1-evaluation-report", "source_type": "tech_report", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "Python", "sampling": "pass@1 unless source row states otherwise", "harness": "Meta MSL general-capability evaluation", "temperature": "1.0 for Muse; provider/source setting for comparisons", "context": "benchmark-specific", "configuration": "default" }, "notes": "Figure 44 · CharXiv Reasoning w/code execution; metric=headline_metric. Exact printed value in the general-capability summary." } ] }, { "model_id": "muse-spark", "benchmark_id": "erqa", "score": 64.7, "reference_url": "https://ai.meta.com/blog/introducing-muse-spark-msl/", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1 (single-pass)", "judge": "rule-based", "harness": "internal (Meta MSL)", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "rule_ids": [ "R5b" ], "notes": "Meta MSL blog main table: ERQA 64.7, Thinking mode. 925 questions with GPT-4.1-genai judge.", "candidates": [] }, { "model_id": "muse-spark", "benchmark_id": "simplevqa", "score": 71.3, "reference_url": "https://ai.meta.com/blog/introducing-muse-spark-msl/", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1 (single-pass)", "judge": "rule-based", "harness": "internal (Meta MSL)", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "rule_ids": [ "R5b" ], "notes": "Meta MSL blog main table: SimpleVQA 71.3, Thinking mode.", "candidates": [] }, { "model_id": "muse-spark", "benchmark_id": "screenspot_pro", "score": 84.1, "reference_url": "https://ai.meta.com/blog/introducing-muse-spark-msl/", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "python (cropping tool)", "sampling": "pass@1 (single-pass)", "judge": "rule-based", "harness": "internal (Meta MSL)", "prompt_style": "default", "temperature": "default", "notes": "with-Python variant per methodology; canonical BP is no-tools" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "rule_ids": [ "R5b" ], "notes": "Meta MSL blog: ScreenSpot Pro 84.1 with Python cropping tool. matches_canonical=false: canonical uses no tools.", "candidates": [] }, { "model_id": "muse-spark", "benchmark_id": "zerobench_tools", "score": 33.0, "reference_url": "https://ai.meta.com/blog/introducing-muse-spark-msl/", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "python", "sampling": "pass@5", "judge": "rule-based", "harness": "internal (Meta MSL)", "prompt_style": "default", "temperature": "default", "notes": "pass@5 per zerobench leaderboard convention; with Python" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "rule_ids": [ "R5b" ], "notes": "Meta MSL blog: ZeroBench 33.0, pass@5 with Python. matches_canonical=false: canonical BP uses pass@1.", "candidates": [] }, { "model_id": "muse-spark", "benchmark_id": "hle_tools", "score": 50.4, "reference_url": "https://ai.meta.com/blog/introducing-muse-spark-msl/", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "bash + browser (blocklist: hf.co, kaggle, pwc, lmsys)", "sampling": "pass@1 (single-pass)", "judge": "gpt-o3-mini (LLM-as-judge)", "harness": "internal (Meta MSL)", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "rule_ids": [ "R5b" ], "notes": "Meta MSL blog: HLE (With Tools) 50.4, Thinking mode. Bash and browser tools enabled per methodology.", "candidates": [] }, { "model_id": "muse-spark", "benchmark_id": "healthbench_hard", "score": 42.8, "reference_url": "https://ai.meta.com/blog/introducing-muse-spark-msl/", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1 (single-pass)", "judge": "gpt-4.1-genai (LLM-as-judge)", "harness": "internal (Meta MSL)", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "rule_ids": [ "R5b" ], "notes": "Meta MSL blog: HealthBench Hard 42.8, Thinking mode. 1,000 prompts, GPT-4.1-genai judge.", "candidates": [] }, { "model_id": "muse-spark", "benchmark_id": "medxpertqa_text", "score": 52.6, "reference_url": "https://ai.meta.com/blog/introducing-muse-spark-msl/", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1 (single-pass)", "judge": "gpt-oss-120b", "harness": "internal (Meta MSL)", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "rule_ids": [ "R5b" ], "notes": "Meta MSL blog: MedXpertQA Text 52.6, Thinking mode. 2,450 prompts, 10 choices, gpt-oss-120b judge.", "candidates": [] }, { "model_id": "muse-spark", "benchmark_id": "medxpertqa_mm", "score": 78.4, "reference_url": "https://ai.meta.com/blog/introducing-muse-spark-msl/", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1 (single-pass)", "judge": "gpt-oss-120b", "harness": "internal (Meta MSL)", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "rule_ids": [ "R5b" ], "notes": "Meta MSL blog: MedXpertQA MM 78.4, Thinking mode. 2,000 multimodal medical questions.", "candidates": [] }, { "model_id": "muse-spark", "benchmark_id": "deepsearchqa_f1", "score": 74.8, "reference_url": "https://ai.meta.com/blog/introducing-muse-spark-msl/", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "agentic (browser: search, open, find)", "sampling": "pass@1 (single-pass)", "judge": "gpt-oss-120b (LLM-as-judge for F1)", "harness": "internal (Meta MSL)", "prompt_style": "default", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "rule_ids": [ "R5b" ], "notes": "Meta MSL blog: DeepSearchQA F1 74.8, Thinking mode. Agentic browser tools, F1 metric.", "candidates": [ { "score": 76.8, "reference_url": "https://ai.meta.com/static-resource/muse-spark-1.1-evaluation-report", "source_type": "tech_report", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "search/browser", "sampling": "pass@1 unless source row states otherwise", "harness": "Meta MSL general-capability evaluation", "temperature": "1.0 for Muse; provider/source setting for comparisons", "context": "benchmark-specific", "configuration": "default" }, "notes": "Figure 44 · DeepSearchQA F1; metric=headline_metric. Exact printed value in the general-capability summary." } ] }, { "model_id": "muse-spark", "benchmark_id": "tau2_bench_telecom", "score": 91.5, "reference_url": "https://ai.meta.com/blog/introducing-muse-spark-msl/", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1 (single-pass)", "judge": "rule-based", "harness": "Artificial Analysis τ²-Bench Telecom Leaderboard", "prompt_style": "default", "temperature": "default", "notes": "Results from Artificial Analysis leaderboard per Meta methodology" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "rule_ids": [ "R5b" ], "notes": "Meta MSL blog: τ²-bench Telecom 91.5. Sourced from Artificial Analysis leaderboard.", "candidates": [] }, { "model_id": "o1-mini", "benchmark_id": "humaneval", "score": 92.4, "reference_url": "https://llm-stats.com/benchmarks/humaneval", "reported_setting": { "mode": "unknown", "harness": "unknown (llm-stats aggregator)", "sampling": "unknown", "judge": "unknown", "notes": "Setting not disclosed by aggregator" }, "audit_status": "verified_third_party", "source_type": "third_party_aggregator", "notes": "From llm-stats.com HumanEval leaderboard (display name 'o1-mini', rank 7, accessed 2026-04-29). Setting unknown." }, { "model_id": "qwen-2.5-32b", "benchmark_id": "humaneval", "score": 88.4, "reference_url": "https://llm-stats.com/benchmarks/humaneval", "reported_setting": { "mode": "unknown", "harness": "unknown (llm-stats aggregator)", "sampling": "unknown", "judge": "unknown", "notes": "Setting not disclosed by aggregator" }, "audit_status": "verified_third_party", "source_type": "third_party_aggregator", "notes": "From llm-stats.com HumanEval leaderboard (display name 'Qwen2.5 32B Instruct', rank 21, accessed 2026-04-29). Setting unknown." }, { "model_id": "gemma-3-12b", "benchmark_id": "humaneval", "score": 85.4, "reference_url": "https://llm-stats.com/benchmarks/humaneval", "reported_setting": { "mode": "unknown", "harness": "unknown (llm-stats aggregator)", "sampling": "unknown", "judge": "unknown", "notes": "Setting not disclosed by aggregator" }, "audit_status": "verified_third_party", "source_type": "third_party_aggregator", "notes": "From llm-stats.com HumanEval leaderboard (display name 'Gemma 3 12B', rank 34, accessed 2026-04-29). Setting unknown." }, { "model_id": "qwen-2.5-7b", "benchmark_id": "humaneval", "score": 84.8, "reference_url": "https://llm-stats.com/benchmarks/humaneval", "reported_setting": { "mode": "unknown", "harness": "unknown (llm-stats aggregator)", "sampling": "unknown", "judge": "unknown", "notes": "Setting not disclosed by aggregator" }, "audit_status": "verified_third_party", "source_type": "third_party_aggregator", "notes": "From llm-stats.com HumanEval leaderboard (display name 'Qwen2.5 7B Instruct', rank 37, accessed 2026-04-29). Setting unknown." }, { "model_id": "gemini-1.5-pro", "benchmark_id": "humaneval", "score": 84.1, "reference_url": "https://llm-stats.com/benchmarks/humaneval", "reported_setting": { "mode": "unknown", "harness": "unknown (llm-stats aggregator)", "sampling": "unknown", "judge": "unknown", "notes": "Setting not disclosed by aggregator" }, "audit_status": "verified_third_party", "source_type": "third_party_aggregator", "notes": "From llm-stats.com HumanEval leaderboard (display name 'Gemini 1.5 Pro', rank 39, accessed 2026-04-29). Setting unknown." }, { "model_id": "llama-3.1-70b", "benchmark_id": "humaneval", "score": 80.5, "reference_url": "https://llm-stats.com/benchmarks/humaneval", "reported_setting": { "mode": "unknown", "harness": "unknown (llm-stats aggregator)", "sampling": "unknown", "judge": "unknown", "notes": "Setting not disclosed by aggregator" }, "audit_status": "verified_third_party", "source_type": "third_party_aggregator", "notes": "From llm-stats.com HumanEval leaderboard (display name 'Llama 3.1 70B Instruct', rank 45, accessed 2026-04-29). Setting unknown." }, { "model_id": "gemini-1.5-flash", "benchmark_id": "humaneval", "score": 74.3, "reference_url": "https://llm-stats.com/benchmarks/humaneval", "reported_setting": { "mode": "unknown", "harness": "unknown (llm-stats aggregator)", "sampling": "unknown", "judge": "unknown", "notes": "Setting not disclosed by aggregator" }, "audit_status": "verified_third_party", "source_type": "third_party_aggregator", "notes": "From llm-stats.com HumanEval leaderboard (display name 'Gemini 1.5 Flash', rank 51, accessed 2026-04-29). Setting unknown." }, { "model_id": "llama-3.1-8b", "benchmark_id": "humaneval", "score": 72.6, "reference_url": "https://llm-stats.com/benchmarks/humaneval", "reported_setting": { "mode": "unknown", "harness": "unknown (llm-stats aggregator)", "sampling": "unknown", "judge": "unknown", "notes": "Setting not disclosed by aggregator" }, "audit_status": "verified_third_party", "source_type": "third_party_aggregator", "notes": "From llm-stats.com HumanEval leaderboard (display name 'Llama 3.1 8B Instruct', rank 54, accessed 2026-04-29). Setting unknown." }, { "model_id": "gemma-3-4b", "benchmark_id": "humaneval", "score": 71.3, "reference_url": "https://llm-stats.com/benchmarks/humaneval", "reported_setting": { "mode": "unknown", "harness": "unknown (llm-stats aggregator)", "sampling": "unknown", "judge": "unknown", "notes": "Setting not disclosed by aggregator" }, "audit_status": "verified_third_party", "source_type": "third_party_aggregator", "notes": "From llm-stats.com HumanEval leaderboard (display name 'Gemma 3 4B', rank 56, accessed 2026-04-29). Setting unknown." }, { "model_id": "gpt-3.5-turbo", "benchmark_id": "humaneval", "score": 68.0, "reference_url": "https://llm-stats.com/benchmarks/humaneval", "reported_setting": { "mode": "unknown", "harness": "unknown (llm-stats aggregator)", "sampling": "unknown", "judge": "unknown", "notes": "Setting not disclosed by aggregator" }, "audit_status": "verified_third_party", "source_type": "third_party_aggregator", "notes": "From llm-stats.com HumanEval leaderboard (display name 'GPT-3.5 Turbo', rank 58, accessed 2026-04-29). Setting unknown." }, { "model_id": "gemma-2-27b", "benchmark_id": "humaneval", "score": 51.8, "reference_url": "https://llm-stats.com/benchmarks/humaneval", "reported_setting": { "mode": "unknown", "harness": "unknown (llm-stats aggregator)", "sampling": "unknown", "judge": "unknown", "notes": "Setting not disclosed by aggregator" }, "audit_status": "verified_third_party", "source_type": "third_party_aggregator", "notes": "From llm-stats.com HumanEval leaderboard (display name 'Gemma 2 27B', rank 63, accessed 2026-04-29). Setting unknown." }, { "model_id": "gemma-3-1b", "benchmark_id": "humaneval", "score": 41.5, "reference_url": "https://llm-stats.com/benchmarks/humaneval", "reported_setting": { "mode": "unknown", "harness": "unknown (llm-stats aggregator)", "sampling": "unknown", "judge": "unknown", "notes": "Setting not disclosed by aggregator" }, "audit_status": "verified_third_party", "source_type": "third_party_aggregator", "notes": "From llm-stats.com HumanEval leaderboard (display name 'Gemma 3 1B', rank 64, accessed 2026-04-29). Setting unknown." }, { "model_id": "gemma-2-9b", "benchmark_id": "humaneval", "score": 40.2, "reference_url": "https://llm-stats.com/benchmarks/humaneval", "reported_setting": { "mode": "unknown", "harness": "unknown (llm-stats aggregator)", "sampling": "unknown", "judge": "unknown", "notes": "Setting not disclosed by aggregator" }, "audit_status": "verified_third_party", "source_type": "third_party_aggregator", "notes": "From llm-stats.com HumanEval leaderboard (display name 'Gemma 2 9B', rank 65, accessed 2026-04-29). Setting unknown." }, { "model_id": "ministral-8b", "benchmark_id": "humaneval", "score": 34.8, "reference_url": "https://llm-stats.com/benchmarks/humaneval", "reported_setting": { "mode": "unknown", "harness": "unknown (llm-stats aggregator)", "sampling": "unknown", "judge": "unknown", "notes": "Setting not disclosed by aggregator" }, "audit_status": "verified_third_party", "source_type": "third_party_aggregator", "notes": "From llm-stats.com HumanEval leaderboard (display name 'Ministral 8B Instruct', rank 66, accessed 2026-04-29). Setting unknown." }, { "model_id": "minicpm-sala", "benchmark_id": "humaneval", "score": 95.1, "reference_url": "https://llm-stats.com/benchmarks/humaneval", "reported_setting": { "mode": "unknown", "harness": "unknown (llm-stats aggregator)", "sampling": "unknown", "judge": "unknown", "notes": "Setting not disclosed by aggregator" }, "audit_status": "verified_third_party", "source_type": "third_party_aggregator", "notes": "From llm-stats.com HumanEval leaderboard (display name 'MiniCPM-SALA', rank 1, accessed 2026-04-29). Setting unknown." }, { "model_id": "qwen2.5-coder-32b-instruct", "benchmark_id": "humaneval", "score": 92.7, "reference_url": "https://llm-stats.com/benchmarks/humaneval", "reported_setting": { "mode": "unknown", "harness": "unknown (llm-stats aggregator)", "sampling": "unknown", "judge": "unknown", "notes": "Setting not disclosed by aggregator" }, "audit_status": "verified_third_party", "source_type": "third_party_aggregator", "notes": "From llm-stats.com HumanEval leaderboard (display name 'Qwen2.5-Coder 32B Instruct', rank 6, accessed 2026-04-29). Setting unknown." }, { "model_id": "sarvam-30b", "benchmark_id": "humaneval", "score": 92.1, "reference_url": "https://llm-stats.com/benchmarks/humaneval", "reported_setting": { "mode": "unknown", "harness": "unknown (llm-stats aggregator)", "sampling": "unknown", "judge": "unknown", "notes": "Setting not disclosed by aggregator" }, "audit_status": "verified_third_party", "source_type": "third_party_aggregator", "notes": "From llm-stats.com HumanEval leaderboard (display name 'Sarvam-30B', rank 8, accessed 2026-04-29). Setting unknown." }, { "model_id": "qwen2.5-vl-32b-instruct", "benchmark_id": "humaneval", "score": 91.5, "reference_url": "https://llm-stats.com/benchmarks/humaneval", "reported_setting": { "mode": "unknown", "harness": "unknown (llm-stats aggregator)", "sampling": "unknown", "judge": "unknown", "notes": "Setting not disclosed by aggregator" }, "audit_status": "verified_third_party", "source_type": "third_party_aggregator", "notes": "From llm-stats.com HumanEval leaderboard (display name 'Qwen2.5 VL 32B Instruct', rank 11, accessed 2026-04-29). Setting unknown." }, { "model_id": "granite-3.3-8b-instruct", "benchmark_id": "humaneval", "score": 89.7, "reference_url": "https://llm-stats.com/benchmarks/humaneval", "reported_setting": { "mode": "unknown", "harness": "unknown (llm-stats aggregator)", "sampling": "unknown", "judge": "unknown", "notes": "Setting not disclosed by aggregator" }, "audit_status": "verified_third_party", "source_type": "third_party_aggregator", "notes": "From llm-stats.com HumanEval leaderboard (display name 'Granite 3.3 8B Instruct', rank 13, accessed 2026-04-29). Setting unknown." }, { "model_id": "granite-3.3-8b-base", "benchmark_id": "humaneval", "score": 89.7, "reference_url": "https://llm-stats.com/benchmarks/humaneval", "reported_setting": { "mode": "unknown", "harness": "unknown (llm-stats aggregator)", "sampling": "unknown", "judge": "unknown", "notes": "Setting not disclosed by aggregator" }, "audit_status": "verified_third_party", "source_type": "third_party_aggregator", "notes": "From llm-stats.com HumanEval leaderboard (display name 'Granite 3.3 8B Base', rank 13, accessed 2026-04-29). Setting unknown." }, { "model_id": "gemini-diffusion", "benchmark_id": "humaneval", "score": 89.6, "reference_url": "https://llm-stats.com/benchmarks/humaneval", "reported_setting": { "mode": "unknown", "harness": "unknown (llm-stats aggregator)", "sampling": "unknown", "judge": "unknown", "notes": "Setting not disclosed by aggregator" }, "audit_status": "verified_third_party", "source_type": "third_party_aggregator", "notes": "From llm-stats.com HumanEval leaderboard (display name 'Gemini Diffusion', rank 15, accessed 2026-04-29). Setting unknown." }, { "model_id": "longcat-flash-chat", "benchmark_id": "humaneval", "score": 88.4, "reference_url": "https://llm-stats.com/benchmarks/humaneval", "reported_setting": { "mode": "unknown", "harness": "unknown (llm-stats aggregator)", "sampling": "unknown", "judge": "unknown", "notes": "Setting not disclosed by aggregator" }, "audit_status": "verified_third_party", "source_type": "third_party_aggregator", "notes": "From llm-stats.com HumanEval leaderboard (display name 'LongCat-Flash-Chat', rank 19, accessed 2026-04-29). Setting unknown." }, { "model_id": "qwen2.5-coder-7b-instruct", "benchmark_id": "humaneval", "score": 88.4, "reference_url": "https://llm-stats.com/benchmarks/humaneval", "reported_setting": { "mode": "unknown", "harness": "unknown (llm-stats aggregator)", "sampling": "unknown", "judge": "unknown", "notes": "Setting not disclosed by aggregator" }, "audit_status": "verified_third_party", "source_type": "third_party_aggregator", "notes": "From llm-stats.com HumanEval leaderboard (display name 'Qwen2.5-Coder 7B Instruct', rank 21, accessed 2026-04-29). Setting unknown." }, { "model_id": "grok-2", "benchmark_id": "humaneval", "score": 88.4, "reference_url": "https://llm-stats.com/benchmarks/humaneval", "reported_setting": { "mode": "unknown", "harness": "unknown (llm-stats aggregator)", "sampling": "unknown", "judge": "unknown", "notes": "Setting not disclosed by aggregator" }, "audit_status": "verified_third_party", "source_type": "third_party_aggregator", "notes": "From llm-stats.com HumanEval leaderboard (display name 'Grok-2', rank 21, accessed 2026-04-29). Setting unknown." }, { "model_id": "claude-3.5-haiku", "benchmark_id": "humaneval", "score": 88.1, "reference_url": "https://llm-stats.com/benchmarks/humaneval", "reported_setting": { "mode": "unknown", "harness": "unknown (llm-stats aggregator)", "sampling": "unknown", "judge": "unknown", "notes": "Setting not disclosed by aggregator" }, "audit_status": "verified_third_party", "source_type": "third_party_aggregator", "notes": "From llm-stats.com HumanEval leaderboard (display name 'Claude 3.5 Haiku', rank 25, accessed 2026-04-29). Setting unknown." }, { "model_id": "gpt-4-turbo", "benchmark_id": "humaneval", "score": 87.1, "reference_url": "https://llm-stats.com/benchmarks/humaneval", "reported_setting": { "mode": "unknown", "harness": "unknown (llm-stats aggregator)", "sampling": "unknown", "judge": "unknown", "notes": "Setting not disclosed by aggregator" }, "audit_status": "verified_third_party", "source_type": "third_party_aggregator", "notes": "From llm-stats.com HumanEval leaderboard (display name 'GPT-4 Turbo', rank 30, accessed 2026-04-29). Setting unknown." }, { "model_id": "qwen2-72b-instruct", "benchmark_id": "humaneval", "score": 86.0, "reference_url": "https://llm-stats.com/benchmarks/humaneval", "reported_setting": { "mode": "unknown", "harness": "unknown (llm-stats aggregator)", "sampling": "unknown", "judge": "unknown", "notes": "Setting not disclosed by aggregator" }, "audit_status": "verified_third_party", "source_type": "third_party_aggregator", "notes": "From llm-stats.com HumanEval leaderboard (display name 'Qwen2 72B Instruct', rank 32, accessed 2026-04-29). Setting unknown." }, { "model_id": "grok-2-mini", "benchmark_id": "humaneval", "score": 85.7, "reference_url": "https://llm-stats.com/benchmarks/humaneval", "reported_setting": { "mode": "unknown", "harness": "unknown (llm-stats aggregator)", "sampling": "unknown", "judge": "unknown", "notes": "Setting not disclosed by aggregator" }, "audit_status": "verified_third_party", "source_type": "third_party_aggregator", "notes": "From llm-stats.com HumanEval leaderboard (display name 'Grok-2 mini', rank 33, accessed 2026-04-29). Setting unknown." }, { "model_id": "nova-lite", "benchmark_id": "humaneval", "score": 85.4, "reference_url": "https://llm-stats.com/benchmarks/humaneval", "reported_setting": { "mode": "unknown", "harness": "unknown (llm-stats aggregator)", "sampling": "unknown", "judge": "unknown", "notes": "Setting not disclosed by aggregator" }, "audit_status": "verified_third_party", "source_type": "third_party_aggregator", "notes": "From llm-stats.com HumanEval leaderboard (display name 'Nova Lite', rank 34, accessed 2026-04-29). Setting unknown." }, { "model_id": "claude-3-opus", "benchmark_id": "humaneval", "score": 84.9, "reference_url": "https://llm-stats.com/benchmarks/humaneval", "reported_setting": { "mode": "unknown", "harness": "unknown (llm-stats aggregator)", "sampling": "unknown", "judge": "unknown", "notes": "Setting not disclosed by aggregator" }, "audit_status": "verified_third_party", "source_type": "third_party_aggregator", "notes": "From llm-stats.com HumanEval leaderboard (display name 'Claude 3 Opus', rank 36, accessed 2026-04-29). Setting unknown." }, { "model_id": "mistral-small-3-24b-instruct", "benchmark_id": "humaneval", "score": 84.8, "reference_url": "https://llm-stats.com/benchmarks/humaneval", "reported_setting": { "mode": "unknown", "harness": "unknown (llm-stats aggregator)", "sampling": "unknown", "judge": "unknown", "notes": "Setting not disclosed by aggregator" }, "audit_status": "verified_third_party", "source_type": "third_party_aggregator", "notes": "From llm-stats.com HumanEval leaderboard (display name 'Mistral Small 3 24B Instruct', rank 37, accessed 2026-04-29). Setting unknown." }, { "model_id": "ibm-granite-4.0-tiny-preview", "benchmark_id": "humaneval", "score": 82.4, "reference_url": "https://llm-stats.com/benchmarks/humaneval", "reported_setting": { "mode": "unknown", "harness": "unknown (llm-stats aggregator)", "sampling": "unknown", "judge": "unknown", "notes": "Setting not disclosed by aggregator" }, "audit_status": "verified_third_party", "source_type": "third_party_aggregator", "notes": "From llm-stats.com HumanEval leaderboard (display name 'IBM Granite 4.0 Tiny Preview', rank 42, accessed 2026-04-29). Setting unknown." }, { "model_id": "nova-micro", "benchmark_id": "humaneval", "score": 81.1, "reference_url": "https://llm-stats.com/benchmarks/humaneval", "reported_setting": { "mode": "unknown", "harness": "unknown (llm-stats aggregator)", "sampling": "unknown", "judge": "unknown", "notes": "Setting not disclosed by aggregator" }, "audit_status": "verified_third_party", "source_type": "third_party_aggregator", "notes": "From llm-stats.com HumanEval leaderboard (display name 'Nova Micro', rank 43, accessed 2026-04-29). Setting unknown." }, { "model_id": "codestral-22b", "benchmark_id": "humaneval", "score": 81.1, "reference_url": "https://llm-stats.com/benchmarks/humaneval", "reported_setting": { "mode": "unknown", "harness": "unknown (llm-stats aggregator)", "sampling": "unknown", "judge": "unknown", "notes": "Setting not disclosed by aggregator" }, "audit_status": "verified_third_party", "source_type": "third_party_aggregator", "notes": "From llm-stats.com HumanEval leaderboard (display name 'Codestral-22B', rank 43, accessed 2026-04-29). Setting unknown." }, { "model_id": "qwen2-7b-instruct", "benchmark_id": "humaneval", "score": 79.9, "reference_url": "https://llm-stats.com/benchmarks/humaneval", "reported_setting": { "mode": "unknown", "harness": "unknown (llm-stats aggregator)", "sampling": "unknown", "judge": "unknown", "notes": "Setting not disclosed by aggregator" }, "audit_status": "verified_third_party", "source_type": "third_party_aggregator", "notes": "From llm-stats.com HumanEval leaderboard (display name 'Qwen2 7B Instruct', rank 46, accessed 2026-04-29). Setting unknown." }, { "model_id": "qwen2.5-omni-7b", "benchmark_id": "humaneval", "score": 78.7, "reference_url": "https://llm-stats.com/benchmarks/humaneval", "reported_setting": { "mode": "unknown", "harness": "unknown (llm-stats aggregator)", "sampling": "unknown", "judge": "unknown", "notes": "Setting not disclosed by aggregator" }, "audit_status": "verified_third_party", "source_type": "third_party_aggregator", "notes": "From llm-stats.com HumanEval leaderboard (display name 'Qwen2.5-Omni-7B', rank 47, accessed 2026-04-29). Setting unknown." }, { "model_id": "claude-3-haiku", "benchmark_id": "humaneval", "score": 75.9, "reference_url": "https://llm-stats.com/benchmarks/humaneval", "reported_setting": { "mode": "unknown", "harness": "unknown (llm-stats aggregator)", "sampling": "unknown", "judge": "unknown", "notes": "Setting not disclosed by aggregator" }, "audit_status": "verified_third_party", "source_type": "third_party_aggregator", "notes": "From llm-stats.com HumanEval leaderboard (display name 'Claude 3 Haiku', rank 48, accessed 2026-04-29). Setting unknown." }, { "model_id": "gemma-3n-e4b-instructed-litert-preview", "benchmark_id": "humaneval", "score": 75.0, "reference_url": "https://llm-stats.com/benchmarks/humaneval", "reported_setting": { "mode": "unknown", "harness": "unknown (llm-stats aggregator)", "sampling": "unknown", "judge": "unknown", "notes": "Setting not disclosed by aggregator" }, "audit_status": "verified_third_party", "source_type": "third_party_aggregator", "notes": "From llm-stats.com HumanEval leaderboard (display name 'Gemma 3n E4B Instructed LiteRT Preview', rank 49, accessed 2026-04-29). Setting unknown." }, { "model_id": "gemma-3n-e4b-instructed", "benchmark_id": "humaneval", "score": 75.0, "reference_url": "https://llm-stats.com/benchmarks/humaneval", "reported_setting": { "mode": "unknown", "harness": "unknown (llm-stats aggregator)", "sampling": "unknown", "judge": "unknown", "notes": "Setting not disclosed by aggregator" }, "audit_status": "verified_third_party", "source_type": "third_party_aggregator", "notes": "From llm-stats.com HumanEval leaderboard (display name 'Gemma 3n E4B Instructed', rank 49, accessed 2026-04-29). Setting unknown." }, { "model_id": "grok-1.5", "benchmark_id": "humaneval", "score": 74.1, "reference_url": "https://llm-stats.com/benchmarks/humaneval", "reported_setting": { "mode": "unknown", "harness": "unknown (llm-stats aggregator)", "sampling": "unknown", "judge": "unknown", "notes": "Setting not disclosed by aggregator" }, "audit_status": "verified_third_party", "source_type": "third_party_aggregator", "notes": "From llm-stats.com HumanEval leaderboard (display name 'Grok-1.5', rank 52, accessed 2026-04-29). Setting unknown." }, { "model_id": "claude-3-sonnet", "benchmark_id": "humaneval", "score": 73.0, "reference_url": "https://llm-stats.com/benchmarks/humaneval", "reported_setting": { "mode": "unknown", "harness": "unknown (llm-stats aggregator)", "sampling": "unknown", "judge": "unknown", "notes": "Setting not disclosed by aggregator" }, "audit_status": "verified_third_party", "source_type": "third_party_aggregator", "notes": "From llm-stats.com HumanEval leaderboard (display name 'Claude 3 Sonnet', rank 53, accessed 2026-04-29). Setting unknown." }, { "model_id": "pixtral-12b", "benchmark_id": "humaneval", "score": 72.0, "reference_url": "https://llm-stats.com/benchmarks/humaneval", "reported_setting": { "mode": "unknown", "harness": "unknown (llm-stats aggregator)", "sampling": "unknown", "judge": "unknown", "notes": "Setting not disclosed by aggregator" }, "audit_status": "verified_third_party", "source_type": "third_party_aggregator", "notes": "From llm-stats.com HumanEval leaderboard (display name 'Pixtral-12B', rank 55, accessed 2026-04-29). Setting unknown." }, { "model_id": "phi-3.5-moe-instruct", "benchmark_id": "humaneval", "score": 70.7, "reference_url": "https://llm-stats.com/benchmarks/humaneval", "reported_setting": { "mode": "unknown", "harness": "unknown (llm-stats aggregator)", "sampling": "unknown", "judge": "unknown", "notes": "Setting not disclosed by aggregator" }, "audit_status": "verified_third_party", "source_type": "third_party_aggregator", "notes": "From llm-stats.com HumanEval leaderboard (display name 'Phi-3.5-MoE-instruct', rank 57, accessed 2026-04-29). Setting unknown." }, { "model_id": "gemma-3n-e2b-instructed-litert-preview", "benchmark_id": "humaneval", "score": 66.5, "reference_url": "https://llm-stats.com/benchmarks/humaneval", "reported_setting": { "mode": "unknown", "harness": "unknown (llm-stats aggregator)", "sampling": "unknown", "judge": "unknown", "notes": "Setting not disclosed by aggregator" }, "audit_status": "verified_third_party", "source_type": "third_party_aggregator", "notes": "From llm-stats.com HumanEval leaderboard (display name 'Gemma 3n E2B Instructed LiteRT (Preview)', rank 60, accessed 2026-04-29). Setting unknown." }, { "model_id": "gemma-3n-e2b-instructed", "benchmark_id": "humaneval", "score": 66.5, "reference_url": "https://llm-stats.com/benchmarks/humaneval", "reported_setting": { "mode": "unknown", "harness": "unknown (llm-stats aggregator)", "sampling": "unknown", "judge": "unknown", "notes": "Setting not disclosed by aggregator" }, "audit_status": "verified_third_party", "source_type": "third_party_aggregator", "notes": "From llm-stats.com HumanEval leaderboard (display name 'Gemma 3n E2B Instructed', rank 60, accessed 2026-04-29). Setting unknown." }, { "model_id": "phi-3.5-mini-instruct", "benchmark_id": "humaneval", "score": 62.8, "reference_url": "https://llm-stats.com/benchmarks/humaneval", "reported_setting": { "mode": "unknown", "harness": "unknown (llm-stats aggregator)", "sampling": "unknown", "judge": "unknown", "notes": "Setting not disclosed by aggregator" }, "audit_status": "verified_third_party", "source_type": "third_party_aggregator", "notes": "From llm-stats.com HumanEval leaderboard (display name 'Phi-3.5-mini-instruct', rank 62, accessed 2026-04-29). Setting unknown." }, { "model_id": "claude-3.5-sonnet", "benchmark_id": "ai2d", "score": 94.7, "reference_url": "https://llm-stats.com/benchmarks/ai2d", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default", "context": "default", "notes": "Per DeepSeek V3 paper Table 6." }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=Claude 3.5 Sonnet, slug=claude-3-5-sonnet-20241022, provider=Anthropic" }, { "model_id": "qwen3.6-plus", "benchmark_id": "ai2d", "score": 94.4, "reference_url": "https://llm-stats.com/benchmarks/ai2d", "reported_setting": {}, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=Qwen3.6 Plus, slug=qwen3.6-plus, provider=Alibaba Cloud / Qwen Team" }, { "model_id": "gemma-3-27b", "benchmark_id": "ai2d", "score": 84.5, "reference_url": "https://llm-stats.com/benchmarks/ai2d", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default", "context": "default (128K)", "notes": "Per Gemma 3 tech report (arxiv:2503.19786) Table 6 IT column for 27B." }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=Gemma 3 27B, slug=gemma-3-27b-it, provider=Google" }, { "model_id": "gemma-3-12b", "benchmark_id": "ai2d", "score": 84.2, "reference_url": "https://llm-stats.com/benchmarks/ai2d", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default", "context": "default", "notes": "Per Gemma 3 tech report Table 6 IT instruction-tuned column." }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=Gemma 3 12B, slug=gemma-3-12b-it, provider=Google" }, { "model_id": "qwen2.5-omni-7b", "benchmark_id": "ai2d", "score": 83.2, "reference_url": "https://llm-stats.com/benchmarks/ai2d", "reported_setting": { "mode": "unknown", "effort": "n/a", "tools": "unknown", "sampling": "unknown", "judge": "unknown", "harness": "unknown", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown", "notes": "Model added from third-party aggregator (llm-stats.com); canonical setting not established." }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=Qwen2.5-Omni-7B, slug=qwen2.5-omni-7b, provider=Alibaba Cloud / Qwen Team" }, { "model_id": "deepseek-v3.2", "benchmark_id": "aider_polyglot_diff", "score": 74.5, "reference_url": "https://llm-stats.com/benchmarks/aider-polyglot", "reported_setting": { "variant": "exp" }, "matches_canonical": false, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=DeepSeek-V3.2-Exp, slug=deepseek-v3.2-exp, provider=DeepSeek" }, { "model_id": "gemini-2.5-flash", "benchmark_id": "aider_polyglot_diff", "score": 61.9, "reference_url": "https://llm-stats.com/benchmarks/aider-polyglot", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none (per-bench override)", "sampling": "pass@1 (single attempt; multi-trial avg for smaller bench)", "judge": "rule-based", "harness": "AI Studio API (gemini-2.5-flash GA)", "prompt_style": "default", "temperature": "default", "context": "default (1M)", "notes": "Per Gemini 2.5 Flash Model Card (Dec 2025) GA Thinking column." }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=Gemini 2.5 Flash, slug=gemini-2.5-flash, provider=Google" }, { "model_id": "qwen3-235b", "benchmark_id": "aider_polyglot_diff", "score": 57.3, "reference_url": "https://llm-stats.com/benchmarks/aider-polyglot", "reported_setting": { "mode": "instruct", "snapshot": "2507" }, "matches_canonical": false, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=Qwen3-235B-A22B-Instruct-2507, slug=qwen3-235b-a22b-instruct-2507, provider=Alibaba Cloud / Qwen Team" }, { "model_id": "gpt-5", "benchmark_id": "aider_polyglot_whole", "score": 88.0, "reference_url": "https://llm-stats.com/benchmarks/aider-polyglot", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none (per-bench override)", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default", "context": "default", "notes": "Per OpenAI GPT-5 dev blog. high reasoning effort." }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=GPT-5, slug=gpt-5-2025-08-07, provider=OpenAI" }, { "model_id": "gemini-2.5-pro", "benchmark_id": "aider_polyglot_whole", "score": 82.2, "reference_url": "https://llm-stats.com/benchmarks/aider-polyglot", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1 (single attempt)", "judge": "rule-based", "harness": "AI Studio API", "prompt_style": "default", "temperature": "default", "context": "default", "notes": "Per Gemini 2.5 Pro Model Card (GA, June 27 2025): pass@1 single attempt, AI Studio API model-id gemini-2.5-pro-preview-06-05 / gemini-2.5-pro GA, default sampling. Release date set to GA date 2025-06-27." }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=Gemini 2.5 Pro Preview 06-05, slug=gemini-2.5-pro-preview-06-05, provider=Google" }, { "model_id": "o3-high", "benchmark_id": "aider_polyglot_whole", "score": 81.3, "reference_url": "https://llm-stats.com/benchmarks/aider-polyglot", "reported_setting": { "effort": "default (o3 base, not high)" }, "matches_canonical": false, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=o3, slug=o3-2025-04-16, provider=OpenAI" }, { "model_id": "deepseek-v3.2", "benchmark_id": "aider_polyglot_whole", "score": 74.5, "reference_url": "https://llm-stats.com/benchmarks/aider-polyglot", "reported_setting": { "variant": "exp" }, "matches_canonical": false, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=DeepSeek-V3.2-Exp, slug=deepseek-v3.2-exp, provider=DeepSeek" }, { "model_id": "deepseek-r1", "benchmark_id": "aider_polyglot_whole", "score": 71.6, "reference_url": "https://llm-stats.com/benchmarks/aider-polyglot", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none (per-bench override)", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default", "context": "default", "notes": "Per DeepSeek R1-0528 model card." }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=DeepSeek-R1-0528, slug=deepseek-r1-0528, provider=DeepSeek" }, { "model_id": "o4-mini-high", "benchmark_id": "aider_polyglot_whole", "score": 68.9, "reference_url": "https://llm-stats.com/benchmarks/aider-polyglot", "reported_setting": { "effort": "default (o4-mini base, not high)" }, "matches_canonical": false, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=o4-mini, slug=o4-mini, provider=OpenAI" }, { "model_id": "kimi-k2", "benchmark_id": "aider_polyglot_whole", "score": 60.0, "reference_url": "https://llm-stats.com/benchmarks/aider-polyglot", "reported_setting": { "variant": "instruct", "snapshot": "0905" }, "matches_canonical": false, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=Kimi K2-Instruct-0905, slug=kimi-k2-instruct-0905, provider=Moonshot AI" }, { "model_id": "qwen3-235b", "benchmark_id": "aider_polyglot_whole", "score": 57.3, "reference_url": "https://llm-stats.com/benchmarks/aider-polyglot", "reported_setting": { "mode": "instruct", "snapshot": "2507" }, "matches_canonical": false, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=Qwen3-235B-A22B-Instruct-2507, slug=qwen3-235b-a22b-instruct-2507, provider=Alibaba Cloud / Qwen Team" }, { "model_id": "deepseek-v3", "benchmark_id": "aider_polyglot_whole", "score": 49.6, "reference_url": "https://llm-stats.com/benchmarks/aider-polyglot", "reported_setting": {}, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=DeepSeek-V3, slug=deepseek-v3, provider=DeepSeek" }, { "model_id": "ministral-8b", "benchmark_id": "aime_2024", "score": 86.0, "reference_url": "https://llm-stats.com/benchmarks/aime-2024", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "OLMES (Allen AI)", "prompt_style": "default", "temperature": "default", "context": "default", "notes": "Per OLMo 2 paper Table 7 Instruct (instruction-tuned variant)." }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=Ministral 3 (8B Reasoning 2512), slug=ministral-8b-latest, provider=Mistral AI" }, { "model_id": "minicpm-sala", "benchmark_id": "aime_2024", "score": 83.8, "reference_url": "https://llm-stats.com/benchmarks/aime-2024", "reported_setting": { "mode": "unknown", "effort": "n/a", "tools": "unknown", "sampling": "unknown", "judge": "unknown", "harness": "unknown", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown", "notes": "Model added from third-party aggregator (llm-stats.com); canonical setting not established." }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=MiniCPM-SALA, slug=minicpm-sala, provider=OpenBMB" }, { "model_id": "granite-3.3-8b-base", "benchmark_id": "aime_2024", "score": 81.2, "reference_url": "https://llm-stats.com/benchmarks/aime-2024", "reported_setting": { "mode": "unknown", "effort": "n/a", "tools": "unknown", "sampling": "unknown", "judge": "unknown", "harness": "unknown", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown", "notes": "Model added from third-party aggregator (llm-stats.com); canonical setting not established." }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=Granite 3.3 8B Base, slug=granite-3.3-8b-base, provider=IBM" }, { "model_id": "granite-3.3-8b-instruct", "benchmark_id": "aime_2024", "score": 81.2, "reference_url": "https://llm-stats.com/benchmarks/aime-2024", "reported_setting": { "mode": "unknown", "effort": "n/a", "tools": "unknown", "sampling": "unknown", "judge": "unknown", "harness": "unknown", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown", "notes": "Model added from third-party aggregator (llm-stats.com); canonical setting not established." }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=Granite 3.3 8B Instruct, slug=granite-3.3-8b-instruct, provider=IBM" }, { "model_id": "o4-mini-high", "benchmark_id": "mt_aime_2024", "score": 93.4, "reference_url": "https://llm-stats.com/benchmarks/aime-2024", "reported_setting": { "effort": "default (o4-mini base, not high)" }, "matches_canonical": false, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=o4-mini, slug=o4-mini, provider=OpenAI" }, { "model_id": "grok-3-beta", "benchmark_id": "mt_aime_2024", "score": 93.3, "reference_url": "https://llm-stats.com/benchmarks/aime-2024", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none (per-bench override)", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default", "context": "default", "notes": "Per xAI Grok 3 blog: canonical = non-reasoning mode (the cross-model table without 'Think' label). Reasoning ON values not used." }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=Grok-3, slug=grok-3, provider=xAI" }, { "model_id": "gemini-2.5-pro", "benchmark_id": "mt_aime_2024", "score": 92.0, "reference_url": "https://llm-stats.com/benchmarks/aime-2024", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1 (single attempt)", "judge": "rule-based", "harness": "AI Studio API", "prompt_style": "default", "temperature": "default", "context": "default", "notes": "Per Gemini 2.5 Pro Model Card (GA, June 27 2025): pass@1 single attempt, AI Studio API model-id gemini-2.5-pro-preview-06-05 / gemini-2.5-pro GA, default sampling. Release date set to GA date 2025-06-27." }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=Gemini 2.5 Pro, slug=gemini-2.5-pro, provider=Google" }, { "model_id": "o3-high", "benchmark_id": "mt_aime_2024", "score": 91.6, "reference_url": "https://llm-stats.com/benchmarks/aime-2024", "reported_setting": { "effort": "default (o3 base, not high)" }, "matches_canonical": false, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=o3, slug=o3-2025-04-16, provider=OpenAI" }, { "model_id": "deepseek-r1", "benchmark_id": "mt_aime_2024", "score": 91.4, "reference_url": "https://llm-stats.com/benchmarks/aime-2024", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none (per-bench override)", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default", "context": "default", "notes": "Per DeepSeek R1-0528 model card." }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=DeepSeek-R1-0528, slug=deepseek-r1-0528, provider=DeepSeek" }, { "model_id": "gemini-2.5-flash", "benchmark_id": "mt_aime_2024", "score": 88.0, "reference_url": "https://llm-stats.com/benchmarks/aime-2024", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none (per-bench override)", "sampling": "pass@1 (single attempt; multi-trial avg for smaller bench)", "judge": "rule-based", "harness": "AI Studio API (gemini-2.5-flash GA)", "prompt_style": "default", "temperature": "default", "context": "default (1M)", "notes": "Per Gemini 2.5 Flash Model Card (Dec 2025) GA Thinking column." }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=Gemini 2.5 Flash, slug=gemini-2.5-flash, provider=Google" }, { "model_id": "ministral-8b", "benchmark_id": "mt_aime_2024", "score": 86.0, "reference_url": "https://llm-stats.com/benchmarks/aime-2024", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "OLMES (Allen AI)", "prompt_style": "default", "temperature": "default", "context": "default", "notes": "Per OLMo 2 paper Table 7 Instruct (instruction-tuned variant)." }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=Ministral 3 (8B Reasoning 2512), slug=ministral-8b-latest, provider=Mistral AI" }, { "model_id": "minicpm-sala", "benchmark_id": "mt_aime_2024", "score": 83.8, "reference_url": "https://llm-stats.com/benchmarks/aime-2024", "reported_setting": { "mode": "unknown", "effort": "n/a", "tools": "unknown", "sampling": "unknown", "judge": "unknown", "harness": "unknown", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown", "notes": "Model added from third-party aggregator (llm-stats.com); canonical setting not established." }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=MiniCPM-SALA, slug=minicpm-sala, provider=OpenBMB" }, { "model_id": "deepseek-r1-distill-qwen-32b", "benchmark_id": "mt_aime_2024", "score": 83.3, "reference_url": "https://llm-stats.com/benchmarks/aime-2024", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default", "context": "default", "notes": "Per DeepSeek R1 paper Table 15 (distilled model)." }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=DeepSeek R1 Distill Qwen 32B, slug=deepseek-r1-distill-qwen-32b, provider=DeepSeek" }, { "model_id": "deepseek-r1-distill-qwen-7b", "benchmark_id": "mt_aime_2024", "score": 83.3, "reference_url": "https://llm-stats.com/benchmarks/aime-2024", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default", "context": "default", "notes": "Per DeepSeek R1 paper Table 15 (distilled model)." }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=DeepSeek R1 Distill Qwen 7B, slug=deepseek-r1-distill-qwen-7b, provider=DeepSeek" }, { "model_id": "phi-4-reasoning-plus", "benchmark_id": "mt_aime_2024", "score": 81.3, "reference_url": "https://llm-stats.com/benchmarks/aime-2024", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1 (avg of 5)", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "0.8", "context": "default (32K)", "notes": "Per Phi-4-reasoning paper. Same as phi-4-reasoning + GRPO RL on top." }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=Phi 4 Reasoning Plus, slug=phi-4-reasoning-plus, provider=Microsoft" }, { "model_id": "granite-3.3-8b-base", "benchmark_id": "mt_aime_2024", "score": 81.2, "reference_url": "https://llm-stats.com/benchmarks/aime-2024", "reported_setting": { "mode": "unknown", "effort": "n/a", "tools": "unknown", "sampling": "unknown", "judge": "unknown", "harness": "unknown", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown", "notes": "Model added from third-party aggregator (llm-stats.com); canonical setting not established." }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=Granite 3.3 8B Base, slug=granite-3.3-8b-base, provider=IBM" }, { "model_id": "granite-3.3-8b-instruct", "benchmark_id": "mt_aime_2024", "score": 81.2, "reference_url": "https://llm-stats.com/benchmarks/aime-2024", "reported_setting": { "mode": "unknown", "effort": "n/a", "tools": "unknown", "sampling": "unknown", "judge": "unknown", "harness": "unknown", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown", "notes": "Model added from third-party aggregator (llm-stats.com); canonical setting not established." }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=Granite 3.3 8B Instruct, slug=granite-3.3-8b-instruct, provider=IBM" }, { "model_id": "deepseek-r1-distill-llama-8b", "benchmark_id": "mt_aime_2024", "score": 80.0, "reference_url": "https://llm-stats.com/benchmarks/aime-2024", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default", "context": "default", "notes": "Per DeepSeek R1 paper Table 15 (distilled model)." }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=DeepSeek R1 Distill Llama 8B, slug=deepseek-r1-distill-llama-8b, provider=DeepSeek" }, { "model_id": "deepseek-r1-distill-qwen-14b", "benchmark_id": "mt_aime_2024", "score": 80.0, "reference_url": "https://llm-stats.com/benchmarks/aime-2024", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default", "context": "default", "notes": "Per DeepSeek R1 paper Table 15 (distilled model)." }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=DeepSeek R1 Distill Qwen 14B, slug=deepseek-r1-distill-qwen-14b, provider=DeepSeek" }, { "model_id": "claude-3.7-sonnet", "benchmark_id": "mt_aime_2024", "score": 80.0, "reference_url": "https://llm-stats.com/benchmarks/aime-2024", "reported_setting": { "mode": "thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "zeroshot-cot", "temperature": 0.0, "context": "default", "notes": "Per Claude 4 blog cross-model table: with extended thinking for GPQA/MMMU/AIME/TAU; no extended thinking for SWE/Terminal/MMMLU." }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=Claude 3.7 Sonnet, slug=claude-3-7-sonnet-20250219, provider=Anthropic" }, { "model_id": "phi-4-reasoning", "benchmark_id": "mt_aime_2024", "score": 75.3, "reference_url": "https://llm-stats.com/benchmarks/aime-2024", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1 (avg of 5)", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "0.8", "context": "default (32K)", "notes": "Per Phi-4-reasoning paper (arxiv:2504.21318) Tables 1-2. Thinking model, temp 0.8, max_seq_len 31k." }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=Phi 4 Reasoning, slug=phi-4-reasoning, provider=Microsoft" }, { "model_id": "o1-high", "benchmark_id": "mt_aime_2024", "score": 74.3, "reference_url": "https://llm-stats.com/benchmarks/aime-2024", "reported_setting": { "effort": "default (o1 base, not high)" }, "matches_canonical": false, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=o1, slug=o1-2024-12-17, provider=OpenAI" }, { "model_id": "gemini-2.0-flash", "benchmark_id": "mt_aime_2024", "score": 73.3, "reference_url": "https://llm-stats.com/benchmarks/aime-2024", "reported_setting": { "effort": "thinking" }, "matches_canonical": false, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=Gemini 2.0 Flash Thinking, slug=gemini-2.0-flash-thinking, provider=Google" }, { "model_id": "kimi-k2", "benchmark_id": "mt_aime_2024", "score": 72.0, "reference_url": "https://llm-stats.com/benchmarks/aime-2024", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none (per-bench override)", "sampling": "pass@1 (avg of 4-64 trials per benchmark)", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default", "context": "default (256K)", "notes": "Per Kimi K2 tech report (arxiv:2507.20534) Table 3 Kimi-K2-Instruct column. Non-thinking model." }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=Kimi K2 0905, slug=kimi-k2-0905, provider=Moonshot AI" }, { "model_id": "deepseek-v3", "benchmark_id": "mt_aime_2024", "score": 59.4, "reference_url": "https://llm-stats.com/benchmarks/aime-2024", "reported_setting": {}, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=DeepSeek-V3 0324, slug=deepseek-v3-0324, provider=DeepSeek" }, { "model_id": "deepseek-r1-distill-qwen-1.5b", "benchmark_id": "mt_aime_2024", "score": 52.7, "reference_url": "https://llm-stats.com/benchmarks/aime-2024", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default", "context": "default", "notes": "Per DeepSeek R1 paper Table 15 (distilled model)." }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=DeepSeek R1 Distill Qwen 1.5B, slug=deepseek-r1-distill-qwen-1.5b, provider=DeepSeek" }, { "model_id": "gpt-4.1-mini", "benchmark_id": "mt_aime_2024", "score": 49.6, "reference_url": "https://llm-stats.com/benchmarks/aime-2024", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none (per-bench override)", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default", "context": "default", "notes": "Per OpenAI GPT-5 dev blog. non-reasoning model." }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=GPT-4.1 mini, slug=gpt-4.1-mini-2025-04-14, provider=OpenAI" }, { "model_id": "gpt-4.1", "benchmark_id": "mt_aime_2024", "score": 48.1, "reference_url": "https://llm-stats.com/benchmarks/aime-2024", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none (per-bench override)", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default", "context": "default", "notes": "Per OpenAI GPT-5 dev blog. non-reasoning model." }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=GPT-4.1, slug=gpt-4.1-2025-04-14, provider=OpenAI" }, { "model_id": "gpt-4.5", "benchmark_id": "mt_aime_2024", "score": 36.7, "reference_url": "https://llm-stats.com/benchmarks/aime-2024", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none (per-bench override)", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default", "context": "default", "notes": "Per OpenAI GPT-4.1 blog table." }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=GPT-4.5, slug=gpt-4.5, provider=OpenAI" }, { "model_id": "gpt-4.1-nano", "benchmark_id": "mt_aime_2024", "score": 29.4, "reference_url": "https://llm-stats.com/benchmarks/aime-2024", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none (per-bench override)", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default", "context": "default", "notes": "Per OpenAI GPT-5 dev blog. non-reasoning model." }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=GPT-4.1 nano, slug=gpt-4.1-nano-2025-04-14, provider=OpenAI" }, { "model_id": "gpt-4o", "benchmark_id": "mt_aime_2024", "score": 13.1, "reference_url": "https://llm-stats.com/benchmarks/aime-2024", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none (per-bench override)", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default", "context": "default", "notes": "Per OpenAI GPT-4.1 blog table." }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=GPT-4o, slug=gpt-4o-2024-08-06, provider=OpenAI" }, { "model_id": "sarvam-30b", "benchmark_id": "aime_2025", "score": 96.7, "reference_url": "https://llm-stats.com/benchmarks/aime-2025", "reported_setting": { "mode": "unknown", "effort": "n/a", "tools": "unknown", "sampling": "unknown", "judge": "unknown", "harness": "unknown", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown", "notes": "Model added from third-party aggregator (llm-stats.com); canonical setting not established." }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=Sarvam-30B, slug=sarvam-30b, provider=Sarvam AI" }, { "model_id": "ministral-8b", "benchmark_id": "aime_2025", "score": 78.7, "reference_url": "https://llm-stats.com/benchmarks/aime-2025", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "OLMES (Allen AI)", "prompt_style": "default", "temperature": "default", "context": "default", "notes": "Per OLMo 2 paper Table 7 Instruct (instruction-tuned variant)." }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=Ministral 3 (8B Reasoning 2512), slug=ministral-8b-latest, provider=Mistral AI" }, { "model_id": "minicpm-sala", "benchmark_id": "aime_2025", "score": 78.3, "reference_url": "https://llm-stats.com/benchmarks/aime-2025", "reported_setting": { "mode": "unknown", "effort": "n/a", "tools": "unknown", "sampling": "unknown", "judge": "unknown", "harness": "unknown", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown", "notes": "Model added from third-party aggregator (llm-stats.com); canonical setting not established." }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=MiniCPM-SALA, slug=minicpm-sala, provider=OpenBMB" }, { "model_id": "longcat-flash-chat", "benchmark_id": "aime_2025", "score": 61.3, "reference_url": "https://llm-stats.com/benchmarks/aime-2025", "reported_setting": { "mode": "unknown", "effort": "n/a", "tools": "unknown", "sampling": "unknown", "judge": "unknown", "harness": "unknown", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown", "notes": "Model added from third-party aggregator (llm-stats.com); canonical setting not established." }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=LongCat-Flash-Chat, slug=longcat-flash-chat, provider=Meituan" }, { "model_id": "gemini-diffusion", "benchmark_id": "aime_2025", "score": 23.3, "reference_url": "https://llm-stats.com/benchmarks/aime-2025", "reported_setting": { "mode": "unknown", "effort": "n/a", "tools": "unknown", "sampling": "unknown", "judge": "unknown", "harness": "unknown", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown", "notes": "Model added from third-party aggregator (llm-stats.com); canonical setting not established." }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=Gemini Diffusion, slug=gemini-diffusion, provider=Google" }, { "model_id": "mistral-small-24b", "benchmark_id": "arena_hard", "score": 87.6, "reference_url": "https://llm-stats.com/benchmarks/arena-hard", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "OLMES (Allen AI)", "prompt_style": "default", "temperature": "default", "context": "default", "notes": "Per OLMo 2 paper Table 7 Instruct (instruction-tuned variant)." }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=Mistral Small 3 24B Instruct, slug=mistral-small-24b-instruct-2501, provider=Mistral AI" }, { "model_id": "deepseek-v2.5-0905", "benchmark_id": "arena_hard", "score": 76.2, "reference_url": "https://llm-stats.com/benchmarks/arena-hard", "reported_setting": { "snapshot": "0905" }, "matches_canonical": false, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=DeepSeek-V2.5, slug=deepseek-v2.5, provider=DeepSeek" }, { "model_id": "granite-3.3-8b-base", "benchmark_id": "arena_hard", "score": 57.6, "reference_url": "https://llm-stats.com/benchmarks/arena-hard", "reported_setting": { "mode": "unknown", "effort": "n/a", "tools": "unknown", "sampling": "unknown", "judge": "unknown", "harness": "unknown", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown", "notes": "Model added from third-party aggregator (llm-stats.com); canonical setting not established." }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=Granite 3.3 8B Base, slug=granite-3.3-8b-base, provider=IBM" }, { "model_id": "granite-3.3-8b-instruct", "benchmark_id": "arena_hard", "score": 57.6, "reference_url": "https://llm-stats.com/benchmarks/arena-hard", "reported_setting": { "mode": "unknown", "effort": "n/a", "tools": "unknown", "sampling": "unknown", "judge": "unknown", "harness": "unknown", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown", "notes": "Model added from third-party aggregator (llm-stats.com); canonical setting not established." }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=Granite 3.3 8B Instruct, slug=granite-3.3-8b-instruct, provider=IBM" }, { "model_id": "mistral-large-3", "benchmark_id": "arena_hard", "score": 66.7, "reference_url": "https://huggingface.co/mistralai/Mistral-Small-4-119B-2603/resolve/a11f36bebf709121056b1dbcc943d1c6afbe494d/images/image3.png?download=true", "reported_setting": { "mode": "non-thinking", "effort": "none", "tools": "none", "sampling": "pass@1; repeat count undisclosed", "judge": "Arena-Hard automatic judge", "harness": "Mistral official model-card evaluation", "prompt_style": "official/default or source does not state", "temperature": "source does not state", "context": "source does not state" }, "matches_canonical": false, "source_type": "official_model_card", "audit_status": "verified", "notes": "Displayed exactly: '66.7'. Research observation: small-image3.png:4:4:score:reported.", "candidates": [ { "score": 55.1, "reference_url": "https://llm-stats.com/benchmarks/arena-hard", "source_type": "third_party_aggregator", "reported_setting": {}, "notes": "llm-stats.com row: model=Mistral Large 3, slug=mistral-large-3-2509, provider=Mistral AI" } ] }, { "model_id": "qwen-2.5-7b", "benchmark_id": "arena_hard", "score": 52.0, "reference_url": "https://llm-stats.com/benchmarks/arena-hard", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "OLMES (Allen AI)", "prompt_style": "default", "temperature": "default", "context": "default", "notes": "Per OLMo 2 paper Table 7 Instruct (instruction-tuned variant)." }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=Qwen2.5 7B Instruct, slug=qwen-2.5-7b-instruct, provider=Alibaba Cloud / Qwen Team" }, { "model_id": "phi-3.5-moe-instruct", "benchmark_id": "arena_hard", "score": 37.9, "reference_url": "https://llm-stats.com/benchmarks/arena-hard", "reported_setting": { "mode": "unknown", "effort": "n/a", "tools": "unknown", "sampling": "unknown", "judge": "unknown", "harness": "unknown", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown", "notes": "Model added from third-party aggregator (llm-stats.com); canonical setting not established." }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=Phi-3.5-MoE-instruct, slug=phi-3.5-moe-instruct, provider=Microsoft" }, { "model_id": "phi-3.5-mini-instruct", "benchmark_id": "arena_hard", "score": 37.0, "reference_url": "https://llm-stats.com/benchmarks/arena-hard", "reported_setting": { "mode": "unknown", "effort": "n/a", "tools": "unknown", "sampling": "unknown", "judge": "unknown", "harness": "unknown", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown", "notes": "Model added from third-party aggregator (llm-stats.com); canonical setting not established." }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=Phi-3.5-mini-instruct, slug=phi-3.5-mini-instruct, provider=Microsoft" }, { "model_id": "phi-4-mini", "benchmark_id": "arena_hard", "score": 32.8, "reference_url": "https://llm-stats.com/benchmarks/arena-hard", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default", "context": "default (128K)", "notes": "Per Phi-4-mini-instruct HF model card." }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=Phi 4 Mini, slug=phi-4-mini, provider=Microsoft" }, { "model_id": "sarvam-30b", "benchmark_id": "beyond_aime", "score": 58.3, "reference_url": "https://llm-stats.com/benchmarks/beyond-aime", "reported_setting": { "mode": "unknown", "effort": "n/a", "tools": "unknown", "sampling": "unknown", "judge": "unknown", "harness": "unknown", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown", "notes": "Model added from third-party aggregator (llm-stats.com); canonical setting not established." }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=Sarvam-30B, slug=sarvam-30b, provider=Sarvam AI" }, { "model_id": "minimax-m2.5", "benchmark_id": "bfcl_v3_multiturn", "score": 76.8, "reference_url": "https://llm-stats.com/benchmarks/bfcl-v3-multiturn", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none (per-bench override)", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default", "context": "default", "notes": "Per MiniMax M2.5 model card." }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=MiniMax M2.5, slug=minimax-m2.5, provider=MiniMax" }, { "model_id": "llama-3.1-405b-instruct", "benchmark_id": "bfcl", "score": 88.5, "reference_url": "https://llm-stats.com/benchmarks/bfcl", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default", "context": "default", "notes": "Per DeepSeek V3 paper Table 6." }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=Llama 3.1 405B Instruct, slug=llama-3.1-405b-instruct, provider=Meta" }, { "model_id": "llama-3.1-70b", "benchmark_id": "bfcl", "score": 84.8, "reference_url": "https://llm-stats.com/benchmarks/bfcl", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "OLMES (Allen AI)", "prompt_style": "default", "temperature": "default", "context": "default", "notes": "Per OLMo 2 paper Table 7 Instruct (instruction-tuned variant)." }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=Llama 3.1 70B Instruct, slug=llama-3.1-70b-instruct, provider=Meta" }, { "model_id": "qwen3-235b", "benchmark_id": "bfcl", "score": 70.8, "reference_url": "https://llm-stats.com/benchmarks/bfcl", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official (Qwen-Agent / sglang)", "prompt_style": "default (enable_thinking=True)", "temperature": "default", "context": "default", "notes": "Per Qwen3 tech report (arxiv:2505.09388) Table 11 Thinking column. Output length 32768 (38912 for AIME)." }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=Qwen3 235B A22B, slug=qwen3-235b-a22b, provider=Alibaba Cloud / Qwen Team" }, { "model_id": "qwen3-32b", "benchmark_id": "bfcl", "score": 70.3, "reference_url": "https://llm-stats.com/benchmarks/bfcl", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official (Qwen-Agent / sglang)", "prompt_style": "default (enable_thinking=True)", "temperature": "default", "context": "default", "notes": "Per Qwen3 tech report (arxiv:2505.09388) Table 13/15/17/19 Thinking column." }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=Qwen3 32B, slug=qwen3-32b, provider=Alibaba Cloud / Qwen Team" }, { "model_id": "qwen3-30b-a3b", "benchmark_id": "bfcl", "score": 69.1, "reference_url": "https://llm-stats.com/benchmarks/bfcl", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official (Qwen-Agent / sglang)", "prompt_style": "default (enable_thinking=True)", "temperature": "default", "context": "default", "notes": "Per Qwen3 tech report (arxiv:2505.09388) Table 13/15/17/19 Thinking column." }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=Qwen3 30B A3B, slug=qwen3-30b-a3b, provider=Alibaba Cloud / Qwen Team" }, { "model_id": "nova-lite", "benchmark_id": "bfcl", "score": 66.6, "reference_url": "https://llm-stats.com/benchmarks/bfcl", "reported_setting": { "mode": "unknown", "effort": "n/a", "tools": "unknown", "sampling": "unknown", "judge": "unknown", "harness": "unknown", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown", "notes": "Model added from third-party aggregator (llm-stats.com); canonical setting not established." }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=Nova Lite, slug=nova-lite, provider=Amazon" }, { "model_id": "qwq-32b", "benchmark_id": "bfcl", "score": 66.4, "reference_url": "https://llm-stats.com/benchmarks/bfcl", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default", "context": "default (32K)", "notes": "Per QwQ-32B release blog (qwenlm.github.io/blog/qwq-32b/) + Qwen3 tech report Table 13 baseline column." }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=QwQ-32B, slug=qwq-32b, provider=Alibaba Cloud / Qwen Team" }, { "model_id": "nova-micro", "benchmark_id": "bfcl", "score": 56.2, "reference_url": "https://llm-stats.com/benchmarks/bfcl", "reported_setting": { "mode": "unknown", "effort": "n/a", "tools": "unknown", "sampling": "unknown", "judge": "unknown", "harness": "unknown", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown", "notes": "Model added from third-party aggregator (llm-stats.com); canonical setting not established." }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=Nova Micro, slug=nova-micro, provider=Amazon" }, { "model_id": "gemma-3-27b", "benchmark_id": "bigbench_extra_hard", "score": 19.3, "reference_url": "https://arxiv.org/abs/2607.02770", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "provider-reported", "judge": "benchmark-specified", "harness": "official Google evaluation", "prompt_style": "official", "temperature": "not disclosed", "context": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Gemma 4 Technical Report Table 5; Gemma 4 thinking unless explicitly stated, Gemma 3 27B non-thinking." }, { "model_id": "gemma-3-12b", "benchmark_id": "bigbench_extra_hard", "score": 16.3, "reference_url": "https://llm-stats.com/benchmarks/big-bench-extra-hard", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default", "context": "default", "notes": "Per Gemma 3 tech report Table 6 IT instruction-tuned column." }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=Gemma 3 12B, slug=gemma-3-12b-it, provider=Google" }, { "model_id": "gemini-diffusion", "benchmark_id": "bigbench_extra_hard", "score": 15.0, "reference_url": "https://llm-stats.com/benchmarks/big-bench-extra-hard", "reported_setting": { "mode": "unknown", "effort": "n/a", "tools": "unknown", "sampling": "unknown", "judge": "unknown", "harness": "unknown", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown", "notes": "Model added from third-party aggregator (llm-stats.com); canonical setting not established." }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=Gemini Diffusion, slug=gemini-diffusion, provider=Google" }, { "model_id": "gemma-3-1b-it", "benchmark_id": "bigbench_extra_hard", "score": 7.2, "reference_url": "https://llm-stats.com/benchmarks/big-bench-extra-hard", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default", "context": "32k", "notes": "Google Gemma 3 1B instruction-tuned." }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=Gemma 3 1B, slug=gemma-3-1b-it, provider=Google" }, { "model_id": "claude-3.5-sonnet", "benchmark_id": "bigbench_hard", "score": 93.1, "reference_url": "https://llm-stats.com/benchmarks/big-bench-hard", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default", "context": "default", "notes": "Per DeepSeek V3 paper Table 6." }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=Claude 3.5 Sonnet, slug=claude-3-5-sonnet-20241022, provider=Anthropic" }, { "model_id": "gemini-1.5-pro", "benchmark_id": "bigbench_hard", "score": 89.2, "reference_url": "https://llm-stats.com/benchmarks/big-bench-hard", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default", "context": "default", "notes": "Per Gemini 2.0 Flash Model Card (April 15 2025): non-thinking generation, default sampling. Same canonical setting applies to 1.5 / 2.0 family." }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=Gemini 1.5 Pro, slug=gemini-1.5-pro, provider=Google" }, { "model_id": "claude-3-opus", "benchmark_id": "bigbench_hard", "score": 86.8, "reference_url": "https://llm-stats.com/benchmarks/big-bench-hard", "reported_setting": { "mode": "unknown", "effort": "n/a", "tools": "unknown", "sampling": "unknown", "judge": "unknown", "harness": "unknown", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown", "notes": "Model added from third-party aggregator (llm-stats.com); canonical setting not established." }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=Claude 3 Opus, slug=claude-3-opus-20240229, provider=Anthropic" }, { "model_id": "gemma-3-12b", "benchmark_id": "bigbench_hard", "score": 85.7, "reference_url": "https://llm-stats.com/benchmarks/big-bench-hard", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default", "context": "default", "notes": "Per Gemma 3 tech report Table 6 IT instruction-tuned column." }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=Gemma 3 12B, slug=gemma-3-12b-it, provider=Google" }, { "model_id": "gemini-1.5-flash", "benchmark_id": "bigbench_hard", "score": 85.5, "reference_url": "https://llm-stats.com/benchmarks/big-bench-hard", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default", "context": "default", "notes": "Per Gemini 2.0 Flash Model Card (April 15 2025): non-thinking generation, default sampling. Same canonical setting applies to 1.5 / 2.0 family." }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=Gemini 1.5 Flash, slug=gemini-1.5-flash, provider=Google" }, { "model_id": "claude-3-sonnet", "benchmark_id": "bigbench_hard", "score": 82.9, "reference_url": "https://llm-stats.com/benchmarks/big-bench-hard", "reported_setting": { "mode": "unknown", "effort": "n/a", "tools": "unknown", "sampling": "unknown", "judge": "unknown", "harness": "unknown", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown", "notes": "Model added from third-party aggregator (llm-stats.com); canonical setting not established." }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=Claude 3 Sonnet, slug=claude-3-sonnet-20240229, provider=Anthropic" }, { "model_id": "phi-3.5-moe-instruct", "benchmark_id": "bigbench_hard", "score": 79.1, "reference_url": "https://llm-stats.com/benchmarks/big-bench-hard", "reported_setting": { "mode": "unknown", "effort": "n/a", "tools": "unknown", "sampling": "unknown", "judge": "unknown", "harness": "unknown", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown", "notes": "Model added from third-party aggregator (llm-stats.com); canonical setting not established." }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=Phi-3.5-MoE-instruct, slug=phi-3.5-moe-instruct, provider=Microsoft" }, { "model_id": "claude-3-haiku", "benchmark_id": "bigbench_hard", "score": 73.7, "reference_url": "https://llm-stats.com/benchmarks/big-bench-hard", "reported_setting": { "mode": "unknown", "effort": "n/a", "tools": "unknown", "sampling": "unknown", "judge": "unknown", "harness": "unknown", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown", "notes": "Model added from third-party aggregator (llm-stats.com); canonical setting not established." }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=Claude 3 Haiku, slug=claude-3-haiku-20240307, provider=Anthropic" }, { "model_id": "granite-3.3-8b-instruct", "benchmark_id": "bigbench_hard", "score": 69.1, "reference_url": "https://llm-stats.com/benchmarks/big-bench-hard", "reported_setting": { "mode": "unknown", "effort": "n/a", "tools": "unknown", "sampling": "unknown", "judge": "unknown", "harness": "unknown", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown", "notes": "Model added from third-party aggregator (llm-stats.com); canonical setting not established." }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=Granite 3.3 8B Instruct, slug=granite-3.3-8b-instruct, provider=IBM" }, { "model_id": "granite-3.3-8b-base", "benchmark_id": "bigbench_hard", "score": 69.1, "reference_url": "https://llm-stats.com/benchmarks/big-bench-hard", "reported_setting": { "mode": "unknown", "effort": "n/a", "tools": "unknown", "sampling": "unknown", "judge": "unknown", "harness": "unknown", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown", "notes": "Model added from third-party aggregator (llm-stats.com); canonical setting not established." }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=Granite 3.3 8B Base, slug=granite-3.3-8b-base, provider=IBM" }, { "model_id": "phi-3.5-mini-instruct", "benchmark_id": "bigbench_hard", "score": 69.0, "reference_url": "https://llm-stats.com/benchmarks/big-bench-hard", "reported_setting": { "mode": "unknown", "effort": "n/a", "tools": "unknown", "sampling": "unknown", "judge": "unknown", "harness": "unknown", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown", "notes": "Model added from third-party aggregator (llm-stats.com); canonical setting not established." }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=Phi-3.5-mini-instruct, slug=phi-3.5-mini-instruct, provider=Microsoft" }, { "model_id": "gemma-3-1b-it", "benchmark_id": "bigbench_hard", "score": 39.1, "reference_url": "https://llm-stats.com/benchmarks/big-bench-hard", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default", "context": "32k", "notes": "Google Gemma 3 1B instruction-tuned." }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=Gemma 3 1B, slug=gemma-3-1b-it, provider=Google" }, { "model_id": "gemini-diffusion", "benchmark_id": "bigcodebench", "score": 45.4, "reference_url": "https://llm-stats.com/benchmarks/bigcodebench", "reported_setting": { "mode": "unknown", "effort": "n/a", "tools": "unknown", "sampling": "unknown", "judge": "unknown", "harness": "unknown", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown", "notes": "Model added from third-party aggregator (llm-stats.com); canonical setting not established." }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=Gemini Diffusion, slug=gemini-diffusion, provider=Google" }, { "model_id": "gemma-3-1b-it", "benchmark_id": "bird_sql", "score": 6.4, "reference_url": "https://llm-stats.com/benchmarks/bird-sql-(dev)", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default", "context": "32k", "notes": "Google Gemma 3 1B instruction-tuned." }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=Gemma 3 1B, slug=gemma-3-1b-it, provider=Google" }, { "model_id": "gpt-5.2", "benchmark_id": "browsecomp_long_context_128k", "score": 92.0, "reference_url": "https://llm-stats.com/benchmarks/browsecomp-long-128k", "reported_setting": { "mode": "thinking", "effort": "xhigh", "tools": "none (per-bench override)", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default (with prompt adjustment for τ²-bench)", "temperature": "default", "context": "default", "notes": "Per OpenAI gpt-5.2 blog: reasoning effort=xhigh, research environment." }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=GPT-5.2, slug=gpt-5.2-2025-12-11, provider=OpenAI" }, { "model_id": "gpt-5.2", "benchmark_id": "browsecomp_long_context_256k", "score": 89.8, "reference_url": "https://llm-stats.com/benchmarks/browsecomp-long-256k", "reported_setting": { "mode": "thinking", "effort": "xhigh", "tools": "none (per-bench override)", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default (with prompt adjustment for τ²-bench)", "temperature": "default", "context": "default", "notes": "Per OpenAI gpt-5.2 blog: reasoning effort=xhigh, research environment." }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=GPT-5.2, slug=gpt-5.2-2025-12-11, provider=OpenAI" }, { "model_id": "gpt-5", "benchmark_id": "browsecomp_long_context_256k", "score": 88.8, "reference_url": "https://llm-stats.com/benchmarks/browsecomp-long-256k", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none (per-bench override)", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default", "context": "default", "notes": "Per OpenAI GPT-5 dev blog. high reasoning effort." }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=GPT-5, slug=gpt-5-2025-08-07, provider=OpenAI" }, { "model_id": "deepseek-r1", "benchmark_id": "browsecomp_zh", "score": 35.7, "reference_url": "https://llm-stats.com/benchmarks/browsecomp-zh", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none (per-bench override)", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default", "context": "default", "notes": "Per DeepSeek R1-0528 model card." }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=DeepSeek-R1-0528, slug=deepseek-r1-0528, provider=DeepSeek" }, { "model_id": "minimax-m2.5", "benchmark_id": "browsecomp", "score": 76.3, "reference_url": "https://llm-stats.com/benchmarks/browsecomp", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none (per-bench override)", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default", "context": "default", "notes": "Per MiniMax M2.5 model card." }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=MiniMax M2.5, slug=minimax-m2.5, provider=MiniMax" }, { "model_id": "minimax-m2.1", "benchmark_id": "browsecomp", "score": 62.0, "reference_url": "https://llm-stats.com/benchmarks/browsecomp", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none (per-bench override)", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default", "context": "default", "notes": "Per MiniMax M2.5 model card (referenced as predecessor)." }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=MiniMax M2.1, slug=minimax-m2.1, provider=MiniMax" }, { "model_id": "o4-mini-high", "benchmark_id": "browsecomp", "score": 51.5, "reference_url": "https://llm-stats.com/benchmarks/browsecomp", "reported_setting": { "effort": "default (o4-mini base, not high)" }, "matches_canonical": false, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=o4-mini, slug=o4-mini, provider=OpenAI" }, { "model_id": "sarvam-30b", "benchmark_id": "browsecomp", "score": 35.5, "reference_url": "https://llm-stats.com/benchmarks/browsecomp", "reported_setting": { "mode": "unknown", "effort": "n/a", "tools": "unknown", "sampling": "unknown", "judge": "unknown", "harness": "unknown", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown", "notes": "Model added from third-party aggregator (llm-stats.com); canonical setting not established." }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=Sarvam-30B, slug=sarvam-30b, provider=Sarvam AI" }, { "model_id": "deepseek-r1", "benchmark_id": "browsecomp", "score": 8.9, "reference_url": "https://llm-stats.com/benchmarks/browsecomp", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none (per-bench override)", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default", "context": "default", "notes": "Per DeepSeek R1-0528 model card." }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=DeepSeek-R1-0528, slug=deepseek-r1-0528, provider=DeepSeek" }, { "model_id": "qwen3.6-plus", "benchmark_id": "c_eval", "score": 93.3, "reference_url": "https://llm-stats.com/benchmarks/c-eval", "reported_setting": {}, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=Qwen3.6 Plus, slug=qwen3.6-plus, provider=Alibaba Cloud / Qwen Team" }, { "model_id": "qwen2-72b-instruct", "benchmark_id": "c_eval", "score": 83.8, "reference_url": "https://llm-stats.com/benchmarks/c-eval", "reported_setting": { "mode": "unknown", "effort": "n/a", "tools": "unknown", "sampling": "unknown", "judge": "unknown", "harness": "unknown", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown", "notes": "Model added from third-party aggregator (llm-stats.com); canonical setting not established." }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=Qwen2 72B Instruct, slug=qwen2-72b-instruct, provider=Alibaba Cloud / Qwen Team" }, { "model_id": "qwen2-7b-instruct", "benchmark_id": "c_eval", "score": 77.2, "reference_url": "https://llm-stats.com/benchmarks/c-eval", "reported_setting": { "mode": "unknown", "effort": "n/a", "tools": "unknown", "sampling": "unknown", "judge": "unknown", "harness": "unknown", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown", "notes": "Model added from third-party aggregator (llm-stats.com); canonical setting not established." }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=Qwen2 7B Instruct, slug=qwen2-7b-instruct, provider=Alibaba Cloud / Qwen Team" }, { "model_id": "claude-3.5-sonnet", "benchmark_id": "chartqa", "score": 90.8, "reference_url": "https://llm-stats.com/benchmarks/chartqa", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default", "context": "default", "notes": "Per DeepSeek V3 paper Table 6." }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=Claude 3.5 Sonnet, slug=claude-3-5-sonnet-20241022, provider=Anthropic" }, { "model_id": "llama-4-scout", "benchmark_id": "chartqa", "score": 88.8, "reference_url": "https://llm-stats.com/benchmarks/chartqa", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official-meta", "prompt_style": "zeroshot", "temperature": "0.0", "context": "default", "multimodal_input": true }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=Llama 4 Scout, slug=llama-4-scout, provider=Meta" }, { "model_id": "nova-lite", "benchmark_id": "chartqa", "score": 86.8, "reference_url": "https://llm-stats.com/benchmarks/chartqa", "reported_setting": { "mode": "unknown", "effort": "n/a", "tools": "unknown", "sampling": "unknown", "judge": "unknown", "harness": "unknown", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown", "notes": "Model added from third-party aggregator (llm-stats.com); canonical setting not established." }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=Nova Lite, slug=nova-lite, provider=Amazon" }, { "model_id": "qwen2.5-omni-7b", "benchmark_id": "chartqa", "score": 85.3, "reference_url": "https://llm-stats.com/benchmarks/chartqa", "reported_setting": { "mode": "unknown", "effort": "n/a", "tools": "unknown", "sampling": "unknown", "judge": "unknown", "harness": "unknown", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown", "notes": "Model added from third-party aggregator (llm-stats.com); canonical setting not established." }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=Qwen2.5-Omni-7B, slug=qwen2.5-omni-7b, provider=Alibaba Cloud / Qwen Team" }, { "model_id": "pixtral-12b", "benchmark_id": "chartqa", "score": 81.8, "reference_url": "https://llm-stats.com/benchmarks/chartqa", "reported_setting": { "mode": "unknown", "effort": "n/a", "tools": "unknown", "sampling": "unknown", "judge": "unknown", "harness": "unknown", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown", "notes": "Model added from third-party aggregator (llm-stats.com); canonical setting not established." }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=Pixtral-12B, slug=pixtral-12b-2409, provider=Mistral AI" }, { "model_id": "gemma-3-27b", "benchmark_id": "chartqa", "score": 78.0, "reference_url": "https://llm-stats.com/benchmarks/chartqa", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default", "context": "default (128K)", "notes": "Per Gemma 3 tech report (arxiv:2503.19786) Table 6 IT column for 27B." }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=Gemma 3 27B, slug=gemma-3-27b-it, provider=Google" }, { "model_id": "gemma-3-12b", "benchmark_id": "chartqa", "score": 75.7, "reference_url": "https://llm-stats.com/benchmarks/chartqa", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default", "context": "default", "notes": "Per Gemma 3 tech report Table 6 IT instruction-tuned column." }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=Gemma 3 12B, slug=gemma-3-12b-it, provider=Google" }, { "model_id": "qwen2-72b-instruct", "benchmark_id": "cmmlu", "score": 90.1, "reference_url": "https://llm-stats.com/benchmarks/cmmlu", "reported_setting": { "mode": "unknown", "effort": "n/a", "tools": "unknown", "sampling": "unknown", "judge": "unknown", "harness": "unknown", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown", "notes": "Model added from third-party aggregator (llm-stats.com); canonical setting not established." }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=Qwen2 72B Instruct, slug=qwen2-72b-instruct, provider=Alibaba Cloud / Qwen Team" }, { "model_id": "longcat-flash-chat", "benchmark_id": "cmmlu", "score": 84.3, "reference_url": "https://llm-stats.com/benchmarks/cmmlu", "reported_setting": { "mode": "unknown", "effort": "n/a", "tools": "unknown", "sampling": "unknown", "judge": "unknown", "harness": "unknown", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown", "notes": "Model added from third-party aggregator (llm-stats.com); canonical setting not established." }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=LongCat-Flash-Chat, slug=longcat-flash-chat, provider=Meituan" }, { "model_id": "minicpm-sala", "benchmark_id": "cmmlu", "score": 81.6, "reference_url": "https://llm-stats.com/benchmarks/cmmlu", "reported_setting": { "mode": "unknown", "effort": "n/a", "tools": "unknown", "sampling": "unknown", "judge": "unknown", "harness": "unknown", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown", "notes": "Model added from third-party aggregator (llm-stats.com); canonical setting not established." }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=MiniCPM-SALA, slug=minicpm-sala, provider=OpenBMB" }, { "model_id": "deepseek-v4-pro", "benchmark_id": "codeforces_avg8", "score": 100.0, "reference_url": "https://llm-stats.com/benchmarks/codeforces", "reported_setting": { "effort": "max" }, "matches_canonical": false, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=DeepSeek-V4-Pro-Max, slug=deepseek-v4-pro-max, provider=DeepSeek" }, { "model_id": "deepseek-v4-flash", "benchmark_id": "codeforces_avg8", "score": 100.0, "reference_url": "https://llm-stats.com/benchmarks/codeforces", "reported_setting": { "effort": "max" }, "matches_canonical": false, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=DeepSeek-V4-Flash-Max, slug=deepseek-v4-flash-max, provider=DeepSeek" }, { "model_id": "deepseek-v3.2-speciale", "benchmark_id": "codeforces_avg8", "score": 90.0, "reference_url": "https://llm-stats.com/benchmarks/codeforces", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "none (per-bench override)", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "1.0", "context": "128K", "notes": "DS V3.2-Speciale: high-compute reasoning variant of V3.2. Per V3.2 tech report Table 3." }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=DeepSeek-V3.2-Speciale, slug=deepseek-v3.2-speciale, provider=DeepSeek" }, { "model_id": "gpt-oss-120b", "benchmark_id": "codeforces_avg8", "score": 82.1, "reference_url": "https://llm-stats.com/benchmarks/codeforces", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "with tools (per-bench override; harmony chat format default)", "sampling": "pass@1", "judge": "rule-based", "harness": "official (harmony chat format)", "prompt_style": "default", "temperature": "default", "context": "default (128K)", "notes": "Per OpenAI gpt-oss model card (arxiv:2508.10925) Table 3, high reasoning effort. Default with-tools for agentic; AIME/GPQA also has no-tools variant in same table." }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=GPT OSS 120B, slug=gpt-oss-120b, provider=OpenAI" }, { "model_id": "deepseek-v3.2", "benchmark_id": "codeforces_avg8", "score": 79.5, "reference_url": "https://llm-stats.com/benchmarks/codeforces", "reported_setting": { "mode": "thinking (DeepSeek-Reasoner endpoint)" }, "matches_canonical": false, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=DeepSeek-V3.2 (Thinking), slug=deepseek-reasoner, provider=DeepSeek" }, { "model_id": "gpt-oss-20b", "benchmark_id": "codeforces_avg8", "score": 74.3, "reference_url": "https://llm-stats.com/benchmarks/codeforces", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "with tools (per-bench override; harmony chat format default)", "sampling": "pass@1", "judge": "rule-based", "harness": "official (harmony chat format)", "prompt_style": "default", "temperature": "default", "context": "default (128K)", "notes": "Per OpenAI gpt-oss model card (arxiv:2508.10925) Table 3, high reasoning effort. Default with-tools for agentic; AIME/GPQA also has no-tools variant in same table." }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=GPT OSS 20B, slug=gpt-oss-20b, provider=OpenAI" }, { "model_id": "qwen3-32b", "benchmark_id": "codeforces_avg8", "score": 65.9, "reference_url": "https://llm-stats.com/benchmarks/codeforces", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official (Qwen-Agent / sglang)", "prompt_style": "default (enable_thinking=True)", "temperature": "default", "context": "default", "notes": "Per Qwen3 tech report (arxiv:2505.09388) Table 13/15/17/19 Thinking column." }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=Qwen3 32B, slug=qwen3-32b, provider=Alibaba Cloud / Qwen Team" }, { "model_id": "deepseek-v4-pro", "benchmark_id": "codeforces_pass8", "score": 100.0, "reference_url": "https://llm-stats.com/benchmarks/codeforces", "reported_setting": { "effort": "max" }, "matches_canonical": false, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=DeepSeek-V4-Pro-Max, slug=deepseek-v4-pro-max, provider=DeepSeek" }, { "model_id": "deepseek-v4-flash", "benchmark_id": "codeforces_pass8", "score": 100.0, "reference_url": "https://llm-stats.com/benchmarks/codeforces", "reported_setting": { "effort": "max" }, "matches_canonical": false, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=DeepSeek-V4-Flash-Max, slug=deepseek-v4-flash-max, provider=DeepSeek" }, { "model_id": "deepseek-v3.2-speciale", "benchmark_id": "codeforces_pass8", "score": 90.0, "reference_url": "https://llm-stats.com/benchmarks/codeforces", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "none (per-bench override)", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "1.0", "context": "128K", "notes": "DS V3.2-Speciale: high-compute reasoning variant of V3.2. Per V3.2 tech report Table 3." }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=DeepSeek-V3.2-Speciale, slug=deepseek-v3.2-speciale, provider=DeepSeek" }, { "model_id": "gpt-oss-120b", "benchmark_id": "codeforces_pass8", "score": 82.1, "reference_url": "https://llm-stats.com/benchmarks/codeforces", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "with tools (per-bench override; harmony chat format default)", "sampling": "pass@1", "judge": "rule-based", "harness": "official (harmony chat format)", "prompt_style": "default", "temperature": "default", "context": "default (128K)", "notes": "Per OpenAI gpt-oss model card (arxiv:2508.10925) Table 3, high reasoning effort. Default with-tools for agentic; AIME/GPQA also has no-tools variant in same table." }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=GPT OSS 120B, slug=gpt-oss-120b, provider=OpenAI" }, { "model_id": "deepseek-v3.2", "benchmark_id": "codeforces_pass8", "score": 79.5, "reference_url": "https://llm-stats.com/benchmarks/codeforces", "reported_setting": { "mode": "thinking (DeepSeek-Reasoner endpoint)" }, "matches_canonical": false, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=DeepSeek-V3.2 (Thinking), slug=deepseek-reasoner, provider=DeepSeek" }, { "model_id": "gpt-oss-20b", "benchmark_id": "codeforces_pass8", "score": 74.3, "reference_url": "https://llm-stats.com/benchmarks/codeforces", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "with tools (per-bench override; harmony chat format default)", "sampling": "pass@1", "judge": "rule-based", "harness": "official (harmony chat format)", "prompt_style": "default", "temperature": "default", "context": "default (128K)", "notes": "Per OpenAI gpt-oss model card (arxiv:2508.10925) Table 3, high reasoning effort. Default with-tools for agentic; AIME/GPQA also has no-tools variant in same table." }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=GPT OSS 20B, slug=gpt-oss-20b, provider=OpenAI" }, { "model_id": "qwen3-32b", "benchmark_id": "codeforces_pass8", "score": 65.9, "reference_url": "https://llm-stats.com/benchmarks/codeforces", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official (Qwen-Agent / sglang)", "prompt_style": "default (enable_thinking=True)", "temperature": "default", "context": "default", "notes": "Per Qwen3 tech report (arxiv:2505.09388) Table 13/15/17/19 Thinking column." }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=Qwen3 32B, slug=qwen3-32b, provider=Alibaba Cloud / Qwen Team" }, { "model_id": "qwen3.6-plus", "benchmark_id": "countbench", "score": 97.6, "reference_url": "https://llm-stats.com/benchmarks/countbench", "reported_setting": {}, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=Qwen3.6 Plus, slug=qwen3.6-plus, provider=Alibaba Cloud / Qwen Team" }, { "model_id": "qwen3-235b", "benchmark_id": "creative_writing_v3", "score": 87.5, "reference_url": "https://llm-stats.com/benchmarks/creative-writing-v3", "reported_setting": { "mode": "instruct", "snapshot": "2507" }, "matches_canonical": false, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=Qwen3-235B-A22B-Instruct-2507, slug=qwen3-235b-a22b-instruct-2507, provider=Alibaba Cloud / Qwen Team" }, { "model_id": "claude-mythos", "benchmark_id": "cybench", "score": 100.0, "reference_url": "https://llm-stats.com/benchmarks/cybench", "reported_setting": { "mode": "thinking", "effort": "adaptive thinking at max effort", "tools": "none (per-bench override)", "sampling": "pass@1 (avg 5 trials)", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default", "context": "up to 1M", "notes": "Per Mythos System Card Table 6.3.A: standard config is adaptive thinking max effort, default sampling (temperature, top_p), avg 5 trials, context up to 1M." }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=Claude Mythos Preview, slug=claude-mythos-preview, provider=Anthropic" }, { "model_id": "grok-4.1", "benchmark_id": "cybench", "score": 39.0, "reference_url": "https://llm-stats.com/benchmarks/cybench", "reported_setting": { "effort": "thinking" }, "matches_canonical": false, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=Grok-4.1 Thinking, slug=grok-4.1-thinking-2025-11-17, provider=xAI" }, { "model_id": "claude-3.5-sonnet", "benchmark_id": "docvqa", "score": 95.2, "reference_url": "https://llm-stats.com/benchmarks/docvqa", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default", "context": "default", "notes": "Per DeepSeek V3 paper Table 6." }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=Claude 3.5 Sonnet, slug=claude-3-5-sonnet-20241022, provider=Anthropic" }, { "model_id": "qwen2.5-omni-7b", "benchmark_id": "docvqa", "score": 95.2, "reference_url": "https://llm-stats.com/benchmarks/docvqa", "reported_setting": { "mode": "unknown", "effort": "n/a", "tools": "unknown", "sampling": "unknown", "judge": "unknown", "harness": "unknown", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown", "notes": "Model added from third-party aggregator (llm-stats.com); canonical setting not established." }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=Qwen2.5-Omni-7B, slug=qwen2.5-omni-7b, provider=Alibaba Cloud / Qwen Team" }, { "model_id": "llama-4-scout", "benchmark_id": "docvqa", "score": 94.4, "reference_url": "https://llm-stats.com/benchmarks/docvqa", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official-meta", "prompt_style": "zeroshot", "temperature": "0.0", "context": "default", "multimodal_input": true }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=Llama 4 Scout, slug=llama-4-scout, provider=Meta" }, { "model_id": "grok-2", "benchmark_id": "docvqa", "score": 93.6, "reference_url": "https://llm-stats.com/benchmarks/docvqa", "reported_setting": { "mode": "unknown", "effort": "n/a", "tools": "unknown", "sampling": "unknown", "judge": "unknown", "harness": "unknown", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown", "notes": "Model added from third-party aggregator (llm-stats.com); canonical setting not established." }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=Grok-2, slug=grok-2, provider=xAI" }, { "model_id": "grok-2-mini", "benchmark_id": "docvqa", "score": 93.2, "reference_url": "https://llm-stats.com/benchmarks/docvqa", "reported_setting": { "mode": "unknown", "effort": "n/a", "tools": "unknown", "sampling": "unknown", "judge": "unknown", "harness": "unknown", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown", "notes": "Model added from third-party aggregator (llm-stats.com); canonical setting not established." }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=Grok-2 mini, slug=grok-2-mini, provider=xAI" }, { "model_id": "nova-lite", "benchmark_id": "docvqa", "score": 92.4, "reference_url": "https://llm-stats.com/benchmarks/docvqa", "reported_setting": { "mode": "unknown", "effort": "n/a", "tools": "unknown", "sampling": "unknown", "judge": "unknown", "harness": "unknown", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown", "notes": "Model added from third-party aggregator (llm-stats.com); canonical setting not established." }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=Nova Lite, slug=nova-lite, provider=Amazon" }, { "model_id": "pixtral-12b", "benchmark_id": "docvqa", "score": 90.7, "reference_url": "https://llm-stats.com/benchmarks/docvqa", "reported_setting": { "mode": "unknown", "effort": "n/a", "tools": "unknown", "sampling": "unknown", "judge": "unknown", "harness": "unknown", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown", "notes": "Model added from third-party aggregator (llm-stats.com); canonical setting not established." }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=Pixtral-12B, slug=pixtral-12b-2409, provider=Mistral AI" }, { "model_id": "gemma-3-12b", "benchmark_id": "docvqa", "score": 87.1, "reference_url": "https://llm-stats.com/benchmarks/docvqa", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default", "context": "default", "notes": "Per Gemma 3 tech report Table 6 IT instruction-tuned column." }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=Gemma 3 12B, slug=gemma-3-12b-it, provider=Google" }, { "model_id": "gemma-3-27b", "benchmark_id": "docvqa", "score": 86.6, "reference_url": "https://llm-stats.com/benchmarks/docvqa", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default", "context": "default (128K)", "notes": "Per Gemma 3 tech report (arxiv:2503.19786) Table 6 IT column for 27B." }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=Gemma 3 27B, slug=gemma-3-27b-it, provider=Google" }, { "model_id": "grok-1.5", "benchmark_id": "docvqa", "score": 85.6, "reference_url": "https://llm-stats.com/benchmarks/docvqa", "reported_setting": { "mode": "unknown", "effort": "n/a", "tools": "unknown", "sampling": "unknown", "judge": "unknown", "harness": "unknown", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown", "notes": "Model added from third-party aggregator (llm-stats.com); canonical setting not established." }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=Grok-1.5, slug=grok-1.5, provider=xAI" }, { "model_id": "gpt-4-turbo", "benchmark_id": "drop", "score": 86.0, "reference_url": "https://llm-stats.com/benchmarks/drop", "reported_setting": { "mode": "unknown", "effort": "n/a", "tools": "unknown", "sampling": "unknown", "judge": "unknown", "harness": "unknown", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown", "notes": "Model added from third-party aggregator (llm-stats.com); canonical setting not established." }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=GPT-4 Turbo, slug=gpt-4-turbo-2024-04-09, provider=OpenAI" }, { "model_id": "gpt-4o", "benchmark_id": "drop", "score": 83.4, "reference_url": "https://llm-stats.com/benchmarks/drop", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none (per-bench override)", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default", "context": "default", "notes": "Per OpenAI GPT-4.1 blog table." }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=GPT-4o, slug=gpt-4o-2024-05-13, provider=OpenAI" }, { "model_id": "claude-3.5-haiku", "benchmark_id": "drop", "score": 83.1, "reference_url": "https://llm-stats.com/benchmarks/drop", "reported_setting": { "mode": "unknown", "effort": "n/a", "tools": "unknown", "sampling": "unknown", "judge": "unknown", "harness": "unknown", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown", "notes": "Model added from third-party aggregator (llm-stats.com); canonical setting not established." }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=Claude 3.5 Haiku, slug=claude-3-5-haiku-20241022, provider=Anthropic" }, { "model_id": "claude-3-opus", "benchmark_id": "drop", "score": 83.1, "reference_url": "https://llm-stats.com/benchmarks/drop", "reported_setting": { "mode": "unknown", "effort": "n/a", "tools": "unknown", "sampling": "unknown", "judge": "unknown", "harness": "unknown", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown", "notes": "Model added from third-party aggregator (llm-stats.com); canonical setting not established." }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=Claude 3 Opus, slug=claude-3-opus-20240229, provider=Anthropic" }, { "model_id": "nova-lite", "benchmark_id": "drop", "score": 80.2, "reference_url": "https://llm-stats.com/benchmarks/drop", "reported_setting": { "mode": "unknown", "effort": "n/a", "tools": "unknown", "sampling": "unknown", "judge": "unknown", "harness": "unknown", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown", "notes": "Model added from third-party aggregator (llm-stats.com); canonical setting not established." }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=Nova Lite, slug=nova-lite, provider=Amazon" }, { "model_id": "nova-micro", "benchmark_id": "drop", "score": 79.3, "reference_url": "https://llm-stats.com/benchmarks/drop", "reported_setting": { "mode": "unknown", "effort": "n/a", "tools": "unknown", "sampling": "unknown", "judge": "unknown", "harness": "unknown", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown", "notes": "Model added from third-party aggregator (llm-stats.com); canonical setting not established." }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=Nova Micro, slug=nova-micro, provider=Amazon" }, { "model_id": "longcat-flash-chat", "benchmark_id": "drop", "score": 79.1, "reference_url": "https://llm-stats.com/benchmarks/drop", "reported_setting": { "mode": "unknown", "effort": "n/a", "tools": "unknown", "sampling": "unknown", "judge": "unknown", "harness": "unknown", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown", "notes": "Model added from third-party aggregator (llm-stats.com); canonical setting not established." }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=LongCat-Flash-Chat, slug=longcat-flash-chat, provider=Meituan" }, { "model_id": "claude-3-sonnet", "benchmark_id": "drop", "score": 78.9, "reference_url": "https://llm-stats.com/benchmarks/drop", "reported_setting": { "mode": "unknown", "effort": "n/a", "tools": "unknown", "sampling": "unknown", "judge": "unknown", "harness": "unknown", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown", "notes": "Model added from third-party aggregator (llm-stats.com); canonical setting not established." }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=Claude 3 Sonnet, slug=claude-3-sonnet-20240229, provider=Anthropic" }, { "model_id": "claude-3-haiku", "benchmark_id": "drop", "score": 78.4, "reference_url": "https://llm-stats.com/benchmarks/drop", "reported_setting": { "mode": "unknown", "effort": "n/a", "tools": "unknown", "sampling": "unknown", "judge": "unknown", "harness": "unknown", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown", "notes": "Model added from third-party aggregator (llm-stats.com); canonical setting not established." }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=Claude 3 Haiku, slug=claude-3-haiku-20240307, provider=Anthropic" }, { "model_id": "gemini-1.5-pro", "benchmark_id": "drop", "score": 74.9, "reference_url": "https://llm-stats.com/benchmarks/drop", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default", "context": "default", "notes": "Per Gemini 2.0 Flash Model Card (April 15 2025): non-thinking generation, default sampling. Same canonical setting applies to 1.5 / 2.0 family." }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=Gemini 1.5 Pro, slug=gemini-1.5-pro, provider=Google" }, { "model_id": "granite-3.3-8b-instruct", "benchmark_id": "drop", "score": 59.4, "reference_url": "https://llm-stats.com/benchmarks/drop", "reported_setting": { "mode": "unknown", "effort": "n/a", "tools": "unknown", "sampling": "unknown", "judge": "unknown", "harness": "unknown", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown", "notes": "Model added from third-party aggregator (llm-stats.com); canonical setting not established." }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=Granite 3.3 8B Instruct, slug=granite-3.3-8b-instruct, provider=IBM" }, { "model_id": "granite-3.3-8b-base", "benchmark_id": "drop", "score": 36.1, "reference_url": "https://llm-stats.com/benchmarks/drop", "reported_setting": { "mode": "unknown", "effort": "n/a", "tools": "unknown", "sampling": "unknown", "judge": "unknown", "harness": "unknown", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown", "notes": "Model added from third-party aggregator (llm-stats.com); canonical setting not established." }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=Granite 3.3 8B Base, slug=granite-3.3-8b-base, provider=IBM" }, { "model_id": "qwen3.6-plus", "benchmark_id": "dynamath", "score": 88.0, "reference_url": "https://llm-stats.com/benchmarks/dynamath", "reported_setting": {}, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=Qwen3.6 Plus, slug=qwen3.6-plus, provider=Alibaba Cloud / Qwen Team" }, { "model_id": "gpt-4o", "benchmark_id": "egoschema", "score": 72.2, "reference_url": "https://llm-stats.com/benchmarks/egoschema", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none (per-bench override)", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default", "context": "default", "notes": "Per OpenAI GPT-4.1 blog table." }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=GPT-4o, slug=gpt-4o-2024-08-06, provider=OpenAI" }, { "model_id": "nova-lite", "benchmark_id": "egoschema", "score": 71.4, "reference_url": "https://llm-stats.com/benchmarks/egoschema", "reported_setting": { "mode": "unknown", "effort": "n/a", "tools": "unknown", "sampling": "unknown", "judge": "unknown", "harness": "unknown", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown", "notes": "Model added from third-party aggregator (llm-stats.com); canonical setting not established." }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=Nova Lite, slug=nova-lite, provider=Amazon" }, { "model_id": "qwen2.5-omni-7b", "benchmark_id": "egoschema", "score": 68.6, "reference_url": "https://llm-stats.com/benchmarks/egoschema", "reported_setting": { "mode": "unknown", "effort": "n/a", "tools": "unknown", "sampling": "unknown", "judge": "unknown", "harness": "unknown", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown", "notes": "Model added from third-party aggregator (llm-stats.com); canonical setting not established." }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=Qwen2.5-Omni-7B, slug=qwen2.5-omni-7b, provider=Alibaba Cloud / Qwen Team" }, { "model_id": "qwen3.6-plus", "benchmark_id": "erqa", "score": 65.7, "reference_url": "https://llm-stats.com/benchmarks/erqa", "reported_setting": {}, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=Qwen3.6 Plus, slug=qwen3.6-plus, provider=Alibaba Cloud / Qwen Team" }, { "model_id": "gpt-4o", "benchmark_id": "erqa", "score": 35.2, "reference_url": "https://llm-stats.com/benchmarks/erqa", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none (per-bench override)", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default", "context": "default", "notes": "Per OpenAI GPT-4.1 blog table." }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=GPT-4o, slug=gpt-4o-2024-08-06, provider=OpenAI" }, { "model_id": "gemini-3-pro", "benchmark_id": "facts_grounding", "score": 70.5, "reference_url": "https://llm-stats.com/benchmarks/facts-grounding", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default", "context": "default", "notes": "Per Google DeepMind /models/gemini/pro/ table: Gemini 3 Pro Thinking (High), no tools by default." }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=Gemini 3 Pro, slug=gemini-3-pro-preview, provider=Google" }, { "model_id": "gemini-3-flash", "benchmark_id": "facts_grounding", "score": 61.9, "reference_url": "https://llm-stats.com/benchmarks/facts-grounding", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default", "context": "default", "notes": "Per Google DeepMind /models/gemini/flash/ table: Gemini 3 Flash Thinking, no tools by default." }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=Gemini 3 Flash, slug=gemini-3-flash-preview, provider=Google" }, { "model_id": "gemma-3-1b-it", "benchmark_id": "facts_grounding", "score": 36.4, "reference_url": "https://llm-stats.com/benchmarks/facts-grounding", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default", "context": "32k", "notes": "Google Gemma 3 1B instruction-tuned." }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=Gemma 3 1B, slug=gemma-3-1b-it, provider=Google" }, { "model_id": "grok-4.1", "benchmark_id": "factscore", "score": 97.0, "reference_url": "https://llm-stats.com/benchmarks/factscore", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none (per-bench override)", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default", "context": "default", "notes": "Per xAI Grok 4.1 blog: standard chart values. Hallucination/FActScore use with-web-search but those are not stored in BP." }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=Grok-4.1, slug=grok-4.1-2025-11-17, provider=xAI" }, { "model_id": "gemini-1.5-flash-8b", "benchmark_id": "fleurs", "score": 86.4, "reference_url": "https://llm-stats.com/benchmarks/fleurs", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default", "context": "1M", "notes": "Per Command A paper baseline." }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=Gemini 1.5 Flash 8B, slug=gemini-1.5-flash-8b, provider=Google" }, { "model_id": "gemini-1.5-flash", "benchmark_id": "fleurs", "score": 9.6, "reference_url": "https://llm-stats.com/benchmarks/fleurs", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default", "context": "default", "notes": "Per Gemini 2.0 Flash Model Card (April 15 2025): non-thinking generation, default sampling. Same canonical setting applies to 1.5 / 2.0 family." }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=Gemini 1.5 Flash, slug=gemini-1.5-flash, provider=Google" }, { "model_id": "gemini-1.5-pro", "benchmark_id": "fleurs", "score": 6.7, "reference_url": "https://llm-stats.com/benchmarks/fleurs", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default", "context": "default", "notes": "Per Gemini 2.0 Flash Model Card (April 15 2025): non-thinking generation, default sampling. Same canonical setting applies to 1.5 / 2.0 family." }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=Gemini 1.5 Pro, slug=gemini-1.5-pro, provider=Google" }, { "model_id": "qwen2.5-omni-7b", "benchmark_id": "fleurs", "score": 4.1, "reference_url": "https://llm-stats.com/benchmarks/fleurs", "reported_setting": { "mode": "unknown", "effort": "n/a", "tools": "unknown", "sampling": "unknown", "judge": "unknown", "harness": "unknown", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown", "notes": "Model added from third-party aggregator (llm-stats.com); canonical setting not established." }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=Qwen2.5-Omni-7B, slug=qwen2.5-omni-7b, provider=Alibaba Cloud / Qwen Team" }, { "model_id": "o1-high", "benchmark_id": "frontiermath", "score": 5.5, "reference_url": "https://llm-stats.com/benchmarks/frontiermath", "reported_setting": { "effort": "default (o1 base, not high)" }, "matches_canonical": false, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=o1, slug=o1-2024-12-17, provider=OpenAI" }, { "model_id": "gpt-5", "benchmark_id": "frontiermath_tier4", "score": 26.3, "reference_url": "https://llm-stats.com/benchmarks/frontiermath", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none (per-bench override)", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default", "context": "default", "notes": "Per OpenAI GPT-5 dev blog. high reasoning effort." }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=GPT-5, slug=gpt-5-2025-08-07, provider=OpenAI" }, { "model_id": "gpt-5-mini", "benchmark_id": "frontiermath_tier4", "score": 22.1, "reference_url": "https://llm-stats.com/benchmarks/frontiermath", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none (per-bench override)", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default", "context": "default", "notes": "Per OpenAI GPT-5 developer blog https://openai.com/index/introducing-gpt-5-for-developers/" }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=GPT-5 mini, slug=gpt-5-mini-2025-08-07, provider=OpenAI" }, { "model_id": "o3-high", "benchmark_id": "frontiermath_tier4", "score": 15.8, "reference_url": "https://llm-stats.com/benchmarks/frontiermath", "reported_setting": { "effort": "default (o3 base, not high)" }, "matches_canonical": false, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=o3, slug=o3-2025-04-16, provider=OpenAI" }, { "model_id": "gpt-5-nano", "benchmark_id": "frontiermath_tier4", "score": 9.6, "reference_url": "https://llm-stats.com/benchmarks/frontiermath", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none (per-bench override)", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default", "context": "default", "notes": "Per OpenAI GPT-5 developer blog https://openai.com/index/introducing-gpt-5-for-developers/" }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=GPT-5 nano, slug=gpt-5-nano-2025-08-07, provider=OpenAI" }, { "model_id": "o1-high", "benchmark_id": "frontiermath_tier4", "score": 5.5, "reference_url": "https://llm-stats.com/benchmarks/frontiermath", "reported_setting": { "effort": "default (o1 base, not high)" }, "matches_canonical": false, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=o1, slug=o1-2024-12-17, provider=OpenAI" }, { "model_id": "muse-spark", "benchmark_id": "frontier_science_research", "score": 38.3, "reference_url": "https://llm-stats.com/benchmarks/frontierscience-research", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none (per-benchmark override where tools are required)", "sampling": "pass@1 (single-pass)", "judge": "rule-based or gpt-oss-120b (per benchmark)", "harness": "internal (Meta MSL)", "prompt_style": "default", "temperature": "default", "notes": "Per Meta MSL blog https://ai.meta.com/blog/introducing-muse-spark-msl/. Standard Thinking mode. Contemplating mode results flagged for review." }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=Muse Spark, slug=muse-spark, provider=Meta" }, { "model_id": "gemini-diffusion", "benchmark_id": "global_mmlu_lite", "score": 69.1, "reference_url": "https://llm-stats.com/benchmarks/global-mmlu-lite", "reported_setting": { "mode": "unknown", "effort": "n/a", "tools": "unknown", "sampling": "unknown", "judge": "unknown", "harness": "unknown", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown", "notes": "Model added from third-party aggregator (llm-stats.com); canonical setting not established." }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=Gemini Diffusion, slug=gemini-diffusion, provider=Google" }, { "model_id": "gemma-3-1b-it", "benchmark_id": "global_mmlu_lite", "score": 34.2, "reference_url": "https://llm-stats.com/benchmarks/global-mmlu-lite", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default", "context": "32k", "notes": "Google Gemma 3 1B instruction-tuned." }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=Gemma 3 1B, slug=gemma-3-1b-it, provider=Google" }, { "model_id": "qwen3.6-plus", "benchmark_id": "global_piqa", "score": 89.8, "reference_url": "https://llm-stats.com/benchmarks/global-piqa", "reported_setting": {}, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=Qwen3.6 Plus, slug=qwen3.6-plus, provider=Alibaba Cloud / Qwen Team" }, { "model_id": "longcat-flash-chat", "benchmark_id": "gpqa_diamond", "score": 73.2, "reference_url": "https://llm-stats.com/benchmarks/gpqa", "reported_setting": { "mode": "unknown", "effort": "n/a", "tools": "unknown", "sampling": "unknown", "judge": "unknown", "harness": "unknown", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown", "notes": "Model added from third-party aggregator (llm-stats.com); canonical setting not established." }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=LongCat-Flash-Chat, slug=longcat-flash-chat, provider=Meituan" }, { "model_id": "ministral-8b", "benchmark_id": "gpqa_diamond", "score": 66.8, "reference_url": "https://llm-stats.com/benchmarks/gpqa", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "OLMES (Allen AI)", "prompt_style": "default", "temperature": "default", "context": "default", "notes": "Per OLMo 2 paper Table 7 Instruct (instruction-tuned variant)." }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=Ministral 3 (8B Reasoning 2512), slug=ministral-8b-latest, provider=Mistral AI" }, { "model_id": "sarvam-30b", "benchmark_id": "gpqa_diamond", "score": 66.5, "reference_url": "https://llm-stats.com/benchmarks/gpqa", "reported_setting": { "mode": "unknown", "effort": "n/a", "tools": "unknown", "sampling": "unknown", "judge": "unknown", "harness": "unknown", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown", "notes": "Model added from third-party aggregator (llm-stats.com); canonical setting not established." }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=Sarvam-30B, slug=sarvam-30b, provider=Sarvam AI" }, { "model_id": "grok-2", "benchmark_id": "gpqa_diamond", "score": 56.0, "reference_url": "https://llm-stats.com/benchmarks/gpqa", "reported_setting": { "mode": "unknown", "effort": "n/a", "tools": "unknown", "sampling": "unknown", "judge": "unknown", "harness": "unknown", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown", "notes": "Model added from third-party aggregator (llm-stats.com); canonical setting not established." }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=Grok-2, slug=grok-2, provider=xAI" }, { "model_id": "grok-2-mini", "benchmark_id": "gpqa_diamond", "score": 51.0, "reference_url": "https://llm-stats.com/benchmarks/gpqa", "reported_setting": { "mode": "unknown", "effort": "n/a", "tools": "unknown", "sampling": "unknown", "judge": "unknown", "harness": "unknown", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown", "notes": "Model added from third-party aggregator (llm-stats.com); canonical setting not established." }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=Grok-2 mini, slug=grok-2-mini, provider=xAI" }, { "model_id": "claude-3-opus", "benchmark_id": "gpqa_diamond", "score": 50.4, "reference_url": "https://llm-stats.com/benchmarks/gpqa", "reported_setting": { "mode": "unknown", "effort": "n/a", "tools": "unknown", "sampling": "unknown", "judge": "unknown", "harness": "unknown", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown", "notes": "Model added from third-party aggregator (llm-stats.com); canonical setting not established." }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=Claude 3 Opus, slug=claude-3-opus-20240229, provider=Anthropic" }, { "model_id": "qwen-2.5-32b", "benchmark_id": "gpqa_diamond", "score": 49.5, "reference_url": "https://llm-stats.com/benchmarks/gpqa", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "OLMES (Allen AI)", "prompt_style": "default", "temperature": "default", "context": "default", "notes": "Per OLMo 2 paper Table 7 Instruct (instruction-tuned variant)." }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=Qwen2.5 32B Instruct, slug=qwen-2.5-32b-instruct, provider=Alibaba Cloud / Qwen Team" }, { "model_id": "gpt-4-turbo", "benchmark_id": "gpqa_diamond", "score": 48.0, "reference_url": "https://llm-stats.com/benchmarks/gpqa", "reported_setting": { "mode": "unknown", "effort": "n/a", "tools": "unknown", "sampling": "unknown", "judge": "unknown", "harness": "unknown", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown", "notes": "Model added from third-party aggregator (llm-stats.com); canonical setting not established." }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=GPT-4 Turbo, slug=gpt-4-turbo-2024-04-09, provider=OpenAI" }, { "model_id": "mistral-small-24b", "benchmark_id": "gpqa_diamond", "score": 45.3, "reference_url": "https://llm-stats.com/benchmarks/gpqa", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "OLMES (Allen AI)", "prompt_style": "default", "temperature": "default", "context": "default", "notes": "Per OLMo 2 paper Table 7 Instruct (instruction-tuned variant)." }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=Mistral Small 3 24B Instruct, slug=mistral-small-24b-instruct-2501, provider=Mistral AI" }, { "model_id": "qwen2-72b-instruct", "benchmark_id": "gpqa_diamond", "score": 42.4, "reference_url": "https://llm-stats.com/benchmarks/gpqa", "reported_setting": { "mode": "unknown", "effort": "n/a", "tools": "unknown", "sampling": "unknown", "judge": "unknown", "harness": "unknown", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown", "notes": "Model added from third-party aggregator (llm-stats.com); canonical setting not established." }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=Qwen2 72B Instruct, slug=qwen2-72b-instruct, provider=Alibaba Cloud / Qwen Team" }, { "model_id": "nova-lite", "benchmark_id": "gpqa_diamond", "score": 42.0, "reference_url": "https://llm-stats.com/benchmarks/gpqa", "reported_setting": { "mode": "unknown", "effort": "n/a", "tools": "unknown", "sampling": "unknown", "judge": "unknown", "harness": "unknown", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown", "notes": "Model added from third-party aggregator (llm-stats.com); canonical setting not established." }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=Nova Lite, slug=nova-lite, provider=Amazon" }, { "model_id": "llama-3.1-70b", "benchmark_id": "gpqa_diamond", "score": 41.7, "reference_url": "https://llm-stats.com/benchmarks/gpqa", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "OLMES (Allen AI)", "prompt_style": "default", "temperature": "default", "context": "default", "notes": "Per OLMo 2 paper Table 7 Instruct (instruction-tuned variant)." }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=Llama 3.1 70B Instruct, slug=llama-3.1-70b-instruct, provider=Meta" }, { "model_id": "claude-3.5-haiku", "benchmark_id": "gpqa_diamond", "score": 41.6, "reference_url": "https://llm-stats.com/benchmarks/gpqa", "reported_setting": { "mode": "unknown", "effort": "n/a", "tools": "unknown", "sampling": "unknown", "judge": "unknown", "harness": "unknown", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown", "notes": "Model added from third-party aggregator (llm-stats.com); canonical setting not established." }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=Claude 3.5 Haiku, slug=claude-3-5-haiku-20241022, provider=Anthropic" }, { "model_id": "claude-3-sonnet", "benchmark_id": "gpqa_diamond", "score": 40.4, "reference_url": "https://llm-stats.com/benchmarks/gpqa", "reported_setting": { "mode": "unknown", "effort": "n/a", "tools": "unknown", "sampling": "unknown", "judge": "unknown", "harness": "unknown", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown", "notes": "Model added from third-party aggregator (llm-stats.com); canonical setting not established." }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=Claude 3 Sonnet, slug=claude-3-sonnet-20240229, provider=Anthropic" }, { "model_id": "gemini-diffusion", "benchmark_id": "gpqa_diamond", "score": 40.4, "reference_url": "https://llm-stats.com/benchmarks/gpqa", "reported_setting": { "mode": "unknown", "effort": "n/a", "tools": "unknown", "sampling": "unknown", "judge": "unknown", "harness": "unknown", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown", "notes": "Model added from third-party aggregator (llm-stats.com); canonical setting not established." }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=Gemini Diffusion, slug=gemini-diffusion, provider=Google" }, { "model_id": "nova-micro", "benchmark_id": "gpqa_diamond", "score": 40.0, "reference_url": "https://llm-stats.com/benchmarks/gpqa", "reported_setting": { "mode": "unknown", "effort": "n/a", "tools": "unknown", "sampling": "unknown", "judge": "unknown", "harness": "unknown", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown", "notes": "Model added from third-party aggregator (llm-stats.com); canonical setting not established." }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=Nova Micro, slug=nova-micro, provider=Amazon" }, { "model_id": "gemini-1.5-flash-8b", "benchmark_id": "gpqa_diamond", "score": 38.4, "reference_url": "https://llm-stats.com/benchmarks/gpqa", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default", "context": "1M", "notes": "Per Command A paper baseline." }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=Gemini 1.5 Flash 8B, slug=gemini-1.5-flash-8b, provider=Google" }, { "model_id": "phi-3.5-moe-instruct", "benchmark_id": "gpqa_diamond", "score": 36.8, "reference_url": "https://llm-stats.com/benchmarks/gpqa", "reported_setting": { "mode": "unknown", "effort": "n/a", "tools": "unknown", "sampling": "unknown", "judge": "unknown", "harness": "unknown", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown", "notes": "Model added from third-party aggregator (llm-stats.com); canonical setting not established." }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=Phi-3.5-MoE-instruct, slug=phi-3.5-moe-instruct, provider=Microsoft" }, { "model_id": "qwen-2.5-7b", "benchmark_id": "gpqa_diamond", "score": 36.4, "reference_url": "https://llm-stats.com/benchmarks/gpqa", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "OLMES (Allen AI)", "prompt_style": "default", "temperature": "default", "context": "default", "notes": "Per OLMo 2 paper Table 7 Instruct (instruction-tuned variant)." }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=Qwen2.5 7B Instruct, slug=qwen-2.5-7b-instruct, provider=Alibaba Cloud / Qwen Team" }, { "model_id": "grok-1.5", "benchmark_id": "gpqa_diamond", "score": 35.9, "reference_url": "https://llm-stats.com/benchmarks/gpqa", "reported_setting": { "mode": "unknown", "effort": "n/a", "tools": "unknown", "sampling": "unknown", "judge": "unknown", "harness": "unknown", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown", "notes": "Model added from third-party aggregator (llm-stats.com); canonical setting not established." }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=Grok-1.5, slug=grok-1.5, provider=xAI" }, { "model_id": "claude-3-haiku", "benchmark_id": "gpqa_diamond", "score": 33.3, "reference_url": "https://llm-stats.com/benchmarks/gpqa", "reported_setting": { "mode": "unknown", "effort": "n/a", "tools": "unknown", "sampling": "unknown", "judge": "unknown", "harness": "unknown", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown", "notes": "Model added from third-party aggregator (llm-stats.com); canonical setting not established." }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=Claude 3 Haiku, slug=claude-3-haiku-20240307, provider=Anthropic" }, { "model_id": "gpt-3.5-turbo", "benchmark_id": "gpqa_diamond", "score": 30.8, "reference_url": "https://llm-stats.com/benchmarks/gpqa", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "OLMES (Allen AI)", "prompt_style": "default", "temperature": "default", "context": "default", "notes": "Per OLMo 2 paper Table 7 Instruct (instruction-tuned variant)." }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=GPT-3.5 Turbo, slug=gpt-3.5-turbo-0125, provider=OpenAI" }, { "model_id": "qwen2.5-omni-7b", "benchmark_id": "gpqa_diamond", "score": 30.8, "reference_url": "https://llm-stats.com/benchmarks/gpqa", "reported_setting": { "mode": "unknown", "effort": "n/a", "tools": "unknown", "sampling": "unknown", "judge": "unknown", "harness": "unknown", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown", "notes": "Model added from third-party aggregator (llm-stats.com); canonical setting not established." }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=Qwen2.5-Omni-7B, slug=qwen2.5-omni-7b, provider=Alibaba Cloud / Qwen Team" }, { "model_id": "phi-3.5-mini-instruct", "benchmark_id": "gpqa_diamond", "score": 30.4, "reference_url": "https://llm-stats.com/benchmarks/gpqa", "reported_setting": { "mode": "unknown", "effort": "n/a", "tools": "unknown", "sampling": "unknown", "judge": "unknown", "harness": "unknown", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown", "notes": "Model added from third-party aggregator (llm-stats.com); canonical setting not established." }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=Phi-3.5-mini-instruct, slug=phi-3.5-mini-instruct, provider=Microsoft" }, { "model_id": "qwen2-7b-instruct", "benchmark_id": "gpqa_diamond", "score": 25.3, "reference_url": "https://llm-stats.com/benchmarks/gpqa", "reported_setting": { "mode": "unknown", "effort": "n/a", "tools": "unknown", "sampling": "unknown", "judge": "unknown", "harness": "unknown", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown", "notes": "Model added from third-party aggregator (llm-stats.com); canonical setting not established." }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=Qwen2 7B Instruct, slug=qwen2-7b-instruct, provider=Alibaba Cloud / Qwen Team" }, { "model_id": "claude-mythos", "benchmark_id": "gpqa_main", "score": 94.6, "reference_url": "https://llm-stats.com/benchmarks/gpqa", "reported_setting": { "mode": "thinking", "effort": "adaptive thinking at max effort", "tools": "none (per-bench override)", "sampling": "pass@1 (avg 5 trials)", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default", "context": "up to 1M", "notes": "Per Mythos System Card Table 6.3.A: standard config is adaptive thinking max effort, default sampling (temperature, top_p), avg 5 trials, context up to 1M." }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=Claude Mythos Preview, slug=claude-mythos-preview, provider=Anthropic" }, { "model_id": "gemini-3.1-pro", "benchmark_id": "gpqa_main", "score": 94.3, "reference_url": "https://llm-stats.com/benchmarks/gpqa", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default", "context": "default", "notes": "Per Google DeepMind /models/gemini/pro/ table: Gemini 3.1 Pro Thinking (High), no tools by default." }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=Gemini 3.1 Pro, slug=gemini-3.1-pro-preview, provider=Google" }, { "model_id": "claude-opus-4.7", "benchmark_id": "gpqa_main", "score": 94.2, "reference_url": "https://llm-stats.com/benchmarks/gpqa", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none (per-bench override)", "sampling": "pass@1 (avg 5 trials)", "judge": "rule-based", "harness": "Terminus-2 (for terminal-bench, thinking disabled)", "prompt_style": "default", "temperature": "default", "context": "default", "notes": "Per Anthropic Opus 4.7 announcement table. Default thinking mode with extended thinking; Terminal-Bench uses Terminus-2 harness with thinking disabled (1× guaranteed/3× ceiling, avg 5 attempts)." }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=Claude Opus 4.7, slug=claude-opus-4-7, provider=Anthropic" }, { "model_id": "gpt-5.5", "benchmark_id": "gpqa_main", "score": 93.6, "reference_url": "https://llm-stats.com/benchmarks/gpqa", "reported_setting": { "mode": "thinking", "effort": "xhigh", "tools": "none (per-bench override)", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default", "context": "default", "notes": "Per OpenAI gpt-5.5 blog: reasoning effort=xhigh, research environment (may differ from production ChatGPT)." }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=GPT-5.5, slug=gpt-5.5, provider=OpenAI" }, { "model_id": "gpt-5.2", "benchmark_id": "gpqa_main", "score": 93.2, "reference_url": "https://llm-stats.com/benchmarks/gpqa", "reported_setting": { "effort": "pro" }, "matches_canonical": false, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=GPT-5.2 Pro, slug=gpt-5.2-pro-2025-12-11, provider=OpenAI" }, { "model_id": "gpt-5.4", "benchmark_id": "gpqa_main", "score": 92.8, "reference_url": "https://llm-stats.com/benchmarks/gpqa", "reported_setting": { "mode": "thinking", "effort": "xhigh", "tools": "none (per-bench override)", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default", "context": "default", "notes": "Per OpenAI gpt-5.4 blog: reasoning effort=xhigh, research environment (may differ from production ChatGPT)." }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=GPT-5.4, slug=gpt-5.4, provider=OpenAI" }, { "model_id": "gemini-3-pro", "benchmark_id": "gpqa_main", "score": 91.9, "reference_url": "https://llm-stats.com/benchmarks/gpqa", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default", "context": "default", "notes": "Per Google DeepMind /models/gemini/pro/ table: Gemini 3 Pro Thinking (High), no tools by default." }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=Gemini 3 Pro, slug=gemini-3-pro-preview, provider=Google" }, { "model_id": "claude-opus-4.6", "benchmark_id": "gpqa_main", "score": 91.3, "reference_url": "https://llm-stats.com/benchmarks/gpqa", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "none (per-bench override)", "sampling": "pass@1 (avg 5 trials)", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default", "context": "up to 1M", "notes": "Per Mythos System Card Table 6.3.A: standard config = adaptive thinking max effort, default sampling, avg 5 trials, context up to 1M." }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=Claude Opus 4.6, slug=claude-opus-4-6, provider=Anthropic" }, { "model_id": "kimi-k2.6", "benchmark_id": "gpqa_main", "score": 90.5, "reference_url": "https://llm-stats.com/benchmarks/gpqa", "reported_setting": { "mode": "thinking", "effort": "max (98304 generation tokens)", "tools": "none (per-bench override)", "sampling": "pass@1 (avg@10 coding, avg@3 vision)", "judge": "rule-based", "harness": "official (in-house SWE-agent-derived for SWE-Bench)", "prompt_style": "default", "temperature": "1.0", "context": "262144 (256K)", "notes": "Per Kimi K2.6 model card (HF moonshotai/Kimi-K2.6). Thinking mode, max effort, t=1.0 top_p=1.0." }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=Kimi K2.6, slug=kimi-k2.6, provider=Moonshot AI" }, { "model_id": "gemini-3-flash", "benchmark_id": "gpqa_main", "score": 90.4, "reference_url": "https://llm-stats.com/benchmarks/gpqa", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default", "context": "default", "notes": "Per Google DeepMind /models/gemini/flash/ table: Gemini 3 Flash Thinking, no tools by default." }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=Gemini 3 Flash, slug=gemini-3-flash-preview, provider=Google" }, { "model_id": "qwen3.6-plus", "benchmark_id": "gpqa_main", "score": 90.4, "reference_url": "https://llm-stats.com/benchmarks/gpqa", "reported_setting": {}, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=Qwen3.6 Plus, slug=qwen3.6-plus, provider=Alibaba Cloud / Qwen Team" }, { "model_id": "deepseek-v4-pro", "benchmark_id": "gpqa_main", "score": 90.1, "reference_url": "https://llm-stats.com/benchmarks/gpqa", "reported_setting": { "effort": "max" }, "matches_canonical": false, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=DeepSeek-V4-Pro-Max, slug=deepseek-v4-pro-max, provider=DeepSeek" }, { "model_id": "claude-sonnet-4.6", "benchmark_id": "gpqa_main", "score": 89.9, "reference_url": "https://llm-stats.com/benchmarks/gpqa", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "none (per-bench override)", "sampling": "pass@1 (avg 5-15 trials)", "judge": "rule-based", "harness": "Terminus-2 (terminal-bench, thinking off); official otherwise", "prompt_style": "default", "temperature": "default", "context": "default", "notes": "Per Anthropic Sonnet 4.6 announcement: max thinking effort default; Terminal-Bench=thinking off, Terminus-2; SWE-bench=avg 10 trials; BrowseComp/HLE-with-tools have specific tool configs (web search/fetch + 50k context compaction)." }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=Claude Sonnet 4.6, slug=claude-sonnet-4-6, provider=Anthropic" }, { "model_id": "muse-spark", "benchmark_id": "gpqa_main", "score": 89.5, "reference_url": "https://llm-stats.com/benchmarks/gpqa", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none (per-benchmark override where tools are required)", "sampling": "pass@1 (single-pass)", "judge": "rule-based or gpt-oss-120b (per benchmark)", "harness": "internal (Meta MSL)", "prompt_style": "default", "temperature": "default", "notes": "Per Meta MSL blog https://ai.meta.com/blog/introducing-muse-spark-msl/. Standard Thinking mode. Contemplating mode results flagged for review." }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=Muse Spark, slug=muse-spark, provider=Meta" }, { "model_id": "grok-4", "benchmark_id": "gpqa_main", "score": 88.4, "reference_url": "https://llm-stats.com/benchmarks/gpqa", "reported_setting": { "effort": "heavy" }, "matches_canonical": false, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=Grok-4 Heavy, slug=grok-4-heavy, provider=xAI" }, { "model_id": "gpt-5.1", "benchmark_id": "gpqa_main", "score": 88.1, "reference_url": "https://llm-stats.com/benchmarks/gpqa", "reported_setting": { "effort": "high" }, "matches_canonical": false, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=GPT-5.1 High, slug=gpt-5.1-high-2025-11-12, provider=OpenAI" }, { "model_id": "deepseek-v4-flash", "benchmark_id": "gpqa_main", "score": 88.1, "reference_url": "https://llm-stats.com/benchmarks/gpqa", "reported_setting": { "effort": "max" }, "matches_canonical": false, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=DeepSeek-V4-Flash-Max, slug=deepseek-v4-flash-max, provider=DeepSeek" }, { "model_id": "gpt-5", "benchmark_id": "gpqa_main", "score": 88.1, "reference_url": "https://llm-stats.com/benchmarks/gpqa", "reported_setting": { "effort": "medium" }, "matches_canonical": false, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=GPT-5 Medium, slug=gpt-5-medium-2025-08-07, provider=OpenAI" }, { "model_id": "kimi-k2.5", "benchmark_id": "gpqa_main", "score": 87.6, "reference_url": "https://llm-stats.com/benchmarks/gpqa", "reported_setting": { "mode": "thinking", "effort": "default (96k for reasoning, 64k for vision)", "tools": "none (per-bench override)", "sampling": "pass@1 (avg@32 AIME/HMMT, avg@8 GPQA, avg@3 vision, avg@4 Seal/WideSearch)", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "1.0", "context": "256K", "notes": "Per Kimi K2.5 model card (HF moonshotai/Kimi-K2.5). Thinking mode, t=1.0 top_p=0.95." }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=Kimi K2.5, slug=kimi-k2.5, provider=Moonshot AI" }, { "model_id": "claude-opus-4.5", "benchmark_id": "gpqa_main", "score": 87.0, "reference_url": "https://llm-stats.com/benchmarks/gpqa", "reported_setting": { "mode": "thinking", "effort": "high (default)", "tools": "none (per-bench override)", "sampling": "pass@1 (avg 5 trials)", "judge": "rule-based", "harness": "Terminus-2 (for terminal-bench); official otherwise", "prompt_style": "interleaved scratchpads", "temperature": "default (top_p, temperature)", "context": "200K (default)", "notes": "Per Anthropic Opus 4.5 announcement: 64K thinking budget, interleaved scratchpads, 200K context, default effort=high, default sampling, avg 5 trials. Exceptions: SWE-bench Verified=no thinking; Terminal-Bench=128K thinking budget." }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=Claude Opus 4.5, slug=claude-opus-4-5-20251101, provider=Anthropic" }, { "model_id": "gemini-2.5-pro", "benchmark_id": "gpqa_main", "score": 86.4, "reference_url": "https://llm-stats.com/benchmarks/gpqa", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1 (single attempt)", "judge": "rule-based", "harness": "AI Studio API", "prompt_style": "default", "temperature": "default", "context": "default", "notes": "Per Gemini 2.5 Pro Model Card (GA, June 27 2025): pass@1 single attempt, AI Studio API model-id gemini-2.5-pro-preview-06-05 / gemini-2.5-pro GA, default sampling. Release date set to GA date 2025-06-27." }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=Gemini 2.5 Pro Preview 06-05, slug=gemini-2.5-pro-preview-06-05, provider=Google" }, { "model_id": "glm-5.1", "benchmark_id": "gpqa_main", "score": 86.2, "reference_url": "https://llm-stats.com/benchmarks/gpqa", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none (per-bench override)", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default", "context": "default", "notes": "Per GLM-5.1 blog." }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=GLM-5.1, slug=glm-5.1, provider=Zhipu AI" }, { "model_id": "glm-4.7", "benchmark_id": "gpqa_main", "score": 85.7, "reference_url": "https://llm-stats.com/benchmarks/gpqa", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none (per-bench override)", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default", "context": "default", "notes": "Per GLM-5 blog." }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=GLM-4.7, slug=glm-4.7, provider=Zhipu AI" }, { "model_id": "claude-3.7-sonnet", "benchmark_id": "gpqa_main", "score": 84.8, "reference_url": "https://llm-stats.com/benchmarks/gpqa", "reported_setting": { "mode": "thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "zeroshot-cot", "temperature": 0.0, "context": "default", "notes": "Per Claude 4 blog cross-model table: with extended thinking for GPQA/MMMU/AIME/TAU; no extended thinking for SWE/Terminal/MMMLU." }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=Claude 3.7 Sonnet, slug=claude-3-7-sonnet-20250219, provider=Anthropic" }, { "model_id": "grok-3-beta", "benchmark_id": "gpqa_main", "score": 84.6, "reference_url": "https://llm-stats.com/benchmarks/gpqa", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none (per-bench override)", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default", "context": "default", "notes": "Per xAI Grok 3 blog: canonical = non-reasoning mode (the cross-model table without 'Think' label). Reasoning ON values not used." }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=Grok-3, slug=grok-3, provider=xAI" }, { "model_id": "kimi-k2-thinking", "benchmark_id": "gpqa_main", "score": 84.5, "reference_url": "https://llm-stats.com/benchmarks/gpqa", "reported_setting": { "mode": "thinking", "effort": "default (96k thinking-token budget for HLE/AIME/HMMT/GPQA; 128k for IMO/LCB/OJ)", "tools": "none (per-bench override)", "sampling": "pass@1 (avg@32 for AIME/HMMT no-tools)", "judge": "rule-based (HLE uses o3-mini)", "harness": "official", "prompt_style": "default", "temperature": "default", "context": "default (256K)", "notes": "Per Kimi K2 Thinking blog (moonshotai.github.io/Kimi-K2/thinking.html). Thinking model." }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=Kimi K2-Thinking-0905, slug=kimi-k2-thinking-0905, provider=Moonshot AI" }, { "model_id": "claude-sonnet-4.5", "benchmark_id": "gpqa_main", "score": 83.4, "reference_url": "https://llm-stats.com/benchmarks/gpqa", "reported_setting": { "mode": "thinking", "effort": "high (default)", "tools": "none (per-bench override)", "sampling": "pass@1 (avg 5 trials)", "judge": "rule-based", "harness": "Terminus-2 (for terminal-bench); official otherwise", "prompt_style": "interleaved scratchpads", "temperature": "default (top_p, temperature)", "context": "200K (default)", "notes": "Per Anthropic Opus 4.5 announcement: 64K thinking budget, interleaved scratchpads, 200K context, default effort=high, default sampling, avg 5 trials. Exceptions: SWE-bench Verified=no thinking; Terminal-Bench=128K thinking budget." }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=Claude Sonnet 4.5, slug=claude-sonnet-4-5-20250929, provider=Anthropic" }, { "model_id": "o3-high", "benchmark_id": "gpqa_main", "score": 83.3, "reference_url": "https://llm-stats.com/benchmarks/gpqa", "reported_setting": { "effort": "default (o3 base, not high)" }, "matches_canonical": false, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=o3, slug=o3-2025-04-16, provider=OpenAI" }, { "model_id": "gemini-2.5-flash", "benchmark_id": "gpqa_main", "score": 82.8, "reference_url": "https://llm-stats.com/benchmarks/gpqa", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none (per-bench override)", "sampling": "pass@1 (single attempt; multi-trial avg for smaller bench)", "judge": "rule-based", "harness": "AI Studio API (gemini-2.5-flash GA)", "prompt_style": "default", "temperature": "default", "context": "default (1M)", "notes": "Per Gemini 2.5 Flash Model Card (Dec 2025) GA Thinking column." }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=Gemini 2.5 Flash, slug=gemini-2.5-flash, provider=Google" }, { "model_id": "deepseek-v3.2", "benchmark_id": "gpqa_main", "score": 82.4, "reference_url": "https://llm-stats.com/benchmarks/gpqa", "reported_setting": { "mode": "thinking (DeepSeek-Reasoner endpoint)" }, "matches_canonical": false, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=DeepSeek-V3.2 (Thinking), slug=deepseek-reasoner, provider=DeepSeek" }, { "model_id": "gpt-5-mini", "benchmark_id": "gpqa_main", "score": 82.3, "reference_url": "https://llm-stats.com/benchmarks/gpqa", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none (per-bench override)", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default", "context": "default", "notes": "Per OpenAI GPT-5 developer blog https://openai.com/index/introducing-gpt-5-for-developers/" }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=GPT-5 mini, slug=gpt-5-mini-2025-08-07, provider=OpenAI" }, { "model_id": "o4-mini-high", "benchmark_id": "gpqa_main", "score": 81.4, "reference_url": "https://llm-stats.com/benchmarks/gpqa", "reported_setting": { "effort": "default (o4-mini base, not high)" }, "matches_canonical": false, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=o4-mini, slug=o4-mini, provider=OpenAI" }, { "model_id": "qwen3-235b", "benchmark_id": "gpqa_main", "score": 81.1, "reference_url": "https://llm-stats.com/benchmarks/gpqa", "reported_setting": { "mode": "thinking", "snapshot": "2507" }, "matches_canonical": false, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=Qwen3-235B-A22B-Thinking-2507, slug=qwen3-235b-a22b-thinking-2507, provider=Alibaba Cloud / Qwen Team" }, { "model_id": "glm-4.6", "benchmark_id": "gpqa_main", "score": 81.0, "reference_url": "https://llm-stats.com/benchmarks/gpqa", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none (per-bench override)", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default", "context": "default", "notes": "Per GLM-4.7 blog." }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=GLM-4.6, slug=glm-4.6, provider=Zhipu AI" }, { "model_id": "deepseek-r1", "benchmark_id": "gpqa_main", "score": 81.0, "reference_url": "https://llm-stats.com/benchmarks/gpqa", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none (per-bench override)", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default", "context": "default", "notes": "Per DeepSeek R1-0528 model card." }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=DeepSeek-R1-0528, slug=deepseek-r1-0528, provider=DeepSeek" }, { "model_id": "minimax-m2.1", "benchmark_id": "gpqa_main", "score": 81.0, "reference_url": "https://llm-stats.com/benchmarks/gpqa", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none (per-bench override)", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default", "context": "default", "notes": "Per MiniMax M2.5 model card (referenced as predecessor)." }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=MiniMax M2.1, slug=minimax-m2.1, provider=MiniMax" }, { "model_id": "claude-opus-4.1", "benchmark_id": "gpqa_main", "score": 80.9, "reference_url": "https://llm-stats.com/benchmarks/gpqa", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none (per-bench override)", "sampling": "pass@1 (single-pass)", "judge": "rule-based", "harness": "Terminus 1 (for agentic benchmarks)", "prompt_style": "default", "temperature": "top_p=0.95", "context": "default", "notes": "Per Anthropic Opus 4.1 announcement table. Single-pass values; harness=Terminus 1 for Terminal-Bench (note: differs from Claude 4 blog which used Claude Code framework)." }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=Claude Opus 4.1, slug=claude-opus-4-1-20250805, provider=Anthropic" }, { "model_id": "gpt-oss-120b", "benchmark_id": "gpqa_main", "score": 80.9, "reference_url": "https://llm-stats.com/benchmarks/gpqa", "reported_setting": { "effort": "high" }, "matches_canonical": false, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=GPT OSS 120B High, slug=gpt-oss-120b-high, provider=OpenAI" }, { "model_id": "claude-opus-4", "benchmark_id": "gpqa_main", "score": 79.6, "reference_url": "https://llm-stats.com/benchmarks/gpqa", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none (per-bench override)", "sampling": "pass@1 (single-pass)", "judge": "rule-based", "harness": "Claude Code agent framework (for agentic benchmarks)", "prompt_style": "default", "temperature": "top_p=0.95", "context": "default", "notes": "Per Anthropic Claude 4 announcement table: single-pass values; no extended thinking for SWE/Terminal/MMMLU; extended thinking up to 64K for GPQA/MMMU/AIME/TAU; tools=agentic for SWE/Terminal/TAU/OSWorld. Agentic harness=Claude Code per Claude 4 blog." }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=Claude Opus 4, slug=claude-opus-4-20250514, provider=Anthropic" }, { "model_id": "minimax-m2", "benchmark_id": "gpqa_main", "score": 78.0, "reference_url": "https://llm-stats.com/benchmarks/gpqa", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none (per-bench override)", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default", "context": "default", "notes": "Per MiniMax M2 model card." }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=MiniMax M2, slug=minimax-m2, provider=MiniMax" }, { "model_id": "o1-high", "benchmark_id": "gpqa_main", "score": 78.0, "reference_url": "https://llm-stats.com/benchmarks/gpqa", "reported_setting": { "effort": "default (o1 base, not high)" }, "matches_canonical": false, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=o1, slug=o1-2024-12-17, provider=OpenAI" }, { "model_id": "nemotron-ultra-253b", "benchmark_id": "gpqa_main", "score": 76.0, "reference_url": "https://llm-stats.com/benchmarks/gpqa", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none (benchmark-specified)", "sampling": "pass@1 (avg of up to 16 benchmark runs)", "judge": "rule-based", "harness": "NVIDIA eval pipeline (transformers/vLLM)", "prompt_style": "custom prompt templates per benchmark", "temperature": "0.6 (thinking mode)", "context": "32k", "notes": "Per NVIDIA HF model card https://huggingface.co/nvidia/Llama-3_1-Nemotron-Ultra-253B-v1: Reasoning On mode, temperature=0.6, top_p=0.95, 32k context, benchmarks run up to 16x and averaged." }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=Llama 3.1 Nemotron Ultra 253B v1, slug=llama-3.1-nemotron-ultra-253b-v1, provider=NVIDIA" }, { "model_id": "kimi-k2", "benchmark_id": "gpqa_main", "score": 75.8, "reference_url": "https://llm-stats.com/benchmarks/gpqa", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none (per-bench override)", "sampling": "pass@1 (avg of 4-64 trials per benchmark)", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default", "context": "default (256K)", "notes": "Per Kimi K2 tech report (arxiv:2507.20534) Table 3 Kimi-K2-Instruct column. Non-thinking model." }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=Kimi K2 0905, slug=kimi-k2-0905, provider=Moonshot AI" }, { "model_id": "claude-sonnet-4", "benchmark_id": "gpqa_main", "score": 75.4, "reference_url": "https://llm-stats.com/benchmarks/gpqa", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none (per-bench override)", "sampling": "pass@1 (single-pass)", "judge": "rule-based", "harness": "Claude Code agent framework (for agentic benchmarks)", "prompt_style": "default", "temperature": "top_p=0.95", "context": "default", "notes": "Per Anthropic Claude 4 announcement table: single-pass values; no extended thinking for SWE/Terminal/MMMLU; extended thinking up to 64K for GPQA/MMMU/AIME/TAU; tools=agentic for SWE/Terminal/TAU/OSWorld. Agentic harness=Claude Code per Claude 4 blog." }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=Claude Sonnet 4, slug=claude-sonnet-4-20250514, provider=Anthropic" }, { "model_id": "gpt-oss-20b", "benchmark_id": "gpqa_main", "score": 74.2, "reference_url": "https://llm-stats.com/benchmarks/gpqa", "reported_setting": { "effort": "high" }, "matches_canonical": false, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=GPT OSS 20B High, slug=gpt-oss-20b-high, provider=OpenAI" }, { "model_id": "gemini-2.0-flash", "benchmark_id": "gpqa_main", "score": 74.2, "reference_url": "https://llm-stats.com/benchmarks/gpqa", "reported_setting": { "effort": "thinking" }, "matches_canonical": false, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=Gemini 2.0 Flash Thinking, slug=gemini-2.0-flash-thinking, provider=Google" }, { "model_id": "longcat-flash-chat", "benchmark_id": "gpqa_main", "score": 73.2, "reference_url": "https://llm-stats.com/benchmarks/gpqa", "reported_setting": { "mode": "unknown", "effort": "n/a", "tools": "unknown", "sampling": "unknown", "judge": "unknown", "harness": "unknown", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown", "notes": "Model added from third-party aggregator (llm-stats.com); canonical setting not established." }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=LongCat-Flash-Chat, slug=longcat-flash-chat, provider=Meituan" }, { "model_id": "claude-haiku-4.5", "benchmark_id": "gpqa_main", "score": 73.0, "reference_url": "https://llm-stats.com/benchmarks/gpqa", "reported_setting": { "mode": "thinking", "effort": "128K thinking budget", "tools": "none (per-bench override for agentic benchmarks)", "sampling": "pass@1 (avg over ≥10 runs unless noted)", "judge": "official harness", "harness": "official", "prompt_style": "default", "temperature": "default (Anthropic default)" }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=Claude Haiku 4.5, slug=claude-haiku-4-5-20251001, provider=Anthropic" }, { "model_id": "gpt-5-nano", "benchmark_id": "gpqa_main", "score": 71.2, "reference_url": "https://llm-stats.com/benchmarks/gpqa", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none (per-bench override)", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default", "context": "default", "notes": "Per OpenAI GPT-5 developer blog https://openai.com/index/introducing-gpt-5-for-developers/" }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=GPT-5 nano, slug=gpt-5-nano-2025-08-07, provider=OpenAI" }, { "model_id": "gpt-4o", "benchmark_id": "gpqa_main", "score": 70.1, "reference_url": "https://llm-stats.com/benchmarks/gpqa", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none (per-bench override)", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default", "context": "default", "notes": "Per OpenAI GPT-4.1 blog table." }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=GPT-4o, slug=gpt-4o-2024-08-06, provider=OpenAI" }, { "model_id": "llama-4-maverick", "benchmark_id": "gpqa_main", "score": 69.8, "reference_url": "https://llm-stats.com/benchmarks/gpqa", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official-meta", "prompt_style": "zeroshot", "temperature": "0.0", "context": "default", "multimodal_input": true }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=Llama 4 Maverick, slug=llama-4-maverick, provider=Meta" }, { "model_id": "gpt-4.5", "benchmark_id": "gpqa_main", "score": 69.5, "reference_url": "https://llm-stats.com/benchmarks/gpqa", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none (per-bench override)", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default", "context": "default", "notes": "Per OpenAI GPT-4.1 blog table." }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=GPT-4.5, slug=gpt-4.5, provider=OpenAI" }, { "model_id": "phi-4-reasoning-plus", "benchmark_id": "gpqa_main", "score": 68.9, "reference_url": "https://llm-stats.com/benchmarks/gpqa", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1 (avg of 5)", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "0.8", "context": "default (32K)", "notes": "Per Phi-4-reasoning paper. Same as phi-4-reasoning + GRPO RL on top." }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=Phi 4 Reasoning Plus, slug=phi-4-reasoning-plus, provider=Microsoft" }, { "model_id": "sarvam-30b", "benchmark_id": "gpqa_main", "score": 66.5, "reference_url": "https://llm-stats.com/benchmarks/gpqa", "reported_setting": { "mode": "unknown", "effort": "n/a", "tools": "unknown", "sampling": "unknown", "judge": "unknown", "harness": "unknown", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown", "notes": "Model added from third-party aggregator (llm-stats.com); canonical setting not established." }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=Sarvam-30B, slug=sarvam-30b, provider=Sarvam AI" }, { "model_id": "gpt-4.1", "benchmark_id": "gpqa_main", "score": 66.3, "reference_url": "https://llm-stats.com/benchmarks/gpqa", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none (per-bench override)", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default", "context": "default", "notes": "Per OpenAI GPT-5 dev blog. non-reasoning model." }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=GPT-4.1, slug=gpt-4.1-2025-04-14, provider=OpenAI" }, { "model_id": "phi-4-reasoning", "benchmark_id": "gpqa_main", "score": 65.8, "reference_url": "https://llm-stats.com/benchmarks/gpqa", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1 (avg of 5)", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "0.8", "context": "default (32K)", "notes": "Per Phi-4-reasoning paper (arxiv:2504.21318) Tables 1-2. Thinking model, temp 0.8, max_seq_len 31k." }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=Phi 4 Reasoning, slug=phi-4-reasoning, provider=Microsoft" }, { "model_id": "qwen3-30b-a3b", "benchmark_id": "gpqa_main", "score": 65.8, "reference_url": "https://llm-stats.com/benchmarks/gpqa", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official (Qwen-Agent / sglang)", "prompt_style": "default (enable_thinking=True)", "temperature": "default", "context": "default", "notes": "Per Qwen3 tech report (arxiv:2505.09388) Table 13/15/17/19 Thinking column." }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=Qwen3 30B A3B, slug=qwen3-30b-a3b, provider=Alibaba Cloud / Qwen Team" }, { "model_id": "deepseek-r1-distill-llama-70b", "benchmark_id": "gpqa_main", "score": 65.2, "reference_url": "https://llm-stats.com/benchmarks/gpqa", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official (DeepSeek SGLang/vLLM)", "prompt_style": "default", "temperature": "default", "context": "default", "notes": "Distill of DeepSeek-R1 onto Llama-70B base. Cells in BP currently primarily sourced from Qwen3 tech report Table 13 third-party self-test (matches_canonical=false). DeepSeek own paper audit pending — canonical_setting may need refinement." }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=DeepSeek R1 Distill Llama 70B, slug=deepseek-r1-distill-llama-70b, provider=DeepSeek" }, { "model_id": "qwq-32b", "benchmark_id": "gpqa_main", "score": 65.2, "reference_url": "https://llm-stats.com/benchmarks/gpqa", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default", "context": "default (32K)", "notes": "Per QwQ-32B release blog (qwenlm.github.io/blog/qwq-32b/) + Qwen3 tech report Table 13 baseline column." }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=QwQ-32B, slug=qwq-32b, provider=Alibaba Cloud / Qwen Team" }, { "model_id": "gpt-4.1-mini", "benchmark_id": "gpqa_main", "score": 65.0, "reference_url": "https://llm-stats.com/benchmarks/gpqa", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none (per-bench override)", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default", "context": "default", "notes": "Per OpenAI GPT-5 dev blog. non-reasoning model." }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=GPT-4.1 mini, slug=gpt-4.1-mini-2025-04-14, provider=OpenAI" }, { "model_id": "deepseek-r1-distill-qwen-32b", "benchmark_id": "gpqa_main", "score": 62.1, "reference_url": "https://llm-stats.com/benchmarks/gpqa", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default", "context": "default", "notes": "Per DeepSeek R1 paper Table 15 (distilled model)." }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=DeepSeek R1 Distill Qwen 32B, slug=deepseek-r1-distill-qwen-32b, provider=DeepSeek" }, { "model_id": "o1-mini", "benchmark_id": "gpqa_main", "score": 60.0, "reference_url": "https://llm-stats.com/benchmarks/gpqa", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default", "context": "default", "notes": "Per Phi-4-reasoning paper Table 1 baseline (Microsoft self-test)." }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=o1-mini, slug=o1-mini, provider=OpenAI" }, { "model_id": "deepseek-r1-distill-qwen-14b", "benchmark_id": "gpqa_main", "score": 59.1, "reference_url": "https://llm-stats.com/benchmarks/gpqa", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default", "context": "default", "notes": "Per DeepSeek R1 paper Table 15 (distilled model)." }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=DeepSeek R1 Distill Qwen 14B, slug=deepseek-r1-distill-qwen-14b, provider=DeepSeek" }, { "model_id": "gemini-1.5-pro", "benchmark_id": "gpqa_main", "score": 59.1, "reference_url": "https://llm-stats.com/benchmarks/gpqa", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default", "context": "default", "notes": "Per Gemini 2.0 Flash Model Card (April 15 2025): non-thinking generation, default sampling. Same canonical setting applies to 1.5 / 2.0 family." }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=Gemini 1.5 Pro, slug=gemini-1.5-pro, provider=Google" }, { "model_id": "llama-4-scout", "benchmark_id": "gpqa_main", "score": 57.2, "reference_url": "https://llm-stats.com/benchmarks/gpqa", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official-meta", "prompt_style": "zeroshot", "temperature": "0.0", "context": "default", "multimodal_input": true }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=Llama 4 Scout, slug=llama-4-scout, provider=Meta" }, { "model_id": "phi-4", "benchmark_id": "gpqa_main", "score": 56.1, "reference_url": "https://llm-stats.com/benchmarks/gpqa", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "OpenAI simple-evals", "prompt_style": "default", "temperature": "0.5", "context": "default (16K)", "notes": "Per Phi-4 paper (arxiv:2412.08905) Table 1, simple-evals framework, temperature 0.5." }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=Phi 4, slug=phi-4, provider=Microsoft" }, { "model_id": "grok-2", "benchmark_id": "gpqa_main", "score": 56.0, "reference_url": "https://llm-stats.com/benchmarks/gpqa", "reported_setting": { "mode": "unknown", "effort": "n/a", "tools": "unknown", "sampling": "unknown", "judge": "unknown", "harness": "unknown", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown", "notes": "Model added from third-party aggregator (llm-stats.com); canonical setting not established." }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=Grok-2, slug=grok-2, provider=xAI" }, { "model_id": "gemini-2.0-flash-lite", "benchmark_id": "gpqa_main", "score": 51.5, "reference_url": "https://llm-stats.com/benchmarks/gpqa", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default", "context": "default", "notes": "Per Gemini 2.0 Flash Model Card (April 15 2025): non-thinking generation, default sampling. Same canonical setting applies to 1.5 / 2.0 family." }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=Gemini 2.0 Flash-Lite, slug=gemini-2.0-flash-lite, provider=Google" }, { "model_id": "grok-2-mini", "benchmark_id": "gpqa_main", "score": 51.0, "reference_url": "https://llm-stats.com/benchmarks/gpqa", "reported_setting": { "mode": "unknown", "effort": "n/a", "tools": "unknown", "sampling": "unknown", "judge": "unknown", "harness": "unknown", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown", "notes": "Model added from third-party aggregator (llm-stats.com); canonical setting not established." }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=Grok-2 mini, slug=grok-2-mini, provider=xAI" }, { "model_id": "gemini-1.5-flash", "benchmark_id": "gpqa_main", "score": 51.0, "reference_url": "https://llm-stats.com/benchmarks/gpqa", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default", "context": "default", "notes": "Per Gemini 2.0 Flash Model Card (April 15 2025): non-thinking generation, default sampling. Same canonical setting applies to 1.5 / 2.0 family." }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=Gemini 1.5 Flash, slug=gemini-1.5-flash, provider=Google" }, { "model_id": "claude-3-opus", "benchmark_id": "gpqa_main", "score": 50.4, "reference_url": "https://llm-stats.com/benchmarks/gpqa", "reported_setting": { "mode": "unknown", "effort": "n/a", "tools": "unknown", "sampling": "unknown", "judge": "unknown", "harness": "unknown", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown", "notes": "Model added from third-party aggregator (llm-stats.com); canonical setting not established." }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=Claude 3 Opus, slug=claude-3-opus-20240229, provider=Anthropic" }, { "model_id": "gpt-4.1-nano", "benchmark_id": "gpqa_main", "score": 50.3, "reference_url": "https://llm-stats.com/benchmarks/gpqa", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none (per-bench override)", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default", "context": "default", "notes": "Per OpenAI GPT-5 dev blog. non-reasoning model." }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=GPT-4.1 nano, slug=gpt-4.1-nano-2025-04-14, provider=OpenAI" }, { "model_id": "qwen-2.5-32b", "benchmark_id": "gpqa_main", "score": 49.5, "reference_url": "https://llm-stats.com/benchmarks/gpqa", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "OLMES (Allen AI)", "prompt_style": "default", "temperature": "default", "context": "default", "notes": "Per OLMo 2 paper Table 7 Instruct (instruction-tuned variant)." }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=Qwen2.5 32B Instruct, slug=qwen-2.5-32b-instruct, provider=Alibaba Cloud / Qwen Team" }, { "model_id": "deepseek-r1-distill-qwen-7b", "benchmark_id": "gpqa_main", "score": 49.1, "reference_url": "https://llm-stats.com/benchmarks/gpqa", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default", "context": "default", "notes": "Per DeepSeek R1 paper Table 15 (distilled model)." }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=DeepSeek R1 Distill Qwen 7B, slug=deepseek-r1-distill-qwen-7b, provider=DeepSeek" }, { "model_id": "deepseek-r1-distill-llama-8b", "benchmark_id": "gpqa_main", "score": 49.0, "reference_url": "https://llm-stats.com/benchmarks/gpqa", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default", "context": "default", "notes": "Per DeepSeek R1 paper Table 15 (distilled model)." }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=DeepSeek R1 Distill Llama 8B, slug=deepseek-r1-distill-llama-8b, provider=DeepSeek" }, { "model_id": "gpt-4-turbo", "benchmark_id": "gpqa_main", "score": 48.0, "reference_url": "https://llm-stats.com/benchmarks/gpqa", "reported_setting": { "mode": "unknown", "effort": "n/a", "tools": "unknown", "sampling": "unknown", "judge": "unknown", "harness": "unknown", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown", "notes": "Model added from third-party aggregator (llm-stats.com); canonical setting not established." }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=GPT-4 Turbo, slug=gpt-4-turbo-2024-04-09, provider=OpenAI" }, { "model_id": "mistral-small-3.1", "benchmark_id": "gpqa_main", "score": 46.0, "reference_url": "https://llm-stats.com/benchmarks/gpqa", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "mistral-eval", "prompt_style": "mixed (5-shot CoT for MMLU/Pro/GPQA per HF MC; default for others)", "temperature": "0.0", "context": "default", "multimodal_input": true }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=Mistral Small 3.1 24B Instruct, slug=mistral-small-3.1-24b-instruct-2503, provider=Mistral AI" }, { "model_id": "qwen-2.5-14b-instruct", "benchmark_id": "gpqa_main", "score": 45.5, "reference_url": "https://llm-stats.com/benchmarks/gpqa", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default", "context": "default (128K)", "notes": "Per Phi-4 paper Table 1 baseline." }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=Qwen2.5 14B Instruct, slug=qwen-2.5-14b-instruct, provider=Alibaba Cloud / Qwen Team" }, { "model_id": "mistral-small-24b", "benchmark_id": "gpqa_main", "score": 45.3, "reference_url": "https://llm-stats.com/benchmarks/gpqa", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "OLMES (Allen AI)", "prompt_style": "default", "temperature": "default", "context": "default", "notes": "Per OLMo 2 paper Table 7 Instruct (instruction-tuned variant)." }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=Mistral Small 3 24B Instruct, slug=mistral-small-24b-instruct-2501, provider=Mistral AI" }, { "model_id": "qwen2-72b-instruct", "benchmark_id": "gpqa_main", "score": 42.4, "reference_url": "https://llm-stats.com/benchmarks/gpqa", "reported_setting": { "mode": "unknown", "effort": "n/a", "tools": "unknown", "sampling": "unknown", "judge": "unknown", "harness": "unknown", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown", "notes": "Model added from third-party aggregator (llm-stats.com); canonical setting not established." }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=Qwen2 72B Instruct, slug=qwen2-72b-instruct, provider=Alibaba Cloud / Qwen Team" }, { "model_id": "gemma-3-27b", "benchmark_id": "gpqa_main", "score": 42.4, "reference_url": "https://llm-stats.com/benchmarks/gpqa", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default", "context": "default (128K)", "notes": "Per Gemma 3 tech report (arxiv:2503.19786) Table 6 IT column for 27B." }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=Gemma 3 27B, slug=gemma-3-27b-it, provider=Google" }, { "model_id": "nova-lite", "benchmark_id": "gpqa_main", "score": 42.0, "reference_url": "https://llm-stats.com/benchmarks/gpqa", "reported_setting": { "mode": "unknown", "effort": "n/a", "tools": "unknown", "sampling": "unknown", "judge": "unknown", "harness": "unknown", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown", "notes": "Model added from third-party aggregator (llm-stats.com); canonical setting not established." }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=Nova Lite, slug=nova-lite, provider=Amazon" }, { "model_id": "llama-3.1-70b", "benchmark_id": "gpqa_main", "score": 41.7, "reference_url": "https://llm-stats.com/benchmarks/gpqa", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "OLMES (Allen AI)", "prompt_style": "default", "temperature": "default", "context": "default", "notes": "Per OLMo 2 paper Table 7 Instruct (instruction-tuned variant)." }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=Llama 3.1 70B Instruct, slug=llama-3.1-70b-instruct, provider=Meta" }, { "model_id": "claude-3.5-haiku", "benchmark_id": "gpqa_main", "score": 41.6, "reference_url": "https://llm-stats.com/benchmarks/gpqa", "reported_setting": { "mode": "unknown", "effort": "n/a", "tools": "unknown", "sampling": "unknown", "judge": "unknown", "harness": "unknown", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown", "notes": "Model added from third-party aggregator (llm-stats.com); canonical setting not established." }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=Claude 3.5 Haiku, slug=claude-3-5-haiku-20241022, provider=Anthropic" }, { "model_id": "gemma-3-12b", "benchmark_id": "gpqa_main", "score": 40.9, "reference_url": "https://llm-stats.com/benchmarks/gpqa", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default", "context": "default", "notes": "Per Gemma 3 tech report Table 6 IT instruction-tuned column." }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=Gemma 3 12B, slug=gemma-3-12b-it, provider=Google" }, { "model_id": "claude-3-sonnet", "benchmark_id": "gpqa_main", "score": 40.4, "reference_url": "https://llm-stats.com/benchmarks/gpqa", "reported_setting": { "mode": "unknown", "effort": "n/a", "tools": "unknown", "sampling": "unknown", "judge": "unknown", "harness": "unknown", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown", "notes": "Model added from third-party aggregator (llm-stats.com); canonical setting not established." }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=Claude 3 Sonnet, slug=claude-3-sonnet-20240229, provider=Anthropic" }, { "model_id": "gemini-diffusion", "benchmark_id": "gpqa_main", "score": 40.4, "reference_url": "https://llm-stats.com/benchmarks/gpqa", "reported_setting": { "mode": "unknown", "effort": "n/a", "tools": "unknown", "sampling": "unknown", "judge": "unknown", "harness": "unknown", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown", "notes": "Model added from third-party aggregator (llm-stats.com); canonical setting not established." }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=Gemini Diffusion, slug=gemini-diffusion, provider=Google" }, { "model_id": "gpt-4o-mini", "benchmark_id": "gpqa_main", "score": 40.2, "reference_url": "https://llm-stats.com/benchmarks/gpqa", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none (per-bench override)", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default", "context": "default", "notes": "Per OpenAI GPT-4.1 blog table." }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=GPT-4o mini, slug=gpt-4o-mini-2024-07-18, provider=OpenAI" }, { "model_id": "nova-micro", "benchmark_id": "gpqa_main", "score": 40.0, "reference_url": "https://llm-stats.com/benchmarks/gpqa", "reported_setting": { "mode": "unknown", "effort": "n/a", "tools": "unknown", "sampling": "unknown", "judge": "unknown", "harness": "unknown", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown", "notes": "Model added from third-party aggregator (llm-stats.com); canonical setting not established." }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=Nova Micro, slug=nova-micro, provider=Amazon" }, { "model_id": "phi-3.5-moe-instruct", "benchmark_id": "gpqa_main", "score": 36.8, "reference_url": "https://llm-stats.com/benchmarks/gpqa", "reported_setting": { "mode": "unknown", "effort": "n/a", "tools": "unknown", "sampling": "unknown", "judge": "unknown", "harness": "unknown", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown", "notes": "Model added from third-party aggregator (llm-stats.com); canonical setting not established." }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=Phi-3.5-MoE-instruct, slug=phi-3.5-moe-instruct, provider=Microsoft" }, { "model_id": "qwen-2.5-7b", "benchmark_id": "gpqa_main", "score": 36.4, "reference_url": "https://llm-stats.com/benchmarks/gpqa", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "OLMES (Allen AI)", "prompt_style": "default", "temperature": "default", "context": "default", "notes": "Per OLMo 2 paper Table 7 Instruct (instruction-tuned variant)." }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=Qwen2.5 7B Instruct, slug=qwen-2.5-7b-instruct, provider=Alibaba Cloud / Qwen Team" }, { "model_id": "grok-1.5", "benchmark_id": "gpqa_main", "score": 35.9, "reference_url": "https://llm-stats.com/benchmarks/gpqa", "reported_setting": { "mode": "unknown", "effort": "n/a", "tools": "unknown", "sampling": "unknown", "judge": "unknown", "harness": "unknown", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown", "notes": "Model added from third-party aggregator (llm-stats.com); canonical setting not established." }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=Grok-1.5, slug=grok-1.5, provider=xAI" }, { "model_id": "deepseek-r1-distill-qwen-1.5b", "benchmark_id": "gpqa_main", "score": 33.8, "reference_url": "https://llm-stats.com/benchmarks/gpqa", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default", "context": "default", "notes": "Per DeepSeek R1 paper Table 15 (distilled model)." }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=DeepSeek R1 Distill Qwen 1.5B, slug=deepseek-r1-distill-qwen-1.5b, provider=DeepSeek" }, { "model_id": "claude-3-haiku", "benchmark_id": "gpqa_main", "score": 33.3, "reference_url": "https://llm-stats.com/benchmarks/gpqa", "reported_setting": { "mode": "unknown", "effort": "n/a", "tools": "unknown", "sampling": "unknown", "judge": "unknown", "harness": "unknown", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown", "notes": "Model added from third-party aggregator (llm-stats.com); canonical setting not established." }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=Claude 3 Haiku, slug=claude-3-haiku-20240307, provider=Anthropic" }, { "model_id": "gpt-3.5-turbo", "benchmark_id": "gpqa_main", "score": 30.8, "reference_url": "https://llm-stats.com/benchmarks/gpqa", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "OLMES (Allen AI)", "prompt_style": "default", "temperature": "default", "context": "default", "notes": "Per OLMo 2 paper Table 7 Instruct (instruction-tuned variant)." }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=GPT-3.5 Turbo, slug=gpt-3.5-turbo-0125, provider=OpenAI" }, { "model_id": "qwen2.5-omni-7b", "benchmark_id": "gpqa_main", "score": 30.8, "reference_url": "https://llm-stats.com/benchmarks/gpqa", "reported_setting": { "mode": "unknown", "effort": "n/a", "tools": "unknown", "sampling": "unknown", "judge": "unknown", "harness": "unknown", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown", "notes": "Model added from third-party aggregator (llm-stats.com); canonical setting not established." }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=Qwen2.5-Omni-7B, slug=qwen2.5-omni-7b, provider=Alibaba Cloud / Qwen Team" }, { "model_id": "phi-3.5-mini-instruct", "benchmark_id": "gpqa_main", "score": 30.4, "reference_url": "https://llm-stats.com/benchmarks/gpqa", "reported_setting": { "mode": "unknown", "effort": "n/a", "tools": "unknown", "sampling": "unknown", "judge": "unknown", "harness": "unknown", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown", "notes": "Model added from third-party aggregator (llm-stats.com); canonical setting not established." }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=Phi-3.5-mini-instruct, slug=phi-3.5-mini-instruct, provider=Microsoft" }, { "model_id": "qwen2-7b-instruct", "benchmark_id": "gpqa_main", "score": 25.3, "reference_url": "https://llm-stats.com/benchmarks/gpqa", "reported_setting": { "mode": "unknown", "effort": "n/a", "tools": "unknown", "sampling": "unknown", "judge": "unknown", "harness": "unknown", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown", "notes": "Model added from third-party aggregator (llm-stats.com); canonical setting not established." }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=Qwen2 7B Instruct, slug=qwen2-7b-instruct, provider=Alibaba Cloud / Qwen Team" }, { "model_id": "gemma-3-1b-it", "benchmark_id": "gpqa_main", "score": 19.2, "reference_url": "https://llm-stats.com/benchmarks/gpqa", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default", "context": "32k", "notes": "Google Gemma 3 1B instruction-tuned." }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=Gemma 3 1B, slug=gemma-3-1b-it, provider=Google" }, { "model_id": "claude-mythos", "benchmark_id": "graphwalks_bfs_128k_plus", "score": 80.0, "reference_url": "https://llm-stats.com/benchmarks/graphwalks-bfs->128k", "reported_setting": { "mode": "thinking", "effort": "adaptive thinking at max effort", "tools": "none (per-bench override)", "sampling": "pass@1 (avg 5 trials)", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default", "context": "up to 1M", "notes": "Per Mythos System Card Table 6.3.A: standard config is adaptive thinking max effort, default sampling (temperature, top_p), avg 5 trials, context up to 1M." }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=Claude Mythos Preview, slug=claude-mythos-preview, provider=Anthropic" }, { "model_id": "claude-opus-4.6", "benchmark_id": "graphwalks_bfs_128k_plus", "score": 61.5, "reference_url": "https://llm-stats.com/benchmarks/graphwalks-bfs->128k", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "none (per-bench override)", "sampling": "pass@1 (avg 5 trials)", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default", "context": "up to 1M", "notes": "Per Mythos System Card Table 6.3.A: standard config = adaptive thinking max effort, default sampling, avg 5 trials, context up to 1M." }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=Claude Opus 4.6, slug=claude-opus-4-6, provider=Anthropic" }, { "model_id": "gpt-5.5", "benchmark_id": "graphwalks_bfs_128k_plus", "score": 45.4, "reference_url": "https://llm-stats.com/benchmarks/graphwalks-bfs->128k", "reported_setting": { "mode": "thinking", "effort": "xhigh", "tools": "none (per-bench override)", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default", "context": "default", "notes": "Per OpenAI gpt-5.5 blog: reasoning effort=xhigh, research environment (may differ from production ChatGPT)." }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=GPT-5.5, slug=gpt-5.5, provider=OpenAI" }, { "model_id": "gpt-5.4", "benchmark_id": "graphwalks_bfs_128k_plus", "score": 21.4, "reference_url": "https://llm-stats.com/benchmarks/graphwalks-bfs->128k", "reported_setting": { "mode": "thinking", "effort": "xhigh", "tools": "none (per-bench override)", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default", "context": "default", "notes": "Per OpenAI gpt-5.4 blog: reasoning effort=xhigh, research environment (may differ from production ChatGPT)." }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=GPT-5.4, slug=gpt-5.4, provider=OpenAI" }, { "model_id": "claude-opus-4.6", "benchmark_id": "graphwalks_parents_128k_plus", "score": 95.4, "reference_url": "https://llm-stats.com/benchmarks/graphwalks-parents->128k", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "none (per-bench override)", "sampling": "pass@1 (avg 5 trials)", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default", "context": "up to 1M", "notes": "Per Mythos System Card Table 6.3.A: standard config = adaptive thinking max effort, default sampling, avg 5 trials, context up to 1M." }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=Claude Opus 4.6, slug=claude-opus-4-6, provider=Anthropic" }, { "model_id": "gpt-5.5", "benchmark_id": "graphwalks_parents_128k_plus", "score": 58.5, "reference_url": "https://llm-stats.com/benchmarks/graphwalks-parents->128k", "reported_setting": { "mode": "thinking", "effort": "xhigh", "tools": "none (per-bench override)", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default", "context": "default", "notes": "Per OpenAI gpt-5.5 blog: reasoning effort=xhigh, research environment (may differ from production ChatGPT)." }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=GPT-5.5, slug=gpt-5.5, provider=OpenAI" }, { "model_id": "gpt-5.4", "benchmark_id": "graphwalks_parents_128k_plus", "score": 32.4, "reference_url": "https://llm-stats.com/benchmarks/graphwalks-parents->128k", "reported_setting": { "mode": "thinking", "effort": "xhigh", "tools": "none (per-bench override)", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default", "context": "default", "notes": "Per OpenAI gpt-5.4 blog: reasoning effort=xhigh, research environment (may differ from production ChatGPT)." }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=GPT-5.4, slug=gpt-5.4, provider=OpenAI" }, { "model_id": "kimi-k2", "benchmark_id": "gsm8k", "score": 97.3, "reference_url": "https://llm-stats.com/benchmarks/gsm8k", "reported_setting": { "variant": "instruct (BP kimi-k2 is base)" }, "matches_canonical": false, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=Kimi K2 Instruct, slug=kimi-k2-instruct, provider=Moonshot AI", "candidates": [ { "score": 91.05, "reference_url": "https://research.nvidia.com/labs/nemotron/files/NVIDIA-Nemotron-3-Ultra-Technical-Report.pdf", "source_type": "tech_report", "reported_setting": { "version_metric": "8-shot CoT exact match", "tools": "none", "harness": "NeMo Evaluator", "sampling": "pass@1", "checkpoint": "base", "mode": "pretrained" }, "notes": "NVIDIA Nemotron 3 Ultra technical report, Base Table 2: Kimi K2 Base: gsm8k=91.05. Base checkpoint score mapped to the canonical post-trained model row as checkpoint=base." }, { "score": 92.1, "reference_url": "https://huggingface.co/XiaomiMiMo/MiMo-V2.5-Pro/resolve/21d1ecfecd7bd70f31be25ca49d7edd21f003659/README.md", "source_type": "official_model_card_base_checkpoint", "reported_setting": { "mode": "base pretrained checkpoint", "effort": "not applicable", "tools": "none", "sampling": "8-shot; single completion per prompt", "judge": "benchmark-native rule-based evaluator", "harness": "Xiaomi MiMo base-checkpoint evaluation", "prompt_style": "8-shot few-shot continuation", "temperature": "source does not state", "context": "source does not state", "input_modalities": "text", "trials": "one completion per item", "checkpoint": "base" }, "notes": "Base checkpoint observation attached to the canonical post-trained model ID. Base result is attached to the canonical post-trained model ID. The Xiaomi table does not state this comparator Base checkpoint's context length, and no comparator Base config is locked; no Pro context value or conflict is copied onto this observation." } ] }, { "model_id": "o1-high", "benchmark_id": "gsm8k", "score": 97.1, "reference_url": "https://llm-stats.com/benchmarks/gsm8k", "reported_setting": { "effort": "default (o1 base, not high)" }, "matches_canonical": false, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=o1, slug=o1-2024-12-17, provider=OpenAI" }, { "model_id": "gpt-4.5", "benchmark_id": "gsm8k", "score": 97.0, "reference_url": "https://llm-stats.com/benchmarks/gsm8k", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none (per-bench override)", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default", "context": "default", "notes": "Per OpenAI GPT-4.1 blog table." }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=GPT-4.5, slug=gpt-4.5, provider=OpenAI" }, { "model_id": "llama-3.1-405b-instruct", "benchmark_id": "gsm8k", "score": 96.8, "reference_url": "https://llm-stats.com/benchmarks/gsm8k", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default", "context": "default", "notes": "Per DeepSeek V3 paper Table 6." }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=Llama 3.1 405B Instruct, slug=llama-3.1-405b-instruct, provider=Meta" }, { "model_id": "claude-3.5-sonnet", "benchmark_id": "gsm8k", "score": 96.4, "reference_url": "https://llm-stats.com/benchmarks/gsm8k", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default", "context": "default", "notes": "Per DeepSeek V3 paper Table 6." }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=Claude 3.5 Sonnet, slug=claude-3-5-sonnet-20241022, provider=Anthropic" }, { "model_id": "deepseek-v2.5-0905", "benchmark_id": "gsm8k", "score": 95.1, "reference_url": "https://llm-stats.com/benchmarks/gsm8k", "reported_setting": { "snapshot": "0905" }, "matches_canonical": false, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=DeepSeek-V2.5, slug=deepseek-v2.5, provider=DeepSeek" }, { "model_id": "claude-3-opus", "benchmark_id": "gsm8k", "score": 95.0, "reference_url": "https://llm-stats.com/benchmarks/gsm8k", "reported_setting": { "mode": "unknown", "effort": "n/a", "tools": "unknown", "sampling": "unknown", "judge": "unknown", "harness": "unknown", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown", "notes": "Model added from third-party aggregator (llm-stats.com); canonical setting not established." }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=Claude 3 Opus, slug=claude-3-opus-20240229, provider=Anthropic" }, { "model_id": "qwen-2.5-14b-instruct", "benchmark_id": "gsm8k", "score": 94.8, "reference_url": "https://llm-stats.com/benchmarks/gsm8k", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default", "context": "default (128K)", "notes": "Per Phi-4 paper Table 1 baseline." }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=Qwen2.5 14B Instruct, slug=qwen-2.5-14b-instruct, provider=Alibaba Cloud / Qwen Team" }, { "model_id": "nova-lite", "benchmark_id": "gsm8k", "score": 94.5, "reference_url": "https://llm-stats.com/benchmarks/gsm8k", "reported_setting": { "mode": "unknown", "effort": "n/a", "tools": "unknown", "sampling": "unknown", "judge": "unknown", "harness": "unknown", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown", "notes": "Model added from third-party aggregator (llm-stats.com); canonical setting not established." }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=Nova Lite, slug=nova-lite, provider=Amazon" }, { "model_id": "gemma-3-12b", "benchmark_id": "gsm8k", "score": 94.4, "reference_url": "https://llm-stats.com/benchmarks/gsm8k", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default", "context": "default", "notes": "Per Gemma 3 tech report Table 6 IT instruction-tuned column." }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=Gemma 3 12B, slug=gemma-3-12b-it, provider=Google" }, { "model_id": "mistral-large-2", "benchmark_id": "gsm8k", "score": 93.0, "reference_url": "https://llm-stats.com/benchmarks/gsm8k", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default", "context": "128K", "notes": "Per Command A paper (third-party Cohere reproduction baseline)." }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=Mistral Large 2, slug=mistral-large-2-2407, provider=Mistral AI" }, { "model_id": "claude-3-sonnet", "benchmark_id": "gsm8k", "score": 92.3, "reference_url": "https://llm-stats.com/benchmarks/gsm8k", "reported_setting": { "mode": "unknown", "effort": "n/a", "tools": "unknown", "sampling": "unknown", "judge": "unknown", "harness": "unknown", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown", "notes": "Model added from third-party aggregator (llm-stats.com); canonical setting not established." }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=Claude 3 Sonnet, slug=claude-3-sonnet-20240229, provider=Anthropic" }, { "model_id": "nova-micro", "benchmark_id": "gsm8k", "score": 92.3, "reference_url": "https://llm-stats.com/benchmarks/gsm8k", "reported_setting": { "mode": "unknown", "effort": "n/a", "tools": "unknown", "sampling": "unknown", "judge": "unknown", "harness": "unknown", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown", "notes": "Model added from third-party aggregator (llm-stats.com); canonical setting not established." }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=Nova Micro, slug=nova-micro, provider=Amazon" }, { "model_id": "qwen2-72b-instruct", "benchmark_id": "gsm8k", "score": 91.1, "reference_url": "https://llm-stats.com/benchmarks/gsm8k", "reported_setting": { "mode": "unknown", "effort": "n/a", "tools": "unknown", "sampling": "unknown", "judge": "unknown", "harness": "unknown", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown", "notes": "Model added from third-party aggregator (llm-stats.com); canonical setting not established." }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=Qwen2 72B Instruct, slug=qwen2-72b-instruct, provider=Alibaba Cloud / Qwen Team" }, { "model_id": "gemini-1.5-pro", "benchmark_id": "gsm8k", "score": 90.8, "reference_url": "https://llm-stats.com/benchmarks/gsm8k", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default", "context": "default", "notes": "Per Gemini 2.0 Flash Model Card (April 15 2025): non-thinking generation, default sampling. Same canonical setting applies to 1.5 / 2.0 family." }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=Gemini 1.5 Pro, slug=gemini-1.5-pro, provider=Google" }, { "model_id": "grok-1.5", "benchmark_id": "gsm8k", "score": 90.0, "reference_url": "https://llm-stats.com/benchmarks/gsm8k", "reported_setting": { "mode": "unknown", "effort": "n/a", "tools": "unknown", "sampling": "unknown", "judge": "unknown", "harness": "unknown", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown", "notes": "Model added from third-party aggregator (llm-stats.com); canonical setting not established." }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=Grok-1.5, slug=grok-1.5, provider=xAI" }, { "model_id": "claude-3-haiku", "benchmark_id": "gsm8k", "score": 88.9, "reference_url": "https://llm-stats.com/benchmarks/gsm8k", "reported_setting": { "mode": "unknown", "effort": "n/a", "tools": "unknown", "sampling": "unknown", "judge": "unknown", "harness": "unknown", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown", "notes": "Model added from third-party aggregator (llm-stats.com); canonical setting not established." }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=Claude 3 Haiku, slug=claude-3-haiku-20240307, provider=Anthropic" }, { "model_id": "qwen2.5-omni-7b", "benchmark_id": "gsm8k", "score": 88.7, "reference_url": "https://llm-stats.com/benchmarks/gsm8k", "reported_setting": { "mode": "unknown", "effort": "n/a", "tools": "unknown", "sampling": "unknown", "judge": "unknown", "harness": "unknown", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown", "notes": "Model added from third-party aggregator (llm-stats.com); canonical setting not established." }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=Qwen2.5-Omni-7B, slug=qwen2.5-omni-7b, provider=Alibaba Cloud / Qwen Team" }, { "model_id": "phi-3.5-moe-instruct", "benchmark_id": "gsm8k", "score": 88.7, "reference_url": "https://llm-stats.com/benchmarks/gsm8k", "reported_setting": { "mode": "unknown", "effort": "n/a", "tools": "unknown", "sampling": "unknown", "judge": "unknown", "harness": "unknown", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown", "notes": "Model added from third-party aggregator (llm-stats.com); canonical setting not established." }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=Phi-3.5-MoE-instruct, slug=phi-3.5-moe-instruct, provider=Microsoft" }, { "model_id": "gemini-1.5-flash", "benchmark_id": "gsm8k", "score": 86.2, "reference_url": "https://llm-stats.com/benchmarks/gsm8k", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default", "context": "default", "notes": "Per Gemini 2.0 Flash Model Card (April 15 2025): non-thinking generation, default sampling. Same canonical setting applies to 1.5 / 2.0 family." }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=Gemini 1.5 Flash, slug=gemini-1.5-flash, provider=Google" }, { "model_id": "phi-3.5-mini-instruct", "benchmark_id": "gsm8k", "score": 86.2, "reference_url": "https://llm-stats.com/benchmarks/gsm8k", "reported_setting": { "mode": "unknown", "effort": "n/a", "tools": "unknown", "sampling": "unknown", "judge": "unknown", "harness": "unknown", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown", "notes": "Model added from third-party aggregator (llm-stats.com); canonical setting not established." }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=Phi-3.5-mini-instruct, slug=phi-3.5-mini-instruct, provider=Microsoft" }, { "model_id": "qwen2-7b-instruct", "benchmark_id": "gsm8k", "score": 82.3, "reference_url": "https://llm-stats.com/benchmarks/gsm8k", "reported_setting": { "mode": "unknown", "effort": "n/a", "tools": "unknown", "sampling": "unknown", "judge": "unknown", "harness": "unknown", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown", "notes": "Model added from third-party aggregator (llm-stats.com); canonical setting not established." }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=Qwen2 7B Instruct, slug=qwen2-7b-instruct, provider=Alibaba Cloud / Qwen Team" }, { "model_id": "granite-3.3-8b-instruct", "benchmark_id": "gsm8k", "score": 80.9, "reference_url": "https://llm-stats.com/benchmarks/gsm8k", "reported_setting": { "mode": "unknown", "effort": "n/a", "tools": "unknown", "sampling": "unknown", "judge": "unknown", "harness": "unknown", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown", "notes": "Model added from third-party aggregator (llm-stats.com); canonical setting not established." }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=Granite 3.3 8B Instruct, slug=granite-3.3-8b-instruct, provider=IBM" }, { "model_id": "granite-3.3-8b-base", "benchmark_id": "gsm8k", "score": 59.0, "reference_url": "https://llm-stats.com/benchmarks/gsm8k", "reported_setting": { "mode": "unknown", "effort": "n/a", "tools": "unknown", "sampling": "unknown", "judge": "unknown", "harness": "unknown", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown", "notes": "Model added from third-party aggregator (llm-stats.com); canonical setting not established." }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=Granite 3.3 8B Base, slug=granite-3.3-8b-base, provider=IBM" }, { "model_id": "gpt-5", "benchmark_id": "healthbench_hard", "score": 1.6, "reference_url": "https://llm-stats.com/benchmarks/healthbench-hard", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none (per-bench override)", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default", "context": "default", "notes": "Per OpenAI GPT-5 dev blog. high reasoning effort." }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=GPT-5, slug=gpt-5-2025-08-07, provider=OpenAI", "candidates": [ { "score": 41.6, "reference_url": "https://deploymentsafety.openai.com/gpt-5-6", "source_type": "model_card", "reported_setting": { "effort": "source does not state", "tools": "none", "harness": "OpenAI official HealthBench scoring" }, "notes": "OpenAI GPT-5.6 System Card table (HealthBench length-adjusted scores): gpt-5; HealthBench Hard unadjusted=41.6. System Card Table 6. Adjusted and unadjusted metrics are kept as distinct benchmark identities." } ] }, { "model_id": "claude-3-opus", "benchmark_id": "hellaswag", "score": 95.4, "reference_url": "https://llm-stats.com/benchmarks/hellaswag", "reported_setting": { "mode": "unknown", "effort": "n/a", "tools": "unknown", "sampling": "unknown", "judge": "unknown", "harness": "unknown", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown", "notes": "Model added from third-party aggregator (llm-stats.com); canonical setting not established." }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=Claude 3 Opus, slug=claude-3-opus-20240229, provider=Anthropic" }, { "model_id": "gemini-1.5-pro", "benchmark_id": "hellaswag", "score": 93.3, "reference_url": "https://llm-stats.com/benchmarks/hellaswag", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default", "context": "default", "notes": "Per Gemini 2.0 Flash Model Card (April 15 2025): non-thinking generation, default sampling. Same canonical setting applies to 1.5 / 2.0 family." }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=Gemini 1.5 Pro, slug=gemini-1.5-pro, provider=Google" }, { "model_id": "claude-3-sonnet", "benchmark_id": "hellaswag", "score": 89.0, "reference_url": "https://llm-stats.com/benchmarks/hellaswag", "reported_setting": { "mode": "unknown", "effort": "n/a", "tools": "unknown", "sampling": "unknown", "judge": "unknown", "harness": "unknown", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown", "notes": "Model added from third-party aggregator (llm-stats.com); canonical setting not established." }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=Claude 3 Sonnet, slug=claude-3-sonnet-20240229, provider=Anthropic" }, { "model_id": "qwen2-72b-instruct", "benchmark_id": "hellaswag", "score": 87.6, "reference_url": "https://llm-stats.com/benchmarks/hellaswag", "reported_setting": { "mode": "unknown", "effort": "n/a", "tools": "unknown", "sampling": "unknown", "judge": "unknown", "harness": "unknown", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown", "notes": "Model added from third-party aggregator (llm-stats.com); canonical setting not established." }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=Qwen2 72B Instruct, slug=qwen2-72b-instruct, provider=Alibaba Cloud / Qwen Team" }, { "model_id": "gemini-1.5-flash", "benchmark_id": "hellaswag", "score": 86.5, "reference_url": "https://llm-stats.com/benchmarks/hellaswag", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default", "context": "default", "notes": "Per Gemini 2.0 Flash Model Card (April 15 2025): non-thinking generation, default sampling. Same canonical setting applies to 1.5 / 2.0 family." }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=Gemini 1.5 Flash, slug=gemini-1.5-flash, provider=Google" }, { "model_id": "claude-3-haiku", "benchmark_id": "hellaswag", "score": 85.9, "reference_url": "https://llm-stats.com/benchmarks/hellaswag", "reported_setting": { "mode": "unknown", "effort": "n/a", "tools": "unknown", "sampling": "unknown", "judge": "unknown", "harness": "unknown", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown", "notes": "Model added from third-party aggregator (llm-stats.com); canonical setting not established." }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=Claude 3 Haiku, slug=claude-3-haiku-20240307, provider=Anthropic" }, { "model_id": "qwen-2.5-32b", "benchmark_id": "hellaswag", "score": 85.2, "reference_url": "https://llm-stats.com/benchmarks/hellaswag", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "OLMES (Allen AI)", "prompt_style": "default", "temperature": "default", "context": "default", "notes": "Per OLMo 2 paper Table 7 Instruct (instruction-tuned variant)." }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=Qwen2.5 32B Instruct, slug=qwen-2.5-32b-instruct, provider=Alibaba Cloud / Qwen Team" }, { "model_id": "phi-3.5-moe-instruct", "benchmark_id": "hellaswag", "score": 83.8, "reference_url": "https://llm-stats.com/benchmarks/hellaswag", "reported_setting": { "mode": "unknown", "effort": "n/a", "tools": "unknown", "sampling": "unknown", "judge": "unknown", "harness": "unknown", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown", "notes": "Model added from third-party aggregator (llm-stats.com); canonical setting not established." }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=Phi-3.5-MoE-instruct, slug=phi-3.5-moe-instruct, provider=Microsoft" }, { "model_id": "granite-3.3-8b-base", "benchmark_id": "hellaswag", "score": 80.1, "reference_url": "https://llm-stats.com/benchmarks/hellaswag", "reported_setting": { "mode": "unknown", "effort": "n/a", "tools": "unknown", "sampling": "unknown", "judge": "unknown", "harness": "unknown", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown", "notes": "Model added from third-party aggregator (llm-stats.com); canonical setting not established." }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=Granite 3.3 8B Base, slug=granite-3.3-8b-base, provider=IBM" }, { "model_id": "phi-3.5-mini-instruct", "benchmark_id": "hellaswag", "score": 69.4, "reference_url": "https://llm-stats.com/benchmarks/hellaswag", "reported_setting": { "mode": "unknown", "effort": "n/a", "tools": "unknown", "sampling": "unknown", "judge": "unknown", "harness": "unknown", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown", "notes": "Model added from third-party aggregator (llm-stats.com); canonical setting not established." }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=Phi-3.5-mini-instruct, slug=phi-3.5-mini-instruct, provider=Microsoft" }, { "model_id": "phi-4-mini", "benchmark_id": "hellaswag", "score": 69.1, "reference_url": "https://llm-stats.com/benchmarks/hellaswag", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default", "context": "default (128K)", "notes": "Per Phi-4-mini-instruct HF model card." }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=Phi 4 Mini, slug=phi-4-mini, provider=Microsoft" }, { "model_id": "gemini-1.5-flash-8b", "benchmark_id": "hiddenmath", "score": 32.8, "reference_url": "https://llm-stats.com/benchmarks/hiddenmath", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default", "context": "1M", "notes": "Per Command A paper baseline." }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=Gemini 1.5 Flash 8B, slug=gemini-1.5-flash-8b, provider=Google" }, { "model_id": "gemma-3-1b-it", "benchmark_id": "hiddenmath", "score": 15.8, "reference_url": "https://llm-stats.com/benchmarks/hiddenmath", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default", "context": "32k", "notes": "Google Gemma 3 1B instruction-tuned." }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=Gemma 3 1B, slug=gemma-3-1b-it, provider=Google" }, { "model_id": "qwen3.6-plus", "benchmark_id": "hmmt_feb_2025", "score": 96.7, "reference_url": "https://llm-stats.com/benchmarks/hmmt-2025", "reported_setting": {}, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=Qwen3.6 Plus, slug=qwen3.6-plus, provider=Alibaba Cloud / Qwen Team" }, { "model_id": "glm-5.1", "benchmark_id": "hmmt_feb_2025", "score": 94.0, "reference_url": "https://llm-stats.com/benchmarks/hmmt-2025", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none (per-bench override)", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default", "context": "default", "notes": "Per GLM-5.1 blog." }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=GLM-5.1, slug=glm-5.1, provider=Zhipu AI" }, { "model_id": "sarvam-30b", "benchmark_id": "hmmt_feb_2025", "score": 73.3, "reference_url": "https://llm-stats.com/benchmarks/hmmt-2025", "reported_setting": { "mode": "unknown", "effort": "n/a", "tools": "unknown", "sampling": "unknown", "judge": "unknown", "harness": "unknown", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown", "notes": "Model added from third-party aggregator (llm-stats.com); canonical setting not established." }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=Sarvam-30B, slug=sarvam-30b, provider=Sarvam AI" }, { "model_id": "granite-3.3-8b-base", "benchmark_id": "humaneval_plus", "score": 86.1, "reference_url": "https://llm-stats.com/benchmarks/humaneval+", "reported_setting": { "mode": "unknown", "effort": "n/a", "tools": "unknown", "sampling": "unknown", "judge": "unknown", "harness": "unknown", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown", "notes": "Model added from third-party aggregator (llm-stats.com); canonical setting not established." }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=Granite 3.3 8B Base, slug=granite-3.3-8b-base, provider=IBM" }, { "model_id": "granite-3.3-8b-instruct", "benchmark_id": "humaneval_plus", "score": 86.1, "reference_url": "https://llm-stats.com/benchmarks/humaneval+", "reported_setting": { "mode": "unknown", "effort": "n/a", "tools": "unknown", "sampling": "unknown", "judge": "unknown", "harness": "unknown", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown", "notes": "Model added from third-party aggregator (llm-stats.com); canonical setting not established." }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=Granite 3.3 8B Instruct, slug=granite-3.3-8b-instruct, provider=IBM" }, { "model_id": "qwen-2.5-32b", "benchmark_id": "humaneval_plus", "score": 52.4, "reference_url": "https://llm-stats.com/benchmarks/humaneval+", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "OLMES (Allen AI)", "prompt_style": "default", "temperature": "default", "context": "default", "notes": "Per OLMo 2 paper Table 7 Instruct (instruction-tuned variant)." }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=Qwen2.5 32B Instruct, slug=qwen-2.5-32b-instruct, provider=Alibaba Cloud / Qwen Team" }, { "model_id": "o1-high", "benchmark_id": "humaneval", "score": 88.1, "reference_url": "https://llm-stats.com/benchmarks/humaneval", "reported_setting": { "effort": "default (o1 base, not high)" }, "matches_canonical": false, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=o1, slug=o1-2024-12-17, provider=OpenAI" }, { "model_id": "mistral-small-24b", "benchmark_id": "humaneval", "score": 84.8, "reference_url": "https://llm-stats.com/benchmarks/humaneval", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "OLMES (Allen AI)", "prompt_style": "default", "temperature": "default", "context": "default", "notes": "Per OLMo 2 paper Table 7 Instruct (instruction-tuned variant)." }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=Mistral Small 3 24B Instruct, slug=mistral-small-24b-instruct-2501, provider=Mistral AI" }, { "model_id": "gemma-3-1b-it", "benchmark_id": "humaneval", "score": 41.5, "reference_url": "https://llm-stats.com/benchmarks/humaneval", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default", "context": "32k", "notes": "Google Gemma 3 1B instruction-tuned." }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=Gemma 3 1B, slug=gemma-3-1b-it, provider=Google" }, { "model_id": "gpt-4o", "benchmark_id": "hle", "score": 5.3, "reference_url": "https://llm-stats.com/benchmarks/humanity's-last-exam", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none (per-bench override)", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default", "context": "default", "notes": "Per OpenAI GPT-4.1 blog table." }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=GPT-4o, slug=gpt-4o-2024-08-06, provider=OpenAI" }, { "model_id": "claude-mythos", "benchmark_id": "hle_text", "score": 64.7, "reference_url": "https://llm-stats.com/benchmarks/humanity's-last-exam", "reported_setting": { "mode": "thinking", "effort": "adaptive thinking at max effort", "tools": "none (per-bench override)", "sampling": "pass@1 (avg 5 trials)", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default", "context": "up to 1M", "notes": "Per Mythos System Card Table 6.3.A: standard config is adaptive thinking max effort, default sampling (temperature, top_p), avg 5 trials, context up to 1M." }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=Claude Mythos Preview, slug=claude-mythos-preview, provider=Anthropic" }, { "model_id": "muse-spark", "benchmark_id": "hle_text", "score": 58.4, "reference_url": "https://llm-stats.com/benchmarks/humanity's-last-exam", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none (per-benchmark override where tools are required)", "sampling": "pass@1 (single-pass)", "judge": "rule-based or gpt-oss-120b (per benchmark)", "harness": "internal (Meta MSL)", "prompt_style": "default", "temperature": "default", "notes": "Per Meta MSL blog https://ai.meta.com/blog/introducing-muse-spark-msl/. Standard Thinking mode. Contemplating mode results flagged for review." }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=Muse Spark, slug=muse-spark, provider=Meta" }, { "model_id": "gpt-5.5", "benchmark_id": "hle_text", "score": 46.9, "reference_url": "https://huggingface.co/tencent/Hy3", "reported_setting": { "mode": "thinking", "effort": "high / highest available", "tools": "none", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official benchmark or reported provider harness", "prompt_style": "default", "temperature": "0.9 for Hy3 family; otherwise source/provider setting", "context": "256000 for Hy3 family; otherwise source/provider setting", "notes": "Tencent own test" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Hy3 official model-card appendix: HLE (no tools, text-only) = 46.9*. Tencent own testing.", "candidates": [ { "score": 57.2, "reference_url": "https://llm-stats.com/benchmarks/humanity's-last-exam", "source_type": "third_party_aggregator", "reported_setting": { "effort": "pro" }, "notes": "llm-stats.com row: model=GPT-5.5 Pro, slug=gpt-5.5-pro, provider=OpenAI [Displaced by approved Hy3 official-source audit.]" } ] }, { "model_id": "claude-opus-4.7", "benchmark_id": "hle_text", "score": 54.7, "reference_url": "https://llm-stats.com/benchmarks/humanity's-last-exam", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none (per-bench override)", "sampling": "pass@1 (avg 5 trials)", "judge": "rule-based", "harness": "Terminus-2 (for terminal-bench, thinking disabled)", "prompt_style": "default", "temperature": "default", "context": "default", "notes": "Per Anthropic Opus 4.7 announcement table. Default thinking mode with extended thinking; Terminal-Bench uses Terminus-2 harness with thinking disabled (1× guaranteed/3× ceiling, avg 5 attempts)." }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=Claude Opus 4.7, slug=claude-opus-4-7, provider=Anthropic" }, { "model_id": "claude-opus-4.6", "benchmark_id": "hle_text", "score": 53.1, "reference_url": "https://llm-stats.com/benchmarks/humanity's-last-exam", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "none (per-bench override)", "sampling": "pass@1 (avg 5 trials)", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default", "context": "up to 1M", "notes": "Per Mythos System Card Table 6.3.A: standard config = adaptive thinking max effort, default sampling, avg 5 trials, context up to 1M." }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=Claude Opus 4.6, slug=claude-opus-4-6, provider=Anthropic" }, { "model_id": "glm-5.1", "benchmark_id": "hle_text", "score": 31.0, "reference_url": "https://z.ai/blog/glm-5.2", "reported_setting": { "mode": "thinking", "effort": "official reported", "sampling": "pass@1", "prompt_style": "default", "tools": "none", "temperature": "1.0", "top_p": "0.95", "context": "default", "max_output_tokens": "163840", "judge": "GPT-5.5 medium for AIME/HMMT/IMO; benchmark-specified otherwise", "harness": "official reasoning evaluation" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "GLM-5.2 official release source: GLM-5.1 / HLE = 31.0.", "candidates": [ { "score": 34.7, "reference_url": "https://huggingface.co/deepseek-ai/DeepSeek-V4-Pro", "source_type": "model_card", "reported_setting": { "mode": "thinking", "effort": "official reported", "sampling": "pass@1", "prompt_style": "default", "tools": "none", "temperature": "1.0", "top_p": "0.95", "context": "default", "max_output_tokens": "163840", "judge": "GPT-5.5 medium for AIME/HMMT/IMO; benchmark-specified otherwise", "harness": "official reasoning evaluation" }, "notes": "DeepSeek V4-Pro cross-table reports GLM-5.1 HLE Text = 34.7 at high effort; moved from the wrong full-HLE cell." }, { "score": 27.2, "reference_url": "https://research.nvidia.com/labs/nemotron/files/NVIDIA-Nemotron-3-Ultra-Technical-Report.pdf", "source_type": "tech_report", "reported_setting": { "version_metric": "text-only / no tools / %", "tools": "none", "harness": "NeMo Evaluator", "sampling": "pass@1", "variant": "no tools" }, "notes": "NVIDIA Nemotron 3 Ultra technical report, Final Table 10: GLM-5.1: hle_text=27.2, variant=no tools." }, { "score": 50.4, "reference_url": "https://research.nvidia.com/labs/nemotron/files/NVIDIA-Nemotron-3-Ultra-Technical-Report.pdf", "source_type": "tech_report", "reported_setting": { "version_metric": "text-only / with tools / %", "tools": "benchmark tool environment", "harness": "NeMo Evaluator", "sampling": "pass@1", "variant": "with tools" }, "notes": "NVIDIA Nemotron 3 Ultra technical report, Final Table 10: GLM-5.1: hle_text=50.4, variant=with tools." } ] }, { "model_id": "gemini-3.1-pro", "benchmark_id": "hle_text", "score": 45.0, "reference_url": "https://z.ai/blog/glm-5.2", "reported_setting": { "mode": "thinking", "effort": "high", "sampling": "pass@1", "prompt_style": "default", "tools": "none", "temperature": "1.0", "top_p": "0.95", "context": "default", "max_output_tokens": "163840", "judge": "GPT-5.5 medium for AIME/HMMT/IMO; benchmark-specified otherwise", "harness": "official reasoning evaluation" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "GLM-5.2 official release source: Gemini 3.1 Pro / HLE = 45.0.", "candidates": [] }, { "model_id": "kimi-k2-thinking", "benchmark_id": "hle_text", "score": 51.0, "reference_url": "https://llm-stats.com/benchmarks/humanity's-last-exam", "reported_setting": { "mode": "thinking", "effort": "default (96k thinking-token budget for HLE/AIME/HMMT/GPQA; 128k for IMO/LCB/OJ)", "tools": "none (per-bench override)", "sampling": "pass@1 (avg@32 for AIME/HMMT no-tools)", "judge": "rule-based (HLE uses o3-mini)", "harness": "official", "prompt_style": "default", "temperature": "default", "context": "default (256K)", "notes": "Per Kimi K2 Thinking blog (moonshotai.github.io/Kimi-K2/thinking.html). Thinking model." }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=Kimi K2-Thinking-0905, slug=kimi-k2-thinking-0905, provider=Moonshot AI" }, { "model_id": "grok-4", "benchmark_id": "hle_text", "score": 50.7, "reference_url": "https://llm-stats.com/benchmarks/humanity's-last-exam", "reported_setting": { "effort": "heavy" }, "matches_canonical": false, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=Grok-4 Heavy, slug=grok-4-heavy, provider=xAI" }, { "model_id": "kimi-k2.5", "benchmark_id": "hle_text", "score": 50.2, "reference_url": "https://llm-stats.com/benchmarks/humanity's-last-exam", "reported_setting": { "mode": "thinking", "effort": "default (96k for reasoning, 64k for vision)", "tools": "none (per-bench override)", "sampling": "pass@1 (avg@32 AIME/HMMT, avg@8 GPQA, avg@3 vision, avg@4 Seal/WideSearch)", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "1.0", "context": "256K", "notes": "Per Kimi K2.5 model card (HF moonshotai/Kimi-K2.5). Thinking mode, t=1.0 top_p=0.95." }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=Kimi K2.5, slug=kimi-k2.5, provider=Moonshot AI" }, { "model_id": "claude-sonnet-4.6", "benchmark_id": "hle_text", "score": 49.0, "reference_url": "https://llm-stats.com/benchmarks/humanity's-last-exam", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "none (per-bench override)", "sampling": "pass@1 (avg 5-15 trials)", "judge": "rule-based", "harness": "Terminus-2 (terminal-bench, thinking off); official otherwise", "prompt_style": "default", "temperature": "default", "context": "default", "notes": "Per Anthropic Sonnet 4.6 announcement: max thinking effort default; Terminal-Bench=thinking off, Terminus-2; SWE-bench=avg 10 trials; BrowseComp/HLE-with-tools have specific tool configs (web search/fetch + 50k context compaction)." }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=Claude Sonnet 4.6, slug=claude-sonnet-4-6, provider=Anthropic" }, { "model_id": "deepseek-v4-pro", "benchmark_id": "hle_text", "score": 37.7, "reference_url": "https://huggingface.co/deepseek-ai/DeepSeek-V4-Pro", "reported_setting": { "mode": "thinking", "effort": "Max", "sampling": "pass@1", "prompt_style": "default", "tools": "none", "temperature": "1.0", "top_p": "0.95", "context": "default", "max_output_tokens": "163840", "judge": "GPT-5.5 medium for AIME/HMMT/IMO; benchmark-specified otherwise", "harness": "official reasoning evaluation" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "DeepSeek-V4-Pro first-party source: HLE = 37.7; GLM-5.2 table used as cross-check.", "candidates": [ { "score": 48.2, "reference_url": "https://research.nvidia.com/labs/nemotron/files/NVIDIA-Nemotron-3-Ultra-Technical-Report.pdf", "source_type": "tech_report", "reported_setting": { "version_metric": "text-only / with tools / %", "tools": "benchmark tool environment", "harness": "NeMo Evaluator", "sampling": "pass@1", "variant": "with tools" }, "notes": "NVIDIA Nemotron 3 Ultra technical report, Final Table 10: DeepSeek-V4-Pro: hle_text=48.2, variant=with tools." }, { "score": 34.5, "reference_url": "https://research.nvidia.com/labs/nemotron/files/NVIDIA-Nemotron-3-Ultra-Technical-Report.pdf", "source_type": "tech_report", "reported_setting": { "version_metric": "no tools / %", "tools": "none", "harness": "NeMo Evaluator", "sampling": "pass@1", "variant": "High", "effort": "High" }, "notes": "NVIDIA Nemotron 3 Ultra technical report, Internal Table 3: DeepSeek V4 Pro (High): hle_text=34.5, variant=High." } ] }, { "model_id": "deepseek-v4-flash", "benchmark_id": "hle_text", "score": 34.8, "reference_url": "https://huggingface.co/tencent/Hy3", "reported_setting": { "mode": "thinking", "effort": "high / highest available", "tools": "none", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official benchmark or reported provider harness", "prompt_style": "default", "temperature": "0.9 for Hy3 family; otherwise source/provider setting", "context": "256000 for Hy3 family; otherwise source/provider setting", "notes": "Comparator value not marked as Tencent own test" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Hy3 official model-card appendix: HLE (no tools, text-only) = 34.8. Comparator-reported value.", "candidates": [ { "score": 45.1, "reference_url": "https://llm-stats.com/benchmarks/humanity's-last-exam", "source_type": "third_party_aggregator", "reported_setting": { "effort": "max" }, "notes": "llm-stats.com row: model=DeepSeek-V4-Flash-Max, slug=deepseek-v4-flash-max, provider=DeepSeek [Displaced by approved Hy3 official-source audit.]" }, { "score": 32.2, "reference_url": "https://research.nvidia.com/labs/nemotron/files/NVIDIA-Nemotron-3-Ultra-Technical-Report.pdf", "source_type": "tech_report", "reported_setting": { "version_metric": "text-only / no tools / %", "tools": "none", "harness": "NeMo Evaluator", "sampling": "pass@1", "variant": "no tools" }, "notes": "NVIDIA Nemotron 3 Ultra technical report, Final Table 10: DeepSeek-V4-Flash Preview: hle_text=32.2, variant=no tools." } ] }, { "model_id": "gemini-3-flash", "benchmark_id": "hle_text", "score": 43.5, "reference_url": "https://llm-stats.com/benchmarks/humanity's-last-exam", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default", "context": "default", "notes": "Per Google DeepMind /models/gemini/flash/ table: Gemini 3 Flash Thinking, no tools by default." }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=Gemini 3 Flash, slug=gemini-3-flash-preview, provider=Google" }, { "model_id": "glm-4.7", "benchmark_id": "hle_text", "score": 42.8, "reference_url": "https://llm-stats.com/benchmarks/humanity's-last-exam", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none (per-bench override)", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default", "context": "default", "notes": "Per GLM-5 blog." }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=GLM-4.7, slug=glm-4.7, provider=Zhipu AI" }, { "model_id": "deepseek-v3.2", "benchmark_id": "hle_text", "score": 40.8, "reference_url": "https://llm-stats.com/benchmarks/humanity's-last-exam", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none (per-bench override)", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "1.0", "context": "128K", "notes": "Per DeepSeek V3.2 tech report: temperature=1.0, context=128K, thinking mode for tool-use." }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=DeepSeek-V3.2, slug=deepseek-v3.2, provider=DeepSeek" }, { "model_id": "gpt-5.4", "benchmark_id": "hle_text", "score": 39.8, "reference_url": "https://llm-stats.com/benchmarks/humanity's-last-exam", "reported_setting": { "mode": "thinking", "effort": "xhigh", "tools": "none (per-bench override)", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default", "context": "default", "notes": "Per OpenAI gpt-5.4 blog: reasoning effort=xhigh, research environment (may differ from production ChatGPT)." }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=GPT-5.4, slug=gpt-5.4, provider=OpenAI" }, { "model_id": "kimi-k2.6", "benchmark_id": "hle_text", "score": 36.4, "reference_url": "https://llm-stats.com/benchmarks/humanity's-last-exam", "reported_setting": { "mode": "thinking", "effort": "max (98304 generation tokens)", "tools": "none (per-bench override)", "sampling": "pass@1 (avg@10 coding, avg@3 vision)", "judge": "rule-based", "harness": "official (in-house SWE-agent-derived for SWE-Bench)", "prompt_style": "default", "temperature": "1.0", "context": "262144 (256K)", "notes": "Per Kimi K2.6 model card (HF moonshotai/Kimi-K2.6). Thinking mode, max effort, t=1.0 top_p=1.0." }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=Kimi K2.6, slug=kimi-k2.6, provider=Moonshot AI", "candidates": [ { "score": 34.8, "reference_url": "https://research.nvidia.com/labs/nemotron/files/NVIDIA-Nemotron-3-Ultra-Technical-Report.pdf", "source_type": "tech_report", "reported_setting": { "version_metric": "text-only / no tools / %", "tools": "none", "harness": "NeMo Evaluator", "sampling": "pass@1", "variant": "no tools" }, "notes": "NVIDIA Nemotron 3 Ultra technical report, Final Table 10: Kimi K2.6: hle_text=34.8, variant=no tools." }, { "score": 54.0, "reference_url": "https://research.nvidia.com/labs/nemotron/files/NVIDIA-Nemotron-3-Ultra-Technical-Report.pdf", "source_type": "tech_report", "reported_setting": { "version_metric": "text-only / with tools / %", "tools": "benchmark tool environment", "harness": "NeMo Evaluator", "sampling": "pass@1", "variant": "with tools" }, "notes": "NVIDIA Nemotron 3 Ultra technical report, Final Table 10: Kimi K2.6: hle_text=54, variant=with tools." } ] }, { "model_id": "deepseek-v3.2-speciale", "benchmark_id": "hle_text", "score": 30.6, "reference_url": "https://llm-stats.com/benchmarks/humanity's-last-exam", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "none (per-bench override)", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "1.0", "context": "128K", "notes": "DS V3.2-Speciale: high-compute reasoning variant of V3.2. Per V3.2 tech report Table 3." }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=DeepSeek-V3.2-Speciale, slug=deepseek-v3.2-speciale, provider=DeepSeek" }, { "model_id": "qwen3.6-plus", "benchmark_id": "hle_text", "score": 28.8, "reference_url": "https://llm-stats.com/benchmarks/humanity's-last-exam", "reported_setting": {}, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=Qwen3.6 Plus, slug=qwen3.6-plus, provider=Alibaba Cloud / Qwen Team" }, { "model_id": "gpt-5", "benchmark_id": "hle_text", "score": 24.8, "reference_url": "https://llm-stats.com/benchmarks/humanity's-last-exam", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none (per-bench override)", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default", "context": "default", "notes": "Per OpenAI GPT-5 dev blog. high reasoning effort." }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=GPT-5, slug=gpt-5-2025-08-07, provider=OpenAI" }, { "model_id": "minimax-m2.1", "benchmark_id": "hle_text", "score": 22.0, "reference_url": "https://llm-stats.com/benchmarks/humanity's-last-exam", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none (per-bench override)", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default", "context": "default", "notes": "Per MiniMax M2.5 model card (referenced as predecessor)." }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=MiniMax M2.1, slug=minimax-m2.1, provider=MiniMax" }, { "model_id": "gemini-2.5-pro", "benchmark_id": "hle_text", "score": 21.6, "reference_url": "https://llm-stats.com/benchmarks/humanity's-last-exam", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1 (single attempt)", "judge": "rule-based", "harness": "AI Studio API", "prompt_style": "default", "temperature": "default", "context": "default", "notes": "Per Gemini 2.5 Pro Model Card (GA, June 27 2025): pass@1 single attempt, AI Studio API model-id gemini-2.5-pro-preview-06-05 / gemini-2.5-pro GA, default sampling. Release date set to GA date 2025-06-27." }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=Gemini 2.5 Pro Preview 06-05, slug=gemini-2.5-pro-preview-06-05, provider=Google" }, { "model_id": "qwen3-235b", "benchmark_id": "hle_text", "score": 18.2, "reference_url": "https://llm-stats.com/benchmarks/humanity's-last-exam", "reported_setting": { "mode": "thinking", "snapshot": "2507" }, "matches_canonical": false, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=Qwen3-235B-A22B-Thinking-2507, slug=qwen3-235b-a22b-thinking-2507, provider=Alibaba Cloud / Qwen Team" }, { "model_id": "deepseek-r1", "benchmark_id": "hle_text", "score": 17.7, "reference_url": "https://llm-stats.com/benchmarks/humanity's-last-exam", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none (per-bench override)", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default", "context": "default", "notes": "Per DeepSeek R1-0528 model card." }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=DeepSeek-R1-0528, slug=deepseek-r1-0528, provider=DeepSeek" }, { "model_id": "glm-4.6", "benchmark_id": "hle_text", "score": 17.2, "reference_url": "https://llm-stats.com/benchmarks/humanity's-last-exam", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none (per-bench override)", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default", "context": "default", "notes": "Per GLM-4.7 blog." }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=GLM-4.6, slug=glm-4.6, provider=Zhipu AI" }, { "model_id": "gpt-5-mini", "benchmark_id": "hle_text", "score": 16.7, "reference_url": "https://llm-stats.com/benchmarks/humanity's-last-exam", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none (per-bench override)", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default", "context": "default", "notes": "Per OpenAI GPT-5 developer blog https://openai.com/index/introducing-gpt-5-for-developers/" }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=GPT-5 mini, slug=gpt-5-mini-2025-08-07, provider=OpenAI" }, { "model_id": "gpt-oss-120b", "benchmark_id": "hle_text", "score": 14.9, "reference_url": "https://llm-stats.com/benchmarks/humanity's-last-exam", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "with tools (per-bench override; harmony chat format default)", "sampling": "pass@1", "judge": "rule-based", "harness": "official (harmony chat format)", "prompt_style": "default", "temperature": "default", "context": "default (128K)", "notes": "Per OpenAI gpt-oss model card (arxiv:2508.10925) Table 3, high reasoning effort. Default with-tools for agentic; AIME/GPQA also has no-tools variant in same table." }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=GPT OSS 120B, slug=gpt-oss-120b, provider=OpenAI" }, { "model_id": "o4-mini-high", "benchmark_id": "hle_text", "score": 14.7, "reference_url": "https://llm-stats.com/benchmarks/humanity's-last-exam", "reported_setting": { "effort": "default (o4-mini base, not high)" }, "matches_canonical": false, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=o4-mini, slug=o4-mini, provider=OpenAI" }, { "model_id": "o3-high", "benchmark_id": "hle_text", "score": 14.7, "reference_url": "https://llm-stats.com/benchmarks/humanity's-last-exam", "reported_setting": { "effort": "default (o3 base, not high)" }, "matches_canonical": false, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=o3, slug=o3-2025-04-16, provider=OpenAI" }, { "model_id": "minimax-m2", "benchmark_id": "hle_text", "score": 12.5, "reference_url": "https://llm-stats.com/benchmarks/humanity's-last-exam", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none (per-bench override)", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default", "context": "default", "notes": "Per MiniMax M2 model card." }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=MiniMax M2, slug=minimax-m2, provider=MiniMax" }, { "model_id": "gemini-2.5-flash", "benchmark_id": "hle_text", "score": 11.0, "reference_url": "https://llm-stats.com/benchmarks/humanity's-last-exam", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none (per-bench override)", "sampling": "pass@1 (single attempt; multi-trial avg for smaller bench)", "judge": "rule-based", "harness": "AI Studio API (gemini-2.5-flash GA)", "prompt_style": "default", "temperature": "default", "context": "default (1M)", "notes": "Per Gemini 2.5 Flash Model Card (Dec 2025) GA Thinking column." }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=Gemini 2.5 Flash, slug=gemini-2.5-flash, provider=Google" }, { "model_id": "gpt-oss-20b", "benchmark_id": "hle_text", "score": 10.9, "reference_url": "https://llm-stats.com/benchmarks/humanity's-last-exam", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "with tools (per-bench override; harmony chat format default)", "sampling": "pass@1", "judge": "rule-based", "harness": "official (harmony chat format)", "prompt_style": "default", "temperature": "default", "context": "default (128K)", "notes": "Per OpenAI gpt-oss model card (arxiv:2508.10925) Table 3, high reasoning effort. Default with-tools for agentic; AIME/GPQA also has no-tools variant in same table." }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=GPT OSS 20B, slug=gpt-oss-20b, provider=OpenAI" }, { "model_id": "gpt-5-nano", "benchmark_id": "hle_text", "score": 8.7, "reference_url": "https://llm-stats.com/benchmarks/humanity's-last-exam", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none (per-bench override)", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default", "context": "default", "notes": "Per OpenAI GPT-5 developer blog https://openai.com/index/introducing-gpt-5-for-developers/" }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=GPT-5 nano, slug=gpt-5-nano-2025-08-07, provider=OpenAI" }, { "model_id": "gpt-4.1", "benchmark_id": "hle_text", "score": 5.4, "reference_url": "https://llm-stats.com/benchmarks/humanity's-last-exam", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none (per-bench override)", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default", "context": "default", "notes": "Per OpenAI GPT-5 dev blog. non-reasoning model." }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=GPT-4.1, slug=gpt-4.1-2025-04-14, provider=OpenAI" }, { "model_id": "gpt-4o", "benchmark_id": "hle_text", "score": 5.3, "reference_url": "https://llm-stats.com/benchmarks/humanity's-last-exam", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none (per-bench override)", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default", "context": "default", "notes": "Per OpenAI GPT-4.1 blog table." }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=GPT-4o, slug=gpt-4o-2024-08-06, provider=OpenAI" }, { "model_id": "kimi-k2", "benchmark_id": "hle_text", "score": 4.7, "reference_url": "https://llm-stats.com/benchmarks/humanity's-last-exam", "reported_setting": { "variant": "instruct", "snapshot": "0905" }, "matches_canonical": false, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=Kimi K2-Instruct-0905, slug=kimi-k2-instruct-0905, provider=Moonshot AI" }, { "model_id": "gpt-4.1-mini", "benchmark_id": "hle_text", "score": 3.7, "reference_url": "https://llm-stats.com/benchmarks/humanity's-last-exam", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none (per-bench override)", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default", "context": "default", "notes": "Per OpenAI GPT-5 dev blog. non-reasoning model." }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=GPT-4.1 mini, slug=gpt-4.1-mini-2025-04-14, provider=OpenAI" }, { "model_id": "gpt-oss-120b", "benchmark_id": "ifbench", "score": 69.5, "reference_url": "https://llm-stats.com/benchmarks/ifbench", "reported_setting": { "effort": "high" }, "matches_canonical": false, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=GPT OSS 120B High, slug=gpt-oss-120b-high, provider=OpenAI" }, { "model_id": "qwen3.6-plus", "benchmark_id": "ifeval", "score": 94.3, "reference_url": "https://llm-stats.com/benchmarks/ifeval", "reported_setting": {}, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=Qwen3.6 Plus, slug=qwen3.6-plus, provider=Alibaba Cloud / Qwen Team" }, { "model_id": "nova-lite", "benchmark_id": "ifeval", "score": 89.7, "reference_url": "https://llm-stats.com/benchmarks/ifeval", "reported_setting": { "mode": "unknown", "effort": "n/a", "tools": "unknown", "sampling": "unknown", "judge": "unknown", "harness": "unknown", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown", "notes": "Model added from third-party aggregator (llm-stats.com); canonical setting not established." }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=Nova Lite, slug=nova-lite, provider=Amazon" }, { "model_id": "longcat-flash-chat", "benchmark_id": "ifeval", "score": 89.6, "reference_url": "https://llm-stats.com/benchmarks/ifeval", "reported_setting": { "mode": "unknown", "effort": "n/a", "tools": "unknown", "sampling": "unknown", "judge": "unknown", "harness": "unknown", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown", "notes": "Model added from third-party aggregator (llm-stats.com); canonical setting not established." }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=LongCat-Flash-Chat, slug=longcat-flash-chat, provider=Meituan" }, { "model_id": "gemma-3-12b", "benchmark_id": "ifeval", "score": 88.9, "reference_url": "https://llm-stats.com/benchmarks/ifeval", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default", "context": "default", "notes": "Per Gemma 3 tech report Table 6 IT instruction-tuned column." }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=Gemma 3 12B, slug=gemma-3-12b-it, provider=Google" }, { "model_id": "nova-micro", "benchmark_id": "ifeval", "score": 87.2, "reference_url": "https://llm-stats.com/benchmarks/ifeval", "reported_setting": { "mode": "unknown", "effort": "n/a", "tools": "unknown", "sampling": "unknown", "judge": "unknown", "harness": "unknown", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown", "notes": "Model added from third-party aggregator (llm-stats.com); canonical setting not established." }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=Nova Micro, slug=nova-micro, provider=Amazon" }, { "model_id": "minicpm-sala", "benchmark_id": "ifeval", "score": 76.3, "reference_url": "https://llm-stats.com/benchmarks/ifeval", "reported_setting": { "mode": "unknown", "effort": "n/a", "tools": "unknown", "sampling": "unknown", "judge": "unknown", "harness": "unknown", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown", "notes": "Model added from third-party aggregator (llm-stats.com); canonical setting not established." }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=MiniCPM-SALA, slug=minicpm-sala, provider=OpenBMB" }, { "model_id": "granite-3.3-8b-instruct", "benchmark_id": "ifeval", "score": 74.8, "reference_url": "https://llm-stats.com/benchmarks/ifeval", "reported_setting": { "mode": "unknown", "effort": "n/a", "tools": "unknown", "sampling": "unknown", "judge": "unknown", "harness": "unknown", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown", "notes": "Model added from third-party aggregator (llm-stats.com); canonical setting not established." }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=Granite 3.3 8B Instruct, slug=granite-3.3-8b-instruct, provider=IBM" }, { "model_id": "granite-3.3-8b-base", "benchmark_id": "ifeval", "score": 74.8, "reference_url": "https://llm-stats.com/benchmarks/ifeval", "reported_setting": { "mode": "unknown", "effort": "n/a", "tools": "unknown", "sampling": "unknown", "judge": "unknown", "harness": "unknown", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown", "notes": "Model added from third-party aggregator (llm-stats.com); canonical setting not established." }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=Granite 3.3 8B Base, slug=granite-3.3-8b-base, provider=IBM" }, { "model_id": "pixtral-12b", "benchmark_id": "ifeval", "score": 61.3, "reference_url": "https://llm-stats.com/benchmarks/ifeval", "reported_setting": { "mode": "unknown", "effort": "n/a", "tools": "unknown", "sampling": "unknown", "judge": "unknown", "harness": "unknown", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown", "notes": "Model added from third-party aggregator (llm-stats.com); canonical setting not established." }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=Pixtral-12B, slug=pixtral-12b-2409, provider=Mistral AI" }, { "model_id": "gemma-3-27b", "benchmark_id": "infovqa", "score": 70.6, "reference_url": "https://llm-stats.com/benchmarks/infovqa", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default", "context": "default (128K)", "notes": "Per Gemma 3 tech report (arxiv:2503.19786) Table 6 IT column for 27B." }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=Gemma 3 27B, slug=gemma-3-27b-it, provider=Google" }, { "model_id": "gemma-3-12b", "benchmark_id": "infovqa", "score": 64.9, "reference_url": "https://llm-stats.com/benchmarks/infovqa", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default", "context": "default", "notes": "Per Gemma 3 tech report Table 6 IT instruction-tuned column." }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=Gemma 3 12B, slug=gemma-3-12b-it, provider=Google" }, { "model_id": "o1-high", "benchmark_id": "livebench", "score": 67.0, "reference_url": "https://llm-stats.com/benchmarks/livebench", "reported_setting": { "effort": "default (o1 base, not high)" }, "matches_canonical": false, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=o1, slug=o1-2024-12-17, provider=OpenAI" }, { "model_id": "qwen-2.5-7b", "benchmark_id": "livebench", "score": 35.9, "reference_url": "https://llm-stats.com/benchmarks/livebench", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "OLMES (Allen AI)", "prompt_style": "default", "temperature": "default", "context": "default", "notes": "Per OLMo 2 paper Table 7 Instruct (instruction-tuned variant)." }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=Qwen2.5 7B Instruct, slug=qwen-2.5-7b-instruct, provider=Alibaba Cloud / Qwen Team" }, { "model_id": "qwen2.5-omni-7b", "benchmark_id": "livebench", "score": 29.6, "reference_url": "https://llm-stats.com/benchmarks/livebench", "reported_setting": { "mode": "unknown", "effort": "n/a", "tools": "unknown", "sampling": "unknown", "judge": "unknown", "harness": "unknown", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown", "notes": "Model added from third-party aggregator (llm-stats.com); canonical setting not established." }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=Qwen2.5-Omni-7B, slug=qwen2.5-omni-7b, provider=Alibaba Cloud / Qwen Team" }, { "model_id": "muse-spark", "benchmark_id": "livecodebench_pro", "score": 80.0, "reference_url": "https://llm-stats.com/benchmarks/livecodebench-pro", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none (per-benchmark override where tools are required)", "sampling": "pass@1 (single-pass)", "judge": "rule-based or gpt-oss-120b (per benchmark)", "harness": "internal (Meta MSL)", "prompt_style": "default", "temperature": "default", "notes": "Per Meta MSL blog https://ai.meta.com/blog/introducing-muse-spark-msl/. Standard Thinking mode. Contemplating mode results flagged for review." }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=Muse Spark, slug=muse-spark, provider=Meta" }, { "model_id": "gemini-2.5-flash", "benchmark_id": "livecodebench_v5", "score": 63.9, "reference_url": "https://llm-stats.com/benchmarks/livecodebench-v5", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none (per-bench override)", "sampling": "pass@1 (single attempt; multi-trial avg for smaller bench)", "judge": "rule-based", "harness": "AI Studio API (gemini-2.5-flash GA)", "prompt_style": "default", "temperature": "default", "context": "default (1M)", "notes": "Per Gemini 2.5 Flash Model Card (Dec 2025) GA Thinking column." }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=Gemini 2.5 Flash, slug=gemini-2.5-flash, provider=Google" }, { "model_id": "minicpm-sala", "benchmark_id": "livecodebench_v5", "score": 60.5, "reference_url": "https://llm-stats.com/benchmarks/livecodebench-v5", "reported_setting": { "mode": "unknown", "effort": "n/a", "tools": "unknown", "sampling": "unknown", "judge": "unknown", "harness": "unknown", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown", "notes": "Model added from third-party aggregator (llm-stats.com); canonical setting not established." }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=MiniCPM-SALA, slug=minicpm-sala, provider=OpenBMB" }, { "model_id": "gemini-2.0-flash-lite", "benchmark_id": "livecodebench_v5", "score": 28.9, "reference_url": "https://llm-stats.com/benchmarks/livecodebench-v5", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default", "context": "default", "notes": "Per Gemini 2.0 Flash Model Card (April 15 2025): non-thinking generation, default sampling. Same canonical setting applies to 1.5 / 2.0 family." }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=Gemini 2.0 Flash-Lite, slug=gemini-2.0-flash-lite, provider=Google" }, { "model_id": "qwen3.6-plus", "benchmark_id": "livecodebench_v6", "score": 87.1, "reference_url": "https://llm-stats.com/benchmarks/livecodebench-v6", "reported_setting": {}, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=Qwen3.6 Plus, slug=qwen3.6-plus, provider=Alibaba Cloud / Qwen Team" }, { "model_id": "gpt-oss-120b", "benchmark_id": "livecodebench_v6", "score": 81.9, "reference_url": "https://llm-stats.com/benchmarks/livecodebench-v6", "reported_setting": { "effort": "high" }, "matches_canonical": false, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=GPT OSS 120B High, slug=gpt-oss-120b-high, provider=OpenAI" }, { "model_id": "qwen3-235b", "benchmark_id": "livecodebench_v6", "score": 74.1, "reference_url": "https://llm-stats.com/benchmarks/livecodebench-v6", "reported_setting": { "mode": "thinking", "snapshot": "2507" }, "matches_canonical": false, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=Qwen3-235B-A22B-Thinking-2507, slug=qwen3-235b-a22b-thinking-2507, provider=Alibaba Cloud / Qwen Team" }, { "model_id": "sarvam-30b", "benchmark_id": "livecodebench_v6", "score": 70.0, "reference_url": "https://llm-stats.com/benchmarks/livecodebench-v6", "reported_setting": { "mode": "unknown", "effort": "n/a", "tools": "unknown", "sampling": "unknown", "judge": "unknown", "harness": "unknown", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown", "notes": "Model added from third-party aggregator (llm-stats.com); canonical setting not established." }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=Sarvam-30B, slug=sarvam-30b, provider=Sarvam AI" }, { "model_id": "minicpm-sala", "benchmark_id": "livecodebench_v6", "score": 52.0, "reference_url": "https://llm-stats.com/benchmarks/livecodebench-v6", "reported_setting": { "mode": "unknown", "effort": "n/a", "tools": "unknown", "sampling": "unknown", "judge": "unknown", "harness": "unknown", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown", "notes": "Model added from third-party aggregator (llm-stats.com); canonical setting not established." }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=MiniCPM-SALA, slug=minicpm-sala, provider=OpenBMB" }, { "model_id": "minimax-m2.1", "benchmark_id": "livecodebench", "score": 78.0, "reference_url": "https://llm-stats.com/benchmarks/livecodebench", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none (per-bench override)", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default", "context": "default", "notes": "Per MiniMax M2.5 model card (referenced as predecessor)." }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=MiniMax M2.1, slug=minimax-m2.1, provider=MiniMax" }, { "model_id": "ministral-8b", "benchmark_id": "livecodebench", "score": 61.6, "reference_url": "https://llm-stats.com/benchmarks/livecodebench", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "OLMES (Allen AI)", "prompt_style": "default", "temperature": "default", "context": "default", "notes": "Per OLMo 2 paper Table 7 Instruct (instruction-tuned variant)." }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=Ministral 3 (8B Reasoning 2512), slug=ministral-8b-latest, provider=Mistral AI" }, { "model_id": "longcat-flash-chat", "benchmark_id": "livecodebench", "score": 48.0, "reference_url": "https://llm-stats.com/benchmarks/livecodebench", "reported_setting": { "mode": "unknown", "effort": "n/a", "tools": "unknown", "sampling": "unknown", "judge": "unknown", "harness": "unknown", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown", "notes": "Model added from third-party aggregator (llm-stats.com); canonical setting not established." }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=LongCat-Flash-Chat, slug=longcat-flash-chat, provider=Meituan" }, { "model_id": "gemini-diffusion", "benchmark_id": "livecodebench", "score": 30.9, "reference_url": "https://llm-stats.com/benchmarks/livecodebench", "reported_setting": { "mode": "unknown", "effort": "n/a", "tools": "unknown", "sampling": "unknown", "judge": "unknown", "harness": "unknown", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown", "notes": "Model added from third-party aggregator (llm-stats.com); canonical setting not established." }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=Gemini Diffusion, slug=gemini-diffusion, provider=Google" }, { "model_id": "qwen-2.5-7b", "benchmark_id": "livecodebench", "score": 28.7, "reference_url": "https://llm-stats.com/benchmarks/livecodebench", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "OLMES (Allen AI)", "prompt_style": "default", "temperature": "default", "context": "default", "notes": "Per OLMo 2 paper Table 7 Instruct (instruction-tuned variant)." }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=Qwen2.5 7B Instruct, slug=qwen-2.5-7b-instruct, provider=Alibaba Cloud / Qwen Team" }, { "model_id": "qwen2-7b-instruct", "benchmark_id": "livecodebench", "score": 26.6, "reference_url": "https://llm-stats.com/benchmarks/livecodebench", "reported_setting": { "mode": "unknown", "effort": "n/a", "tools": "unknown", "sampling": "unknown", "judge": "unknown", "harness": "unknown", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown", "notes": "Model added from third-party aggregator (llm-stats.com); canonical setting not established." }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=Qwen2 7B Instruct, slug=qwen2-7b-instruct, provider=Alibaba Cloud / Qwen Team" }, { "model_id": "gemma-3-1b-it", "benchmark_id": "livecodebench", "score": 1.9, "reference_url": "https://llm-stats.com/benchmarks/livecodebench", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default", "context": "32k", "notes": "Google Gemma 3 1B instruction-tuned." }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=Gemma 3 1B, slug=gemma-3-1b-it, provider=Google" }, { "model_id": "qwen3.6-plus", "benchmark_id": "longbench_v2", "score": 62.0, "reference_url": "https://llm-stats.com/benchmarks/longbench-v2", "reported_setting": {}, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=Qwen3.6 Plus, slug=qwen3.6-plus, provider=Alibaba Cloud / Qwen Team" }, { "model_id": "nova-lite", "benchmark_id": "lvbench", "score": 40.4, "reference_url": "https://llm-stats.com/benchmarks/lvbench", "reported_setting": { "mode": "unknown", "effort": "n/a", "tools": "unknown", "sampling": "unknown", "judge": "unknown", "harness": "unknown", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown", "notes": "Model added from third-party aggregator (llm-stats.com); canonical setting not established." }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=Nova Lite, slug=nova-lite, provider=Amazon" }, { "model_id": "sarvam-30b", "benchmark_id": "math_500", "score": 97.0, "reference_url": "https://llm-stats.com/benchmarks/math-500", "reported_setting": { "mode": "unknown", "effort": "n/a", "tools": "unknown", "sampling": "unknown", "judge": "unknown", "harness": "unknown", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown", "notes": "Model added from third-party aggregator (llm-stats.com); canonical setting not established." }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=Sarvam-30B, slug=sarvam-30b, provider=Sarvam AI" }, { "model_id": "longcat-flash-chat", "benchmark_id": "math_500", "score": 96.4, "reference_url": "https://llm-stats.com/benchmarks/math-500", "reported_setting": { "mode": "unknown", "effort": "n/a", "tools": "unknown", "sampling": "unknown", "judge": "unknown", "harness": "unknown", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown", "notes": "Model added from third-party aggregator (llm-stats.com); canonical setting not established." }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=LongCat-Flash-Chat, slug=longcat-flash-chat, provider=Meituan" }, { "model_id": "o1-mini", "benchmark_id": "math_500", "score": 90.0, "reference_url": "https://llm-stats.com/benchmarks/math-500", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default", "context": "default", "notes": "Per Phi-4-reasoning paper Table 1 baseline (Microsoft self-test)." }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=o1-mini, slug=o1-mini, provider=OpenAI" }, { "model_id": "granite-3.3-8b-instruct", "benchmark_id": "math_500", "score": 69.0, "reference_url": "https://llm-stats.com/benchmarks/math-500", "reported_setting": { "mode": "unknown", "effort": "n/a", "tools": "unknown", "sampling": "unknown", "judge": "unknown", "harness": "unknown", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown", "notes": "Model added from third-party aggregator (llm-stats.com); canonical setting not established." }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=Granite 3.3 8B Instruct, slug=granite-3.3-8b-instruct, provider=IBM" }, { "model_id": "granite-3.3-8b-base", "benchmark_id": "math_500", "score": 69.0, "reference_url": "https://llm-stats.com/benchmarks/math-500", "reported_setting": { "mode": "unknown", "effort": "n/a", "tools": "unknown", "sampling": "unknown", "judge": "unknown", "harness": "unknown", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown", "notes": "Model added from third-party aggregator (llm-stats.com); canonical setting not established." }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=Granite 3.3 8B Base, slug=granite-3.3-8b-base, provider=IBM" }, { "model_id": "o1-high", "benchmark_id": "math", "score": 96.4, "reference_url": "https://llm-stats.com/benchmarks/math", "reported_setting": { "effort": "default (o1 base, not high)" }, "matches_canonical": false, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=o1, slug=o1-2024-12-17, provider=OpenAI" }, { "model_id": "mistral-large-3", "benchmark_id": "math", "score": 90.4, "reference_url": "https://llm-stats.com/benchmarks/math", "reported_setting": {}, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=Mistral Large 3, slug=mistral-large-3-2509, provider=Mistral AI" }, { "model_id": "kimi-k2", "benchmark_id": "math", "score": 89.1, "reference_url": "https://llm-stats.com/benchmarks/math", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none (per-bench override)", "sampling": "pass@1 (avg of 4-64 trials per benchmark)", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default", "context": "default (256K)", "notes": "Per Kimi K2 tech report (arxiv:2507.20534) Table 3 Kimi-K2-Instruct column. Non-thinking model." }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=Kimi K2 0905, slug=kimi-k2-0905, provider=Moonshot AI", "candidates": [ { "score": 70.2, "reference_url": "https://huggingface.co/XiaomiMiMo/MiMo-V2.5-Pro/resolve/21d1ecfecd7bd70f31be25ca49d7edd21f003659/README.md", "source_type": "official_model_card_base_checkpoint", "reported_setting": { "mode": "base pretrained checkpoint", "effort": "not applicable", "tools": "none", "sampling": "4-shot; single completion per prompt", "judge": "benchmark-native rule-based evaluator", "harness": "Xiaomi MiMo base-checkpoint evaluation", "prompt_style": "4-shot few-shot continuation", "temperature": "source does not state", "context": "source does not state", "input_modalities": "text", "trials": "one completion per item", "checkpoint": "base" }, "notes": "Base checkpoint observation attached to the canonical post-trained model ID. Base result is attached to the canonical post-trained model ID. The Xiaomi table does not state this comparator Base checkpoint's context length, and no comparator Base config is locked; no Pro context value or conflict is copied onto this observation." } ] }, { "model_id": "gpt-5", "benchmark_id": "math", "score": 84.7, "reference_url": "https://llm-stats.com/benchmarks/math", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none (per-bench override)", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default", "context": "default", "notes": "Per OpenAI GPT-5 dev blog. high reasoning effort." }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=GPT-5, slug=gpt-5-2025-08-07, provider=OpenAI" }, { "model_id": "phi-4", "benchmark_id": "math", "score": 80.4, "reference_url": "https://llm-stats.com/benchmarks/math", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "OpenAI simple-evals", "prompt_style": "default", "temperature": "0.5", "context": "default (16K)", "notes": "Per Phi-4 paper (arxiv:2412.08905) Table 1, simple-evals framework, temperature 0.5." }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=Phi 4, slug=phi-4, provider=Microsoft" }, { "model_id": "qwen-2.5-14b-instruct", "benchmark_id": "math", "score": 80.0, "reference_url": "https://llm-stats.com/benchmarks/math", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default", "context": "default (128K)", "notes": "Per Phi-4 paper Table 1 baseline." }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=Qwen2.5 14B Instruct, slug=qwen-2.5-14b-instruct, provider=Alibaba Cloud / Qwen Team" }, { "model_id": "claude-3.5-sonnet", "benchmark_id": "math", "score": 78.3, "reference_url": "https://llm-stats.com/benchmarks/math", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default", "context": "default", "notes": "Per DeepSeek V3 paper Table 6." }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=Claude 3.5 Sonnet, slug=claude-3-5-sonnet-20241022, provider=Anthropic" }, { "model_id": "gpt-4o", "benchmark_id": "math", "score": 76.6, "reference_url": "https://llm-stats.com/benchmarks/math", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none (per-bench override)", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default", "context": "default", "notes": "Per OpenAI GPT-4.1 blog table." }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=GPT-4o, slug=gpt-4o-2024-05-13, provider=OpenAI" }, { "model_id": "grok-2", "benchmark_id": "math", "score": 76.1, "reference_url": "https://llm-stats.com/benchmarks/math", "reported_setting": { "mode": "unknown", "effort": "n/a", "tools": "unknown", "sampling": "unknown", "judge": "unknown", "harness": "unknown", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown", "notes": "Model added from third-party aggregator (llm-stats.com); canonical setting not established." }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=Grok-2, slug=grok-2, provider=xAI" }, { "model_id": "deepseek-v2.5-0905", "benchmark_id": "math", "score": 74.7, "reference_url": "https://llm-stats.com/benchmarks/math", "reported_setting": { "snapshot": "0905" }, "matches_canonical": false, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=DeepSeek-V2.5, slug=deepseek-v2.5, provider=DeepSeek" }, { "model_id": "llama-3.1-405b-instruct", "benchmark_id": "math", "score": 73.8, "reference_url": "https://llm-stats.com/benchmarks/math", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default", "context": "default", "notes": "Per DeepSeek V3 paper Table 6." }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=Llama 3.1 405B Instruct, slug=llama-3.1-405b-instruct, provider=Meta" }, { "model_id": "nova-lite", "benchmark_id": "math", "score": 73.3, "reference_url": "https://llm-stats.com/benchmarks/math", "reported_setting": { "mode": "unknown", "effort": "n/a", "tools": "unknown", "sampling": "unknown", "judge": "unknown", "harness": "unknown", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown", "notes": "Model added from third-party aggregator (llm-stats.com); canonical setting not established." }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=Nova Lite, slug=nova-lite, provider=Amazon" }, { "model_id": "grok-2-mini", "benchmark_id": "math", "score": 73.0, "reference_url": "https://llm-stats.com/benchmarks/math", "reported_setting": { "mode": "unknown", "effort": "n/a", "tools": "unknown", "sampling": "unknown", "judge": "unknown", "harness": "unknown", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown", "notes": "Model added from third-party aggregator (llm-stats.com); canonical setting not established." }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=Grok-2 mini, slug=grok-2-mini, provider=xAI" }, { "model_id": "gpt-4-turbo", "benchmark_id": "math", "score": 72.6, "reference_url": "https://llm-stats.com/benchmarks/math", "reported_setting": { "mode": "unknown", "effort": "n/a", "tools": "unknown", "sampling": "unknown", "judge": "unknown", "harness": "unknown", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown", "notes": "Model added from third-party aggregator (llm-stats.com); canonical setting not established." }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=GPT-4 Turbo, slug=gpt-4-turbo-2024-04-09, provider=OpenAI" }, { "model_id": "qwen3-235b", "benchmark_id": "math", "score": 71.8, "reference_url": "https://llm-stats.com/benchmarks/math", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official (Qwen-Agent / sglang)", "prompt_style": "default (enable_thinking=True)", "temperature": "default", "context": "default", "notes": "Per Qwen3 tech report (arxiv:2505.09388) Table 11 Thinking column. Output length 32768 (38912 for AIME)." }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=Qwen3 235B A22B, slug=qwen3-235b-a22b, provider=Alibaba Cloud / Qwen Team" }, { "model_id": "qwen2.5-omni-7b", "benchmark_id": "math", "score": 71.5, "reference_url": "https://llm-stats.com/benchmarks/math", "reported_setting": { "mode": "unknown", "effort": "n/a", "tools": "unknown", "sampling": "unknown", "judge": "unknown", "harness": "unknown", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown", "notes": "Model added from third-party aggregator (llm-stats.com); canonical setting not established." }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=Qwen2.5-Omni-7B, slug=qwen2.5-omni-7b, provider=Alibaba Cloud / Qwen Team" }, { "model_id": "claude-3.5-haiku", "benchmark_id": "math", "score": 69.4, "reference_url": "https://llm-stats.com/benchmarks/math", "reported_setting": { "mode": "unknown", "effort": "n/a", "tools": "unknown", "sampling": "unknown", "judge": "unknown", "harness": "unknown", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown", "notes": "Model added from third-party aggregator (llm-stats.com); canonical setting not established." }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=Claude 3.5 Haiku, slug=claude-3-5-haiku-20241022, provider=Anthropic" }, { "model_id": "nova-micro", "benchmark_id": "math", "score": 69.3, "reference_url": "https://llm-stats.com/benchmarks/math", "reported_setting": { "mode": "unknown", "effort": "n/a", "tools": "unknown", "sampling": "unknown", "judge": "unknown", "harness": "unknown", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown", "notes": "Model added from third-party aggregator (llm-stats.com); canonical setting not established." }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=Nova Micro, slug=nova-micro, provider=Amazon" }, { "model_id": "llama-4-maverick", "benchmark_id": "math", "score": 61.2, "reference_url": "https://llm-stats.com/benchmarks/math", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official-meta", "prompt_style": "zeroshot", "temperature": "0.0", "context": "default", "multimodal_input": true }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=Llama 4 Maverick, slug=llama-4-maverick, provider=Meta" }, { "model_id": "claude-3-opus", "benchmark_id": "math", "score": 60.1, "reference_url": "https://llm-stats.com/benchmarks/math", "reported_setting": { "mode": "unknown", "effort": "n/a", "tools": "unknown", "sampling": "unknown", "judge": "unknown", "harness": "unknown", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown", "notes": "Model added from third-party aggregator (llm-stats.com); canonical setting not established." }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=Claude 3 Opus, slug=claude-3-opus-20240229, provider=Anthropic" }, { "model_id": "qwen2-72b-instruct", "benchmark_id": "math", "score": 59.7, "reference_url": "https://llm-stats.com/benchmarks/math", "reported_setting": { "mode": "unknown", "effort": "n/a", "tools": "unknown", "sampling": "unknown", "judge": "unknown", "harness": "unknown", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown", "notes": "Model added from third-party aggregator (llm-stats.com); canonical setting not established." }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=Qwen2 72B Instruct, slug=qwen2-72b-instruct, provider=Alibaba Cloud / Qwen Team" }, { "model_id": "phi-3.5-moe-instruct", "benchmark_id": "math", "score": 59.5, "reference_url": "https://llm-stats.com/benchmarks/math", "reported_setting": { "mode": "unknown", "effort": "n/a", "tools": "unknown", "sampling": "unknown", "judge": "unknown", "harness": "unknown", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown", "notes": "Model added from third-party aggregator (llm-stats.com); canonical setting not established." }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=Phi-3.5-MoE-instruct, slug=phi-3.5-moe-instruct, provider=Microsoft" }, { "model_id": "gemini-1.5-flash-8b", "benchmark_id": "math", "score": 58.7, "reference_url": "https://llm-stats.com/benchmarks/math", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default", "context": "1M", "notes": "Per Command A paper baseline." }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=Gemini 1.5 Flash 8B, slug=gemini-1.5-flash-8b, provider=Google" }, { "model_id": "grok-1.5", "benchmark_id": "math", "score": 50.6, "reference_url": "https://llm-stats.com/benchmarks/math", "reported_setting": { "mode": "unknown", "effort": "n/a", "tools": "unknown", "sampling": "unknown", "judge": "unknown", "harness": "unknown", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown", "notes": "Model added from third-party aggregator (llm-stats.com); canonical setting not established." }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=Grok-1.5, slug=grok-1.5, provider=xAI" }, { "model_id": "llama-4-scout", "benchmark_id": "math", "score": 50.3, "reference_url": "https://llm-stats.com/benchmarks/math", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official-meta", "prompt_style": "zeroshot", "temperature": "0.0", "context": "default", "multimodal_input": true }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=Llama 4 Scout, slug=llama-4-scout, provider=Meta" }, { "model_id": "qwen2-7b-instruct", "benchmark_id": "math", "score": 49.6, "reference_url": "https://llm-stats.com/benchmarks/math", "reported_setting": { "mode": "unknown", "effort": "n/a", "tools": "unknown", "sampling": "unknown", "judge": "unknown", "harness": "unknown", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown", "notes": "Model added from third-party aggregator (llm-stats.com); canonical setting not established." }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=Qwen2 7B Instruct, slug=qwen2-7b-instruct, provider=Alibaba Cloud / Qwen Team" }, { "model_id": "phi-3.5-mini-instruct", "benchmark_id": "math", "score": 48.5, "reference_url": "https://llm-stats.com/benchmarks/math", "reported_setting": { "mode": "unknown", "effort": "n/a", "tools": "unknown", "sampling": "unknown", "judge": "unknown", "harness": "unknown", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown", "notes": "Model added from third-party aggregator (llm-stats.com); canonical setting not established." }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=Phi-3.5-mini-instruct, slug=phi-3.5-mini-instruct, provider=Microsoft" }, { "model_id": "pixtral-12b", "benchmark_id": "math", "score": 48.1, "reference_url": "https://llm-stats.com/benchmarks/math", "reported_setting": { "mode": "unknown", "effort": "n/a", "tools": "unknown", "sampling": "unknown", "judge": "unknown", "harness": "unknown", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown", "notes": "Model added from third-party aggregator (llm-stats.com); canonical setting not established." }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=Pixtral-12B, slug=pixtral-12b-2409, provider=Mistral AI" }, { "model_id": "gemma-3-1b-it", "benchmark_id": "math", "score": 48.0, "reference_url": "https://llm-stats.com/benchmarks/math", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default", "context": "32k", "notes": "Google Gemma 3 1B instruction-tuned." }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=Gemma 3 1B, slug=gemma-3-1b-it, provider=Google" }, { "model_id": "claude-3-sonnet", "benchmark_id": "math", "score": 43.1, "reference_url": "https://llm-stats.com/benchmarks/math", "reported_setting": { "mode": "unknown", "effort": "n/a", "tools": "unknown", "sampling": "unknown", "judge": "unknown", "harness": "unknown", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown", "notes": "Model added from third-party aggregator (llm-stats.com); canonical setting not established." }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=Claude 3 Sonnet, slug=claude-3-sonnet-20240229, provider=Anthropic" }, { "model_id": "claude-3-haiku", "benchmark_id": "math", "score": 38.9, "reference_url": "https://llm-stats.com/benchmarks/math", "reported_setting": { "mode": "unknown", "effort": "n/a", "tools": "unknown", "sampling": "unknown", "judge": "unknown", "harness": "unknown", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown", "notes": "Model added from third-party aggregator (llm-stats.com); canonical setting not established." }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=Claude 3 Haiku, slug=claude-3-haiku-20240307, provider=Anthropic" }, { "model_id": "qwen3.6-plus", "benchmark_id": "mathvision", "score": 88.0, "reference_url": "https://llm-stats.com/benchmarks/mathvision", "reported_setting": {}, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=Qwen3.6 Plus, slug=qwen3.6-plus, provider=Alibaba Cloud / Qwen Team" }, { "model_id": "qwen2.5-omni-7b", "benchmark_id": "mathvision", "score": 25.0, "reference_url": "https://llm-stats.com/benchmarks/mathvision", "reported_setting": { "mode": "unknown", "effort": "n/a", "tools": "unknown", "sampling": "unknown", "judge": "unknown", "harness": "unknown", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown", "notes": "Model added from third-party aggregator (llm-stats.com); canonical setting not established." }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=Qwen2.5-Omni-7B, slug=qwen2.5-omni-7b, provider=Alibaba Cloud / Qwen Team" }, { "model_id": "gemma-3-27b", "benchmark_id": "mathvista_mini", "score": 67.6, "reference_url": "https://llm-stats.com/benchmarks/mathvista-mini", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default", "context": "default (128K)", "notes": "Per Gemma 3 tech report (arxiv:2503.19786) Table 6 IT column for 27B." }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=Gemma 3 27B, slug=gemma-3-27b-it, provider=Google" }, { "model_id": "gemma-3-12b", "benchmark_id": "mathvista_mini", "score": 62.9, "reference_url": "https://llm-stats.com/benchmarks/mathvista-mini", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default", "context": "default", "notes": "Per Gemma 3 tech report Table 6 IT instruction-tuned column." }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=Gemma 3 12B, slug=gemma-3-12b-it, provider=Google" }, { "model_id": "o3-high", "benchmark_id": "mathvista", "score": 86.8, "reference_url": "https://llm-stats.com/benchmarks/mathvista", "reported_setting": { "effort": "default (o3 base, not high)" }, "matches_canonical": false, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=o3, slug=o3-2025-04-16, provider=OpenAI" }, { "model_id": "o4-mini-high", "benchmark_id": "mathvista", "score": 84.3, "reference_url": "https://llm-stats.com/benchmarks/mathvista", "reported_setting": { "effort": "default (o4-mini base, not high)" }, "matches_canonical": false, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=o4-mini, slug=o4-mini, provider=OpenAI" }, { "model_id": "llama-4-maverick", "benchmark_id": "mathvista", "score": 73.7, "reference_url": "https://llm-stats.com/benchmarks/mathvista", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official-meta", "prompt_style": "zeroshot", "temperature": "0.0", "context": "default", "multimodal_input": true }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=Llama 4 Maverick, slug=llama-4-maverick, provider=Meta" }, { "model_id": "llama-4-scout", "benchmark_id": "mathvista", "score": 70.7, "reference_url": "https://llm-stats.com/benchmarks/mathvista", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official-meta", "prompt_style": "zeroshot", "temperature": "0.0", "context": "default", "multimodal_input": true }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=Llama 4 Scout, slug=llama-4-scout, provider=Meta" }, { "model_id": "grok-2", "benchmark_id": "mathvista", "score": 69.0, "reference_url": "https://llm-stats.com/benchmarks/mathvista", "reported_setting": { "mode": "unknown", "effort": "n/a", "tools": "unknown", "sampling": "unknown", "judge": "unknown", "harness": "unknown", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown", "notes": "Model added from third-party aggregator (llm-stats.com); canonical setting not established." }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=Grok-2, slug=grok-2, provider=xAI" }, { "model_id": "grok-2-mini", "benchmark_id": "mathvista", "score": 68.1, "reference_url": "https://llm-stats.com/benchmarks/mathvista", "reported_setting": { "mode": "unknown", "effort": "n/a", "tools": "unknown", "sampling": "unknown", "judge": "unknown", "harness": "unknown", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown", "notes": "Model added from third-party aggregator (llm-stats.com); canonical setting not established." }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=Grok-2 mini, slug=grok-2-mini, provider=xAI" }, { "model_id": "gemini-1.5-pro", "benchmark_id": "mathvista", "score": 68.1, "reference_url": "https://llm-stats.com/benchmarks/mathvista", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default", "context": "default", "notes": "Per Gemini 2.0 Flash Model Card (April 15 2025): non-thinking generation, default sampling. Same canonical setting applies to 1.5 / 2.0 family." }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=Gemini 1.5 Pro, slug=gemini-1.5-pro, provider=Google" }, { "model_id": "qwen2.5-omni-7b", "benchmark_id": "mathvista", "score": 67.9, "reference_url": "https://llm-stats.com/benchmarks/mathvista", "reported_setting": { "mode": "unknown", "effort": "n/a", "tools": "unknown", "sampling": "unknown", "judge": "unknown", "harness": "unknown", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown", "notes": "Model added from third-party aggregator (llm-stats.com); canonical setting not established." }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=Qwen2.5-Omni-7B, slug=qwen2.5-omni-7b, provider=Alibaba Cloud / Qwen Team" }, { "model_id": "claude-3.5-sonnet", "benchmark_id": "mathvista", "score": 67.7, "reference_url": "https://llm-stats.com/benchmarks/mathvista", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default", "context": "default", "notes": "Per DeepSeek V3 paper Table 6." }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=Claude 3.5 Sonnet, slug=claude-3-5-sonnet-20241022, provider=Anthropic" }, { "model_id": "gemini-1.5-flash", "benchmark_id": "mathvista", "score": 65.8, "reference_url": "https://llm-stats.com/benchmarks/mathvista", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default", "context": "default", "notes": "Per Gemini 2.0 Flash Model Card (April 15 2025): non-thinking generation, default sampling. Same canonical setting applies to 1.5 / 2.0 family." }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=Gemini 1.5 Flash, slug=gemini-1.5-flash, provider=Google" }, { "model_id": "pixtral-12b", "benchmark_id": "mathvista", "score": 58.0, "reference_url": "https://llm-stats.com/benchmarks/mathvista", "reported_setting": { "mode": "unknown", "effort": "n/a", "tools": "unknown", "sampling": "unknown", "judge": "unknown", "harness": "unknown", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown", "notes": "Model added from third-party aggregator (llm-stats.com); canonical setting not established." }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=Pixtral-12B, slug=pixtral-12b-2409, provider=Mistral AI" }, { "model_id": "gemini-1.5-flash-8b", "benchmark_id": "mathvista", "score": 54.7, "reference_url": "https://llm-stats.com/benchmarks/mathvista", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default", "context": "1M", "notes": "Per Command A paper baseline." }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=Gemini 1.5 Flash 8B, slug=gemini-1.5-flash-8b, provider=Google" }, { "model_id": "grok-1.5", "benchmark_id": "mathvista", "score": 52.8, "reference_url": "https://llm-stats.com/benchmarks/mathvista", "reported_setting": { "mode": "unknown", "effort": "n/a", "tools": "unknown", "sampling": "unknown", "judge": "unknown", "harness": "unknown", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown", "notes": "Model added from third-party aggregator (llm-stats.com); canonical setting not established." }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=Grok-1.5, slug=grok-1.5, provider=xAI" }, { "model_id": "gpt-3.5-turbo", "benchmark_id": "mathvista", "score": 0.0, "reference_url": "https://llm-stats.com/benchmarks/mathvista", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "OLMES (Allen AI)", "prompt_style": "default", "temperature": "default", "context": "default", "notes": "Per OLMo 2 paper Table 7 Instruct (instruction-tuned variant)." }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=GPT-3.5 Turbo, slug=gpt-3.5-turbo-0125, provider=OpenAI" }, { "model_id": "qwen-2.5-32b", "benchmark_id": "mbpp_plus", "score": 67.2, "reference_url": "https://llm-stats.com/benchmarks/mbpp+", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "OLMES (Allen AI)", "prompt_style": "default", "temperature": "default", "context": "default", "notes": "Per OLMo 2 paper Table 7 Instruct (instruction-tuned variant)." }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=Qwen2.5 32B Instruct, slug=qwen-2.5-32b-instruct, provider=Alibaba Cloud / Qwen Team" }, { "model_id": "qwen-2.5-14b-instruct", "benchmark_id": "mbpp_plus", "score": 63.2, "reference_url": "https://llm-stats.com/benchmarks/mbpp+", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default", "context": "default (128K)", "notes": "Per Phi-4 paper Table 1 baseline." }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=Qwen2.5 14B Instruct, slug=qwen-2.5-14b-instruct, provider=Alibaba Cloud / Qwen Team" }, { "model_id": "qwen3.6-plus", "benchmark_id": "mcpmark", "score": 48.2, "reference_url": "https://llm-stats.com/benchmarks/mcp-mark", "reported_setting": {}, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=Qwen3.6 Plus, slug=qwen3.6-plus, provider=Alibaba Cloud / Qwen Team" }, { "model_id": "llama-4-maverick", "benchmark_id": "mgsm", "score": 92.3, "reference_url": "https://llm-stats.com/benchmarks/mgsm", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official-meta", "prompt_style": "zeroshot", "temperature": "0.0", "context": "default", "multimodal_input": true }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=Llama 4 Maverick, slug=llama-4-maverick, provider=Meta" }, { "model_id": "claude-3.5-sonnet", "benchmark_id": "mgsm", "score": 91.6, "reference_url": "https://llm-stats.com/benchmarks/mgsm", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default", "context": "default", "notes": "Per DeepSeek V3 paper Table 6." }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=Claude 3.5 Sonnet, slug=claude-3-5-sonnet-20240620, provider=Anthropic" }, { "model_id": "claude-3-opus", "benchmark_id": "mgsm", "score": 90.7, "reference_url": "https://llm-stats.com/benchmarks/mgsm", "reported_setting": { "mode": "unknown", "effort": "n/a", "tools": "unknown", "sampling": "unknown", "judge": "unknown", "harness": "unknown", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown", "notes": "Model added from third-party aggregator (llm-stats.com); canonical setting not established." }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=Claude 3 Opus, slug=claude-3-opus-20240229, provider=Anthropic" }, { "model_id": "llama-4-scout", "benchmark_id": "mgsm", "score": 90.6, "reference_url": "https://llm-stats.com/benchmarks/mgsm", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official-meta", "prompt_style": "zeroshot", "temperature": "0.0", "context": "default", "multimodal_input": true }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=Llama 4 Scout, slug=llama-4-scout, provider=Meta" }, { "model_id": "gpt-4o", "benchmark_id": "mgsm", "score": 90.5, "reference_url": "https://llm-stats.com/benchmarks/mgsm", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none (per-bench override)", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default", "context": "default", "notes": "Per OpenAI GPT-4.1 blog table." }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=GPT-4o, slug=gpt-4o-2024-05-13, provider=OpenAI" }, { "model_id": "o1-high", "benchmark_id": "mgsm", "score": 89.3, "reference_url": "https://llm-stats.com/benchmarks/mgsm", "reported_setting": { "effort": "default (o1 base, not high)" }, "matches_canonical": false, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=o1, slug=o1-2024-12-17, provider=OpenAI" }, { "model_id": "gpt-4-turbo", "benchmark_id": "mgsm", "score": 88.5, "reference_url": "https://llm-stats.com/benchmarks/mgsm", "reported_setting": { "mode": "unknown", "effort": "n/a", "tools": "unknown", "sampling": "unknown", "judge": "unknown", "harness": "unknown", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown", "notes": "Model added from third-party aggregator (llm-stats.com); canonical setting not established." }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=GPT-4 Turbo, slug=gpt-4-turbo-2024-04-09, provider=OpenAI" }, { "model_id": "gemini-1.5-pro", "benchmark_id": "mgsm", "score": 87.5, "reference_url": "https://llm-stats.com/benchmarks/mgsm", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default", "context": "default", "notes": "Per Gemini 2.0 Flash Model Card (April 15 2025): non-thinking generation, default sampling. Same canonical setting applies to 1.5 / 2.0 family." }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=Gemini 1.5 Pro, slug=gemini-1.5-pro, provider=Google" }, { "model_id": "claude-3.5-haiku", "benchmark_id": "mgsm", "score": 85.6, "reference_url": "https://llm-stats.com/benchmarks/mgsm", "reported_setting": { "mode": "unknown", "effort": "n/a", "tools": "unknown", "sampling": "unknown", "judge": "unknown", "harness": "unknown", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown", "notes": "Model added from third-party aggregator (llm-stats.com); canonical setting not established." }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=Claude 3.5 Haiku, slug=claude-3-5-haiku-20241022, provider=Anthropic" }, { "model_id": "qwen3-235b", "benchmark_id": "mgsm", "score": 83.5, "reference_url": "https://llm-stats.com/benchmarks/mgsm", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official (Qwen-Agent / sglang)", "prompt_style": "default (enable_thinking=True)", "temperature": "default", "context": "default", "notes": "Per Qwen3 tech report (arxiv:2505.09388) Table 11 Thinking column. Output length 32768 (38912 for AIME)." }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=Qwen3 235B A22B, slug=qwen3-235b-a22b, provider=Alibaba Cloud / Qwen Team" }, { "model_id": "claude-3-sonnet", "benchmark_id": "mgsm", "score": 83.5, "reference_url": "https://llm-stats.com/benchmarks/mgsm", "reported_setting": { "mode": "unknown", "effort": "n/a", "tools": "unknown", "sampling": "unknown", "judge": "unknown", "harness": "unknown", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown", "notes": "Model added from third-party aggregator (llm-stats.com); canonical setting not established." }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=Claude 3 Sonnet, slug=claude-3-sonnet-20240229, provider=Anthropic" }, { "model_id": "gemini-1.5-flash", "benchmark_id": "mgsm", "score": 82.6, "reference_url": "https://llm-stats.com/benchmarks/mgsm", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default", "context": "default", "notes": "Per Gemini 2.0 Flash Model Card (April 15 2025): non-thinking generation, default sampling. Same canonical setting applies to 1.5 / 2.0 family." }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=Gemini 1.5 Flash, slug=gemini-1.5-flash, provider=Google" }, { "model_id": "claude-3-haiku", "benchmark_id": "mgsm", "score": 75.1, "reference_url": "https://llm-stats.com/benchmarks/mgsm", "reported_setting": { "mode": "unknown", "effort": "n/a", "tools": "unknown", "sampling": "unknown", "judge": "unknown", "harness": "unknown", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown", "notes": "Model added from third-party aggregator (llm-stats.com); canonical setting not established." }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=Claude 3 Haiku, slug=claude-3-haiku-20240307, provider=Anthropic" }, { "model_id": "phi-3.5-moe-instruct", "benchmark_id": "mgsm", "score": 58.7, "reference_url": "https://llm-stats.com/benchmarks/mgsm", "reported_setting": { "mode": "unknown", "effort": "n/a", "tools": "unknown", "sampling": "unknown", "judge": "unknown", "harness": "unknown", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown", "notes": "Model added from third-party aggregator (llm-stats.com); canonical setting not established." }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=Phi-3.5-MoE-instruct, slug=phi-3.5-moe-instruct, provider=Microsoft" }, { "model_id": "gpt-3.5-turbo", "benchmark_id": "mgsm", "score": 56.3, "reference_url": "https://llm-stats.com/benchmarks/mgsm", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "OLMES (Allen AI)", "prompt_style": "default", "temperature": "default", "context": "default", "notes": "Per OLMo 2 paper Table 7 Instruct (instruction-tuned variant)." }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=GPT-3.5 Turbo, slug=gpt-3.5-turbo-0125, provider=OpenAI" }, { "model_id": "phi-3.5-mini-instruct", "benchmark_id": "mgsm", "score": 47.9, "reference_url": "https://llm-stats.com/benchmarks/mgsm", "reported_setting": { "mode": "unknown", "effort": "n/a", "tools": "unknown", "sampling": "unknown", "judge": "unknown", "harness": "unknown", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown", "notes": "Model added from third-party aggregator (llm-stats.com); canonical setting not established." }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=Phi-3.5-mini-instruct, slug=phi-3.5-mini-instruct, provider=Microsoft" }, { "model_id": "qwen3.6-plus", "benchmark_id": "mmlongbench_doc", "score": 62.0, "reference_url": "https://llm-stats.com/benchmarks/mmlongbench-doc", "reported_setting": {}, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=Qwen3.6 Plus, slug=qwen3.6-plus, provider=Alibaba Cloud / Qwen Team" }, { "model_id": "minimax-m2.1", "benchmark_id": "mmlu_pro", "score": 88.0, "reference_url": "https://llm-stats.com/benchmarks/mmlu-pro", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none (per-bench override)", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default", "context": "default", "notes": "Per MiniMax M2.5 model card (referenced as predecessor)." }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=MiniMax M2.1, slug=minimax-m2.1, provider=MiniMax" }, { "model_id": "longcat-flash-chat", "benchmark_id": "mmlu_pro", "score": 82.7, "reference_url": "https://llm-stats.com/benchmarks/mmlu-pro", "reported_setting": { "mode": "unknown", "effort": "n/a", "tools": "unknown", "sampling": "unknown", "judge": "unknown", "harness": "unknown", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown", "notes": "Model added from third-party aggregator (llm-stats.com); canonical setting not established." }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=LongCat-Flash-Chat, slug=longcat-flash-chat, provider=Meituan" }, { "model_id": "sarvam-30b", "benchmark_id": "mmlu_pro", "score": 80.0, "reference_url": "https://llm-stats.com/benchmarks/mmlu-pro", "reported_setting": { "mode": "unknown", "effort": "n/a", "tools": "unknown", "sampling": "unknown", "judge": "unknown", "harness": "unknown", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown", "notes": "Model added from third-party aggregator (llm-stats.com); canonical setting not established." }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=Sarvam-30B, slug=sarvam-30b, provider=Sarvam AI" }, { "model_id": "grok-2", "benchmark_id": "mmlu_pro", "score": 75.5, "reference_url": "https://llm-stats.com/benchmarks/mmlu-pro", "reported_setting": { "mode": "unknown", "effort": "n/a", "tools": "unknown", "sampling": "unknown", "judge": "unknown", "harness": "unknown", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown", "notes": "Model added from third-party aggregator (llm-stats.com); canonical setting not established." }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=Grok-2, slug=grok-2, provider=xAI" }, { "model_id": "grok-2-mini", "benchmark_id": "mmlu_pro", "score": 72.0, "reference_url": "https://llm-stats.com/benchmarks/mmlu-pro", "reported_setting": { "mode": "unknown", "effort": "n/a", "tools": "unknown", "sampling": "unknown", "judge": "unknown", "harness": "unknown", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown", "notes": "Model added from third-party aggregator (llm-stats.com); canonical setting not established." }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=Grok-2 mini, slug=grok-2-mini, provider=xAI" }, { "model_id": "qwen-2.5-32b", "benchmark_id": "mmlu_pro", "score": 69.0, "reference_url": "https://llm-stats.com/benchmarks/mmlu-pro", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "OLMES (Allen AI)", "prompt_style": "default", "temperature": "default", "context": "default", "notes": "Per OLMo 2 paper Table 7 Instruct (instruction-tuned variant)." }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=Qwen2.5 32B Instruct, slug=qwen-2.5-32b-instruct, provider=Alibaba Cloud / Qwen Team" }, { "model_id": "claude-3-opus", "benchmark_id": "mmlu_pro", "score": 68.5, "reference_url": "https://llm-stats.com/benchmarks/mmlu-pro", "reported_setting": { "mode": "unknown", "effort": "n/a", "tools": "unknown", "sampling": "unknown", "judge": "unknown", "harness": "unknown", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown", "notes": "Model added from third-party aggregator (llm-stats.com); canonical setting not established." }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=Claude 3 Opus, slug=claude-3-opus-20240229, provider=Anthropic" }, { "model_id": "minicpm-sala", "benchmark_id": "mmlu_pro", "score": 67.0, "reference_url": "https://llm-stats.com/benchmarks/mmlu-pro", "reported_setting": { "mode": "unknown", "effort": "n/a", "tools": "unknown", "sampling": "unknown", "judge": "unknown", "harness": "unknown", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown", "notes": "Model added from third-party aggregator (llm-stats.com); canonical setting not established." }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=MiniCPM-SALA, slug=minicpm-sala, provider=OpenBMB" }, { "model_id": "llama-3.1-70b", "benchmark_id": "mmlu_pro", "score": 66.4, "reference_url": "https://llm-stats.com/benchmarks/mmlu-pro", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "OLMES (Allen AI)", "prompt_style": "default", "temperature": "default", "context": "default", "notes": "Per OLMo 2 paper Table 7 Instruct (instruction-tuned variant)." }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=Llama 3.1 70B Instruct, slug=llama-3.1-70b-instruct, provider=Meta" }, { "model_id": "mistral-small-24b", "benchmark_id": "mmlu_pro", "score": 66.3, "reference_url": "https://llm-stats.com/benchmarks/mmlu-pro", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "OLMES (Allen AI)", "prompt_style": "default", "temperature": "default", "context": "default", "notes": "Per OLMo 2 paper Table 7 Instruct (instruction-tuned variant)." }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=Mistral Small 3 24B Instruct, slug=mistral-small-24b-instruct-2501, provider=Mistral AI" }, { "model_id": "claude-3.5-haiku", "benchmark_id": "mmlu_pro", "score": 65.0, "reference_url": "https://llm-stats.com/benchmarks/mmlu-pro", "reported_setting": { "mode": "unknown", "effort": "n/a", "tools": "unknown", "sampling": "unknown", "judge": "unknown", "harness": "unknown", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown", "notes": "Model added from third-party aggregator (llm-stats.com); canonical setting not established." }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=Claude 3.5 Haiku, slug=claude-3-5-haiku-20241022, provider=Anthropic" }, { "model_id": "qwen2-72b-instruct", "benchmark_id": "mmlu_pro", "score": 64.4, "reference_url": "https://llm-stats.com/benchmarks/mmlu-pro", "reported_setting": { "mode": "unknown", "effort": "n/a", "tools": "unknown", "sampling": "unknown", "judge": "unknown", "harness": "unknown", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown", "notes": "Model added from third-party aggregator (llm-stats.com); canonical setting not established." }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=Qwen2 72B Instruct, slug=qwen2-72b-instruct, provider=Alibaba Cloud / Qwen Team" }, { "model_id": "claude-3-sonnet", "benchmark_id": "mmlu_pro", "score": 56.8, "reference_url": "https://llm-stats.com/benchmarks/mmlu-pro", "reported_setting": { "mode": "unknown", "effort": "n/a", "tools": "unknown", "sampling": "unknown", "judge": "unknown", "harness": "unknown", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown", "notes": "Model added from third-party aggregator (llm-stats.com); canonical setting not established." }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=Claude 3 Sonnet, slug=claude-3-sonnet-20240229, provider=Anthropic" }, { "model_id": "qwen-2.5-7b", "benchmark_id": "mmlu_pro", "score": 56.3, "reference_url": "https://llm-stats.com/benchmarks/mmlu-pro", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "OLMES (Allen AI)", "prompt_style": "default", "temperature": "default", "context": "default", "notes": "Per OLMo 2 paper Table 7 Instruct (instruction-tuned variant)." }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=Qwen2.5 7B Instruct, slug=qwen-2.5-7b-instruct, provider=Alibaba Cloud / Qwen Team" }, { "model_id": "grok-1.5", "benchmark_id": "mmlu_pro", "score": 51.0, "reference_url": "https://llm-stats.com/benchmarks/mmlu-pro", "reported_setting": { "mode": "unknown", "effort": "n/a", "tools": "unknown", "sampling": "unknown", "judge": "unknown", "harness": "unknown", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown", "notes": "Model added from third-party aggregator (llm-stats.com); canonical setting not established." }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=Grok-1.5, slug=grok-1.5, provider=xAI" }, { "model_id": "phi-3.5-mini-instruct", "benchmark_id": "mmlu_pro", "score": 47.4, "reference_url": "https://llm-stats.com/benchmarks/mmlu-pro", "reported_setting": { "mode": "unknown", "effort": "n/a", "tools": "unknown", "sampling": "unknown", "judge": "unknown", "harness": "unknown", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown", "notes": "Model added from third-party aggregator (llm-stats.com); canonical setting not established." }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=Phi-3.5-mini-instruct, slug=phi-3.5-mini-instruct, provider=Microsoft" }, { "model_id": "qwen2.5-omni-7b", "benchmark_id": "mmlu_pro", "score": 47.0, "reference_url": "https://llm-stats.com/benchmarks/mmlu-pro", "reported_setting": { "mode": "unknown", "effort": "n/a", "tools": "unknown", "sampling": "unknown", "judge": "unknown", "harness": "unknown", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown", "notes": "Model added from third-party aggregator (llm-stats.com); canonical setting not established." }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=Qwen2.5-Omni-7B, slug=qwen2.5-omni-7b, provider=Alibaba Cloud / Qwen Team" }, { "model_id": "phi-3.5-moe-instruct", "benchmark_id": "mmlu_pro", "score": 45.3, "reference_url": "https://llm-stats.com/benchmarks/mmlu-pro", "reported_setting": { "mode": "unknown", "effort": "n/a", "tools": "unknown", "sampling": "unknown", "judge": "unknown", "harness": "unknown", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown", "notes": "Model added from third-party aggregator (llm-stats.com); canonical setting not established." }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=Phi-3.5-MoE-instruct, slug=phi-3.5-moe-instruct, provider=Microsoft" }, { "model_id": "qwen2-7b-instruct", "benchmark_id": "mmlu_pro", "score": 44.1, "reference_url": "https://llm-stats.com/benchmarks/mmlu-pro", "reported_setting": { "mode": "unknown", "effort": "n/a", "tools": "unknown", "sampling": "unknown", "judge": "unknown", "harness": "unknown", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown", "notes": "Model added from third-party aggregator (llm-stats.com); canonical setting not established." }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=Qwen2 7B Instruct, slug=qwen2-7b-instruct, provider=Alibaba Cloud / Qwen Team" }, { "model_id": "qwen3.6-plus", "benchmark_id": "mmlu_redux", "score": 94.5, "reference_url": "https://llm-stats.com/benchmarks/mmlu-redux", "reported_setting": {}, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=Qwen3.6 Plus, slug=qwen3.6-plus, provider=Alibaba Cloud / Qwen Team" }, { "model_id": "qwen-2.5-32b", "benchmark_id": "mmlu_redux", "score": 83.9, "reference_url": "https://llm-stats.com/benchmarks/mmlu-redux", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "OLMES (Allen AI)", "prompt_style": "default", "temperature": "default", "context": "default", "notes": "Per OLMo 2 paper Table 7 Instruct (instruction-tuned variant)." }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=Qwen2.5 32B Instruct, slug=qwen-2.5-32b-instruct, provider=Alibaba Cloud / Qwen Team" }, { "model_id": "mistral-large-3", "benchmark_id": "mmlu_redux", "score": 82.0, "reference_url": "https://llm-stats.com/benchmarks/mmlu-redux", "reported_setting": {}, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=Mistral Large 3, slug=mistral-large-3-2509, provider=Mistral AI" }, { "model_id": "qwen-2.5-14b-instruct", "benchmark_id": "mmlu_redux", "score": 80.0, "reference_url": "https://llm-stats.com/benchmarks/mmlu-redux", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default", "context": "default (128K)", "notes": "Per Phi-4 paper Table 1 baseline." }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=Qwen2.5 14B Instruct, slug=qwen-2.5-14b-instruct, provider=Alibaba Cloud / Qwen Team" }, { "model_id": "qwen-2.5-7b", "benchmark_id": "mmlu_redux", "score": 75.4, "reference_url": "https://llm-stats.com/benchmarks/mmlu-redux", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "OLMES (Allen AI)", "prompt_style": "default", "temperature": "default", "context": "default", "notes": "Per OLMo 2 paper Table 7 Instruct (instruction-tuned variant)." }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=Qwen2.5 7B Instruct, slug=qwen-2.5-7b-instruct, provider=Alibaba Cloud / Qwen Team" }, { "model_id": "qwen2.5-omni-7b", "benchmark_id": "mmlu_redux", "score": 71.0, "reference_url": "https://llm-stats.com/benchmarks/mmlu-redux", "reported_setting": { "mode": "unknown", "effort": "n/a", "tools": "unknown", "sampling": "unknown", "judge": "unknown", "harness": "unknown", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown", "notes": "Model added from third-party aggregator (llm-stats.com); canonical setting not established." }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=Qwen2.5-Omni-7B, slug=qwen2.5-omni-7b, provider=Alibaba Cloud / Qwen Team" }, { "model_id": "longcat-flash-chat", "benchmark_id": "mmlu", "score": 89.7, "reference_url": "https://llm-stats.com/benchmarks/mmlu", "reported_setting": { "mode": "unknown", "effort": "n/a", "tools": "unknown", "sampling": "unknown", "judge": "unknown", "harness": "unknown", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown", "notes": "Model added from third-party aggregator (llm-stats.com); canonical setting not established." }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=LongCat-Flash-Chat, slug=longcat-flash-chat, provider=Meituan" }, { "model_id": "grok-2", "benchmark_id": "mmlu", "score": 87.5, "reference_url": "https://llm-stats.com/benchmarks/mmlu", "reported_setting": { "mode": "unknown", "effort": "n/a", "tools": "unknown", "sampling": "unknown", "judge": "unknown", "harness": "unknown", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown", "notes": "Model added from third-party aggregator (llm-stats.com); canonical setting not established." }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=Grok-2, slug=grok-2, provider=xAI" }, { "model_id": "claude-3-opus", "benchmark_id": "mmlu", "score": 86.8, "reference_url": "https://llm-stats.com/benchmarks/mmlu", "reported_setting": { "mode": "unknown", "effort": "n/a", "tools": "unknown", "sampling": "unknown", "judge": "unknown", "harness": "unknown", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown", "notes": "Model added from third-party aggregator (llm-stats.com); canonical setting not established." }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=Claude 3 Opus, slug=claude-3-opus-20240229, provider=Anthropic" }, { "model_id": "gpt-4-turbo", "benchmark_id": "mmlu", "score": 86.5, "reference_url": "https://llm-stats.com/benchmarks/mmlu", "reported_setting": { "mode": "unknown", "effort": "n/a", "tools": "unknown", "sampling": "unknown", "judge": "unknown", "harness": "unknown", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown", "notes": "Model added from third-party aggregator (llm-stats.com); canonical setting not established." }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=GPT-4 Turbo, slug=gpt-4-turbo-2024-04-09, provider=OpenAI" }, { "model_id": "grok-2-mini", "benchmark_id": "mmlu", "score": 86.2, "reference_url": "https://llm-stats.com/benchmarks/mmlu", "reported_setting": { "mode": "unknown", "effort": "n/a", "tools": "unknown", "sampling": "unknown", "judge": "unknown", "harness": "unknown", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown", "notes": "Model added from third-party aggregator (llm-stats.com); canonical setting not established." }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=Grok-2 mini, slug=grok-2-mini, provider=xAI" }, { "model_id": "gemini-1.5-pro", "benchmark_id": "mmlu", "score": 85.9, "reference_url": "https://llm-stats.com/benchmarks/mmlu", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default", "context": "default", "notes": "Per Gemini 2.0 Flash Model Card (April 15 2025): non-thinking generation, default sampling. Same canonical setting applies to 1.5 / 2.0 family." }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=Gemini 1.5 Pro, slug=gemini-1.5-pro, provider=Google" }, { "model_id": "o1-mini", "benchmark_id": "mmlu", "score": 85.2, "reference_url": "https://llm-stats.com/benchmarks/mmlu", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default", "context": "default", "notes": "Per Phi-4-reasoning paper Table 1 baseline (Microsoft self-test)." }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=o1-mini, slug=o1-mini, provider=OpenAI" }, { "model_id": "sarvam-30b", "benchmark_id": "mmlu", "score": 85.1, "reference_url": "https://llm-stats.com/benchmarks/mmlu", "reported_setting": { "mode": "unknown", "effort": "n/a", "tools": "unknown", "sampling": "unknown", "judge": "unknown", "harness": "unknown", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown", "notes": "Model added from third-party aggregator (llm-stats.com); canonical setting not established." }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=Sarvam-30B, slug=sarvam-30b, provider=Sarvam AI" }, { "model_id": "qwen2-72b-instruct", "benchmark_id": "mmlu", "score": 82.3, "reference_url": "https://llm-stats.com/benchmarks/mmlu", "reported_setting": { "mode": "unknown", "effort": "n/a", "tools": "unknown", "sampling": "unknown", "judge": "unknown", "harness": "unknown", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown", "notes": "Model added from third-party aggregator (llm-stats.com); canonical setting not established." }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=Qwen2 72B Instruct, slug=qwen2-72b-instruct, provider=Alibaba Cloud / Qwen Team" }, { "model_id": "grok-1.5", "benchmark_id": "mmlu", "score": 81.3, "reference_url": "https://llm-stats.com/benchmarks/mmlu", "reported_setting": { "mode": "unknown", "effort": "n/a", "tools": "unknown", "sampling": "unknown", "judge": "unknown", "harness": "unknown", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown", "notes": "Model added from third-party aggregator (llm-stats.com); canonical setting not established." }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=Grok-1.5, slug=grok-1.5, provider=xAI" }, { "model_id": "nova-lite", "benchmark_id": "mmlu", "score": 80.5, "reference_url": "https://llm-stats.com/benchmarks/mmlu", "reported_setting": { "mode": "unknown", "effort": "n/a", "tools": "unknown", "sampling": "unknown", "judge": "unknown", "harness": "unknown", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown", "notes": "Model added from third-party aggregator (llm-stats.com); canonical setting not established." }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=Nova Lite, slug=nova-lite, provider=Amazon" }, { "model_id": "claude-3-sonnet", "benchmark_id": "mmlu", "score": 79.0, "reference_url": "https://llm-stats.com/benchmarks/mmlu", "reported_setting": { "mode": "unknown", "effort": "n/a", "tools": "unknown", "sampling": "unknown", "judge": "unknown", "harness": "unknown", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown", "notes": "Model added from third-party aggregator (llm-stats.com); canonical setting not established." }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=Claude 3 Sonnet, slug=claude-3-sonnet-20240229, provider=Anthropic" }, { "model_id": "gemini-1.5-flash", "benchmark_id": "mmlu", "score": 78.9, "reference_url": "https://llm-stats.com/benchmarks/mmlu", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default", "context": "default", "notes": "Per Gemini 2.0 Flash Model Card (April 15 2025): non-thinking generation, default sampling. Same canonical setting applies to 1.5 / 2.0 family." }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=Gemini 1.5 Flash, slug=gemini-1.5-flash, provider=Google" }, { "model_id": "phi-3.5-moe-instruct", "benchmark_id": "mmlu", "score": 78.9, "reference_url": "https://llm-stats.com/benchmarks/mmlu", "reported_setting": { "mode": "unknown", "effort": "n/a", "tools": "unknown", "sampling": "unknown", "judge": "unknown", "harness": "unknown", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown", "notes": "Model added from third-party aggregator (llm-stats.com); canonical setting not established." }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=Phi-3.5-MoE-instruct, slug=phi-3.5-moe-instruct, provider=Microsoft" }, { "model_id": "nova-micro", "benchmark_id": "mmlu", "score": 77.6, "reference_url": "https://llm-stats.com/benchmarks/mmlu", "reported_setting": { "mode": "unknown", "effort": "n/a", "tools": "unknown", "sampling": "unknown", "judge": "unknown", "harness": "unknown", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown", "notes": "Model added from third-party aggregator (llm-stats.com); canonical setting not established." }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=Nova Micro, slug=nova-micro, provider=Amazon" }, { "model_id": "claude-3-haiku", "benchmark_id": "mmlu", "score": 75.2, "reference_url": "https://llm-stats.com/benchmarks/mmlu", "reported_setting": { "mode": "unknown", "effort": "n/a", "tools": "unknown", "sampling": "unknown", "judge": "unknown", "harness": "unknown", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown", "notes": "Model added from third-party aggregator (llm-stats.com); canonical setting not established." }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=Claude 3 Haiku, slug=claude-3-haiku-20240307, provider=Anthropic" }, { "model_id": "qwen2-7b-instruct", "benchmark_id": "mmlu", "score": 70.5, "reference_url": "https://llm-stats.com/benchmarks/mmlu", "reported_setting": { "mode": "unknown", "effort": "n/a", "tools": "unknown", "sampling": "unknown", "judge": "unknown", "harness": "unknown", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown", "notes": "Model added from third-party aggregator (llm-stats.com); canonical setting not established." }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=Qwen2 7B Instruct, slug=qwen2-7b-instruct, provider=Alibaba Cloud / Qwen Team" }, { "model_id": "pixtral-12b", "benchmark_id": "mmlu", "score": 69.2, "reference_url": "https://llm-stats.com/benchmarks/mmlu", "reported_setting": { "mode": "unknown", "effort": "n/a", "tools": "unknown", "sampling": "unknown", "judge": "unknown", "harness": "unknown", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown", "notes": "Model added from third-party aggregator (llm-stats.com); canonical setting not established." }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=Pixtral-12B, slug=pixtral-12b-2409, provider=Mistral AI" }, { "model_id": "phi-3.5-mini-instruct", "benchmark_id": "mmlu", "score": 69.0, "reference_url": "https://llm-stats.com/benchmarks/mmlu", "reported_setting": { "mode": "unknown", "effort": "n/a", "tools": "unknown", "sampling": "unknown", "judge": "unknown", "harness": "unknown", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown", "notes": "Model added from third-party aggregator (llm-stats.com); canonical setting not established." }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=Phi-3.5-mini-instruct, slug=phi-3.5-mini-instruct, provider=Microsoft" }, { "model_id": "granite-3.3-8b-instruct", "benchmark_id": "mmlu", "score": 65.5, "reference_url": "https://llm-stats.com/benchmarks/mmlu", "reported_setting": { "mode": "unknown", "effort": "n/a", "tools": "unknown", "sampling": "unknown", "judge": "unknown", "harness": "unknown", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown", "notes": "Model added from third-party aggregator (llm-stats.com); canonical setting not established." }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=Granite 3.3 8B Instruct, slug=granite-3.3-8b-instruct, provider=IBM" }, { "model_id": "granite-3.3-8b-base", "benchmark_id": "mmlu", "score": 63.9, "reference_url": "https://llm-stats.com/benchmarks/mmlu", "reported_setting": { "mode": "unknown", "effort": "n/a", "tools": "unknown", "sampling": "unknown", "judge": "unknown", "harness": "unknown", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown", "notes": "Model added from third-party aggregator (llm-stats.com); canonical setting not established." }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=Granite 3.3 8B Base, slug=granite-3.3-8b-base, provider=IBM" }, { "model_id": "qwen3.6-plus", "benchmark_id": "mmmlu", "score": 89.5, "reference_url": "https://llm-stats.com/benchmarks/mmmlu", "reported_setting": {}, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=Qwen3.6 Plus, slug=qwen3.6-plus, provider=Alibaba Cloud / Qwen Team" }, { "model_id": "mistral-large-3", "benchmark_id": "mmmlu", "score": 74.2, "reference_url": "https://llm-stats.com/benchmarks/mmmlu", "reported_setting": {}, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=Mistral Large 3, slug=mistral-large-3-2509, provider=Mistral AI" }, { "model_id": "phi-3.5-moe-instruct", "benchmark_id": "mmmlu", "score": 69.9, "reference_url": "https://llm-stats.com/benchmarks/mmmlu", "reported_setting": { "mode": "unknown", "effort": "n/a", "tools": "unknown", "sampling": "unknown", "judge": "unknown", "harness": "unknown", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown", "notes": "Model added from third-party aggregator (llm-stats.com); canonical setting not established." }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=Phi-3.5-MoE-instruct, slug=phi-3.5-moe-instruct, provider=Microsoft" }, { "model_id": "phi-3.5-mini-instruct", "benchmark_id": "mmmlu", "score": 55.4, "reference_url": "https://llm-stats.com/benchmarks/mmmlu", "reported_setting": { "mode": "unknown", "effort": "n/a", "tools": "unknown", "sampling": "unknown", "judge": "unknown", "harness": "unknown", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown", "notes": "Model added from third-party aggregator (llm-stats.com); canonical setting not established." }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=Phi-3.5-mini-instruct, slug=phi-3.5-mini-instruct, provider=Microsoft" }, { "model_id": "qwen3.6-plus", "benchmark_id": "mmmu_pro", "score": 78.8, "reference_url": "https://llm-stats.com/benchmarks/mmmu-pro", "reported_setting": {}, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=Qwen3.6 Plus, slug=qwen3.6-plus, provider=Alibaba Cloud / Qwen Team" }, { "model_id": "gpt-4o", "benchmark_id": "mmmu_pro", "score": 59.9, "reference_url": "https://llm-stats.com/benchmarks/mmmu-pro", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none (per-bench override)", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default", "context": "default", "notes": "Per OpenAI GPT-4.1 blog table." }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=GPT-4o, slug=gpt-4o-2024-08-06, provider=OpenAI" }, { "model_id": "llama-4-maverick", "benchmark_id": "mmmu_pro", "score": 59.6, "reference_url": "https://llm-stats.com/benchmarks/mmmu-pro", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official-meta", "prompt_style": "zeroshot", "temperature": "0.0", "context": "default", "multimodal_input": true }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=Llama 4 Maverick, slug=llama-4-maverick, provider=Meta" }, { "model_id": "qwen2.5-omni-7b", "benchmark_id": "mmmu_pro", "score": 36.6, "reference_url": "https://llm-stats.com/benchmarks/mmmu-pro", "reported_setting": { "mode": "unknown", "effort": "n/a", "tools": "unknown", "sampling": "unknown", "judge": "unknown", "harness": "unknown", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown", "notes": "Model added from third-party aggregator (llm-stats.com); canonical setting not established." }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=Qwen2.5-Omni-7B, slug=qwen2.5-omni-7b, provider=Alibaba Cloud / Qwen Team" }, { "model_id": "qwen3.6-plus", "benchmark_id": "mmmu", "score": 86.0, "reference_url": "https://llm-stats.com/benchmarks/mmmu", "reported_setting": {}, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=Qwen3.6 Plus, slug=qwen3.6-plus, provider=Alibaba Cloud / Qwen Team" }, { "model_id": "grok-2", "benchmark_id": "mmmu", "score": 66.1, "reference_url": "https://llm-stats.com/benchmarks/mmmu", "reported_setting": { "mode": "unknown", "effort": "n/a", "tools": "unknown", "sampling": "unknown", "judge": "unknown", "harness": "unknown", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown", "notes": "Model added from third-party aggregator (llm-stats.com); canonical setting not established." }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=Grok-2, slug=grok-2, provider=xAI" }, { "model_id": "grok-2-mini", "benchmark_id": "mmmu", "score": 63.2, "reference_url": "https://llm-stats.com/benchmarks/mmmu", "reported_setting": { "mode": "unknown", "effort": "n/a", "tools": "unknown", "sampling": "unknown", "judge": "unknown", "harness": "unknown", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown", "notes": "Model added from third-party aggregator (llm-stats.com); canonical setting not established." }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=Grok-2 mini, slug=grok-2-mini, provider=xAI" }, { "model_id": "qwen2.5-omni-7b", "benchmark_id": "mmmu", "score": 59.2, "reference_url": "https://llm-stats.com/benchmarks/mmmu", "reported_setting": { "mode": "unknown", "effort": "n/a", "tools": "unknown", "sampling": "unknown", "judge": "unknown", "harness": "unknown", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown", "notes": "Model added from third-party aggregator (llm-stats.com); canonical setting not established." }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=Qwen2.5-Omni-7B, slug=qwen2.5-omni-7b, provider=Alibaba Cloud / Qwen Team" }, { "model_id": "nova-lite", "benchmark_id": "mmmu", "score": 56.2, "reference_url": "https://llm-stats.com/benchmarks/mmmu", "reported_setting": { "mode": "unknown", "effort": "n/a", "tools": "unknown", "sampling": "unknown", "judge": "unknown", "harness": "unknown", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown", "notes": "Model added from third-party aggregator (llm-stats.com); canonical setting not established." }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=Nova Lite, slug=nova-lite, provider=Amazon" }, { "model_id": "gemini-1.5-flash-8b", "benchmark_id": "mmmu", "score": 53.7, "reference_url": "https://llm-stats.com/benchmarks/mmmu", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default", "context": "1M", "notes": "Per Command A paper baseline." }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=Gemini 1.5 Flash 8B, slug=gemini-1.5-flash-8b, provider=Google" }, { "model_id": "grok-1.5", "benchmark_id": "mmmu", "score": 53.6, "reference_url": "https://llm-stats.com/benchmarks/mmmu", "reported_setting": { "mode": "unknown", "effort": "n/a", "tools": "unknown", "sampling": "unknown", "judge": "unknown", "harness": "unknown", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown", "notes": "Model added from third-party aggregator (llm-stats.com); canonical setting not established." }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=Grok-1.5, slug=grok-1.5, provider=xAI" }, { "model_id": "pixtral-12b", "benchmark_id": "mmmu", "score": 52.5, "reference_url": "https://llm-stats.com/benchmarks/mmmu", "reported_setting": { "mode": "unknown", "effort": "n/a", "tools": "unknown", "sampling": "unknown", "judge": "unknown", "harness": "unknown", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown", "notes": "Model added from third-party aggregator (llm-stats.com); canonical setting not established." }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=Pixtral-12B, slug=pixtral-12b-2409, provider=Mistral AI" }, { "model_id": "gpt-3.5-turbo", "benchmark_id": "mmmu", "score": 0.0, "reference_url": "https://llm-stats.com/benchmarks/mmmu", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "OLMES (Allen AI)", "prompt_style": "default", "temperature": "default", "context": "default", "notes": "Per OLMo 2 paper Table 7 Instruct (instruction-tuned variant)." }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=GPT-3.5 Turbo, slug=gpt-3.5-turbo-0125, provider=OpenAI" }, { "model_id": "qwen3.6-plus", "benchmark_id": "mmstar", "score": 83.3, "reference_url": "https://llm-stats.com/benchmarks/mmstar", "reported_setting": {}, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=Qwen3.6 Plus, slug=qwen3.6-plus, provider=Alibaba Cloud / Qwen Team" }, { "model_id": "qwen2.5-omni-7b", "benchmark_id": "mmstar", "score": 64.0, "reference_url": "https://llm-stats.com/benchmarks/mmstar", "reported_setting": { "mode": "unknown", "effort": "n/a", "tools": "unknown", "sampling": "unknown", "judge": "unknown", "harness": "unknown", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown", "notes": "Model added from third-party aggregator (llm-stats.com); canonical setting not established." }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=Qwen2.5-Omni-7B, slug=qwen2.5-omni-7b, provider=Alibaba Cloud / Qwen Team" }, { "model_id": "deepseek-v4-pro", "benchmark_id": "mrcr_v2_2needle_1m", "score": 83.5, "reference_url": "https://llm-stats.com/benchmarks/mrcr-1m", "reported_setting": { "effort": "max" }, "matches_canonical": false, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=DeepSeek-V4-Pro-Max, slug=deepseek-v4-pro-max, provider=DeepSeek" }, { "model_id": "deepseek-v4-flash", "benchmark_id": "mrcr_v2_2needle_1m", "score": 78.7, "reference_url": "https://llm-stats.com/benchmarks/mrcr-1m", "reported_setting": { "effort": "max" }, "matches_canonical": false, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=DeepSeek-V4-Flash-Max, slug=deepseek-v4-flash-max, provider=DeepSeek" }, { "model_id": "gemini-2.0-flash-lite", "benchmark_id": "mrcr_v2_2needle_1m", "score": 58.0, "reference_url": "https://llm-stats.com/benchmarks/mrcr-1m", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default", "context": "default", "notes": "Per Gemini 2.0 Flash Model Card (April 15 2025): non-thinking generation, default sampling. Same canonical setting applies to 1.5 / 2.0 family." }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=Gemini 2.0 Flash-Lite, slug=gemini-2.0-flash-lite, provider=Google" }, { "model_id": "claude-opus-4.6", "benchmark_id": "mrcr_v2_8needle", "score": 93.0, "reference_url": "https://llm-stats.com/benchmarks/mrcr-v2-(8-needle)", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "none (per-bench override)", "sampling": "pass@1 (avg 5 trials)", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default", "context": "up to 1M", "notes": "Per Mythos System Card Table 6.3.A: standard config = adaptive thinking max effort, default sampling, avg 5 trials, context up to 1M." }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=Claude Opus 4.6, slug=claude-opus-4-6, provider=Anthropic" }, { "model_id": "gpt-5.5", "benchmark_id": "mrcr_v2_8needle", "score": 74.0, "reference_url": "https://llm-stats.com/benchmarks/mrcr-v2-(8-needle)", "reported_setting": { "mode": "thinking", "effort": "xhigh", "tools": "none (per-bench override)", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default", "context": "default", "notes": "Per OpenAI gpt-5.5 blog: reasoning effort=xhigh, research environment (may differ from production ChatGPT)." }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=GPT-5.5, slug=gpt-5.5, provider=OpenAI" }, { "model_id": "gemini-3.1-pro", "benchmark_id": "mrcr_v2_8needle", "score": 26.3, "reference_url": "https://llm-stats.com/benchmarks/mrcr-v2-(8-needle)", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default", "context": "default", "notes": "Per Google DeepMind /models/gemini/pro/ table: Gemini 3.1 Pro Thinking (High), no tools by default." }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=Gemini 3.1 Pro, slug=gemini-3.1-pro-preview, provider=Google" }, { "model_id": "gemini-3-flash", "benchmark_id": "mrcr_v2_8needle", "score": 22.1, "reference_url": "https://llm-stats.com/benchmarks/mrcr-v2-(8-needle)", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default", "context": "default", "notes": "Per Google DeepMind /models/gemini/flash/ table: Gemini 3 Flash Thinking, no tools by default." }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=Gemini 3 Flash, slug=gemini-3-flash-preview, provider=Google" }, { "model_id": "gemini-2.5-pro", "benchmark_id": "mrcr_v2_8needle", "score": 16.4, "reference_url": "https://llm-stats.com/benchmarks/mrcr-v2-(8-needle)", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1 (single attempt)", "judge": "rule-based", "harness": "AI Studio API", "prompt_style": "default", "temperature": "default", "context": "default", "notes": "Per Gemini 2.5 Pro Model Card (GA, June 27 2025): pass@1 single attempt, AI Studio API model-id gemini-2.5-pro-preview-06-05 / gemini-2.5-pro GA, default sampling. Release date set to GA date 2025-06-27." }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=Gemini 2.5 Pro Preview 06-05, slug=gemini-2.5-pro-preview-06-05, provider=Google" }, { "model_id": "gemini-2.5-pro", "benchmark_id": "mrcr_v1", "score": 93.0, "reference_url": "https://llm-stats.com/benchmarks/mrcr", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1 (single attempt)", "judge": "rule-based", "harness": "AI Studio API", "prompt_style": "default", "temperature": "default", "context": "default", "notes": "Per Gemini 2.5 Pro Model Card (GA, June 27 2025): pass@1 single attempt, AI Studio API model-id gemini-2.5-pro-preview-06-05 / gemini-2.5-pro GA, default sampling. Release date set to GA date 2025-06-27." }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=Gemini 2.5 Pro, slug=gemini-2.5-pro, provider=Google" }, { "model_id": "gemini-1.5-flash-8b", "benchmark_id": "mrcr_v1", "score": 54.7, "reference_url": "https://llm-stats.com/benchmarks/mrcr", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default", "context": "1M", "notes": "Per Command A paper baseline." }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=Gemini 1.5 Flash 8B, slug=gemini-1.5-flash-8b, provider=Google" }, { "model_id": "deepseek-v2.5-0905", "benchmark_id": "mt_bench_101", "score": 90.2, "reference_url": "https://llm-stats.com/benchmarks/mt-bench", "reported_setting": { "snapshot": "0905" }, "matches_canonical": false, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=DeepSeek-V2.5, slug=deepseek-v2.5, provider=DeepSeek" }, { "model_id": "qwen-2.5-7b", "benchmark_id": "mt_bench_101", "score": 87.5, "reference_url": "https://llm-stats.com/benchmarks/mt-bench", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "OLMES (Allen AI)", "prompt_style": "default", "temperature": "default", "context": "default", "notes": "Per OLMo 2 paper Table 7 Instruct (instruction-tuned variant)." }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=Qwen2.5 7B Instruct, slug=qwen-2.5-7b-instruct, provider=Alibaba Cloud / Qwen Team" }, { "model_id": "mistral-large-2", "benchmark_id": "mt_bench_101", "score": 86.3, "reference_url": "https://llm-stats.com/benchmarks/mt-bench", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default", "context": "128K", "notes": "Per Command A paper (third-party Cohere reproduction baseline)." }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=Mistral Large 2, slug=mistral-large-2-2407, provider=Mistral AI" }, { "model_id": "qwen2-7b-instruct", "benchmark_id": "mt_bench_101", "score": 84.1, "reference_url": "https://llm-stats.com/benchmarks/mt-bench", "reported_setting": { "mode": "unknown", "effort": "n/a", "tools": "unknown", "sampling": "unknown", "judge": "unknown", "harness": "unknown", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown", "notes": "Model added from third-party aggregator (llm-stats.com); canonical setting not established." }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=Qwen2 7B Instruct, slug=qwen2-7b-instruct, provider=Alibaba Cloud / Qwen Team" }, { "model_id": "mistral-small-24b", "benchmark_id": "mt_bench_101", "score": 83.5, "reference_url": "https://llm-stats.com/benchmarks/mt-bench", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "OLMES (Allen AI)", "prompt_style": "default", "temperature": "default", "context": "default", "notes": "Per OLMo 2 paper Table 7 Instruct (instruction-tuned variant)." }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=Mistral Small 3 24B Instruct, slug=mistral-small-24b-instruct-2501, provider=Mistral AI" }, { "model_id": "pixtral-12b", "benchmark_id": "mt_bench_101", "score": 76.8, "reference_url": "https://llm-stats.com/benchmarks/mt-bench", "reported_setting": { "mode": "unknown", "effort": "n/a", "tools": "unknown", "sampling": "unknown", "judge": "unknown", "harness": "unknown", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown", "notes": "Model added from third-party aggregator (llm-stats.com); canonical setting not established." }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=Pixtral-12B, slug=pixtral-12b-2409, provider=Mistral AI" }, { "model_id": "qwen2.5-omni-7b", "benchmark_id": "muirbench", "score": 59.2, "reference_url": "https://llm-stats.com/benchmarks/muirbench", "reported_setting": { "mode": "unknown", "effort": "n/a", "tools": "unknown", "sampling": "unknown", "judge": "unknown", "harness": "unknown", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown", "notes": "Model added from third-party aggregator (llm-stats.com); canonical setting not established." }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=Qwen2.5-Omni-7B, slug=qwen2.5-omni-7b, provider=Alibaba Cloud / Qwen Team" }, { "model_id": "minimax-m2.5", "benchmark_id": "multi_swe_bench", "score": 51.3, "reference_url": "https://llm-stats.com/benchmarks/multi-swe-bench", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none (per-bench override)", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default", "context": "default", "notes": "Per MiniMax M2.5 model card." }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=MiniMax M2.5, slug=minimax-m2.5, provider=MiniMax" }, { "model_id": "minimax-m2.1", "benchmark_id": "multi_swe_bench", "score": 49.4, "reference_url": "https://llm-stats.com/benchmarks/multi-swe-bench", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none (per-bench override)", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default", "context": "default", "notes": "Per MiniMax M2.5 model card (referenced as predecessor)." }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=MiniMax M2.1, slug=minimax-m2.1, provider=MiniMax" }, { "model_id": "gpt-5", "benchmark_id": "multichallenge", "score": 69.6, "reference_url": "https://llm-stats.com/benchmarks/multichallenge", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none (per-bench override)", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default", "context": "default", "notes": "Per OpenAI GPT-5 dev blog. high reasoning effort." }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=GPT-5, slug=gpt-5-2025-08-07, provider=OpenAI" }, { "model_id": "o3-high", "benchmark_id": "multichallenge", "score": 60.4, "reference_url": "https://llm-stats.com/benchmarks/multichallenge", "reported_setting": { "effort": "default (o3 base, not high)" }, "matches_canonical": false, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=o3, slug=o3-2025-04-16, provider=OpenAI" }, { "model_id": "kimi-k2", "benchmark_id": "multichallenge", "score": 54.1, "reference_url": "https://llm-stats.com/benchmarks/multichallenge", "reported_setting": { "variant": "instruct", "snapshot": "0905" }, "matches_canonical": false, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=Kimi K2-Instruct-0905, slug=kimi-k2-instruct-0905, provider=Moonshot AI" }, { "model_id": "o4-mini-high", "benchmark_id": "multichallenge", "score": 43.0, "reference_url": "https://llm-stats.com/benchmarks/multichallenge", "reported_setting": { "effort": "default (o4-mini base, not high)" }, "matches_canonical": false, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=o4-mini, slug=o4-mini, provider=OpenAI" }, { "model_id": "qwen3-235b", "benchmark_id": "multipl_e_avg", "score": 87.9, "reference_url": "https://llm-stats.com/benchmarks/multipl-e", "reported_setting": { "mode": "instruct", "snapshot": "2507" }, "matches_canonical": false, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=Qwen3-235B-A22B-Instruct-2507, slug=qwen3-235b-a22b-instruct-2507, provider=Alibaba Cloud / Qwen Team" }, { "model_id": "kimi-k2", "benchmark_id": "multipl_e_avg", "score": 85.7, "reference_url": "https://llm-stats.com/benchmarks/multipl-e", "reported_setting": { "variant": "instruct (BP kimi-k2 is base)" }, "matches_canonical": false, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=Kimi K2 Instruct, slug=kimi-k2-instruct, provider=Moonshot AI" }, { "model_id": "qwen-2.5-32b", "benchmark_id": "multipl_e_avg", "score": 75.4, "reference_url": "https://llm-stats.com/benchmarks/multipl-e", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "OLMES (Allen AI)", "prompt_style": "default", "temperature": "default", "context": "default", "notes": "Per OLMo 2 paper Table 7 Instruct (instruction-tuned variant)." }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=Qwen2.5 32B Instruct, slug=qwen-2.5-32b-instruct, provider=Alibaba Cloud / Qwen Team" }, { "model_id": "qwen-2.5-14b-instruct", "benchmark_id": "multipl_e_avg", "score": 72.8, "reference_url": "https://llm-stats.com/benchmarks/multipl-e", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default", "context": "default (128K)", "notes": "Per Phi-4 paper Table 1 baseline." }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=Qwen2.5 14B Instruct, slug=qwen-2.5-14b-instruct, provider=Alibaba Cloud / Qwen Team" }, { "model_id": "qwen-2.5-7b", "benchmark_id": "multipl_e_avg", "score": 70.4, "reference_url": "https://llm-stats.com/benchmarks/multipl-e", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "OLMES (Allen AI)", "prompt_style": "default", "temperature": "default", "context": "default", "notes": "Per OLMo 2 paper Table 7 Instruct (instruction-tuned variant)." }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=Qwen2.5 7B Instruct, slug=qwen-2.5-7b-instruct, provider=Alibaba Cloud / Qwen Team" }, { "model_id": "qwen2-72b-instruct", "benchmark_id": "multipl_e_avg", "score": 69.2, "reference_url": "https://llm-stats.com/benchmarks/multipl-e", "reported_setting": { "mode": "unknown", "effort": "n/a", "tools": "unknown", "sampling": "unknown", "judge": "unknown", "harness": "unknown", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown", "notes": "Model added from third-party aggregator (llm-stats.com); canonical setting not established." }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=Qwen2 72B Instruct, slug=qwen2-72b-instruct, provider=Alibaba Cloud / Qwen Team" }, { "model_id": "qwen2.5-omni-7b", "benchmark_id": "multipl_e_avg", "score": 65.8, "reference_url": "https://llm-stats.com/benchmarks/multipl-e", "reported_setting": { "mode": "unknown", "effort": "n/a", "tools": "unknown", "sampling": "unknown", "judge": "unknown", "harness": "unknown", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown", "notes": "Model added from third-party aggregator (llm-stats.com); canonical setting not established." }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=Qwen2.5-Omni-7B, slug=qwen2.5-omni-7b, provider=Alibaba Cloud / Qwen Team" }, { "model_id": "qwen2-7b-instruct", "benchmark_id": "multipl_e_avg", "score": 59.1, "reference_url": "https://llm-stats.com/benchmarks/multipl-e", "reported_setting": { "mode": "unknown", "effort": "n/a", "tools": "unknown", "sampling": "unknown", "judge": "unknown", "harness": "unknown", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown", "notes": "Model added from third-party aggregator (llm-stats.com); canonical setting not established." }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=Qwen2 7B Instruct, slug=qwen2-7b-instruct, provider=Alibaba Cloud / Qwen Team" }, { "model_id": "qwen2.5-omni-7b", "benchmark_id": "ocrbench_v2", "score": 57.8, "reference_url": "https://llm-stats.com/benchmarks/ocrbench-v2", "reported_setting": { "mode": "unknown", "effort": "n/a", "tools": "unknown", "sampling": "unknown", "judge": "unknown", "harness": "unknown", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown", "notes": "Model added from third-party aggregator (llm-stats.com); canonical setting not established." }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=Qwen2.5-Omni-7B, slug=qwen2.5-omni-7b, provider=Alibaba Cloud / Qwen Team" }, { "model_id": "qwen3-235b", "benchmark_id": "ojbench", "score": 32.5, "reference_url": "https://llm-stats.com/benchmarks/ojbench", "reported_setting": { "mode": "thinking", "snapshot": "2507" }, "matches_canonical": false, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=Qwen3-235B-A22B-Thinking-2507, slug=qwen3-235b-a22b-thinking-2507, provider=Alibaba Cloud / Qwen Team" }, { "model_id": "qwen3.6-plus", "benchmark_id": "omnidocbench_1.5", "score": 91.2, "reference_url": "https://llm-stats.com/benchmarks/omnidocbench-1.5", "reported_setting": {}, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=Qwen3.6 Plus, slug=qwen3.6-plus, provider=Alibaba Cloud / Qwen Team" }, { "model_id": "gpt-5.4", "benchmark_id": "omnidocbench_1.5", "score": 89.1, "reference_url": "https://llm-stats.com/benchmarks/omnidocbench-1.5", "reported_setting": { "mode": "thinking", "effort": "xhigh", "tools": "none (per-bench override)", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default", "context": "default", "notes": "Per OpenAI gpt-5.4 blog: reasoning effort=xhigh, research environment (may differ from production ChatGPT)." }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=GPT-5.4, slug=gpt-5.4, provider=OpenAI" }, { "model_id": "granite-3.3-8b-base", "benchmark_id": "popqa", "score": 26.2, "reference_url": "https://llm-stats.com/benchmarks/popqa", "reported_setting": { "mode": "unknown", "effort": "n/a", "tools": "unknown", "sampling": "unknown", "judge": "unknown", "harness": "unknown", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown", "notes": "Model added from third-party aggregator (llm-stats.com); canonical setting not established." }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=Granite 3.3 8B Base, slug=granite-3.3-8b-base, provider=IBM" }, { "model_id": "granite-3.3-8b-instruct", "benchmark_id": "popqa", "score": 26.2, "reference_url": "https://llm-stats.com/benchmarks/popqa", "reported_setting": { "mode": "unknown", "effort": "n/a", "tools": "unknown", "sampling": "unknown", "judge": "unknown", "harness": "unknown", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown", "notes": "Model added from third-party aggregator (llm-stats.com); canonical setting not established." }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=Granite 3.3 8B Instruct, slug=granite-3.3-8b-instruct, provider=IBM" }, { "model_id": "qwen3.6-plus", "benchmark_id": "realworldqa", "score": 85.4, "reference_url": "https://llm-stats.com/benchmarks/realworldqa", "reported_setting": {}, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=Qwen3.6 Plus, slug=qwen3.6-plus, provider=Alibaba Cloud / Qwen Team" }, { "model_id": "qwen2.5-omni-7b", "benchmark_id": "realworldqa", "score": 70.3, "reference_url": "https://llm-stats.com/benchmarks/realworldqa", "reported_setting": { "mode": "unknown", "effort": "n/a", "tools": "unknown", "sampling": "unknown", "judge": "unknown", "harness": "unknown", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown", "notes": "Model added from third-party aggregator (llm-stats.com); canonical setting not established." }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=Qwen2.5-Omni-7B, slug=qwen2.5-omni-7b, provider=Alibaba Cloud / Qwen Team" }, { "model_id": "phi-3.5-moe-instruct", "benchmark_id": "repoqa", "score": 85.0, "reference_url": "https://llm-stats.com/benchmarks/repoqa", "reported_setting": { "mode": "unknown", "effort": "n/a", "tools": "unknown", "sampling": "unknown", "judge": "unknown", "harness": "unknown", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown", "notes": "Model added from third-party aggregator (llm-stats.com); canonical setting not established." }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=Phi-3.5-MoE-instruct, slug=phi-3.5-moe-instruct, provider=Microsoft" }, { "model_id": "phi-3.5-mini-instruct", "benchmark_id": "repoqa", "score": 77.0, "reference_url": "https://llm-stats.com/benchmarks/repoqa", "reported_setting": { "mode": "unknown", "effort": "n/a", "tools": "unknown", "sampling": "unknown", "judge": "unknown", "harness": "unknown", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown", "notes": "Model added from third-party aggregator (llm-stats.com); canonical setting not established." }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=Phi-3.5-mini-instruct, slug=phi-3.5-mini-instruct, provider=Microsoft" }, { "model_id": "minicpm-sala", "benchmark_id": "ruler_128k", "score": 89.4, "reference_url": "https://llm-stats.com/benchmarks/ruler-128k", "reported_setting": { "mode": "unknown", "effort": "n/a", "tools": "unknown", "sampling": "unknown", "judge": "unknown", "harness": "unknown", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown", "notes": "Model added from third-party aggregator (llm-stats.com); canonical setting not established." }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=MiniCPM-SALA, slug=minicpm-sala, provider=OpenBMB" }, { "model_id": "qwen3.6-plus", "benchmark_id": "screenspot_pro", "score": 68.2, "reference_url": "https://llm-stats.com/benchmarks/screenspot-pro", "reported_setting": {}, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=Qwen3.6 Plus, slug=qwen3.6-plus, provider=Alibaba Cloud / Qwen Team" }, { "model_id": "kimi-k2-thinking", "benchmark_id": "seal_0", "score": 56.3, "reference_url": "https://llm-stats.com/benchmarks/seal-0", "reported_setting": { "mode": "thinking", "effort": "default (96k thinking-token budget for HLE/AIME/HMMT/GPQA; 128k for IMO/LCB/OJ)", "tools": "none (per-bench override)", "sampling": "pass@1 (avg@32 for AIME/HMMT no-tools)", "judge": "rule-based (HLE uses o3-mini)", "harness": "official", "prompt_style": "default", "temperature": "default", "context": "default (256K)", "notes": "Per Kimi K2 Thinking blog (moonshotai.github.io/Kimi-K2/thinking.html). Thinking model." }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=Kimi K2-Thinking-0905, slug=kimi-k2-thinking-0905, provider=Moonshot AI" }, { "model_id": "deepseek-v4-pro", "benchmark_id": "simpleqa", "score": 57.9, "reference_url": "https://llm-stats.com/benchmarks/simpleqa", "reported_setting": { "effort": "max" }, "matches_canonical": false, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=DeepSeek-V4-Pro-Max, slug=deepseek-v4-pro-max, provider=DeepSeek" }, { "model_id": "o1-high", "benchmark_id": "simpleqa", "score": 47.0, "reference_url": "https://llm-stats.com/benchmarks/simpleqa", "reported_setting": { "effort": "default (o1 base, not high)" }, "matches_canonical": false, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=o1, slug=o1-2024-12-17, provider=OpenAI" }, { "model_id": "gpt-4o", "benchmark_id": "simpleqa", "score": 38.2, "reference_url": "https://llm-stats.com/benchmarks/simpleqa", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none (per-bench override)", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default", "context": "default", "notes": "Per OpenAI GPT-4.1 blog table." }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=GPT-4o, slug=gpt-4o-2024-08-06, provider=OpenAI" }, { "model_id": "deepseek-v4-flash", "benchmark_id": "simpleqa", "score": 34.1, "reference_url": "https://llm-stats.com/benchmarks/simpleqa", "reported_setting": { "effort": "max" }, "matches_canonical": false, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=DeepSeek-V4-Flash-Max, slug=deepseek-v4-flash-max, provider=DeepSeek" }, { "model_id": "gemma-3-1b-it", "benchmark_id": "simpleqa", "score": 2.2, "reference_url": "https://llm-stats.com/benchmarks/simpleqa", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default", "context": "32k", "notes": "Google Gemma 3 1B instruction-tuned." }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=Gemma 3 1B, slug=gemma-3-1b-it, provider=Google" }, { "model_id": "qwen3.6-plus", "benchmark_id": "simplevqa", "score": 67.3, "reference_url": "https://llm-stats.com/benchmarks/simplevqa", "reported_setting": {}, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=Qwen3.6 Plus, slug=qwen3.6-plus, provider=Alibaba Cloud / Qwen Team" }, { "model_id": "qwen3.6-plus", "benchmark_id": "supergpqa", "score": 71.6, "reference_url": "https://llm-stats.com/benchmarks/supergpqa", "reported_setting": {}, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=Qwen3.6 Plus, slug=qwen3.6-plus, provider=Alibaba Cloud / Qwen Team" }, { "model_id": "qwen3-235b", "benchmark_id": "supergpqa", "score": 64.9, "reference_url": "https://llm-stats.com/benchmarks/supergpqa", "reported_setting": { "mode": "thinking", "snapshot": "2507" }, "matches_canonical": false, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=Qwen3-235B-A22B-Thinking-2507, slug=qwen3-235b-a22b-thinking-2507, provider=Alibaba Cloud / Qwen Team" }, { "model_id": "qwen3.6-plus", "benchmark_id": "swe_bench_multilingual", "score": 73.8, "reference_url": "https://llm-stats.com/benchmarks/swe-bench-multilingual", "reported_setting": {}, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=Qwen3.6 Plus, slug=qwen3.6-plus, provider=Alibaba Cloud / Qwen Team" }, { "model_id": "minimax-m2.1", "benchmark_id": "swe_bench_multilingual", "score": 72.5, "reference_url": "https://llm-stats.com/benchmarks/swe-bench-multilingual", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none (per-bench override)", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default", "context": "default", "notes": "Per MiniMax M2.5 model card (referenced as predecessor)." }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=MiniMax M2.1, slug=minimax-m2.1, provider=MiniMax" }, { "model_id": "deepseek-r1", "benchmark_id": "swe_bench_multilingual", "score": 30.5, "reference_url": "https://llm-stats.com/benchmarks/swe-bench-multilingual", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none (per-bench override)", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default", "context": "default", "notes": "Per DeepSeek R1-0528 model card." }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=DeepSeek-R1-0528, slug=deepseek-r1-0528, provider=DeepSeek" }, { "model_id": "minimax-m2.5", "benchmark_id": "swe_bench_pro", "score": 55.4, "reference_url": "https://llm-stats.com/benchmarks/swe-bench-pro", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none (per-bench override)", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default", "context": "default", "notes": "Per MiniMax M2.5 model card." }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=MiniMax M2.5, slug=minimax-m2.5, provider=MiniMax" }, { "model_id": "minimax-m2.5", "benchmark_id": "swe_bench_verified", "score": 80.2, "reference_url": "https://llm-stats.com/benchmarks/swe-bench-verified", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none (per-bench override)", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default", "context": "default", "notes": "Per MiniMax M2.5 model card." }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=MiniMax M2.5, slug=minimax-m2.5, provider=MiniMax" }, { "model_id": "minimax-m2.1", "benchmark_id": "swe_bench_verified", "score": 67.0, "reference_url": "https://llm-stats.com/benchmarks/swe-bench-verified", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none (per-bench override)", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default", "context": "default", "notes": "Per MiniMax M2.5 model card (referenced as predecessor)." }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=MiniMax M2.1, slug=minimax-m2.1, provider=MiniMax" }, { "model_id": "longcat-flash-chat", "benchmark_id": "swe_bench_verified", "score": 60.4, "reference_url": "https://llm-stats.com/benchmarks/swe-bench-verified", "reported_setting": { "mode": "unknown", "effort": "n/a", "tools": "unknown", "sampling": "unknown", "judge": "unknown", "harness": "unknown", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown", "notes": "Model added from third-party aggregator (llm-stats.com); canonical setting not established." }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=LongCat-Flash-Chat, slug=longcat-flash-chat, provider=Meituan" }, { "model_id": "claude-3.5-haiku", "benchmark_id": "swe_bench_verified", "score": 40.6, "reference_url": "https://llm-stats.com/benchmarks/swe-bench-verified", "reported_setting": { "mode": "unknown", "effort": "n/a", "tools": "unknown", "sampling": "unknown", "judge": "unknown", "harness": "unknown", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown", "notes": "Model added from third-party aggregator (llm-stats.com); canonical setting not established." }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=Claude 3.5 Haiku, slug=claude-3-5-haiku-20241022, provider=Anthropic" }, { "model_id": "sarvam-30b", "benchmark_id": "swe_bench_verified", "score": 34.0, "reference_url": "https://llm-stats.com/benchmarks/swe-bench-verified", "reported_setting": { "mode": "unknown", "effort": "n/a", "tools": "unknown", "sampling": "unknown", "judge": "unknown", "harness": "unknown", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown", "notes": "Model added from third-party aggregator (llm-stats.com); canonical setting not established." }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=Sarvam-30B, slug=sarvam-30b, provider=Sarvam AI" }, { "model_id": "gemini-diffusion", "benchmark_id": "swe_bench_verified", "score": 22.9, "reference_url": "https://llm-stats.com/benchmarks/swe-bench-verified", "reported_setting": { "mode": "unknown", "effort": "n/a", "tools": "unknown", "sampling": "unknown", "judge": "unknown", "harness": "unknown", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown", "notes": "Model added from third-party aggregator (llm-stats.com); canonical setting not established." }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=Gemini Diffusion, slug=gemini-diffusion, provider=Google" }, { "model_id": "gpt-4.5", "benchmark_id": "swelancer", "score": 37.3, "reference_url": "https://llm-stats.com/benchmarks/swe-lancer", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none (per-bench override)", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default", "context": "default", "notes": "Per OpenAI GPT-4.1 blog table." }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=GPT-4.5, slug=gpt-4.5, provider=OpenAI" }, { "model_id": "gpt-4o", "benchmark_id": "swelancer", "score": 32.6, "reference_url": "https://llm-stats.com/benchmarks/swe-lancer", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none (per-bench override)", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default", "context": "default", "notes": "Per OpenAI GPT-4.1 blog table." }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=GPT-4o, slug=gpt-4o-2024-08-06, provider=OpenAI" }, { "model_id": "claude-sonnet-4.5", "benchmark_id": "tau_bench_airline", "score": 70.0, "reference_url": "https://llm-stats.com/benchmarks/tau-bench-airline", "reported_setting": { "mode": "thinking", "effort": "high (default)", "tools": "none (per-bench override)", "sampling": "pass@1 (avg 5 trials)", "judge": "rule-based", "harness": "Terminus-2 (for terminal-bench); official otherwise", "prompt_style": "interleaved scratchpads", "temperature": "default (top_p, temperature)", "context": "200K (default)", "notes": "Per Anthropic Opus 4.5 announcement: 64K thinking budget, interleaved scratchpads, 200K context, default effort=high, default sampling, avg 5 trials. Exceptions: SWE-bench Verified=no thinking; Terminal-Bench=128K thinking budget." }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=Claude Sonnet 4.5, slug=claude-sonnet-4-5-20250929, provider=Anthropic" }, { "model_id": "o4-mini-high", "benchmark_id": "tau_bench_airline", "score": 49.2, "reference_url": "https://llm-stats.com/benchmarks/tau-bench-airline", "reported_setting": { "effort": "default (o4-mini base, not high)" }, "matches_canonical": false, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=o4-mini, slug=o4-mini, provider=OpenAI" }, { "model_id": "qwen3-235b", "benchmark_id": "tau_bench_airline", "score": 46.0, "reference_url": "https://llm-stats.com/benchmarks/tau-bench-airline", "reported_setting": { "mode": "thinking", "snapshot": "2507" }, "matches_canonical": false, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=Qwen3-235B-A22B-Thinking-2507, slug=qwen3-235b-a22b-thinking-2507, provider=Alibaba Cloud / Qwen Team" }, { "model_id": "claude-3.5-haiku", "benchmark_id": "tau_bench_airline", "score": 22.8, "reference_url": "https://llm-stats.com/benchmarks/tau-bench-airline", "reported_setting": { "mode": "unknown", "effort": "n/a", "tools": "unknown", "sampling": "unknown", "judge": "unknown", "harness": "unknown", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown", "notes": "Model added from third-party aggregator (llm-stats.com); canonical setting not established." }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=Claude 3.5 Haiku, slug=claude-3-5-haiku-20241022, provider=Anthropic" }, { "model_id": "o4-mini-high", "benchmark_id": "tau_bench_retail", "score": 71.8, "reference_url": "https://llm-stats.com/benchmarks/tau-bench-retail", "reported_setting": { "effort": "default (o4-mini base, not high)" }, "matches_canonical": false, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=o4-mini, slug=o4-mini, provider=OpenAI" }, { "model_id": "qwen3-235b", "benchmark_id": "tau_bench_retail", "score": 67.8, "reference_url": "https://llm-stats.com/benchmarks/tau-bench-retail", "reported_setting": { "mode": "thinking", "snapshot": "2507" }, "matches_canonical": false, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=Qwen3-235B-A22B-Thinking-2507, slug=qwen3-235b-a22b-thinking-2507, provider=Alibaba Cloud / Qwen Team" }, { "model_id": "claude-3.5-haiku", "benchmark_id": "tau_bench_retail", "score": 51.0, "reference_url": "https://llm-stats.com/benchmarks/tau-bench-retail", "reported_setting": { "mode": "unknown", "effort": "n/a", "tools": "unknown", "sampling": "unknown", "judge": "unknown", "harness": "unknown", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown", "notes": "Model added from third-party aggregator (llm-stats.com); canonical setting not established." }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=Claude 3.5 Haiku, slug=claude-3-5-haiku-20241022, provider=Anthropic" }, { "model_id": "glm-4.7", "benchmark_id": "tau1_bench_avg", "score": 87.4, "reference_url": "https://llm-stats.com/benchmarks/tau-bench", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none (per-bench override)", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default", "context": "default", "notes": "Per GLM-5 blog." }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=GLM-4.7, slug=glm-4.7, provider=Zhipu AI" }, { "model_id": "minimax-m2", "benchmark_id": "tau1_bench_avg", "score": 77.2, "reference_url": "https://llm-stats.com/benchmarks/tau-bench", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none (per-bench override)", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default", "context": "default", "notes": "Per MiniMax M2 model card." }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=MiniMax M2, slug=minimax-m2, provider=MiniMax" }, { "model_id": "o3-high", "benchmark_id": "tau1_bench_avg", "score": 63.0, "reference_url": "https://llm-stats.com/benchmarks/tau-bench", "reported_setting": { "effort": "default (o3 base, not high)" }, "matches_canonical": false, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=o3, slug=o3-2025-04-16, provider=OpenAI" }, { "model_id": "deepseek-v3.2-speciale", "benchmark_id": "terminal_bench", "score": 46.4, "reference_url": "https://llm-stats.com/benchmarks/terminal-bench-2", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "none (per-bench override)", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "1.0", "context": "128K", "notes": "DS V3.2-Speciale: high-compute reasoning variant of V3.2. Per V3.2 tech report Table 3." }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=DeepSeek-V3.2-Speciale, slug=deepseek-v3.2-speciale, provider=DeepSeek" }, { "model_id": "minimax-m2.1", "benchmark_id": "toolathlon", "score": 43.5, "reference_url": "https://llm-stats.com/benchmarks/toolathlon", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none (per-bench override)", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default", "context": "default", "notes": "Per MiniMax M2.5 model card (referenced as predecessor)." }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=MiniMax M2.1, slug=minimax-m2.1, provider=MiniMax" }, { "model_id": "deepseek-v3.2-speciale", "benchmark_id": "toolathlon", "score": 35.2, "reference_url": "https://llm-stats.com/benchmarks/toolathlon", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "none (per-bench override)", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "1.0", "context": "128K", "notes": "DS V3.2-Speciale: high-compute reasoning variant of V3.2. Per V3.2 tech report Table 3." }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=DeepSeek-V3.2-Speciale, slug=deepseek-v3.2-speciale, provider=DeepSeek" }, { "model_id": "phi-3.5-moe-instruct", "benchmark_id": "truthfulqa", "score": 77.5, "reference_url": "https://llm-stats.com/benchmarks/truthfulqa", "reported_setting": { "mode": "unknown", "effort": "n/a", "tools": "unknown", "sampling": "unknown", "judge": "unknown", "harness": "unknown", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown", "notes": "Model added from third-party aggregator (llm-stats.com); canonical setting not established." }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=Phi-3.5-MoE-instruct, slug=phi-3.5-moe-instruct, provider=Microsoft" }, { "model_id": "granite-3.3-8b-instruct", "benchmark_id": "truthfulqa", "score": 66.9, "reference_url": "https://llm-stats.com/benchmarks/truthfulqa", "reported_setting": { "mode": "unknown", "effort": "n/a", "tools": "unknown", "sampling": "unknown", "judge": "unknown", "harness": "unknown", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown", "notes": "Model added from third-party aggregator (llm-stats.com); canonical setting not established." }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=Granite 3.3 8B Instruct, slug=granite-3.3-8b-instruct, provider=IBM" }, { "model_id": "phi-4-mini", "benchmark_id": "truthfulqa", "score": 66.4, "reference_url": "https://llm-stats.com/benchmarks/truthfulqa", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default", "context": "default (128K)", "notes": "Per Phi-4-mini-instruct HF model card." }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=Phi 4 Mini, slug=phi-4-mini, provider=Microsoft" }, { "model_id": "phi-3.5-mini-instruct", "benchmark_id": "truthfulqa", "score": 64.0, "reference_url": "https://llm-stats.com/benchmarks/truthfulqa", "reported_setting": { "mode": "unknown", "effort": "n/a", "tools": "unknown", "sampling": "unknown", "judge": "unknown", "harness": "unknown", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown", "notes": "Model added from third-party aggregator (llm-stats.com); canonical setting not established." }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=Phi-3.5-mini-instruct, slug=phi-3.5-mini-instruct, provider=Microsoft" }, { "model_id": "qwen-2.5-14b-instruct", "benchmark_id": "truthfulqa", "score": 58.4, "reference_url": "https://llm-stats.com/benchmarks/truthfulqa", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default", "context": "default (128K)", "notes": "Per Phi-4 paper Table 1 baseline." }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=Qwen2.5 14B Instruct, slug=qwen-2.5-14b-instruct, provider=Alibaba Cloud / Qwen Team" }, { "model_id": "qwen2-72b-instruct", "benchmark_id": "truthfulqa", "score": 54.8, "reference_url": "https://llm-stats.com/benchmarks/truthfulqa", "reported_setting": { "mode": "unknown", "effort": "n/a", "tools": "unknown", "sampling": "unknown", "judge": "unknown", "harness": "unknown", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown", "notes": "Model added from third-party aggregator (llm-stats.com); canonical setting not established." }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=Qwen2 72B Instruct, slug=qwen2-72b-instruct, provider=Alibaba Cloud / Qwen Team" }, { "model_id": "granite-3.3-8b-base", "benchmark_id": "truthfulqa", "score": 52.1, "reference_url": "https://llm-stats.com/benchmarks/truthfulqa", "reported_setting": { "mode": "unknown", "effort": "n/a", "tools": "unknown", "sampling": "unknown", "judge": "unknown", "harness": "unknown", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown", "notes": "Model added from third-party aggregator (llm-stats.com); canonical setting not established." }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=Granite 3.3 8B Base, slug=granite-3.3-8b-base, provider=IBM" }, { "model_id": "gemini-2.0-flash", "benchmark_id": "vibe_eval", "score": 56.3, "reference_url": "https://llm-stats.com/benchmarks/vibe-eval", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default", "context": "default", "notes": "Per Gemini 2.0 Flash Model Card (April 15 2025): non-thinking generation, default sampling. Same canonical setting applies to 1.5 / 2.0 family." }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=Gemini 2.0 Flash, slug=gemini-2.0-flash, provider=Google" }, { "model_id": "gemini-1.5-pro", "benchmark_id": "vibe_eval", "score": 53.9, "reference_url": "https://llm-stats.com/benchmarks/vibe-eval", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default", "context": "default", "notes": "Per Gemini 2.0 Flash Model Card (April 15 2025): non-thinking generation, default sampling. Same canonical setting applies to 1.5 / 2.0 family." }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=Gemini 1.5 Pro, slug=gemini-1.5-pro, provider=Google" }, { "model_id": "gemini-1.5-flash", "benchmark_id": "vibe_eval", "score": 48.9, "reference_url": "https://llm-stats.com/benchmarks/vibe-eval", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default", "context": "default", "notes": "Per Gemini 2.0 Flash Model Card (April 15 2025): non-thinking generation, default sampling. Same canonical setting applies to 1.5 / 2.0 family." }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=Gemini 1.5 Flash, slug=gemini-1.5-flash, provider=Google" }, { "model_id": "gemini-1.5-flash-8b", "benchmark_id": "vibe_eval", "score": 40.9, "reference_url": "https://llm-stats.com/benchmarks/vibe-eval", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default", "context": "1M", "notes": "Per Command A paper baseline." }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=Gemini 1.5 Flash 8B, slug=gemini-1.5-flash-8b, provider=Google" }, { "model_id": "qwen3.6-plus", "benchmark_id": "video_mme", "score": 84.2, "reference_url": "https://llm-stats.com/benchmarks/video-mme", "reported_setting": {}, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=Qwen3.6 Plus, slug=qwen3.6-plus, provider=Alibaba Cloud / Qwen Team" }, { "model_id": "gemini-1.5-pro", "benchmark_id": "video_mme", "score": 78.6, "reference_url": "https://llm-stats.com/benchmarks/video-mme", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default", "context": "default", "notes": "Per Gemini 2.0 Flash Model Card (April 15 2025): non-thinking generation, default sampling. Same canonical setting applies to 1.5 / 2.0 family." }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=Gemini 1.5 Pro, slug=gemini-1.5-pro, provider=Google" }, { "model_id": "gemini-1.5-flash", "benchmark_id": "video_mme", "score": 76.1, "reference_url": "https://llm-stats.com/benchmarks/video-mme", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default", "context": "default", "notes": "Per Gemini 2.0 Flash Model Card (April 15 2025): non-thinking generation, default sampling. Same canonical setting applies to 1.5 / 2.0 family." }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=Gemini 1.5 Flash, slug=gemini-1.5-flash, provider=Google" }, { "model_id": "gemini-1.5-flash-8b", "benchmark_id": "video_mme", "score": 66.2, "reference_url": "https://llm-stats.com/benchmarks/video-mme", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default", "context": "1M", "notes": "Per Command A paper baseline." }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=Gemini 1.5 Flash 8B, slug=gemini-1.5-flash-8b, provider=Google" }, { "model_id": "qwen3.6-plus", "benchmark_id": "video_mmmu", "score": 84.0, "reference_url": "https://llm-stats.com/benchmarks/videommmu", "reported_setting": {}, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=Qwen3.6 Plus, slug=qwen3.6-plus, provider=Alibaba Cloud / Qwen Team" }, { "model_id": "gpt-4o", "benchmark_id": "video_mmmu", "score": 61.2, "reference_url": "https://llm-stats.com/benchmarks/videommmu", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none (per-bench override)", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default", "context": "default", "notes": "Per OpenAI GPT-4.1 blog table." }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=GPT-4o, slug=gpt-4o-2024-08-06, provider=OpenAI" }, { "model_id": "qwen3.6-plus", "benchmark_id": "vitabench", "score": 44.3, "reference_url": "https://llm-stats.com/benchmarks/vita-bench", "reported_setting": {}, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=Qwen3.6 Plus, slug=qwen3.6-plus, provider=Alibaba Cloud / Qwen Team" }, { "model_id": "qwen3.6-plus", "benchmark_id": "widesearch", "score": 74.3, "reference_url": "https://llm-stats.com/benchmarks/widesearch", "reported_setting": {}, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=Qwen3.6 Plus, slug=qwen3.6-plus, provider=Alibaba Cloud / Qwen Team" }, { "model_id": "mistral-large-3", "benchmark_id": "wildbench", "score": 68.5, "reference_url": "https://llm-stats.com/benchmarks/wild-bench", "reported_setting": {}, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=Mistral Large 3, slug=mistral-large-3-2509, provider=Mistral AI" }, { "model_id": "mistral-small-24b", "benchmark_id": "wildbench", "score": 52.2, "reference_url": "https://llm-stats.com/benchmarks/wild-bench", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "pass@1", "judge": "rule-based", "harness": "OLMES (Allen AI)", "prompt_style": "default", "temperature": "default", "context": "default", "notes": "Per OLMo 2 paper Table 7 Instruct (instruction-tuned variant)." }, "matches_canonical": true, "source_type": "third_party_aggregator", "audit_status": "verified_third_party", "notes": "llm-stats.com row: model=Mistral Small 3 24B Instruct, slug=mistral-small-24b-instruct-2501, provider=Mistral AI" }, { "model_id": "kimi-k2.6", "benchmark_id": "kimi_code_bench_v2", "score": 50.9, "reference_url": "https://huggingface.co/moonshotai/Kimi-K2.7-Code", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "benchmark-specified", "sampling": "pass@1", "judge": "Moonshot in-house evaluator", "harness": "Kimi Code CLI + official benchmark harness", "prompt_style": "default", "temperature": "1.0; top_p=0.95", "context": "262144 (256K)" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Kimi K2.7 Code model card Evaluation Results table: 50.9. This source uses top_p=0.95; the existing Kimi K2.6 canonical setting uses top_p=1.0." }, { "model_id": "kimi-k2.7-code", "benchmark_id": "kimi_code_bench_v2", "score": 62.0, "reference_url": "https://huggingface.co/moonshotai/Kimi-K2.7-Code", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "benchmark-specified", "sampling": "pass@1", "judge": "Moonshot in-house evaluator", "harness": "Kimi Code CLI + official benchmark harness", "prompt_style": "default", "temperature": "1.0; top_p=0.95", "context": "262144 (256K)" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Kimi K2.7 Code model card Evaluation Results table: 62.0." }, { "model_id": "gpt-5.5", "benchmark_id": "kimi_code_bench_v2", "score": 69.0, "reference_url": "https://huggingface.co/moonshotai/Kimi-K2.7-Code", "reported_setting": { "mode": "thinking", "effort": "xhigh", "tools": "benchmark-specified", "sampling": "pass@1", "judge": "Moonshot in-house evaluator", "harness": "Codex + official benchmark harness", "prompt_style": "default", "temperature": "default", "context": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Kimi K2.7 Code model card Evaluation Results table: 69.0. Cross-model value reported by Moonshot AI.", "candidates": [ { "score": 66.0, "reference_url": "https://github.com/MoonshotAI/Kimi-K3/blob/main/k3_tech_report.pdf", "source_type": "tech_report", "reported_setting": { "mode": "thinking", "effort": "xhigh", "tools": "benchmark-specified", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "Kimi Code harness", "prompt_style": "default", "temperature": "default", "context": "default" }, "notes": "Kimi K3 technical report: Kimi Code Bench 2.0 (Kimi Code) = 66.0. Source setting: Kimi Code harness. Origin: Moonshot evaluation." } ] }, { "model_id": "claude-opus-4.8", "benchmark_id": "kimi_code_bench_v2", "score": 71.7, "reference_url": "https://github.com/MoonshotAI/Kimi-K3/blob/main/k3_tech_report.pdf", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "benchmark-specified", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "agentic coding harness per row", "prompt_style": "default", "temperature": "default", "context": "default" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Kimi K3 technical report: Kimi Code Bench 2.0 = 71.7. Source setting: agentic coding harness per row. Origin: Moonshot evaluation.", "candidates": [ { "score": 67.4, "reference_url": "https://huggingface.co/moonshotai/Kimi-K2.7-Code", "source_type": "model_card", "reported_setting": { "mode": "thinking", "effort": "xhigh", "tools": "benchmark-specified", "sampling": "pass@1", "judge": "Moonshot in-house evaluator", "harness": "Claude Code + official benchmark harness", "prompt_style": "default", "temperature": "default", "context": "default" }, "notes": "Kimi K2.7 Code model card Evaluation Results table: 67.4. Cross-model value reported by Moonshot AI. This source uses xhigh effort, or max for MLS-Bench-Lite; Anthropic's model default is high effort. [Displaced verified primary preserved during Kimi K3 audit.]" } ] }, { "model_id": "kimi-k2.6", "benchmark_id": "program_bench", "score": 48.3, "reference_url": "https://huggingface.co/moonshotai/Kimi-K2.7-Code", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "benchmark-specified", "sampling": "pass@1", "judge": "248,000+ fuzz-generated behavioral tests", "harness": "Kimi Code CLI + official benchmark harness", "prompt_style": "default", "temperature": "1.0; top_p=0.95", "context": "262144 (256K)" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Kimi K2.7 Code model card Evaluation Results table: 48.3. This source uses top_p=0.95; the existing Kimi K2.6 canonical setting uses top_p=1.0." }, { "model_id": "kimi-k2.7-code", "benchmark_id": "program_bench", "score": 53.6, "reference_url": "https://huggingface.co/moonshotai/Kimi-K2.7-Code", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "benchmark-specified", "sampling": "pass@1", "judge": "248,000+ fuzz-generated behavioral tests", "harness": "Kimi Code CLI + official benchmark harness", "prompt_style": "default", "temperature": "1.0; top_p=0.95", "context": "262144 (256K)" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Kimi K2.7 Code model card Evaluation Results table: 53.6." }, { "model_id": "gpt-5.5", "benchmark_id": "program_bench", "score": 70.8, "reference_url": "https://github.com/MoonshotAI/Kimi-K3/blob/main/k3_tech_report.pdf", "reported_setting": { "mode": "thinking", "effort": "xhigh", "tools": "benchmark-specified", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "agentic coding harness per row", "prompt_style": "default", "temperature": "default", "context": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Kimi K3 technical report: ProgramBench = 70.8. Source setting: agentic coding harness per row. Origin: Moonshot K3 run; competitor values from Vals AI.", "candidates": [ { "score": 69.1, "reference_url": "https://huggingface.co/moonshotai/Kimi-K2.7-Code", "source_type": "model_card", "reported_setting": { "mode": "thinking", "effort": "xhigh", "tools": "benchmark-specified", "sampling": "pass@1", "judge": "248,000+ fuzz-generated behavioral tests", "harness": "Codex + official benchmark harness", "prompt_style": "default", "temperature": "default", "context": "default" }, "notes": "Kimi K2.7 Code model card Evaluation Results table: 69.1. Cross-model value reported by Moonshot AI. [Displaced verified primary preserved during Kimi K3 audit.]" }, { "score": 60.2, "reference_url": "https://media.x.ai/v1/website/4p5-5184fdf9.pdf", "source_type": "model_card", "reported_setting": { "mode": "thinking", "effort": "xhigh", "tools": "compiled executable and documentation; no internet or decompilation", "sampling": "unknown", "judge": "248,000+ fuzz-generated behavioral tests", "harness": "official ProgramBench sandbox", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown" }, "notes": "card45 ProgramBench Tests-passed rate (%); source effort=xhigh; benchmark protocol matches canonical=True. Unknown source fields remain unknown." }, { "score": 65.9, "reference_url": "https://seed.bytedance.com/en/blog/seed2-1-officially-released-advancing-ai-productivity", "source_type": "official_blog", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "Trae or benchmark-specific coding-agent environment", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official", "prompt_style": "official default", "temperature": "unknown" }, "notes": "Release blog EN: Coding chart (https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/user-upload/4xfa4mqq4wlbt.png)." } ] }, { "model_id": "claude-opus-4.8", "benchmark_id": "program_bench", "score": 71.9, "reference_url": "https://github.com/MoonshotAI/Kimi-K3/blob/main/k3_tech_report.pdf", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "benchmark-specified", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "agentic coding harness per row", "prompt_style": "default", "temperature": "default", "context": "default" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Kimi K3 technical report: ProgramBench = 71.9. Source setting: agentic coding harness per row. Origin: Moonshot K3 run; competitor values from Vals AI.", "candidates": [ { "score": 63.8, "reference_url": "https://huggingface.co/moonshotai/Kimi-K2.7-Code", "source_type": "model_card", "reported_setting": { "mode": "thinking", "effort": "xhigh", "tools": "benchmark-specified", "sampling": "pass@1", "judge": "248,000+ fuzz-generated behavioral tests", "harness": "Claude Code + official benchmark harness", "prompt_style": "default", "temperature": "default", "context": "default" }, "notes": "Kimi K2.7 Code model card Evaluation Results table: 63.8. Cross-model value reported by Moonshot AI. This source uses xhigh effort, or max for MLS-Bench-Lite; Anthropic's model default is high effort. [Displaced verified primary preserved during Kimi K3 audit.]" }, { "score": 57.1, "reference_url": "https://media.x.ai/v1/website/4p5-5184fdf9.pdf", "source_type": "model_card", "reported_setting": { "mode": "thinking", "effort": "xhigh", "tools": "compiled executable and documentation; no internet or decompilation", "sampling": "unknown", "judge": "248,000+ fuzz-generated behavioral tests", "harness": "official ProgramBench sandbox", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown" }, "notes": "card45 ProgramBench Tests-passed rate (%); source effort=xhigh; benchmark protocol matches canonical=True. Unknown source fields remain unknown." } ] }, { "model_id": "kimi-k2.6", "benchmark_id": "mls_bench_lite", "score": 26.7, "reference_url": "https://huggingface.co/moonshotai/Kimi-K2.7-Code", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "benchmark-specified", "sampling": "pass@1", "judge": "official MLS-Bench scoring", "harness": "Kimi Code CLI + official benchmark harness", "prompt_style": "default", "temperature": "1.0; top_p=0.95", "context": "262144 (256K)" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Kimi K2.7 Code model card Evaluation Results table: 26.7. This source uses top_p=0.95; the existing Kimi K2.6 canonical setting uses top_p=1.0." }, { "model_id": "kimi-k2.7-code", "benchmark_id": "mls_bench_lite", "score": 35.1, "reference_url": "https://huggingface.co/moonshotai/Kimi-K2.7-Code", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "benchmark-specified", "sampling": "pass@1", "judge": "official MLS-Bench scoring", "harness": "Kimi Code CLI + official benchmark harness", "prompt_style": "default", "temperature": "1.0; top_p=0.95", "context": "262144 (256K)" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Kimi K2.7 Code model card Evaluation Results table: 35.1." }, { "model_id": "gpt-5.5", "benchmark_id": "mls_bench_lite", "score": 35.5, "reference_url": "https://huggingface.co/moonshotai/Kimi-K2.7-Code", "reported_setting": { "mode": "thinking", "effort": "xhigh", "tools": "benchmark-specified", "sampling": "pass@1", "judge": "official MLS-Bench scoring", "harness": "Codex + official benchmark harness", "prompt_style": "default", "temperature": "default", "context": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Kimi K2.7 Code model card Evaluation Results table: 35.5. Cross-model value reported by Moonshot AI." }, { "model_id": "claude-opus-4.8", "benchmark_id": "mls_bench_lite", "score": 42.8, "reference_url": "https://huggingface.co/moonshotai/Kimi-K2.7-Code", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "benchmark-specified", "sampling": "pass@1", "judge": "official MLS-Bench scoring", "harness": "Claude Code + official benchmark harness", "prompt_style": "default", "temperature": "default", "context": "default" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Kimi K2.7 Code model card Evaluation Results table: 42.8. Cross-model value reported by Moonshot AI. This source uses xhigh effort, or max for MLS-Bench-Lite; Anthropic's model default is high effort." }, { "model_id": "kimi-k2.6", "benchmark_id": "kimi_claw_24_7", "score": 42.9, "reference_url": "https://huggingface.co/moonshotai/Kimi-K2.7-Code", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "benchmark-specified", "sampling": "avg-of-3", "judge": "evaluation-point pass rate", "harness": "OpenClaw", "prompt_style": "default", "temperature": "1.0; top_p=0.95", "context": "262144 (256K)" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Kimi K2.7 Code model card Evaluation Results table: 42.9. This source uses top_p=0.95; the existing Kimi K2.6 canonical setting uses top_p=1.0." }, { "model_id": "kimi-k2.7-code", "benchmark_id": "kimi_claw_24_7", "score": 46.9, "reference_url": "https://huggingface.co/moonshotai/Kimi-K2.7-Code", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "benchmark-specified", "sampling": "avg-of-3", "judge": "evaluation-point pass rate", "harness": "OpenClaw", "prompt_style": "default", "temperature": "1.0; top_p=0.95", "context": "262144 (256K)" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Kimi K2.7 Code model card Evaluation Results table: 46.9." }, { "model_id": "gpt-5.5", "benchmark_id": "kimi_claw_24_7", "score": 52.8, "reference_url": "https://huggingface.co/moonshotai/Kimi-K2.7-Code", "reported_setting": { "mode": "thinking", "effort": "xhigh", "tools": "benchmark-specified", "sampling": "avg-of-3", "judge": "evaluation-point pass rate", "harness": "OpenClaw", "prompt_style": "default", "temperature": "default", "context": "default" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Kimi K2.7 Code model card Evaluation Results table: 52.8. Cross-model value reported by Moonshot AI. The source averages three OpenClaw runs, while the GPT-5.5 canonical sampling is pass@1." }, { "model_id": "claude-opus-4.8", "benchmark_id": "kimi_claw_24_7", "score": 50.4, "reference_url": "https://huggingface.co/moonshotai/Kimi-K2.7-Code", "reported_setting": { "mode": "thinking", "effort": "xhigh", "tools": "benchmark-specified", "sampling": "avg-of-3", "judge": "evaluation-point pass rate", "harness": "OpenClaw", "prompt_style": "default", "temperature": "default", "context": "default" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Kimi K2.7 Code model card Evaluation Results table: 50.4. Cross-model value reported by Moonshot AI. This source uses xhigh effort, or max for MLS-Bench-Lite; Anthropic's model default is high effort." }, { "model_id": "kimi-k2.6", "benchmark_id": "mcpmark_verified", "score": 72.8, "reference_url": "https://huggingface.co/moonshotai/Kimi-K2.7-Code", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "benchmark-specified", "sampling": "avg-of-3", "judge": "programmatic verification on human-verified tasks", "harness": "official MCPMark; 100 steps; 32K max tokens per step", "prompt_style": "default", "temperature": "1.0; top_p=0.95", "context": "262144 (256K)" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Kimi K2.7 Code model card Evaluation Results table: 72.8. This source uses top_p=0.95; the existing Kimi K2.6 canonical setting uses top_p=1.0." }, { "model_id": "kimi-k2.7-code", "benchmark_id": "mcpmark_verified", "score": 81.1, "reference_url": "https://huggingface.co/moonshotai/Kimi-K2.7-Code", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "benchmark-specified", "sampling": "avg-of-3", "judge": "programmatic verification on human-verified tasks", "harness": "official MCPMark; 100 steps; 32K max tokens per step", "prompt_style": "default", "temperature": "1.0; top_p=0.95", "context": "262144 (256K)" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Kimi K2.7 Code model card Evaluation Results table: 81.1." }, { "model_id": "gpt-5.5", "benchmark_id": "mcpmark_verified", "score": 92.9, "reference_url": "https://huggingface.co/moonshotai/Kimi-K2.7-Code", "reported_setting": { "mode": "thinking", "effort": "xhigh", "tools": "benchmark-specified", "sampling": "avg-of-3", "judge": "programmatic verification on human-verified tasks", "harness": "official MCPMark; 100 steps; 32K max tokens per step", "prompt_style": "default", "temperature": "default", "context": "default" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Kimi K2.7 Code model card Evaluation Results table: 92.9. Cross-model value reported by Moonshot AI. The source averages three MCPMark-Verified runs, while the GPT-5.5 canonical sampling is pass@1." }, { "model_id": "claude-opus-4.8", "benchmark_id": "mcpmark_verified", "score": 76.4, "reference_url": "https://huggingface.co/moonshotai/Kimi-K2.7-Code", "reported_setting": { "mode": "thinking", "effort": "xhigh", "tools": "benchmark-specified", "sampling": "avg-of-3", "judge": "programmatic verification on human-verified tasks", "harness": "official MCPMark; 100 steps; 32K max tokens per step", "prompt_style": "default", "temperature": "default", "context": "default" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Kimi K2.7 Code model card Evaluation Results table: 76.4. Cross-model value reported by Moonshot AI. This source uses xhigh effort, or max for MLS-Bench-Lite; Anthropic's model default is high effort." }, { "model_id": "kimi-k2.7-code", "benchmark_id": "mcpatlas", "score": 76.0, "reference_url": "https://huggingface.co/moonshotai/Kimi-K2.7-Code", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "benchmark-specified", "sampling": "avg-of-3", "judge": "official MCP-Atlas claims-based judge", "harness": "official MCP-Atlas; 100 tool calls; 32K max tokens per step", "prompt_style": "default", "temperature": "1.0; top_p=0.95", "context": "262144 (256K)" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Kimi K2.7 Code model card Evaluation Results table: 76.0. The source uses 100 tool calls and 32K max tokens per step and does not identify whether the public 500-task subset was used; the current canonical row uses the public release and maxTurns=20." }, { "model_id": "claude-opus-4.8", "benchmark_id": "mcpatlas", "score": 83.6, "reference_url": "https://github.com/MoonshotAI/Kimi-K3/blob/main/k3_tech_report.pdf", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "benchmark-specified", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "public 500; maxTurns=100; Gemini 3.1 Pro judge", "prompt_style": "default", "temperature": "default", "context": "default" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Kimi K3 technical report: MCP-Atlas = 83.6. Source setting: public 500; maxTurns=100; Gemini 3.1 Pro judge. Origin: Moonshot evaluation.", "candidates": [ { "score": 81.3, "reference_url": "https://huggingface.co/moonshotai/Kimi-K2.7-Code", "source_type": "model_card", "reported_setting": { "mode": "thinking", "effort": "xhigh", "tools": "benchmark-specified", "sampling": "avg-of-3", "judge": "official MCP-Atlas claims-based judge", "harness": "official MCP-Atlas; 100 tool calls; 32K max tokens per step", "prompt_style": "default", "temperature": "default", "context": "default" }, "notes": "Kimi K2.7 Code model card Evaluation Results table: 81.3. Cross-model value reported by Moonshot AI. This source uses xhigh effort, or max for MLS-Bench-Lite; Anthropic's model default is high effort. The source uses 100 tool calls and 32K max tokens per step and does not identify whether the public 500-task subset was used; the current canonical row uses the public release and maxTurns=20. [Displaced verified primary preserved during Kimi K3 audit.]" }, { "score": 77.8, "reference_url": "https://z.ai/blog/glm-5.2", "source_type": "official_blog", "reported_setting": { "mode": "thinking", "effort": "max (source; model default is high)", "sampling": "pass@1", "prompt_style": "default", "tools": "agentic MCP servers", "context": "default", "judge": "Gemini 3.0 Pro", "harness": "public 500-task set; 10-minute timeout per task", "temperature": "default" }, "notes": "Z.ai public-set alternative." }, { "score": 82.2, "reference_url": "https://huggingface.co/tencent/Hy3", "source_type": "model_card", "reported_setting": { "mode": "thinking", "effort": "high / highest available", "tools": "MCP tool servers", "sampling": "pass@1", "judge": "Gemini 2.5 Pro", "harness": "Scale April 2026 methodology; public 500; 100 tool-call budget", "prompt_style": "default", "temperature": "0.9 for Hy3 family; otherwise source/provider setting", "context": "256000 for Hy3 family; otherwise source/provider setting", "notes": "Comparator value not marked as Tencent own test" }, "notes": "Hy3 official model-card appendix: MCP atlas (public) = 82.2. Comparator-reported value." }, { "score": 84.1, "reference_url": "https://huggingface.co/tencent/Hy3", "source_type": "model_card", "reported_setting": { "mode": "thinking", "effort": "high / highest available", "tools": "MCP tool servers", "sampling": "pass@1", "judge": "Gemini 2.5 Pro", "harness": "Scale April 2026 methodology; public 500; 100 tool-call budget", "prompt_style": "default", "temperature": "0.9 for Hy3 family; otherwise source/provider setting", "context": "256000 for Hy3 family; otherwise source/provider setting", "notes": "Tencent own test" }, "notes": "Hy3 official model-card appendix: MCP atlas (public) = 84.1*. Tencent own testing." } ] }, { "model_id": "kimi-k3", "benchmark_id": "gpqa_diamond", "score": 93.5, "reference_url": "https://github.com/MoonshotAI/Kimi-K3/blob/main/k3_tech_report.pdf", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "none", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "max/xhigh; no tools", "prompt_style": "default", "temperature": "1.0; top_p=0.95 for reasoning/no-tool vision, top_p=1.0 for coding/agentic", "context": "1048576 (1M)" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Kimi K3 technical report: GPQA Diamond = 93.5. Source setting: max/xhigh; no tools. Origin: Moonshot evaluation." }, { "model_id": "kimi-k3", "benchmark_id": "critpt", "score": 23.4, "reference_url": "https://github.com/MoonshotAI/Kimi-K3/blob/main/k3_tech_report.pdf", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "none", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "max/xhigh; no tools", "prompt_style": "default", "temperature": "1.0; top_p=0.95 for reasoning/no-tool vision, top_p=1.0 for coding/agentic", "context": "1048576 (1M)" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Kimi K3 technical report: CritPt = 23.4. Source setting: max/xhigh; no tools. Origin: Artificial Analysis cited by Kimi report." }, { "model_id": "kimi-k3", "benchmark_id": "aa_lcr", "score": 74.7, "reference_url": "https://github.com/MoonshotAI/Kimi-K3/blob/main/k3_tech_report.pdf", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "none", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "max/xhigh; no tools", "prompt_style": "default", "temperature": "1.0; top_p=0.95 for reasoning/no-tool vision, top_p=1.0 for coding/agentic", "context": "1048576 (1M)" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Kimi K3 technical report: AA-LCR = 74.7. Source setting: max/xhigh; no tools. Origin: Artificial Analysis cited by Kimi report." }, { "model_id": "kimi-k3", "benchmark_id": "hle", "score": 43.5, "reference_url": "https://github.com/MoonshotAI/Kimi-K3/blob/main/k3_tech_report.pdf", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "none", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "no tools; K3 max/temp=1/top-p=.95", "prompt_style": "default", "temperature": "1.0; top_p=0.95 for reasoning/no-tool vision, top_p=1.0 for coding/agentic", "context": "1048576 (1M)" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Kimi K3 technical report: HLE-Full (no tools) = 43.5. Source setting: no tools; K3 max/temp=1/top-p=.95. Origin: Moonshot evaluation." }, { "model_id": "kimi-k3", "benchmark_id": "hle_tools", "score": 56.0, "reference_url": "https://github.com/MoonshotAI/Kimi-K3/blob/main/k3_tech_report.pdf", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "general tools", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "general tools; pass@1", "prompt_style": "default", "temperature": "1.0; top_p=0.95 for reasoning/no-tool vision, top_p=1.0 for coding/agentic", "context": "1048576 (1M)" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Kimi K3 technical report: HLE-Full (with tools) = 56.0. Source setting: general tools; pass@1. Origin: Moonshot evaluation." }, { "model_id": "kimi-k3", "benchmark_id": "deep_swe_v1_1", "score": 67.5, "reference_url": "https://github.com/MoonshotAI/Kimi-K3/blob/main/k3_tech_report.pdf", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "benchmark-specified", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "agentic coding harness per row", "prompt_style": "default", "temperature": "1.0; top_p=0.95 for reasoning/no-tool vision, top_p=1.0 for coding/agentic", "context": "1048576 (1M)" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Kimi K3 technical report: DeepSWE = 67.5. Source setting: agentic coding harness per row. Origin: Moonshot K3 run; competitors from DeepSWE leaderboard / GLM blog.", "candidates": [ { "score": 67.3, "reference_url": "https://github.com/MoonshotAI/Kimi-K3/blob/main/k3_tech_report.pdf", "source_type": "tech_report", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "benchmark-specified", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official mini-SWE-agent leaderboard", "prompt_style": "default", "temperature": "1.0; top_p=0.95 for reasoning/no-tool vision, top_p=1.0 for coding/agentic", "context": "1048576 (1M)" }, "notes": "Kimi K3 technical report: DeepSWE (mini-SWE-agent) = 67.3. Source setting: official mini-SWE-agent leaderboard. Origin: DeepSWE leaderboard." }, { "score": 69.0, "reference_url": "https://media.x.ai/v1/website/card-4p6-4cd2dc57.pdf", "source_type": "model_card", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "agentic repository shell/editor", "sampling": "pass@1", "judge": "isolated repository verifier", "harness": "mini-swe-agent run by Datacurve", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown" }, "notes": "card46 DeepSWE v1.1 Pass@1 (%); source effort=max; benchmark protocol matches canonical=True. Unknown source fields remain unknown." } ] }, { "model_id": "kimi-k3", "benchmark_id": "program_bench", "score": 77.8, "reference_url": "https://github.com/MoonshotAI/Kimi-K3/blob/main/k3_tech_report.pdf", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "benchmark-specified", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "agentic coding harness per row", "prompt_style": "default", "temperature": "1.0; top_p=0.95 for reasoning/no-tool vision, top_p=1.0 for coding/agentic", "context": "1048576 (1M)" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Kimi K3 technical report: ProgramBench = 77.8. Source setting: agentic coding harness per row. Origin: Moonshot K3 run; competitor values from Vals AI." }, { "model_id": "kimi-k3", "benchmark_id": "terminal_bench_2_1", "score": 88.3, "reference_url": "https://github.com/MoonshotAI/Kimi-K3/blob/main/k3_tech_report.pdf", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "benchmark-specified", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "agentic coding harness per row", "prompt_style": "default", "temperature": "1.0; top_p=0.95 for reasoning/no-tool vision, top_p=1.0 for coding/agentic", "context": "1048576 (1M)" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Kimi K3 technical report: Terminal-Bench 2.1 = 88.3. Source setting: agentic coding harness per row. Origin: mixed official sources documented by Kimi report." }, { "model_id": "kimi-k3", "benchmark_id": "frontier_swe", "score": 81.2, "reference_url": "https://github.com/MoonshotAI/Kimi-K3/blob/main/k3_tech_report.pdf", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "benchmark-specified", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "agentic coding harness per row", "prompt_style": "default", "temperature": "1.0; top_p=0.95 for reasoning/no-tool vision, top_p=1.0 for coding/agentic", "context": "1048576 (1M)" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Kimi K3 technical report: FrontierSWE = 81.2. Source setting: agentic coding harness per row. Origin: Moonshot/OpenAI runs plus official leaderboard." }, { "model_id": "kimi-k3", "benchmark_id": "swe_marathon_h20_2026_07_09", "score": 42.0, "reference_url": "https://github.com/MoonshotAI/Kimi-K3/blob/main/k3_tech_report.pdf", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "benchmark-specified", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "H20-calibrated pre-final-v1.1 branch (2026-07-09)", "prompt_style": "default", "temperature": "1.0; top_p=0.95 for reasoning/no-tool vision, top_p=1.0 for coding/agentic", "context": "1048576 (1M)" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Kimi K3 technical report: SWE-Marathon = 42.0. Source setting: H20-calibrated pre-final-v1.1 branch (2026-07-09). Origin: Moonshot evaluation." }, { "model_id": "kimi-k3", "benchmark_id": "posttrain_bench", "score": 36.6, "reference_url": "https://github.com/MoonshotAI/Kimi-K3/blob/main/k3_tech_report.pdf", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "benchmark-specified", "sampling": "avg-of-3", "judge": "benchmark-specified", "harness": "H20 avg@3 for K3/Fable/Sol; official H100 for Opus/GPT-5.5/GLM", "prompt_style": "default", "temperature": "1.0; top_p=0.95 for reasoning/no-tool vision, top_p=1.0 for coding/agentic", "context": "1048576 (1M)" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Kimi K3 technical report: PostTrainBench = 36.6. Source setting: H20 avg@3 for K3/Fable/Sol; official H100 for Opus/GPT-5.5/GLM. Origin: Moonshot H20 runs plus official H100 leaderboard values." }, { "model_id": "kimi-k3", "benchmark_id": "mls_bench_lite", "score": 48.3, "reference_url": "https://github.com/MoonshotAI/Kimi-K3/blob/main/k3_tech_report.pdf", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "benchmark-specified", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "agentic coding harness per row", "prompt_style": "default", "temperature": "1.0; top_p=0.95 for reasoning/no-tool vision, top_p=1.0 for coding/agentic", "context": "1048576 (1M)" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Kimi K3 technical report: MLS-Bench-Lite = 48.3. Source setting: agentic coding harness per row. Origin: Moonshot evaluation." }, { "model_id": "kimi-k3", "benchmark_id": "scicode", "score": 58.7, "reference_url": "https://github.com/MoonshotAI/Kimi-K3/blob/main/k3_tech_report.pdf", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "benchmark-specified", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "agentic coding harness per row", "prompt_style": "default", "temperature": "1.0; top_p=0.95 for reasoning/no-tool vision, top_p=1.0 for coding/agentic", "context": "1048576 (1M)" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Kimi K3 technical report: SciCode = 58.7. Source setting: agentic coding harness per row. Origin: Artificial Analysis cited by Kimi report." }, { "model_id": "kimi-k3", "benchmark_id": "kimi_code_bench_v2", "score": 72.9, "reference_url": "https://github.com/MoonshotAI/Kimi-K3/blob/main/k3_tech_report.pdf", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "benchmark-specified", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "agentic coding harness per row", "prompt_style": "default", "temperature": "1.0; top_p=0.95 for reasoning/no-tool vision, top_p=1.0 for coding/agentic", "context": "1048576 (1M)" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Kimi K3 technical report: Kimi Code Bench 2.0 = 72.9. Source setting: agentic coding harness per row. Origin: Moonshot evaluation.", "candidates": [ { "score": 73.7, "reference_url": "https://github.com/MoonshotAI/Kimi-K3/blob/main/k3_tech_report.pdf", "source_type": "tech_report", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "benchmark-specified", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "Claude Code harness", "prompt_style": "default", "temperature": "1.0; top_p=0.95 for reasoning/no-tool vision, top_p=1.0 for coding/agentic", "context": "1048576 (1M)" }, "notes": "Kimi K3 technical report: Kimi Code Bench 2.0 (Claude Code) = 73.7. Source setting: Claude Code harness. Origin: Moonshot evaluation." } ] }, { "model_id": "kimi-k3", "benchmark_id": "browsecomp", "score": 91.2, "reference_url": "https://github.com/MoonshotAI/Kimi-K3/blob/main/k3_tech_report.pdf", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "benchmark-specified", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "agentic benchmark harness", "prompt_style": "default", "temperature": "1.0; top_p=0.95 for reasoning/no-tool vision, top_p=1.0 for coding/agentic", "context": "1048576 (1M)" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Kimi K3 technical report: BrowseComp = 91.2. Source setting: agentic benchmark harness. Origin: Moonshot K3 run; competitor provider sources.", "candidates": [ { "score": 90.4, "reference_url": "https://github.com/MoonshotAI/Kimi-K3/blob/main/k3_tech_report.pdf", "source_type": "tech_report", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "benchmark-specified", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "full 1M context; no compaction", "prompt_style": "default", "temperature": "1.0; top_p=0.95 for reasoning/no-tool vision, top_p=1.0 for coding/agentic", "context": "1048576 (1M)" }, "notes": "Kimi K3 technical report: BrowseComp (no context management) = 90.4. Source setting: full 1M context; no compaction. Origin: Moonshot evaluation." } ] }, { "model_id": "kimi-k3", "benchmark_id": "deepsearchqa_f1", "score": 95.0, "reference_url": "https://github.com/MoonshotAI/Kimi-K3/blob/main/k3_tech_report.pdf", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "benchmark-specified", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "agentic benchmark harness", "prompt_style": "default", "temperature": "1.0; top_p=0.95 for reasoning/no-tool vision, top_p=1.0 for coding/agentic", "context": "1048576 (1M)" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Kimi K3 technical report: DeepSearchQA (F1) = 95.0. Source setting: agentic benchmark harness. Origin: Moonshot evaluation." }, { "model_id": "kimi-k3", "benchmark_id": "researchrubrics", "score": 76.2, "reference_url": "https://github.com/MoonshotAI/Kimi-K3/blob/main/k3_tech_report.pdf", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "benchmark-specified", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "agentic benchmark harness", "prompt_style": "default", "temperature": "1.0; top_p=0.95 for reasoning/no-tool vision, top_p=1.0 for coding/agentic", "context": "1048576 (1M)" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Kimi K3 technical report: ResearchRubrics = 76.2. Source setting: agentic benchmark harness. Origin: Moonshot evaluation." }, { "model_id": "kimi-k3", "benchmark_id": "gdpval_aa_elo", "score": 1686.0, "reference_url": "https://github.com/MoonshotAI/Kimi-K3/blob/main/k3_tech_report.pdf", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "benchmark-specified", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "agentic benchmark harness", "prompt_style": "default", "temperature": "1.0; top_p=0.95 for reasoning/no-tool vision, top_p=1.0 for coding/agentic", "context": "1048576 (1M)" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Kimi K3 technical report: GDPval-AA v2 (Elo) = 1686.0. Source setting: agentic benchmark harness. Origin: Artificial Analysis cited by Kimi report." }, { "model_id": "kimi-k3", "benchmark_id": "toolathlon_verified", "score": 76.5, "reference_url": "https://github.com/MoonshotAI/Kimi-K3/blob/main/k3_tech_report.pdf", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "benchmark-specified", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "Verified edition; official leaderboard snapshot", "prompt_style": "default", "temperature": "1.0; top_p=0.95 for reasoning/no-tool vision, top_p=1.0 for coding/agentic", "context": "1048576 (1M)" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Kimi K3 technical report: Toolathlon-Verified = 76.5. Source setting: Verified edition; official leaderboard snapshot. Origin: official Toolathlon leaderboard cited by Kimi report." }, { "model_id": "kimi-k3", "benchmark_id": "mcpmark_verified", "score": 94.5, "reference_url": "https://github.com/MoonshotAI/Kimi-K3/blob/main/k3_tech_report.pdf", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "benchmark-specified", "sampling": "not restated", "judge": "benchmark-specified", "harness": "verified tasks; sampling not restated in K3 report", "prompt_style": "default", "temperature": "1.0; top_p=0.95 for reasoning/no-tool vision, top_p=1.0 for coding/agentic", "context": "1048576 (1M)" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Kimi K3 technical report: MCPMark-Verified = 94.5. Source setting: verified tasks; sampling not restated in K3 report. Origin: Moonshot evaluation. Canonical compatibility remains under review; stored as best available primary with matches_canonical=false." }, { "model_id": "kimi-k3", "benchmark_id": "mcpatlas", "score": 84.2, "reference_url": "https://github.com/MoonshotAI/Kimi-K3/blob/main/k3_tech_report.pdf", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "benchmark-specified", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "public 500; maxTurns=100; Gemini 3.1 Pro judge", "prompt_style": "default", "temperature": "1.0; top_p=0.95 for reasoning/no-tool vision, top_p=1.0 for coding/agentic", "context": "1048576 (1M)" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Kimi K3 technical report: MCP-Atlas = 84.2. Source setting: public 500; maxTurns=100; Gemini 3.1 Pro judge. Origin: Moonshot evaluation." }, { "model_id": "kimi-k3", "benchmark_id": "automation_bench", "score": 30.8, "reference_url": "https://github.com/MoonshotAI/Kimi-K3/blob/main/k3_tech_report.pdf", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "benchmark-specified", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "agentic benchmark harness", "prompt_style": "default", "temperature": "1.0; top_p=0.95 for reasoning/no-tool vision, top_p=1.0 for coding/agentic", "context": "1048576 (1M)" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Kimi K3 technical report: AutomationBench = 30.8. Source setting: agentic benchmark harness. Origin: Moonshot evaluation." }, { "model_id": "kimi-k3", "benchmark_id": "job_bench", "score": 54.3, "reference_url": "https://github.com/MoonshotAI/Kimi-K3/blob/main/k3_tech_report.pdf", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "benchmark-specified", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "agentic benchmark harness", "prompt_style": "default", "temperature": "1.0; top_p=0.95 for reasoning/no-tool vision, top_p=1.0 for coding/agentic", "context": "1048576 (1M)" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Kimi K3 technical report: JobBench = 54.3. Source setting: agentic benchmark harness. Origin: official JobBench leaderboard cited by Kimi report." }, { "model_id": "kimi-k3", "benchmark_id": "aa_briefcase_elo", "score": 1548.0, "reference_url": "https://github.com/MoonshotAI/Kimi-K3/blob/main/k3_tech_report.pdf", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "benchmark-specified", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "agentic benchmark harness", "prompt_style": "default", "temperature": "1.0; top_p=0.95 for reasoning/no-tool vision, top_p=1.0 for coding/agentic", "context": "1048576 (1M)" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Kimi K3 technical report: AA-Briefcase (Elo) = 1548.0. Source setting: agentic benchmark harness. Origin: Artificial Analysis cited by Kimi report.", "candidates": [ { "score": 1541, "reference_url": "https://media.x.ai/v1/website/card-4p6-4cd2dc57.pdf", "source_type": "model_card", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "offline multi-file knowledge-work tools", "sampling": "unknown", "judge": "rubric plus analytical and presentation pairwise Elo", "harness": "Artificial Analysis AA-Briefcase harness", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown" }, "notes": "card46 AA Briefcase AA-Briefcase Elo; source effort=max; benchmark protocol matches canonical=True. Unknown source fields remain unknown." } ] }, { "model_id": "kimi-k3", "benchmark_id": "agents_last_exam", "score": 28.3, "reference_url": "https://github.com/MoonshotAI/Kimi-K3/blob/main/k3_tech_report.pdf", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "benchmark-specified", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "agentic benchmark harness", "prompt_style": "default", "temperature": "1.0; top_p=0.95 for reasoning/no-tool vision, top_p=1.0 for coding/agentic", "context": "1048576 (1M)" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Kimi K3 technical report: Agents' Last Exam = 28.3. Source setting: agentic benchmark harness. Origin: official leaderboard cited by Kimi report." }, { "model_id": "kimi-k3", "benchmark_id": "apex_agents", "score": 41.0, "reference_url": "https://github.com/MoonshotAI/Kimi-K3/blob/main/k3_tech_report.pdf", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "benchmark-specified", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "agentic benchmark harness", "prompt_style": "default", "temperature": "1.0; top_p=0.95 for reasoning/no-tool vision, top_p=1.0 for coding/agentic", "context": "1048576 (1M)" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Kimi K3 technical report: APEX-Agents = 41.0. Source setting: agentic benchmark harness. Origin: Artificial Analysis cited by Kimi report.", "candidates": [ { "score": 55.4, "reference_url": "https://media.x.ai/v1/website/card-4p6-4cd2dc57.pdf", "source_type": "model_card", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "multi-application professional tools", "sampling": "pass@1", "judge": "expert all-criteria-pass rubric", "harness": "Mercor APEX-Agents harness", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown" }, "notes": "card46 APEX-Agents Pass@1 (%); source effort=max; benchmark protocol matches canonical=True. Unknown source fields remain unknown." } ] }, { "model_id": "kimi-k3", "benchmark_id": "officeqa_pro", "score": 63.3, "reference_url": "https://github.com/MoonshotAI/Kimi-K3/blob/main/k3_tech_report.pdf", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "benchmark-specified", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "PDF corpus rendered as images; no machine-readable text", "prompt_style": "default", "temperature": "1.0; top_p=0.95 for reasoning/no-tool vision, top_p=1.0 for coding/agentic", "context": "1048576 (1M)" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Kimi K3 technical report: OfficeQA Pro = 63.3. Source setting: PDF corpus rendered as images; no machine-readable text. Origin: Moonshot evaluation." }, { "model_id": "kimi-k3", "benchmark_id": "spreadsheetbench_2", "score": 34.8, "reference_url": "https://github.com/MoonshotAI/Kimi-K3/blob/main/k3_tech_report.pdf", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "benchmark-specified", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "agentic benchmark harness", "prompt_style": "default", "temperature": "1.0; top_p=0.95 for reasoning/no-tool vision, top_p=1.0 for coding/agentic", "context": "1048576 (1M)" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Kimi K3 technical report: SpreadsheetBench 2 = 34.8. Source setting: agentic benchmark harness. Origin: Moonshot evaluation." }, { "model_id": "kimi-k3", "benchmark_id": "osworld_verified", "score": 84.8, "reference_url": "https://github.com/MoonshotAI/Kimi-K3/blob/main/k3_tech_report.pdf", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "benchmark-specified", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "agentic benchmark harness", "prompt_style": "default", "temperature": "1.0; top_p=0.95 for reasoning/no-tool vision, top_p=1.0 for coding/agentic", "context": "1048576 (1M)" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Kimi K3 technical report: OSWorld-Verified = 84.8. Source setting: agentic benchmark harness. Origin: Moonshot evaluation." }, { "model_id": "kimi-k3", "benchmark_id": "osworld_2_0", "score": 58.3, "reference_url": "https://github.com/MoonshotAI/Kimi-K3/blob/main/k3_tech_report.pdf", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "benchmark-specified", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "agentic benchmark harness", "prompt_style": "default", "temperature": "1.0; top_p=0.95 for reasoning/no-tool vision, top_p=1.0 for coding/agentic", "context": "1048576 (1M)" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Kimi K3 technical report: OSWorld 2.0 = 58.3. Source setting: agentic benchmark harness. Origin: Moonshot evaluation." }, { "model_id": "kimi-k3", "benchmark_id": "saas_bench", "score": 60.1, "reference_url": "https://github.com/MoonshotAI/Kimi-K3/blob/main/k3_tech_report.pdf", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "benchmark-specified", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "agentic benchmark harness", "prompt_style": "default", "temperature": "1.0; top_p=0.95 for reasoning/no-tool vision, top_p=1.0 for coding/agentic", "context": "1048576 (1M)" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Kimi K3 technical report: SaaS-Bench = 60.1. Source setting: agentic benchmark harness. Origin: Moonshot evaluation." }, { "model_id": "kimi-k3", "benchmark_id": "tau3_banking", "score": 33.4, "reference_url": "https://github.com/MoonshotAI/Kimi-K3/blob/main/k3_tech_report.pdf", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "benchmark-specified", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "agentic benchmark harness", "prompt_style": "default", "temperature": "1.0; top_p=0.95 for reasoning/no-tool vision, top_p=1.0 for coding/agentic", "context": "1048576 (1M)" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Kimi K3 technical report: τ³-Banking = 33.4. Source setting: agentic benchmark harness. Origin: Artificial Analysis cited by Kimi report." }, { "model_id": "kimi-k3", "benchmark_id": "harvey_lab_aa", "score": 94.6, "reference_url": "https://github.com/MoonshotAI/Kimi-K3/blob/main/k3_tech_report.pdf", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "benchmark-specified", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "agentic benchmark harness", "prompt_style": "default", "temperature": "1.0; top_p=0.95 for reasoning/no-tool vision, top_p=1.0 for coding/agentic", "context": "1048576 (1M)" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Kimi K3 technical report: Harvey Lab-AA = 94.6. Source setting: agentic benchmark harness. Origin: Artificial Analysis cited by Kimi report." }, { "model_id": "kimi-k3", "benchmark_id": "corpfin_v2", "score": 71.6, "reference_url": "https://github.com/MoonshotAI/Kimi-K3/blob/main/k3_tech_report.pdf", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "benchmark-specified", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "agentic benchmark harness", "prompt_style": "default", "temperature": "1.0; top_p=0.95 for reasoning/no-tool vision, top_p=1.0 for coding/agentic", "context": "1048576 (1M)" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Kimi K3 technical report: CorpFin v2 = 71.6. Source setting: agentic benchmark harness. Origin: Vals AI cited by Kimi report." }, { "model_id": "kimi-k3", "benchmark_id": "finance_agent_v2", "score": 54.4, "reference_url": "https://github.com/MoonshotAI/Kimi-K3/blob/main/k3_tech_report.pdf", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "benchmark-specified", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "agentic benchmark harness", "prompt_style": "default", "temperature": "1.0; top_p=0.95 for reasoning/no-tool vision, top_p=1.0 for coding/agentic", "context": "1048576 (1M)" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Kimi K3 technical report: Finance Agent v2 = 54.4. Source setting: agentic benchmark harness. Origin: Vals AI cited by Kimi report." }, { "model_id": "kimi-k3", "benchmark_id": "legal_research_bench", "score": 44.2, "reference_url": "https://github.com/MoonshotAI/Kimi-K3/blob/main/k3_tech_report.pdf", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "benchmark-specified", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "agentic benchmark harness", "prompt_style": "default", "temperature": "1.0; top_p=0.95 for reasoning/no-tool vision, top_p=1.0 for coding/agentic", "context": "1048576 (1M)" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Kimi K3 technical report: Legal Research Bench = 44.2. Source setting: agentic benchmark harness. Origin: Vals AI cited by Kimi report." }, { "model_id": "kimi-k3", "benchmark_id": "worldvqa", "score": 51.0, "reference_url": "https://github.com/MoonshotAI/Kimi-K3/blob/main/k3_tech_report.pdf", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "none", "sampling": "avg-of-3", "judge": "benchmark-specified", "harness": "forced-answer prompt; avg 3 runs", "prompt_style": "default", "temperature": "1.0; top_p=0.95 for reasoning/no-tool vision, top_p=1.0 for coding/agentic", "context": "1048576 (1M)" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Kimi K3 technical report: WorldVQA ForceAnswer = 51.0. Source setting: forced-answer prompt; avg 3 runs. Origin: Moonshot evaluation." }, { "model_id": "kimi-k3", "benchmark_id": "omnidocbench", "score": 0.089, "reference_url": "https://github.com/MoonshotAI/Kimi-K3/blob/main/k3_tech_report.pdf", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "benchmark-specified", "sampling": "avg-of-3", "judge": "benchmark-specified", "harness": "avg 3 runs; source 1-NED score normalized to NED", "prompt_style": "default", "temperature": "1.0; top_p=0.95 for reasoning/no-tool vision, top_p=1.0 for coding/agentic", "context": "1048576 (1M)" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Kimi K3 technical report: OmniDocBench = 0.089 NED. Source setting: avg 3 runs; source 1-NED score normalized to NED. Origin: Moonshot evaluation. The source reports 1-NED=91.1%; stored value is normalized NED=0.089." }, { "model_id": "kimi-k3", "benchmark_id": "perception_bench", "score": 58.5, "reference_url": "https://github.com/MoonshotAI/Kimi-K3/blob/main/k3_tech_report.pdf", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "benchmark-specified", "sampling": "avg-of-3", "judge": "benchmark-specified", "harness": "multimodal; avg 3 runs", "prompt_style": "default", "temperature": "1.0; top_p=0.95 for reasoning/no-tool vision, top_p=1.0 for coding/agentic", "context": "1048576 (1M)" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Kimi K3 technical report: PerceptionBench = 58.5. Source setting: multimodal; avg 3 runs. Origin: Moonshot evaluation." }, { "model_id": "kimi-k3", "benchmark_id": "video_mme", "score": 90.0, "reference_url": "https://github.com/MoonshotAI/Kimi-K3/blob/main/k3_tech_report.pdf", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "none", "sampling": "avg-of-3", "judge": "benchmark-specified", "harness": "audio+visual+subtitles; avg 3 runs", "prompt_style": "default", "temperature": "1.0; top_p=0.95 for reasoning/no-tool vision, top_p=1.0 for coding/agentic", "context": "1048576 (1M)" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Kimi K3 technical report: Video-MME (w. sub) = 90.0. Source setting: audio+visual+subtitles; avg 3 runs. Origin: Moonshot evaluation." }, { "model_id": "kimi-k3", "benchmark_id": "mmvu", "score": 82.1, "reference_url": "https://github.com/MoonshotAI/Kimi-K3/blob/main/k3_tech_report.pdf", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "benchmark-specified", "sampling": "avg-of-3", "judge": "benchmark-specified", "harness": "avg 3 runs", "prompt_style": "default", "temperature": "1.0; top_p=0.95 for reasoning/no-tool vision, top_p=1.0 for coding/agentic", "context": "1048576 (1M)" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Kimi K3 technical report: MMVU = 82.1. Source setting: avg 3 runs. Origin: Moonshot evaluation." }, { "model_id": "kimi-k3", "benchmark_id": "babyvision", "score": 85.7, "reference_url": "https://github.com/MoonshotAI/Kimi-K3/blob/main/k3_tech_report.pdf", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "python", "sampling": "avg-of-3", "judge": "benchmark-specified", "harness": "Python tool; avg 3 runs", "prompt_style": "default", "temperature": "1.0; top_p=0.95 for reasoning/no-tool vision, top_p=1.0 for coding/agentic", "context": "1048576 (1M)" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Kimi K3 technical report: BabyVision w/ python = 85.7. Source setting: Python tool; avg 3 runs. Origin: Moonshot evaluation." }, { "model_id": "kimi-k3", "benchmark_id": "mmmu_pro", "score": 81.6, "reference_url": "https://github.com/MoonshotAI/Kimi-K3/blob/main/k3_tech_report.pdf", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "none", "sampling": "avg-of-3", "judge": "benchmark-specified", "harness": "vision; avg 3 runs", "prompt_style": "default", "temperature": "1.0; top_p=0.95 for reasoning/no-tool vision, top_p=1.0 for coding/agentic", "context": "1048576 (1M)" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Kimi K3 technical report: MMMU-Pro (no Python) = 81.6. Source setting: vision; avg 3 runs. Origin: Moonshot evaluation.", "candidates": [ { "score": 83.4, "reference_url": "https://github.com/MoonshotAI/Kimi-K3/blob/main/k3_tech_report.pdf", "source_type": "tech_report", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "python", "sampling": "avg-of-3", "judge": "benchmark-specified", "harness": "Python tool; avg 3 runs", "prompt_style": "default", "temperature": "1.0; top_p=0.95 for reasoning/no-tool vision, top_p=1.0 for coding/agentic", "context": "1048576 (1M)" }, "notes": "Kimi K3 technical report: MMMU-Pro (with Python) = 83.4. Source setting: Python tool; avg 3 runs. Origin: Moonshot evaluation." } ] }, { "model_id": "kimi-k3", "benchmark_id": "charxiv_reasoning", "score": 84.8, "reference_url": "https://github.com/MoonshotAI/Kimi-K3/blob/main/k3_tech_report.pdf", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "none", "sampling": "avg-of-3", "judge": "benchmark-specified", "harness": "vision; avg 3 runs", "prompt_style": "default", "temperature": "1.0; top_p=0.95 for reasoning/no-tool vision, top_p=1.0 for coding/agentic", "context": "1048576 (1M)" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Kimi K3 technical report: CharXiv (RQ) (no Python) = 84.8. Source setting: vision; avg 3 runs. Origin: Moonshot evaluation.", "candidates": [ { "score": 91.3, "reference_url": "https://github.com/MoonshotAI/Kimi-K3/blob/main/k3_tech_report.pdf", "source_type": "tech_report", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "python", "sampling": "avg-of-3", "judge": "benchmark-specified", "harness": "Python tool; avg 3 runs", "prompt_style": "default", "temperature": "1.0; top_p=0.95 for reasoning/no-tool vision, top_p=1.0 for coding/agentic", "context": "1048576 (1M)" }, "notes": "Kimi K3 technical report: CharXiv (RQ) (with Python) = 91.3. Source setting: Python tool; avg 3 runs. Origin: Moonshot evaluation." } ] }, { "model_id": "kimi-k3", "benchmark_id": "mathvision", "score": 94.3, "reference_url": "https://github.com/MoonshotAI/Kimi-K3/blob/main/k3_tech_report.pdf", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "none", "sampling": "avg-of-3", "judge": "benchmark-specified", "harness": "vision; avg 3 runs", "prompt_style": "default", "temperature": "1.0; top_p=0.95 for reasoning/no-tool vision, top_p=1.0 for coding/agentic", "context": "1048576 (1M)" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Kimi K3 technical report: MathVision (no Python) = 94.3. Source setting: vision; avg 3 runs. Origin: Moonshot evaluation.", "candidates": [ { "score": 97.8, "reference_url": "https://github.com/MoonshotAI/Kimi-K3/blob/main/k3_tech_report.pdf", "source_type": "tech_report", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "python", "sampling": "avg-of-3", "judge": "benchmark-specified", "harness": "Python tool; avg 3 runs", "prompt_style": "default", "temperature": "1.0; top_p=0.95 for reasoning/no-tool vision, top_p=1.0 for coding/agentic", "context": "1048576 (1M)" }, "notes": "Kimi K3 technical report: MathVision (with Python) = 97.8. Source setting: Python tool; avg 3 runs. Origin: Moonshot evaluation." } ] }, { "model_id": "kimi-k3", "benchmark_id": "zerobench_main", "score": 23.0, "reference_url": "https://github.com/MoonshotAI/Kimi-K3/blob/main/k3_tech_report.pdf", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "benchmark-specified", "sampling": "pass@5", "judge": "benchmark-specified", "harness": "no Python; pass@5 / 5 runs", "prompt_style": "default", "temperature": "1.0; top_p=0.95 for reasoning/no-tool vision, top_p=1.0 for coding/agentic", "context": "1048576 (1M)" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Kimi K3 technical report: ZeroBench (pass@5) (no Python) = 23.0. Source setting: no Python; pass@5 / 5 runs. Origin: Moonshot evaluation. Canonical compatibility remains under review; stored as best available primary with matches_canonical=false." }, { "model_id": "kimi-k3", "benchmark_id": "zerobench_tools", "score": 41.0, "reference_url": "https://github.com/MoonshotAI/Kimi-K3/blob/main/k3_tech_report.pdf", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "python", "sampling": "pass@5", "judge": "benchmark-specified", "harness": "Python tool; pass@5 / 5 runs", "prompt_style": "default", "temperature": "1.0; top_p=0.95 for reasoning/no-tool vision, top_p=1.0 for coding/agentic", "context": "1048576 (1M)" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Kimi K3 technical report: ZeroBench (pass@5) (with Python) = 41.0. Source setting: Python tool; pass@5 / 5 runs. Origin: Moonshot evaluation. Canonical compatibility remains under review; stored as best available primary with matches_canonical=false." }, { "model_id": "kimi-k3", "benchmark_id": "coding_experience", "score": 56.6, "reference_url": "https://github.com/MoonshotAI/Kimi-K3/blob/main/k3_tech_report.pdf", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "benchmark-specified", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "provider-aligned harness: Kimi Code / Claude Code / Codex", "prompt_style": "default", "temperature": "1.0; top_p=0.95 for reasoning/no-tool vision, top_p=1.0 for coding/agentic", "context": "1048576 (1M)" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Kimi K3 technical report: Coding Experience = 56.6. Source setting: provider-aligned harness: Kimi Code / Claude Code / Codex. Origin: Moonshot evaluation.", "candidates": [ { "score": 59.9, "reference_url": "https://github.com/MoonshotAI/Kimi-K3/blob/main/k3_tech_report.pdf", "source_type": "tech_report", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "benchmark-specified", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "Claude Code harness", "prompt_style": "default", "temperature": "1.0; top_p=0.95 for reasoning/no-tool vision, top_p=1.0 for coding/agentic", "context": "1048576 (1M)" }, "notes": "Kimi K3 technical report: Coding Experience (Claude Code) = 59.9. Source setting: Claude Code harness. Origin: Moonshot evaluation." } ] }, { "model_id": "kimi-k3", "benchmark_id": "clawbench_2_0", "score": 48.3, "reference_url": "https://github.com/MoonshotAI/Kimi-K3/blob/main/k3_tech_report.pdf", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "benchmark-specified", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "OpenClaw harness", "prompt_style": "default", "temperature": "1.0; top_p=0.95 for reasoning/no-tool vision, top_p=1.0 for coding/agentic", "context": "1048576 (1M)" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Kimi K3 technical report: 24/7 ClawBench 2.0 = 48.3. Source setting: OpenClaw harness. Origin: Moonshot evaluation." }, { "model_id": "kimi-k3", "benchmark_id": "mira_bench", "score": 64.1, "reference_url": "https://github.com/MoonshotAI/Kimi-K3/blob/main/k3_tech_report.pdf", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "benchmark-specified", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "MIRA harness", "prompt_style": "default", "temperature": "1.0; top_p=0.95 for reasoning/no-tool vision, top_p=1.0 for coding/agentic", "context": "1048576 (1M)" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Kimi K3 technical report: MIRA Bench = 64.1. Source setting: MIRA harness. Origin: Moonshot evaluation." }, { "model_id": "kimi-k3", "benchmark_id": "kaet", "score": 83.5, "reference_url": "https://github.com/MoonshotAI/Kimi-K3/blob/main/k3_tech_report.pdf", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "benchmark-specified", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "Kimi Code harness", "prompt_style": "default", "temperature": "1.0; top_p=0.95 for reasoning/no-tool vision, top_p=1.0 for coding/agentic", "context": "1048576 (1M)" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Kimi K3 technical report: KAET = 83.5. Source setting: Kimi Code harness. Origin: Moonshot evaluation." }, { "model_id": "kimi-k3", "benchmark_id": "clif_bench", "score": 52.4, "reference_url": "https://github.com/MoonshotAI/Kimi-K3/blob/main/k3_tech_report.pdf", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "benchmark-specified", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "Kimi Code harness", "prompt_style": "default", "temperature": "1.0; top_p=0.95 for reasoning/no-tool vision, top_p=1.0 for coding/agentic", "context": "1048576 (1M)" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Kimi K3 technical report: CLIF Bench = 52.4. Source setting: Kimi Code harness. Origin: Moonshot evaluation." }, { "model_id": "kimi-k3", "benchmark_id": "agentic_vision_bench", "score": 78.3, "reference_url": "https://github.com/MoonshotAI/Kimi-K3/blob/main/k3_tech_report.pdf", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "benchmark-specified", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "Kimi Code harness", "prompt_style": "default", "temperature": "1.0; top_p=0.95 for reasoning/no-tool vision, top_p=1.0 for coding/agentic", "context": "1048576 (1M)" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Kimi K3 technical report: Agentic Vision Bench = 78.3. Source setting: Kimi Code harness. Origin: Moonshot evaluation." }, { "model_id": "kimi-k3", "benchmark_id": "swarm_bench", "score": 76.3, "reference_url": "https://github.com/MoonshotAI/Kimi-K3/blob/main/k3_tech_report.pdf", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "benchmark-specified", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "Kimi Agent harness", "prompt_style": "default", "temperature": "1.0; top_p=0.95 for reasoning/no-tool vision, top_p=1.0 for coding/agentic", "context": "1048576 (1M)" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Kimi K3 technical report: SwarmBench = 76.3. Source setting: Kimi Agent harness. Origin: Moonshot evaluation." }, { "model_id": "kimi-k3", "benchmark_id": "online_experience", "score": 77.9, "reference_url": "https://github.com/MoonshotAI/Kimi-K3/blob/main/k3_tech_report.pdf", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "benchmark-specified", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "Kimi Agent harness", "prompt_style": "default", "temperature": "1.0; top_p=0.95 for reasoning/no-tool vision, top_p=1.0 for coding/agentic", "context": "1048576 (1M)" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Kimi K3 technical report: Online Experience = 77.9. Source setting: Kimi Agent harness. Origin: Moonshot evaluation." }, { "model_id": "kimi-k3", "benchmark_id": "deepresearchbench", "score": 90.0, "reference_url": "https://github.com/MoonshotAI/Kimi-K3/blob/main/k3_tech_report.pdf", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "benchmark-specified", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "Kimi Agent harness", "prompt_style": "default", "temperature": "1.0; top_p=0.95 for reasoning/no-tool vision, top_p=1.0 for coding/agentic", "context": "1048576 (1M)" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Kimi K3 technical report: DeepResearchBench = 90.0. Source setting: Kimi Agent harness. Origin: Moonshot evaluation." }, { "model_id": "kimi-k3", "benchmark_id": "finance_bench", "score": 62.6, "reference_url": "https://github.com/MoonshotAI/Kimi-K3/blob/main/k3_tech_report.pdf", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "benchmark-specified", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "N/A harness", "prompt_style": "default", "temperature": "1.0; top_p=0.95 for reasoning/no-tool vision, top_p=1.0 for coding/agentic", "context": "1048576 (1M)" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Kimi K3 technical report: Finance Bench = 62.6. Source setting: N/A harness. Origin: Moonshot evaluation." }, { "model_id": "kimi-k3", "benchmark_id": "kwv_bench", "score": 64.7, "reference_url": "https://github.com/MoonshotAI/Kimi-K3/blob/main/k3_tech_report.pdf", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "benchmark-specified", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "N/A harness", "prompt_style": "default", "temperature": "1.0; top_p=0.95 for reasoning/no-tool vision, top_p=1.0 for coding/agentic", "context": "1048576 (1M)" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Kimi K3 technical report: KWVBench = 64.7. Source setting: N/A harness. Origin: Moonshot evaluation." }, { "model_id": "kimi-k3", "benchmark_id": "deck_bench", "score": 73.5, "reference_url": "https://github.com/MoonshotAI/Kimi-K3/blob/main/k3_tech_report.pdf", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "benchmark-specified", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "N/A harness", "prompt_style": "default", "temperature": "1.0; top_p=0.95 for reasoning/no-tool vision, top_p=1.0 for coding/agentic", "context": "1048576 (1M)" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Kimi K3 technical report: DECKBench = 73.5. Source setting: N/A harness. Origin: Moonshot evaluation." }, { "model_id": "kimi-k3", "benchmark_id": "agent_behavior_bench", "score": 65.0, "reference_url": "https://github.com/MoonshotAI/Kimi-K3/blob/main/k3_tech_report.pdf", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "benchmark-specified", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "Kimi Work harness", "prompt_style": "default", "temperature": "1.0; top_p=0.95 for reasoning/no-tool vision, top_p=1.0 for coding/agentic", "context": "1048576 (1M)" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Kimi K3 technical report: Agent Behavior Bench = 65.0. Source setting: Kimi Work harness. Origin: Moonshot evaluation." }, { "model_id": "kimi-k3", "benchmark_id": "faithfulness", "score": 85.5, "reference_url": "https://github.com/MoonshotAI/Kimi-K3/blob/main/k3_tech_report.pdf", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "benchmark-specified", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "N/A harness", "prompt_style": "default", "temperature": "1.0; top_p=0.95 for reasoning/no-tool vision, top_p=1.0 for coding/agentic", "context": "1048576 (1M)" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Kimi K3 technical report: Faithfulness (1 - hallucination rate) = 85.5. Source setting: N/A harness. Origin: Moonshot evaluation." }, { "model_id": "kimi-k3", "benchmark_id": "chat_all_in_one_bench", "score": 85.2, "reference_url": "https://github.com/MoonshotAI/Kimi-K3/blob/main/k3_tech_report.pdf", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "benchmark-specified", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "Kimi Work harness", "prompt_style": "default", "temperature": "1.0; top_p=0.95 for reasoning/no-tool vision, top_p=1.0 for coding/agentic", "context": "1048576 (1M)" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Kimi K3 technical report: Chat All-in-One Bench = 85.2. Source setting: Kimi Work harness. Origin: Moonshot evaluation." }, { "model_id": "kimi-k3", "benchmark_id": "aa_intelligence_index_v4_1", "score": 57.1, "reference_url": "https://github.com/MoonshotAI/Kimi-K3/blob/main/k3_tech_report.pdf", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "benchmark-specified", "sampling": "leaderboard snapshot 2026-07-23", "judge": "benchmark-specified", "harness": "live leaderboard snapshot as of 2026-07-23", "prompt_style": "default", "temperature": "1.0; top_p=0.95 for reasoning/no-tool vision, top_p=1.0 for coding/agentic", "context": "1048576 (1M)" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Kimi K3 technical report: AA Intelligence Index v4.1 = 57.1. Source setting: live leaderboard snapshot as of 2026-07-23. Origin: third-party leaderboard cited by Kimi report." }, { "model_id": "kimi-k3", "benchmark_id": "vals_index", "score": 74.7, "reference_url": "https://github.com/MoonshotAI/Kimi-K3/blob/main/k3_tech_report.pdf", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "benchmark-specified", "sampling": "leaderboard snapshot 2026-07-23", "judge": "benchmark-specified", "harness": "live leaderboard snapshot as of 2026-07-23", "prompt_style": "default", "temperature": "1.0; top_p=0.95 for reasoning/no-tool vision, top_p=1.0 for coding/agentic", "context": "1048576 (1M)" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Kimi K3 technical report: Vals Index = 74.7. Source setting: live leaderboard snapshot as of 2026-07-23. Origin: third-party leaderboard cited by Kimi report.", "candidates": [ { "score": 74.7, "reference_url": "https://media.x.ai/v1/website/card-4p6-4cd2dc57.pdf", "source_type": "model_card", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "benchmark-specified professional tools", "sampling": "unknown", "judge": "Vals weighted suite score", "harness": "Vals Index live harness", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown" }, "notes": "card46 Vals Index Suite score (%); source effort=max; benchmark protocol matches canonical=True. Unknown source fields remain unknown." } ] }, { "model_id": "kimi-k3", "benchmark_id": "webdev_arena_elo", "score": 1678.0, "reference_url": "https://github.com/MoonshotAI/Kimi-K3/blob/main/k3_tech_report.pdf", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "benchmark-specified", "sampling": "leaderboard snapshot 2026-07-23", "judge": "benchmark-specified", "harness": "live leaderboard snapshot as of 2026-07-23", "prompt_style": "default", "temperature": "1.0; top_p=0.95 for reasoning/no-tool vision, top_p=1.0 for coding/agentic", "context": "1048576 (1M)" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Kimi K3 technical report: WebDevArena Elo = 1678. Source setting: live leaderboard snapshot as of 2026-07-23. Origin: third-party leaderboard cited by Kimi report." }, { "model_id": "kimi-k3", "benchmark_id": "text_arena_elo", "score": 1486.0, "reference_url": "https://github.com/MoonshotAI/Kimi-K3/blob/main/k3_tech_report.pdf", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "benchmark-specified", "sampling": "leaderboard snapshot 2026-07-23", "judge": "benchmark-specified", "harness": "live leaderboard snapshot as of 2026-07-23", "prompt_style": "default", "temperature": "1.0; top_p=0.95 for reasoning/no-tool vision, top_p=1.0 for coding/agentic", "context": "1048576 (1M)" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Kimi K3 technical report: TextArena Elo = 1486. Source setting: live leaderboard snapshot as of 2026-07-23. Origin: third-party leaderboard cited by Kimi report." }, { "model_id": "kimi-k3", "benchmark_id": "agent_arena", "score": 9.1, "reference_url": "https://github.com/MoonshotAI/Kimi-K3/blob/main/k3_tech_report.pdf", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "benchmark-specified", "sampling": "leaderboard snapshot 2026-07-23", "judge": "benchmark-specified", "harness": "live leaderboard snapshot as of 2026-07-23", "prompt_style": "default", "temperature": "1.0; top_p=0.95 for reasoning/no-tool vision, top_p=1.0 for coding/agentic", "context": "1048576 (1M)" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Kimi K3 technical report: AgentArena = 9.1. Source setting: live leaderboard snapshot as of 2026-07-23. Origin: third-party leaderboard cited by Kimi report." }, { "model_id": "kimi-k3", "benchmark_id": "moonshot_exploit_development_suite", "score": 38.9, "reference_url": "https://github.com/MoonshotAI/Kimi-K3/blob/main/k3_tech_report.pdf", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "benchmark-specified", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "36-task in-house exploit suite", "prompt_style": "default", "temperature": "1.0; top_p=0.95 for reasoning/no-tool vision, top_p=1.0 for coding/agentic", "context": "1048576 (1M)" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Kimi K3 technical report: Moonshot exploit-development suite = 38.9. Source setting: 36-task in-house exploit suite. Origin: Moonshot evaluation." }, { "model_id": "kimi-k3", "benchmark_id": "exploitbench", "score": 32.0, "reference_url": "https://github.com/MoonshotAI/Kimi-K3/blob/main/k3_tech_report.pdf", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "benchmark-specified", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "independent joint assessment", "prompt_style": "default", "temperature": "1.0; top_p=0.95 for reasoning/no-tool vision, top_p=1.0 for coding/agentic", "context": "1048576 (1M)" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Kimi K3 technical report: ExploitBench (UK AISI / NIST CAISI) = 32.0. Source setting: independent joint assessment. Origin: UK AISI / NIST CAISI cited by Kimi report." }, { "model_id": "claude-fable-5", "benchmark_id": "gpqa_diamond", "score": 92.6, "reference_url": "https://github.com/MoonshotAI/Kimi-K3/blob/main/k3_tech_report.pdf", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "none", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "max/xhigh; no tools", "prompt_style": "default; product fallback enabled", "temperature": "default", "context": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Kimi K3 technical report: GPQA Diamond = 92.6. Source setting: max/xhigh; no tools. Origin: Moonshot evaluation.", "candidates": [ { "score": 94.1, "reference_url": "https://openai.com/index/gpt-5-6/", "source_type": "official_blog", "reported_setting": { "mode": "thinking", "effort": "source does not state", "tools": "none", "sampling": "source does not state", "judge": "benchmark-specified", "harness": "official release evaluation", "prompt_style": "source does not state", "temperature": "source does not state", "context": "source does not state", "deployment": "claude-mythos-5", "safeguards": "lifted for Project Glasswing", "notes": "Anthropic documents Mythos 5 as the same underlying model and capabilities as Fable 5, without Fable's safety classifiers." }, "notes": "OpenAI GPT-5.6 release table (table 7): Claude Mythos 5; GPQA Diamond=94.1." } ] }, { "model_id": "claude-fable-5", "benchmark_id": "critpt", "score": 28.6, "reference_url": "https://github.com/MoonshotAI/Kimi-K3/blob/main/k3_tech_report.pdf", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "none", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "max/xhigh; no tools", "prompt_style": "default; product fallback enabled", "temperature": "default", "context": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Kimi K3 technical report: CritPt = 28.6. Source setting: max/xhigh; no tools. Origin: Artificial Analysis cited by Kimi report." }, { "model_id": "claude-fable-5", "benchmark_id": "aa_lcr", "score": 70.0, "reference_url": "https://github.com/MoonshotAI/Kimi-K3/blob/main/k3_tech_report.pdf", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "none", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "max/xhigh; no tools", "prompt_style": "default; product fallback enabled", "temperature": "default", "context": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Kimi K3 technical report: AA-LCR = 70.0. Source setting: max/xhigh; no tools. Origin: Artificial Analysis cited by Kimi report." }, { "model_id": "claude-fable-5", "benchmark_id": "hle", "score": 56.5, "reference_url": "https://www.anthropic.com/claude-opus-5-system-card", "reported_setting": { "effort": "max", "tools": "benchmark-specific" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Figure 8.10.1.B / Claude Fable 5 / max: 2,500 questions; 1M token cap; no tools and no context compaction.", "candidates": [ { "score": 59.0, "reference_url": "https://www.anthropic.com/claude-sonnet-5-system-card", "source_type": "model_card", "reported_setting": { "effort": "max", "tools": "none", "sampling": "pass@1", "harness": "hle", "deployment": "claude-mythos-5" }, "notes": "Anthropic Claude Sonnet 5 System Card Figure 8.9.1.A, page 122: Claude Mythos 5; Humanity’s Last Exam [without tools]=59.0%. Source setting: effort=max; tools=none; sampling=pass@1; harness=hle; deployment=claude-mythos-5." }, { "score": 53.3, "reference_url": "https://github.com/MoonshotAI/Kimi-K3/blob/main/k3_tech_report.pdf", "source_type": "tech_report", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "none", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "no tools; K3 max/temp=1/top-p=.95", "prompt_style": "default; product fallback enabled", "temperature": "default", "context": "default" }, "notes": "Kimi K3 technical report: HLE-Full (no tools) = 53.3. Source setting: no tools; K3 max/temp=1/top-p=.95. Origin: Moonshot evaluation." }, { "score": 50.3, "reference_url": "https://www.anthropic.com/claude-opus-5-system-card", "source_type": "model_card", "reported_setting": { "effort": "low", "tools": "benchmark-specific" }, "notes": "Figure 8.10.1.B / Claude Fable 5 / low: 2,500 questions; 1M token cap; no tools and no context compaction." }, { "score": 53.6, "reference_url": "https://www.anthropic.com/claude-opus-5-system-card", "source_type": "model_card", "reported_setting": { "effort": "medium", "tools": "benchmark-specific" }, "notes": "Figure 8.10.1.B / Claude Fable 5 / medium: 2,500 questions; 1M token cap; no tools and no context compaction." }, { "score": 54.4, "reference_url": "https://www.anthropic.com/claude-opus-5-system-card", "source_type": "model_card", "reported_setting": { "effort": "high", "tools": "benchmark-specific" }, "notes": "Figure 8.10.1.B / Claude Fable 5 / high: 2,500 questions; 1M token cap; no tools and no context compaction." }, { "score": 57.8, "reference_url": "https://www.anthropic.com/claude-opus-5-system-card", "source_type": "model_card", "reported_setting": { "effort": "xhigh", "tools": "benchmark-specific" }, "notes": "Figure 8.10.1.B / Claude Fable 5 / xhigh: 2,500 questions; 1M token cap; no tools and no context compaction." } ] }, { "model_id": "claude-fable-5", "benchmark_id": "hle_tools", "score": 63.9, "reference_url": "https://www.anthropic.com/claude-opus-5-system-card", "reported_setting": { "effort": "max", "tools": "benchmark-specific" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Figure 8.10.1.A / Claude Fable 5 / max: 2,500 questions; 1M token cap; web/programmatic/code tools; no context compaction.", "candidates": [ { "score": 64.5, "reference_url": "https://www.anthropic.com/claude-sonnet-5-system-card", "source_type": "model_card", "reported_setting": { "effort": "max", "tools": "python / benchmark tools", "sampling": "pass@1", "harness": "hle", "deployment": "claude-mythos-5" }, "notes": "Anthropic Claude Sonnet 5 System Card Figure 8.9.1.A, page 122: Claude Mythos 5; Humanity’s Last Exam [with tools]=64.5%. Source setting: effort=max; tools=python / benchmark tools; sampling=pass@1; harness=hle; deployment=claude-mythos-5." }, { "score": 59.8, "reference_url": "https://www.anthropic.com/claude-sonnet-5-system-card", "source_type": "model_card", "reported_setting": { "effort": "low", "tools": "web search, fetch, and code execution", "sampling": "pass@1", "harness": "hle", "deployment": "claude-mythos-5" }, "notes": "Anthropic Claude Sonnet 5 System Card Figure 8.9.1.B, page 123: Claude Mythos 5; Humanity’s Last Exam [low]=59.8%. Source setting: effort=low; tools=web search, fetch, and code execution; sampling=pass@1; harness=hle; deployment=claude-mythos-5." }, { "score": 62.6, "reference_url": "https://www.anthropic.com/claude-sonnet-5-system-card", "source_type": "model_card", "reported_setting": { "effort": "medium", "tools": "web search, fetch, and code execution", "sampling": "pass@1", "harness": "hle", "deployment": "claude-mythos-5" }, "notes": "Anthropic Claude Sonnet 5 System Card Figure 8.9.1.B, page 123: Claude Mythos 5; Humanity’s Last Exam [med]=62.6%. Source setting: effort=medium; tools=web search, fetch, and code execution; sampling=pass@1; harness=hle; deployment=claude-mythos-5." }, { "score": 63.4, "reference_url": "https://www.anthropic.com/claude-sonnet-5-system-card", "source_type": "model_card", "reported_setting": { "effort": "high", "tools": "web search, fetch, and code execution", "sampling": "pass@1", "harness": "hle", "deployment": "claude-mythos-5" }, "notes": "Anthropic Claude Sonnet 5 System Card Figure 8.9.1.B, page 123: Claude Mythos 5; Humanity’s Last Exam [high]=63.4%. Source setting: effort=high; tools=web search, fetch, and code execution; sampling=pass@1; harness=hle; deployment=claude-mythos-5." }, { "score": 64.2, "reference_url": "https://www.anthropic.com/claude-sonnet-5-system-card", "source_type": "model_card", "reported_setting": { "effort": "xhigh", "tools": "web search, fetch, and code execution", "sampling": "pass@1", "harness": "hle", "deployment": "claude-mythos-5" }, "notes": "Anthropic Claude Sonnet 5 System Card Figure 8.9.1.B, page 123: Claude Mythos 5; Humanity’s Last Exam [xhigh]=64.2%. Source setting: effort=xhigh; tools=web search, fetch, and code execution; sampling=pass@1; harness=hle; deployment=claude-mythos-5." }, { "score": 64.7, "reference_url": "https://www.anthropic.com/claude-sonnet-5-system-card", "source_type": "model_card", "reported_setting": { "effort": "max", "tools": "web search, fetch, and code execution", "sampling": "pass@1", "harness": "hle", "deployment": "claude-mythos-5" }, "notes": "Anthropic Claude Sonnet 5 System Card Figure 8.9.1.B, page 123: Claude Mythos 5; Humanity’s Last Exam [max]=64.7%. Source setting: effort=max; tools=web search, fetch, and code execution; sampling=pass@1; harness=hle; deployment=claude-mythos-5." }, { "score": 63.0, "reference_url": "https://github.com/MoonshotAI/Kimi-K3/blob/main/k3_tech_report.pdf", "source_type": "tech_report", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "general tools", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "general tools; pass@1", "prompt_style": "default; product fallback enabled", "temperature": "default", "context": "default" }, "notes": "Kimi K3 technical report: HLE-Full (with tools) = 63.0. Source setting: general tools; pass@1. Origin: Moonshot evaluation." }, { "score": 58.2, "reference_url": "https://www.anthropic.com/claude-opus-5-system-card", "source_type": "model_card", "reported_setting": { "effort": "low", "tools": "benchmark-specific" }, "notes": "Figure 8.10.1.A / Claude Fable 5 / low: 2,500 questions; 1M token cap; web/programmatic/code tools; no context compaction." }, { "score": 61.4, "reference_url": "https://www.anthropic.com/claude-opus-5-system-card", "source_type": "model_card", "reported_setting": { "effort": "medium", "tools": "benchmark-specific" }, "notes": "Figure 8.10.1.A / Claude Fable 5 / medium: 2,500 questions; 1M token cap; web/programmatic/code tools; no context compaction." }, { "score": 61.8, "reference_url": "https://www.anthropic.com/claude-opus-5-system-card", "source_type": "model_card", "reported_setting": { "effort": "high", "tools": "benchmark-specific" }, "notes": "Figure 8.10.1.A / Claude Fable 5 / high: 2,500 questions; 1M token cap; web/programmatic/code tools; no context compaction." }, { "score": 63.1, "reference_url": "https://www.anthropic.com/claude-opus-5-system-card", "source_type": "model_card", "reported_setting": { "effort": "xhigh", "tools": "benchmark-specific" }, "notes": "Figure 8.10.1.A / Claude Fable 5 / xhigh: 2,500 questions; 1M token cap; web/programmatic/code tools; no context compaction." } ] }, { "model_id": "claude-fable-5", "benchmark_id": "deep_swe_v1_1", "score": 70.0, "reference_url": "https://media.x.ai/v1/website/card-4p6-4cd2dc57.pdf", "reported_setting": { "mode": "thinking", "effort": "max, with fallback", "tools": "agentic repository shell/editor", "sampling": "pass@1", "judge": "isolated repository verifier", "harness": "mini-swe-agent run by Datacurve", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "card46 DeepSWE v1.1 Pass@1 (%); source effort=max, with fallback; benchmark protocol matches canonical=True. Unknown source fields remain unknown.", "candidates": [ { "score": 69.7, "reference_url": "https://openai.com/index/gpt-5-6/", "source_type": "official_blog", "reported_setting": { "effort": "max", "eval_variant": null, "code_mode": "false" }, "notes": "OpenAI GPT-5.6 release table (table 1): Claude Fable 5; DeepSWE v1.1=69.7. Exact effort/variant resolved from the rendered chart point." }, { "score": 59.58, "reference_url": "https://openai.com/index/gpt-5-6/", "source_type": "official_blog", "reported_setting": { "effort": "low", "eval_variant": null, "code_mode": "false" }, "notes": "OpenAI GPT-5.6 release figure (DeepSWE v1.1): Claude Fable 5; DeepSWE v1.1=59.58. Figure 1: DeepSWE v1.1. Cost, latency, and output-token coordinates are retained as review metadata, not score-matrix metrics." }, { "score": 65.37, "reference_url": "https://openai.com/index/gpt-5-6/", "source_type": "official_blog", "reported_setting": { "effort": "medium", "eval_variant": null, "code_mode": "false" }, "notes": "OpenAI GPT-5.6 release figure (DeepSWE v1.1): Claude Fable 5; DeepSWE v1.1=65.37. Figure 1: DeepSWE v1.1. Cost, latency, and output-token coordinates are retained as review metadata, not score-matrix metrics." }, { "score": 68.6, "reference_url": "https://openai.com/index/gpt-5-6/", "source_type": "official_blog", "reported_setting": { "effort": "high", "eval_variant": null, "code_mode": "false" }, "notes": "OpenAI GPT-5.6 release figure (DeepSWE v1.1): Claude Fable 5; DeepSWE v1.1=68.6. Figure 1: DeepSWE v1.1. Cost, latency, and output-token coordinates are retained as review metadata, not score-matrix metrics." }, { "score": 69.91, "reference_url": "https://openai.com/index/gpt-5-6/", "source_type": "official_blog", "reported_setting": { "effort": "xhigh", "eval_variant": null, "code_mode": "false" }, "notes": "OpenAI GPT-5.6 release figure (DeepSWE v1.1): Claude Fable 5; DeepSWE v1.1=69.91. Figure 1: DeepSWE v1.1. Cost, latency, and output-token coordinates are retained as review metadata, not score-matrix metrics." }, { "score": 70.0, "reference_url": "https://github.com/MoonshotAI/Kimi-K3/blob/main/k3_tech_report.pdf", "source_type": "tech_report", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "benchmark-specified", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "agentic coding harness per row", "prompt_style": "default; product fallback enabled", "temperature": "default", "context": "default" }, "notes": "Kimi K3 technical report: DeepSWE = 70.0. Source setting: agentic coding harness per row. Origin: Moonshot K3 run; competitors from DeepSWE leaderboard / GLM blog." }, { "score": 59.6, "reference_url": "https://www.anthropic.com/claude-opus-5-system-card", "source_type": "model_card", "reported_setting": { "effort": "low", "tools": "benchmark-specific" }, "notes": "Figure 8.3.A / Claude Fable 5 / low: 113 tasks; five trials." }, { "score": 65.4, "reference_url": "https://www.anthropic.com/claude-opus-5-system-card", "source_type": "model_card", "reported_setting": { "effort": "medium", "tools": "benchmark-specific" }, "notes": "Figure 8.3.A / Claude Fable 5 / medium: 113 tasks; five trials." }, { "score": 68.6, "reference_url": "https://www.anthropic.com/claude-opus-5-system-card", "source_type": "model_card", "reported_setting": { "effort": "high", "tools": "benchmark-specific" }, "notes": "Figure 8.3.A / Claude Fable 5 / high: 113 tasks; five trials." }, { "score": 69.9, "reference_url": "https://www.anthropic.com/claude-opus-5-system-card", "source_type": "model_card", "reported_setting": { "effort": "xhigh", "tools": "benchmark-specific" }, "notes": "Figure 8.3.A / Claude Fable 5 / xhigh: 113 tasks; five trials." }, { "score": 69.7, "reference_url": "https://www.anthropic.com/claude-opus-5-system-card", "source_type": "model_card", "reported_setting": { "effort": "max", "tools": "benchmark-specific" }, "notes": "Figure 8.3.A / Claude Fable 5 / max: 113 tasks; five trials." }, { "score": 70.0, "reference_url": "https://x.ai/news/grok-4-6", "source_type": "official_blog", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "agentic repository shell/editor", "sampling": "pass@1", "judge": "isolated repository verifier", "harness": "mini-swe-agent run by Datacurve", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown" }, "notes": "release46 DeepSWE v1.1 Pass@1 (%); source effort=max; benchmark protocol matches canonical=True. Unknown source fields remain unknown." } ] }, { "model_id": "claude-fable-5", "benchmark_id": "program_bench", "score": 76.8, "reference_url": "https://github.com/MoonshotAI/Kimi-K3/blob/main/k3_tech_report.pdf", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "benchmark-specified", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "agentic coding harness per row", "prompt_style": "default; product fallback enabled", "temperature": "default", "context": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Kimi K3 technical report: ProgramBench = 76.8. Source setting: agentic coding harness per row. Origin: Moonshot K3 run; competitor values from Vals AI." }, { "model_id": "claude-fable-5", "benchmark_id": "terminal_bench_2_1", "score": 84.3, "reference_url": "https://media.x.ai/v1/website/4p5-5184fdf9.pdf", "reported_setting": { "mode": "thinking", "effort": "max, with fallback", "tools": "container terminal", "sampling": "pass@1", "judge": "verified task-success evaluator", "harness": "Grok Build for Grok; source-reported peer harnesses", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "card45 Terminal-Bench 2.1 Task success rate (%); source effort=max, with fallback; benchmark protocol matches canonical=True. Unknown source fields remain unknown.", "candidates": [ { "score": 83.1, "reference_url": "https://openai.com/index/gpt-5-6/", "source_type": "official_blog", "reported_setting": { "effort": "reference", "eval_variant": null, "code_mode": null }, "notes": "OpenAI GPT-5.6 release table (table 1): Claude Fable 5; Terminal-Bench 2.1=83.1. Exact effort/variant resolved from the rendered chart point." }, { "score": 88.0, "reference_url": "https://github.com/MoonshotAI/Kimi-K3/blob/main/k3_tech_report.pdf", "source_type": "tech_report", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "benchmark-specified", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "agentic coding harness per row", "prompt_style": "default; product fallback enabled", "temperature": "default", "context": "default" }, "notes": "Kimi K3 technical report: Terminal-Bench 2.1 = 88.0. Source setting: agentic coding harness per row. Origin: mixed official sources documented by Kimi report." }, { "score": 84.3, "reference_url": "https://x.ai/news/grok-4-5", "source_type": "official_blog", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "container terminal", "sampling": "pass@1", "judge": "verified task-success evaluator", "harness": "Grok Build for Grok; source-reported peer harnesses", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown" }, "notes": "release45 Terminal-Bench 2.1 Task success rate (%); source effort=max; benchmark protocol matches canonical=True. Unknown source fields remain unknown." } ] }, { "model_id": "claude-fable-5", "benchmark_id": "frontier_swe", "score": 89.0, "reference_url": "https://media.x.ai/v1/website/4p5-5184fdf9.pdf", "reported_setting": { "mode": "thinking", "effort": "max, with fallback", "tools": "agentic code execution", "sampling": "mean@5", "judge": "continuous partial-credit task scoring and dominance", "harness": "public FrontierSWE harness; Grok CLI for Grok 4.5", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "card45 FrontierSWE Dominance (%); source effort=max, with fallback; benchmark protocol matches canonical=True. Unknown source fields remain unknown.", "candidates": [ { "score": 86.6, "reference_url": "https://github.com/MoonshotAI/Kimi-K3/blob/main/k3_tech_report.pdf", "source_type": "tech_report", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "benchmark-specified", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "agentic coding harness per row", "prompt_style": "default; product fallback enabled", "temperature": "default", "context": "default" }, "notes": "Kimi K3 technical report: FrontierSWE = 86.6. Source setting: agentic coding harness per row. Origin: Moonshot/OpenAI runs plus official leaderboard." } ] }, { "model_id": "claude-fable-5", "benchmark_id": "swe_marathon_h20_2026_07_09", "score": 35.0, "reference_url": "https://github.com/MoonshotAI/Kimi-K3/blob/main/k3_tech_report.pdf", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "benchmark-specified", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "H20-calibrated pre-final-v1.1 branch (2026-07-09)", "prompt_style": "default; product fallback enabled", "temperature": "default", "context": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Kimi K3 technical report: SWE-Marathon = 35.0. Source setting: H20-calibrated pre-final-v1.1 branch (2026-07-09). Origin: Moonshot evaluation." }, { "model_id": "claude-fable-5", "benchmark_id": "posttrain_bench", "score": 41.4, "reference_url": "https://github.com/MoonshotAI/Kimi-K3/blob/main/k3_tech_report.pdf", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "benchmark-specified", "sampling": "avg-of-3", "judge": "benchmark-specified", "harness": "H20 avg@3 for K3/Fable/Sol; official H100 for Opus/GPT-5.5/GLM", "prompt_style": "default; product fallback enabled", "temperature": "default", "context": "default" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Kimi K3 technical report: PostTrainBench = 41.4. Source setting: H20 avg@3 for K3/Fable/Sol; official H100 for Opus/GPT-5.5/GLM. Origin: Moonshot H20 runs plus official H100 leaderboard values." }, { "model_id": "claude-fable-5", "benchmark_id": "mls_bench_lite", "score": 49.9, "reference_url": "https://github.com/MoonshotAI/Kimi-K3/blob/main/k3_tech_report.pdf", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "benchmark-specified", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "agentic coding harness per row", "prompt_style": "default; product fallback enabled", "temperature": "default", "context": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Kimi K3 technical report: MLS-Bench-Lite = 49.9. Source setting: agentic coding harness per row. Origin: Moonshot evaluation." }, { "model_id": "claude-fable-5", "benchmark_id": "scicode", "score": 60.2, "reference_url": "https://github.com/MoonshotAI/Kimi-K3/blob/main/k3_tech_report.pdf", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "benchmark-specified", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "agentic coding harness per row", "prompt_style": "default; product fallback enabled", "temperature": "default", "context": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Kimi K3 technical report: SciCode = 60.2. Source setting: agentic coding harness per row. Origin: Artificial Analysis cited by Kimi report." }, { "model_id": "claude-fable-5", "benchmark_id": "kimi_code_bench_v2", "score": 76.9, "reference_url": "https://github.com/MoonshotAI/Kimi-K3/blob/main/k3_tech_report.pdf", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "benchmark-specified", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "agentic coding harness per row", "prompt_style": "default; product fallback enabled", "temperature": "default", "context": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Kimi K3 technical report: Kimi Code Bench 2.0 = 76.9. Source setting: agentic coding harness per row. Origin: Moonshot evaluation." }, { "model_id": "claude-fable-5", "benchmark_id": "browsecomp", "score": 88.0, "reference_url": "https://github.com/MoonshotAI/Kimi-K3/blob/main/k3_tech_report.pdf", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "benchmark-specified", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "agentic benchmark harness", "prompt_style": "default; product fallback enabled", "temperature": "default", "context": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Kimi K3 technical report: BrowseComp = 88.0. Source setting: agentic benchmark harness. Origin: Moonshot K3 run; competitor provider sources.", "candidates": [ { "score": 86.3, "reference_url": "https://www.anthropic.com/claude-sonnet-5-system-card", "source_type": "model_card", "reported_setting": { "effort": "max", "tools": "web search, fetch, programmatic tools, code execution", "sampling": "pass@1", "harness": "browse", "deployment": "claude-mythos-5", "token_cap": "1M" }, "notes": "Anthropic Claude Sonnet 5 System Card Figure 8.9.2.A, page 124: Claude Mythos 5; BrowseComp [1M]=86.3%. Source setting: effort=max; tools=web search, fetch, programmatic tools, code execution; sampling=pass@1; harness=browse; deployment=claude-mythos-5; token_cap=1M." }, { "score": 86.5, "reference_url": "https://www.anthropic.com/claude-sonnet-5-system-card", "source_type": "model_card", "reported_setting": { "effort": "max", "tools": "web search, fetch, programmatic tools, code execution", "sampling": "pass@1", "harness": "browse", "deployment": "claude-mythos-5", "token_cap": "3M" }, "notes": "Anthropic Claude Sonnet 5 System Card Figure 8.9.2.A, page 124: Claude Mythos 5; BrowseComp [3M]=86.5%. Source setting: effort=max; tools=web search, fetch, programmatic tools, code execution; sampling=pass@1; harness=browse; deployment=claude-mythos-5; token_cap=3M." }, { "score": 88.0, "reference_url": "https://www.anthropic.com/claude-sonnet-5-system-card", "source_type": "model_card", "reported_setting": { "effort": "max", "tools": "web search, fetch, programmatic tools, code execution", "sampling": "pass@1", "harness": "browse", "deployment": "claude-mythos-5", "token_cap": "10M" }, "notes": "Anthropic Claude Sonnet 5 System Card Figure 8.9.2.A, page 124: Claude Mythos 5; BrowseComp [10M]=88.0%. Source setting: effort=max; tools=web search, fetch, programmatic tools, code execution; sampling=pass@1; harness=browse; deployment=claude-mythos-5; token_cap=10M." }, { "score": 87.4, "reference_url": "https://www.anthropic.com/claude-opus-5-system-card", "source_type": "model_card", "reported_setting": { "effort": "max", "harness": "official or cited leaderboard" }, "notes": "Table 8.1.A / BrowseComp: Unless noted, Anthropic values use adaptive thinking at max effort, default sampling, mean over five trials. Table 8.1.A card summary." }, { "score": 83.0, "reference_url": "https://www.anthropic.com/claude-opus-5-system-card", "source_type": "model_card", "reported_setting": { "effort": "low", "tools": "benchmark-specific" }, "notes": "Figure 8.10.2.B / Claude Fable 5 / low: Fixed 10M token budget with web/programmatic/code tools and context compaction." }, { "score": 87.1, "reference_url": "https://www.anthropic.com/claude-opus-5-system-card", "source_type": "model_card", "reported_setting": { "effort": "medium", "tools": "benchmark-specific" }, "notes": "Figure 8.10.2.B / Claude Fable 5 / medium: Fixed 10M token budget with web/programmatic/code tools and context compaction." }, { "score": 87.3, "reference_url": "https://www.anthropic.com/claude-opus-5-system-card", "source_type": "model_card", "reported_setting": { "effort": "high", "tools": "benchmark-specific" }, "notes": "Figure 8.10.2.B / Claude Fable 5 / high: Fixed 10M token budget with web/programmatic/code tools and context compaction." }, { "score": 86.3, "reference_url": "https://www.anthropic.com/claude-opus-5-system-card", "source_type": "model_card", "reported_setting": { "effort": "max", "tools": "web/programmatic/code", "harness": "token-budget scaling", "deployment": "claude-mythos-5" }, "notes": "Figure 8.10.2.A / Claude Mythos 5 / 1M: Token-budget scaling; source reports exact labels at 1M, 3M, and 10M. Budget variants are candidates; fixed 10M effort ladder supplies the preferred canonical-setting value." }, { "score": 86.5, "reference_url": "https://www.anthropic.com/claude-opus-5-system-card", "source_type": "model_card", "reported_setting": { "effort": "max", "tools": "web/programmatic/code", "harness": "token-budget scaling", "deployment": "claude-mythos-5" }, "notes": "Figure 8.10.2.A / Claude Mythos 5 / 3M: Token-budget scaling; source reports exact labels at 1M, 3M, and 10M. Budget variants are candidates; fixed 10M effort ladder supplies the preferred canonical-setting value." } ] }, { "model_id": "claude-fable-5", "benchmark_id": "deepsearchqa_f1", "score": 94.7, "reference_url": "https://www.anthropic.com/claude-opus-5-system-card", "reported_setting": { "effort": "max", "tools": "benchmark-specific" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Figure 8.10.3.A / Claude Fable 5 / max: Mean F1 over 900 prompts in 17 fields; web tools; nominal 1M token budget.", "candidates": [ { "score": 94.2, "reference_url": "https://github.com/MoonshotAI/Kimi-K3/blob/main/k3_tech_report.pdf", "source_type": "tech_report", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "benchmark-specified", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "agentic benchmark harness", "prompt_style": "default; product fallback enabled", "temperature": "default", "context": "default" }, "notes": "Kimi K3 technical report: DeepSearchQA (F1) = 94.2. Source setting: agentic benchmark harness. Origin: Moonshot evaluation." }, { "score": 92.5, "reference_url": "https://www.anthropic.com/claude-opus-5-system-card", "source_type": "model_card", "reported_setting": { "effort": "low", "tools": "benchmark-specific" }, "notes": "Figure 8.10.3.A / Claude Fable 5 / low: Mean F1 over 900 prompts in 17 fields; web tools; nominal 1M token budget." }, { "score": 93.4, "reference_url": "https://www.anthropic.com/claude-opus-5-system-card", "source_type": "model_card", "reported_setting": { "effort": "medium", "tools": "benchmark-specific" }, "notes": "Figure 8.10.3.A / Claude Fable 5 / medium: Mean F1 over 900 prompts in 17 fields; web tools; nominal 1M token budget." }, { "score": 94.4, "reference_url": "https://www.anthropic.com/claude-opus-5-system-card", "source_type": "model_card", "reported_setting": { "effort": "xhigh", "tools": "benchmark-specific" }, "notes": "Figure 8.10.3.A / Claude Fable 5 / xhigh: Mean F1 over 900 prompts in 17 fields; web tools; nominal 1M token budget." } ] }, { "model_id": "claude-fable-5", "benchmark_id": "gdpval_aa_elo", "score": 1747.0, "reference_url": "https://github.com/MoonshotAI/Kimi-K3/blob/main/k3_tech_report.pdf", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "benchmark-specified", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "agentic benchmark harness", "prompt_style": "default; product fallback enabled", "temperature": "default", "context": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Kimi K3 technical report: GDPval-AA v2 (Elo) = 1747.0. Source setting: agentic benchmark harness. Origin: Artificial Analysis cited by Kimi report.", "candidates": [ { "score": 1759.6, "reference_url": "https://openai.com/index/gpt-5-6/", "source_type": "official_blog", "reported_setting": { "effort": "max", "eval_variant": null, "code_mode": "false" }, "notes": "OpenAI GPT-5.6 release table (table 0): Claude Fable 5; GDPval-AA v2=1759.6. OpenAI row explicitly identifies GDPval-AA v2. Exact effort/variant resolved from the rendered chart point." }, { "score": 1783.0, "reference_url": "https://www.anthropic.com/claude-sonnet-5-system-card", "source_type": "model_card", "reported_setting": { "effort": "source does not state", "tools": "none", "sampling": "pass@1", "harness": "gdpval" }, "notes": "Anthropic Claude Sonnet 5 System Card Section 8.11.4, page 134: Claude Fable 5; GDPval-AA v2 [summary]=1783. Source setting: effort=source does not state; tools=none; sampling=pass@1; harness=gdpval." } ] }, { "model_id": "claude-fable-5", "benchmark_id": "toolathlon_verified", "score": 77.9, "reference_url": "https://github.com/MoonshotAI/Kimi-K3/blob/main/k3_tech_report.pdf", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "benchmark-specified", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "Verified edition; official leaderboard snapshot", "prompt_style": "default; product fallback enabled", "temperature": "default", "context": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Kimi K3 technical report: Toolathlon-Verified = 77.9. Source setting: Verified edition; official leaderboard snapshot. Origin: official Toolathlon leaderboard cited by Kimi report.", "candidates": [ { "score": 79.3, "reference_url": "https://www.anthropic.com/claude-opus-5-system-card", "source_type": "model_card", "reported_setting": { "effort": "max", "tools": "600+ tools across 32 applications", "deployment": "claude-mythos-5" }, "notes": "Table 8.13.6.A / Claude Mythos 5 / Pass@1: 108 tasks; three trials; internal table excludes null attempts. Opus 4.8 and Sonnet 5 internal values are candidates because official published leaderboard values include null attempts." } ] }, { "model_id": "claude-fable-5", "benchmark_id": "mcpmark_verified", "score": 87.4, "reference_url": "https://github.com/MoonshotAI/Kimi-K3/blob/main/k3_tech_report.pdf", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "benchmark-specified", "sampling": "not restated", "judge": "benchmark-specified", "harness": "verified tasks; sampling not restated in K3 report", "prompt_style": "default; product fallback enabled", "temperature": "default", "context": "default" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Kimi K3 technical report: MCPMark-Verified = 87.4. Source setting: verified tasks; sampling not restated in K3 report. Origin: Moonshot evaluation. Canonical compatibility remains under review; stored as best available primary with matches_canonical=false." }, { "model_id": "claude-fable-5", "benchmark_id": "mcpatlas", "score": 84.7, "reference_url": "https://github.com/MoonshotAI/Kimi-K3/blob/main/k3_tech_report.pdf", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "benchmark-specified", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "public 500; maxTurns=100; Gemini 3.1 Pro judge", "prompt_style": "default; product fallback enabled", "temperature": "default", "context": "default" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Kimi K3 technical report: MCP-Atlas = 84.7. Source setting: public 500; maxTurns=100; Gemini 3.1 Pro judge. Origin: Moonshot evaluation." }, { "model_id": "claude-fable-5", "benchmark_id": "automation_bench", "score": 29.1, "reference_url": "https://github.com/MoonshotAI/Kimi-K3/blob/main/k3_tech_report.pdf", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "benchmark-specified", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "agentic benchmark harness", "prompt_style": "default; product fallback enabled", "temperature": "default", "context": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Kimi K3 technical report: AutomationBench = 29.1. Source setting: agentic benchmark harness. Origin: Moonshot evaluation.", "candidates": [ { "score": 17.4, "reference_url": "https://openai.com/index/gpt-5-6/", "source_type": "official_blog", "reported_setting": { "effort": "max", "eval_variant": null, "code_mode": "false" }, "notes": "OpenAI GPT-5.6 release table (table 8): Claude Fable 5; AutomationBench=17.4. Exact effort/variant resolved from the rendered chart point." } ] }, { "model_id": "claude-fable-5", "benchmark_id": "job_bench", "score": 57.4, "reference_url": "https://github.com/MoonshotAI/Kimi-K3/blob/main/k3_tech_report.pdf", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "benchmark-specified", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "agentic benchmark harness", "prompt_style": "default; product fallback enabled", "temperature": "default", "context": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Kimi K3 technical report: JobBench = 57.4. Source setting: agentic benchmark harness. Origin: official JobBench leaderboard cited by Kimi report." }, { "model_id": "claude-fable-5", "benchmark_id": "aa_briefcase_elo", "score": 1574, "reference_url": "https://media.x.ai/v1/website/card-4p6-4cd2dc57.pdf", "reported_setting": { "mode": "thinking", "effort": "max, with fallback", "tools": "offline multi-file knowledge-work tools", "sampling": "unknown", "judge": "rubric plus analytical and presentation pairwise Elo", "harness": "Artificial Analysis AA-Briefcase harness", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "card46 AA Briefcase AA-Briefcase Elo; source effort=max, with fallback; benchmark protocol matches canonical=True. Unknown source fields remain unknown.", "candidates": [ { "score": 1586.0, "reference_url": "https://www.anthropic.com/claude-sonnet-5-system-card", "source_type": "model_card", "reported_setting": { "effort": "source does not state", "tools": "none", "sampling": "pass@1", "harness": "briefcase" }, "notes": "Anthropic Claude Sonnet 5 System Card Section 8.11.7, page 137: Claude Fable 5; AA-Briefcase [summary]=1586. Source setting: effort=source does not state; tools=none; sampling=pass@1; harness=briefcase." }, { "score": 1583.0, "reference_url": "https://github.com/MoonshotAI/Kimi-K3/blob/main/k3_tech_report.pdf", "source_type": "tech_report", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "benchmark-specified", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "agentic benchmark harness", "prompt_style": "default; product fallback enabled", "temperature": "default", "context": "default" }, "notes": "Kimi K3 technical report: AA-Briefcase (Elo) = 1583.0. Source setting: agentic benchmark harness. Origin: Artificial Analysis cited by Kimi report." }, { "score": 1574, "reference_url": "https://www.anthropic.com/claude-opus-5-system-card", "source_type": "model_card", "reported_setting": { "effort": "max", "harness": "official or cited leaderboard" }, "notes": "Table 8.1.A / AA-Briefcase: Unless noted, Anthropic values use adaptive thinking at max effort, default sampling, mean over five trials. Table 8.1.A card summary." }, { "score": 1574, "reference_url": "https://x.ai/news/grok-4-6", "source_type": "official_blog", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "offline multi-file knowledge-work tools", "sampling": "unknown", "judge": "rubric plus analytical and presentation pairwise Elo", "harness": "Artificial Analysis AA-Briefcase harness", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown" }, "notes": "release46 AA-Briefcase Elo; source effort=max; benchmark protocol matches canonical=True. Unknown source fields remain unknown." } ] }, { "model_id": "claude-fable-5", "benchmark_id": "agents_last_exam", "score": 25.7, "reference_url": "https://github.com/MoonshotAI/Kimi-K3/blob/main/k3_tech_report.pdf", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "benchmark-specified", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "agentic benchmark harness", "prompt_style": "default; product fallback enabled", "temperature": "default", "context": "default" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Kimi K3 technical report: Agents' Last Exam = 25.7. Source setting: agentic benchmark harness. Origin: official leaderboard cited by Kimi report.", "candidates": [ { "score": 40.5, "reference_url": "https://openai.com/index/gpt-5-6/", "source_type": "official_blog", "reported_setting": { "effort": "adaptive", "eval_variant": null, "code_mode": "true" }, "notes": "OpenAI GPT-5.6 release table (table 0): Claude Fable 5; Agents' Last Exam=40.5. Exact effort/variant resolved from the rendered chart point." } ] }, { "model_id": "claude-fable-5", "benchmark_id": "apex_agents", "score": 59.2, "reference_url": "https://media.x.ai/v1/website/card-4p6-4cd2dc57.pdf", "reported_setting": { "mode": "thinking", "effort": "max, with fallback", "tools": "multi-application professional tools", "sampling": "pass@1", "judge": "expert all-criteria-pass rubric", "harness": "Mercor APEX-Agents harness", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "card46 APEX-Agents Pass@1 (%); source effort=max, with fallback; benchmark protocol matches canonical=True. Unknown source fields remain unknown.", "candidates": [ { "score": 43.3, "reference_url": "https://github.com/MoonshotAI/Kimi-K3/blob/main/k3_tech_report.pdf", "source_type": "tech_report", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "benchmark-specified", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "agentic benchmark harness", "prompt_style": "default; product fallback enabled", "temperature": "default", "context": "default" }, "notes": "Kimi K3 technical report: APEX-Agents = 43.3. Source setting: agentic benchmark harness. Origin: Artificial Analysis cited by Kimi report." }, { "score": 59.2, "reference_url": "https://x.ai/news/grok-4-6", "source_type": "official_blog", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "multi-application professional tools", "sampling": "pass@1", "judge": "expert all-criteria-pass rubric", "harness": "Mercor APEX-Agents harness", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown" }, "notes": "release46 APEX-Agents Pass@1 (%); source effort=max; benchmark protocol matches canonical=True. Unknown source fields remain unknown." } ] }, { "model_id": "claude-fable-5", "benchmark_id": "officeqa_pro", "score": 60.9, "reference_url": "https://media.x.ai/v1/website/card-4p6-4cd2dc57.pdf", "reported_setting": { "mode": "thinking", "effort": "max, with fallback", "tools": "document and rendered-image analysis", "sampling": "unknown", "judge": "OfficeQA Pro exact-match accuracy", "harness": "OfficeQA Pro official harness", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "card46 OfficeQA Pro Accuracy (%); source effort=max, with fallback; benchmark protocol matches canonical=True. Unknown source fields remain unknown.", "candidates": [ { "score": 67.1, "reference_url": "https://www.anthropic.com/claude-opus-5-system-card", "source_type": "model_card", "reported_setting": { "effort": "max", "deployment": "claude-mythos-5" }, "notes": "OfficeQA Pro narrative / Claude Mythos 5: 133-question subset; public Messages API for Opus 5." }, { "score": 69.9, "reference_url": "https://github.com/MoonshotAI/Kimi-K3/blob/main/k3_tech_report.pdf", "source_type": "tech_report", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "benchmark-specified", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "PDF corpus rendered as images; no machine-readable text", "prompt_style": "default; product fallback enabled", "temperature": "default", "context": "default" }, "notes": "Kimi K3 technical report: OfficeQA Pro = 69.9. Source setting: PDF corpus rendered as images; no machine-readable text. Origin: Moonshot evaluation." } ] }, { "model_id": "claude-fable-5", "benchmark_id": "spreadsheetbench_2", "score": 34.7, "reference_url": "https://github.com/MoonshotAI/Kimi-K3/blob/main/k3_tech_report.pdf", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "benchmark-specified", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "agentic benchmark harness", "prompt_style": "default; product fallback enabled", "temperature": "default", "context": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Kimi K3 technical report: SpreadsheetBench 2 = 34.7. Source setting: agentic benchmark harness. Origin: Moonshot evaluation." }, { "model_id": "claude-fable-5", "benchmark_id": "osworld_verified", "score": 85.0, "reference_url": "https://www.anthropic.com/claude-sonnet-5-system-card", "reported_setting": { "effort": "max", "tools": "GUI computer-use harness", "sampling": "avg 5 trials unless source states otherwise", "harness": "osworld" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Anthropic Claude Sonnet 5 System Card Figure 8.10.2.A, page 126: Claude Fable 5; OSWorld-Verified [summary]=85.0%. Source setting: effort=max; tools=GUI computer-use harness; sampling=avg 5 trials unless source states otherwise; harness=osworld.", "candidates": [ { "score": 85.0, "reference_url": "https://github.com/MoonshotAI/Kimi-K3/blob/main/k3_tech_report.pdf", "source_type": "tech_report", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "benchmark-specified", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "agentic benchmark harness", "prompt_style": "default; product fallback enabled", "temperature": "default", "context": "default" }, "notes": "Kimi K3 technical report: OSWorld-Verified = 85.0. Source setting: agentic benchmark harness. Origin: Moonshot evaluation." } ] }, { "model_id": "claude-fable-5", "benchmark_id": "osworld_2_0", "score": 66.1, "reference_url": "https://github.com/MoonshotAI/Kimi-K3/blob/main/k3_tech_report.pdf", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "benchmark-specified", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "agentic benchmark harness", "prompt_style": "default; product fallback enabled", "temperature": "default", "context": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Kimi K3 technical report: OSWorld 2.0 = 66.1. Source setting: agentic benchmark harness. Origin: Moonshot evaluation." }, { "model_id": "claude-fable-5", "benchmark_id": "tau3_banking", "score": 26.8, "reference_url": "https://media.x.ai/v1/website/4p5-5184fdf9.pdf", "reported_setting": { "mode": "thinking", "effort": "max, with fallback", "tools": "banking APIs and simulated user", "sampling": "pass@1", "judge": "final-state task accuracy", "harness": "Artificial Analysis tau3-banking harness", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "card45 tau3-banking Accuracy (%); source effort=max, with fallback; benchmark protocol matches canonical=True. Unknown source fields remain unknown.", "candidates": [ { "score": 26.8, "reference_url": "https://github.com/MoonshotAI/Kimi-K3/blob/main/k3_tech_report.pdf", "source_type": "tech_report", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "benchmark-specified", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "agentic benchmark harness", "prompt_style": "default; product fallback enabled", "temperature": "default", "context": "default" }, "notes": "Kimi K3 technical report: τ³-Banking = 26.8. Source setting: agentic benchmark harness. Origin: Artificial Analysis cited by Kimi report." } ] }, { "model_id": "claude-fable-5", "benchmark_id": "harvey_lab_aa", "score": 93.6, "reference_url": "https://github.com/MoonshotAI/Kimi-K3/blob/main/k3_tech_report.pdf", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "benchmark-specified", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "agentic benchmark harness", "prompt_style": "default; product fallback enabled", "temperature": "default", "context": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Kimi K3 technical report: Harvey Lab-AA = 93.6. Source setting: agentic benchmark harness. Origin: Artificial Analysis cited by Kimi report." }, { "model_id": "claude-fable-5", "benchmark_id": "corpfin_v2", "score": 71.8, "reference_url": "https://github.com/MoonshotAI/Kimi-K3/blob/main/k3_tech_report.pdf", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "benchmark-specified", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "agentic benchmark harness", "prompt_style": "default; product fallback enabled", "temperature": "default", "context": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Kimi K3 technical report: CorpFin v2 = 71.8. Source setting: agentic benchmark harness. Origin: Vals AI cited by Kimi report." }, { "model_id": "claude-fable-5", "benchmark_id": "finance_agent_v2", "score": 56.3, "reference_url": "https://github.com/MoonshotAI/Kimi-K3/blob/main/k3_tech_report.pdf", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "benchmark-specified", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "agentic benchmark harness", "prompt_style": "default; product fallback enabled", "temperature": "default", "context": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Kimi K3 technical report: Finance Agent v2 = 56.3. Source setting: agentic benchmark harness. Origin: Vals AI cited by Kimi report." }, { "model_id": "claude-fable-5", "benchmark_id": "legal_research_bench", "score": 49.5, "reference_url": "https://github.com/MoonshotAI/Kimi-K3/blob/main/k3_tech_report.pdf", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "benchmark-specified", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "agentic benchmark harness", "prompt_style": "default; product fallback enabled", "temperature": "default", "context": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Kimi K3 technical report: Legal Research Bench = 49.5. Source setting: agentic benchmark harness. Origin: Vals AI cited by Kimi report." }, { "model_id": "claude-fable-5", "benchmark_id": "worldvqa", "score": 56.7, "reference_url": "https://github.com/MoonshotAI/Kimi-K3/blob/main/k3_tech_report.pdf", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "none", "sampling": "avg-of-3", "judge": "benchmark-specified", "harness": "forced-answer prompt; avg 3 runs", "prompt_style": "default; product fallback enabled", "temperature": "default", "context": "default" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Kimi K3 technical report: WorldVQA ForceAnswer = 56.7. Source setting: forced-answer prompt; avg 3 runs. Origin: Moonshot evaluation." }, { "model_id": "claude-fable-5", "benchmark_id": "omnidocbench", "score": 0.102, "reference_url": "https://github.com/MoonshotAI/Kimi-K3/blob/main/k3_tech_report.pdf", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "benchmark-specified", "sampling": "avg-of-3", "judge": "benchmark-specified", "harness": "avg 3 runs; source 1-NED score normalized to NED", "prompt_style": "default; product fallback enabled", "temperature": "default", "context": "default" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Kimi K3 technical report: OmniDocBench = 0.102 NED. Source setting: avg 3 runs; source 1-NED score normalized to NED. Origin: Moonshot evaluation. The source reports 1-NED=89.8%; stored value is normalized NED=0.102." }, { "model_id": "claude-fable-5", "benchmark_id": "perception_bench", "score": 57.2, "reference_url": "https://github.com/MoonshotAI/Kimi-K3/blob/main/k3_tech_report.pdf", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "benchmark-specified", "sampling": "avg-of-3", "judge": "benchmark-specified", "harness": "multimodal; avg 3 runs", "prompt_style": "default; product fallback enabled", "temperature": "default", "context": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Kimi K3 technical report: PerceptionBench = 57.2. Source setting: multimodal; avg 3 runs. Origin: Moonshot evaluation." }, { "model_id": "claude-fable-5", "benchmark_id": "babyvision", "score": 90.5, "reference_url": "https://github.com/MoonshotAI/Kimi-K3/blob/main/k3_tech_report.pdf", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "python", "sampling": "avg-of-3", "judge": "benchmark-specified", "harness": "Python tool; avg 3 runs", "prompt_style": "default; product fallback enabled", "temperature": "default", "context": "default" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Kimi K3 technical report: BabyVision w/ python = 90.5. Source setting: Python tool; avg 3 runs. Origin: Moonshot evaluation." }, { "model_id": "claude-fable-5", "benchmark_id": "mmmu_pro", "score": 81.2, "reference_url": "https://github.com/MoonshotAI/Kimi-K3/blob/main/k3_tech_report.pdf", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "none", "sampling": "avg-of-3", "judge": "benchmark-specified", "harness": "vision; avg 3 runs", "prompt_style": "default; product fallback enabled", "temperature": "default", "context": "default" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Kimi K3 technical report: MMMU-Pro (no Python) = 81.2. Source setting: vision; avg 3 runs. Origin: Moonshot evaluation.", "candidates": [ { "score": 86.5, "reference_url": "https://github.com/MoonshotAI/Kimi-K3/blob/main/k3_tech_report.pdf", "source_type": "tech_report", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "python", "sampling": "avg-of-3", "judge": "benchmark-specified", "harness": "Python tool; avg 3 runs", "prompt_style": "default; product fallback enabled", "temperature": "default", "context": "default" }, "notes": "Kimi K3 technical report: MMMU-Pro (with Python) = 86.5. Source setting: Python tool; avg 3 runs. Origin: Moonshot evaluation." } ] }, { "model_id": "claude-fable-5", "benchmark_id": "charxiv_reasoning", "score": 88.9, "reference_url": "https://github.com/MoonshotAI/Kimi-K3/blob/main/k3_tech_report.pdf", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "none", "sampling": "avg-of-3", "judge": "benchmark-specified", "harness": "vision; avg 3 runs", "prompt_style": "default; product fallback enabled", "temperature": "default", "context": "default" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Kimi K3 technical report: CharXiv (RQ) (no Python) = 88.9. Source setting: vision; avg 3 runs. Origin: Moonshot evaluation.", "candidates": [ { "score": 93.5, "reference_url": "https://github.com/MoonshotAI/Kimi-K3/blob/main/k3_tech_report.pdf", "source_type": "tech_report", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "python", "sampling": "avg-of-3", "judge": "benchmark-specified", "harness": "Python tool; avg 3 runs", "prompt_style": "default; product fallback enabled", "temperature": "default", "context": "default" }, "notes": "Kimi K3 technical report: CharXiv (RQ) (with Python) = 93.5. Source setting: Python tool; avg 3 runs. Origin: Moonshot evaluation." } ] }, { "model_id": "claude-fable-5", "benchmark_id": "mathvision", "score": 94.8, "reference_url": "https://github.com/MoonshotAI/Kimi-K3/blob/main/k3_tech_report.pdf", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "none", "sampling": "avg-of-3", "judge": "benchmark-specified", "harness": "vision; avg 3 runs", "prompt_style": "default; product fallback enabled", "temperature": "default", "context": "default" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Kimi K3 technical report: MathVision (no Python) = 94.8. Source setting: vision; avg 3 runs. Origin: Moonshot evaluation.", "candidates": [ { "score": 98.6, "reference_url": "https://github.com/MoonshotAI/Kimi-K3/blob/main/k3_tech_report.pdf", "source_type": "tech_report", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "python", "sampling": "avg-of-3", "judge": "benchmark-specified", "harness": "Python tool; avg 3 runs", "prompt_style": "default; product fallback enabled", "temperature": "default", "context": "default" }, "notes": "Kimi K3 technical report: MathVision (with Python) = 98.6. Source setting: Python tool; avg 3 runs. Origin: Moonshot evaluation." } ] }, { "model_id": "claude-fable-5", "benchmark_id": "zerobench_main", "score": 23.0, "reference_url": "https://github.com/MoonshotAI/Kimi-K3/blob/main/k3_tech_report.pdf", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "benchmark-specified", "sampling": "pass@5", "judge": "benchmark-specified", "harness": "no Python; pass@5 / 5 runs", "prompt_style": "default; product fallback enabled", "temperature": "default", "context": "default" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Kimi K3 technical report: ZeroBench (pass@5) (no Python) = 23.0. Source setting: no Python; pass@5 / 5 runs. Origin: Moonshot evaluation. Canonical compatibility remains under review; stored as best available primary with matches_canonical=false." }, { "model_id": "claude-fable-5", "benchmark_id": "zerobench_tools", "score": 46.0, "reference_url": "https://github.com/MoonshotAI/Kimi-K3/blob/main/k3_tech_report.pdf", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "python", "sampling": "pass@5", "judge": "benchmark-specified", "harness": "Python tool; pass@5 / 5 runs", "prompt_style": "default; product fallback enabled", "temperature": "default", "context": "default" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Kimi K3 technical report: ZeroBench (pass@5) (with Python) = 46.0. Source setting: Python tool; pass@5 / 5 runs. Origin: Moonshot evaluation. Canonical compatibility remains under review; stored as best available primary with matches_canonical=false." }, { "model_id": "claude-fable-5", "benchmark_id": "coding_experience", "score": 59.8, "reference_url": "https://github.com/MoonshotAI/Kimi-K3/blob/main/k3_tech_report.pdf", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "benchmark-specified", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "provider-aligned harness: Kimi Code / Claude Code / Codex", "prompt_style": "default; product fallback enabled", "temperature": "default", "context": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Kimi K3 technical report: Coding Experience = 59.8. Source setting: provider-aligned harness: Kimi Code / Claude Code / Codex. Origin: Moonshot evaluation." }, { "model_id": "claude-fable-5", "benchmark_id": "clawbench_2_0", "score": 47.4, "reference_url": "https://github.com/MoonshotAI/Kimi-K3/blob/main/k3_tech_report.pdf", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "benchmark-specified", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "OpenClaw harness", "prompt_style": "default; product fallback enabled", "temperature": "default", "context": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Kimi K3 technical report: 24/7 ClawBench 2.0 = 47.4. Source setting: OpenClaw harness. Origin: Moonshot evaluation." }, { "model_id": "claude-fable-5", "benchmark_id": "mira_bench", "score": 72.9, "reference_url": "https://github.com/MoonshotAI/Kimi-K3/blob/main/k3_tech_report.pdf", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "benchmark-specified", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "MIRA harness", "prompt_style": "default; product fallback enabled", "temperature": "default", "context": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Kimi K3 technical report: MIRA Bench = 72.9. Source setting: MIRA harness. Origin: Moonshot evaluation." }, { "model_id": "claude-fable-5", "benchmark_id": "agentic_vision_bench", "score": 81.1, "reference_url": "https://github.com/MoonshotAI/Kimi-K3/blob/main/k3_tech_report.pdf", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "benchmark-specified", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "Kimi Code harness", "prompt_style": "default; product fallback enabled", "temperature": "default", "context": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Kimi K3 technical report: Agentic Vision Bench = 81.1. Source setting: Kimi Code harness. Origin: Moonshot evaluation." }, { "model_id": "claude-fable-5", "benchmark_id": "online_experience", "score": 74.2, "reference_url": "https://github.com/MoonshotAI/Kimi-K3/blob/main/k3_tech_report.pdf", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "benchmark-specified", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "Kimi Agent harness", "prompt_style": "default; product fallback enabled", "temperature": "default", "context": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Kimi K3 technical report: Online Experience = 74.2. Source setting: Kimi Agent harness. Origin: Moonshot evaluation." }, { "model_id": "claude-fable-5", "benchmark_id": "kwv_bench", "score": 63.6, "reference_url": "https://github.com/MoonshotAI/Kimi-K3/blob/main/k3_tech_report.pdf", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "benchmark-specified", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "N/A harness", "prompt_style": "default; product fallback enabled", "temperature": "default", "context": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Kimi K3 technical report: KWVBench = 63.6. Source setting: N/A harness. Origin: Moonshot evaluation." }, { "model_id": "claude-fable-5", "benchmark_id": "deck_bench", "score": 73.0, "reference_url": "https://github.com/MoonshotAI/Kimi-K3/blob/main/k3_tech_report.pdf", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "benchmark-specified", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "N/A harness", "prompt_style": "default; product fallback enabled", "temperature": "default", "context": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Kimi K3 technical report: DECKBench = 73.0. Source setting: N/A harness. Origin: Moonshot evaluation." }, { "model_id": "claude-fable-5", "benchmark_id": "agent_behavior_bench", "score": 75.5, "reference_url": "https://github.com/MoonshotAI/Kimi-K3/blob/main/k3_tech_report.pdf", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "benchmark-specified", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "Kimi Work harness", "prompt_style": "default; product fallback enabled", "temperature": "default", "context": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Kimi K3 technical report: Agent Behavior Bench = 75.5. Source setting: Kimi Work harness. Origin: Moonshot evaluation." }, { "model_id": "claude-fable-5", "benchmark_id": "chat_all_in_one_bench", "score": 88.0, "reference_url": "https://github.com/MoonshotAI/Kimi-K3/blob/main/k3_tech_report.pdf", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "benchmark-specified", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "Kimi Work harness", "prompt_style": "default; product fallback enabled", "temperature": "default", "context": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Kimi K3 technical report: Chat All-in-One Bench = 88.0. Source setting: Kimi Work harness. Origin: Moonshot evaluation." }, { "model_id": "claude-fable-5", "benchmark_id": "aa_intelligence_index_v4_1", "score": 59.9, "reference_url": "https://github.com/MoonshotAI/Kimi-K3/blob/main/k3_tech_report.pdf", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "benchmark-specified", "sampling": "leaderboard snapshot 2026-07-23", "judge": "benchmark-specified", "harness": "live leaderboard snapshot as of 2026-07-23", "prompt_style": "default; product fallback enabled", "temperature": "default", "context": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Kimi K3 technical report: AA Intelligence Index v4.1 = 59.9. Source setting: live leaderboard snapshot as of 2026-07-23. Origin: third-party leaderboard cited by Kimi report.", "candidates": [ { "score": 62, "reference_url": "https://x.ai/news/grok-4-6", "source_type": "official_blog", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "benchmark-specified", "sampling": "unknown", "judge": "Artificial Analysis weighted composite", "harness": "Artificial Analysis live v4.1 harness", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown" }, "notes": "release46 AA Intelligence Index Composite score; source effort=max; benchmark protocol matches canonical=True. Unknown source fields remain unknown." } ] }, { "model_id": "claude-fable-5", "benchmark_id": "vals_index", "score": 75.1, "reference_url": "https://media.x.ai/v1/website/card-4p6-4cd2dc57.pdf", "reported_setting": { "mode": "thinking", "effort": "max, with fallback", "tools": "benchmark-specified professional tools", "sampling": "unknown", "judge": "Vals weighted suite score", "harness": "Vals Index live harness", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "card46 Vals Index Suite score (%); source effort=max, with fallback; benchmark protocol matches canonical=True. Unknown source fields remain unknown.", "candidates": [ { "score": 75.1, "reference_url": "https://github.com/MoonshotAI/Kimi-K3/blob/main/k3_tech_report.pdf", "source_type": "tech_report", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "benchmark-specified", "sampling": "leaderboard snapshot 2026-07-23", "judge": "benchmark-specified", "harness": "live leaderboard snapshot as of 2026-07-23", "prompt_style": "default; product fallback enabled", "temperature": "default", "context": "default" }, "notes": "Kimi K3 technical report: Vals Index = 75.1. Source setting: live leaderboard snapshot as of 2026-07-23. Origin: third-party leaderboard cited by Kimi report." } ] }, { "model_id": "claude-fable-5", "benchmark_id": "webdev_arena_elo", "score": 1634.0, "reference_url": "https://github.com/MoonshotAI/Kimi-K3/blob/main/k3_tech_report.pdf", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "benchmark-specified", "sampling": "leaderboard snapshot 2026-07-23", "judge": "benchmark-specified", "harness": "live leaderboard snapshot as of 2026-07-23", "prompt_style": "default; product fallback enabled", "temperature": "default", "context": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Kimi K3 technical report: WebDevArena Elo = 1634. Source setting: live leaderboard snapshot as of 2026-07-23. Origin: third-party leaderboard cited by Kimi report." }, { "model_id": "claude-fable-5", "benchmark_id": "text_arena_elo", "score": 1507.0, "reference_url": "https://github.com/MoonshotAI/Kimi-K3/blob/main/k3_tech_report.pdf", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "benchmark-specified", "sampling": "leaderboard snapshot 2026-07-23", "judge": "benchmark-specified", "harness": "live leaderboard snapshot as of 2026-07-23", "prompt_style": "default; product fallback enabled", "temperature": "default", "context": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Kimi K3 technical report: TextArena Elo = 1507. Source setting: live leaderboard snapshot as of 2026-07-23. Origin: third-party leaderboard cited by Kimi report.", "candidates": [ { "score": 1508, "reference_url": "https://arxiv.org/abs/2607.02770", "source_type": "tech_report", "reported_setting": { "mode": "unknown", "effort": "unknown", "tools": "none", "sampling": "live human pairwise votes", "judge": "blind human raters", "harness": "Arena Text", "prompt_style": "live user prompts", "temperature": "unknown", "context": "default", "snapshot_date": "2026-06-19" }, "notes": "Gemma 4 Technical Report Table 4 dated Arena Text Elo snapshot." } ] }, { "model_id": "claude-fable-5", "benchmark_id": "agent_arena", "score": 12.7, "reference_url": "https://github.com/MoonshotAI/Kimi-K3/blob/main/k3_tech_report.pdf", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "benchmark-specified", "sampling": "leaderboard snapshot 2026-07-23", "judge": "benchmark-specified", "harness": "live leaderboard snapshot as of 2026-07-23", "prompt_style": "default; product fallback enabled", "temperature": "default", "context": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Kimi K3 technical report: AgentArena = 12.7. Source setting: live leaderboard snapshot as of 2026-07-23. Origin: third-party leaderboard cited by Kimi report." }, { "model_id": "gpt-5.6-sol", "benchmark_id": "gpqa_diamond", "score": 94.1, "reference_url": "https://github.com/MoonshotAI/Kimi-K3/blob/main/k3_tech_report.pdf", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "none", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "max/xhigh; no tools", "prompt_style": "default", "temperature": "default", "context": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Kimi K3 technical report: GPQA Diamond = 94.1. Source setting: max/xhigh; no tools. Origin: Moonshot evaluation.", "candidates": [ { "score": 94.6, "reference_url": "https://openai.com/index/gpt-5-6/", "source_type": "official_blog", "reported_setting": { "mode": "thinking", "effort": "source does not state", "tools": "none", "sampling": "source does not state", "judge": "benchmark-specified", "harness": "official release evaluation", "prompt_style": "source does not state", "temperature": "source does not state", "context": "source does not state" }, "notes": "OpenAI GPT-5.6 release table (table 7): GPT‑5.6 Sol; GPQA Diamond=94.6." } ] }, { "model_id": "gpt-5.6-sol", "benchmark_id": "critpt", "score": 32.3, "reference_url": "https://github.com/MoonshotAI/Kimi-K3/blob/main/k3_tech_report.pdf", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "none", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "max/xhigh; no tools", "prompt_style": "default", "temperature": "default", "context": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Kimi K3 technical report: CritPt = 32.3. Source setting: max/xhigh; no tools. Origin: Artificial Analysis cited by Kimi report." }, { "model_id": "gpt-5.6-sol", "benchmark_id": "aa_lcr", "score": 73.7, "reference_url": "https://github.com/MoonshotAI/Kimi-K3/blob/main/k3_tech_report.pdf", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "none", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "max/xhigh; no tools", "prompt_style": "default", "temperature": "default", "context": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Kimi K3 technical report: AA-LCR = 73.7. Source setting: max/xhigh; no tools. Origin: Artificial Analysis cited by Kimi report." }, { "model_id": "gpt-5.6-sol", "benchmark_id": "hle", "score": 44.5, "reference_url": "https://github.com/MoonshotAI/Kimi-K3/blob/main/k3_tech_report.pdf", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "none", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "no tools; K3 max/temp=1/top-p=.95", "prompt_style": "default", "temperature": "default", "context": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Kimi K3 technical report: HLE-Full (no tools) = 44.5. Source setting: no tools; K3 max/temp=1/top-p=.95. Origin: Moonshot evaluation." }, { "model_id": "gpt-5.6-sol", "benchmark_id": "hle_tools", "score": 58.0, "reference_url": "https://github.com/MoonshotAI/Kimi-K3/blob/main/k3_tech_report.pdf", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "general tools", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "general tools; pass@1", "prompt_style": "default", "temperature": "default", "context": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Kimi K3 technical report: HLE-Full (with tools) = 58.0. Source setting: general tools; pass@1. Origin: Moonshot evaluation." }, { "model_id": "gpt-5.6-sol", "benchmark_id": "deep_swe_v1_1", "score": 73.0, "reference_url": "https://github.com/MoonshotAI/Kimi-K3/blob/main/k3_tech_report.pdf", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "benchmark-specified", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "agentic coding harness per row", "prompt_style": "default", "temperature": "default", "context": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Kimi K3 technical report: DeepSWE = 73.0. Source setting: agentic coding harness per row. Origin: Moonshot K3 run; competitors from DeepSWE leaderboard / GLM blog.", "candidates": [ { "score": 72.7, "reference_url": "https://openai.com/index/gpt-5-6/", "source_type": "official_blog", "reported_setting": { "effort": "max", "eval_variant": null, "code_mode": "false" }, "notes": "OpenAI GPT-5.6 release table (table 1): GPT‑5.6 Sol; DeepSWE v1.1=72.7. Exact effort/variant resolved from the rendered chart point." }, { "score": 44.47, "reference_url": "https://openai.com/index/gpt-5-6/", "source_type": "official_blog", "reported_setting": { "effort": "none", "eval_variant": null, "code_mode": "false" }, "notes": "OpenAI GPT-5.6 release figure (DeepSWE v1.1): GPT-5.6 Sol; DeepSWE v1.1=44.47. Figure 1: DeepSWE v1.1. Cost, latency, and output-token coordinates are retained as review metadata, not score-matrix metrics." }, { "score": 45.35, "reference_url": "https://openai.com/index/gpt-5-6/", "source_type": "official_blog", "reported_setting": { "effort": "low", "eval_variant": null, "code_mode": "false" }, "notes": "OpenAI GPT-5.6 release figure (DeepSWE v1.1): GPT-5.6 Sol; DeepSWE v1.1=45.35. Figure 1: DeepSWE v1.1. Cost, latency, and output-token coordinates are retained as review metadata, not score-matrix metrics." }, { "score": 61.06, "reference_url": "https://openai.com/index/gpt-5-6/", "source_type": "official_blog", "reported_setting": { "effort": "medium", "eval_variant": null, "code_mode": "false" }, "notes": "OpenAI GPT-5.6 release figure (DeepSWE v1.1): GPT-5.6 Sol; DeepSWE v1.1=61.06. Figure 1: DeepSWE v1.1. Cost, latency, and output-token coordinates are retained as review metadata, not score-matrix metrics." }, { "score": 69.4, "reference_url": "https://openai.com/index/gpt-5-6/", "source_type": "official_blog", "reported_setting": { "effort": "high", "eval_variant": null, "code_mode": "false" }, "notes": "OpenAI GPT-5.6 release figure (DeepSWE v1.1): GPT-5.6 Sol; DeepSWE v1.1=69.4. Figure 1: DeepSWE v1.1. Cost, latency, and output-token coordinates are retained as review metadata, not score-matrix metrics." }, { "score": 70.73, "reference_url": "https://openai.com/index/gpt-5-6/", "source_type": "official_blog", "reported_setting": { "effort": "xhigh", "eval_variant": null, "code_mode": "false" }, "notes": "OpenAI GPT-5.6 release figure (DeepSWE v1.1): GPT-5.6 Sol; DeepSWE v1.1=70.73. Figure 1: DeepSWE v1.1. Cost, latency, and output-token coordinates are retained as review metadata, not score-matrix metrics." }, { "score": 72.7, "reference_url": "https://www.anthropic.com/claude-opus-5-system-card", "source_type": "model_card", "reported_setting": { "effort": "max", "harness": "official or cited leaderboard" }, "notes": "Table 8.1.A / DeepSWE v1.1: Unless noted, Anthropic values use adaptive thinking at max effort, default sampling, mean over five trials. Table 8.1.A card summary." }, { "score": 73.0, "reference_url": "https://x.ai/news/grok-4-6", "source_type": "official_blog", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "agentic repository shell/editor", "sampling": "pass@1", "judge": "isolated repository verifier", "harness": "mini-swe-agent run by Datacurve", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown" }, "notes": "release46 DeepSWE v1.1 Pass@1 (%); source effort=max; benchmark protocol matches canonical=True. Unknown source fields remain unknown." } ] }, { "model_id": "gpt-5.6-sol", "benchmark_id": "program_bench", "score": 77.6, "reference_url": "https://github.com/MoonshotAI/Kimi-K3/blob/main/k3_tech_report.pdf", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "benchmark-specified", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "agentic coding harness per row", "prompt_style": "default", "temperature": "default", "context": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Kimi K3 technical report: ProgramBench = 77.6. Source setting: agentic coding harness per row. Origin: Moonshot K3 run; competitor values from Vals AI." }, { "model_id": "gpt-5.6-sol", "benchmark_id": "terminal_bench_2_1", "score": 88.8, "reference_url": "https://github.com/MoonshotAI/Kimi-K3/blob/main/k3_tech_report.pdf", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "benchmark-specified", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "agentic coding harness per row", "prompt_style": "default", "temperature": "default", "context": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Kimi K3 technical report: Terminal-Bench 2.1 = 88.8. Source setting: agentic coding harness per row. Origin: mixed official sources documented by Kimi report.", "candidates": [ { "score": 91.9, "reference_url": "https://openai.com/index/gpt-5-6/", "source_type": "official_blog", "reported_setting": { "effort": "max", "eval_variant": null, "code_mode": "true", "multi_agent": "4 agents" }, "notes": "OpenAI GPT-5.6 release table (table 1): GPT‑5.6 Sol Ultra; Terminal-Bench 2.1=91.9. Exact effort/variant resolved from the rendered chart point." }, { "score": 76.4, "reference_url": "https://openai.com/index/gpt-5-6/", "source_type": "official_blog", "reported_setting": { "effort": "low", "eval_variant": null, "code_mode": "true" }, "notes": "OpenAI GPT-5.6 release figure (Terminal-Bench 2.1): GPT-5.6 Sol; Terminal-Bench 2.1=76.4. Figure 1: Terminal-Bench 2.1. Cost, latency, and output-token coordinates are retained as review metadata, not score-matrix metrics." }, { "score": 81.8, "reference_url": "https://openai.com/index/gpt-5-6/", "source_type": "official_blog", "reported_setting": { "effort": "medium", "eval_variant": null, "code_mode": "true" }, "notes": "OpenAI GPT-5.6 release figure (Terminal-Bench 2.1): GPT-5.6 Sol; Terminal-Bench 2.1=81.8. Figure 1: Terminal-Bench 2.1. Cost, latency, and output-token coordinates are retained as review metadata, not score-matrix metrics." }, { "score": 84.72, "reference_url": "https://openai.com/index/gpt-5-6/", "source_type": "official_blog", "reported_setting": { "effort": "high", "eval_variant": null, "code_mode": "true" }, "notes": "OpenAI GPT-5.6 release figure (Terminal-Bench 2.1): GPT-5.6 Sol; Terminal-Bench 2.1=84.72. Figure 1: Terminal-Bench 2.1. Cost, latency, and output-token coordinates are retained as review metadata, not score-matrix metrics." }, { "score": 84.94, "reference_url": "https://openai.com/index/gpt-5-6/", "source_type": "official_blog", "reported_setting": { "effort": "xhigh", "eval_variant": null, "code_mode": "true" }, "notes": "OpenAI GPT-5.6 release figure (Terminal-Bench 2.1): GPT-5.6 Sol; Terminal-Bench 2.1=84.94. Figure 1: Terminal-Bench 2.1. Cost, latency, and output-token coordinates are retained as review metadata, not score-matrix metrics." }, { "score": 76.4, "reference_url": "https://openai.com/index/gpt-5-6/", "source_type": "official_blog", "reported_setting": { "effort": "low", "eval_variant": "1 agent", "code_mode": "true" }, "notes": "OpenAI GPT-5.6 release figure (Terminal-Bench 2.1 (Multi-Agent)): GPT-5.6 Sol · 1 agent; Terminal-Bench 2.1=76.4. Figure 2: Score-latency/cost/token frontier for 1 and 4 Sol agents.. Cost, latency, and output-token coordinates are retained as review metadata, not score-matrix metrics." }, { "score": 81.8, "reference_url": "https://openai.com/index/gpt-5-6/", "source_type": "official_blog", "reported_setting": { "effort": "medium", "eval_variant": "1 agent", "code_mode": "true" }, "notes": "OpenAI GPT-5.6 release figure (Terminal-Bench 2.1 (Multi-Agent)): GPT-5.6 Sol · 1 agent; Terminal-Bench 2.1=81.8. Figure 2: Score-latency/cost/token frontier for 1 and 4 Sol agents.. Cost, latency, and output-token coordinates are retained as review metadata, not score-matrix metrics." }, { "score": 84.72, "reference_url": "https://openai.com/index/gpt-5-6/", "source_type": "official_blog", "reported_setting": { "effort": "high", "eval_variant": "1 agent", "code_mode": "true" }, "notes": "OpenAI GPT-5.6 release figure (Terminal-Bench 2.1 (Multi-Agent)): GPT-5.6 Sol · 1 agent; Terminal-Bench 2.1=84.72. Figure 2: Score-latency/cost/token frontier for 1 and 4 Sol agents.. Cost, latency, and output-token coordinates are retained as review metadata, not score-matrix metrics." }, { "score": 84.94, "reference_url": "https://openai.com/index/gpt-5-6/", "source_type": "official_blog", "reported_setting": { "effort": "xhigh", "eval_variant": "1 agent", "code_mode": "true" }, "notes": "OpenAI GPT-5.6 release figure (Terminal-Bench 2.1 (Multi-Agent)): GPT-5.6 Sol · 1 agent; Terminal-Bench 2.1=84.94. Figure 2: Score-latency/cost/token frontier for 1 and 4 Sol agents.. Cost, latency, and output-token coordinates are retained as review metadata, not score-matrix metrics." }, { "score": 79.78, "reference_url": "https://openai.com/index/gpt-5-6/", "source_type": "official_blog", "reported_setting": { "effort": "low", "eval_variant": "4 agents", "code_mode": "true" }, "notes": "OpenAI GPT-5.6 release figure (Terminal-Bench 2.1 (Multi-Agent)): GPT-5.6 Sol · 4 agents; Terminal-Bench 2.1=79.78. Figure 2: Score-latency/cost/token frontier for 1 and 4 Sol agents.. Cost, latency, and output-token coordinates are retained as review metadata, not score-matrix metrics." }, { "score": 85.39, "reference_url": "https://openai.com/index/gpt-5-6/", "source_type": "official_blog", "reported_setting": { "effort": "medium", "eval_variant": "4 agents", "code_mode": "true" }, "notes": "OpenAI GPT-5.6 release figure (Terminal-Bench 2.1 (Multi-Agent)): GPT-5.6 Sol · 4 agents; Terminal-Bench 2.1=85.39. Figure 2: Score-latency/cost/token frontier for 1 and 4 Sol agents.. Cost, latency, and output-token coordinates are retained as review metadata, not score-matrix metrics." }, { "score": 86.74, "reference_url": "https://openai.com/index/gpt-5-6/", "source_type": "official_blog", "reported_setting": { "effort": "high", "eval_variant": "4 agents", "code_mode": "true" }, "notes": "OpenAI GPT-5.6 release figure (Terminal-Bench 2.1 (Multi-Agent)): GPT-5.6 Sol · 4 agents; Terminal-Bench 2.1=86.74. Figure 2: Score-latency/cost/token frontier for 1 and 4 Sol agents.. Cost, latency, and output-token coordinates are retained as review metadata, not score-matrix metrics." }, { "score": 89.21, "reference_url": "https://openai.com/index/gpt-5-6/", "source_type": "official_blog", "reported_setting": { "effort": "xhigh", "eval_variant": "4 agents", "code_mode": "true" }, "notes": "OpenAI GPT-5.6 release figure (Terminal-Bench 2.1 (Multi-Agent)): GPT-5.6 Sol · 4 agents; Terminal-Bench 2.1=89.21. Figure 2: Score-latency/cost/token frontier for 1 and 4 Sol agents.. Cost, latency, and output-token coordinates are retained as review metadata, not score-matrix metrics." } ] }, { "model_id": "gpt-5.6-sol", "benchmark_id": "frontier_swe", "score": 71.3, "reference_url": "https://github.com/MoonshotAI/Kimi-K3/blob/main/k3_tech_report.pdf", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "benchmark-specified", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "agentic coding harness per row", "prompt_style": "default", "temperature": "default", "context": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Kimi K3 technical report: FrontierSWE = 71.3. Source setting: agentic coding harness per row. Origin: Moonshot/OpenAI runs plus official leaderboard." }, { "model_id": "gpt-5.6-sol", "benchmark_id": "swe_marathon_h20_2026_07_09", "score": 39.0, "reference_url": "https://github.com/MoonshotAI/Kimi-K3/blob/main/k3_tech_report.pdf", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "benchmark-specified", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "H20-calibrated pre-final-v1.1 branch (2026-07-09)", "prompt_style": "default", "temperature": "default", "context": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Kimi K3 technical report: SWE-Marathon = 39.0. Source setting: H20-calibrated pre-final-v1.1 branch (2026-07-09). Origin: Moonshot evaluation." }, { "model_id": "gpt-5.6-sol", "benchmark_id": "posttrain_bench", "score": 34.6, "reference_url": "https://github.com/MoonshotAI/Kimi-K3/blob/main/k3_tech_report.pdf", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "benchmark-specified", "sampling": "avg-of-3", "judge": "benchmark-specified", "harness": "H20 avg@3 for K3/Fable/Sol; official H100 for Opus/GPT-5.5/GLM", "prompt_style": "default", "temperature": "default", "context": "default" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Kimi K3 technical report: PostTrainBench = 34.6. Source setting: H20 avg@3 for K3/Fable/Sol; official H100 for Opus/GPT-5.5/GLM. Origin: Moonshot H20 runs plus official H100 leaderboard values." }, { "model_id": "gpt-5.6-sol", "benchmark_id": "mls_bench_lite", "score": 46.2, "reference_url": "https://github.com/MoonshotAI/Kimi-K3/blob/main/k3_tech_report.pdf", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "benchmark-specified", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "agentic coding harness per row", "prompt_style": "default", "temperature": "default", "context": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Kimi K3 technical report: MLS-Bench-Lite = 46.2. Source setting: agentic coding harness per row. Origin: Moonshot evaluation." }, { "model_id": "gpt-5.6-sol", "benchmark_id": "scicode", "score": 56.1, "reference_url": "https://github.com/MoonshotAI/Kimi-K3/blob/main/k3_tech_report.pdf", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "benchmark-specified", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "agentic coding harness per row", "prompt_style": "default", "temperature": "default", "context": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Kimi K3 technical report: SciCode = 56.1. Source setting: agentic coding harness per row. Origin: Artificial Analysis cited by Kimi report." }, { "model_id": "gpt-5.6-sol", "benchmark_id": "kimi_code_bench_v2", "score": 64.8, "reference_url": "https://github.com/MoonshotAI/Kimi-K3/blob/main/k3_tech_report.pdf", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "benchmark-specified", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "agentic coding harness per row", "prompt_style": "default", "temperature": "default", "context": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Kimi K3 technical report: Kimi Code Bench 2.0 = 64.8. Source setting: agentic coding harness per row. Origin: Moonshot evaluation." }, { "model_id": "gpt-5.6-sol", "benchmark_id": "browsecomp", "score": 90.4, "reference_url": "https://github.com/MoonshotAI/Kimi-K3/blob/main/k3_tech_report.pdf", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "benchmark-specified", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "agentic benchmark harness", "prompt_style": "default", "temperature": "default", "context": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Kimi K3 technical report: BrowseComp = 90.4. Source setting: agentic benchmark harness. Origin: Moonshot K3 run; competitor provider sources.", "candidates": [ { "score": 92.2, "reference_url": "https://openai.com/index/gpt-5-6/", "source_type": "official_blog", "reported_setting": { "effort": "max", "eval_variant": "4 agents", "code_mode": "true" }, "notes": "OpenAI GPT-5.6 release table (table 3): GPT‑5.6 Sol Ultra; BrowseComp=92.2. Exact effort/variant resolved from the rendered chart point." }, { "score": 69.04, "reference_url": "https://openai.com/index/gpt-5-6/", "source_type": "official_blog", "reported_setting": { "effort": "low", "eval_variant": "1 agent", "code_mode": "true" }, "notes": "OpenAI GPT-5.6 release figure (BrowseComp (Multi-Agent)): GPT-5.6 Sol · 1 agent; BrowseComp=69.04. Figure 2: Score-latency/cost/token frontier for 1, 4, and 16 Sol agents.. Cost, latency, and output-token coordinates are retained as review metadata, not score-matrix metrics." }, { "score": 84.6, "reference_url": "https://openai.com/index/gpt-5-6/", "source_type": "official_blog", "reported_setting": { "effort": "medium", "eval_variant": "1 agent", "code_mode": "true" }, "notes": "OpenAI GPT-5.6 release figure (BrowseComp (Multi-Agent)): GPT-5.6 Sol · 1 agent; BrowseComp=84.6. Figure 2: Score-latency/cost/token frontier for 1, 4, and 16 Sol agents.. Cost, latency, and output-token coordinates are retained as review metadata, not score-matrix metrics." }, { "score": 88.23, "reference_url": "https://openai.com/index/gpt-5-6/", "source_type": "official_blog", "reported_setting": { "effort": "high", "eval_variant": "1 agent", "code_mode": "true" }, "notes": "OpenAI GPT-5.6 release figure (BrowseComp (Multi-Agent)): GPT-5.6 Sol · 1 agent; BrowseComp=88.23. Figure 2: Score-latency/cost/token frontier for 1, 4, and 16 Sol agents.. Cost, latency, and output-token coordinates are retained as review metadata, not score-matrix metrics." }, { "score": 89.42, "reference_url": "https://openai.com/index/gpt-5-6/", "source_type": "official_blog", "reported_setting": { "effort": "xhigh", "eval_variant": "1 agent", "code_mode": "true" }, "notes": "OpenAI GPT-5.6 release figure (BrowseComp (Multi-Agent)): GPT-5.6 Sol · 1 agent; BrowseComp=89.42. Figure 2: Score-latency/cost/token frontier for 1, 4, and 16 Sol agents.. Cost, latency, and output-token coordinates are retained as review metadata, not score-matrix metrics." }, { "score": 90.84, "reference_url": "https://openai.com/index/gpt-5-6/", "source_type": "official_blog", "reported_setting": { "effort": "max", "eval_variant": "1 agent", "code_mode": "true" }, "notes": "OpenAI GPT-5.6 release figure (BrowseComp (Multi-Agent)): GPT-5.6 Sol · 1 agent; BrowseComp=90.84. Figure 2: Score-latency/cost/token frontier for 1, 4, and 16 Sol agents.. Cost, latency, and output-token coordinates are retained as review metadata, not score-matrix metrics." }, { "score": 81.75, "reference_url": "https://openai.com/index/gpt-5-6/", "source_type": "official_blog", "reported_setting": { "effort": "low", "eval_variant": "4 agents", "code_mode": "true" }, "notes": "OpenAI GPT-5.6 release figure (BrowseComp (Multi-Agent)): GPT-5.6 Sol · 4 agents; BrowseComp=81.75. Figure 2: Score-latency/cost/token frontier for 1, 4, and 16 Sol agents.. Cost, latency, and output-token coordinates are retained as review metadata, not score-matrix metrics." }, { "score": 89.18, "reference_url": "https://openai.com/index/gpt-5-6/", "source_type": "official_blog", "reported_setting": { "effort": "medium", "eval_variant": "4 agents", "code_mode": "true" }, "notes": "OpenAI GPT-5.6 release figure (BrowseComp (Multi-Agent)): GPT-5.6 Sol · 4 agents; BrowseComp=89.18. Figure 2: Score-latency/cost/token frontier for 1, 4, and 16 Sol agents.. Cost, latency, and output-token coordinates are retained as review metadata, not score-matrix metrics." }, { "score": 90.76, "reference_url": "https://openai.com/index/gpt-5-6/", "source_type": "official_blog", "reported_setting": { "effort": "high", "eval_variant": "4 agents", "code_mode": "true" }, "notes": "OpenAI GPT-5.6 release figure (BrowseComp (Multi-Agent)): GPT-5.6 Sol · 4 agents; BrowseComp=90.76. Figure 2: Score-latency/cost/token frontier for 1, 4, and 16 Sol agents.. Cost, latency, and output-token coordinates are retained as review metadata, not score-matrix metrics." }, { "score": 91.15, "reference_url": "https://openai.com/index/gpt-5-6/", "source_type": "official_blog", "reported_setting": { "effort": "xhigh", "eval_variant": "4 agents", "code_mode": "true" }, "notes": "OpenAI GPT-5.6 release figure (BrowseComp (Multi-Agent)): GPT-5.6 Sol · 4 agents; BrowseComp=91.15. Figure 2: Score-latency/cost/token frontier for 1, 4, and 16 Sol agents.. Cost, latency, and output-token coordinates are retained as review metadata, not score-matrix metrics." }, { "score": 86.41, "reference_url": "https://openai.com/index/gpt-5-6/", "source_type": "official_blog", "reported_setting": { "effort": "low", "eval_variant": "16 agents", "code_mode": "true" }, "notes": "OpenAI GPT-5.6 release figure (BrowseComp (Multi-Agent)): GPT-5.6 Sol · 16 agents; BrowseComp=86.41. Figure 2: Score-latency/cost/token frontier for 1, 4, and 16 Sol agents.. Cost, latency, and output-token coordinates are retained as review metadata, not score-matrix metrics." }, { "score": 91.07, "reference_url": "https://openai.com/index/gpt-5-6/", "source_type": "official_blog", "reported_setting": { "effort": "medium", "eval_variant": "16 agents", "code_mode": "true" }, "notes": "OpenAI GPT-5.6 release figure (BrowseComp (Multi-Agent)): GPT-5.6 Sol · 16 agents; BrowseComp=91.07. Figure 2: Score-latency/cost/token frontier for 1, 4, and 16 Sol agents.. Cost, latency, and output-token coordinates are retained as review metadata, not score-matrix metrics." }, { "score": 92.18, "reference_url": "https://openai.com/index/gpt-5-6/", "source_type": "official_blog", "reported_setting": { "effort": "high", "eval_variant": "16 agents", "code_mode": "true" }, "notes": "OpenAI GPT-5.6 release figure (BrowseComp (Multi-Agent)): GPT-5.6 Sol · 16 agents; BrowseComp=92.18. Figure 2: Score-latency/cost/token frontier for 1, 4, and 16 Sol agents.. Cost, latency, and output-token coordinates are retained as review metadata, not score-matrix metrics." }, { "score": 93.36, "reference_url": "https://openai.com/index/gpt-5-6/", "source_type": "official_blog", "reported_setting": { "effort": "xhigh", "eval_variant": "16 agents", "code_mode": "true" }, "notes": "OpenAI GPT-5.6 release figure (BrowseComp (Multi-Agent)): GPT-5.6 Sol · 16 agents; BrowseComp=93.36. Figure 2: Score-latency/cost/token frontier for 1, 4, and 16 Sol agents.. Cost, latency, and output-token coordinates are retained as review metadata, not score-matrix metrics." }, { "score": 93.29, "reference_url": "https://openai.com/index/gpt-5-6/", "source_type": "official_blog", "reported_setting": { "effort": "max", "eval_variant": "16 agents", "code_mode": "true" }, "notes": "OpenAI GPT-5.6 release figure (BrowseComp (Multi-Agent)): GPT-5.6 Sol · 16 agents; BrowseComp=93.29. Figure 2: Score-latency/cost/token frontier for 1, 4, and 16 Sol agents.. Cost, latency, and output-token coordinates are retained as review metadata, not score-matrix metrics." }, { "score": 69.67, "reference_url": "https://openai.com/index/gpt-5-6/", "source_type": "official_blog", "reported_setting": { "effort": "none", "eval_variant": null, "code_mode": "false" }, "notes": "OpenAI GPT-5.6 release figure (BrowseComp): GPT-5.6 Sol; BrowseComp=69.67. Figure 15: Agentic browsing tasks.. Cost, latency, and output-token coordinates are retained as review metadata, not score-matrix metrics." }, { "score": 65.32, "reference_url": "https://openai.com/index/gpt-5-6/", "source_type": "official_blog", "reported_setting": { "effort": "low", "eval_variant": null, "code_mode": "false" }, "notes": "OpenAI GPT-5.6 release figure (BrowseComp): GPT-5.6 Sol; BrowseComp=65.32. Figure 15: Agentic browsing tasks.. Cost, latency, and output-token coordinates are retained as review metadata, not score-matrix metrics." }, { "score": 83.41, "reference_url": "https://openai.com/index/gpt-5-6/", "source_type": "official_blog", "reported_setting": { "effort": "medium", "eval_variant": null, "code_mode": "false" }, "notes": "OpenAI GPT-5.6 release figure (BrowseComp): GPT-5.6 Sol; BrowseComp=83.41. Figure 15: Agentic browsing tasks.. Cost, latency, and output-token coordinates are retained as review metadata, not score-matrix metrics." }, { "score": 87.52, "reference_url": "https://openai.com/index/gpt-5-6/", "source_type": "official_blog", "reported_setting": { "effort": "high", "eval_variant": null, "code_mode": "false" }, "notes": "OpenAI GPT-5.6 release figure (BrowseComp): GPT-5.6 Sol; BrowseComp=87.52. Figure 15: Agentic browsing tasks.. Cost, latency, and output-token coordinates are retained as review metadata, not score-matrix metrics." }, { "score": 88.78, "reference_url": "https://openai.com/index/gpt-5-6/", "source_type": "official_blog", "reported_setting": { "effort": "xhigh", "eval_variant": null, "code_mode": "false" }, "notes": "OpenAI GPT-5.6 release figure (BrowseComp): GPT-5.6 Sol; BrowseComp=88.78. Figure 15: Agentic browsing tasks.. Cost, latency, and output-token coordinates are retained as review metadata, not score-matrix metrics." } ] }, { "model_id": "gpt-5.6-sol", "benchmark_id": "researchrubrics", "score": 73.8, "reference_url": "https://github.com/MoonshotAI/Kimi-K3/blob/main/k3_tech_report.pdf", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "benchmark-specified", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "agentic benchmark harness", "prompt_style": "default", "temperature": "default", "context": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Kimi K3 technical report: ResearchRubrics = 73.8. Source setting: agentic benchmark harness. Origin: Moonshot evaluation." }, { "model_id": "gpt-5.6-sol", "benchmark_id": "gdpval_aa_elo", "score": 1736.0, "reference_url": "https://github.com/MoonshotAI/Kimi-K3/blob/main/k3_tech_report.pdf", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "benchmark-specified", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "agentic benchmark harness", "prompt_style": "default", "temperature": "default", "context": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Kimi K3 technical report: GDPval-AA v2 (Elo) = 1736.0. Source setting: agentic benchmark harness. Origin: Artificial Analysis cited by Kimi report.", "candidates": [ { "score": 1747.8, "reference_url": "https://openai.com/index/gpt-5-6/", "source_type": "official_blog", "reported_setting": { "effort": "max", "eval_variant": null, "code_mode": "false" }, "notes": "OpenAI GPT-5.6 release table (table 0): GPT‑5.6 Sol; GDPval-AA v2=1747.8. OpenAI row explicitly identifies GDPval-AA v2. Exact effort/variant resolved from the rendered chart point." }, { "score": 1384.0, "reference_url": "https://openai.com/index/gpt-5-6/", "source_type": "official_blog", "reported_setting": { "effort": "none", "eval_variant": null, "code_mode": "false" }, "notes": "OpenAI GPT-5.6 release figure (GDPval-AA v2): GPT-5.6 Sol; GDPval-AA v2=1384.0. Figure 15: GDPval-AA v2. Cost, latency, and output-token coordinates are retained as review metadata, not score-matrix metrics." }, { "score": 1445.0, "reference_url": "https://openai.com/index/gpt-5-6/", "source_type": "official_blog", "reported_setting": { "effort": "low", "eval_variant": null, "code_mode": "false" }, "notes": "OpenAI GPT-5.6 release figure (GDPval-AA v2): GPT-5.6 Sol; GDPval-AA v2=1445.0. Figure 15: GDPval-AA v2. Cost, latency, and output-token coordinates are retained as review metadata, not score-matrix metrics." }, { "score": 1562.0, "reference_url": "https://openai.com/index/gpt-5-6/", "source_type": "official_blog", "reported_setting": { "effort": "medium", "eval_variant": null, "code_mode": "false" }, "notes": "OpenAI GPT-5.6 release figure (GDPval-AA v2): GPT-5.6 Sol; GDPval-AA v2=1562.0. Figure 15: GDPval-AA v2. Cost, latency, and output-token coordinates are retained as review metadata, not score-matrix metrics." }, { "score": 1630.0, "reference_url": "https://openai.com/index/gpt-5-6/", "source_type": "official_blog", "reported_setting": { "effort": "high", "eval_variant": null, "code_mode": "false" }, "notes": "OpenAI GPT-5.6 release figure (GDPval-AA v2): GPT-5.6 Sol; GDPval-AA v2=1630.0. Figure 15: GDPval-AA v2. Cost, latency, and output-token coordinates are retained as review metadata, not score-matrix metrics." }, { "score": 1702.0, "reference_url": "https://openai.com/index/gpt-5-6/", "source_type": "official_blog", "reported_setting": { "effort": "xhigh", "eval_variant": null, "code_mode": "false" }, "notes": "OpenAI GPT-5.6 release figure (GDPval-AA v2): GPT-5.6 Sol; GDPval-AA v2=1702.0. Figure 15: GDPval-AA v2. Cost, latency, and output-token coordinates are retained as review metadata, not score-matrix metrics." } ] }, { "model_id": "gpt-5.6-sol", "benchmark_id": "toolathlon_verified", "score": 74.9, "reference_url": "https://github.com/MoonshotAI/Kimi-K3/blob/main/k3_tech_report.pdf", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "benchmark-specified", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "Verified edition; official leaderboard snapshot", "prompt_style": "default", "temperature": "default", "context": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Kimi K3 technical report: Toolathlon-Verified = 74.9. Source setting: Verified edition; official leaderboard snapshot. Origin: official Toolathlon leaderboard cited by Kimi report." }, { "model_id": "gpt-5.6-sol", "benchmark_id": "mcpmark_verified", "score": 92.9, "reference_url": "https://github.com/MoonshotAI/Kimi-K3/blob/main/k3_tech_report.pdf", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "benchmark-specified", "sampling": "not restated", "judge": "benchmark-specified", "harness": "verified tasks; sampling not restated in K3 report", "prompt_style": "default", "temperature": "default", "context": "default" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Kimi K3 technical report: MCPMark-Verified = 92.9. Source setting: verified tasks; sampling not restated in K3 report. Origin: Moonshot evaluation. Canonical compatibility remains under review; stored as best available primary with matches_canonical=false." }, { "model_id": "gpt-5.6-sol", "benchmark_id": "mcpatlas", "score": 83.6, "reference_url": "https://github.com/MoonshotAI/Kimi-K3/blob/main/k3_tech_report.pdf", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "benchmark-specified", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "public 500; maxTurns=100; Gemini 3.1 Pro judge", "prompt_style": "default", "temperature": "default", "context": "default" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Kimi K3 technical report: MCP-Atlas = 83.6. Source setting: public 500; maxTurns=100; Gemini 3.1 Pro judge. Origin: Moonshot evaluation." }, { "model_id": "gpt-5.6-sol", "benchmark_id": "automation_bench", "score": 29.7, "reference_url": "https://github.com/MoonshotAI/Kimi-K3/blob/main/k3_tech_report.pdf", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "benchmark-specified", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "agentic benchmark harness", "prompt_style": "default", "temperature": "default", "context": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Kimi K3 technical report: AutomationBench = 29.7. Source setting: agentic benchmark harness. Origin: Moonshot evaluation.", "candidates": [ { "score": 18.1, "reference_url": "https://openai.com/index/gpt-5-6/", "source_type": "official_blog", "reported_setting": { "effort": "max", "eval_variant": null, "code_mode": "false" }, "notes": "OpenAI GPT-5.6 release table (table 8): GPT‑5.6 Sol; AutomationBench=18.1. Exact effort/variant resolved from the rendered chart point." }, { "score": 9.7, "reference_url": "https://openai.com/index/gpt-5-6/", "source_type": "official_blog", "reported_setting": { "effort": "none", "eval_variant": null, "code_mode": "false" }, "notes": "OpenAI GPT-5.6 release figure (AutomationBench): GPT-5.6 Sol; AutomationBench=9.7. Figure 15: AutomationBench. Cost, latency, and output-token coordinates are retained as review metadata, not score-matrix metrics." }, { "score": 9.6, "reference_url": "https://openai.com/index/gpt-5-6/", "source_type": "official_blog", "reported_setting": { "effort": "low", "eval_variant": null, "code_mode": "false" }, "notes": "OpenAI GPT-5.6 release figure (AutomationBench): GPT-5.6 Sol; AutomationBench=9.6. Figure 15: AutomationBench. Cost, latency, and output-token coordinates are retained as review metadata, not score-matrix metrics." }, { "score": 12.6, "reference_url": "https://openai.com/index/gpt-5-6/", "source_type": "official_blog", "reported_setting": { "effort": "medium", "eval_variant": null, "code_mode": "false" }, "notes": "OpenAI GPT-5.6 release figure (AutomationBench): GPT-5.6 Sol; AutomationBench=12.6. Figure 15: AutomationBench. Cost, latency, and output-token coordinates are retained as review metadata, not score-matrix metrics." }, { "score": 12.3, "reference_url": "https://openai.com/index/gpt-5-6/", "source_type": "official_blog", "reported_setting": { "effort": "high", "eval_variant": null, "code_mode": "false" }, "notes": "OpenAI GPT-5.6 release figure (AutomationBench): GPT-5.6 Sol; AutomationBench=12.3. Figure 15: AutomationBench. Cost, latency, and output-token coordinates are retained as review metadata, not score-matrix metrics." }, { "score": 17.0, "reference_url": "https://openai.com/index/gpt-5-6/", "source_type": "official_blog", "reported_setting": { "effort": "xhigh", "eval_variant": null, "code_mode": "false" }, "notes": "OpenAI GPT-5.6 release figure (AutomationBench): GPT-5.6 Sol; AutomationBench=17.0. Figure 15: AutomationBench. Cost, latency, and output-token coordinates are retained as review metadata, not score-matrix metrics." } ] }, { "model_id": "gpt-5.6-sol", "benchmark_id": "job_bench", "score": 45.4, "reference_url": "https://github.com/MoonshotAI/Kimi-K3/blob/main/k3_tech_report.pdf", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "benchmark-specified", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "agentic benchmark harness", "prompt_style": "default", "temperature": "default", "context": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Kimi K3 technical report: JobBench = 45.4. Source setting: agentic benchmark harness. Origin: official JobBench leaderboard cited by Kimi report." }, { "model_id": "gpt-5.6-sol", "benchmark_id": "aa_briefcase_elo", "score": 1495.0, "reference_url": "https://github.com/MoonshotAI/Kimi-K3/blob/main/k3_tech_report.pdf", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "benchmark-specified", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "agentic benchmark harness", "prompt_style": "default", "temperature": "default", "context": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Kimi K3 technical report: AA-Briefcase (Elo) = 1495.0. Source setting: agentic benchmark harness. Origin: Artificial Analysis cited by Kimi report.", "candidates": [ { "score": 1505, "reference_url": "https://www.anthropic.com/claude-opus-5-system-card", "source_type": "model_card", "reported_setting": { "effort": "max", "harness": "official or cited leaderboard" }, "notes": "Table 8.1.A / AA-Briefcase: Unless noted, Anthropic values use adaptive thinking at max effort, default sampling, mean over five trials. Table 8.1.A card summary." }, { "score": 1502, "reference_url": "https://x.ai/news/grok-4-6", "source_type": "official_blog", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "offline multi-file knowledge-work tools", "sampling": "unknown", "judge": "rubric plus analytical and presentation pairwise Elo", "harness": "Artificial Analysis AA-Briefcase harness", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown" }, "notes": "release46 AA-Briefcase Elo; source effort=max; benchmark protocol matches canonical=True. Unknown source fields remain unknown." } ] }, { "model_id": "gpt-5.6-sol", "benchmark_id": "agents_last_exam", "score": 29.6, "reference_url": "https://github.com/MoonshotAI/Kimi-K3/blob/main/k3_tech_report.pdf", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "benchmark-specified", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "agentic benchmark harness", "prompt_style": "default", "temperature": "default", "context": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Kimi K3 technical report: Agents' Last Exam = 29.6. Source setting: agentic benchmark harness. Origin: official leaderboard cited by Kimi report.", "candidates": [ { "score": 52.7, "reference_url": "https://openai.com/index/gpt-5-6/", "source_type": "official_blog", "reported_setting": { "effort": "max", "eval_variant": null, "code_mode": "true" }, "notes": "OpenAI GPT-5.6 release table (table 0): GPT‑5.6 Sol; Agents' Last Exam=52.7. Exact effort/variant resolved from the rendered chart point." }, { "score": 44.8, "reference_url": "https://openai.com/index/gpt-5-6/", "source_type": "official_blog", "reported_setting": { "effort": "low", "eval_variant": null, "code_mode": "true" }, "notes": "OpenAI GPT-5.6 release figure (Agents' Last Exam): GPT-5.6 Sol; Agents' Last Exam=44.8. Figure 0: Long-horizon agentic workflows across professional domains.. Cost, latency, and output-token coordinates are retained as review metadata, not score-matrix metrics." }, { "score": 51.9, "reference_url": "https://openai.com/index/gpt-5-6/", "source_type": "official_blog", "reported_setting": { "effort": "medium", "eval_variant": null, "code_mode": "true" }, "notes": "OpenAI GPT-5.6 release figure (Agents' Last Exam): GPT-5.6 Sol; Agents' Last Exam=51.9. Figure 0: Long-horizon agentic workflows across professional domains.. Cost, latency, and output-token coordinates are retained as review metadata, not score-matrix metrics." }, { "score": 52.1, "reference_url": "https://openai.com/index/gpt-5-6/", "source_type": "official_blog", "reported_setting": { "effort": "high", "eval_variant": null, "code_mode": "true" }, "notes": "OpenAI GPT-5.6 release figure (Agents' Last Exam): GPT-5.6 Sol; Agents' Last Exam=52.1. Figure 0: Long-horizon agentic workflows across professional domains.. Cost, latency, and output-token coordinates are retained as review metadata, not score-matrix metrics." }, { "score": 53.6, "reference_url": "https://openai.com/index/gpt-5-6/", "source_type": "official_blog", "reported_setting": { "effort": "xhigh", "eval_variant": null, "code_mode": "true" }, "notes": "OpenAI GPT-5.6 release figure (Agents' Last Exam): GPT-5.6 Sol; Agents' Last Exam=53.6. Figure 0: Long-horizon agentic workflows across professional domains.. Cost, latency, and output-token coordinates are retained as review metadata, not score-matrix metrics." } ] }, { "model_id": "gpt-5.6-sol", "benchmark_id": "apex_agents", "score": 39.9, "reference_url": "https://github.com/MoonshotAI/Kimi-K3/blob/main/k3_tech_report.pdf", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "benchmark-specified", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "agentic benchmark harness", "prompt_style": "default", "temperature": "default", "context": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Kimi K3 technical report: APEX-Agents = 39.9. Source setting: agentic benchmark harness. Origin: Artificial Analysis cited by Kimi report.", "candidates": [ { "score": 56.7, "reference_url": "https://x.ai/news/grok-4-6", "source_type": "official_blog", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "multi-application professional tools", "sampling": "pass@1", "judge": "expert all-criteria-pass rubric", "harness": "Mercor APEX-Agents harness", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown" }, "notes": "release46 APEX-Agents Pass@1 (%); source effort=max; benchmark protocol matches canonical=True. Unknown source fields remain unknown." } ] }, { "model_id": "gpt-5.6-sol", "benchmark_id": "officeqa_pro", "score": 63.2, "reference_url": "https://github.com/MoonshotAI/Kimi-K3/blob/main/k3_tech_report.pdf", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "benchmark-specified", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "PDF corpus rendered as images; no machine-readable text", "prompt_style": "default", "temperature": "default", "context": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Kimi K3 technical report: OfficeQA Pro = 63.2. Source setting: PDF corpus rendered as images; no machine-readable text. Origin: Moonshot evaluation.", "candidates": [ { "score": 60.2, "reference_url": "https://media.x.ai/v1/website/card-4p6-4cd2dc57.pdf", "source_type": "model_card", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "document and rendered-image analysis", "sampling": "unknown", "judge": "OfficeQA Pro exact-match accuracy", "harness": "OfficeQA Pro official harness", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown" }, "notes": "card46 OfficeQA Pro Accuracy (%); source effort=max; benchmark protocol matches canonical=True. Unknown source fields remain unknown." } ] }, { "model_id": "gpt-5.6-sol", "benchmark_id": "spreadsheetbench_2", "score": 32.4, "reference_url": "https://github.com/MoonshotAI/Kimi-K3/blob/main/k3_tech_report.pdf", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "benchmark-specified", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "agentic benchmark harness", "prompt_style": "default", "temperature": "default", "context": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Kimi K3 technical report: SpreadsheetBench 2 = 32.4. Source setting: agentic benchmark harness. Origin: Moonshot evaluation." }, { "model_id": "gpt-5.6-sol", "benchmark_id": "osworld_verified", "score": 83.0, "reference_url": "https://github.com/MoonshotAI/Kimi-K3/blob/main/k3_tech_report.pdf", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "benchmark-specified", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "agentic benchmark harness", "prompt_style": "default", "temperature": "default", "context": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Kimi K3 technical report: OSWorld-Verified = 83.0. Source setting: agentic benchmark harness. Origin: Moonshot evaluation." }, { "model_id": "gpt-5.6-sol", "benchmark_id": "osworld_2_0", "score": 62.6, "reference_url": "https://github.com/MoonshotAI/Kimi-K3/blob/main/k3_tech_report.pdf", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "benchmark-specified", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "agentic benchmark harness", "prompt_style": "default", "temperature": "default", "context": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Kimi K3 technical report: OSWorld 2.0 = 62.6. Source setting: agentic benchmark harness. Origin: Moonshot evaluation.", "candidates": [ { "score": 10.02, "reference_url": "https://openai.com/index/gpt-5-6/", "source_type": "official_blog", "reported_setting": { "effort": "none", "eval_variant": null, "code_mode": "false" }, "notes": "OpenAI GPT-5.6 release figure (OSWorld 2.0): GPT-5.6 Sol; OSWorld 2.0=10.02. Figure 15: OSWorld 2.0. Cost, latency, and output-token coordinates are retained as review metadata, not score-matrix metrics." }, { "score": 20.73, "reference_url": "https://openai.com/index/gpt-5-6/", "source_type": "official_blog", "reported_setting": { "effort": "low", "eval_variant": null, "code_mode": "false" }, "notes": "OpenAI GPT-5.6 release figure (OSWorld 2.0): GPT-5.6 Sol; OSWorld 2.0=20.73. Figure 15: OSWorld 2.0. Cost, latency, and output-token coordinates are retained as review metadata, not score-matrix metrics." }, { "score": 42.21, "reference_url": "https://openai.com/index/gpt-5-6/", "source_type": "official_blog", "reported_setting": { "effort": "medium", "eval_variant": null, "code_mode": "false" }, "notes": "OpenAI GPT-5.6 release figure (OSWorld 2.0): GPT-5.6 Sol; OSWorld 2.0=42.21. Figure 15: OSWorld 2.0. Cost, latency, and output-token coordinates are retained as review metadata, not score-matrix metrics." }, { "score": 49.75, "reference_url": "https://openai.com/index/gpt-5-6/", "source_type": "official_blog", "reported_setting": { "effort": "high", "eval_variant": null, "code_mode": "false" }, "notes": "OpenAI GPT-5.6 release figure (OSWorld 2.0): GPT-5.6 Sol; OSWorld 2.0=49.75. Figure 15: OSWorld 2.0. Cost, latency, and output-token coordinates are retained as review metadata, not score-matrix metrics." }, { "score": 55.82, "reference_url": "https://openai.com/index/gpt-5-6/", "source_type": "official_blog", "reported_setting": { "effort": "xhigh", "eval_variant": null, "code_mode": "false" }, "notes": "OpenAI GPT-5.6 release figure (OSWorld 2.0): GPT-5.6 Sol; OSWorld 2.0=55.82. Figure 15: OSWorld 2.0. Cost, latency, and output-token coordinates are retained as review metadata, not score-matrix metrics." }, { "score": 30.51, "reference_url": "https://openai.com/index/gpt-5-6/", "source_type": "official_blog", "reported_setting": { "effort": "none", "eval_variant": null, "code_mode": "true" }, "notes": "OpenAI GPT-5.6 release figure (OSWorld 2.0): GPT-5.6 Sol; OSWorld 2.0=30.51. Figure 15: OSWorld 2.0. Cost, latency, and output-token coordinates are retained as review metadata, not score-matrix metrics." }, { "score": 49.94, "reference_url": "https://openai.com/index/gpt-5-6/", "source_type": "official_blog", "reported_setting": { "effort": "low", "eval_variant": null, "code_mode": "true" }, "notes": "OpenAI GPT-5.6 release figure (OSWorld 2.0): GPT-5.6 Sol; OSWorld 2.0=49.94. Figure 15: OSWorld 2.0. Cost, latency, and output-token coordinates are retained as review metadata, not score-matrix metrics." }, { "score": 53.17, "reference_url": "https://openai.com/index/gpt-5-6/", "source_type": "official_blog", "reported_setting": { "effort": "medium", "eval_variant": null, "code_mode": "true" }, "notes": "OpenAI GPT-5.6 release figure (OSWorld 2.0): GPT-5.6 Sol; OSWorld 2.0=53.17. Figure 15: OSWorld 2.0. Cost, latency, and output-token coordinates are retained as review metadata, not score-matrix metrics." }, { "score": 53.55, "reference_url": "https://openai.com/index/gpt-5-6/", "source_type": "official_blog", "reported_setting": { "effort": "high", "eval_variant": null, "code_mode": "true" }, "notes": "OpenAI GPT-5.6 release figure (OSWorld 2.0): GPT-5.6 Sol; OSWorld 2.0=53.55. Figure 15: OSWorld 2.0. Cost, latency, and output-token coordinates are retained as review metadata, not score-matrix metrics." }, { "score": 59.22, "reference_url": "https://openai.com/index/gpt-5-6/", "source_type": "official_blog", "reported_setting": { "effort": "xhigh", "eval_variant": null, "code_mode": "true" }, "notes": "OpenAI GPT-5.6 release figure (OSWorld 2.0): GPT-5.6 Sol; OSWorld 2.0=59.22. Figure 15: OSWorld 2.0. Cost, latency, and output-token coordinates are retained as review metadata, not score-matrix metrics." }, { "score": 60.4, "reference_url": "https://openai.com/index/gpt-5-6/", "source_type": "official_blog", "reported_setting": { "effort": "max", "eval_variant": null, "code_mode": "true" }, "notes": "OpenAI GPT-5.6 release figure (OSWorld 2.0): GPT-5.6 Sol; OSWorld 2.0=60.4. Figure 15: OSWorld 2.0. Cost, latency, and output-token coordinates are retained as review metadata, not score-matrix metrics." } ] }, { "model_id": "gpt-5.6-sol", "benchmark_id": "saas_bench", "score": 61.4, "reference_url": "https://github.com/MoonshotAI/Kimi-K3/blob/main/k3_tech_report.pdf", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "benchmark-specified", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "agentic benchmark harness", "prompt_style": "default", "temperature": "default", "context": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Kimi K3 technical report: SaaS-Bench = 61.4. Source setting: agentic benchmark harness. Origin: Moonshot evaluation." }, { "model_id": "gpt-5.6-sol", "benchmark_id": "tau3_banking", "score": 33.0, "reference_url": "https://github.com/MoonshotAI/Kimi-K3/blob/main/k3_tech_report.pdf", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "benchmark-specified", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "agentic benchmark harness", "prompt_style": "default", "temperature": "default", "context": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Kimi K3 technical report: τ³-Banking = 33.0. Source setting: agentic benchmark harness. Origin: Artificial Analysis cited by Kimi report.", "candidates": [ { "score": 33.0, "reference_url": "https://media.x.ai/v1/website/4p5-5184fdf9.pdf", "source_type": "model_card", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "banking APIs and simulated user", "sampling": "pass@1", "judge": "final-state task accuracy", "harness": "Artificial Analysis tau3-banking harness", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown" }, "notes": "card45 tau3-banking Accuracy (%); source effort=max; benchmark protocol matches canonical=True. Unknown source fields remain unknown." } ] }, { "model_id": "gpt-5.6-sol", "benchmark_id": "harvey_lab_aa", "score": 87.2, "reference_url": "https://github.com/MoonshotAI/Kimi-K3/blob/main/k3_tech_report.pdf", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "benchmark-specified", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "agentic benchmark harness", "prompt_style": "default", "temperature": "default", "context": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Kimi K3 technical report: Harvey Lab-AA = 87.2. Source setting: agentic benchmark harness. Origin: Artificial Analysis cited by Kimi report." }, { "model_id": "gpt-5.6-sol", "benchmark_id": "corpfin_v2", "score": 64.4, "reference_url": "https://github.com/MoonshotAI/Kimi-K3/blob/main/k3_tech_report.pdf", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "benchmark-specified", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "agentic benchmark harness", "prompt_style": "default", "temperature": "default", "context": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Kimi K3 technical report: CorpFin v2 = 64.4. Source setting: agentic benchmark harness. Origin: Vals AI cited by Kimi report." }, { "model_id": "gpt-5.6-sol", "benchmark_id": "finance_agent_v2", "score": 53.8, "reference_url": "https://github.com/MoonshotAI/Kimi-K3/blob/main/k3_tech_report.pdf", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "benchmark-specified", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "agentic benchmark harness", "prompt_style": "default", "temperature": "default", "context": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Kimi K3 technical report: Finance Agent v2 = 53.8. Source setting: agentic benchmark harness. Origin: Vals AI cited by Kimi report." }, { "model_id": "gpt-5.6-sol", "benchmark_id": "legal_research_bench", "score": 48.1, "reference_url": "https://github.com/MoonshotAI/Kimi-K3/blob/main/k3_tech_report.pdf", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "benchmark-specified", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "agentic benchmark harness", "prompt_style": "default", "temperature": "default", "context": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Kimi K3 technical report: Legal Research Bench = 48.1. Source setting: agentic benchmark harness. Origin: Vals AI cited by Kimi report." }, { "model_id": "gpt-5.6-sol", "benchmark_id": "worldvqa", "score": 41.8, "reference_url": "https://github.com/MoonshotAI/Kimi-K3/blob/main/k3_tech_report.pdf", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "none", "sampling": "avg-of-3", "judge": "benchmark-specified", "harness": "forced-answer prompt; avg 3 runs", "prompt_style": "default", "temperature": "default", "context": "default" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Kimi K3 technical report: WorldVQA ForceAnswer = 41.8. Source setting: forced-answer prompt; avg 3 runs. Origin: Moonshot evaluation." }, { "model_id": "gpt-5.6-sol", "benchmark_id": "omnidocbench", "score": 0.142, "reference_url": "https://github.com/MoonshotAI/Kimi-K3/blob/main/k3_tech_report.pdf", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "benchmark-specified", "sampling": "avg-of-3", "judge": "benchmark-specified", "harness": "avg 3 runs; source 1-NED score normalized to NED", "prompt_style": "default", "temperature": "default", "context": "default" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Kimi K3 technical report: OmniDocBench = 0.142 NED. Source setting: avg 3 runs; source 1-NED score normalized to NED. Origin: Moonshot evaluation. The source reports 1-NED=85.8%; stored value is normalized NED=0.142." }, { "model_id": "gpt-5.6-sol", "benchmark_id": "perception_bench", "score": 59.7, "reference_url": "https://github.com/MoonshotAI/Kimi-K3/blob/main/k3_tech_report.pdf", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "benchmark-specified", "sampling": "avg-of-3", "judge": "benchmark-specified", "harness": "multimodal; avg 3 runs", "prompt_style": "default", "temperature": "default", "context": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Kimi K3 technical report: PerceptionBench = 59.7. Source setting: multimodal; avg 3 runs. Origin: Moonshot evaluation." }, { "model_id": "gpt-5.6-sol", "benchmark_id": "video_mme", "score": 89.5, "reference_url": "https://github.com/MoonshotAI/Kimi-K3/blob/main/k3_tech_report.pdf", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "none", "sampling": "avg-of-3", "judge": "benchmark-specified", "harness": "audio+visual+subtitles; avg 3 runs", "prompt_style": "default", "temperature": "default", "context": "default" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Kimi K3 technical report: Video-MME (w. sub) = 89.5. Source setting: audio+visual+subtitles; avg 3 runs. Origin: Moonshot evaluation." }, { "model_id": "gpt-5.6-sol", "benchmark_id": "mmvu", "score": 81.2, "reference_url": "https://github.com/MoonshotAI/Kimi-K3/blob/main/k3_tech_report.pdf", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "benchmark-specified", "sampling": "avg-of-3", "judge": "benchmark-specified", "harness": "avg 3 runs", "prompt_style": "default", "temperature": "default", "context": "default" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Kimi K3 technical report: MMVU = 81.2. Source setting: avg 3 runs. Origin: Moonshot evaluation." }, { "model_id": "gpt-5.6-sol", "benchmark_id": "babyvision", "score": 88.9, "reference_url": "https://github.com/MoonshotAI/Kimi-K3/blob/main/k3_tech_report.pdf", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "python", "sampling": "avg-of-3", "judge": "benchmark-specified", "harness": "Python tool; avg 3 runs", "prompt_style": "default", "temperature": "default", "context": "default" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Kimi K3 technical report: BabyVision w/ python = 88.9. Source setting: Python tool; avg 3 runs. Origin: Moonshot evaluation." }, { "model_id": "gpt-5.6-sol", "benchmark_id": "mmmu_pro", "score": 83.0, "reference_url": "https://github.com/MoonshotAI/Kimi-K3/blob/main/k3_tech_report.pdf", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "none", "sampling": "avg-of-3", "judge": "benchmark-specified", "harness": "vision; avg 3 runs", "prompt_style": "default", "temperature": "default", "context": "default" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Kimi K3 technical report: MMMU-Pro (no Python) = 83.0. Source setting: vision; avg 3 runs. Origin: Moonshot evaluation.", "candidates": [ { "score": 84.6, "reference_url": "https://github.com/MoonshotAI/Kimi-K3/blob/main/k3_tech_report.pdf", "source_type": "tech_report", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "python", "sampling": "avg-of-3", "judge": "benchmark-specified", "harness": "Python tool; avg 3 runs", "prompt_style": "default", "temperature": "default", "context": "default" }, "notes": "Kimi K3 technical report: MMMU-Pro (with Python) = 84.6. Source setting: Python tool; avg 3 runs. Origin: Moonshot evaluation." } ] }, { "model_id": "gpt-5.6-sol", "benchmark_id": "charxiv_reasoning", "score": 84.6, "reference_url": "https://github.com/MoonshotAI/Kimi-K3/blob/main/k3_tech_report.pdf", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "none", "sampling": "avg-of-3", "judge": "benchmark-specified", "harness": "vision; avg 3 runs", "prompt_style": "default", "temperature": "default", "context": "default" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Kimi K3 technical report: CharXiv (RQ) (no Python) = 84.6. Source setting: vision; avg 3 runs. Origin: Moonshot evaluation.", "candidates": [ { "score": 89.1, "reference_url": "https://github.com/MoonshotAI/Kimi-K3/blob/main/k3_tech_report.pdf", "source_type": "tech_report", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "python", "sampling": "avg-of-3", "judge": "benchmark-specified", "harness": "Python tool; avg 3 runs", "prompt_style": "default", "temperature": "default", "context": "default" }, "notes": "Kimi K3 technical report: CharXiv (RQ) (with Python) = 89.1. Source setting: Python tool; avg 3 runs. Origin: Moonshot evaluation." } ] }, { "model_id": "gpt-5.6-sol", "benchmark_id": "mathvision", "score": 95.8, "reference_url": "https://github.com/MoonshotAI/Kimi-K3/blob/main/k3_tech_report.pdf", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "none", "sampling": "avg-of-3", "judge": "benchmark-specified", "harness": "vision; avg 3 runs", "prompt_style": "default", "temperature": "default", "context": "default" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Kimi K3 technical report: MathVision (no Python) = 95.8. Source setting: vision; avg 3 runs. Origin: Moonshot evaluation.", "candidates": [ { "score": 97.8, "reference_url": "https://github.com/MoonshotAI/Kimi-K3/blob/main/k3_tech_report.pdf", "source_type": "tech_report", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "python", "sampling": "avg-of-3", "judge": "benchmark-specified", "harness": "Python tool; avg 3 runs", "prompt_style": "default", "temperature": "default", "context": "default" }, "notes": "Kimi K3 technical report: MathVision (with Python) = 97.8. Source setting: Python tool; avg 3 runs. Origin: Moonshot evaluation." } ] }, { "model_id": "gpt-5.6-sol", "benchmark_id": "zerobench_main", "score": 17.0, "reference_url": "https://github.com/MoonshotAI/Kimi-K3/blob/main/k3_tech_report.pdf", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "benchmark-specified", "sampling": "pass@5", "judge": "benchmark-specified", "harness": "no Python; pass@5 / 5 runs", "prompt_style": "default", "temperature": "default", "context": "default" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Kimi K3 technical report: ZeroBench (pass@5) (no Python) = 17.0. Source setting: no Python; pass@5 / 5 runs. Origin: Moonshot evaluation. Canonical compatibility remains under review; stored as best available primary with matches_canonical=false." }, { "model_id": "gpt-5.6-sol", "benchmark_id": "zerobench_tools", "score": 35.0, "reference_url": "https://github.com/MoonshotAI/Kimi-K3/blob/main/k3_tech_report.pdf", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "python", "sampling": "pass@5", "judge": "benchmark-specified", "harness": "Python tool; pass@5 / 5 runs", "prompt_style": "default", "temperature": "default", "context": "default" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Kimi K3 technical report: ZeroBench (pass@5) (with Python) = 35.0. Source setting: Python tool; pass@5 / 5 runs. Origin: Moonshot evaluation. Canonical compatibility remains under review; stored as best available primary with matches_canonical=false." }, { "model_id": "gpt-5.6-sol", "benchmark_id": "coding_experience", "score": 59.3, "reference_url": "https://github.com/MoonshotAI/Kimi-K3/blob/main/k3_tech_report.pdf", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "benchmark-specified", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "provider-aligned harness: Kimi Code / Claude Code / Codex", "prompt_style": "default", "temperature": "default", "context": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Kimi K3 technical report: Coding Experience = 59.3. Source setting: provider-aligned harness: Kimi Code / Claude Code / Codex. Origin: Moonshot evaluation." }, { "model_id": "gpt-5.6-sol", "benchmark_id": "clawbench_2_0", "score": 52.0, "reference_url": "https://github.com/MoonshotAI/Kimi-K3/blob/main/k3_tech_report.pdf", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "benchmark-specified", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "OpenClaw harness", "prompt_style": "default", "temperature": "default", "context": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Kimi K3 technical report: 24/7 ClawBench 2.0 = 52.0. Source setting: OpenClaw harness. Origin: Moonshot evaluation." }, { "model_id": "gpt-5.6-sol", "benchmark_id": "mira_bench", "score": 62.2, "reference_url": "https://github.com/MoonshotAI/Kimi-K3/blob/main/k3_tech_report.pdf", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "benchmark-specified", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "MIRA harness", "prompt_style": "default", "temperature": "default", "context": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Kimi K3 technical report: MIRA Bench = 62.2. Source setting: MIRA harness. Origin: Moonshot evaluation." }, { "model_id": "gpt-5.6-sol", "benchmark_id": "kaet", "score": 85.4, "reference_url": "https://github.com/MoonshotAI/Kimi-K3/blob/main/k3_tech_report.pdf", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "benchmark-specified", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "Kimi Code harness", "prompt_style": "default", "temperature": "default", "context": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Kimi K3 technical report: KAET = 85.4. Source setting: Kimi Code harness. Origin: Moonshot evaluation." }, { "model_id": "gpt-5.6-sol", "benchmark_id": "clif_bench", "score": 50.6, "reference_url": "https://github.com/MoonshotAI/Kimi-K3/blob/main/k3_tech_report.pdf", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "benchmark-specified", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "Kimi Code harness", "prompt_style": "default", "temperature": "default", "context": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Kimi K3 technical report: CLIF Bench = 50.6. Source setting: Kimi Code harness. Origin: Moonshot evaluation." }, { "model_id": "gpt-5.6-sol", "benchmark_id": "agentic_vision_bench", "score": 82.9, "reference_url": "https://github.com/MoonshotAI/Kimi-K3/blob/main/k3_tech_report.pdf", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "benchmark-specified", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "Kimi Code harness", "prompt_style": "default", "temperature": "default", "context": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Kimi K3 technical report: Agentic Vision Bench = 82.9. Source setting: Kimi Code harness. Origin: Moonshot evaluation." }, { "model_id": "gpt-5.6-sol", "benchmark_id": "swarm_bench", "score": 73.2, "reference_url": "https://github.com/MoonshotAI/Kimi-K3/blob/main/k3_tech_report.pdf", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "benchmark-specified", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "Kimi Agent harness", "prompt_style": "default", "temperature": "default", "context": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Kimi K3 technical report: SwarmBench = 73.2. Source setting: Kimi Agent harness. Origin: Moonshot evaluation." }, { "model_id": "gpt-5.6-sol", "benchmark_id": "online_experience", "score": 84.0, "reference_url": "https://github.com/MoonshotAI/Kimi-K3/blob/main/k3_tech_report.pdf", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "benchmark-specified", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "Kimi Agent harness", "prompt_style": "default", "temperature": "default", "context": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Kimi K3 technical report: Online Experience = 84.0. Source setting: Kimi Agent harness. Origin: Moonshot evaluation." }, { "model_id": "gpt-5.6-sol", "benchmark_id": "deepresearchbench", "score": 85.3, "reference_url": "https://github.com/MoonshotAI/Kimi-K3/blob/main/k3_tech_report.pdf", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "benchmark-specified", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "Kimi Agent harness", "prompt_style": "default", "temperature": "default", "context": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Kimi K3 technical report: DeepResearchBench = 85.3. Source setting: Kimi Agent harness. Origin: Moonshot evaluation." }, { "model_id": "gpt-5.6-sol", "benchmark_id": "finance_bench", "score": 62.7, "reference_url": "https://github.com/MoonshotAI/Kimi-K3/blob/main/k3_tech_report.pdf", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "benchmark-specified", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "N/A harness", "prompt_style": "default", "temperature": "default", "context": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Kimi K3 technical report: Finance Bench = 62.7. Source setting: N/A harness. Origin: Moonshot evaluation." }, { "model_id": "gpt-5.6-sol", "benchmark_id": "kwv_bench", "score": 66.9, "reference_url": "https://github.com/MoonshotAI/Kimi-K3/blob/main/k3_tech_report.pdf", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "benchmark-specified", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "N/A harness", "prompt_style": "default", "temperature": "default", "context": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Kimi K3 technical report: KWVBench = 66.9. Source setting: N/A harness. Origin: Moonshot evaluation." }, { "model_id": "gpt-5.6-sol", "benchmark_id": "deck_bench", "score": 74.7, "reference_url": "https://github.com/MoonshotAI/Kimi-K3/blob/main/k3_tech_report.pdf", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "benchmark-specified", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "N/A harness", "prompt_style": "default", "temperature": "default", "context": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Kimi K3 technical report: DECKBench = 74.7. Source setting: N/A harness. Origin: Moonshot evaluation." }, { "model_id": "gpt-5.6-sol", "benchmark_id": "agent_behavior_bench", "score": 76.4, "reference_url": "https://github.com/MoonshotAI/Kimi-K3/blob/main/k3_tech_report.pdf", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "benchmark-specified", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "Kimi Work harness", "prompt_style": "default", "temperature": "default", "context": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Kimi K3 technical report: Agent Behavior Bench = 76.4. Source setting: Kimi Work harness. Origin: Moonshot evaluation." }, { "model_id": "gpt-5.6-sol", "benchmark_id": "faithfulness", "score": 84.8, "reference_url": "https://github.com/MoonshotAI/Kimi-K3/blob/main/k3_tech_report.pdf", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "benchmark-specified", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "N/A harness", "prompt_style": "default", "temperature": "default", "context": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Kimi K3 technical report: Faithfulness (1 - hallucination rate) = 84.8. Source setting: N/A harness. Origin: Moonshot evaluation." }, { "model_id": "gpt-5.6-sol", "benchmark_id": "chat_all_in_one_bench", "score": 79.0, "reference_url": "https://github.com/MoonshotAI/Kimi-K3/blob/main/k3_tech_report.pdf", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "benchmark-specified", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "Kimi Work harness", "prompt_style": "default", "temperature": "default", "context": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Kimi K3 technical report: Chat All-in-One Bench = 79.0. Source setting: Kimi Work harness. Origin: Moonshot evaluation." }, { "model_id": "gpt-5.6-sol", "benchmark_id": "aa_intelligence_index_v4_1", "score": 58.9, "reference_url": "https://github.com/MoonshotAI/Kimi-K3/blob/main/k3_tech_report.pdf", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "benchmark-specified", "sampling": "leaderboard snapshot 2026-07-23", "judge": "benchmark-specified", "harness": "live leaderboard snapshot as of 2026-07-23", "prompt_style": "default", "temperature": "default", "context": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Kimi K3 technical report: AA Intelligence Index v4.1 = 58.9. Source setting: live leaderboard snapshot as of 2026-07-23. Origin: third-party leaderboard cited by Kimi report.", "candidates": [ { "score": 41.2, "reference_url": "https://openai.com/index/gpt-5-6/", "source_type": "official_blog", "reported_setting": { "effort": "none", "eval_variant": null, "code_mode": "false" }, "notes": "OpenAI GPT-5.6 release figure (Artificial Analysis Intelligence Index v4.1): GPT-5.6 Sol; Artificial Analysis Intelligence Index v4.1=41.2. Figure 0: Composite of nine independent evaluations spanning agentic work, coding, scientific reasoning, and general capabilities.. Cost, latency, and output-token coordinates are retained as review metadata, not score-matrix metrics." }, { "score": 49.44, "reference_url": "https://openai.com/index/gpt-5-6/", "source_type": "official_blog", "reported_setting": { "effort": "low", "eval_variant": null, "code_mode": "false" }, "notes": "OpenAI GPT-5.6 release figure (Artificial Analysis Intelligence Index v4.1): GPT-5.6 Sol; Artificial Analysis Intelligence Index v4.1=49.44. Figure 0: Composite of nine independent evaluations spanning agentic work, coding, scientific reasoning, and general capabilities.. Cost, latency, and output-token coordinates are retained as review metadata, not score-matrix metrics." }, { "score": 53.59, "reference_url": "https://openai.com/index/gpt-5-6/", "source_type": "official_blog", "reported_setting": { "effort": "medium", "eval_variant": null, "code_mode": "false" }, "notes": "OpenAI GPT-5.6 release figure (Artificial Analysis Intelligence Index v4.1): GPT-5.6 Sol; Artificial Analysis Intelligence Index v4.1=53.59. Figure 0: Composite of nine independent evaluations spanning agentic work, coding, scientific reasoning, and general capabilities.. Cost, latency, and output-token coordinates are retained as review metadata, not score-matrix metrics." }, { "score": 55.87, "reference_url": "https://openai.com/index/gpt-5-6/", "source_type": "official_blog", "reported_setting": { "effort": "high", "eval_variant": null, "code_mode": "false" }, "notes": "OpenAI GPT-5.6 release figure (Artificial Analysis Intelligence Index v4.1): GPT-5.6 Sol; Artificial Analysis Intelligence Index v4.1=55.87. Figure 0: Composite of nine independent evaluations spanning agentic work, coding, scientific reasoning, and general capabilities.. Cost, latency, and output-token coordinates are retained as review metadata, not score-matrix metrics." }, { "score": 57.65, "reference_url": "https://openai.com/index/gpt-5-6/", "source_type": "official_blog", "reported_setting": { "effort": "xhigh", "eval_variant": null, "code_mode": "false" }, "notes": "OpenAI GPT-5.6 release figure (Artificial Analysis Intelligence Index v4.1): GPT-5.6 Sol; Artificial Analysis Intelligence Index v4.1=57.65. Figure 0: Composite of nine independent evaluations spanning agentic work, coding, scientific reasoning, and general capabilities.. Cost, latency, and output-token coordinates are retained as review metadata, not score-matrix metrics." }, { "score": 61, "reference_url": "https://x.ai/news/grok-4-6", "source_type": "official_blog", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "benchmark-specified", "sampling": "unknown", "judge": "Artificial Analysis weighted composite", "harness": "Artificial Analysis live v4.1 harness", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown" }, "notes": "release46 AA Intelligence Index Composite score; source effort=max; benchmark protocol matches canonical=True. Unknown source fields remain unknown." } ] }, { "model_id": "gpt-5.6-sol", "benchmark_id": "vals_index", "score": 73.1, "reference_url": "https://github.com/MoonshotAI/Kimi-K3/blob/main/k3_tech_report.pdf", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "benchmark-specified", "sampling": "leaderboard snapshot 2026-07-23", "judge": "benchmark-specified", "harness": "live leaderboard snapshot as of 2026-07-23", "prompt_style": "default", "temperature": "default", "context": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Kimi K3 technical report: Vals Index = 73.1. Source setting: live leaderboard snapshot as of 2026-07-23. Origin: third-party leaderboard cited by Kimi report.", "candidates": [ { "score": 73.1, "reference_url": "https://media.x.ai/v1/website/card-4p6-4cd2dc57.pdf", "source_type": "model_card", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "benchmark-specified professional tools", "sampling": "unknown", "judge": "Vals weighted suite score", "harness": "Vals Index live harness", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown" }, "notes": "card46 Vals Index Suite score (%); source effort=max; benchmark protocol matches canonical=True. Unknown source fields remain unknown." } ] }, { "model_id": "gpt-5.6-sol", "benchmark_id": "webdev_arena_elo", "score": 1630.0, "reference_url": "https://github.com/MoonshotAI/Kimi-K3/blob/main/k3_tech_report.pdf", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "benchmark-specified", "sampling": "leaderboard snapshot 2026-07-23", "judge": "benchmark-specified", "harness": "live leaderboard snapshot as of 2026-07-23", "prompt_style": "default", "temperature": "default", "context": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Kimi K3 technical report: WebDevArena Elo = 1630. Source setting: live leaderboard snapshot as of 2026-07-23. Origin: third-party leaderboard cited by Kimi report." }, { "model_id": "gpt-5.6-sol", "benchmark_id": "text_arena_elo", "score": 1485.0, "reference_url": "https://github.com/MoonshotAI/Kimi-K3/blob/main/k3_tech_report.pdf", "reported_setting": { "mode": "thinking", "effort": "xhigh", "tools": "benchmark-specified", "sampling": "leaderboard snapshot 2026-07-23", "judge": "benchmark-specified", "harness": "live leaderboard snapshot as of 2026-07-23", "prompt_style": "default", "temperature": "default", "context": "default" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Kimi K3 technical report: TextArena Elo = 1485. Source setting: live leaderboard snapshot as of 2026-07-23. Origin: third-party leaderboard cited by Kimi report." }, { "model_id": "gpt-5.6-sol", "benchmark_id": "agent_arena", "score": 10.1, "reference_url": "https://github.com/MoonshotAI/Kimi-K3/blob/main/k3_tech_report.pdf", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "benchmark-specified", "sampling": "leaderboard snapshot 2026-07-23", "judge": "benchmark-specified", "harness": "live leaderboard snapshot as of 2026-07-23", "prompt_style": "default", "temperature": "default", "context": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Kimi K3 technical report: AgentArena = 10.1. Source setting: live leaderboard snapshot as of 2026-07-23. Origin: third-party leaderboard cited by Kimi report." }, { "model_id": "claude-opus-4.8", "benchmark_id": "gpqa_diamond", "score": 91.0, "reference_url": "https://github.com/MoonshotAI/Kimi-K3/blob/main/k3_tech_report.pdf", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "none", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "max/xhigh; no tools", "prompt_style": "default", "temperature": "default", "context": "default" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Kimi K3 technical report: GPQA Diamond = 91.0. Source setting: max/xhigh; no tools. Origin: Moonshot evaluation.", "candidates": [ { "score": 93.6, "reference_url": "https://z.ai/blog/glm-5.2", "source_type": "official_blog", "reported_setting": { "mode": "thinking", "effort": "max (source; model default is high)", "sampling": "pass@1", "prompt_style": "default", "tools": "none", "temperature": "1.0", "top_p": "0.95", "context": "default", "max_output_tokens": "163840", "judge": "GPT-5.5 medium for AIME/HMMT/IMO; benchmark-specified otherwise", "harness": "official reasoning evaluation" }, "notes": "Z.ai cross-table max-effort alternative." }, { "score": 92.4, "reference_url": "https://longcat.chat/blog/longcat-2.0/", "source_type": "official_blog", "reported_setting": { "mode": "thinking", "effort": "source does not state", "tools": "none", "sampling": "source does not state", "judge": "benchmark-specified", "harness": "Meituan unified in-house harness", "prompt_style": "default", "temperature": "0.7; top_k=-1; top_p=0.95", "context": "source does not state", "notes": "Measured in-house by Meituan unless otherwise noted." }, "notes": "LongCat-2.0 official tech blog: GPQA-diamond = 92.4. Measured in-house by Meituan under the reported unified harness." }, { "score": 92.0, "reference_url": "https://openai.com/index/gpt-5-6/", "source_type": "official_blog", "reported_setting": { "mode": "thinking", "effort": "source does not state", "tools": "none", "sampling": "source does not state", "judge": "benchmark-specified", "harness": "official release evaluation", "prompt_style": "source does not state", "temperature": "source does not state", "context": "source does not state" }, "notes": "OpenAI GPT-5.6 release table (table 7): Claude Opus 4.8; GPQA Diamond=92.0." } ] }, { "model_id": "claude-opus-4.8", "benchmark_id": "critpt", "score": 20.9, "reference_url": "https://github.com/MoonshotAI/Kimi-K3/blob/main/k3_tech_report.pdf", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "none", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "max/xhigh; no tools", "prompt_style": "default", "temperature": "default", "context": "default" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Kimi K3 technical report: CritPt = 20.9. Source setting: max/xhigh; no tools. Origin: Artificial Analysis cited by Kimi report." }, { "model_id": "claude-opus-4.8", "benchmark_id": "aa_lcr", "score": 67.7, "reference_url": "https://github.com/MoonshotAI/Kimi-K3/blob/main/k3_tech_report.pdf", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "none", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "max/xhigh; no tools", "prompt_style": "default", "temperature": "default", "context": "default" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Kimi K3 technical report: AA-LCR = 67.7. Source setting: max/xhigh; no tools. Origin: Artificial Analysis cited by Kimi report.", "candidates": [ { "score": 72.2, "reference_url": "https://huggingface.co/tencent/Hy3", "source_type": "model_card", "reported_setting": { "mode": "thinking", "effort": "high / highest available", "tools": "none", "sampling": "repeats=3 per question", "judge": "Equality Checker LLM", "harness": "Artificial Analysis AA-LCR, 100 questions", "prompt_style": "default", "temperature": "0.9 for Hy3 family; otherwise source/provider setting", "context": "256000 for Hy3 family; otherwise source/provider setting", "notes": "Tencent own test" }, "notes": "Hy3 official model-card appendix: AA-LCR = 72.2*. Tencent own testing." } ] }, { "model_id": "claude-opus-4.8", "benchmark_id": "hle", "score": 47.1, "reference_url": "https://www.anthropic.com/claude-opus-5-system-card", "reported_setting": { "effort": "high", "tools": "benchmark-specific" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Figure 8.10.1.B / Claude Opus 4.8 / high: 2,500 questions; 1M token cap; no tools and no context compaction.", "candidates": [ { "score": 49.8, "reference_url": "https://github.com/MoonshotAI/Kimi-K3/blob/main/k3_tech_report.pdf", "source_type": "tech_report", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "none", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "no tools; K3 max/temp=1/top-p=.95", "prompt_style": "default", "temperature": "default", "context": "default" }, "notes": "Kimi K3 technical report: HLE-Full (no tools) = 49.8. Source setting: no tools; K3 max/temp=1/top-p=.95. Origin: Moonshot evaluation." }, { "score": 49.8, "reference_url": "https://www.anthropic.com/claude-sonnet-5-system-card", "source_type": "model_card", "reported_setting": { "effort": "max", "tools": "none", "sampling": "pass@1", "harness": "hle" }, "notes": "Anthropic Claude Sonnet 5 System Card Figure 8.9.1.A, page 122: Claude Opus 4.8; Humanity’s Last Exam [without tools]=49.8%. Source setting: effort=max; tools=none; sampling=pass@1; harness=hle." }, { "score": 42.8, "reference_url": "https://www.anthropic.com/claude-opus-5-system-card", "source_type": "model_card", "reported_setting": { "effort": "low", "tools": "benchmark-specific" }, "notes": "Figure 8.10.1.B / Claude Opus 4.8 / low: 2,500 questions; 1M token cap; no tools and no context compaction." }, { "score": 45.0, "reference_url": "https://www.anthropic.com/claude-opus-5-system-card", "source_type": "model_card", "reported_setting": { "effort": "medium", "tools": "benchmark-specific" }, "notes": "Figure 8.10.1.B / Claude Opus 4.8 / medium: 2,500 questions; 1M token cap; no tools and no context compaction." }, { "score": 49.6, "reference_url": "https://www.anthropic.com/claude-opus-5-system-card", "source_type": "model_card", "reported_setting": { "effort": "xhigh", "tools": "benchmark-specific" }, "notes": "Figure 8.10.1.B / Claude Opus 4.8 / xhigh: 2,500 questions; 1M token cap; no tools and no context compaction." } ] }, { "model_id": "claude-opus-4.8", "benchmark_id": "hle_tools", "score": 55.7, "reference_url": "https://www.anthropic.com/claude-sonnet-5-system-card", "reported_setting": { "effort": "high", "tools": "web search, fetch, and code execution", "sampling": "pass@1", "harness": "hle" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Anthropic Claude Sonnet 5 System Card Figure 8.9.1.B, page 123: Claude Opus 4.8; Humanity’s Last Exam [high]=55.7%. Source setting: effort=high; tools=web search, fetch, and code execution; sampling=pass@1; harness=hle.", "candidates": [ { "score": 57.9, "reference_url": "https://github.com/MoonshotAI/Kimi-K3/blob/main/k3_tech_report.pdf", "source_type": "tech_report", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "general tools", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "general tools; pass@1", "prompt_style": "default", "temperature": "default", "context": "default" }, "notes": "Kimi K3 technical report: HLE-Full (with tools) = 57.9. Source setting: general tools; pass@1. Origin: Moonshot evaluation." }, { "score": 50.2, "reference_url": "https://www.anthropic.com/claude-sonnet-5-system-card", "source_type": "model_card", "reported_setting": { "effort": "low", "tools": "web search, fetch, and code execution", "sampling": "pass@1", "harness": "hle" }, "notes": "Anthropic Claude Sonnet 5 System Card Figure 8.9.1.B, page 123: Claude Opus 4.8; Humanity’s Last Exam [low]=50.2%. Source setting: effort=low; tools=web search, fetch, and code execution; sampling=pass@1; harness=hle." }, { "score": 55.2, "reference_url": "https://www.anthropic.com/claude-sonnet-5-system-card", "source_type": "model_card", "reported_setting": { "effort": "medium", "tools": "web search, fetch, and code execution", "sampling": "pass@1", "harness": "hle" }, "notes": "Anthropic Claude Sonnet 5 System Card Figure 8.9.1.B, page 123: Claude Opus 4.8; Humanity’s Last Exam [med]=55.2%. Source setting: effort=medium; tools=web search, fetch, and code execution; sampling=pass@1; harness=hle." }, { "score": 57.6, "reference_url": "https://www.anthropic.com/claude-sonnet-5-system-card", "source_type": "model_card", "reported_setting": { "effort": "xhigh", "tools": "web search, fetch, and code execution", "sampling": "pass@1", "harness": "hle" }, "notes": "Anthropic Claude Sonnet 5 System Card Figure 8.9.1.B, page 123: Claude Opus 4.8; Humanity’s Last Exam [xhigh]=57.6%. Source setting: effort=xhigh; tools=web search, fetch, and code execution; sampling=pass@1; harness=hle." }, { "score": 58.0, "reference_url": "https://www.anthropic.com/claude-sonnet-5-system-card", "source_type": "model_card", "reported_setting": { "effort": "max", "tools": "web search, fetch, and code execution", "sampling": "pass@1", "harness": "hle" }, "notes": "Anthropic Claude Sonnet 5 System Card Figure 8.9.1.B, page 123: Claude Opus 4.8; Humanity’s Last Exam [max]=58.0%. Source setting: effort=max; tools=web search, fetch, and code execution; sampling=pass@1; harness=hle." }, { "score": 57.9, "reference_url": "https://www.anthropic.com/claude-opus-5-system-card", "source_type": "model_card", "reported_setting": { "effort": "max", "harness": "official or cited leaderboard" }, "notes": "Table 8.1.A / Humanity's Last Exam (With Tools): Unless noted, Anthropic values use adaptive thinking at max effort, default sampling, mean over five trials. Table 8.1.A card summary." }, { "score": 50.2, "reference_url": "https://www.anthropic.com/claude-opus-5-system-card", "source_type": "model_card", "reported_setting": { "effort": "low", "tools": "benchmark-specific" }, "notes": "Figure 8.10.1.A / Claude Opus 4.8 / low: 2,500 questions; 1M token cap; web/programmatic/code tools; no context compaction." }, { "score": 55.2, "reference_url": "https://www.anthropic.com/claude-opus-5-system-card", "source_type": "model_card", "reported_setting": { "effort": "medium", "tools": "benchmark-specific" }, "notes": "Figure 8.10.1.A / Claude Opus 4.8 / medium: 2,500 questions; 1M token cap; web/programmatic/code tools; no context compaction." }, { "score": 57.6, "reference_url": "https://www.anthropic.com/claude-opus-5-system-card", "source_type": "model_card", "reported_setting": { "effort": "xhigh", "tools": "benchmark-specific" }, "notes": "Figure 8.10.1.A / Claude Opus 4.8 / xhigh: 2,500 questions; 1M token cap; web/programmatic/code tools; no context compaction." }, { "score": 58.0, "reference_url": "https://www.anthropic.com/claude-opus-5-system-card", "source_type": "model_card", "reported_setting": { "effort": "max", "tools": "benchmark-specific" }, "notes": "Figure 8.10.1.A / Claude Opus 4.8 / max: 2,500 questions; 1M token cap; web/programmatic/code tools; no context compaction." } ] }, { "model_id": "claude-opus-4.8", "benchmark_id": "deep_swe_v1_1", "score": 51.8, "reference_url": "https://www.anthropic.com/claude-opus-5-system-card", "reported_setting": { "effort": "high", "tools": "benchmark-specific" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Figure 8.3.A / Claude Opus 4.8 / high: 113 tasks; five trials.", "candidates": [ { "score": 58.0, "reference_url": "https://huggingface.co/deepseek-ai/DeepSeek-V4-Flash-0731", "source_type": "model_card", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "benchmark-specified", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "not restated in DeepSeek-V4-Flash-0731 model card", "prompt_style": "default", "temperature": "unknown", "context": "unknown" }, "notes": "DeepSeek-V4-Flash-0731 official model card: Opus-4.8 / DeepSWE = 58.0. Cross-model settings are marked unknown when the source does not restate them." }, { "score": 62.8, "reference_url": "https://huggingface.co/tencent/Hy3", "source_type": "model_card", "reported_setting": { "mode": "thinking", "effort": "high / highest available", "tools": "agentic code execution", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "mini-swe-agent; 2h/task; 2 CPU/8GB; no internet", "prompt_style": "default", "temperature": "0.9 for Hy3 family; otherwise source/provider setting", "context": "256000 for Hy3 family; otherwise source/provider setting", "notes": "Tencent own test" }, "notes": "Hy3 official model-card appendix: DeepSWE = 62.8*. Tencent own testing." }, { "score": 40.8, "reference_url": "https://openai.com/index/gpt-5-6/", "source_type": "official_blog", "reported_setting": { "effort": "low", "eval_variant": null, "code_mode": "false" }, "notes": "OpenAI GPT-5.6 release figure (DeepSWE v1.1): Claude Opus 4.8; DeepSWE v1.1=40.8. Figure 1: DeepSWE v1.1. Cost, latency, and output-token coordinates are retained as review metadata, not score-matrix metrics." }, { "score": 48.67, "reference_url": "https://openai.com/index/gpt-5-6/", "source_type": "official_blog", "reported_setting": { "effort": "medium", "eval_variant": null, "code_mode": "false" }, "notes": "OpenAI GPT-5.6 release figure (DeepSWE v1.1): Claude Opus 4.8; DeepSWE v1.1=48.67. Figure 1: DeepSWE v1.1. Cost, latency, and output-token coordinates are retained as review metadata, not score-matrix metrics." }, { "score": 51.77, "reference_url": "https://openai.com/index/gpt-5-6/", "source_type": "official_blog", "reported_setting": { "effort": "high", "eval_variant": null, "code_mode": "false" }, "notes": "OpenAI GPT-5.6 release figure (DeepSWE v1.1): Claude Opus 4.8; DeepSWE v1.1=51.77. Figure 1: DeepSWE v1.1. Cost, latency, and output-token coordinates are retained as review metadata, not score-matrix metrics." }, { "score": 54.36, "reference_url": "https://openai.com/index/gpt-5-6/", "source_type": "official_blog", "reported_setting": { "effort": "xhigh", "eval_variant": null, "code_mode": "false" }, "notes": "OpenAI GPT-5.6 release figure (DeepSWE v1.1): Claude Opus 4.8; DeepSWE v1.1=54.36. Figure 1: DeepSWE v1.1. Cost, latency, and output-token coordinates are retained as review metadata, not score-matrix metrics." }, { "score": 59.0, "reference_url": "https://github.com/MoonshotAI/Kimi-K3/blob/main/k3_tech_report.pdf", "source_type": "tech_report", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "benchmark-specified", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "agentic coding harness per row", "prompt_style": "default", "temperature": "default", "context": "default" }, "notes": "Kimi K3 technical report: DeepSWE = 59.0. Source setting: agentic coding harness per row. Origin: Moonshot K3 run; competitors from DeepSWE leaderboard / GLM blog." }, { "score": 40.8, "reference_url": "https://www.anthropic.com/claude-opus-5-system-card", "source_type": "model_card", "reported_setting": { "effort": "low", "tools": "benchmark-specific" }, "notes": "Figure 8.3.A / Claude Opus 4.8 / low: 113 tasks; five trials." }, { "score": 48.7, "reference_url": "https://www.anthropic.com/claude-opus-5-system-card", "source_type": "model_card", "reported_setting": { "effort": "medium", "tools": "benchmark-specific" }, "notes": "Figure 8.3.A / Claude Opus 4.8 / medium: 113 tasks; five trials." }, { "score": 54.4, "reference_url": "https://www.anthropic.com/claude-opus-5-system-card", "source_type": "model_card", "reported_setting": { "effort": "xhigh", "tools": "benchmark-specific" }, "notes": "Figure 8.3.A / Claude Opus 4.8 / xhigh: 113 tasks; five trials." }, { "score": 59.0, "reference_url": "https://x.ai/news/grok-4-5", "source_type": "official_blog", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "agentic repository shell/editor", "sampling": "pass@1", "judge": "isolated repository verifier", "harness": "mini-swe-agent run by Datacurve", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown" }, "notes": "release45 DeepSWE 1.1 Pass@1 (%); source effort=max; benchmark protocol matches canonical=True. Unknown source fields remain unknown." } ] }, { "model_id": "claude-opus-4.8", "benchmark_id": "terminal_bench_2_1", "score": 84.6, "reference_url": "https://github.com/MoonshotAI/Kimi-K3/blob/main/k3_tech_report.pdf", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "benchmark-specified", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "agentic coding harness per row", "prompt_style": "default", "temperature": "default", "context": "default" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Kimi K3 technical report: Terminal-Bench 2.1 = 84.6. Source setting: agentic coding harness per row. Origin: mixed official sources documented by Kimi report.", "candidates": [ { "score": 85.0, "reference_url": "https://z.ai/blog/glm-5.2", "source_type": "official_blog", "reported_setting": { "mode": "thinking", "effort": "max (source; model default is high)", "sampling": "pass@1", "prompt_style": "default", "tools": "terminal agent", "temperature": "1.0", "top_p": "1.0", "context": "256000", "max_output_tokens": "48000", "judge": "official task verifier", "harness": "Terminus-2; parser=json; timeout=4h; max_episodes=500; 4 CPU/8GB" }, "notes": "Z.ai Terminus-2 alternative." }, { "score": 78.9, "reference_url": "https://z.ai/blog/glm-5.2", "source_type": "official_blog", "reported_setting": { "mode": "thinking", "effort": "max (source; model default is high)", "sampling": "pass@1", "prompt_style": "default", "tools": "terminal agent", "temperature": "1.0 for Claude Code; unknown otherwise", "top_p": "0.95 for Claude Code; unknown otherwise", "context": "source-reported", "max_output_tokens": "131072 for Claude Code; source-reported otherwise", "judge": "official task verifier", "harness": "Claude Code 2.1.167; Claude Code values averaged over 5 runs" }, "notes": "Z.ai Claude Code alternative." }, { "score": 85.4, "reference_url": "https://huggingface.co/tencent/Hy3", "source_type": "model_card", "reported_setting": { "mode": "thinking", "effort": "high / highest available", "tools": "terminal/code execution", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "Terminus-2; parser=xml; timeout=4h; CPU=16 cores; memory=32GB; max episodes=500", "prompt_style": "default", "temperature": "0.9 for Hy3 family; otherwise source/provider setting", "context": "256000 for Hy3 family; otherwise source/provider setting", "notes": "Tencent own test" }, "notes": "Hy3 official model-card appendix: Terminal-Bench 2.1 = 85.4*. Tencent own testing." }, { "score": 82.7, "reference_url": "https://ai.meta.com/static-resource/muse-spark-1.1-evaluation-report", "source_type": "tech_report", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "terminal", "sampling": "pass@1 unless source row states otherwise", "harness": "Meta MSL general-capability evaluation", "temperature": "1.0 for Muse; provider/source setting for comparisons", "context": "benchmark-specific", "configuration": "default" }, "notes": "Figure 44 · Terminal-Bench 2.1; metric=headline_metric. Exact printed value in the general-capability summary." }, { "score": 78.9, "reference_url": "https://x.ai/news/grok-4-5", "source_type": "official_blog", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "container terminal", "sampling": "pass@1", "judge": "verified task-success evaluator", "harness": "Grok Build for Grok; source-reported peer harnesses", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown" }, "notes": "release45 Terminal-Bench 2.1 Task success rate (%); source effort=max; benchmark protocol matches canonical=True. Unknown source fields remain unknown." } ] }, { "model_id": "claude-opus-4.8", "benchmark_id": "frontier_swe", "score": 66.7, "reference_url": "https://github.com/MoonshotAI/Kimi-K3/blob/main/k3_tech_report.pdf", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "benchmark-specified", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "agentic coding harness per row", "prompt_style": "default", "temperature": "default", "context": "default" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Kimi K3 technical report: FrontierSWE = 66.7. Source setting: agentic coding harness per row. Origin: Moonshot/OpenAI runs plus official leaderboard.", "candidates": [ { "score": 75.1, "reference_url": "https://z.ai/blog/glm-5.2", "source_type": "official_blog", "reported_setting": { "mode": "thinking", "effort": "max (source; model default is high)", "sampling": "pass@1", "prompt_style": "default", "tools": "agentic code execution", "context": "1000000", "max_output_tokens": "128000", "judge": "FrontierSWE dominance", "harness": "Proximal evaluation; max effort; as of 2026-06-16", "temperature": "default" }, "notes": "Z.ai dated 2026-06-16 alternative." }, { "score": 73.0, "reference_url": "https://media.x.ai/v1/website/4p5-5184fdf9.pdf", "source_type": "model_card", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "agentic code execution", "sampling": "mean@5", "judge": "continuous partial-credit task scoring and dominance", "harness": "public FrontierSWE harness; Grok CLI for Grok 4.5", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown" }, "notes": "card45 FrontierSWE Dominance (%); source effort=max; benchmark protocol matches canonical=True. Unknown source fields remain unknown." } ] }, { "model_id": "claude-opus-4.8", "benchmark_id": "swe_marathon_h20_2026_07_09", "score": 40.0, "reference_url": "https://github.com/MoonshotAI/Kimi-K3/blob/main/k3_tech_report.pdf", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "benchmark-specified", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "H20-calibrated pre-final-v1.1 branch (2026-07-09)", "prompt_style": "default", "temperature": "default", "context": "default" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Kimi K3 technical report: SWE-Marathon = 40.0. Source setting: H20-calibrated pre-final-v1.1 branch (2026-07-09). Origin: Moonshot evaluation." }, { "model_id": "claude-opus-4.8", "benchmark_id": "posttrain_bench", "score": 34.1, "reference_url": "https://github.com/MoonshotAI/Kimi-K3/blob/main/k3_tech_report.pdf", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "benchmark-specified", "sampling": "avg-of-3", "judge": "benchmark-specified", "harness": "H20 avg@3 for K3/Fable/Sol; official H100 for Opus/GPT-5.5/GLM", "prompt_style": "default", "temperature": "default", "context": "default" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Kimi K3 technical report: PostTrainBench = 34.1. Source setting: H20 avg@3 for K3/Fable/Sol; official H100 for Opus/GPT-5.5/GLM. Origin: Moonshot H20 runs plus official H100 leaderboard values.", "candidates": [ { "score": 37.2, "reference_url": "https://z.ai/blog/glm-5.2", "source_type": "official_blog", "reported_setting": { "mode": "thinking", "effort": "max (source; model default is high)", "sampling": "pass@1", "prompt_style": "default", "tools": "agentic post-training", "context": "1000000", "max_output_tokens": "128000", "judge": "official weighted benchmark score", "harness": "PostTrainBench evaluation; max effort", "temperature": "default" }, "notes": "Z.ai official-benchmark alternative." } ] }, { "model_id": "claude-opus-4.8", "benchmark_id": "scicode", "score": 53.5, "reference_url": "https://github.com/MoonshotAI/Kimi-K3/blob/main/k3_tech_report.pdf", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "benchmark-specified", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "agentic coding harness per row", "prompt_style": "default", "temperature": "default", "context": "default" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Kimi K3 technical report: SciCode = 53.5. Source setting: agentic coding harness per row. Origin: Artificial Analysis cited by Kimi report.", "candidates": [ { "score": 56.21, "reference_url": "https://huggingface.co/internlm/Intern-S2-Preview-397B", "source_type": "model_card", "reported_setting": { "mode": "source does not state", "effort": "source does not state", "tools": "code execution", "sampling": "source does not state", "judge": "benchmark-specific; exact version not stated", "harness": "AgentCompass / benchmark evaluator", "prompt_style": "source does not state", "temperature": "source does not state", "context": "source does not state", "notes": "Shanghai AI Lab evaluated all model columns using OpenCompass, VLMEvalKit, and AgentCompass. Per-model mode/effort/sampling are not reported; only Intern-S2 maximum evaluation lengths are disclosed." }, "notes": "Intern-S2-Preview-397B official model card: SciCode = 56.21. This is a Shanghai AI Lab cross-model evaluation; the source does not disclose this model's exact mode, effort, or sampling." } ] }, { "model_id": "claude-opus-4.8", "benchmark_id": "browsecomp", "score": 82.0, "reference_url": "https://www.anthropic.com/claude-opus-5-system-card", "reported_setting": { "effort": "high", "tools": "benchmark-specific" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Figure 8.10.2.B / Claude Opus 4.8 / high: Fixed 10M token budget with web/programmatic/code tools and context compaction.", "candidates": [ { "score": 84.4, "reference_url": "https://openai.com/index/gpt-5-6/", "source_type": "official_blog", "reported_setting": { "effort": "reference", "eval_variant": null, "code_mode": null }, "notes": "OpenAI GPT-5.6 release figure (BrowseComp): Claude Opus 4.8; BrowseComp=84.4. Figure 15: Agentic browsing tasks.. Cost, latency, and output-token coordinates are retained as review metadata, not score-matrix metrics." }, { "score": 84.3, "reference_url": "https://github.com/MoonshotAI/Kimi-K3/blob/main/k3_tech_report.pdf", "source_type": "tech_report", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "benchmark-specified", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "agentic benchmark harness", "prompt_style": "default", "temperature": "default", "context": "default" }, "notes": "Kimi K3 technical report: BrowseComp = 84.3. Source setting: agentic benchmark harness. Origin: Moonshot K3 run; competitor provider sources." }, { "score": 80.6, "reference_url": "https://www.anthropic.com/claude-sonnet-5-system-card", "source_type": "model_card", "reported_setting": { "effort": "max", "tools": "web search, fetch, programmatic tools, code execution", "sampling": "pass@1", "harness": "browse", "token_cap": "1M" }, "notes": "Anthropic Claude Sonnet 5 System Card Figure 8.9.2.A, page 124: Claude Opus 4.8; BrowseComp [1M]=80.6%. Source setting: effort=max; tools=web search, fetch, programmatic tools, code execution; sampling=pass@1; harness=browse; token_cap=1M." }, { "score": 84.0, "reference_url": "https://www.anthropic.com/claude-sonnet-5-system-card", "source_type": "model_card", "reported_setting": { "effort": "max", "tools": "web search, fetch, programmatic tools, code execution", "sampling": "pass@1", "harness": "browse", "token_cap": "3M" }, "notes": "Anthropic Claude Sonnet 5 System Card Figure 8.9.2.A, page 124: Claude Opus 4.8; BrowseComp [3M]=84.0%. Source setting: effort=max; tools=web search, fetch, programmatic tools, code execution; sampling=pass@1; harness=browse; token_cap=3M." }, { "score": 84.3, "reference_url": "https://www.anthropic.com/claude-sonnet-5-system-card", "source_type": "model_card", "reported_setting": { "effort": "max", "tools": "web search, fetch, programmatic tools, code execution", "sampling": "pass@1", "harness": "browse", "token_cap": "10M" }, "notes": "Anthropic Claude Sonnet 5 System Card Figure 8.9.2.A, page 124: Claude Opus 4.8; BrowseComp [10M]=84.3%. Source setting: effort=max; tools=web search, fetch, programmatic tools, code execution; sampling=pass@1; harness=browse; token_cap=10M." }, { "score": 77.4, "reference_url": "https://www.anthropic.com/claude-opus-5-system-card", "source_type": "model_card", "reported_setting": { "effort": "low", "tools": "benchmark-specific" }, "notes": "Figure 8.10.2.B / Claude Opus 4.8 / low: Fixed 10M token budget with web/programmatic/code tools and context compaction." }, { "score": 79.1, "reference_url": "https://www.anthropic.com/claude-opus-5-system-card", "source_type": "model_card", "reported_setting": { "effort": "medium", "tools": "benchmark-specific" }, "notes": "Figure 8.10.2.B / Claude Opus 4.8 / medium: Fixed 10M token budget with web/programmatic/code tools and context compaction." }, { "score": 84.0, "reference_url": "https://www.anthropic.com/claude-opus-5-system-card", "source_type": "model_card", "reported_setting": { "effort": "xhigh", "tools": "benchmark-specific" }, "notes": "Figure 8.10.2.B / Claude Opus 4.8 / xhigh: Fixed 10M token budget with web/programmatic/code tools and context compaction." }, { "score": 80.6, "reference_url": "https://www.anthropic.com/claude-opus-5-system-card", "source_type": "model_card", "reported_setting": { "effort": "max", "tools": "web/programmatic/code", "harness": "token-budget scaling" }, "notes": "Figure 8.10.2.A / Claude Opus 4.8 / 1M: Token-budget scaling; source reports exact labels at 1M, 3M, and 10M. Budget variants are candidates; fixed 10M effort ladder supplies the preferred canonical-setting value." }, { "score": 84.0, "reference_url": "https://www.anthropic.com/claude-opus-5-system-card", "source_type": "model_card", "reported_setting": { "effort": "max", "tools": "web/programmatic/code", "harness": "token-budget scaling" }, "notes": "Figure 8.10.2.A / Claude Opus 4.8 / 3M: Token-budget scaling; source reports exact labels at 1M, 3M, and 10M. Budget variants are candidates; fixed 10M effort ladder supplies the preferred canonical-setting value." } ] }, { "model_id": "claude-opus-4.8", "benchmark_id": "deepsearchqa_f1", "score": 91.0, "reference_url": "https://www.anthropic.com/claude-opus-5-system-card", "reported_setting": { "effort": "high", "tools": "benchmark-specific" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Figure 8.10.3.A / Claude Opus 4.8 / high: Mean F1 over 900 prompts in 17 fields; web tools; nominal 1M token budget.", "candidates": [ { "score": 93.1, "reference_url": "https://github.com/MoonshotAI/Kimi-K3/blob/main/k3_tech_report.pdf", "source_type": "tech_report", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "benchmark-specified", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "agentic benchmark harness", "prompt_style": "default", "temperature": "default", "context": "default" }, "notes": "Kimi K3 technical report: DeepSearchQA (F1) = 93.1. Source setting: agentic benchmark harness. Origin: Moonshot evaluation." }, { "score": 88.6, "reference_url": "https://www.anthropic.com/claude-opus-5-system-card", "source_type": "model_card", "reported_setting": { "effort": "low", "tools": "benchmark-specific" }, "notes": "Figure 8.10.3.A / Claude Opus 4.8 / low: Mean F1 over 900 prompts in 17 fields; web tools; nominal 1M token budget." }, { "score": 90.1, "reference_url": "https://www.anthropic.com/claude-opus-5-system-card", "source_type": "model_card", "reported_setting": { "effort": "medium", "tools": "benchmark-specific" }, "notes": "Figure 8.10.3.A / Claude Opus 4.8 / medium: Mean F1 over 900 prompts in 17 fields; web tools; nominal 1M token budget." }, { "score": 92.1, "reference_url": "https://www.anthropic.com/claude-opus-5-system-card", "source_type": "model_card", "reported_setting": { "effort": "xhigh", "tools": "benchmark-specific" }, "notes": "Figure 8.10.3.A / Claude Opus 4.8 / xhigh: Mean F1 over 900 prompts in 17 fields; web tools; nominal 1M token budget." }, { "score": 84.3, "reference_url": "https://ai.meta.com/static-resource/muse-spark-1.1-evaluation-report", "source_type": "tech_report", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "search/browser", "sampling": "pass@1 unless source row states otherwise", "harness": "Meta MSL general-capability evaluation", "temperature": "1.0 for Muse; provider/source setting for comparisons", "context": "benchmark-specific", "configuration": "default" }, "notes": "Figure 44 · DeepSearchQA F1; metric=headline_metric. Exact printed value in the general-capability summary." } ] }, { "model_id": "claude-opus-4.8", "benchmark_id": "researchrubrics", "score": 73.5, "reference_url": "https://github.com/MoonshotAI/Kimi-K3/blob/main/k3_tech_report.pdf", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "benchmark-specified", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "agentic benchmark harness", "prompt_style": "default", "temperature": "default", "context": "default" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Kimi K3 technical report: ResearchRubrics = 73.5. Source setting: agentic benchmark harness. Origin: Moonshot evaluation." }, { "model_id": "claude-opus-4.8", "benchmark_id": "gdpval_aa_elo", "score": 1593.0, "reference_url": "https://github.com/MoonshotAI/Kimi-K3/blob/main/k3_tech_report.pdf", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "benchmark-specified", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "agentic benchmark harness", "prompt_style": "default", "temperature": "default", "context": "default" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Kimi K3 technical report: GDPval-AA v2 (Elo) = 1593.0. Source setting: agentic benchmark harness. Origin: Artificial Analysis cited by Kimi report.", "candidates": [ { "score": 1600.1, "reference_url": "https://openai.com/index/gpt-5-6/", "source_type": "official_blog", "reported_setting": { "effort": "max", "eval_variant": null, "code_mode": "false" }, "notes": "OpenAI GPT-5.6 release table (table 0): Claude Opus 4.8; GDPval-AA v2=1600.1. OpenAI row explicitly identifies GDPval-AA v2. Exact effort/variant resolved from the rendered chart point." }, { "score": 1615.0, "reference_url": "https://www.anthropic.com/claude-sonnet-5-system-card", "source_type": "model_card", "reported_setting": { "effort": "source does not state", "tools": "none", "sampling": "pass@1", "harness": "gdpval" }, "notes": "Anthropic Claude Sonnet 5 System Card Section 8.11.4, page 134: Claude Opus 4.8; GDPval-AA v2 [summary]=1615. Source setting: effort=source does not state; tools=none; sampling=pass@1; harness=gdpval." } ] }, { "model_id": "claude-opus-4.8", "benchmark_id": "toolathlon_verified", "score": 76.2, "reference_url": "https://github.com/MoonshotAI/Kimi-K3/blob/main/k3_tech_report.pdf", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "benchmark-specified", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "Verified edition; official leaderboard snapshot", "prompt_style": "default", "temperature": "default", "context": "default" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Kimi K3 technical report: Toolathlon-Verified = 76.2. Source setting: Verified edition; official leaderboard snapshot. Origin: official Toolathlon leaderboard cited by Kimi report.", "candidates": [ { "score": 79.9, "reference_url": "https://www.anthropic.com/claude-opus-5-system-card", "source_type": "model_card", "reported_setting": { "effort": "max", "tools": "600+ tools across 32 applications" }, "notes": "Table 8.13.6.A / Claude Opus 4.8 / Pass@1: 108 tasks; three trials; internal table excludes null attempts. Opus 4.8 and Sonnet 5 internal values are candidates because official published leaderboard values include null attempts." } ] }, { "model_id": "claude-opus-4.8", "benchmark_id": "automation_bench", "score": 27.2, "reference_url": "https://github.com/MoonshotAI/Kimi-K3/blob/main/k3_tech_report.pdf", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "benchmark-specified", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "agentic benchmark harness", "prompt_style": "default", "temperature": "default", "context": "default" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Kimi K3 technical report: AutomationBench = 27.2. Source setting: agentic benchmark harness. Origin: Moonshot evaluation.", "candidates": [ { "score": 15.5, "reference_url": "https://openai.com/index/gpt-5-6/", "source_type": "official_blog", "reported_setting": { "effort": "max", "eval_variant": null, "code_mode": "false" }, "notes": "OpenAI GPT-5.6 release table (table 8): Claude Opus 4.8; AutomationBench=15.5. Exact effort/variant resolved from the rendered chart point." } ] }, { "model_id": "claude-opus-4.8", "benchmark_id": "job_bench", "score": 48.4, "reference_url": "https://github.com/MoonshotAI/Kimi-K3/blob/main/k3_tech_report.pdf", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "benchmark-specified", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "agentic benchmark harness", "prompt_style": "default", "temperature": "default", "context": "default" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Kimi K3 technical report: JobBench = 48.4. Source setting: agentic benchmark harness. Origin: official JobBench leaderboard cited by Kimi report." }, { "model_id": "claude-opus-4.8", "benchmark_id": "aa_briefcase_elo", "score": 1354.0, "reference_url": "https://github.com/MoonshotAI/Kimi-K3/blob/main/k3_tech_report.pdf", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "benchmark-specified", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "agentic benchmark harness", "prompt_style": "default", "temperature": "default", "context": "default" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Kimi K3 technical report: AA-Briefcase (Elo) = 1354.0. Source setting: agentic benchmark harness. Origin: Artificial Analysis cited by Kimi report.", "candidates": [ { "score": 1352.0, "reference_url": "https://www.anthropic.com/claude-sonnet-5-system-card", "source_type": "model_card", "reported_setting": { "effort": "source does not state", "tools": "none", "sampling": "pass@1", "harness": "briefcase" }, "notes": "Anthropic Claude Sonnet 5 System Card Section 8.11.7, page 137: Claude Opus 4.8; AA-Briefcase [summary]=1352. Source setting: effort=source does not state; tools=none; sampling=pass@1; harness=briefcase." }, { "score": 1346, "reference_url": "https://www.anthropic.com/claude-opus-5-system-card", "source_type": "model_card", "reported_setting": { "effort": "max", "harness": "official or cited leaderboard" }, "notes": "Table 8.1.A / AA-Briefcase: Unless noted, Anthropic values use adaptive thinking at max effort, default sampling, mean over five trials. Table 8.1.A card summary." }, { "score": 1340, "reference_url": "https://media.x.ai/v1/website/card-4p6-4cd2dc57.pdf", "source_type": "model_card", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "offline multi-file knowledge-work tools", "sampling": "unknown", "judge": "rubric plus analytical and presentation pairwise Elo", "harness": "Artificial Analysis AA-Briefcase harness", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown" }, "notes": "card46 AA Briefcase AA-Briefcase Elo; source effort=max; benchmark protocol matches canonical=True. Unknown source fields remain unknown." } ] }, { "model_id": "claude-opus-4.8", "benchmark_id": "agents_last_exam", "score": 27.0, "reference_url": "https://github.com/MoonshotAI/Kimi-K3/blob/main/k3_tech_report.pdf", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "benchmark-specified", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "agentic benchmark harness", "prompt_style": "default", "temperature": "default", "context": "default" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Kimi K3 technical report: Agents' Last Exam = 27.0. Source setting: agentic benchmark harness. Origin: official leaderboard cited by Kimi report.", "candidates": [ { "score": 25.7, "reference_url": "https://huggingface.co/deepseek-ai/DeepSeek-V4-Flash-0731", "source_type": "model_card", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "benchmark-specified", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "not restated in DeepSeek-V4-Flash-0731 model card", "prompt_style": "default", "temperature": "unknown", "context": "unknown" }, "notes": "DeepSeek-V4-Flash-0731 official model card: Opus-4.8 / Agents' Last Exam = 25.7. Cross-model settings are marked unknown when the source does not restate them." }, { "score": 45.2, "reference_url": "https://openai.com/index/gpt-5-6/", "source_type": "official_blog", "reported_setting": { "effort": "max", "eval_variant": null, "code_mode": "true" }, "notes": "OpenAI GPT-5.6 release table (table 0): Claude Opus 4.8; Agents' Last Exam=45.2. Exact effort/variant resolved from the rendered chart point." }, { "score": 37.4, "reference_url": "https://openai.com/index/gpt-5-6/", "source_type": "official_blog", "reported_setting": { "effort": "low", "eval_variant": null, "code_mode": "true" }, "notes": "OpenAI GPT-5.6 release figure (Agents' Last Exam): Claude Opus 4.8; Agents' Last Exam=37.4. Figure 0: Long-horizon agentic workflows across professional domains.. Cost, latency, and output-token coordinates are retained as review metadata, not score-matrix metrics." }, { "score": 37.8, "reference_url": "https://openai.com/index/gpt-5-6/", "source_type": "official_blog", "reported_setting": { "effort": "medium", "eval_variant": null, "code_mode": "true" }, "notes": "OpenAI GPT-5.6 release figure (Agents' Last Exam): Claude Opus 4.8; Agents' Last Exam=37.8. Figure 0: Long-horizon agentic workflows across professional domains.. Cost, latency, and output-token coordinates are retained as review metadata, not score-matrix metrics." }, { "score": 38.7, "reference_url": "https://openai.com/index/gpt-5-6/", "source_type": "official_blog", "reported_setting": { "effort": "high", "eval_variant": null, "code_mode": "true" }, "notes": "OpenAI GPT-5.6 release figure (Agents' Last Exam): Claude Opus 4.8; Agents' Last Exam=38.7. Figure 0: Long-horizon agentic workflows across professional domains.. Cost, latency, and output-token coordinates are retained as review metadata, not score-matrix metrics." }, { "score": 42.3, "reference_url": "https://openai.com/index/gpt-5-6/", "source_type": "official_blog", "reported_setting": { "effort": "xhigh", "eval_variant": null, "code_mode": "true" }, "notes": "OpenAI GPT-5.6 release figure (Agents' Last Exam): Claude Opus 4.8; Agents' Last Exam=42.3. Figure 0: Long-horizon agentic workflows across professional domains.. Cost, latency, and output-token coordinates are retained as review metadata, not score-matrix metrics." } ] }, { "model_id": "claude-opus-4.8", "benchmark_id": "apex_agents", "score": 39.4, "reference_url": "https://github.com/MoonshotAI/Kimi-K3/blob/main/k3_tech_report.pdf", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "benchmark-specified", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "agentic benchmark harness", "prompt_style": "default", "temperature": "default", "context": "default" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Kimi K3 technical report: APEX-Agents = 39.4. Source setting: agentic benchmark harness. Origin: Artificial Analysis cited by Kimi report.", "candidates": [ { "score": 42.5, "reference_url": "https://huggingface.co/tencent/Hy3", "source_type": "model_card", "reported_setting": { "mode": "thinking", "effort": "high / highest available", "tools": "multiple application/software tools", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "APEX-Agents long-horizon professional task harness", "prompt_style": "default", "temperature": "0.9 for Hy3 family; otherwise source/provider setting", "context": "256000 for Hy3 family; otherwise source/provider setting", "notes": "Comparator value not marked as Tencent own test" }, "notes": "Hy3 official model-card appendix: Apex-Agent (pass@1) = 42.5. Comparator-reported value." }, { "score": 56.2, "reference_url": "https://media.x.ai/v1/website/card-4p6-4cd2dc57.pdf", "source_type": "model_card", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "multi-application professional tools", "sampling": "pass@1", "judge": "expert all-criteria-pass rubric", "harness": "Mercor APEX-Agents harness", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown" }, "notes": "card46 APEX-Agents Pass@1 (%); source effort=max; benchmark protocol matches canonical=True. Unknown source fields remain unknown." } ] }, { "model_id": "claude-opus-4.8", "benchmark_id": "officeqa_pro", "score": 63.9, "reference_url": "https://github.com/MoonshotAI/Kimi-K3/blob/main/k3_tech_report.pdf", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "benchmark-specified", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "PDF corpus rendered as images; no machine-readable text", "prompt_style": "default", "temperature": "default", "context": "default" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Kimi K3 technical report: OfficeQA Pro = 63.9. Source setting: PDF corpus rendered as images; no machine-readable text. Origin: Moonshot evaluation.", "candidates": [ { "score": 66.2, "reference_url": "https://www.anthropic.com/claude-sonnet-5-system-card", "source_type": "model_card", "reported_setting": { "effort": "source does not state", "tools": "agentic benchmark harness", "sampling": "pass@1", "harness": "office" }, "notes": "Anthropic Claude Sonnet 5 System Card Figure 8.11.1.A, page 131: Claude Opus 4.8; OfficeQA / OfficeQA Pro [OfficeQA Pro]=66.2. Source setting: effort=source does not state; tools=agentic benchmark harness; sampling=pass@1; harness=office." }, { "score": 66.2, "reference_url": "https://www.anthropic.com/claude-opus-5-system-card", "source_type": "model_card", "reported_setting": { "effort": "max" }, "notes": "OfficeQA Pro narrative / Claude Opus 4.8: 133-question subset; public Messages API for Opus 5." } ] }, { "model_id": "claude-opus-4.8", "benchmark_id": "spreadsheetbench_2", "score": 31.6, "reference_url": "https://github.com/MoonshotAI/Kimi-K3/blob/main/k3_tech_report.pdf", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "benchmark-specified", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "agentic benchmark harness", "prompt_style": "default", "temperature": "default", "context": "default" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Kimi K3 technical report: SpreadsheetBench 2 = 31.6. Source setting: agentic benchmark harness. Origin: Moonshot evaluation." }, { "model_id": "claude-opus-4.8", "benchmark_id": "osworld_verified", "score": 83.4, "reference_url": "https://www.anthropic.com/claude-sonnet-5-system-card", "reported_setting": { "effort": "max", "tools": "GUI computer-use harness", "sampling": "avg 5 trials unless source states otherwise", "harness": "osworld" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Anthropic Claude Sonnet 5 System Card Figure 8.10.2.A, page 126: Claude Opus 4.8; OSWorld-Verified [summary]=83.4%. Source setting: effort=max; tools=GUI computer-use harness; sampling=avg 5 trials unless source states otherwise; harness=osworld.", "candidates": [ { "score": 83.4, "reference_url": "https://github.com/MoonshotAI/Kimi-K3/blob/main/k3_tech_report.pdf", "source_type": "tech_report", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "benchmark-specified", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "agentic benchmark harness", "prompt_style": "default", "temperature": "default", "context": "default" }, "notes": "Kimi K3 technical report: OSWorld-Verified = 83.4. Source setting: agentic benchmark harness. Origin: Moonshot evaluation." } ] }, { "model_id": "claude-opus-4.8", "benchmark_id": "osworld_2_0", "score": 55.7, "reference_url": "https://github.com/MoonshotAI/Kimi-K3/blob/main/k3_tech_report.pdf", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "benchmark-specified", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "agentic benchmark harness", "prompt_style": "default", "temperature": "default", "context": "default" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Kimi K3 technical report: OSWorld 2.0 = 55.7. Source setting: agentic benchmark harness. Origin: Moonshot evaluation.", "candidates": [ { "score": 54.8, "reference_url": "https://openai.com/index/gpt-5-6/", "source_type": "official_blog", "reported_setting": { "effort": "max", "eval_variant": null, "code_mode": "false" }, "notes": "OpenAI GPT-5.6 release table (table 3): Claude Opus 4.8; OSWorld 2.0=54.8. Exact effort/variant resolved from the rendered chart point." }, { "score": 47.3, "reference_url": "https://openai.com/index/gpt-5-6/", "source_type": "official_blog", "reported_setting": { "effort": "low", "eval_variant": null, "code_mode": "false" }, "notes": "OpenAI GPT-5.6 release figure (OSWorld 2.0): Claude Opus 4.8; OSWorld 2.0=47.3. Figure 15: OSWorld 2.0. Cost, latency, and output-token coordinates are retained as review metadata, not score-matrix metrics." }, { "score": 48.6, "reference_url": "https://openai.com/index/gpt-5-6/", "source_type": "official_blog", "reported_setting": { "effort": "medium", "eval_variant": null, "code_mode": "false" }, "notes": "OpenAI GPT-5.6 release figure (OSWorld 2.0): Claude Opus 4.8; OSWorld 2.0=48.6. Figure 15: OSWorld 2.0. Cost, latency, and output-token coordinates are retained as review metadata, not score-matrix metrics." }, { "score": 49.0, "reference_url": "https://openai.com/index/gpt-5-6/", "source_type": "official_blog", "reported_setting": { "effort": "high", "eval_variant": null, "code_mode": "false" }, "notes": "OpenAI GPT-5.6 release figure (OSWorld 2.0): Claude Opus 4.8; OSWorld 2.0=49.0. Figure 15: OSWorld 2.0. Cost, latency, and output-token coordinates are retained as review metadata, not score-matrix metrics." }, { "score": 49.7, "reference_url": "https://openai.com/index/gpt-5-6/", "source_type": "official_blog", "reported_setting": { "effort": "xhigh", "eval_variant": null, "code_mode": "false" }, "notes": "OpenAI GPT-5.6 release figure (OSWorld 2.0): Claude Opus 4.8; OSWorld 2.0=49.7. Figure 15: OSWorld 2.0. Cost, latency, and output-token coordinates are retained as review metadata, not score-matrix metrics." } ] }, { "model_id": "claude-opus-4.8", "benchmark_id": "saas_bench", "score": 56.1, "reference_url": "https://github.com/MoonshotAI/Kimi-K3/blob/main/k3_tech_report.pdf", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "benchmark-specified", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "agentic benchmark harness", "prompt_style": "default", "temperature": "default", "context": "default" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Kimi K3 technical report: SaaS-Bench = 56.1. Source setting: agentic benchmark harness. Origin: Moonshot evaluation." }, { "model_id": "claude-opus-4.8", "benchmark_id": "tau3_banking", "score": 27.6, "reference_url": "https://github.com/MoonshotAI/Kimi-K3/blob/main/k3_tech_report.pdf", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "benchmark-specified", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "agentic benchmark harness", "prompt_style": "default", "temperature": "default", "context": "default" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Kimi K3 technical report: τ³-Banking = 27.6. Source setting: agentic benchmark harness. Origin: Artificial Analysis cited by Kimi report.", "candidates": [ { "score": 27.6, "reference_url": "https://media.x.ai/v1/website/4p5-5184fdf9.pdf", "source_type": "model_card", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "banking APIs and simulated user", "sampling": "pass@1", "judge": "final-state task accuracy", "harness": "Artificial Analysis tau3-banking harness", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown" }, "notes": "card45 tau3-banking Accuracy (%); source effort=max; benchmark protocol matches canonical=True. Unknown source fields remain unknown." } ] }, { "model_id": "claude-opus-4.8", "benchmark_id": "harvey_lab_aa", "score": 91.1, "reference_url": "https://github.com/MoonshotAI/Kimi-K3/blob/main/k3_tech_report.pdf", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "benchmark-specified", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "agentic benchmark harness", "prompt_style": "default", "temperature": "default", "context": "default" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Kimi K3 technical report: Harvey Lab-AA = 91.1. Source setting: agentic benchmark harness. Origin: Artificial Analysis cited by Kimi report." }, { "model_id": "claude-opus-4.8", "benchmark_id": "corpfin_v2", "score": 66.7, "reference_url": "https://github.com/MoonshotAI/Kimi-K3/blob/main/k3_tech_report.pdf", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "benchmark-specified", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "agentic benchmark harness", "prompt_style": "default", "temperature": "default", "context": "default" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Kimi K3 technical report: CorpFin v2 = 66.7. Source setting: agentic benchmark harness. Origin: Vals AI cited by Kimi report." }, { "model_id": "claude-opus-4.8", "benchmark_id": "finance_agent_v2", "score": 53.9, "reference_url": "https://github.com/MoonshotAI/Kimi-K3/blob/main/k3_tech_report.pdf", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "benchmark-specified", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "agentic benchmark harness", "prompt_style": "default", "temperature": "default", "context": "default" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Kimi K3 technical report: Finance Agent v2 = 53.9. Source setting: agentic benchmark harness. Origin: Vals AI cited by Kimi report." }, { "model_id": "claude-opus-4.8", "benchmark_id": "legal_research_bench", "score": 43.8, "reference_url": "https://github.com/MoonshotAI/Kimi-K3/blob/main/k3_tech_report.pdf", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "benchmark-specified", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "agentic benchmark harness", "prompt_style": "default", "temperature": "default", "context": "default" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Kimi K3 technical report: Legal Research Bench = 43.8. Source setting: agentic benchmark harness. Origin: Vals AI cited by Kimi report." }, { "model_id": "claude-opus-4.8", "benchmark_id": "worldvqa", "score": 39.1, "reference_url": "https://github.com/MoonshotAI/Kimi-K3/blob/main/k3_tech_report.pdf", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "none", "sampling": "avg-of-3", "judge": "benchmark-specified", "harness": "forced-answer prompt; avg 3 runs", "prompt_style": "default", "temperature": "default", "context": "default" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Kimi K3 technical report: WorldVQA ForceAnswer = 39.1. Source setting: forced-answer prompt; avg 3 runs. Origin: Moonshot evaluation." }, { "model_id": "claude-opus-4.8", "benchmark_id": "omnidocbench", "score": 0.121, "reference_url": "https://github.com/MoonshotAI/Kimi-K3/blob/main/k3_tech_report.pdf", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "benchmark-specified", "sampling": "avg-of-3", "judge": "benchmark-specified", "harness": "avg 3 runs; source 1-NED score normalized to NED", "prompt_style": "default", "temperature": "default", "context": "default" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Kimi K3 technical report: OmniDocBench = 0.121 NED. Source setting: avg 3 runs; source 1-NED score normalized to NED. Origin: Moonshot evaluation. The source reports 1-NED=87.9%; stored value is normalized NED=0.121." }, { "model_id": "claude-opus-4.8", "benchmark_id": "perception_bench", "score": 47.2, "reference_url": "https://github.com/MoonshotAI/Kimi-K3/blob/main/k3_tech_report.pdf", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "benchmark-specified", "sampling": "avg-of-3", "judge": "benchmark-specified", "harness": "multimodal; avg 3 runs", "prompt_style": "default", "temperature": "default", "context": "default" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Kimi K3 technical report: PerceptionBench = 47.2. Source setting: multimodal; avg 3 runs. Origin: Moonshot evaluation." }, { "model_id": "claude-opus-4.8", "benchmark_id": "video_mme", "score": 86.0, "reference_url": "https://github.com/MoonshotAI/Kimi-K3/blob/main/k3_tech_report.pdf", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "none", "sampling": "avg-of-3", "judge": "benchmark-specified", "harness": "audio+visual+subtitles; avg 3 runs", "prompt_style": "default", "temperature": "default", "context": "default" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Kimi K3 technical report: Video-MME (w. sub) = 86.0. Source setting: audio+visual+subtitles; avg 3 runs. Origin: Moonshot evaluation." }, { "model_id": "claude-opus-4.8", "benchmark_id": "mmvu", "score": 79.2, "reference_url": "https://github.com/MoonshotAI/Kimi-K3/blob/main/k3_tech_report.pdf", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "benchmark-specified", "sampling": "avg-of-3", "judge": "benchmark-specified", "harness": "avg 3 runs", "prompt_style": "default", "temperature": "default", "context": "default" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Kimi K3 technical report: MMVU = 79.2. Source setting: avg 3 runs. Origin: Moonshot evaluation." }, { "model_id": "claude-opus-4.8", "benchmark_id": "babyvision", "score": 81.2, "reference_url": "https://github.com/MoonshotAI/Kimi-K3/blob/main/k3_tech_report.pdf", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "python", "sampling": "avg-of-3", "judge": "benchmark-specified", "harness": "Python tool; avg 3 runs", "prompt_style": "default", "temperature": "default", "context": "default" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Kimi K3 technical report: BabyVision w/ python = 81.2. Source setting: Python tool; avg 3 runs. Origin: Moonshot evaluation." }, { "model_id": "claude-opus-4.8", "benchmark_id": "mmmu_pro", "score": 78.9, "reference_url": "https://github.com/MoonshotAI/Kimi-K3/blob/main/k3_tech_report.pdf", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "none", "sampling": "avg-of-3", "judge": "benchmark-specified", "harness": "vision; avg 3 runs", "prompt_style": "default", "temperature": "default", "context": "default" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Kimi K3 technical report: MMMU-Pro (no Python) = 78.9. Source setting: vision; avg 3 runs. Origin: Moonshot evaluation.", "candidates": [ { "score": 82.7, "reference_url": "https://github.com/MoonshotAI/Kimi-K3/blob/main/k3_tech_report.pdf", "source_type": "tech_report", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "python", "sampling": "avg-of-3", "judge": "benchmark-specified", "harness": "Python tool; avg 3 runs", "prompt_style": "default", "temperature": "default", "context": "default" }, "notes": "Kimi K3 technical report: MMMU-Pro (with Python) = 82.7. Source setting: Python tool; avg 3 runs. Origin: Moonshot evaluation." }, { "score": 76.88, "reference_url": "https://huggingface.co/internlm/Intern-S2-Preview-397B", "source_type": "model_card", "reported_setting": { "mode": "source does not state", "effort": "source does not state", "tools": "none", "sampling": "source does not state", "judge": "benchmark-specific; exact version not stated", "harness": "VLMEvalKit or AgentCompass; exact adapter not stated", "prompt_style": "source does not state", "temperature": "source does not state", "context": "source does not state", "notes": "Shanghai AI Lab evaluated all model columns using OpenCompass, VLMEvalKit, and AgentCompass. Per-model mode/effort/sampling are not reported; only Intern-S2 maximum evaluation lengths are disclosed." }, "notes": "Intern-S2-Preview-397B official model card: MMMU Pro = 76.88. This is a Shanghai AI Lab cross-model evaluation; the source does not disclose this model's exact mode, effort, or sampling." } ] }, { "model_id": "claude-opus-4.8", "benchmark_id": "charxiv_reasoning", "score": 80.5, "reference_url": "https://www.anthropic.com/claude-sonnet-5-system-card", "reported_setting": { "effort": "max", "tools": "none", "sampling": "avg 5 trials unless source states otherwise", "harness": "vision" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Anthropic Claude Sonnet 5 System Card Figure 8.10.5.A, page 130: Claude Opus 4.8; CharXiv Reasoning [no tools]=80.5. Source setting: effort=max; tools=none; sampling=avg 5 trials unless source states otherwise; harness=vision.", "candidates": [ { "score": 89.9, "reference_url": "https://github.com/MoonshotAI/Kimi-K3/blob/main/k3_tech_report.pdf", "source_type": "tech_report", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "python", "sampling": "avg-of-3", "judge": "benchmark-specified", "harness": "Python tool; avg 3 runs", "prompt_style": "default", "temperature": "default", "context": "default" }, "notes": "Kimi K3 technical report: CharXiv (RQ) (with Python) = 89.9. Source setting: Python tool; avg 3 runs. Origin: Moonshot evaluation." }, { "score": 80.5, "reference_url": "https://github.com/MoonshotAI/Kimi-K3/blob/main/k3_tech_report.pdf", "source_type": "tech_report", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "none", "sampling": "avg-of-3", "judge": "benchmark-specified", "harness": "vision; avg 3 runs", "prompt_style": "default", "temperature": "default", "context": "default" }, "notes": "Kimi K3 technical report: CharXiv (RQ) (no Python) = 80.5. Source setting: vision; avg 3 runs. Origin: Moonshot evaluation." } ] }, { "model_id": "claude-opus-4.8", "benchmark_id": "mathvision", "score": 86.7, "reference_url": "https://github.com/MoonshotAI/Kimi-K3/blob/main/k3_tech_report.pdf", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "none", "sampling": "avg-of-3", "judge": "benchmark-specified", "harness": "vision; avg 3 runs", "prompt_style": "default", "temperature": "default", "context": "default" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Kimi K3 technical report: MathVision (no Python) = 86.7. Source setting: vision; avg 3 runs. Origin: Moonshot evaluation.", "candidates": [ { "score": 97.1, "reference_url": "https://github.com/MoonshotAI/Kimi-K3/blob/main/k3_tech_report.pdf", "source_type": "tech_report", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "python", "sampling": "avg-of-3", "judge": "benchmark-specified", "harness": "Python tool; avg 3 runs", "prompt_style": "default", "temperature": "default", "context": "default" }, "notes": "Kimi K3 technical report: MathVision (with Python) = 97.1. Source setting: Python tool; avg 3 runs. Origin: Moonshot evaluation." } ] }, { "model_id": "claude-opus-4.8", "benchmark_id": "zerobench_main", "score": 17.0, "reference_url": "https://github.com/MoonshotAI/Kimi-K3/blob/main/k3_tech_report.pdf", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "benchmark-specified", "sampling": "pass@5", "judge": "benchmark-specified", "harness": "no Python; pass@5 / 5 runs", "prompt_style": "default", "temperature": "default", "context": "default" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Kimi K3 technical report: ZeroBench (pass@5) (no Python) = 17.0. Source setting: no Python; pass@5 / 5 runs. Origin: Moonshot evaluation." }, { "model_id": "claude-opus-4.8", "benchmark_id": "zerobench_tools", "score": 34.0, "reference_url": "https://github.com/MoonshotAI/Kimi-K3/blob/main/k3_tech_report.pdf", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "python", "sampling": "pass@5", "judge": "benchmark-specified", "harness": "Python tool; pass@5 / 5 runs", "prompt_style": "default", "temperature": "default", "context": "default" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Kimi K3 technical report: ZeroBench (pass@5) (with Python) = 34.0. Source setting: Python tool; pass@5 / 5 runs. Origin: Moonshot evaluation." }, { "model_id": "claude-opus-4.8", "benchmark_id": "coding_experience", "score": 58.0, "reference_url": "https://github.com/MoonshotAI/Kimi-K3/blob/main/k3_tech_report.pdf", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "benchmark-specified", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "provider-aligned harness: Kimi Code / Claude Code / Codex", "prompt_style": "default", "temperature": "default", "context": "default" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Kimi K3 technical report: Coding Experience = 58.0. Source setting: provider-aligned harness: Kimi Code / Claude Code / Codex. Origin: Moonshot evaluation." }, { "model_id": "claude-opus-4.8", "benchmark_id": "clawbench_2_0", "score": 47.2, "reference_url": "https://github.com/MoonshotAI/Kimi-K3/blob/main/k3_tech_report.pdf", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "benchmark-specified", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "OpenClaw harness", "prompt_style": "default", "temperature": "default", "context": "default" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Kimi K3 technical report: 24/7 ClawBench 2.0 = 47.2. Source setting: OpenClaw harness. Origin: Moonshot evaluation." }, { "model_id": "claude-opus-4.8", "benchmark_id": "mira_bench", "score": 59.8, "reference_url": "https://github.com/MoonshotAI/Kimi-K3/blob/main/k3_tech_report.pdf", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "benchmark-specified", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "MIRA harness", "prompt_style": "default", "temperature": "default", "context": "default" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Kimi K3 technical report: MIRA Bench = 59.8. Source setting: MIRA harness. Origin: Moonshot evaluation." }, { "model_id": "claude-opus-4.8", "benchmark_id": "kaet", "score": 78.7, "reference_url": "https://github.com/MoonshotAI/Kimi-K3/blob/main/k3_tech_report.pdf", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "benchmark-specified", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "Kimi Code harness", "prompt_style": "default", "temperature": "default", "context": "default" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Kimi K3 technical report: KAET = 78.7. Source setting: Kimi Code harness. Origin: Moonshot evaluation." }, { "model_id": "claude-opus-4.8", "benchmark_id": "clif_bench", "score": 48.8, "reference_url": "https://github.com/MoonshotAI/Kimi-K3/blob/main/k3_tech_report.pdf", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "benchmark-specified", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "Kimi Code harness", "prompt_style": "default", "temperature": "default", "context": "default" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Kimi K3 technical report: CLIF Bench = 48.8. Source setting: Kimi Code harness. Origin: Moonshot evaluation." }, { "model_id": "claude-opus-4.8", "benchmark_id": "agentic_vision_bench", "score": 82.8, "reference_url": "https://github.com/MoonshotAI/Kimi-K3/blob/main/k3_tech_report.pdf", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "benchmark-specified", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "Kimi Code harness", "prompt_style": "default", "temperature": "default", "context": "default" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Kimi K3 technical report: Agentic Vision Bench = 82.8. Source setting: Kimi Code harness. Origin: Moonshot evaluation." }, { "model_id": "claude-opus-4.8", "benchmark_id": "swarm_bench", "score": 72.6, "reference_url": "https://github.com/MoonshotAI/Kimi-K3/blob/main/k3_tech_report.pdf", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "benchmark-specified", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "Kimi Agent harness", "prompt_style": "default", "temperature": "default", "context": "default" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Kimi K3 technical report: SwarmBench = 72.6. Source setting: Kimi Agent harness. Origin: Moonshot evaluation." }, { "model_id": "claude-opus-4.8", "benchmark_id": "online_experience", "score": 69.4, "reference_url": "https://github.com/MoonshotAI/Kimi-K3/blob/main/k3_tech_report.pdf", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "benchmark-specified", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "Kimi Agent harness", "prompt_style": "default", "temperature": "default", "context": "default" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Kimi K3 technical report: Online Experience = 69.4. Source setting: Kimi Agent harness. Origin: Moonshot evaluation." }, { "model_id": "claude-opus-4.8", "benchmark_id": "deepresearchbench", "score": 87.2, "reference_url": "https://github.com/MoonshotAI/Kimi-K3/blob/main/k3_tech_report.pdf", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "benchmark-specified", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "Kimi Agent harness", "prompt_style": "default", "temperature": "default", "context": "default" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Kimi K3 technical report: DeepResearchBench = 87.2. Source setting: Kimi Agent harness. Origin: Moonshot evaluation." }, { "model_id": "claude-opus-4.8", "benchmark_id": "finance_bench", "score": 60.7, "reference_url": "https://github.com/MoonshotAI/Kimi-K3/blob/main/k3_tech_report.pdf", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "benchmark-specified", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "N/A harness", "prompt_style": "default", "temperature": "default", "context": "default" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Kimi K3 technical report: Finance Bench = 60.7. Source setting: N/A harness. Origin: Moonshot evaluation." }, { "model_id": "claude-opus-4.8", "benchmark_id": "kwv_bench", "score": 61.7, "reference_url": "https://github.com/MoonshotAI/Kimi-K3/blob/main/k3_tech_report.pdf", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "benchmark-specified", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "N/A harness", "prompt_style": "default", "temperature": "default", "context": "default" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Kimi K3 technical report: KWVBench = 61.7. Source setting: N/A harness. Origin: Moonshot evaluation." }, { "model_id": "claude-opus-4.8", "benchmark_id": "deck_bench", "score": 66.9, "reference_url": "https://github.com/MoonshotAI/Kimi-K3/blob/main/k3_tech_report.pdf", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "benchmark-specified", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "N/A harness", "prompt_style": "default", "temperature": "default", "context": "default" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Kimi K3 technical report: DECKBench = 66.9. Source setting: N/A harness. Origin: Moonshot evaluation." }, { "model_id": "claude-opus-4.8", "benchmark_id": "agent_behavior_bench", "score": 65.7, "reference_url": "https://github.com/MoonshotAI/Kimi-K3/blob/main/k3_tech_report.pdf", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "benchmark-specified", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "Kimi Work harness", "prompt_style": "default", "temperature": "default", "context": "default" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Kimi K3 technical report: Agent Behavior Bench = 65.7. Source setting: Kimi Work harness. Origin: Moonshot evaluation." }, { "model_id": "claude-opus-4.8", "benchmark_id": "faithfulness", "score": 83.6, "reference_url": "https://github.com/MoonshotAI/Kimi-K3/blob/main/k3_tech_report.pdf", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "benchmark-specified", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "N/A harness", "prompt_style": "default", "temperature": "default", "context": "default" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Kimi K3 technical report: Faithfulness (1 - hallucination rate) = 83.6. Source setting: N/A harness. Origin: Moonshot evaluation." }, { "model_id": "claude-opus-4.8", "benchmark_id": "chat_all_in_one_bench", "score": 83.8, "reference_url": "https://github.com/MoonshotAI/Kimi-K3/blob/main/k3_tech_report.pdf", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "benchmark-specified", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "Kimi Work harness", "prompt_style": "default", "temperature": "default", "context": "default" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Kimi K3 technical report: Chat All-in-One Bench = 83.8. Source setting: Kimi Work harness. Origin: Moonshot evaluation." }, { "model_id": "claude-opus-4.8", "benchmark_id": "aa_intelligence_index_v4_1", "score": 55.7, "reference_url": "https://github.com/MoonshotAI/Kimi-K3/blob/main/k3_tech_report.pdf", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "benchmark-specified", "sampling": "leaderboard snapshot 2026-07-23", "judge": "benchmark-specified", "harness": "live leaderboard snapshot as of 2026-07-23", "prompt_style": "default", "temperature": "default", "context": "default" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Kimi K3 technical report: AA Intelligence Index v4.1 = 55.7. Source setting: live leaderboard snapshot as of 2026-07-23. Origin: third-party leaderboard cited by Kimi report." }, { "model_id": "claude-opus-4.8", "benchmark_id": "vals_index", "score": 70.4, "reference_url": "https://github.com/MoonshotAI/Kimi-K3/blob/main/k3_tech_report.pdf", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "benchmark-specified", "sampling": "leaderboard snapshot 2026-07-23", "judge": "benchmark-specified", "harness": "live leaderboard snapshot as of 2026-07-23", "prompt_style": "default", "temperature": "default", "context": "default" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Kimi K3 technical report: Vals Index = 70.4. Source setting: live leaderboard snapshot as of 2026-07-23. Origin: third-party leaderboard cited by Kimi report.", "candidates": [ { "score": 70.4, "reference_url": "https://media.x.ai/v1/website/card-4p6-4cd2dc57.pdf", "source_type": "model_card", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "benchmark-specified professional tools", "sampling": "unknown", "judge": "Vals weighted suite score", "harness": "Vals Index live harness", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown" }, "notes": "card46 Vals Index Suite score (%); source effort=max; benchmark protocol matches canonical=True. Unknown source fields remain unknown." } ] }, { "model_id": "claude-opus-4.8", "benchmark_id": "webdev_arena_elo", "score": 1565.0, "reference_url": "https://github.com/MoonshotAI/Kimi-K3/blob/main/k3_tech_report.pdf", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "benchmark-specified", "sampling": "leaderboard snapshot 2026-07-23", "judge": "benchmark-specified", "harness": "live leaderboard snapshot as of 2026-07-23", "prompt_style": "default", "temperature": "default", "context": "default" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Kimi K3 technical report: WebDevArena Elo = 1565. Source setting: live leaderboard snapshot as of 2026-07-23. Origin: third-party leaderboard cited by Kimi report." }, { "model_id": "claude-opus-4.8", "benchmark_id": "text_arena_elo", "score": 1484.0, "reference_url": "https://github.com/MoonshotAI/Kimi-K3/blob/main/k3_tech_report.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "benchmark-specified", "sampling": "leaderboard snapshot 2026-07-23", "judge": "benchmark-specified", "harness": "live leaderboard snapshot as of 2026-07-23", "prompt_style": "default", "temperature": "default", "context": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Kimi K3 technical report: TextArena Elo = 1484. Source setting: live leaderboard snapshot as of 2026-07-23. Origin: third-party leaderboard cited by Kimi report." }, { "model_id": "claude-opus-4.8", "benchmark_id": "agent_arena", "score": 9.8, "reference_url": "https://github.com/MoonshotAI/Kimi-K3/blob/main/k3_tech_report.pdf", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "benchmark-specified", "sampling": "leaderboard snapshot 2026-07-23", "judge": "benchmark-specified", "harness": "live leaderboard snapshot as of 2026-07-23", "prompt_style": "default", "temperature": "default", "context": "default" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Kimi K3 technical report: AgentArena = 9.8. Source setting: live leaderboard snapshot as of 2026-07-23. Origin: third-party leaderboard cited by Kimi report." }, { "model_id": "gpt-5.5", "benchmark_id": "critpt", "score": 27.1, "reference_url": "https://github.com/MoonshotAI/Kimi-K3/blob/main/k3_tech_report.pdf", "reported_setting": { "mode": "thinking", "effort": "xhigh", "tools": "none", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "max/xhigh; no tools", "prompt_style": "default", "temperature": "default", "context": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Kimi K3 technical report: CritPt = 27.1. Source setting: max/xhigh; no tools. Origin: Artificial Analysis cited by Kimi report." }, { "model_id": "gpt-5.5", "benchmark_id": "aa_lcr", "score": 74.3, "reference_url": "https://github.com/MoonshotAI/Kimi-K3/blob/main/k3_tech_report.pdf", "reported_setting": { "mode": "thinking", "effort": "xhigh", "tools": "none", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "max/xhigh; no tools", "prompt_style": "default", "temperature": "default", "context": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Kimi K3 technical report: AA-LCR = 74.3. Source setting: max/xhigh; no tools. Origin: Artificial Analysis cited by Kimi report.", "candidates": [ { "score": 76.4, "reference_url": "https://huggingface.co/tencent/Hy3", "source_type": "model_card", "reported_setting": { "mode": "thinking", "effort": "high / highest available", "tools": "none", "sampling": "repeats=3 per question", "judge": "Equality Checker LLM", "harness": "Artificial Analysis AA-LCR, 100 questions", "prompt_style": "default", "temperature": "0.9 for Hy3 family; otherwise source/provider setting", "context": "256000 for Hy3 family; otherwise source/provider setting", "notes": "Tencent own test" }, "notes": "Hy3 official model-card appendix: AA-LCR = 76.4*. Tencent own testing." } ] }, { "model_id": "gpt-5.5", "benchmark_id": "hle_tools", "score": 52.2, "reference_url": "https://github.com/MoonshotAI/Kimi-K3/blob/main/k3_tech_report.pdf", "reported_setting": { "mode": "thinking", "effort": "xhigh", "tools": "general tools", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "general tools; pass@1", "prompt_style": "default", "temperature": "default", "context": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Kimi K3 technical report: HLE-Full (with tools) = 52.2. Source setting: general tools; pass@1. Origin: Moonshot evaluation.", "candidates": [ { "score": 57.2, "reference_url": "https://www.anthropic.com/claude-sonnet-5-system-card", "source_type": "model_card", "reported_setting": { "effort": "source does not state", "tools": "python / benchmark tools", "sampling": "pass@1", "harness": "hle", "deployment": "gpt-5.5-pro" }, "notes": "Anthropic Claude Sonnet 5 System Card Figure 8.9.1.A, page 122: GPT-5.5 Pro; Humanity’s Last Exam [with tools]=57.2%. Source setting: effort=source does not state; tools=python / benchmark tools; sampling=pass@1; harness=hle; deployment=gpt-5.5-pro." }, { "score": 52.2, "reference_url": "https://static.stepfun.com/blog/step-3.7-flash/", "source_type": "official_blog", "reported_setting": { "mode": "thinking", "effort": "xhigh", "tools": "general tools", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "general tools; pass@1", "prompt_style": "default", "temperature": "default", "context": "default", "multimodal_input": true }, "notes": "Official StepFun launch comparison table. Provider-quoted value is independently locked by the campaign's primary official-source record." } ] }, { "model_id": "gpt-5.5", "benchmark_id": "deep_swe_v1_1", "score": 67.0, "reference_url": "https://github.com/MoonshotAI/Kimi-K3/blob/main/k3_tech_report.pdf", "reported_setting": { "mode": "thinking", "effort": "xhigh", "tools": "benchmark-specified", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "agentic coding harness per row", "prompt_style": "default", "temperature": "default", "context": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Kimi K3 technical report: DeepSWE = 67.0. Source setting: agentic coding harness per row. Origin: Moonshot K3 run; competitors from DeepSWE leaderboard / GLM blog.", "candidates": [ { "score": 70.0, "reference_url": "https://z.ai/blog/glm-5.2", "source_type": "official_blog", "reported_setting": { "mode": "thinking", "effort": "xhigh", "sampling": "pass@1", "prompt_style": "default", "tools": "agentic code execution", "temperature": "1.0", "top_p": "1.0", "context": "400000", "judge": "isolated-container verification", "harness": "official pier + mini-swe-agent; timeout=2h; 2 CPU/8GB; no internet" }, "notes": "Z.ai cross-table alternative." }, { "score": 70.8, "reference_url": "https://huggingface.co/tencent/Hy3", "source_type": "model_card", "reported_setting": { "mode": "thinking", "effort": "high / highest available", "tools": "agentic code execution", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "mini-swe-agent; 2h/task; 2 CPU/8GB; no internet", "prompt_style": "default", "temperature": "0.9 for Hy3 family; otherwise source/provider setting", "context": "256000 for Hy3 family; otherwise source/provider setting", "notes": "Tencent own test" }, "notes": "Hy3 official model-card appendix: DeepSWE = 70.8*. Tencent own testing." }, { "score": 26.99, "reference_url": "https://openai.com/index/gpt-5-6/", "source_type": "official_blog", "reported_setting": { "effort": "low", "eval_variant": null, "code_mode": "false" }, "notes": "OpenAI GPT-5.6 release figure (DeepSWE v1.1): GPT-5.5; DeepSWE v1.1=26.99. Figure 1: DeepSWE v1.1. Cost, latency, and output-token coordinates are retained as review metadata, not score-matrix metrics." }, { "score": 53.98, "reference_url": "https://openai.com/index/gpt-5-6/", "source_type": "official_blog", "reported_setting": { "effort": "medium", "eval_variant": null, "code_mode": "false" }, "notes": "OpenAI GPT-5.6 release figure (DeepSWE v1.1): GPT-5.5; DeepSWE v1.1=53.98. Figure 1: DeepSWE v1.1. Cost, latency, and output-token coordinates are retained as review metadata, not score-matrix metrics." }, { "score": 64.38, "reference_url": "https://openai.com/index/gpt-5-6/", "source_type": "official_blog", "reported_setting": { "effort": "high", "eval_variant": null, "code_mode": "false" }, "notes": "OpenAI GPT-5.6 release figure (DeepSWE v1.1): GPT-5.5; DeepSWE v1.1=64.38. Figure 1: DeepSWE v1.1. Cost, latency, and output-token coordinates are retained as review metadata, not score-matrix metrics." }, { "score": 67.0, "reference_url": "https://x.ai/news/grok-4-5", "source_type": "official_blog", "reported_setting": { "mode": "thinking", "effort": "xhigh", "tools": "agentic repository shell/editor", "sampling": "pass@1", "judge": "isolated repository verifier", "harness": "mini-swe-agent run by Datacurve", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown" }, "notes": "release45 DeepSWE 1.1 Pass@1 (%); source effort=xhigh; benchmark protocol matches canonical=True. Unknown source fields remain unknown." }, { "score": 70.0, "reference_url": "https://seed.bytedance.com/en/blog/seed2-1-officially-released-advancing-ai-productivity", "source_type": "official_blog", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "Trae or benchmark-specific coding-agent environment", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official", "prompt_style": "official default", "temperature": "unknown" }, "notes": "Release blog EN: Coding chart (https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/user-upload/4xfa4mqq4wlbt.png)." } ] }, { "model_id": "gpt-5.5", "benchmark_id": "terminal_bench_2_1", "score": 83.4, "reference_url": "https://github.com/MoonshotAI/Kimi-K3/blob/main/k3_tech_report.pdf", "reported_setting": { "mode": "thinking", "effort": "xhigh", "tools": "benchmark-specified", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "agentic coding harness per row", "prompt_style": "default", "temperature": "default", "context": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Kimi K3 technical report: Terminal-Bench 2.1 = 83.4. Source setting: agentic coding harness per row. Origin: mixed official sources documented by Kimi report.", "candidates": [ { "score": 84.0, "reference_url": "https://z.ai/blog/glm-5.2", "source_type": "official_blog", "reported_setting": { "mode": "thinking", "effort": "xhigh", "sampling": "pass@1", "prompt_style": "default", "tools": "terminal agent", "temperature": "1.0", "top_p": "1.0", "context": "256000", "max_output_tokens": "48000", "judge": "official task verifier", "harness": "Terminus-2; parser=json; timeout=4h; max_episodes=500; 4 CPU/8GB" }, "notes": "Z.ai Terminus-2 alternative." }, { "score": 79.8, "reference_url": "https://huggingface.co/tencent/Hy3", "source_type": "model_card", "reported_setting": { "mode": "thinking", "effort": "high / highest available", "tools": "terminal/code execution", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "Terminus-2; parser=xml; timeout=4h; CPU=16 cores; memory=32GB; max episodes=500", "prompt_style": "default", "temperature": "0.9 for Hy3 family; otherwise source/provider setting", "context": "256000 for Hy3 family; otherwise source/provider setting", "notes": "Tencent own test" }, "notes": "Hy3 official model-card appendix: Terminal-Bench 2.1 = 79.8*. Tencent own testing." }, { "score": 73.8, "reference_url": "https://longcat.chat/blog/longcat-2.0/", "source_type": "official_blog", "reported_setting": { "mode": "thinking", "effort": "xhigh", "tools": "terminal agent", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "provider official report; details not reproduced by LongCat", "prompt_style": "default", "temperature": "provider official setting", "context": "default", "notes": "Asterisk: LongCat cites this value from the model provider's official report rather than measuring it in-house." }, "notes": "LongCat-2.0 official tech blog: Terminal-Bench 2.1 = 73.8*. Asterisked value is cited by LongCat from the model provider's official report." }, { "score": 79.4, "reference_url": "https://huggingface.co/internlm/Intern-S2-Preview-397B", "source_type": "model_card", "reported_setting": { "mode": "source does not state", "effort": "source does not state", "tools": "terminal/code execution", "sampling": "source does not state", "judge": "benchmark-specific; exact version not stated", "harness": "Terminus 2", "prompt_style": "source does not state", "temperature": "source does not state", "context": "source does not state", "notes": "Shanghai AI Lab evaluated all model columns using OpenCompass, VLMEvalKit, and AgentCompass. Per-model mode/effort/sampling are not reported; only Intern-S2 maximum evaluation lengths are disclosed." }, "notes": "Intern-S2-Preview-397B official model card: TerminalBench 2.1 = 79.4. This is a Shanghai AI Lab cross-model evaluation; the source does not disclose this model's exact mode, effort, or sampling." }, { "score": 85.6, "reference_url": "https://openai.com/index/gpt-5-6/", "source_type": "official_blog", "reported_setting": { "effort": "xhigh", "eval_variant": null, "code_mode": "true" }, "notes": "OpenAI GPT-5.6 release table (table 1): GPT‑5.5; Terminal-Bench 2.1=85.6. Exact effort/variant resolved from the rendered chart point." }, { "score": 72.81, "reference_url": "https://openai.com/index/gpt-5-6/", "source_type": "official_blog", "reported_setting": { "effort": "low", "eval_variant": null, "code_mode": "true" }, "notes": "OpenAI GPT-5.6 release figure (Terminal-Bench 2.1): GPT-5.5; Terminal-Bench 2.1=72.81. Figure 1: Terminal-Bench 2.1. Cost, latency, and output-token coordinates are retained as review metadata, not score-matrix metrics." }, { "score": 80.0, "reference_url": "https://openai.com/index/gpt-5-6/", "source_type": "official_blog", "reported_setting": { "effort": "medium", "eval_variant": null, "code_mode": "true" }, "notes": "OpenAI GPT-5.6 release figure (Terminal-Bench 2.1): GPT-5.5; Terminal-Bench 2.1=80.0. Figure 1: Terminal-Bench 2.1. Cost, latency, and output-token coordinates are retained as review metadata, not score-matrix metrics." }, { "score": 83.82, "reference_url": "https://openai.com/index/gpt-5-6/", "source_type": "official_blog", "reported_setting": { "effort": "high", "eval_variant": null, "code_mode": "true" }, "notes": "OpenAI GPT-5.6 release figure (Terminal-Bench 2.1): GPT-5.5; Terminal-Bench 2.1=83.82. Figure 1: Terminal-Bench 2.1. Cost, latency, and output-token coordinates are retained as review metadata, not score-matrix metrics." }, { "score": 78.2, "reference_url": "https://deepmind.google/models/model-cards/gemini-3-5-flash/", "source_type": "official_model_card", "reported_setting": { "mode": "thinking/reasoning", "effort": "xhigh/best available", "tools": "terminal agent", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "Terminus-2; Gemini self-computed, others public leaderboard", "temperature": "default", "snapshot": "May 2026" }, "notes": "Google Gemini 3.5 Flash May 2026 table: Terminal-Bench 2.1; Terminus-2 harness. Settings and provenance are preserved per cell." }, { "score": 83.4, "reference_url": "https://x.ai/news/grok-4-5", "source_type": "official_blog", "reported_setting": { "mode": "thinking", "effort": "xhigh", "tools": "container terminal", "sampling": "pass@1", "judge": "verified task-success evaluator", "harness": "Grok Build for Grok; source-reported peer harnesses", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown" }, "notes": "release45 Terminal-Bench 2.1 Task success rate (%); source effort=xhigh; benchmark protocol matches canonical=True. Unknown source fields remain unknown." }, { "score": 73.8, "reference_url": "https://seed.bytedance.com/en/blog/seed2-1-officially-released-advancing-ai-productivity", "source_type": "official_blog", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "Trae or benchmark-specific coding-agent environment", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official", "prompt_style": "official default", "temperature": "unknown" }, "notes": "Release blog EN: Coding chart (https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/user-upload/4xfa4mqq4wlbt.png)." } ] }, { "model_id": "gpt-5.5", "benchmark_id": "frontier_swe", "score": 64.9, "reference_url": "https://github.com/MoonshotAI/Kimi-K3/blob/main/k3_tech_report.pdf", "reported_setting": { "mode": "thinking", "effort": "xhigh", "tools": "benchmark-specified", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "agentic coding harness per row", "prompt_style": "default", "temperature": "default", "context": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Kimi K3 technical report: FrontierSWE = 64.9. Source setting: agentic coding harness per row. Origin: Moonshot/OpenAI runs plus official leaderboard.", "candidates": [ { "score": 72.6, "reference_url": "https://z.ai/blog/glm-5.2", "source_type": "official_blog", "reported_setting": { "mode": "thinking", "effort": "xhigh", "sampling": "pass@1", "prompt_style": "default", "tools": "agentic code execution", "context": "1000000", "max_output_tokens": "128000", "judge": "FrontierSWE dominance", "harness": "Proximal evaluation; max effort; as of 2026-06-16", "temperature": "default" }, "notes": "Z.ai dated 2026-06-16 alternative." }, { "score": 70.0, "reference_url": "https://media.x.ai/v1/website/4p5-5184fdf9.pdf", "source_type": "model_card", "reported_setting": { "mode": "thinking", "effort": "xhigh", "tools": "agentic code execution", "sampling": "mean@5", "judge": "continuous partial-credit task scoring and dominance", "harness": "public FrontierSWE harness; Grok CLI for Grok 4.5", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown" }, "notes": "card45 FrontierSWE Dominance (%); source effort=xhigh; benchmark protocol matches canonical=True. Unknown source fields remain unknown." } ] }, { "model_id": "gpt-5.5", "benchmark_id": "swe_marathon_h20_2026_07_09", "score": 14.0, "reference_url": "https://github.com/MoonshotAI/Kimi-K3/blob/main/k3_tech_report.pdf", "reported_setting": { "mode": "thinking", "effort": "xhigh", "tools": "benchmark-specified", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "H20-calibrated pre-final-v1.1 branch (2026-07-09)", "prompt_style": "default", "temperature": "default", "context": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Kimi K3 technical report: SWE-Marathon = 14.0. Source setting: H20-calibrated pre-final-v1.1 branch (2026-07-09). Origin: Moonshot evaluation." }, { "model_id": "gpt-5.5", "benchmark_id": "posttrain_bench", "score": 28.4, "reference_url": "https://github.com/MoonshotAI/Kimi-K3/blob/main/k3_tech_report.pdf", "reported_setting": { "mode": "thinking", "effort": "xhigh", "tools": "benchmark-specified", "sampling": "avg-of-3", "judge": "benchmark-specified", "harness": "H20 avg@3 for K3/Fable/Sol; official H100 for Opus/GPT-5.5/GLM", "prompt_style": "default", "temperature": "default", "context": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Kimi K3 technical report: PostTrainBench = 28.4. Source setting: H20 avg@3 for K3/Fable/Sol; official H100 for Opus/GPT-5.5/GLM. Origin: Moonshot H20 runs plus official H100 leaderboard values.", "candidates": [ { "score": 25.0, "reference_url": "https://z.ai/blog/glm-5.2", "source_type": "official_blog", "reported_setting": { "mode": "thinking", "effort": "xhigh", "sampling": "pass@1", "prompt_style": "default", "tools": "agentic post-training", "context": "1000000", "max_output_tokens": "128000", "judge": "official weighted benchmark score", "harness": "PostTrainBench evaluation; max effort", "temperature": "default" }, "notes": "Z.ai chart conflicts with the 28.4 full-table value." }, { "score": 25.0, "reference_url": "https://seed.bytedance.com/en/blog/seed2-1-officially-released-advancing-ai-productivity", "source_type": "official_blog", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "research-oriented agent environment", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official", "prompt_style": "official default", "temperature": "unknown" }, "notes": "Release blog EN: Frontier-research chart (https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/user-upload/4xfa4mqq52cdr.png). Source dashes are not zero." } ] }, { "model_id": "gpt-5.5", "benchmark_id": "scicode", "score": 56.1, "reference_url": "https://github.com/MoonshotAI/Kimi-K3/blob/main/k3_tech_report.pdf", "reported_setting": { "mode": "thinking", "effort": "xhigh", "tools": "benchmark-specified", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "agentic coding harness per row", "prompt_style": "default", "temperature": "default", "context": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Kimi K3 technical report: SciCode = 56.1. Source setting: agentic coding harness per row. Origin: Artificial Analysis cited by Kimi report.", "candidates": [ { "score": 55.92, "reference_url": "https://huggingface.co/internlm/Intern-S2-Preview-397B", "source_type": "model_card", "reported_setting": { "mode": "source does not state", "effort": "source does not state", "tools": "code execution", "sampling": "source does not state", "judge": "benchmark-specific; exact version not stated", "harness": "AgentCompass / benchmark evaluator", "prompt_style": "source does not state", "temperature": "source does not state", "context": "source does not state", "notes": "Shanghai AI Lab evaluated all model columns using OpenCompass, VLMEvalKit, and AgentCompass. Per-model mode/effort/sampling are not reported; only Intern-S2 maximum evaluation lengths are disclosed." }, "notes": "Intern-S2-Preview-397B official model card: SciCode = 55.92. This is a Shanghai AI Lab cross-model evaluation; the source does not disclose this model's exact mode, effort, or sampling." }, { "score": 58.4, "reference_url": "https://seed.bytedance.com/en/blog/seed2-1-officially-released-advancing-ai-productivity", "source_type": "official_blog", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "none", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official", "prompt_style": "official default", "temperature": "unknown" }, "notes": "Release blog EN: Language/search chart (https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/user-upload/4xfa4mqq510uc.png)." } ] }, { "model_id": "gpt-5.5", "benchmark_id": "researchrubrics", "score": 64.0, "reference_url": "https://github.com/MoonshotAI/Kimi-K3/blob/main/k3_tech_report.pdf", "reported_setting": { "mode": "thinking", "effort": "xhigh", "tools": "benchmark-specified", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "agentic benchmark harness", "prompt_style": "default", "temperature": "default", "context": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Kimi K3 technical report: ResearchRubrics = 64.0. Source setting: agentic benchmark harness. Origin: Moonshot evaluation.", "candidates": [ { "score": 61.5, "reference_url": "https://static.stepfun.com/blog/step-3.7-flash/", "source_type": "official_blog", "reported_setting": { "quantization_config": "common release evaluation; artifact precision not established", "configuration": "base Step evaluation configuration", "mode": "source does not state", "effort": "source does not state", "tools_search": "research tools", "shots": "source does not state", "trials_samples": "source does not state", "aggregation_pass_k": "source-reported score", "context": "source does not state", "output_cap": "source does not state", "temperature": "source does not state", "top_p": "source does not state", "judge": "weighted binary rubric compliance evaluation", "harness_agent": "official StepFun evaluation; exact harness not stated", "dataset_version_split": "ResearchRubrics 101-task official release", "multimodal_input": false }, "notes": "Official StepFun launch comparison table." } ] }, { "model_id": "gpt-5.5", "benchmark_id": "gdpval_aa_elo", "score": 1491.0, "reference_url": "https://github.com/MoonshotAI/Kimi-K3/blob/main/k3_tech_report.pdf", "reported_setting": { "mode": "thinking", "effort": "xhigh", "tools": "benchmark-specified", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "agentic benchmark harness", "prompt_style": "default", "temperature": "default", "context": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Kimi K3 technical report: GDPval-AA v2 (Elo) = 1491.0. Source setting: agentic benchmark harness. Origin: Artificial Analysis cited by Kimi report.", "candidates": [ { "score": 1493.7, "reference_url": "https://openai.com/index/gpt-5-6/", "source_type": "official_blog", "reported_setting": { "effort": "xhigh", "eval_variant": null, "code_mode": "false" }, "notes": "OpenAI GPT-5.6 release table (table 0): GPT‑5.5; GDPval-AA v2=1493.7. OpenAI row explicitly identifies GDPval-AA v2. Exact effort/variant resolved from the rendered chart point." }, { "score": 1119.0, "reference_url": "https://openai.com/index/gpt-5-6/", "source_type": "official_blog", "reported_setting": { "effort": "none", "eval_variant": null, "code_mode": "false" }, "notes": "OpenAI GPT-5.6 release figure (GDPval-AA v2): GPT-5.5; GDPval-AA v2=1119.0. Figure 15: GDPval-AA v2. Cost, latency, and output-token coordinates are retained as review metadata, not score-matrix metrics." }, { "score": 1191.0, "reference_url": "https://openai.com/index/gpt-5-6/", "source_type": "official_blog", "reported_setting": { "effort": "low", "eval_variant": null, "code_mode": "false" }, "notes": "OpenAI GPT-5.6 release figure (GDPval-AA v2): GPT-5.5; GDPval-AA v2=1191.0. Figure 15: GDPval-AA v2. Cost, latency, and output-token coordinates are retained as review metadata, not score-matrix metrics." }, { "score": 1375.0, "reference_url": "https://openai.com/index/gpt-5-6/", "source_type": "official_blog", "reported_setting": { "effort": "medium", "eval_variant": null, "code_mode": "false" }, "notes": "OpenAI GPT-5.6 release figure (GDPval-AA v2): GPT-5.5; GDPval-AA v2=1375.0. Figure 15: GDPval-AA v2. Cost, latency, and output-token coordinates are retained as review metadata, not score-matrix metrics." }, { "score": 1471.0, "reference_url": "https://openai.com/index/gpt-5-6/", "source_type": "official_blog", "reported_setting": { "effort": "high", "eval_variant": null, "code_mode": "false" }, "notes": "OpenAI GPT-5.6 release figure (GDPval-AA v2): GPT-5.5; GDPval-AA v2=1471.0. Figure 15: GDPval-AA v2. Cost, latency, and output-token coordinates are retained as review metadata, not score-matrix metrics." }, { "score": 1509.0, "reference_url": "https://www.anthropic.com/claude-sonnet-5-system-card", "source_type": "model_card", "reported_setting": { "effort": "source does not state", "tools": "none", "sampling": "avg 5 trials unless source states otherwise", "harness": "table8" }, "notes": "Anthropic Claude Sonnet 5 System Card Table 8.1.A, page 115: GPT-5.5; GDPval-AA v2 [summary]=1509. Source setting: effort=source does not state; tools=none; sampling=avg 5 trials unless source states otherwise; harness=table8." }, { "score": 1769.0, "reference_url": "https://static.stepfun.com/blog/step-3.7-flash/", "source_type": "official_blog", "reported_setting": { "quantization_config": "common release evaluation; artifact precision not established", "configuration": "base Step evaluation configuration", "mode": "source does not state", "effort": "source does not state", "tools_search": "official Artificial Analysis Stirrup harness", "shots": "source does not state", "trials_samples": "source does not state", "aggregation_pass_k": "source-reported score", "context": "source does not state", "output_cap": "source does not state", "temperature": "source does not state", "top_p": "source does not state", "judge": "pairwise Elo evaluation anchored to human experts", "harness_agent": "official Artificial Analysis Stirrup leaderboard", "dataset_version_split": "GDPval-AA historical 220-task Stirrup leaderboard quoted 2026-05-29", "multimodal_input": false }, "notes": "Official StepFun launch comparison table." } ] }, { "model_id": "gpt-5.5", "benchmark_id": "toolathlon_verified", "score": 73.5, "reference_url": "https://github.com/MoonshotAI/Kimi-K3/blob/main/k3_tech_report.pdf", "reported_setting": { "mode": "thinking", "effort": "xhigh", "tools": "benchmark-specified", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "Verified edition; official leaderboard snapshot", "prompt_style": "default", "temperature": "default", "context": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Kimi K3 technical report: Toolathlon-Verified = 73.5. Source setting: Verified edition; official leaderboard snapshot. Origin: official Toolathlon leaderboard cited by Kimi report." }, { "model_id": "gpt-5.5", "benchmark_id": "automation_bench", "score": 22.7, "reference_url": "https://github.com/MoonshotAI/Kimi-K3/blob/main/k3_tech_report.pdf", "reported_setting": { "mode": "thinking", "effort": "xhigh", "tools": "benchmark-specified", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "agentic benchmark harness", "prompt_style": "default", "temperature": "default", "context": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Kimi K3 technical report: AutomationBench = 22.7. Source setting: agentic benchmark harness. Origin: Moonshot evaluation.", "candidates": [ { "score": 12.9, "reference_url": "https://openai.com/index/gpt-5-6/", "source_type": "official_blog", "reported_setting": { "effort": "xhigh", "eval_variant": null, "code_mode": "false" }, "notes": "OpenAI GPT-5.6 release table (table 8): GPT‑5.5; AutomationBench=12.9. Exact effort/variant resolved from the rendered chart point." }, { "score": 3.3, "reference_url": "https://openai.com/index/gpt-5-6/", "source_type": "official_blog", "reported_setting": { "effort": "none", "eval_variant": null, "code_mode": "false" }, "notes": "OpenAI GPT-5.6 release figure (AutomationBench): GPT-5.5; AutomationBench=3.3. Figure 15: AutomationBench. Cost, latency, and output-token coordinates are retained as review metadata, not score-matrix metrics." }, { "score": 6.7, "reference_url": "https://openai.com/index/gpt-5-6/", "source_type": "official_blog", "reported_setting": { "effort": "low", "eval_variant": null, "code_mode": "false" }, "notes": "OpenAI GPT-5.6 release figure (AutomationBench): GPT-5.5; AutomationBench=6.7. Figure 15: AutomationBench. Cost, latency, and output-token coordinates are retained as review metadata, not score-matrix metrics." }, { "score": 8.5, "reference_url": "https://openai.com/index/gpt-5-6/", "source_type": "official_blog", "reported_setting": { "effort": "medium", "eval_variant": null, "code_mode": "false" }, "notes": "OpenAI GPT-5.6 release figure (AutomationBench): GPT-5.5; AutomationBench=8.5. Figure 15: AutomationBench. Cost, latency, and output-token coordinates are retained as review metadata, not score-matrix metrics." }, { "score": 11.3, "reference_url": "https://openai.com/index/gpt-5-6/", "source_type": "official_blog", "reported_setting": { "effort": "high", "eval_variant": null, "code_mode": "false" }, "notes": "OpenAI GPT-5.6 release figure (AutomationBench): GPT-5.5; AutomationBench=11.3. Figure 15: AutomationBench. Cost, latency, and output-token coordinates are retained as review metadata, not score-matrix metrics." } ] }, { "model_id": "gpt-5.5", "benchmark_id": "job_bench", "score": 38.3, "reference_url": "https://github.com/MoonshotAI/Kimi-K3/blob/main/k3_tech_report.pdf", "reported_setting": { "mode": "thinking", "effort": "xhigh", "tools": "benchmark-specified", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "agentic benchmark harness", "prompt_style": "default", "temperature": "default", "context": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Kimi K3 technical report: JobBench = 38.3. Source setting: agentic benchmark harness. Origin: official JobBench leaderboard cited by Kimi report." }, { "model_id": "gpt-5.5", "benchmark_id": "aa_briefcase_elo", "score": 1158.0, "reference_url": "https://github.com/MoonshotAI/Kimi-K3/blob/main/k3_tech_report.pdf", "reported_setting": { "mode": "thinking", "effort": "xhigh", "tools": "benchmark-specified", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "agentic benchmark harness", "prompt_style": "default", "temperature": "default", "context": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Kimi K3 technical report: AA-Briefcase (Elo) = 1158.0. Source setting: agentic benchmark harness. Origin: Artificial Analysis cited by Kimi report.", "candidates": [ { "score": 1150, "reference_url": "https://media.x.ai/v1/website/card-4p6-4cd2dc57.pdf", "source_type": "model_card", "reported_setting": { "mode": "thinking", "effort": "xhigh", "tools": "offline multi-file knowledge-work tools", "sampling": "unknown", "judge": "rubric plus analytical and presentation pairwise Elo", "harness": "Artificial Analysis AA-Briefcase harness", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown" }, "notes": "card46 AA Briefcase AA-Briefcase Elo; source effort=xhigh; benchmark protocol matches canonical=True. Unknown source fields remain unknown." } ] }, { "model_id": "gpt-5.5", "benchmark_id": "agents_last_exam", "score": 26.6, "reference_url": "https://github.com/MoonshotAI/Kimi-K3/blob/main/k3_tech_report.pdf", "reported_setting": { "mode": "thinking", "effort": "xhigh", "tools": "benchmark-specified", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "agentic benchmark harness", "prompt_style": "default", "temperature": "default", "context": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Kimi K3 technical report: Agents' Last Exam = 26.6. Source setting: agentic benchmark harness. Origin: official leaderboard cited by Kimi report.", "candidates": [ { "score": 46.9, "reference_url": "https://openai.com/index/gpt-5-6/", "source_type": "official_blog", "reported_setting": { "effort": "xhigh", "eval_variant": null, "code_mode": "true" }, "notes": "OpenAI GPT-5.6 release table (table 0): GPT‑5.5; Agents' Last Exam=46.9. Exact effort/variant resolved from the rendered chart point." }, { "score": 37.5, "reference_url": "https://openai.com/index/gpt-5-6/", "source_type": "official_blog", "reported_setting": { "effort": "low", "eval_variant": null, "code_mode": "true" }, "notes": "OpenAI GPT-5.6 release figure (Agents' Last Exam): GPT-5.5; Agents' Last Exam=37.5. Figure 0: Long-horizon agentic workflows across professional domains.. Cost, latency, and output-token coordinates are retained as review metadata, not score-matrix metrics." }, { "score": 41.4, "reference_url": "https://openai.com/index/gpt-5-6/", "source_type": "official_blog", "reported_setting": { "effort": "medium", "eval_variant": null, "code_mode": "true" }, "notes": "OpenAI GPT-5.6 release figure (Agents' Last Exam): GPT-5.5; Agents' Last Exam=41.4. Figure 0: Long-horizon agentic workflows across professional domains.. Cost, latency, and output-token coordinates are retained as review metadata, not score-matrix metrics." }, { "score": 44.9, "reference_url": "https://openai.com/index/gpt-5-6/", "source_type": "official_blog", "reported_setting": { "effort": "high", "eval_variant": null, "code_mode": "true" }, "notes": "OpenAI GPT-5.6 release figure (Agents' Last Exam): GPT-5.5; Agents' Last Exam=44.9. Figure 0: Long-horizon agentic workflows across professional domains.. Cost, latency, and output-token coordinates are retained as review metadata, not score-matrix metrics." }, { "score": 24.0, "reference_url": "https://seed.bytedance.com/en/blog/seed2-1-officially-released-advancing-ai-productivity", "source_type": "official_blog", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "computer-use environment", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official", "prompt_style": "official default", "temperature": "unknown" }, "notes": "Release blog EN: General-agent chart (https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/user-upload/4xfa4mqq4u89c.png). Turbo is a source dash, not zero." } ] }, { "model_id": "gpt-5.5", "benchmark_id": "apex_agents", "score": 38.5, "reference_url": "https://github.com/MoonshotAI/Kimi-K3/blob/main/k3_tech_report.pdf", "reported_setting": { "mode": "thinking", "effort": "xhigh", "tools": "benchmark-specified", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "agentic benchmark harness", "prompt_style": "default", "temperature": "default", "context": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Kimi K3 technical report: APEX-Agents = 38.5. Source setting: agentic benchmark harness. Origin: Artificial Analysis cited by Kimi report.", "candidates": [ { "score": 38.4, "reference_url": "https://huggingface.co/tencent/Hy3", "source_type": "model_card", "reported_setting": { "mode": "thinking", "effort": "high / highest available", "tools": "multiple application/software tools", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "APEX-Agents long-horizon professional task harness", "prompt_style": "default", "temperature": "0.9 for Hy3 family; otherwise source/provider setting", "context": "256000 for Hy3 family; otherwise source/provider setting", "notes": "Comparator value not marked as Tencent own test" }, "notes": "Hy3 official model-card appendix: Apex-Agent (pass@1) = 38.4. Comparator-reported value." }, { "score": 55.5, "reference_url": "https://media.x.ai/v1/website/card-4p6-4cd2dc57.pdf", "source_type": "model_card", "reported_setting": { "mode": "thinking", "effort": "xhigh", "tools": "multi-application professional tools", "sampling": "pass@1", "judge": "expert all-criteria-pass rubric", "harness": "Mercor APEX-Agents harness", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown" }, "notes": "card46 APEX-Agents Pass@1 (%); source effort=xhigh; benchmark protocol matches canonical=True. Unknown source fields remain unknown." }, { "score": 35.4, "reference_url": "https://seed.bytedance.com/en/blog/seed2-1-officially-released-advancing-ai-productivity", "source_type": "official_blog", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "long-horizon professional-task environment", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official", "prompt_style": "official default", "temperature": "unknown" }, "notes": "Release blog EN: General-agent chart (https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/user-upload/4xfa4mqq4u89c.png)." } ] }, { "model_id": "gpt-5.5", "benchmark_id": "spreadsheetbench_2", "score": 29.1, "reference_url": "https://github.com/MoonshotAI/Kimi-K3/blob/main/k3_tech_report.pdf", "reported_setting": { "mode": "thinking", "effort": "xhigh", "tools": "benchmark-specified", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "agentic benchmark harness", "prompt_style": "default", "temperature": "default", "context": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Kimi K3 technical report: SpreadsheetBench 2 = 29.1. Source setting: agentic benchmark harness. Origin: Moonshot evaluation." }, { "model_id": "gpt-5.5", "benchmark_id": "osworld_verified", "score": 79.0, "reference_url": "https://github.com/MoonshotAI/Kimi-K3/blob/main/k3_tech_report.pdf", "reported_setting": { "mode": "thinking", "effort": "xhigh", "tools": "benchmark-specified", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "agentic benchmark harness", "prompt_style": "default", "temperature": "default", "context": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Kimi K3 technical report: OSWorld-Verified = 79.0. Source setting: agentic benchmark harness. Origin: Moonshot evaluation.", "candidates": [ { "score": 78.7, "reference_url": "https://www.anthropic.com/claude-sonnet-5-system-card", "source_type": "model_card", "reported_setting": { "effort": "source does not state", "tools": "none", "sampling": "avg 5 trials unless source states otherwise", "harness": "table8" }, "notes": "Anthropic Claude Sonnet 5 System Card Table 8.1.A, page 115: GPT-5.5; OSWorld-Verified [summary]=78.7. Source setting: effort=source does not state; tools=none; sampling=avg 5 trials unless source states otherwise; harness=table8." }, { "score": 78.7, "reference_url": "https://www.anthropic.com/claude-sonnet-5-system-card", "source_type": "model_card", "reported_setting": { "effort": "source does not state", "tools": "GUI computer-use harness", "sampling": "avg 5 trials unless source states otherwise", "harness": "osworld" }, "notes": "Anthropic Claude Sonnet 5 System Card Figure 8.10.2.A, page 126: GPT-5.5; OSWorld-Verified [summary]=78.7%. Source setting: effort=source does not state; tools=GUI computer-use harness; sampling=avg 5 trials unless source states otherwise; harness=osworld." } ] }, { "model_id": "gpt-5.5", "benchmark_id": "osworld_2_0", "score": 49.5, "reference_url": "https://github.com/MoonshotAI/Kimi-K3/blob/main/k3_tech_report.pdf", "reported_setting": { "mode": "thinking", "effort": "xhigh", "tools": "benchmark-specified", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "agentic benchmark harness", "prompt_style": "default", "temperature": "default", "context": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Kimi K3 technical report: OSWorld 2.0 = 49.5. Source setting: agentic benchmark harness. Origin: Moonshot evaluation.", "candidates": [ { "score": 47.5, "reference_url": "https://openai.com/index/gpt-5-6/", "source_type": "official_blog", "reported_setting": { "effort": "xhigh", "eval_variant": null, "code_mode": "false" }, "notes": "OpenAI GPT-5.6 release table (table 3): GPT‑5.5; OSWorld 2.0=47.5. Exact effort/variant resolved from the rendered chart point." }, { "score": 2.98, "reference_url": "https://openai.com/index/gpt-5-6/", "source_type": "official_blog", "reported_setting": { "effort": "none", "eval_variant": null, "code_mode": "false" }, "notes": "OpenAI GPT-5.6 release figure (OSWorld 2.0): GPT-5.5; OSWorld 2.0=2.98. Figure 15: OSWorld 2.0. Cost, latency, and output-token coordinates are retained as review metadata, not score-matrix metrics." }, { "score": 16.26, "reference_url": "https://openai.com/index/gpt-5-6/", "source_type": "official_blog", "reported_setting": { "effort": "low", "eval_variant": null, "code_mode": "false" }, "notes": "OpenAI GPT-5.6 release figure (OSWorld 2.0): GPT-5.5; OSWorld 2.0=16.26. Figure 15: OSWorld 2.0. Cost, latency, and output-token coordinates are retained as review metadata, not score-matrix metrics." }, { "score": 36.9, "reference_url": "https://openai.com/index/gpt-5-6/", "source_type": "official_blog", "reported_setting": { "effort": "medium", "eval_variant": null, "code_mode": "false" }, "notes": "OpenAI GPT-5.6 release figure (OSWorld 2.0): GPT-5.5; OSWorld 2.0=36.9. Figure 15: OSWorld 2.0. Cost, latency, and output-token coordinates are retained as review metadata, not score-matrix metrics." }, { "score": 40.44, "reference_url": "https://openai.com/index/gpt-5-6/", "source_type": "official_blog", "reported_setting": { "effort": "high", "eval_variant": null, "code_mode": "false" }, "notes": "OpenAI GPT-5.6 release figure (OSWorld 2.0): GPT-5.5; OSWorld 2.0=40.44. Figure 15: OSWorld 2.0. Cost, latency, and output-token coordinates are retained as review metadata, not score-matrix metrics." } ] }, { "model_id": "gpt-5.5", "benchmark_id": "saas_bench", "score": 43.8, "reference_url": "https://github.com/MoonshotAI/Kimi-K3/blob/main/k3_tech_report.pdf", "reported_setting": { "mode": "thinking", "effort": "xhigh", "tools": "benchmark-specified", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "agentic benchmark harness", "prompt_style": "default", "temperature": "default", "context": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Kimi K3 technical report: SaaS-Bench = 43.8. Source setting: agentic benchmark harness. Origin: Moonshot evaluation." }, { "model_id": "gpt-5.5", "benchmark_id": "tau3_banking", "score": 31.3, "reference_url": "https://github.com/MoonshotAI/Kimi-K3/blob/main/k3_tech_report.pdf", "reported_setting": { "mode": "thinking", "effort": "xhigh", "tools": "benchmark-specified", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "agentic benchmark harness", "prompt_style": "default", "temperature": "default", "context": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Kimi K3 technical report: τ³-Banking = 31.3. Source setting: agentic benchmark harness. Origin: Artificial Analysis cited by Kimi report.", "candidates": [ { "score": 31.3, "reference_url": "https://media.x.ai/v1/website/4p5-5184fdf9.pdf", "source_type": "model_card", "reported_setting": { "mode": "thinking", "effort": "xhigh", "tools": "banking APIs and simulated user", "sampling": "pass@1", "judge": "final-state task accuracy", "harness": "Artificial Analysis tau3-banking harness", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown" }, "notes": "card45 tau3-banking Accuracy (%); source effort=xhigh; benchmark protocol matches canonical=True. Unknown source fields remain unknown." } ] }, { "model_id": "gpt-5.5", "benchmark_id": "harvey_lab_aa", "score": 86.3, "reference_url": "https://github.com/MoonshotAI/Kimi-K3/blob/main/k3_tech_report.pdf", "reported_setting": { "mode": "thinking", "effort": "xhigh", "tools": "benchmark-specified", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "agentic benchmark harness", "prompt_style": "default", "temperature": "default", "context": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Kimi K3 technical report: Harvey Lab-AA = 86.3. Source setting: agentic benchmark harness. Origin: Artificial Analysis cited by Kimi report." }, { "model_id": "gpt-5.5", "benchmark_id": "corpfin_v2", "score": 68.4, "reference_url": "https://github.com/MoonshotAI/Kimi-K3/blob/main/k3_tech_report.pdf", "reported_setting": { "mode": "thinking", "effort": "xhigh", "tools": "benchmark-specified", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "agentic benchmark harness", "prompt_style": "default", "temperature": "default", "context": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Kimi K3 technical report: CorpFin v2 = 68.4. Source setting: agentic benchmark harness. Origin: Vals AI cited by Kimi report." }, { "model_id": "gpt-5.5", "benchmark_id": "finance_agent_v2", "score": 51.8, "reference_url": "https://github.com/MoonshotAI/Kimi-K3/blob/main/k3_tech_report.pdf", "reported_setting": { "mode": "thinking", "effort": "xhigh", "tools": "benchmark-specified", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "agentic benchmark harness", "prompt_style": "default", "temperature": "default", "context": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Kimi K3 technical report: Finance Agent v2 = 51.8. Source setting: agentic benchmark harness. Origin: Vals AI cited by Kimi report." }, { "model_id": "gpt-5.5", "benchmark_id": "legal_research_bench", "score": 40.4, "reference_url": "https://github.com/MoonshotAI/Kimi-K3/blob/main/k3_tech_report.pdf", "reported_setting": { "mode": "thinking", "effort": "xhigh", "tools": "benchmark-specified", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "agentic benchmark harness", "prompt_style": "default", "temperature": "default", "context": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Kimi K3 technical report: Legal Research Bench = 40.4. Source setting: agentic benchmark harness. Origin: Vals AI cited by Kimi report." }, { "model_id": "gpt-5.5", "benchmark_id": "worldvqa", "score": 38.5, "reference_url": "https://github.com/MoonshotAI/Kimi-K3/blob/main/k3_tech_report.pdf", "reported_setting": { "mode": "thinking", "effort": "xhigh", "tools": "none", "sampling": "avg-of-3", "judge": "benchmark-specified", "harness": "forced-answer prompt; avg 3 runs", "prompt_style": "default", "temperature": "default", "context": "default" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Kimi K3 technical report: WorldVQA ForceAnswer = 38.5. Source setting: forced-answer prompt; avg 3 runs. Origin: Moonshot evaluation.", "candidates": [ { "score": 34.6, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2.1/Seed2_1_Model_Card.pdf", "source_type": "tech_report", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "none", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official", "prompt_style": "official default", "temperature": "unknown" }, "notes": "Model card Table 9 (p51)." }, { "score": 54.58, "reference_url": "https://static.stepfun.com/blog/step-3.7-flash/", "source_type": "official_blog", "reported_setting": { "quantization_config": "common release evaluation; artifact precision not established", "configuration": "base Step evaluation configuration", "mode": "source does not state", "effort": "source does not state", "tools_search": "Visual Search", "shots": "source does not state", "trials_samples": "source does not state", "aggregation_pass_k": "source-reported score", "context": "source does not state", "output_cap": "source does not state", "temperature": "source does not state", "top_p": "source does not state", "judge": "LLM-as-judge; exact StepFun judge not stated", "harness_agent": "official StepFun evaluation; exact harness not stated", "dataset_version_split": "WorldVQA first-eight-category 3,000-item release", "multimodal_input": true }, "notes": "Official StepFun benchmark-specific tool table." } ] }, { "model_id": "gpt-5.5", "benchmark_id": "omnidocbench", "score": 0.106, "reference_url": "https://github.com/MoonshotAI/Kimi-K3/blob/main/k3_tech_report.pdf", "reported_setting": { "mode": "thinking", "effort": "xhigh", "tools": "benchmark-specified", "sampling": "avg-of-3", "judge": "benchmark-specified", "harness": "avg 3 runs; source 1-NED score normalized to NED", "prompt_style": "default", "temperature": "default", "context": "default" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Kimi K3 technical report: OmniDocBench = 0.106 NED. Source setting: avg 3 runs; source 1-NED score normalized to NED. Origin: Moonshot evaluation. The source reports 1-NED=89.4%; stored value is normalized NED=0.106." }, { "model_id": "gpt-5.5", "benchmark_id": "perception_bench", "score": 55.8, "reference_url": "https://github.com/MoonshotAI/Kimi-K3/blob/main/k3_tech_report.pdf", "reported_setting": { "mode": "thinking", "effort": "xhigh", "tools": "benchmark-specified", "sampling": "avg-of-3", "judge": "benchmark-specified", "harness": "multimodal; avg 3 runs", "prompt_style": "default", "temperature": "default", "context": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Kimi K3 technical report: PerceptionBench = 55.8. Source setting: multimodal; avg 3 runs. Origin: Moonshot evaluation." }, { "model_id": "gpt-5.5", "benchmark_id": "video_mme", "score": 89.3, "reference_url": "https://github.com/MoonshotAI/Kimi-K3/blob/main/k3_tech_report.pdf", "reported_setting": { "mode": "thinking", "effort": "xhigh", "tools": "none", "sampling": "avg-of-3", "judge": "benchmark-specified", "harness": "audio+visual+subtitles; avg 3 runs", "prompt_style": "default", "temperature": "default", "context": "default" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Kimi K3 technical report: Video-MME (w. sub) = 89.3. Source setting: audio+visual+subtitles; avg 3 runs. Origin: Moonshot evaluation." }, { "model_id": "gpt-5.5", "benchmark_id": "mmvu", "score": 81.7, "reference_url": "https://github.com/MoonshotAI/Kimi-K3/blob/main/k3_tech_report.pdf", "reported_setting": { "mode": "thinking", "effort": "xhigh", "tools": "benchmark-specified", "sampling": "avg-of-3", "judge": "benchmark-specified", "harness": "avg 3 runs", "prompt_style": "default", "temperature": "default", "context": "default" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Kimi K3 technical report: MMVU = 81.7. Source setting: avg 3 runs. Origin: Moonshot evaluation." }, { "model_id": "gpt-5.5", "benchmark_id": "babyvision", "score": 83.6, "reference_url": "https://github.com/MoonshotAI/Kimi-K3/blob/main/k3_tech_report.pdf", "reported_setting": { "mode": "thinking", "effort": "xhigh", "tools": "python", "sampling": "avg-of-3", "judge": "benchmark-specified", "harness": "Python tool; avg 3 runs", "prompt_style": "default", "temperature": "default", "context": "default" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Kimi K3 technical report: BabyVision w/ python = 83.6. Source setting: Python tool; avg 3 runs. Origin: Moonshot evaluation.", "candidates": [ { "score": 55.9, "reference_url": "https://seed.bytedance.com/en/blog/seed2-1-officially-released-advancing-ai-productivity", "source_type": "official_blog", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "none", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official", "prompt_style": "official default", "temperature": "unknown" }, "notes": "Release blog EN: Vision capability chart (https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/user-upload/4xfa4mqq4yer3.png)." } ] }, { "model_id": "gpt-5.5", "benchmark_id": "charxiv_reasoning", "score": 84.1, "reference_url": "https://github.com/MoonshotAI/Kimi-K3/blob/main/k3_tech_report.pdf", "reported_setting": { "mode": "thinking", "effort": "xhigh", "tools": "none", "sampling": "avg-of-3", "judge": "benchmark-specified", "harness": "vision; avg 3 runs", "prompt_style": "default", "temperature": "default", "context": "default" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Kimi K3 technical report: CharXiv (RQ) (no Python) = 84.1. Source setting: vision; avg 3 runs. Origin: Moonshot evaluation.", "candidates": [ { "score": 89.0, "reference_url": "https://github.com/MoonshotAI/Kimi-K3/blob/main/k3_tech_report.pdf", "source_type": "tech_report", "reported_setting": { "mode": "thinking", "effort": "xhigh", "tools": "python", "sampling": "avg-of-3", "judge": "benchmark-specified", "harness": "Python tool; avg 3 runs", "prompt_style": "default", "temperature": "default", "context": "default" }, "notes": "Kimi K3 technical report: CharXiv (RQ) (with Python) = 89.0. Source setting: Python tool; avg 3 runs. Origin: Moonshot evaluation." }, { "score": 84.8, "reference_url": "https://ai.meta.com/static-resource/muse-spark-1.1-evaluation-report", "source_type": "tech_report", "reported_setting": { "mode": "thinking", "effort": "xhigh", "tools": "Python", "sampling": "pass@1 unless source row states otherwise", "harness": "Meta MSL general-capability evaluation", "temperature": "1.0 for Muse; provider/source setting for comparisons", "context": "benchmark-specific", "configuration": "default" }, "notes": "Figure 44 · CharXiv Reasoning w/code execution; metric=headline_metric. Exact printed value in the general-capability summary." }, { "score": 83.2, "reference_url": "https://seed.bytedance.com/en/blog/seed2-1-officially-released-advancing-ai-productivity", "source_type": "official_blog", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "none", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official", "prompt_style": "official default", "temperature": "unknown" }, "notes": "Release blog EN: Vision capability chart (https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/user-upload/4xfa4mqq4yer3.png)." } ] }, { "model_id": "gpt-5.5", "benchmark_id": "mathvision", "score": 92.2, "reference_url": "https://github.com/MoonshotAI/Kimi-K3/blob/main/k3_tech_report.pdf", "reported_setting": { "mode": "thinking", "effort": "xhigh", "tools": "none", "sampling": "avg-of-3", "judge": "benchmark-specified", "harness": "vision; avg 3 runs", "prompt_style": "default", "temperature": "default", "context": "default" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Kimi K3 technical report: MathVision (no Python) = 92.2. Source setting: vision; avg 3 runs. Origin: Moonshot evaluation.", "candidates": [ { "score": 96.8, "reference_url": "https://github.com/MoonshotAI/Kimi-K3/blob/main/k3_tech_report.pdf", "source_type": "tech_report", "reported_setting": { "mode": "thinking", "effort": "xhigh", "tools": "python", "sampling": "avg-of-3", "judge": "benchmark-specified", "harness": "Python tool; avg 3 runs", "prompt_style": "default", "temperature": "default", "context": "default" }, "notes": "Kimi K3 technical report: MathVision (with Python) = 96.8. Source setting: Python tool; avg 3 runs. Origin: Moonshot evaluation." }, { "score": 92.2, "reference_url": "https://seed.bytedance.com/en/blog/seed2-1-officially-released-advancing-ai-productivity", "source_type": "official_blog", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "none", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official", "prompt_style": "official default", "temperature": "unknown" }, "notes": "Release blog EN: Vision capability chart (https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/user-upload/4xfa4mqq4yer3.png)." } ] }, { "model_id": "gpt-5.5", "benchmark_id": "zerobench_main", "score": 22.0, "reference_url": "https://github.com/MoonshotAI/Kimi-K3/blob/main/k3_tech_report.pdf", "reported_setting": { "mode": "thinking", "effort": "xhigh", "tools": "benchmark-specified", "sampling": "pass@5", "judge": "benchmark-specified", "harness": "no Python; pass@5 / 5 runs", "prompt_style": "default", "temperature": "default", "context": "default" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Kimi K3 technical report: ZeroBench (pass@5) (no Python) = 22.0. Source setting: no Python; pass@5 / 5 runs. Origin: Moonshot evaluation. Canonical compatibility remains under review; stored as best available primary with matches_canonical=false.", "candidates": [ { "score": 13.0, "reference_url": "https://seed.bytedance.com/en/blog/seed2-1-officially-released-advancing-ai-productivity", "source_type": "official_blog", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "none", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official", "prompt_style": "official default", "temperature": "unknown" }, "notes": "Release blog EN: Vision capability chart (https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/user-upload/4xfa4mqq4yer3.png)." } ] }, { "model_id": "gpt-5.5", "benchmark_id": "zerobench_tools", "score": 41.0, "reference_url": "https://github.com/MoonshotAI/Kimi-K3/blob/main/k3_tech_report.pdf", "reported_setting": { "mode": "thinking", "effort": "xhigh", "tools": "python", "sampling": "pass@5", "judge": "benchmark-specified", "harness": "Python tool; pass@5 / 5 runs", "prompt_style": "default", "temperature": "default", "context": "default" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Kimi K3 technical report: ZeroBench (pass@5) (with Python) = 41.0. Source setting: Python tool; pass@5 / 5 runs. Origin: Moonshot evaluation. Canonical compatibility remains under review; stored as best available primary with matches_canonical=false." }, { "model_id": "gpt-5.5", "benchmark_id": "coding_experience", "score": 56.8, "reference_url": "https://github.com/MoonshotAI/Kimi-K3/blob/main/k3_tech_report.pdf", "reported_setting": { "mode": "thinking", "effort": "xhigh", "tools": "benchmark-specified", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "provider-aligned harness: Kimi Code / Claude Code / Codex", "prompt_style": "default", "temperature": "default", "context": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Kimi K3 technical report: Coding Experience = 56.8. Source setting: provider-aligned harness: Kimi Code / Claude Code / Codex. Origin: Moonshot evaluation." }, { "model_id": "gpt-5.5", "benchmark_id": "clawbench_2_0", "score": 48.5, "reference_url": "https://github.com/MoonshotAI/Kimi-K3/blob/main/k3_tech_report.pdf", "reported_setting": { "mode": "thinking", "effort": "xhigh", "tools": "benchmark-specified", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "OpenClaw harness", "prompt_style": "default", "temperature": "default", "context": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Kimi K3 technical report: 24/7 ClawBench 2.0 = 48.5. Source setting: OpenClaw harness. Origin: Moonshot evaluation." }, { "model_id": "gpt-5.5", "benchmark_id": "mira_bench", "score": 54.6, "reference_url": "https://github.com/MoonshotAI/Kimi-K3/blob/main/k3_tech_report.pdf", "reported_setting": { "mode": "thinking", "effort": "xhigh", "tools": "benchmark-specified", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "MIRA harness", "prompt_style": "default", "temperature": "default", "context": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Kimi K3 technical report: MIRA Bench = 54.6. Source setting: MIRA harness. Origin: Moonshot evaluation." }, { "model_id": "gpt-5.5", "benchmark_id": "kaet", "score": 79.7, "reference_url": "https://github.com/MoonshotAI/Kimi-K3/blob/main/k3_tech_report.pdf", "reported_setting": { "mode": "thinking", "effort": "xhigh", "tools": "benchmark-specified", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "Kimi Code harness", "prompt_style": "default", "temperature": "default", "context": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Kimi K3 technical report: KAET = 79.7. Source setting: Kimi Code harness. Origin: Moonshot evaluation." }, { "model_id": "gpt-5.5", "benchmark_id": "clif_bench", "score": 52.3, "reference_url": "https://github.com/MoonshotAI/Kimi-K3/blob/main/k3_tech_report.pdf", "reported_setting": { "mode": "thinking", "effort": "xhigh", "tools": "benchmark-specified", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "Kimi Code harness", "prompt_style": "default", "temperature": "default", "context": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Kimi K3 technical report: CLIF Bench = 52.3. Source setting: Kimi Code harness. Origin: Moonshot evaluation." }, { "model_id": "gpt-5.5", "benchmark_id": "agentic_vision_bench", "score": 76.9, "reference_url": "https://github.com/MoonshotAI/Kimi-K3/blob/main/k3_tech_report.pdf", "reported_setting": { "mode": "thinking", "effort": "xhigh", "tools": "benchmark-specified", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "Kimi Code harness", "prompt_style": "default", "temperature": "default", "context": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Kimi K3 technical report: Agentic Vision Bench = 76.9. Source setting: Kimi Code harness. Origin: Moonshot evaluation." }, { "model_id": "gpt-5.5", "benchmark_id": "swarm_bench", "score": 61.8, "reference_url": "https://github.com/MoonshotAI/Kimi-K3/blob/main/k3_tech_report.pdf", "reported_setting": { "mode": "thinking", "effort": "xhigh", "tools": "benchmark-specified", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "Kimi Agent harness", "prompt_style": "default", "temperature": "default", "context": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Kimi K3 technical report: SwarmBench = 61.8. Source setting: Kimi Agent harness. Origin: Moonshot evaluation." }, { "model_id": "gpt-5.5", "benchmark_id": "online_experience", "score": 73.7, "reference_url": "https://github.com/MoonshotAI/Kimi-K3/blob/main/k3_tech_report.pdf", "reported_setting": { "mode": "thinking", "effort": "xhigh", "tools": "benchmark-specified", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "Kimi Agent harness", "prompt_style": "default", "temperature": "default", "context": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Kimi K3 technical report: Online Experience = 73.7. Source setting: Kimi Agent harness. Origin: Moonshot evaluation." }, { "model_id": "gpt-5.5", "benchmark_id": "deepresearchbench", "score": 81.9, "reference_url": "https://github.com/MoonshotAI/Kimi-K3/blob/main/k3_tech_report.pdf", "reported_setting": { "mode": "thinking", "effort": "xhigh", "tools": "benchmark-specified", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "Kimi Agent harness", "prompt_style": "default", "temperature": "default", "context": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Kimi K3 technical report: DeepResearchBench = 81.9. Source setting: Kimi Agent harness. Origin: Moonshot evaluation." }, { "model_id": "gpt-5.5", "benchmark_id": "finance_bench", "score": 58.4, "reference_url": "https://github.com/MoonshotAI/Kimi-K3/blob/main/k3_tech_report.pdf", "reported_setting": { "mode": "thinking", "effort": "xhigh", "tools": "benchmark-specified", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "N/A harness", "prompt_style": "default", "temperature": "default", "context": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Kimi K3 technical report: Finance Bench = 58.4. Source setting: N/A harness. Origin: Moonshot evaluation." }, { "model_id": "gpt-5.5", "benchmark_id": "kwv_bench", "score": 65.8, "reference_url": "https://github.com/MoonshotAI/Kimi-K3/blob/main/k3_tech_report.pdf", "reported_setting": { "mode": "thinking", "effort": "xhigh", "tools": "benchmark-specified", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "N/A harness", "prompt_style": "default", "temperature": "default", "context": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Kimi K3 technical report: KWVBench = 65.8. Source setting: N/A harness. Origin: Moonshot evaluation." }, { "model_id": "gpt-5.5", "benchmark_id": "deck_bench", "score": 68.2, "reference_url": "https://github.com/MoonshotAI/Kimi-K3/blob/main/k3_tech_report.pdf", "reported_setting": { "mode": "thinking", "effort": "xhigh", "tools": "benchmark-specified", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "N/A harness", "prompt_style": "default", "temperature": "default", "context": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Kimi K3 technical report: DECKBench = 68.2. Source setting: N/A harness. Origin: Moonshot evaluation." }, { "model_id": "gpt-5.5", "benchmark_id": "agent_behavior_bench", "score": 70.1, "reference_url": "https://github.com/MoonshotAI/Kimi-K3/blob/main/k3_tech_report.pdf", "reported_setting": { "mode": "thinking", "effort": "xhigh", "tools": "benchmark-specified", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "Kimi Work harness", "prompt_style": "default", "temperature": "default", "context": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Kimi K3 technical report: Agent Behavior Bench = 70.1. Source setting: Kimi Work harness. Origin: Moonshot evaluation." }, { "model_id": "gpt-5.5", "benchmark_id": "faithfulness", "score": 86.5, "reference_url": "https://github.com/MoonshotAI/Kimi-K3/blob/main/k3_tech_report.pdf", "reported_setting": { "mode": "thinking", "effort": "xhigh", "tools": "benchmark-specified", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "N/A harness", "prompt_style": "default", "temperature": "default", "context": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Kimi K3 technical report: Faithfulness (1 - hallucination rate) = 86.5. Source setting: N/A harness. Origin: Moonshot evaluation." }, { "model_id": "gpt-5.5", "benchmark_id": "chat_all_in_one_bench", "score": 71.8, "reference_url": "https://github.com/MoonshotAI/Kimi-K3/blob/main/k3_tech_report.pdf", "reported_setting": { "mode": "thinking", "effort": "xhigh", "tools": "benchmark-specified", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "Kimi Work harness", "prompt_style": "default", "temperature": "default", "context": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Kimi K3 technical report: Chat All-in-One Bench = 71.8. Source setting: Kimi Work harness. Origin: Moonshot evaluation." }, { "model_id": "gpt-5.5", "benchmark_id": "aa_intelligence_index_v4_1", "score": 55.0, "reference_url": "https://github.com/MoonshotAI/Kimi-K3/blob/main/k3_tech_report.pdf", "reported_setting": { "mode": "thinking", "effort": "xhigh", "tools": "benchmark-specified", "sampling": "leaderboard snapshot 2026-07-23", "judge": "benchmark-specified", "harness": "live leaderboard snapshot as of 2026-07-23", "prompt_style": "default", "temperature": "default", "context": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Kimi K3 technical report: AA Intelligence Index v4.1 = 55.0. Source setting: live leaderboard snapshot as of 2026-07-23. Origin: third-party leaderboard cited by Kimi report.", "candidates": [ { "score": 54.8, "reference_url": "https://openai.com/index/gpt-5-6/", "source_type": "official_blog", "reported_setting": { "effort": "xhigh", "eval_variant": null, "code_mode": "false" }, "notes": "OpenAI GPT-5.6 release table (table 0): GPT‑5.5; Artificial Analysis Intelligence Index v4.1=54.8. Exact effort/variant resolved from the rendered chart point." }, { "score": 35.38, "reference_url": "https://openai.com/index/gpt-5-6/", "source_type": "official_blog", "reported_setting": { "effort": "none", "eval_variant": null, "code_mode": "false" }, "notes": "OpenAI GPT-5.6 release figure (Artificial Analysis Intelligence Index v4.1): GPT-5.5; Artificial Analysis Intelligence Index v4.1=35.38. Figure 0: Composite of nine independent evaluations spanning agentic work, coding, scientific reasoning, and general capabilities.. Cost, latency, and output-token coordinates are retained as review metadata, not score-matrix metrics." }, { "score": 43.46, "reference_url": "https://openai.com/index/gpt-5-6/", "source_type": "official_blog", "reported_setting": { "effort": "low", "eval_variant": null, "code_mode": "false" }, "notes": "OpenAI GPT-5.6 release figure (Artificial Analysis Intelligence Index v4.1): GPT-5.5; Artificial Analysis Intelligence Index v4.1=43.46. Figure 0: Composite of nine independent evaluations spanning agentic work, coding, scientific reasoning, and general capabilities.. Cost, latency, and output-token coordinates are retained as review metadata, not score-matrix metrics." }, { "score": 50.41, "reference_url": "https://openai.com/index/gpt-5-6/", "source_type": "official_blog", "reported_setting": { "effort": "medium", "eval_variant": null, "code_mode": "false" }, "notes": "OpenAI GPT-5.6 release figure (Artificial Analysis Intelligence Index v4.1): GPT-5.5; Artificial Analysis Intelligence Index v4.1=50.41. Figure 0: Composite of nine independent evaluations spanning agentic work, coding, scientific reasoning, and general capabilities.. Cost, latency, and output-token coordinates are retained as review metadata, not score-matrix metrics." }, { "score": 53.13, "reference_url": "https://openai.com/index/gpt-5-6/", "source_type": "official_blog", "reported_setting": { "effort": "high", "eval_variant": null, "code_mode": "false" }, "notes": "OpenAI GPT-5.6 release figure (Artificial Analysis Intelligence Index v4.1): GPT-5.5; Artificial Analysis Intelligence Index v4.1=53.13. Figure 0: Composite of nine independent evaluations spanning agentic work, coding, scientific reasoning, and general capabilities.. Cost, latency, and output-token coordinates are retained as review metadata, not score-matrix metrics." } ] }, { "model_id": "gpt-5.5", "benchmark_id": "vals_index", "score": 68.0, "reference_url": "https://github.com/MoonshotAI/Kimi-K3/blob/main/k3_tech_report.pdf", "reported_setting": { "mode": "thinking", "effort": "xhigh", "tools": "benchmark-specified", "sampling": "leaderboard snapshot 2026-07-23", "judge": "benchmark-specified", "harness": "live leaderboard snapshot as of 2026-07-23", "prompt_style": "default", "temperature": "default", "context": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Kimi K3 technical report: Vals Index = 68.0. Source setting: live leaderboard snapshot as of 2026-07-23. Origin: third-party leaderboard cited by Kimi report.", "candidates": [ { "score": 68.0, "reference_url": "https://media.x.ai/v1/website/card-4p6-4cd2dc57.pdf", "source_type": "model_card", "reported_setting": { "mode": "thinking", "effort": "xhigh", "tools": "benchmark-specified professional tools", "sampling": "unknown", "judge": "Vals weighted suite score", "harness": "Vals Index live harness", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown" }, "notes": "card46 Vals Index Suite score (%); source effort=xhigh; benchmark protocol matches canonical=True. Unknown source fields remain unknown." } ] }, { "model_id": "gpt-5.5", "benchmark_id": "webdev_arena_elo", "score": 1507.0, "reference_url": "https://github.com/MoonshotAI/Kimi-K3/blob/main/k3_tech_report.pdf", "reported_setting": { "mode": "thinking", "effort": "xhigh", "tools": "benchmark-specified", "sampling": "leaderboard snapshot 2026-07-23", "judge": "benchmark-specified", "harness": "live leaderboard snapshot as of 2026-07-23", "prompt_style": "default", "temperature": "default", "context": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Kimi K3 technical report: WebDevArena Elo = 1507. Source setting: live leaderboard snapshot as of 2026-07-23. Origin: third-party leaderboard cited by Kimi report." }, { "model_id": "gpt-5.5", "benchmark_id": "text_arena_elo", "score": 1482.0, "reference_url": "https://github.com/MoonshotAI/Kimi-K3/blob/main/k3_tech_report.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "benchmark-specified", "sampling": "leaderboard snapshot 2026-07-23", "judge": "benchmark-specified", "harness": "live leaderboard snapshot as of 2026-07-23", "prompt_style": "default", "temperature": "default", "context": "default" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Kimi K3 technical report: TextArena Elo = 1482. Source setting: live leaderboard snapshot as of 2026-07-23. Origin: third-party leaderboard cited by Kimi report." }, { "model_id": "gpt-5.5", "benchmark_id": "agent_arena", "score": 8.8, "reference_url": "https://github.com/MoonshotAI/Kimi-K3/blob/main/k3_tech_report.pdf", "reported_setting": { "mode": "thinking", "effort": "xhigh", "tools": "benchmark-specified", "sampling": "leaderboard snapshot 2026-07-23", "judge": "benchmark-specified", "harness": "live leaderboard snapshot as of 2026-07-23", "prompt_style": "default", "temperature": "default", "context": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Kimi K3 technical report: AgentArena = 8.8. Source setting: live leaderboard snapshot as of 2026-07-23. Origin: third-party leaderboard cited by Kimi report." }, { "model_id": "glm-5.2", "benchmark_id": "gpqa_diamond", "score": 91.2, "reference_url": "https://z.ai/blog/glm-5.2", "reported_setting": { "mode": "thinking", "effort": "max", "sampling": "pass@1", "prompt_style": "default", "tools": "none", "temperature": "1.0", "top_p": "0.95", "context": "default", "max_output_tokens": "163840", "judge": "GPT-5.5 medium for AIME/HMMT/IMO; benchmark-specified otherwise", "harness": "official reasoning evaluation" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "GLM-5.2 official release source: GLM-5.2 / GPQA-Diamond = 91.2.", "candidates": [] }, { "model_id": "glm-5.2", "benchmark_id": "critpt", "score": 20.9, "reference_url": "https://z.ai/blog/glm-5.2", "reported_setting": { "mode": "thinking", "effort": "max", "sampling": "pass@1", "prompt_style": "default", "tools": "none", "temperature": "1.0", "top_p": "0.95", "context": "default", "max_output_tokens": "163840", "judge": "GPT-5.5 medium for AIME/HMMT/IMO; benchmark-specified otherwise", "harness": "official reasoning evaluation" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "GLM-5.2 official release source: GLM-5.2 / CritPt = 20.9.", "candidates": [] }, { "model_id": "glm-5.2", "benchmark_id": "aa_lcr", "score": 71.3, "reference_url": "https://github.com/MoonshotAI/Kimi-K3/blob/main/k3_tech_report.pdf", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "none", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "max/xhigh; no tools", "prompt_style": "default", "temperature": "default", "context": "1048576 (1M)" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Kimi K3 technical report: AA-LCR = 71.3. Source setting: max/xhigh; no tools. Origin: Artificial Analysis cited by Kimi report.", "candidates": [ { "score": 73.4, "reference_url": "https://huggingface.co/tencent/Hy3", "source_type": "model_card", "reported_setting": { "mode": "thinking", "effort": "high / highest available", "tools": "none", "sampling": "repeats=3 per question", "judge": "Equality Checker LLM", "harness": "Artificial Analysis AA-LCR, 100 questions", "prompt_style": "default", "temperature": "0.9 for Hy3 family; otherwise source/provider setting", "context": "256000 for Hy3 family; otherwise source/provider setting", "notes": "Tencent own test" }, "notes": "Hy3 official model-card appendix: AA-LCR = 73.4*. Tencent own testing." } ] }, { "model_id": "glm-5.2", "benchmark_id": "deep_swe_v1_1", "score": 46.2, "reference_url": "https://z.ai/blog/glm-5.2", "reported_setting": { "mode": "thinking", "effort": "max", "sampling": "pass@1", "prompt_style": "default", "tools": "agentic code execution", "temperature": "1.0", "top_p": "1.0", "context": "400000", "judge": "isolated-container verification", "harness": "official pier + mini-swe-agent; timeout=2h; 2 CPU/8GB; no internet" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "GLM-5.2 official release source: GLM-5.2 / DeepSWE = 46.2.", "candidates": [ { "score": 42.5, "reference_url": "https://huggingface.co/tencent/Hy3", "source_type": "model_card", "reported_setting": { "mode": "thinking", "effort": "high / highest available", "tools": "agentic code execution", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "mini-swe-agent; 2h/task; 2 CPU/8GB; no internet", "prompt_style": "default", "temperature": "0.9 for Hy3 family; otherwise source/provider setting", "context": "256000 for Hy3 family; otherwise source/provider setting", "notes": "Tencent own test" }, "notes": "Hy3 official model-card appendix: DeepSWE = 42.5*. Tencent own testing." }, { "score": 44.0, "reference_url": "https://media.x.ai/v1/website/4p5-5184fdf9.pdf", "source_type": "model_card", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "agentic repository shell/editor", "sampling": "pass@1", "judge": "isolated repository verifier", "harness": "mini-swe-agent run by Datacurve", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown" }, "notes": "card45 DeepSWE v1.1 Pass@1 (%); source effort=max; benchmark protocol matches canonical=True. Unknown source fields remain unknown." }, { "score": 44.0, "reference_url": "https://x.ai/news/grok-4-5", "source_type": "official_blog", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "agentic repository shell/editor", "sampling": "pass@1", "judge": "isolated repository verifier", "harness": "mini-swe-agent run by Datacurve", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown" }, "notes": "release45 DeepSWE 1.1 Pass@1 (%); source effort=unknown; benchmark protocol matches canonical=True. Unknown source fields remain unknown." } ] }, { "model_id": "glm-5.2", "benchmark_id": "program_bench", "score": 63.7, "reference_url": "https://z.ai/blog/glm-5.2", "reported_setting": { "mode": "thinking", "effort": "max", "sampling": "pass@1", "prompt_style": "default", "tools": "agentic code execution", "temperature": "1.0", "top_p": "1.0", "context": "400000", "max_output_tokens": "64000", "judge": "248,000+ fuzz-generated behavioral tests", "harness": "Claude Code 2.1.156; max_turns=2000; timeout=6h; 4 CPU/8GB; no internet" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "GLM-5.2 official release source: GLM-5.2 / ProgramBench = 63.7.", "candidates": [] }, { "model_id": "glm-5.2", "benchmark_id": "terminal_bench_2_1", "score": 82.7, "reference_url": "https://z.ai/blog/glm-5.2", "reported_setting": { "mode": "thinking", "effort": "max", "sampling": "pass@1", "prompt_style": "default", "tools": "terminal agent", "temperature": "1.0 for Claude Code; unknown otherwise", "top_p": "0.95 for Claude Code; unknown otherwise", "context": "source-reported", "max_output_tokens": "131072 for Claude Code; source-reported otherwise", "judge": "official task verifier", "harness": "Claude Code 2.1.167; Claude Code values averaged over 5 runs" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "GLM-5.2 official release source: GLM-5.2 / Terminal Bench 2.1 (Best Reported Harness) = 82.7.", "candidates": [ { "score": 81.0, "reference_url": "https://z.ai/blog/glm-5.2", "source_type": "official_blog", "reported_setting": { "mode": "thinking", "effort": "max", "sampling": "pass@1", "prompt_style": "default", "tools": "terminal agent", "temperature": "1.0", "top_p": "1.0", "context": "256000", "max_output_tokens": "48000", "judge": "official task verifier", "harness": "Terminus-2; parser=json; timeout=4h; max_episodes=500; 4 CPU/8GB" }, "notes": "Z.ai first-party Terminus-2 result." }, { "score": 77.3, "reference_url": "https://huggingface.co/tencent/Hy3", "source_type": "model_card", "reported_setting": { "mode": "thinking", "effort": "high / highest available", "tools": "terminal/code execution", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "Terminus-2; parser=xml; timeout=4h; CPU=16 cores; memory=32GB; max episodes=500", "prompt_style": "default", "temperature": "0.9 for Hy3 family; otherwise source/provider setting", "context": "256000 for Hy3 family; otherwise source/provider setting", "notes": "Tencent own test" }, "notes": "Hy3 official model-card appendix: Terminal-Bench 2.1 = 77.3*. Tencent own testing." }, { "score": 77.9, "reference_url": "https://huggingface.co/internlm/Intern-S2-Preview-397B", "source_type": "model_card", "reported_setting": { "mode": "source does not state", "effort": "source does not state", "tools": "terminal/code execution", "sampling": "source does not state", "judge": "benchmark-specific; exact version not stated", "harness": "Terminus 2", "prompt_style": "source does not state", "temperature": "source does not state", "context": "source does not state", "notes": "Shanghai AI Lab evaluated all model columns using OpenCompass, VLMEvalKit, and AgentCompass. Per-model mode/effort/sampling are not reported; only Intern-S2 maximum evaluation lengths are disclosed." }, "notes": "Intern-S2-Preview-397B official model card: TerminalBench 2.1 = 77.9. This is a Shanghai AI Lab cross-model evaluation; the source does not disclose this model's exact mode, effort, or sampling." } ] }, { "model_id": "glm-5.2", "benchmark_id": "frontier_swe", "score": 74.4, "reference_url": "https://z.ai/blog/glm-5.2", "reported_setting": { "mode": "thinking", "effort": "max", "sampling": "pass@1", "prompt_style": "default", "tools": "agentic code execution", "context": "1000000", "max_output_tokens": "128000", "judge": "FrontierSWE dominance", "harness": "Proximal evaluation; max effort; as of 2026-06-16", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "GLM-5.2 official release source: GLM-5.2 / FrontierSWE (Dominance) = 74.4.", "candidates": [ { "score": 67.3, "reference_url": "https://github.com/MoonshotAI/Kimi-K3/blob/main/k3_tech_report.pdf", "source_type": "tech_report", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "benchmark-specified", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "agentic coding harness per row", "prompt_style": "default", "temperature": "default", "context": "1048576 (1M)" }, "notes": "Kimi K3 technical report: FrontierSWE = 67.3. Source setting: agentic coding harness per row. Origin: Moonshot/OpenAI runs plus official leaderboard. [Displaced by GLM-5.2 first-party audit.]" }, { "score": 72.0, "reference_url": "https://media.x.ai/v1/website/4p5-5184fdf9.pdf", "source_type": "model_card", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "agentic code execution", "sampling": "mean@5", "judge": "continuous partial-credit task scoring and dominance", "harness": "public FrontierSWE harness; Grok CLI for Grok 4.5", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown" }, "notes": "card45 FrontierSWE Dominance (%); source effort=max; benchmark protocol matches canonical=True. Unknown source fields remain unknown." } ] }, { "model_id": "glm-5.2", "benchmark_id": "swe_marathon_h20_2026_07_09", "score": 13.0, "reference_url": "https://github.com/MoonshotAI/Kimi-K3/blob/main/k3_tech_report.pdf", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "benchmark-specified", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "H20-calibrated pre-final-v1.1 branch (2026-07-09)", "prompt_style": "default", "temperature": "default", "context": "1048576 (1M)" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Kimi K3 technical report: SWE-Marathon = 13.0. Source setting: H20-calibrated pre-final-v1.1 branch (2026-07-09). Origin: Moonshot evaluation." }, { "model_id": "glm-5.2", "benchmark_id": "posttrain_bench", "score": 34.3, "reference_url": "https://z.ai/blog/glm-5.2", "reported_setting": { "mode": "thinking", "effort": "max", "sampling": "pass@1", "prompt_style": "default", "tools": "agentic post-training", "context": "1000000", "max_output_tokens": "128000", "judge": "official weighted benchmark score", "harness": "PostTrainBench evaluation; max effort", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "GLM-5.2 official release source: GLM-5.2 / PostTrainBench = 34.3.", "candidates": [] }, { "model_id": "glm-5.2", "benchmark_id": "mls_bench_lite", "score": 40.4, "reference_url": "https://github.com/MoonshotAI/Kimi-K3/blob/main/k3_tech_report.pdf", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "benchmark-specified", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "agentic coding harness per row", "prompt_style": "default", "temperature": "default", "context": "1048576 (1M)" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Kimi K3 technical report: MLS-Bench-Lite = 40.4. Source setting: agentic coding harness per row. Origin: Moonshot evaluation." }, { "model_id": "glm-5.2", "benchmark_id": "scicode", "score": 50.5, "reference_url": "https://github.com/MoonshotAI/Kimi-K3/blob/main/k3_tech_report.pdf", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "benchmark-specified", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "agentic coding harness per row", "prompt_style": "default", "temperature": "default", "context": "1048576 (1M)" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Kimi K3 technical report: SciCode = 50.5. Source setting: agentic coding harness per row. Origin: Artificial Analysis cited by Kimi report.", "candidates": [ { "score": 51.97, "reference_url": "https://huggingface.co/internlm/Intern-S2-Preview-397B", "source_type": "model_card", "reported_setting": { "mode": "source does not state", "effort": "source does not state", "tools": "code execution", "sampling": "source does not state", "judge": "benchmark-specific; exact version not stated", "harness": "AgentCompass / benchmark evaluator", "prompt_style": "source does not state", "temperature": "source does not state", "context": "source does not state", "notes": "Shanghai AI Lab evaluated all model columns using OpenCompass, VLMEvalKit, and AgentCompass. Per-model mode/effort/sampling are not reported; only Intern-S2 maximum evaluation lengths are disclosed." }, "notes": "Intern-S2-Preview-397B official model card: SciCode = 51.97. This is a Shanghai AI Lab cross-model evaluation; the source does not disclose this model's exact mode, effort, or sampling." } ] }, { "model_id": "glm-5.2", "benchmark_id": "kimi_code_bench_v2", "score": 64.2, "reference_url": "https://github.com/MoonshotAI/Kimi-K3/blob/main/k3_tech_report.pdf", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "benchmark-specified", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "agentic coding harness per row", "prompt_style": "default", "temperature": "default", "context": "1048576 (1M)" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Kimi K3 technical report: Kimi Code Bench 2.0 = 64.2. Source setting: agentic coding harness per row. Origin: Moonshot evaluation." }, { "model_id": "glm-5.2", "benchmark_id": "researchrubrics", "score": 71.1, "reference_url": "https://github.com/MoonshotAI/Kimi-K3/blob/main/k3_tech_report.pdf", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "benchmark-specified", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "agentic benchmark harness", "prompt_style": "default", "temperature": "default", "context": "1048576 (1M)" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Kimi K3 technical report: ResearchRubrics = 71.1. Source setting: agentic benchmark harness. Origin: Moonshot evaluation." }, { "model_id": "glm-5.2", "benchmark_id": "gdpval_aa_elo", "score": 1510.0, "reference_url": "https://github.com/MoonshotAI/Kimi-K3/blob/main/k3_tech_report.pdf", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "benchmark-specified", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "agentic benchmark harness", "prompt_style": "default", "temperature": "default", "context": "1048576 (1M)" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Kimi K3 technical report: GDPval-AA v2 (Elo) = 1510.0. Source setting: agentic benchmark harness. Origin: Artificial Analysis cited by Kimi report." }, { "model_id": "glm-5.2", "benchmark_id": "toolathlon_verified", "score": 59.9, "reference_url": "https://github.com/MoonshotAI/Kimi-K3/blob/main/k3_tech_report.pdf", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "benchmark-specified", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "Verified edition; official leaderboard snapshot", "prompt_style": "default", "temperature": "default", "context": "1048576 (1M)" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Kimi K3 technical report: Toolathlon-Verified = 59.9. Source setting: Verified edition; official leaderboard snapshot. Origin: official Toolathlon leaderboard cited by Kimi report." }, { "model_id": "glm-5.2", "benchmark_id": "mcpatlas", "score": 76.8, "reference_url": "https://z.ai/blog/glm-5.2", "reported_setting": { "mode": "thinking", "effort": "max", "sampling": "pass@1", "prompt_style": "default", "tools": "agentic MCP servers", "context": "default", "judge": "Gemini 3.0 Pro", "harness": "public 500-task set; 10-minute timeout per task", "temperature": "default" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "GLM-5.2 official release source: GLM-5.2 / MCP-Atlas (Public Set) = 76.8.", "candidates": [ { "score": 82.6, "reference_url": "https://github.com/MoonshotAI/Kimi-K3/blob/main/k3_tech_report.pdf", "source_type": "tech_report", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "benchmark-specified", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "public 500; maxTurns=100; Gemini 3.1 Pro judge", "prompt_style": "default", "temperature": "default", "context": "1048576 (1M)" }, "notes": "Kimi K3 technical report: MCP-Atlas = 82.6. Source setting: public 500; maxTurns=100; Gemini 3.1 Pro judge. Origin: Moonshot evaluation. [Displaced by GLM-5.2 first-party audit.]" }, { "score": 77.0, "reference_url": "https://z.ai/blog/glm-5.2", "source_type": "official_blog", "reported_setting": { "mode": "thinking", "effort": "max", "sampling": "pass@1", "prompt_style": "default", "tools": "agentic MCP servers", "context": "default", "judge": "Gemini 3.0 Pro", "harness": "public 500-task set; 10-minute timeout per task", "temperature": "default" }, "notes": "Z.ai numeric chart conflicts with the 76.8 full-table value." } ] }, { "model_id": "glm-5.2", "benchmark_id": "automation_bench", "score": 12.9, "reference_url": "https://github.com/MoonshotAI/Kimi-K3/blob/main/k3_tech_report.pdf", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "benchmark-specified", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "agentic benchmark harness", "prompt_style": "default", "temperature": "default", "context": "1048576 (1M)" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Kimi K3 technical report: AutomationBench = 12.9. Source setting: agentic benchmark harness. Origin: Moonshot evaluation." }, { "model_id": "glm-5.2", "benchmark_id": "job_bench", "score": 43.4, "reference_url": "https://github.com/MoonshotAI/Kimi-K3/blob/main/k3_tech_report.pdf", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "benchmark-specified", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "agentic benchmark harness", "prompt_style": "default", "temperature": "default", "context": "1048576 (1M)" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Kimi K3 technical report: JobBench = 43.4. Source setting: agentic benchmark harness. Origin: official JobBench leaderboard cited by Kimi report." }, { "model_id": "glm-5.2", "benchmark_id": "aa_briefcase_elo", "score": 1260.0, "reference_url": "https://github.com/MoonshotAI/Kimi-K3/blob/main/k3_tech_report.pdf", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "benchmark-specified", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "agentic benchmark harness", "prompt_style": "default", "temperature": "default", "context": "1048576 (1M)" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Kimi K3 technical report: AA-Briefcase (Elo) = 1260.0. Source setting: agentic benchmark harness. Origin: Artificial Analysis cited by Kimi report." }, { "model_id": "glm-5.2", "benchmark_id": "agents_last_exam", "score": 20.4, "reference_url": "https://github.com/MoonshotAI/Kimi-K3/blob/main/k3_tech_report.pdf", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "benchmark-specified", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "agentic benchmark harness", "prompt_style": "default", "temperature": "default", "context": "1048576 (1M)" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Kimi K3 technical report: Agents' Last Exam = 20.4. Source setting: agentic benchmark harness. Origin: official leaderboard cited by Kimi report.", "candidates": [ { "score": 23.8, "reference_url": "https://huggingface.co/deepseek-ai/DeepSeek-V4-Flash-0731", "source_type": "model_card", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "benchmark-specified", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "not restated in DeepSeek-V4-Flash-0731 model card", "prompt_style": "default", "temperature": "unknown", "context": "unknown" }, "notes": "DeepSeek-V4-Flash-0731 official model card: GLM-5.2 / Agents' Last Exam = 23.8. Cross-model settings are marked unknown when the source does not restate them." } ] }, { "model_id": "glm-5.2", "benchmark_id": "apex_agents", "score": 35.6, "reference_url": "https://github.com/MoonshotAI/Kimi-K3/blob/main/k3_tech_report.pdf", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "benchmark-specified", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "agentic benchmark harness", "prompt_style": "default", "temperature": "default", "context": "1048576 (1M)" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Kimi K3 technical report: APEX-Agents = 35.6. Source setting: agentic benchmark harness. Origin: Artificial Analysis cited by Kimi report.", "candidates": [ { "score": 29.9, "reference_url": "https://huggingface.co/tencent/Hy3", "source_type": "model_card", "reported_setting": { "mode": "thinking", "effort": "high / highest available", "tools": "multiple application/software tools", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "APEX-Agents long-horizon professional task harness", "prompt_style": "default", "temperature": "0.9 for Hy3 family; otherwise source/provider setting", "context": "256000 for Hy3 family; otherwise source/provider setting", "notes": "Tencent own test" }, "notes": "Hy3 official model-card appendix: Apex-Agent (pass@1) = 29.9*. Tencent own testing." } ] }, { "model_id": "glm-5.2", "benchmark_id": "officeqa_pro", "score": 41.4, "reference_url": "https://github.com/MoonshotAI/Kimi-K3/blob/main/k3_tech_report.pdf", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "benchmark-specified", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "PDF corpus rendered as images; no machine-readable text", "prompt_style": "default", "temperature": "default", "context": "1048576 (1M)" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Kimi K3 technical report: OfficeQA Pro = 41.4. Source setting: PDF corpus rendered as images; no machine-readable text. Origin: Moonshot evaluation." }, { "model_id": "glm-5.2", "benchmark_id": "spreadsheetbench_2", "score": 28.1, "reference_url": "https://github.com/MoonshotAI/Kimi-K3/blob/main/k3_tech_report.pdf", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "benchmark-specified", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "agentic benchmark harness", "prompt_style": "default", "temperature": "default", "context": "1048576 (1M)" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Kimi K3 technical report: SpreadsheetBench 2 = 28.1. Source setting: agentic benchmark harness. Origin: Moonshot evaluation." }, { "model_id": "glm-5.2", "benchmark_id": "tau3_banking", "score": 26.8, "reference_url": "https://github.com/MoonshotAI/Kimi-K3/blob/main/k3_tech_report.pdf", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "benchmark-specified", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "agentic benchmark harness", "prompt_style": "default", "temperature": "default", "context": "1048576 (1M)" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Kimi K3 technical report: τ³-Banking = 26.8. Source setting: agentic benchmark harness. Origin: Artificial Analysis cited by Kimi report.", "candidates": [ { "score": 26.8, "reference_url": "https://media.x.ai/v1/website/4p5-5184fdf9.pdf", "source_type": "model_card", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "banking APIs and simulated user", "sampling": "pass@1", "judge": "final-state task accuracy", "harness": "Artificial Analysis tau3-banking harness", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown" }, "notes": "card45 tau3-banking Accuracy (%); source effort=max; benchmark protocol matches canonical=True. Unknown source fields remain unknown." } ] }, { "model_id": "glm-5.2", "benchmark_id": "harvey_lab_aa", "score": 91.0, "reference_url": "https://github.com/MoonshotAI/Kimi-K3/blob/main/k3_tech_report.pdf", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "benchmark-specified", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "agentic benchmark harness", "prompt_style": "default", "temperature": "default", "context": "1048576 (1M)" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Kimi K3 technical report: Harvey Lab-AA = 91.0. Source setting: agentic benchmark harness. Origin: Artificial Analysis cited by Kimi report." }, { "model_id": "glm-5.2", "benchmark_id": "corpfin_v2", "score": 66.1, "reference_url": "https://github.com/MoonshotAI/Kimi-K3/blob/main/k3_tech_report.pdf", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "benchmark-specified", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "agentic benchmark harness", "prompt_style": "default", "temperature": "default", "context": "1048576 (1M)" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Kimi K3 technical report: CorpFin v2 = 66.1. Source setting: agentic benchmark harness. Origin: Vals AI cited by Kimi report." }, { "model_id": "glm-5.2", "benchmark_id": "finance_agent_v2", "score": 49.7, "reference_url": "https://github.com/MoonshotAI/Kimi-K3/blob/main/k3_tech_report.pdf", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "benchmark-specified", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "agentic benchmark harness", "prompt_style": "default", "temperature": "default", "context": "1048576 (1M)" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Kimi K3 technical report: Finance Agent v2 = 49.7. Source setting: agentic benchmark harness. Origin: Vals AI cited by Kimi report." }, { "model_id": "glm-5.2", "benchmark_id": "legal_research_bench", "score": 31.3, "reference_url": "https://github.com/MoonshotAI/Kimi-K3/blob/main/k3_tech_report.pdf", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "benchmark-specified", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "agentic benchmark harness", "prompt_style": "default", "temperature": "default", "context": "1048576 (1M)" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Kimi K3 technical report: Legal Research Bench = 31.3. Source setting: agentic benchmark harness. Origin: Vals AI cited by Kimi report." }, { "model_id": "glm-5.2", "benchmark_id": "coding_experience", "score": 53.3, "reference_url": "https://github.com/MoonshotAI/Kimi-K3/blob/main/k3_tech_report.pdf", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "benchmark-specified", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "provider-aligned harness: Kimi Code / Claude Code / Codex", "prompt_style": "default", "temperature": "default", "context": "1048576 (1M)" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Kimi K3 technical report: Coding Experience = 53.3. Source setting: provider-aligned harness: Kimi Code / Claude Code / Codex. Origin: Moonshot evaluation." }, { "model_id": "glm-5.2", "benchmark_id": "clawbench_2_0", "score": 43.2, "reference_url": "https://github.com/MoonshotAI/Kimi-K3/blob/main/k3_tech_report.pdf", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "benchmark-specified", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "OpenClaw harness", "prompt_style": "default", "temperature": "default", "context": "1048576 (1M)" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Kimi K3 technical report: 24/7 ClawBench 2.0 = 43.2. Source setting: OpenClaw harness. Origin: Moonshot evaluation." }, { "model_id": "glm-5.2", "benchmark_id": "kaet", "score": 74.7, "reference_url": "https://github.com/MoonshotAI/Kimi-K3/blob/main/k3_tech_report.pdf", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "benchmark-specified", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "Kimi Code harness", "prompt_style": "default", "temperature": "default", "context": "1048576 (1M)" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Kimi K3 technical report: KAET = 74.7. Source setting: Kimi Code harness. Origin: Moonshot evaluation." }, { "model_id": "glm-5.2", "benchmark_id": "clif_bench", "score": 39.2, "reference_url": "https://github.com/MoonshotAI/Kimi-K3/blob/main/k3_tech_report.pdf", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "benchmark-specified", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "Kimi Code harness", "prompt_style": "default", "temperature": "default", "context": "1048576 (1M)" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Kimi K3 technical report: CLIF Bench = 39.2. Source setting: Kimi Code harness. Origin: Moonshot evaluation." }, { "model_id": "glm-5.2", "benchmark_id": "swarm_bench", "score": 58.5, "reference_url": "https://github.com/MoonshotAI/Kimi-K3/blob/main/k3_tech_report.pdf", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "benchmark-specified", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "Kimi Agent harness", "prompt_style": "default", "temperature": "default", "context": "1048576 (1M)" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Kimi K3 technical report: SwarmBench = 58.5. Source setting: Kimi Agent harness. Origin: Moonshot evaluation." }, { "model_id": "glm-5.2", "benchmark_id": "online_experience", "score": 64.0, "reference_url": "https://github.com/MoonshotAI/Kimi-K3/blob/main/k3_tech_report.pdf", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "benchmark-specified", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "Kimi Agent harness", "prompt_style": "default", "temperature": "default", "context": "1048576 (1M)" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Kimi K3 technical report: Online Experience = 64.0. Source setting: Kimi Agent harness. Origin: Moonshot evaluation." }, { "model_id": "glm-5.2", "benchmark_id": "deepresearchbench", "score": 84.0, "reference_url": "https://github.com/MoonshotAI/Kimi-K3/blob/main/k3_tech_report.pdf", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "benchmark-specified", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "Kimi Agent harness", "prompt_style": "default", "temperature": "default", "context": "1048576 (1M)" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Kimi K3 technical report: DeepResearchBench = 84.0. Source setting: Kimi Agent harness. Origin: Moonshot evaluation." }, { "model_id": "glm-5.2", "benchmark_id": "finance_bench", "score": 55.4, "reference_url": "https://github.com/MoonshotAI/Kimi-K3/blob/main/k3_tech_report.pdf", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "benchmark-specified", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "N/A harness", "prompt_style": "default", "temperature": "default", "context": "1048576 (1M)" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Kimi K3 technical report: Finance Bench = 55.4. Source setting: N/A harness. Origin: Moonshot evaluation." }, { "model_id": "glm-5.2", "benchmark_id": "deck_bench", "score": 68.6, "reference_url": "https://github.com/MoonshotAI/Kimi-K3/blob/main/k3_tech_report.pdf", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "benchmark-specified", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "N/A harness", "prompt_style": "default", "temperature": "default", "context": "1048576 (1M)" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Kimi K3 technical report: DECKBench = 68.6. Source setting: N/A harness. Origin: Moonshot evaluation." }, { "model_id": "glm-5.2", "benchmark_id": "faithfulness", "score": 74.8, "reference_url": "https://github.com/MoonshotAI/Kimi-K3/blob/main/k3_tech_report.pdf", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "benchmark-specified", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "N/A harness", "prompt_style": "default", "temperature": "default", "context": "1048576 (1M)" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Kimi K3 technical report: Faithfulness (1 - hallucination rate) = 74.8. Source setting: N/A harness. Origin: Moonshot evaluation." }, { "model_id": "glm-5.2", "benchmark_id": "aa_intelligence_index_v4_1", "score": 51.1, "reference_url": "https://github.com/MoonshotAI/Kimi-K3/blob/main/k3_tech_report.pdf", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "benchmark-specified", "sampling": "leaderboard snapshot 2026-07-23", "judge": "benchmark-specified", "harness": "live leaderboard snapshot as of 2026-07-23", "prompt_style": "default", "temperature": "default", "context": "1048576 (1M)" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Kimi K3 technical report: AA Intelligence Index v4.1 = 51.1. Source setting: live leaderboard snapshot as of 2026-07-23. Origin: third-party leaderboard cited by Kimi report." }, { "model_id": "glm-5.2", "benchmark_id": "vals_index", "score": 65.0, "reference_url": "https://github.com/MoonshotAI/Kimi-K3/blob/main/k3_tech_report.pdf", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "benchmark-specified", "sampling": "leaderboard snapshot 2026-07-23", "judge": "benchmark-specified", "harness": "live leaderboard snapshot as of 2026-07-23", "prompt_style": "default", "temperature": "default", "context": "1048576 (1M)" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Kimi K3 technical report: Vals Index = 65.0. Source setting: live leaderboard snapshot as of 2026-07-23. Origin: third-party leaderboard cited by Kimi report." }, { "model_id": "glm-5.2", "benchmark_id": "webdev_arena_elo", "score": 1592.0, "reference_url": "https://github.com/MoonshotAI/Kimi-K3/blob/main/k3_tech_report.pdf", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "benchmark-specified", "sampling": "leaderboard snapshot 2026-07-23", "judge": "benchmark-specified", "harness": "live leaderboard snapshot as of 2026-07-23", "prompt_style": "default", "temperature": "default", "context": "1048576 (1M)" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Kimi K3 technical report: WebDevArena Elo = 1592. Source setting: live leaderboard snapshot as of 2026-07-23. Origin: third-party leaderboard cited by Kimi report." }, { "model_id": "glm-5.2", "benchmark_id": "text_arena_elo", "score": 1469.0, "reference_url": "https://github.com/MoonshotAI/Kimi-K3/blob/main/k3_tech_report.pdf", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "benchmark-specified", "sampling": "leaderboard snapshot 2026-07-23", "judge": "benchmark-specified", "harness": "live leaderboard snapshot as of 2026-07-23", "prompt_style": "default", "temperature": "default", "context": "1048576 (1M)" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Kimi K3 technical report: TextArena Elo = 1469. Source setting: live leaderboard snapshot as of 2026-07-23. Origin: third-party leaderboard cited by Kimi report.", "candidates": [ { "score": 1471, "reference_url": "https://arxiv.org/abs/2607.02770", "source_type": "tech_report", "reported_setting": { "mode": "unknown", "effort": "unknown", "tools": "none", "sampling": "live human pairwise votes", "judge": "blind human raters", "harness": "Arena Text", "prompt_style": "live user prompts", "temperature": "unknown", "context": "default", "snapshot_date": "2026-06-19" }, "notes": "Gemma 4 Technical Report Table 4 dated Arena Text Elo snapshot." } ] }, { "model_id": "glm-5.2", "benchmark_id": "agent_arena", "score": 6.5, "reference_url": "https://github.com/MoonshotAI/Kimi-K3/blob/main/k3_tech_report.pdf", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "benchmark-specified", "sampling": "leaderboard snapshot 2026-07-23", "judge": "benchmark-specified", "harness": "live leaderboard snapshot as of 2026-07-23", "prompt_style": "default", "temperature": "default", "context": "1048576 (1M)" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Kimi K3 technical report: AgentArena = 6.5. Source setting: live leaderboard snapshot as of 2026-07-23. Origin: third-party leaderboard cited by Kimi report." }, { "model_id": "glm-5.2", "benchmark_id": "moonshot_exploit_development_suite", "score": 22.2, "reference_url": "https://github.com/MoonshotAI/Kimi-K3/blob/main/k3_tech_report.pdf", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "benchmark-specified", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "36-task in-house exploit suite", "prompt_style": "default", "temperature": "default", "context": "1048576 (1M)" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Kimi K3 technical report: Moonshot exploit-development suite = 22.2. Source setting: 36-task in-house exploit suite. Origin: Moonshot evaluation." }, { "model_id": "glm-5.2", "benchmark_id": "exploitbench", "score": 24.0, "reference_url": "https://github.com/MoonshotAI/Kimi-K3/blob/main/k3_tech_report.pdf", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "benchmark-specified", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "independent joint assessment", "prompt_style": "default", "temperature": "default", "context": "1048576 (1M)" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Kimi K3 technical report: ExploitBench (UK AISI / NIST CAISI) = 24.0. Source setting: independent joint assessment. Origin: UK AISI / NIST CAISI cited by Kimi report." }, { "model_id": "deepseek-v4-flash-0731", "benchmark_id": "terminal_bench_2_1", "score": 82.7, "reference_url": "https://huggingface.co/deepseek-ai/DeepSeek-V4-Flash-0731", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "benchmark-specified", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "DeepSeek Harness minimal mode for Code Agent tasks", "prompt_style": "default", "temperature": "1.0; top_p=0.95", "context": "1048576 (1M)" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "DeepSeek-V4-Flash-0731 official model card: DeepSeek-V4-Flash-0731 / Terminal Bench 2.1 = 82.7. Cross-model settings are marked unknown when the source does not restate them." }, { "model_id": "deepseek-v4-flash", "benchmark_id": "terminal_bench_2_1", "score": 61.8, "reference_url": "https://huggingface.co/deepseek-ai/DeepSeek-V4-Flash-0731", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "benchmark-specified", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "not restated in DeepSeek-V4-Flash-0731 model card", "prompt_style": "default", "temperature": "unknown", "context": "unknown" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "DeepSeek-V4-Flash-0731 official model card: DeepSeek-V4-Flash (Preview) / Terminal Bench 2.1 = 61.8. Cross-model settings are marked unknown when the source does not restate them.", "candidates": [ { "score": 52.58, "reference_url": "https://huggingface.co/tencent/Hy3", "source_type": "model_card", "reported_setting": { "mode": "thinking", "effort": "high / highest available", "tools": "terminal/code execution", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "Terminus-2; parser=xml; timeout=4h; CPU=16 cores; memory=32GB; max episodes=500", "prompt_style": "default", "temperature": "0.9 for Hy3 family; otherwise source/provider setting", "context": "256000 for Hy3 family; otherwise source/provider setting", "notes": "Tencent own test" }, "notes": "Hy3 official model-card appendix: Terminal-Bench 2.1 = 52.58*. Tencent own testing." }, { "score": 54.2, "reference_url": "https://research.nvidia.com/labs/nemotron/files/NVIDIA-Nemotron-3-Ultra-Technical-Report.pdf", "source_type": "tech_report", "reported_setting": { "version_metric": "v2.1 / % resolved", "tools": "terminal and code execution", "harness": "Harbor / Terminus-2", "sampling": "8-run average" }, "notes": "NVIDIA Nemotron 3 Ultra technical report, Final Table 10: DeepSeek-V4-Flash Preview: terminal_bench_2_1=54.2." }, { "score": 62.0, "reference_url": "https://static.stepfun.com/blog/step-3.7-flash/", "source_type": "official_blog", "reported_setting": { "quantization_config": "common release evaluation; artifact precision not established", "configuration": "base Step evaluation configuration", "mode": "source does not state", "effort": "source does not state", "tools_search": "terminal tools", "shots": "source does not state", "trials_samples": "source does not state", "aggregation_pass_k": "% tasks resolved", "context": "source does not state", "output_cap": "source does not state", "temperature": "source does not state", "top_p": "source does not state", "judge": "benchmark-specified; exact StepFun judge not stated", "harness_agent": "source does not state terminal scaffold", "dataset_version_split": "Terminal-Bench 2.1", "multimodal_input": false }, "notes": "Official StepFun launch comparison table." } ] }, { "model_id": "deepseek-v4-pro", "benchmark_id": "terminal_bench_2_1", "score": 72.1, "reference_url": "https://huggingface.co/deepseek-ai/DeepSeek-V4-Flash-0731", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "benchmark-specified", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "not restated in DeepSeek-V4-Flash-0731 model card", "prompt_style": "default", "temperature": "unknown", "context": "unknown" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "DeepSeek-V4-Flash-0731 official model card: DeepSeek-V4-Pro (Preview) / Terminal Bench 2.1 = 72.1. Cross-model settings are marked unknown when the source does not restate them.", "candidates": [ { "score": 64.0, "reference_url": "https://z.ai/blog/glm-5.2", "source_type": "official_blog", "reported_setting": { "mode": "thinking", "effort": "Max", "sampling": "pass@1", "prompt_style": "default", "tools": "terminal agent", "temperature": "1.0", "top_p": "1.0", "context": "256000", "max_output_tokens": "48000", "judge": "official task verifier", "harness": "Terminus-2; parser=json; timeout=4h; max_episodes=500; 4 CPU/8GB" }, "notes": "GLM-5.2 Terminus-2 alternative; newer DeepSeek-provider 72.1 remains primary." }, { "score": 60.5, "reference_url": "https://huggingface.co/tencent/Hy3", "source_type": "model_card", "reported_setting": { "mode": "thinking", "effort": "high / highest available", "tools": "terminal/code execution", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "Terminus-2; parser=xml; timeout=4h; CPU=16 cores; memory=32GB; max episodes=500", "prompt_style": "default", "temperature": "0.9 for Hy3 family; otherwise source/provider setting", "context": "256000 for Hy3 family; otherwise source/provider setting", "notes": "Tencent own test" }, "notes": "Hy3 official model-card appendix: Terminal-Bench 2.1 = 60.5*. Tencent own testing." }, { "score": 49.2, "reference_url": "https://research.nvidia.com/labs/nemotron/files/NVIDIA-Nemotron-3-Ultra-Technical-Report.pdf", "source_type": "tech_report", "reported_setting": { "version_metric": "v2.1 / % resolved", "tools": "terminal and code execution", "harness": "Harbor / Terminus-2", "sampling": "8-run average" }, "notes": "NVIDIA Nemotron 3 Ultra technical report, Final Table 10: DeepSeek-V4-Pro: terminal_bench_2_1=49.2." } ] }, { "model_id": "deepseek-v4-flash-0731", "benchmark_id": "nl2repo_bench", "score": 54.2, "reference_url": "https://huggingface.co/deepseek-ai/DeepSeek-V4-Flash-0731", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "benchmark-specified", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "DeepSeek Harness minimal mode for Code Agent tasks", "prompt_style": "default", "temperature": "1.0; top_p=0.95", "context": "1048576 (1M)" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "DeepSeek-V4-Flash-0731 official model card: DeepSeek-V4-Flash-0731 / NL2Repo = 54.2. Cross-model settings are marked unknown when the source does not restate them." }, { "model_id": "deepseek-v4-flash", "benchmark_id": "nl2repo_bench", "score": 39.4, "reference_url": "https://huggingface.co/deepseek-ai/DeepSeek-V4-Flash-0731", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "benchmark-specified", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "not restated in DeepSeek-V4-Flash-0731 model card", "prompt_style": "default", "temperature": "unknown", "context": "unknown" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "DeepSeek-V4-Flash-0731 official model card: DeepSeek-V4-Flash (Preview) / NL2Repo = 39.4. Cross-model settings are marked unknown when the source does not restate them." }, { "model_id": "deepseek-v4-pro", "benchmark_id": "nl2repo_bench", "score": 38.5, "reference_url": "https://huggingface.co/deepseek-ai/DeepSeek-V4-Flash-0731", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "benchmark-specified", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "not restated in DeepSeek-V4-Flash-0731 model card", "prompt_style": "default", "temperature": "unknown", "context": "unknown" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "DeepSeek-V4-Flash-0731 official model card: DeepSeek-V4-Pro (Preview) / NL2Repo = 38.5. Cross-model settings are marked unknown when the source does not restate them.", "candidates": [ { "score": 35.5, "reference_url": "https://z.ai/blog/glm-5.2", "source_type": "official_blog", "reported_setting": { "mode": "thinking", "effort": "Max", "sampling": "pass@1", "prompt_style": "default", "tools": "agentic code execution", "temperature": "1.0", "top_p": "1.0", "context": "400000", "max_output_tokens": "48000", "judge": "rule-based + LLM anti-hack judgement", "harness": "official NL2Repo evaluation" }, "notes": "GLM-5.2 cross-table alternative; newer DeepSeek-provider 38.5 remains primary." }, { "score": 41.5, "reference_url": "https://huggingface.co/tencent/Hy3", "source_type": "model_card", "reported_setting": { "mode": "thinking", "effort": "high / highest available", "tools": "Claude Code agentic code execution", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "Claude Code; 250-turn budget; timeout=12000s; 4 CPU/32GB; anti-hacking/tool-call monitoring", "prompt_style": "default", "temperature": "0.9 for Hy3 family; otherwise source/provider setting", "context": "256000 for Hy3 family; otherwise source/provider setting", "notes": "Tencent own test" }, "notes": "Hy3 official model-card appendix: NL2Repo = 41.5*. Tencent own testing." } ] }, { "model_id": "glm-5.2", "benchmark_id": "nl2repo_bench", "score": 48.9, "reference_url": "https://z.ai/blog/glm-5.2", "reported_setting": { "mode": "thinking", "effort": "max", "sampling": "pass@1", "prompt_style": "default", "tools": "agentic code execution", "temperature": "1.0", "top_p": "1.0", "context": "400000", "max_output_tokens": "48000", "judge": "rule-based + LLM anti-hack judgement", "harness": "official NL2Repo evaluation" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "GLM-5.2 official release source: GLM-5.2 / NL2Repo = 48.9.", "candidates": [] }, { "model_id": "claude-opus-4.8", "benchmark_id": "nl2repo_bench", "score": 69.7, "reference_url": "https://huggingface.co/deepseek-ai/DeepSeek-V4-Flash-0731", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "benchmark-specified", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "not restated in DeepSeek-V4-Flash-0731 model card", "prompt_style": "default", "temperature": "unknown", "context": "unknown" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "DeepSeek-V4-Flash-0731 official model card: Opus-4.8 / NL2Repo = 69.7. Cross-model settings are marked unknown when the source does not restate them." }, { "model_id": "deepseek-v4-flash-0731", "benchmark_id": "cybergym", "score": 76.7, "reference_url": "https://huggingface.co/deepseek-ai/DeepSeek-V4-Flash-0731", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "benchmark-specified", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "DeepSeek Harness minimal mode for Code Agent tasks", "prompt_style": "default", "temperature": "1.0; top_p=0.95", "context": "1048576 (1M)" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "DeepSeek-V4-Flash-0731 official model card: DeepSeek-V4-Flash-0731 / Cybergym = 76.7. Cross-model settings are marked unknown when the source does not restate them." }, { "model_id": "deepseek-v4-flash", "benchmark_id": "cybergym", "score": 38.7, "reference_url": "https://huggingface.co/deepseek-ai/DeepSeek-V4-Flash-0731", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "benchmark-specified", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "not restated in DeepSeek-V4-Flash-0731 model card", "prompt_style": "default", "temperature": "unknown", "context": "unknown" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "DeepSeek-V4-Flash-0731 official model card: DeepSeek-V4-Flash (Preview) / Cybergym = 38.7. Cross-model settings are marked unknown when the source does not restate them." }, { "model_id": "deepseek-v4-pro", "benchmark_id": "cybergym", "score": 52.7, "reference_url": "https://huggingface.co/deepseek-ai/DeepSeek-V4-Flash-0731", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "benchmark-specified", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "not restated in DeepSeek-V4-Flash-0731 model card", "prompt_style": "default", "temperature": "unknown", "context": "unknown" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "DeepSeek-V4-Flash-0731 official model card: DeepSeek-V4-Pro (Preview) / Cybergym = 52.7. Cross-model settings are marked unknown when the source does not restate them." }, { "model_id": "claude-opus-4.8", "benchmark_id": "cybergym", "score": 83.1, "reference_url": "https://huggingface.co/deepseek-ai/DeepSeek-V4-Flash-0731", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "benchmark-specified", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "not restated in DeepSeek-V4-Flash-0731 model card", "prompt_style": "default", "temperature": "unknown", "context": "unknown" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "DeepSeek-V4-Flash-0731 official model card: Opus-4.8 / Cybergym = 83.1. Cross-model settings are marked unknown when the source does not restate them.", "candidates": [ { "score": 78.1, "reference_url": "https://www.anthropic.com/claude-sonnet-5-system-card", "source_type": "model_card", "reported_setting": { "effort": "source does not state", "tools": "cyber execution environment", "sampling": "pass@1", "harness": "cybergym" }, "notes": "Anthropic Claude Sonnet 5 System Card Figure 3.2.3.A, page 33: Claude Opus 4.8; CyberGym vulnerability discovery [targeted]=78.1%. Source setting: effort=source does not state; tools=cyber execution environment; sampling=pass@1; harness=cybergym." }, { "score": 78.8, "reference_url": "https://ai.meta.com/static-resource/muse-spark-1.1-evaluation-report", "source_type": "tech_report", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "benchmark-specified", "sampling": "pass@1", "harness": "Meta safety/preparedness evaluation", "temperature": "1.0 for Muse; provider/source setting for comparisons", "context": "source-specific", "configuration": "default", "notes": "Capability scorecard; Muse evaluated without system prompt; high reasoning." }, "notes": "Table 1 · p6; metric=pass@1_percent. Claude cyber-benchmark cells omitted where refusal compromised capability measurement." }, { "score": 78.1, "reference_url": "https://media.x.ai/v1/website/card-4p6-4cd2dc57.pdf", "source_type": "model_card", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "unrestricted agentic code execution", "sampling": "unknown", "judge": "PoC reproduced on vulnerable but not fixed version", "harness": "CyberGym unrestricted capability harness", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown" }, "notes": "card46 CyberGym Mean Reproduced (%); source effort=max; benchmark protocol matches canonical=True. Unknown source fields remain unknown." } ] }, { "model_id": "deepseek-v4-flash-0731", "benchmark_id": "deep_swe_v1_1", "score": 54.4, "reference_url": "https://huggingface.co/deepseek-ai/DeepSeek-V4-Flash-0731", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "benchmark-specified", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "DeepSeek Harness minimal mode for Code Agent tasks", "prompt_style": "default", "temperature": "1.0; top_p=0.95", "context": "1048576 (1M)" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "DeepSeek-V4-Flash-0731 official model card: DeepSeek-V4-Flash-0731 / DeepSWE = 54.4. Cross-model settings are marked unknown when the source does not restate them." }, { "model_id": "deepseek-v4-flash", "benchmark_id": "deep_swe_v1_1", "score": 7.3, "reference_url": "https://huggingface.co/deepseek-ai/DeepSeek-V4-Flash-0731", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "benchmark-specified", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "not restated in DeepSeek-V4-Flash-0731 model card", "prompt_style": "default", "temperature": "unknown", "context": "unknown" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "DeepSeek-V4-Flash-0731 official model card: DeepSeek-V4-Flash (Preview) / DeepSWE = 7.3. Cross-model settings are marked unknown when the source does not restate them.", "candidates": [ { "score": 7.1, "reference_url": "https://huggingface.co/tencent/Hy3", "source_type": "model_card", "reported_setting": { "mode": "thinking", "effort": "high / highest available", "tools": "agentic code execution", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "mini-swe-agent; 2h/task; 2 CPU/8GB; no internet", "prompt_style": "default", "temperature": "0.9 for Hy3 family; otherwise source/provider setting", "context": "256000 for Hy3 family; otherwise source/provider setting", "notes": "Tencent own test" }, "notes": "Hy3 official model-card appendix: DeepSWE = 7.1*. Tencent own testing." } ] }, { "model_id": "deepseek-v4-pro", "benchmark_id": "deep_swe_v1_1", "score": 12.8, "reference_url": "https://huggingface.co/deepseek-ai/DeepSeek-V4-Flash-0731", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "benchmark-specified", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "not restated in DeepSeek-V4-Flash-0731 model card", "prompt_style": "default", "temperature": "unknown", "context": "unknown" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "DeepSeek-V4-Flash-0731 official model card: DeepSeek-V4-Pro (Preview) / DeepSWE = 12.8. Cross-model settings are marked unknown when the source does not restate them.", "candidates": [ { "score": 8.0, "reference_url": "https://z.ai/blog/glm-5.2", "source_type": "official_blog", "reported_setting": { "mode": "thinking", "effort": "Max", "sampling": "pass@1", "prompt_style": "default", "tools": "agentic code execution", "temperature": "1.0", "top_p": "1.0", "context": "400000", "judge": "isolated-container verification", "harness": "official pier + mini-swe-agent; timeout=2h; 2 CPU/8GB; no internet" }, "notes": "GLM-5.2 cross-table alternative; newer DeepSeek-provider 12.8 remains primary." }, { "score": 9.7, "reference_url": "https://huggingface.co/tencent/Hy3", "source_type": "model_card", "reported_setting": { "mode": "thinking", "effort": "high / highest available", "tools": "agentic code execution", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "mini-swe-agent; 2h/task; 2 CPU/8GB; no internet", "prompt_style": "default", "temperature": "0.9 for Hy3 family; otherwise source/provider setting", "context": "256000 for Hy3 family; otherwise source/provider setting", "notes": "Tencent own test" }, "notes": "Hy3 official model-card appendix: DeepSWE = 9.7*. Tencent own testing." } ] }, { "model_id": "deepseek-v4-flash-0731", "benchmark_id": "toolathlon_verified", "score": 70.3, "reference_url": "https://huggingface.co/deepseek-ai/DeepSeek-V4-Flash-0731", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "benchmark-specified", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "not restated in DeepSeek-V4-Flash-0731 model card", "prompt_style": "default", "temperature": "unknown", "context": "unknown" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "DeepSeek-V4-Flash-0731 official model card: DeepSeek-V4-Flash-0731 / Toolathlon-Verified = 70.3. Cross-model settings are marked unknown when the source does not restate them." }, { "model_id": "deepseek-v4-flash", "benchmark_id": "toolathlon_verified", "score": 49.7, "reference_url": "https://huggingface.co/deepseek-ai/DeepSeek-V4-Flash-0731", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "benchmark-specified", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "not restated in DeepSeek-V4-Flash-0731 model card", "prompt_style": "default", "temperature": "unknown", "context": "unknown" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "DeepSeek-V4-Flash-0731 official model card: DeepSeek-V4-Flash (Preview) / Toolathlon-Verified = 49.7. Cross-model settings are marked unknown when the source does not restate them." }, { "model_id": "deepseek-v4-pro", "benchmark_id": "toolathlon_verified", "score": 55.9, "reference_url": "https://huggingface.co/deepseek-ai/DeepSeek-V4-Flash-0731", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "benchmark-specified", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "not restated in DeepSeek-V4-Flash-0731 model card", "prompt_style": "default", "temperature": "unknown", "context": "unknown" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "DeepSeek-V4-Flash-0731 official model card: DeepSeek-V4-Pro (Preview) / Toolathlon-Verified = 55.9. Cross-model settings are marked unknown when the source does not restate them." }, { "model_id": "deepseek-v4-flash-0731", "benchmark_id": "agents_last_exam", "score": 25.2, "reference_url": "https://huggingface.co/deepseek-ai/DeepSeek-V4-Flash-0731", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "benchmark-specified", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "not restated in DeepSeek-V4-Flash-0731 model card", "prompt_style": "default", "temperature": "unknown", "context": "unknown" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "DeepSeek-V4-Flash-0731 official model card: DeepSeek-V4-Flash-0731 / Agents' Last Exam = 25.2. Cross-model settings are marked unknown when the source does not restate them." }, { "model_id": "deepseek-v4-flash", "benchmark_id": "agents_last_exam", "score": 15.8, "reference_url": "https://huggingface.co/deepseek-ai/DeepSeek-V4-Flash-0731", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "benchmark-specified", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "not restated in DeepSeek-V4-Flash-0731 model card", "prompt_style": "default", "temperature": "unknown", "context": "unknown" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "DeepSeek-V4-Flash-0731 official model card: DeepSeek-V4-Flash (Preview) / Agents' Last Exam = 15.8. Cross-model settings are marked unknown when the source does not restate them." }, { "model_id": "deepseek-v4-pro", "benchmark_id": "agents_last_exam", "score": 16.5, "reference_url": "https://huggingface.co/deepseek-ai/DeepSeek-V4-Flash-0731", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "benchmark-specified", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "not restated in DeepSeek-V4-Flash-0731 model card", "prompt_style": "default", "temperature": "unknown", "context": "unknown" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "DeepSeek-V4-Flash-0731 official model card: DeepSeek-V4-Pro (Preview) / Agents' Last Exam = 16.5. Cross-model settings are marked unknown when the source does not restate them." }, { "model_id": "deepseek-v4-flash-0731", "benchmark_id": "automation_bench", "score": 25.1, "reference_url": "https://huggingface.co/deepseek-ai/DeepSeek-V4-Flash-0731", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "benchmark-specified", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "not restated in DeepSeek-V4-Flash-0731 model card", "prompt_style": "default", "temperature": "unknown", "context": "unknown" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "DeepSeek-V4-Flash-0731 official model card: DeepSeek-V4-Flash-0731 / AutomationBench Public = 25.1. Cross-model settings are marked unknown when the source does not restate them." }, { "model_id": "deepseek-v4-flash", "benchmark_id": "automation_bench", "score": 10.8, "reference_url": "https://huggingface.co/deepseek-ai/DeepSeek-V4-Flash-0731", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "benchmark-specified", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "not restated in DeepSeek-V4-Flash-0731 model card", "prompt_style": "default", "temperature": "unknown", "context": "unknown" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "DeepSeek-V4-Flash-0731 official model card: DeepSeek-V4-Flash (Preview) / AutomationBench Public = 10.8. Cross-model settings are marked unknown when the source does not restate them." }, { "model_id": "deepseek-v4-pro", "benchmark_id": "automation_bench", "score": 12.8, "reference_url": "https://huggingface.co/deepseek-ai/DeepSeek-V4-Flash-0731", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "benchmark-specified", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "not restated in DeepSeek-V4-Flash-0731 model card", "prompt_style": "default", "temperature": "unknown", "context": "unknown" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "DeepSeek-V4-Flash-0731 official model card: DeepSeek-V4-Pro (Preview) / AutomationBench Public = 12.8. Cross-model settings are marked unknown when the source does not restate them." }, { "model_id": "deepseek-v4-flash-0731", "benchmark_id": "dsbench_fullstack", "score": 68.7, "reference_url": "https://huggingface.co/deepseek-ai/DeepSeek-V4-Flash-0731", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "benchmark-specified", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "not restated in DeepSeek-V4-Flash-0731 model card", "prompt_style": "default", "temperature": "unknown", "context": "unknown" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "DeepSeek-V4-Flash-0731 official model card: DeepSeek-V4-Flash-0731 / DSBench-FullStack † = 68.7. Cross-model settings are marked unknown when the source does not restate them." }, { "model_id": "deepseek-v4-flash", "benchmark_id": "dsbench_fullstack", "score": 37.0, "reference_url": "https://huggingface.co/deepseek-ai/DeepSeek-V4-Flash-0731", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "benchmark-specified", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "not restated in DeepSeek-V4-Flash-0731 model card", "prompt_style": "default", "temperature": "unknown", "context": "unknown" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "DeepSeek-V4-Flash-0731 official model card: DeepSeek-V4-Flash (Preview) / DSBench-FullStack † = 37.0. Cross-model settings are marked unknown when the source does not restate them." }, { "model_id": "deepseek-v4-pro", "benchmark_id": "dsbench_fullstack", "score": 41.8, "reference_url": "https://huggingface.co/deepseek-ai/DeepSeek-V4-Flash-0731", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "benchmark-specified", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "not restated in DeepSeek-V4-Flash-0731 model card", "prompt_style": "default", "temperature": "unknown", "context": "unknown" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "DeepSeek-V4-Flash-0731 official model card: DeepSeek-V4-Pro (Preview) / DSBench-FullStack † = 41.8. Cross-model settings are marked unknown when the source does not restate them." }, { "model_id": "glm-5.2", "benchmark_id": "dsbench_fullstack", "score": 61.8, "reference_url": "https://huggingface.co/deepseek-ai/DeepSeek-V4-Flash-0731", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "benchmark-specified", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "not restated in DeepSeek-V4-Flash-0731 model card", "prompt_style": "default", "temperature": "unknown", "context": "unknown" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "DeepSeek-V4-Flash-0731 official model card: GLM-5.2 / DSBench-FullStack † = 61.8. Cross-model settings are marked unknown when the source does not restate them." }, { "model_id": "claude-opus-4.8", "benchmark_id": "dsbench_fullstack", "score": 71.6, "reference_url": "https://huggingface.co/deepseek-ai/DeepSeek-V4-Flash-0731", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "benchmark-specified", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "not restated in DeepSeek-V4-Flash-0731 model card", "prompt_style": "default", "temperature": "unknown", "context": "unknown" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "DeepSeek-V4-Flash-0731 official model card: Opus-4.8 / DSBench-FullStack † = 71.6. Cross-model settings are marked unknown when the source does not restate them." }, { "model_id": "deepseek-v4-flash-0731", "benchmark_id": "dsbench_hard", "score": 59.6, "reference_url": "https://huggingface.co/deepseek-ai/DeepSeek-V4-Flash-0731", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "benchmark-specified", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "not restated in DeepSeek-V4-Flash-0731 model card", "prompt_style": "default", "temperature": "unknown", "context": "unknown" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "DeepSeek-V4-Flash-0731 official model card: DeepSeek-V4-Flash-0731 / DSBench-Hard † = 59.6. Cross-model settings are marked unknown when the source does not restate them." }, { "model_id": "deepseek-v4-flash", "benchmark_id": "dsbench_hard", "score": 25.8, "reference_url": "https://huggingface.co/deepseek-ai/DeepSeek-V4-Flash-0731", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "benchmark-specified", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "not restated in DeepSeek-V4-Flash-0731 model card", "prompt_style": "default", "temperature": "unknown", "context": "unknown" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "DeepSeek-V4-Flash-0731 official model card: DeepSeek-V4-Flash (Preview) / DSBench-Hard † = 25.8. Cross-model settings are marked unknown when the source does not restate them." }, { "model_id": "deepseek-v4-pro", "benchmark_id": "dsbench_hard", "score": 31.1, "reference_url": "https://huggingface.co/deepseek-ai/DeepSeek-V4-Flash-0731", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "benchmark-specified", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "not restated in DeepSeek-V4-Flash-0731 model card", "prompt_style": "default", "temperature": "unknown", "context": "unknown" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "DeepSeek-V4-Flash-0731 official model card: DeepSeek-V4-Pro (Preview) / DSBench-Hard † = 31.1. Cross-model settings are marked unknown when the source does not restate them." }, { "model_id": "glm-5.2", "benchmark_id": "dsbench_hard", "score": 54.5, "reference_url": "https://huggingface.co/deepseek-ai/DeepSeek-V4-Flash-0731", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "benchmark-specified", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "not restated in DeepSeek-V4-Flash-0731 model card", "prompt_style": "default", "temperature": "unknown", "context": "unknown" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "DeepSeek-V4-Flash-0731 official model card: GLM-5.2 / DSBench-Hard † = 54.5. Cross-model settings are marked unknown when the source does not restate them." }, { "model_id": "claude-opus-4.8", "benchmark_id": "dsbench_hard", "score": 71.7, "reference_url": "https://huggingface.co/deepseek-ai/DeepSeek-V4-Flash-0731", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "benchmark-specified", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "not restated in DeepSeek-V4-Flash-0731 model card", "prompt_style": "default", "temperature": "unknown", "context": "unknown" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "DeepSeek-V4-Flash-0731 official model card: Opus-4.8 / DSBench-Hard † = 71.7. Cross-model settings are marked unknown when the source does not restate them." }, { "model_id": "glm-5.2", "benchmark_id": "hle_text", "score": 40.5, "reference_url": "https://z.ai/blog/glm-5.2", "reported_setting": { "mode": "thinking", "effort": "max", "sampling": "pass@1", "prompt_style": "default", "tools": "none", "temperature": "1.0", "top_p": "0.95", "context": "default", "max_output_tokens": "163840", "judge": "GPT-5.5 medium for AIME/HMMT/IMO; benchmark-specified otherwise", "harness": "official reasoning evaluation" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "GLM-5.2 official release source: GLM-5.2 / HLE = 40.5.", "candidates": [] }, { "model_id": "glm-5.2", "benchmark_id": "hle_tools_text", "score": 54.7, "reference_url": "https://z.ai/blog/glm-5.2", "reported_setting": { "mode": "thinking", "effort": "max", "sampling": "pass@1", "prompt_style": "default", "tools": "search+code+web", "temperature": "1.0", "top_p": "0.95", "context": "300000", "max_output_tokens": "163840", "judge": "answer-key scoring", "harness": "official; no context-management strategy" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "GLM-5.2 official release source: GLM-5.2 / HLE (w/ Tools) = 54.7.", "candidates": [] }, { "model_id": "glm-5.2", "benchmark_id": "aime_2026", "score": 99.2, "reference_url": "https://z.ai/blog/glm-5.2", "reported_setting": { "mode": "thinking", "effort": "max", "sampling": "pass@1", "prompt_style": "default", "tools": "none", "temperature": "1.0", "top_p": "0.95", "context": "default", "max_output_tokens": "163840", "judge": "GPT-5.5 medium for AIME/HMMT/IMO; benchmark-specified otherwise", "harness": "official reasoning evaluation" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "GLM-5.2 official release source: GLM-5.2 / AIME 2026 = 99.2.", "candidates": [] }, { "model_id": "glm-5.2", "benchmark_id": "hmmt_nov_2025", "score": 94.4, "reference_url": "https://z.ai/blog/glm-5.2", "reported_setting": { "mode": "thinking", "effort": "max", "sampling": "pass@1", "prompt_style": "default", "tools": "none", "temperature": "1.0", "top_p": "0.95", "context": "default", "max_output_tokens": "163840", "judge": "GPT-5.5 medium for AIME/HMMT/IMO; benchmark-specified otherwise", "harness": "official reasoning evaluation" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "GLM-5.2 official release source: GLM-5.2 / HMMT Nov. 2025 = 94.4.", "candidates": [] }, { "model_id": "glm-5.2", "benchmark_id": "hmmt_feb_2026", "score": 92.5, "reference_url": "https://z.ai/blog/glm-5.2", "reported_setting": { "mode": "thinking", "effort": "max", "sampling": "pass@1", "prompt_style": "default", "tools": "none", "temperature": "1.0", "top_p": "0.95", "context": "default", "max_output_tokens": "163840", "judge": "GPT-5.5 medium for AIME/HMMT/IMO; benchmark-specified otherwise", "harness": "official reasoning evaluation" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "GLM-5.2 official release source: GLM-5.2 / HMMT Feb. 2026 = 92.5.", "candidates": [] }, { "model_id": "glm-5.2", "benchmark_id": "imo_answerbench", "score": 91.0, "reference_url": "https://z.ai/blog/glm-5.2", "reported_setting": { "mode": "thinking", "effort": "max", "sampling": "pass@1", "prompt_style": "default", "tools": "none", "temperature": "1.0", "top_p": "0.95", "context": "default", "max_output_tokens": "163840", "judge": "GPT-5.5 medium for AIME/HMMT/IMO; benchmark-specified otherwise", "harness": "official reasoning evaluation" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "GLM-5.2 official release source: GLM-5.2 / IMOAnswerBench = 91.0.", "candidates": [] }, { "model_id": "glm-5.2", "benchmark_id": "swe_bench_pro", "score": 62.1, "reference_url": "https://z.ai/blog/glm-5.2", "reported_setting": { "mode": "thinking", "effort": "max", "sampling": "pass@1", "prompt_style": "default", "tools": "agentic code execution", "temperature": "1.0", "top_p": "1.0", "context": "400000", "max_output_tokens": "32000", "judge": "official tests", "harness": "OpenHands with tailored instruction prompt" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "GLM-5.2 official release source: GLM-5.2 / SWE-bench Pro = 62.1.", "candidates": [ { "score": 62.1, "reference_url": "https://media.x.ai/v1/website/4p5-5184fdf9.pdf", "source_type": "model_card", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "agentic repository shell/editor", "sampling": "pass@1", "judge": "verification reward and repository tests", "harness": "fixed SWE-bench Pro agent scaffold", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown" }, "notes": "card45 SWE-Bench Pro Resolve rate (%); source effort=max; benchmark protocol matches canonical=True. Unknown source fields remain unknown." }, { "score": 62.1, "reference_url": "https://x.ai/news/grok-4-5", "source_type": "official_blog", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "agentic repository shell/editor", "sampling": "pass@1", "judge": "verification reward and repository tests", "harness": "fixed SWE-bench Pro agent scaffold", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown" }, "notes": "release45 SWE-Bench Pro Resolve rate (%); source effort=unknown; benchmark protocol matches canonical=True. Unknown source fields remain unknown." } ] }, { "model_id": "glm-5.2", "benchmark_id": "swe_marathon_v1_0", "score": 13.0, "reference_url": "https://z.ai/blog/glm-5.2", "reported_setting": { "mode": "thinking", "effort": "max", "sampling": "pass@1", "prompt_style": "default", "tools": "agentic code execution", "context": "1000000", "max_output_tokens": "128000", "judge": "official task verifiers", "harness": "Abundant AI SWE-Marathon v1.0; max effort", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "GLM-5.2 official release source: GLM-5.2 / SWE-Marathon = 13.0.", "candidates": [ { "score": 13.0, "reference_url": "https://media.x.ai/v1/website/4p5-5184fdf9.pdf", "source_type": "model_card", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "agentic code execution", "sampling": "pass@1", "judge": "multi-layer resolution verification", "harness": "SWE-Marathon v1.0 harness", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown" }, "notes": "card45 SWE-Marathon v1.0 Resolution rate (%); source effort=max; benchmark protocol matches canonical=True. Unknown source fields remain unknown." } ] }, { "model_id": "glm-5.2", "benchmark_id": "toolathlon", "score": 48.2, "reference_url": "https://z.ai/blog/glm-5.2", "reported_setting": { "mode": "thinking", "effort": "max", "sampling": "pass@1", "prompt_style": "default", "tools": "agentic tool use", "context": "default", "max_output_tokens": "128000", "judge": "official evaluation service", "harness": "original Toolathlon public evaluation service (pre-Verified)", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "GLM-5.2 official release source: GLM-5.2 / Tool-Decathlon = 48.2.", "candidates": [ { "score": 46.6, "reference_url": "https://huggingface.co/tencent/Hy3", "source_type": "model_card", "reported_setting": { "mode": "thinking", "effort": "high / highest available", "tools": "agentic tool use", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "original Toolathlon official evaluation", "prompt_style": "default", "temperature": "0.9 for Hy3 family; otherwise source/provider setting", "context": "256000 for Hy3 family; otherwise source/provider setting", "notes": "Tencent own test" }, "notes": "Hy3 official model-card appendix: Toolathlon = 46.6*. Tencent own testing." } ] }, { "model_id": "glm-5.1", "benchmark_id": "hle_tools_text", "score": 52.3, "reference_url": "https://z.ai/blog/glm-5.2", "reported_setting": { "mode": "thinking", "effort": "official reported", "sampling": "pass@1", "prompt_style": "default", "tools": "search+code+web", "temperature": "1.0", "top_p": "0.95", "context": "300000", "max_output_tokens": "163840", "judge": "answer-key scoring", "harness": "official; no context-management strategy" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "GLM-5.2 official release source: GLM-5.1 / HLE (w/ Tools) = 52.3.", "candidates": [] }, { "model_id": "glm-5.1", "benchmark_id": "critpt", "score": 4.6, "reference_url": "https://z.ai/blog/glm-5.2", "reported_setting": { "mode": "thinking", "effort": "official reported", "sampling": "pass@1", "prompt_style": "default", "tools": "none", "temperature": "1.0", "top_p": "0.95", "context": "default", "max_output_tokens": "163840", "judge": "GPT-5.5 medium for AIME/HMMT/IMO; benchmark-specified otherwise", "harness": "official reasoning evaluation" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "GLM-5.2 official release source: GLM-5.1 / CritPt = 4.6.", "candidates": [ { "score": 3.7, "reference_url": "https://research.nvidia.com/labs/nemotron/files/NVIDIA-Nemotron-3-Ultra-Technical-Report.pdf", "source_type": "tech_report", "reported_setting": { "version_metric": "no tools / %", "tools": "none", "harness": "NeMo Evaluator", "sampling": "5-run average" }, "notes": "NVIDIA Nemotron 3 Ultra technical report, Final Table 10: GLM-5.1: critpt=3.7." } ] }, { "model_id": "glm-5.1", "benchmark_id": "deep_swe_v1_1", "score": 18.0, "reference_url": "https://z.ai/blog/glm-5.2", "reported_setting": { "mode": "thinking", "effort": "official reported", "sampling": "pass@1", "prompt_style": "default", "tools": "agentic code execution", "temperature": "1.0", "top_p": "1.0", "context": "400000", "judge": "isolated-container verification", "harness": "official pier + mini-swe-agent; timeout=2h; 2 CPU/8GB; no internet" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "GLM-5.2 official release source: GLM-5.1 / DeepSWE = 18.0.", "candidates": [] }, { "model_id": "glm-5.1", "benchmark_id": "program_bench", "score": 50.9, "reference_url": "https://z.ai/blog/glm-5.2", "reported_setting": { "mode": "thinking", "effort": "official reported", "sampling": "pass@1", "prompt_style": "default", "tools": "agentic code execution", "temperature": "1.0", "top_p": "1.0", "context": "400000", "max_output_tokens": "64000", "judge": "248,000+ fuzz-generated behavioral tests", "harness": "Claude Code 2.1.156; max_turns=2000; timeout=6h; 4 CPU/8GB; no internet" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "GLM-5.2 official release source: GLM-5.1 / ProgramBench = 50.9.", "candidates": [] }, { "model_id": "glm-5.1", "benchmark_id": "terminal_bench_2_1", "score": 69.0, "reference_url": "https://z.ai/blog/glm-5.2", "reported_setting": { "mode": "thinking", "effort": "official reported", "sampling": "pass@1", "prompt_style": "default", "tools": "terminal agent", "temperature": "1.0 for Claude Code; unknown otherwise", "top_p": "0.95 for Claude Code; unknown otherwise", "context": "source-reported", "max_output_tokens": "131072 for Claude Code; source-reported otherwise", "judge": "official task verifier", "harness": "Claude Code 2.1.167; Claude Code values averaged over 5 runs" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "GLM-5.2 official release source: GLM-5.1 / Terminal Bench 2.1 (Best Reported Harness) = 69.0.", "candidates": [ { "score": 63.5, "reference_url": "https://z.ai/blog/glm-5.2", "source_type": "official_blog", "reported_setting": { "mode": "thinking", "effort": "official reported", "sampling": "pass@1", "prompt_style": "default", "tools": "terminal agent", "temperature": "1.0", "top_p": "1.0", "context": "256000", "max_output_tokens": "48000", "judge": "official task verifier", "harness": "Terminus-2; parser=json; timeout=4h; max_episodes=500; 4 CPU/8GB" }, "notes": "Z.ai first-party Terminus-2 alternative." }, { "score": 60.2, "reference_url": "https://huggingface.co/tencent/Hy3", "source_type": "model_card", "reported_setting": { "mode": "thinking", "effort": "high / highest available", "tools": "terminal/code execution", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "Terminus-2; parser=xml; timeout=4h; CPU=16 cores; memory=32GB; max episodes=500", "prompt_style": "default", "temperature": "0.9 for Hy3 family; otherwise source/provider setting", "context": "256000 for Hy3 family; otherwise source/provider setting", "notes": "Tencent own test" }, "notes": "Hy3 official model-card appendix: Terminal-Bench 2.1 = 60.2*. Tencent own testing." }, { "score": 59.3, "reference_url": "https://research.nvidia.com/labs/nemotron/files/NVIDIA-Nemotron-3-Ultra-Technical-Report.pdf", "source_type": "tech_report", "reported_setting": { "version_metric": "v2.1 / % resolved", "tools": "terminal and code execution", "harness": "Harbor / Terminus-2", "sampling": "8-run average" }, "notes": "NVIDIA Nemotron 3 Ultra technical report, Final Table 10: GLM-5.1: terminal_bench_2_1=59.3." }, { "score": 61.8, "reference_url": "https://research.nvidia.com/labs/nemotron/files/NVIDIA-Nemotron-3-Ultra-Technical-Report.pdf", "source_type": "tech_report", "reported_setting": { "version_metric": "OpenCode scaffold / %", "tools": "agent-specific software-engineering environment", "harness": "OpenCode", "sampling": "single agent result", "variant": "OpenCode", "agent_scaffold": "OpenCode" }, "notes": "NVIDIA Nemotron 3 Ultra technical report, Figure 17: GLM-5.1: terminal_bench_2_1=61.8, variant=OpenCode." }, { "score": 61.4, "reference_url": "https://research.nvidia.com/labs/nemotron/files/NVIDIA-Nemotron-3-Ultra-Technical-Report.pdf", "source_type": "tech_report", "reported_setting": { "version_metric": "Pi scaffold / %", "tools": "agent-specific software-engineering environment", "harness": "Pi", "sampling": "single agent result", "variant": "Pi", "agent_scaffold": "Pi" }, "notes": "NVIDIA Nemotron 3 Ultra technical report, Figure 17: GLM-5.1: terminal_bench_2_1=61.4, variant=Pi." }, { "score": 51.9, "reference_url": "https://research.nvidia.com/labs/nemotron/files/NVIDIA-Nemotron-3-Ultra-Technical-Report.pdf", "source_type": "tech_report", "reported_setting": { "version_metric": "Claude scaffold / %", "tools": "agent-specific software-engineering environment", "harness": "Claude", "sampling": "single agent result", "variant": "Claude", "agent_scaffold": "Claude" }, "notes": "NVIDIA Nemotron 3 Ultra technical report, Figure 17: GLM-5.1: terminal_bench_2_1=51.9, variant=Claude." }, { "score": 71.3, "reference_url": "https://research.nvidia.com/labs/nemotron/files/NVIDIA-Nemotron-3-Ultra-Technical-Report.pdf", "source_type": "tech_report", "reported_setting": { "version_metric": "Hermes scaffold / %", "tools": "agent-specific software-engineering environment", "harness": "Hermes", "sampling": "single agent result", "variant": "Hermes", "agent_scaffold": "Hermes" }, "notes": "NVIDIA Nemotron 3 Ultra technical report, Figure 17: GLM-5.1: terminal_bench_2_1=71.3, variant=Hermes." }, { "score": 55.5, "reference_url": "https://research.nvidia.com/labs/nemotron/files/NVIDIA-Nemotron-3-Ultra-Technical-Report.pdf", "source_type": "tech_report", "reported_setting": { "version_metric": "OpenHands scaffold / %", "tools": "agent-specific software-engineering environment", "harness": "OpenHands", "sampling": "single agent result", "variant": "OpenHands", "agent_scaffold": "OpenHands" }, "notes": "NVIDIA Nemotron 3 Ultra technical report, Figure 17: GLM-5.1: terminal_bench_2_1=55.5, variant=OpenHands." }, { "score": 51.2, "reference_url": "https://research.nvidia.com/labs/nemotron/files/NVIDIA-Nemotron-3-Ultra-Technical-Report.pdf", "source_type": "tech_report", "reported_setting": { "version_metric": "Codex scaffold / %", "tools": "agent-specific software-engineering environment", "harness": "Codex", "sampling": "single agent result", "variant": "Codex", "agent_scaffold": "Codex" }, "notes": "NVIDIA Nemotron 3 Ultra technical report, Figure 17: GLM-5.1: terminal_bench_2_1=51.2, variant=Codex." }, { "score": 58.9, "reference_url": "https://research.nvidia.com/labs/nemotron/files/NVIDIA-Nemotron-3-Ultra-Technical-Report.pdf", "source_type": "tech_report", "reported_setting": { "version_metric": "Average scaffold / %", "tools": "agent-specific software-engineering environment", "harness": "Average", "sampling": "single agent result", "variant": "Average", "agent_scaffold": "Average" }, "notes": "NVIDIA Nemotron 3 Ultra technical report, Figure 17: GLM-5.1: terminal_bench_2_1=58.9, variant=Average." } ] }, { "model_id": "glm-5.1", "benchmark_id": "frontier_swe", "score": 30.5, "reference_url": "https://z.ai/blog/glm-5.2", "reported_setting": { "mode": "thinking", "effort": "official reported", "sampling": "pass@1", "prompt_style": "default", "tools": "agentic code execution", "context": "1000000", "max_output_tokens": "128000", "judge": "FrontierSWE dominance", "harness": "Proximal evaluation; max effort; as of 2026-06-16", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "GLM-5.2 official release source: GLM-5.1 / FrontierSWE (Dominance) = 30.5.", "candidates": [] }, { "model_id": "glm-5.1", "benchmark_id": "posttrain_bench", "score": 20.1, "reference_url": "https://z.ai/blog/glm-5.2", "reported_setting": { "mode": "thinking", "effort": "official reported", "sampling": "pass@1", "prompt_style": "default", "tools": "agentic post-training", "context": "1000000", "max_output_tokens": "128000", "judge": "official weighted benchmark score", "harness": "PostTrainBench evaluation; max effort", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "GLM-5.2 official release source: GLM-5.1 / PostTrainBench = 20.1.", "candidates": [] }, { "model_id": "glm-5.1", "benchmark_id": "swe_marathon_v1_0", "score": 1.0, "reference_url": "https://z.ai/blog/glm-5.2", "reported_setting": { "mode": "thinking", "effort": "official reported", "sampling": "pass@1", "prompt_style": "default", "tools": "agentic code execution", "context": "1000000", "max_output_tokens": "128000", "judge": "official task verifiers", "harness": "Abundant AI SWE-Marathon v1.0; max effort", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "GLM-5.2 official release source: GLM-5.1 / SWE-Marathon = 1.0.", "candidates": [] }, { "model_id": "qwen3.7-max", "benchmark_id": "hle_text", "score": 41.4, "reference_url": "https://www.alibabacloud.com/blog/qwen3-7-the-agent-frontier_603154", "reported_setting": { "mode": "thinking", "effort": "xhigh / maximum thinking", "sampling": "pass@1", "prompt_style": "default", "tools": "none", "temperature": "1.0", "top_p": "0.95", "context": "default", "max_output_tokens": "163840", "judge": "GPT-5.5 medium for AIME/HMMT/IMO; benchmark-specified otherwise", "harness": "official reasoning evaluation" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "Qwen3.7-Max first-party source: HLE = 41.4; GLM-5.2 table used as cross-check.", "candidates": [] }, { "model_id": "qwen3.7-max", "benchmark_id": "hle_tools_text", "score": 53.5, "reference_url": "https://z.ai/blog/glm-5.2", "reported_setting": { "mode": "thinking", "effort": "xhigh / maximum thinking", "sampling": "pass@1", "prompt_style": "default", "tools": "search+code+web", "temperature": "1.0", "top_p": "0.95", "context": "300000", "max_output_tokens": "163840", "judge": "answer-key scoring", "harness": "official; no context-management strategy" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "GLM-5.2 official release source: Qwen3.7-Max / HLE (w/ Tools) = 53.5.", "candidates": [] }, { "model_id": "qwen3.7-max", "benchmark_id": "critpt", "score": 13.4, "reference_url": "https://z.ai/blog/glm-5.2", "reported_setting": { "mode": "thinking", "effort": "xhigh / maximum thinking", "sampling": "pass@1", "prompt_style": "default", "tools": "none", "temperature": "1.0", "top_p": "0.95", "context": "default", "max_output_tokens": "163840", "judge": "GPT-5.5 medium for AIME/HMMT/IMO; benchmark-specified otherwise", "harness": "official reasoning evaluation" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "GLM-5.2 official release source: Qwen3.7-Max / CritPt = 13.4.", "candidates": [] }, { "model_id": "qwen3.7-max", "benchmark_id": "aime_2026", "score": 97.0, "reference_url": "https://z.ai/blog/glm-5.2", "reported_setting": { "mode": "thinking", "effort": "xhigh / maximum thinking", "sampling": "pass@1", "prompt_style": "default", "tools": "none", "temperature": "1.0", "top_p": "0.95", "context": "default", "max_output_tokens": "163840", "judge": "GPT-5.5 medium for AIME/HMMT/IMO; benchmark-specified otherwise", "harness": "official reasoning evaluation" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "GLM-5.2 official release source: Qwen3.7-Max / AIME 2026 = 97.0.", "candidates": [] }, { "model_id": "qwen3.7-max", "benchmark_id": "hmmt_nov_2025", "score": 95.0, "reference_url": "https://z.ai/blog/glm-5.2", "reported_setting": { "mode": "thinking", "effort": "xhigh / maximum thinking", "sampling": "pass@1", "prompt_style": "default", "tools": "none", "temperature": "1.0", "top_p": "0.95", "context": "default", "max_output_tokens": "163840", "judge": "GPT-5.5 medium for AIME/HMMT/IMO; benchmark-specified otherwise", "harness": "official reasoning evaluation" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "GLM-5.2 official release source: Qwen3.7-Max / HMMT Nov. 2025 = 95.0.", "candidates": [] }, { "model_id": "qwen3.7-max", "benchmark_id": "hmmt_feb_2026", "score": 97.1, "reference_url": "https://www.alibabacloud.com/blog/qwen3-7-the-agent-frontier_603154", "reported_setting": { "mode": "thinking", "effort": "xhigh / maximum thinking", "sampling": "pass@1", "prompt_style": "default", "tools": "none", "temperature": "1.0", "top_p": "0.95", "context": "default", "max_output_tokens": "163840", "judge": "GPT-5.5 medium for AIME/HMMT/IMO; benchmark-specified otherwise", "harness": "official reasoning evaluation" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "Qwen3.7-Max first-party source: HMMT Feb. 2026 = 97.1; GLM-5.2 table used as cross-check.", "candidates": [] }, { "model_id": "qwen3.7-max", "benchmark_id": "imo_answerbench", "score": 90.0, "reference_url": "https://www.alibabacloud.com/blog/qwen3-7-the-agent-frontier_603154", "reported_setting": { "mode": "thinking", "effort": "xhigh / maximum thinking", "sampling": "pass@1", "prompt_style": "default", "tools": "none", "temperature": "1.0", "top_p": "0.95", "context": "default", "max_output_tokens": "163840", "judge": "GPT-5.5 medium for AIME/HMMT/IMO; benchmark-specified otherwise", "harness": "official reasoning evaluation" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "Qwen3.7-Max first-party source: IMOAnswerBench = 90.0; GLM-5.2 table used as cross-check.", "candidates": [] }, { "model_id": "qwen3.7-max", "benchmark_id": "gpqa_diamond", "score": 92.4, "reference_url": "https://www.alibabacloud.com/blog/qwen3-7-the-agent-frontier_603154", "reported_setting": { "mode": "thinking", "effort": "xhigh / maximum thinking", "sampling": "pass@1", "prompt_style": "default", "tools": "none", "temperature": "1.0", "top_p": "0.95", "context": "default", "max_output_tokens": "163840", "judge": "GPT-5.5 medium for AIME/HMMT/IMO; benchmark-specified otherwise", "harness": "official reasoning evaluation" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "Qwen3.7-Max first-party source: GPQA-Diamond = 92.4; GLM-5.2 table used as cross-check.", "candidates": [ { "score": 90.0, "reference_url": "https://z.ai/blog/glm-5.2", "source_type": "official_blog", "reported_setting": { "mode": "thinking", "effort": "xhigh / maximum thinking", "sampling": "pass@1", "prompt_style": "default", "tools": "none", "temperature": "1.0", "top_p": "0.95", "context": "default", "max_output_tokens": "163840", "judge": "GPT-5.5 medium for AIME/HMMT/IMO; benchmark-specified otherwise", "harness": "official reasoning evaluation" }, "notes": "GLM-5.2 cross-table value; Qwen first-party 92.4 is primary." } ] }, { "model_id": "qwen3.7-max", "benchmark_id": "swe_bench_pro", "score": 60.6, "reference_url": "https://www.alibabacloud.com/blog/qwen3-7-the-agent-frontier_603154", "reported_setting": { "mode": "thinking", "effort": "xhigh / maximum thinking", "sampling": "pass@1", "prompt_style": "default", "tools": "agentic code execution", "temperature": "1.0", "top_p": "1.0", "context": "400000", "max_output_tokens": "32000", "judge": "official tests", "harness": "OpenHands with tailored instruction prompt" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "Qwen3.7-Max first-party source: SWE-bench Pro = 60.6; GLM-5.2 table used as cross-check.", "candidates": [] }, { "model_id": "qwen3.7-max", "benchmark_id": "nl2repo_bench", "score": 47.2, "reference_url": "https://z.ai/blog/glm-5.2", "reported_setting": { "mode": "thinking", "effort": "xhigh / maximum thinking", "sampling": "pass@1", "prompt_style": "default", "tools": "agentic code execution", "temperature": "1.0", "top_p": "1.0", "context": "400000", "max_output_tokens": "48000", "judge": "rule-based + LLM anti-hack judgement", "harness": "official NL2Repo evaluation" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "GLM-5.2 official release source: Qwen3.7-Max / NL2Repo = 47.2.", "candidates": [] }, { "model_id": "qwen3.7-max", "benchmark_id": "deep_swe_v1_1", "score": 18.0, "reference_url": "https://z.ai/blog/glm-5.2", "reported_setting": { "mode": "thinking", "effort": "xhigh / maximum thinking", "sampling": "pass@1", "prompt_style": "default", "tools": "agentic code execution", "temperature": "1.0", "top_p": "1.0", "context": "400000", "judge": "isolated-container verification", "harness": "official pier + mini-swe-agent; timeout=2h; 2 CPU/8GB; no internet" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "GLM-5.2 official release source: Qwen3.7-Max / DeepSWE = 18.0.", "candidates": [ { "score": 14.2, "reference_url": "https://huggingface.co/tencent/Hy3", "source_type": "model_card", "reported_setting": { "mode": "thinking", "effort": "high / highest available", "tools": "agentic code execution", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "mini-swe-agent; 2h/task; 2 CPU/8GB; no internet", "prompt_style": "default", "temperature": "0.9 for Hy3 family; otherwise source/provider setting", "context": "256000 for Hy3 family; otherwise source/provider setting", "notes": "Tencent own test" }, "notes": "Hy3 official model-card appendix: DeepSWE = 14.2*. Tencent own testing." } ] }, { "model_id": "qwen3.7-max", "benchmark_id": "terminal_bench_2_1", "score": 75.0, "reference_url": "https://z.ai/blog/glm-5.2", "reported_setting": { "mode": "thinking", "effort": "xhigh / maximum thinking", "sampling": "pass@1", "prompt_style": "default", "tools": "terminal agent", "temperature": "1.0", "top_p": "1.0", "context": "256000", "max_output_tokens": "48000", "judge": "official task verifier", "harness": "Terminus-2; parser=json; timeout=4h; max_episodes=500; 4 CPU/8GB" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "GLM-5.2 official release source: Qwen3.7-Max / Terminal Bench 2.1 (Terminus-2) = 75.0.", "candidates": [ { "score": 71.5, "reference_url": "https://huggingface.co/tencent/Hy3", "source_type": "model_card", "reported_setting": { "mode": "thinking", "effort": "high / highest available", "tools": "terminal/code execution", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "Terminus-2; parser=xml; timeout=4h; CPU=16 cores; memory=32GB; max episodes=500", "prompt_style": "default", "temperature": "0.9 for Hy3 family; otherwise source/provider setting", "context": "256000 for Hy3 family; otherwise source/provider setting", "notes": "Tencent own test" }, "notes": "Hy3 official model-card appendix: Terminal-Bench 2.1 = 71.5*. Tencent own testing." } ] }, { "model_id": "qwen3.7-max", "benchmark_id": "mcpatlas", "score": 76.4, "reference_url": "https://www.alibabacloud.com/blog/qwen3-7-the-agent-frontier_603154", "reported_setting": { "mode": "thinking", "effort": "xhigh / maximum thinking", "sampling": "pass@1", "prompt_style": "default", "tools": "agentic MCP servers", "context": "default", "judge": "Gemini 3.0 Pro", "harness": "public 500-task set; 10-minute timeout per task", "temperature": "default" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Qwen3.7-Max first-party source: MCP-Atlas (Public Set) = 76.4; GLM-5.2 table used as cross-check.", "candidates": [ { "score": 79.6, "reference_url": "https://huggingface.co/tencent/Hy3", "source_type": "model_card", "reported_setting": { "mode": "thinking", "effort": "high / highest available", "tools": "MCP tool servers", "sampling": "pass@1", "judge": "Gemini 2.5 Pro", "harness": "Scale April 2026 methodology; public 500; 100 tool-call budget", "prompt_style": "default", "temperature": "0.9 for Hy3 family; otherwise source/provider setting", "context": "256000 for Hy3 family; otherwise source/provider setting", "notes": "Tencent own test" }, "notes": "Hy3 official model-card appendix: MCP atlas (public) = 79.6*. Tencent own testing." } ] }, { "model_id": "minimax-m3", "benchmark_id": "hle_text", "score": 37.0, "reference_url": "https://z.ai/blog/glm-5.2", "reported_setting": { "mode": "thinking", "effort": "thinking enabled/adaptive", "sampling": "pass@1", "prompt_style": "default", "tools": "none", "temperature": "1.0", "top_p": "0.95", "context": "default", "max_output_tokens": "163840", "judge": "GPT-5.5 medium for AIME/HMMT/IMO; benchmark-specified otherwise", "harness": "official reasoning evaluation" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "GLM-5.2 official release source: MiniMax M3 / HLE = 37.0.", "candidates": [] }, { "model_id": "minimax-m3", "benchmark_id": "critpt", "score": 3.7, "reference_url": "https://z.ai/blog/glm-5.2", "reported_setting": { "mode": "thinking", "effort": "thinking enabled/adaptive", "sampling": "pass@1", "prompt_style": "default", "tools": "none", "temperature": "1.0", "top_p": "0.95", "context": "default", "max_output_tokens": "163840", "judge": "GPT-5.5 medium for AIME/HMMT/IMO; benchmark-specified otherwise", "harness": "official reasoning evaluation" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "GLM-5.2 official release source: MiniMax M3 / CritPt = 3.7.", "candidates": [] }, { "model_id": "minimax-m3", "benchmark_id": "hmmt_nov_2025", "score": 84.4, "reference_url": "https://z.ai/blog/glm-5.2", "reported_setting": { "mode": "thinking", "effort": "thinking enabled/adaptive", "sampling": "pass@1", "prompt_style": "default", "tools": "none", "temperature": "1.0", "top_p": "0.95", "context": "default", "max_output_tokens": "163840", "judge": "GPT-5.5 medium for AIME/HMMT/IMO; benchmark-specified otherwise", "harness": "official reasoning evaluation" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "GLM-5.2 official release source: MiniMax M3 / HMMT Nov. 2025 = 84.4.", "candidates": [] }, { "model_id": "minimax-m3", "benchmark_id": "hmmt_feb_2026", "score": 84.4, "reference_url": "https://z.ai/blog/glm-5.2", "reported_setting": { "mode": "thinking", "effort": "thinking enabled/adaptive", "sampling": "pass@1", "prompt_style": "default", "tools": "none", "temperature": "1.0", "top_p": "0.95", "context": "default", "max_output_tokens": "163840", "judge": "GPT-5.5 medium for AIME/HMMT/IMO; benchmark-specified otherwise", "harness": "official reasoning evaluation" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "GLM-5.2 official release source: MiniMax M3 / HMMT Feb. 2026 = 84.4.", "candidates": [] }, { "model_id": "minimax-m3", "benchmark_id": "gpqa_diamond", "score": 93.0, "reference_url": "https://z.ai/blog/glm-5.2", "reported_setting": { "mode": "thinking", "effort": "thinking enabled/adaptive", "sampling": "pass@1", "prompt_style": "default", "tools": "none", "temperature": "1.0", "top_p": "0.95", "context": "default", "max_output_tokens": "163840", "judge": "GPT-5.5 medium for AIME/HMMT/IMO; benchmark-specified otherwise", "harness": "official reasoning evaluation" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "GLM-5.2 official release source: MiniMax M3 / GPQA-Diamond = 93.0.", "candidates": [] }, { "model_id": "minimax-m3", "benchmark_id": "swe_bench_pro", "score": 59.0, "reference_url": "https://www.minimax.io/blog/minimax-m3", "reported_setting": { "mode": "thinking", "effort": "thinking enabled/adaptive", "sampling": "pass@1", "prompt_style": "default", "tools": "agentic code execution", "temperature": "1.0", "top_p": "1.0", "context": "400000", "max_output_tokens": "32000", "judge": "official tests", "harness": "OpenHands with tailored instruction prompt" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "MiniMax M3 first-party source: SWE-bench Pro = 59.0; GLM-5.2 table used as cross-check.", "candidates": [] }, { "model_id": "minimax-m3", "benchmark_id": "nl2repo_bench", "score": 42.1, "reference_url": "https://z.ai/blog/glm-5.2", "reported_setting": { "mode": "thinking", "effort": "thinking enabled/adaptive", "sampling": "pass@1", "prompt_style": "default", "tools": "agentic code execution", "temperature": "1.0", "top_p": "1.0", "context": "400000", "max_output_tokens": "48000", "judge": "rule-based + LLM anti-hack judgement", "harness": "official NL2Repo evaluation" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "GLM-5.2 official release source: MiniMax M3 / NL2Repo = 42.1.", "candidates": [] }, { "model_id": "minimax-m3", "benchmark_id": "deep_swe_v1_1", "score": 20.0, "reference_url": "https://z.ai/blog/glm-5.2", "reported_setting": { "mode": "thinking", "effort": "thinking enabled/adaptive", "sampling": "pass@1", "prompt_style": "default", "tools": "agentic code execution", "temperature": "1.0", "top_p": "1.0", "context": "400000", "judge": "isolated-container verification", "harness": "official pier + mini-swe-agent; timeout=2h; 2 CPU/8GB; no internet" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "GLM-5.2 official release source: MiniMax M3 / DeepSWE = 20.0.", "candidates": [] }, { "model_id": "minimax-m3", "benchmark_id": "terminal_bench_2_1", "score": 66.0, "reference_url": "https://www.minimax.io/blog/minimax-m3", "reported_setting": { "mode": "thinking", "effort": "thinking enabled/adaptive", "sampling": "pass@1", "prompt_style": "default", "tools": "terminal agent", "temperature": "1.0", "top_p": "1.0", "context": "256000", "max_output_tokens": "48000", "judge": "official task verifier", "harness": "Terminus-2; parser=json; timeout=4h; max_episodes=500; 4 CPU/8GB" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "MiniMax M3 first-party source: Terminal Bench 2.1 (Terminus-2) = 66.0; GLM-5.2 table used as cross-check.", "candidates": [ { "score": 65.0, "reference_url": "https://z.ai/blog/glm-5.2", "source_type": "official_blog", "reported_setting": { "mode": "thinking", "effort": "thinking enabled/adaptive", "sampling": "pass@1", "prompt_style": "default", "tools": "terminal agent", "temperature": "1.0", "top_p": "1.0", "context": "256000", "max_output_tokens": "48000", "judge": "official task verifier", "harness": "Terminus-2; parser=json; timeout=4h; max_episodes=500; 4 CPU/8GB" }, "notes": "GLM-5.2 cross-table value; MiniMax first-party 66.0 is primary." } ] }, { "model_id": "minimax-m3", "benchmark_id": "mcpatlas", "score": 74.2, "reference_url": "https://www.minimax.io/blog/minimax-m3", "reported_setting": { "mode": "thinking", "effort": "thinking enabled/adaptive", "sampling": "pass@1", "prompt_style": "default", "tools": "agentic MCP servers", "context": "default", "judge": "Gemini 3.0 Pro", "harness": "public 500-task set; 10-minute timeout per task", "temperature": "default" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "MiniMax M3 first-party source: MCP-Atlas (Public Set) = 74.2; GLM-5.2 table used as cross-check.", "candidates": [] }, { "model_id": "deepseek-v4-pro", "benchmark_id": "hle_tools_text", "score": 48.2, "reference_url": "https://huggingface.co/deepseek-ai/DeepSeek-V4-Pro", "reported_setting": { "mode": "thinking", "effort": "Max", "sampling": "pass@1", "prompt_style": "default", "tools": "search+code+web", "temperature": "1.0", "top_p": "0.95", "context": "300000", "max_output_tokens": "163840", "judge": "answer-key scoring", "harness": "official; no context-management strategy" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "DeepSeek-V4-Pro first-party source: HLE (w/ Tools) = 48.2; GLM-5.2 table used as cross-check.", "candidates": [] }, { "model_id": "deepseek-v4-pro", "benchmark_id": "critpt", "score": 12.9, "reference_url": "https://z.ai/blog/glm-5.2", "reported_setting": { "mode": "thinking", "effort": "Max", "sampling": "pass@1", "prompt_style": "default", "tools": "none", "temperature": "1.0", "top_p": "0.95", "context": "default", "max_output_tokens": "163840", "judge": "GPT-5.5 medium for AIME/HMMT/IMO; benchmark-specified otherwise", "harness": "official reasoning evaluation" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "GLM-5.2 official release source: DeepSeek-V4-Pro / CritPt = 12.9.", "candidates": [ { "score": 14.0, "reference_url": "https://research.nvidia.com/labs/nemotron/files/NVIDIA-Nemotron-3-Ultra-Technical-Report.pdf", "source_type": "tech_report", "reported_setting": { "version_metric": "no tools / %", "tools": "none", "harness": "NeMo Evaluator", "sampling": "5-run average" }, "notes": "NVIDIA Nemotron 3 Ultra technical report, Final Table 10: DeepSeek-V4-Pro: critpt=14." } ] }, { "model_id": "deepseek-v4-pro", "benchmark_id": "aime_2026", "score": 94.6, "reference_url": "https://z.ai/blog/glm-5.2", "reported_setting": { "mode": "thinking", "effort": "Max", "sampling": "pass@1", "prompt_style": "default", "tools": "none", "temperature": "1.0", "top_p": "0.95", "context": "default", "max_output_tokens": "163840", "judge": "GPT-5.5 medium for AIME/HMMT/IMO; benchmark-specified otherwise", "harness": "official reasoning evaluation" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "GLM-5.2 official release source: DeepSeek-V4-Pro / AIME 2026 = 94.6.", "candidates": [ { "score": 92.6, "reference_url": "https://ernie.baidu.com/blog/posts/ernie-5.1-0508-release/benchmark.png", "source_type": "official_blog", "reported_setting": { "tools": "tools enabled (unspecified)", "sampling": "avg@32", "metric": "score" }, "notes": "With-tool avg@32 value is noncanonical against pass@1/no-tools. Research observation obs-002." }, { "score": 95.8, "reference_url": "https://ernie.baidu.com/blog/posts/ernie-5.1-0508-release/benchmark.png", "source_type": "official_blog", "reported_setting": { "tools": "none implied by contrast with w/tools row", "sampling": "avg@32" }, "notes": "No-tool avg@32 value is noncanonical against pass@1. Research observation obs-018. Literal marker ▲ has no source legend. Marker has no legend in image or article." } ] }, { "model_id": "deepseek-v4-pro", "benchmark_id": "hmmt_nov_2025", "score": 94.4, "reference_url": "https://z.ai/blog/glm-5.2", "reported_setting": { "mode": "thinking", "effort": "Max", "sampling": "pass@1", "prompt_style": "default", "tools": "none", "temperature": "1.0", "top_p": "0.95", "context": "default", "max_output_tokens": "163840", "judge": "GPT-5.5 medium for AIME/HMMT/IMO; benchmark-specified otherwise", "harness": "official reasoning evaluation" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "GLM-5.2 official release source: DeepSeek-V4-Pro / HMMT Nov. 2025 = 94.4.", "candidates": [] }, { "model_id": "deepseek-v4-pro", "benchmark_id": "program_bench", "score": 47.8, "reference_url": "https://z.ai/blog/glm-5.2", "reported_setting": { "mode": "thinking", "effort": "Max", "sampling": "pass@1", "prompt_style": "default", "tools": "agentic code execution", "temperature": "1.0", "top_p": "1.0", "context": "400000", "max_output_tokens": "64000", "judge": "248,000+ fuzz-generated behavioral tests", "harness": "Claude Code 2.1.156; max_turns=2000; timeout=6h; 4 CPU/8GB; no internet" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "GLM-5.2 official release source: DeepSeek-V4-Pro / ProgramBench = 47.8.", "candidates": [] }, { "model_id": "deepseek-v4-pro", "benchmark_id": "frontier_swe", "score": 29.0, "reference_url": "https://z.ai/blog/glm-5.2", "reported_setting": { "mode": "thinking", "effort": "Max", "sampling": "pass@1", "prompt_style": "default", "tools": "agentic code execution", "context": "1000000", "max_output_tokens": "128000", "judge": "FrontierSWE dominance", "harness": "Proximal evaluation; max effort; as of 2026-06-16", "temperature": "default" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "GLM-5.2 official release source: DeepSeek-V4-Pro / FrontierSWE (Dominance) = 29.0.", "candidates": [] }, { "model_id": "claude-opus-4.8", "benchmark_id": "aime_2026", "score": 95.7, "reference_url": "https://z.ai/blog/glm-5.2", "reported_setting": { "mode": "thinking", "effort": "max (source; model default is high)", "sampling": "pass@1", "prompt_style": "default", "tools": "none", "temperature": "1.0", "top_p": "0.95", "context": "default", "max_output_tokens": "163840", "judge": "GPT-5.5 medium for AIME/HMMT/IMO; benchmark-specified otherwise", "harness": "official reasoning evaluation" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "GLM-5.2 official release source: Claude Opus 4.8 / AIME 2026 = 95.7.", "candidates": [] }, { "model_id": "claude-opus-4.8", "benchmark_id": "hmmt_nov_2025", "score": 96.5, "reference_url": "https://z.ai/blog/glm-5.2", "reported_setting": { "mode": "thinking", "effort": "max (source; model default is high)", "sampling": "pass@1", "prompt_style": "default", "tools": "none", "temperature": "1.0", "top_p": "0.95", "context": "default", "max_output_tokens": "163840", "judge": "GPT-5.5 medium for AIME/HMMT/IMO; benchmark-specified otherwise", "harness": "official reasoning evaluation" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "GLM-5.2 official release source: Claude Opus 4.8 / HMMT Nov. 2025 = 96.5.", "candidates": [] }, { "model_id": "claude-opus-4.8", "benchmark_id": "hmmt_feb_2026", "score": 96.7, "reference_url": "https://z.ai/blog/glm-5.2", "reported_setting": { "mode": "thinking", "effort": "max (source; model default is high)", "sampling": "pass@1", "prompt_style": "default", "tools": "none", "temperature": "1.0", "top_p": "0.95", "context": "default", "max_output_tokens": "163840", "judge": "GPT-5.5 medium for AIME/HMMT/IMO; benchmark-specified otherwise", "harness": "official reasoning evaluation" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "GLM-5.2 official release source: Claude Opus 4.8 / HMMT Feb. 2026 = 96.7.", "candidates": [ { "score": 95.36, "reference_url": "https://huggingface.co/internlm/Intern-S2-Preview-397B", "source_type": "model_card", "reported_setting": { "mode": "source does not state", "effort": "source does not state", "tools": "none", "sampling": "source does not state", "judge": "benchmark-specific; exact version not stated", "harness": "OpenCompass or AgentCompass; exact adapter not stated", "prompt_style": "source does not state", "temperature": "source does not state", "context": "source does not state", "notes": "Shanghai AI Lab evaluated all model columns using OpenCompass, VLMEvalKit, and AgentCompass. Per-model mode/effort/sampling are not reported; only Intern-S2 maximum evaluation lengths are disclosed." }, "notes": "Intern-S2-Preview-397B official model card: HMMT-2026 = 95.36. This is a Shanghai AI Lab cross-model evaluation; the source does not disclose this model's exact mode, effort, or sampling." } ] }, { "model_id": "claude-opus-4.8", "benchmark_id": "imo_answerbench", "score": 83.5, "reference_url": "https://z.ai/blog/glm-5.2", "reported_setting": { "mode": "thinking", "effort": "max (source; model default is high)", "sampling": "pass@1", "prompt_style": "default", "tools": "none", "temperature": "1.0", "top_p": "0.95", "context": "default", "max_output_tokens": "163840", "judge": "GPT-5.5 medium for AIME/HMMT/IMO; benchmark-specified otherwise", "harness": "official reasoning evaluation" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "GLM-5.2 official release source: Claude Opus 4.8 / IMOAnswerBench = 83.5.", "candidates": [ { "score": 75.3, "reference_url": "https://longcat.chat/blog/longcat-2.0/", "source_type": "official_blog", "reported_setting": { "mode": "thinking", "effort": "source does not state", "tools": "none", "sampling": "source does not state", "judge": "IMO-AnswerBench answer autograder", "harness": "Meituan unified in-house harness", "prompt_style": "default", "temperature": "1.0; top_k=-1; top_p=0.95", "context": "source does not state", "notes": "Measured in-house by Meituan unless otherwise noted." }, "notes": "LongCat-2.0 official tech blog: IMO-AnswerBench = 75.3. Measured in-house by Meituan under the reported unified harness." } ] }, { "model_id": "claude-opus-4.8", "benchmark_id": "swe_bench_pro", "score": 69.2, "reference_url": "https://z.ai/blog/glm-5.2", "reported_setting": { "mode": "thinking", "effort": "max (source; model default is high)", "sampling": "pass@1", "prompt_style": "default", "tools": "agentic code execution", "temperature": "1.0", "top_p": "1.0", "context": "400000", "max_output_tokens": "32000", "judge": "official tests", "harness": "OpenHands with tailored instruction prompt" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "GLM-5.2 official release source: Claude Opus 4.8 / SWE-bench Pro = 69.2.", "candidates": [ { "score": 69.2, "reference_url": "https://x.ai/news/grok-4-5", "source_type": "official_blog", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "agentic repository shell/editor", "sampling": "pass@1", "judge": "verification reward and repository tests", "harness": "fixed SWE-bench Pro agent scaffold", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown" }, "notes": "release45 SWE-Bench Pro Resolve rate (%); source effort=max; benchmark protocol matches canonical=True. Unknown source fields remain unknown." } ] }, { "model_id": "claude-opus-4.8", "benchmark_id": "swe_marathon_v1_0", "score": 26.0, "reference_url": "https://z.ai/blog/glm-5.2", "reported_setting": { "mode": "thinking", "effort": "max (source; model default is high)", "sampling": "pass@1", "prompt_style": "default", "tools": "agentic code execution", "context": "1000000", "max_output_tokens": "128000", "judge": "official task verifiers", "harness": "Abundant AI SWE-Marathon v1.0; max effort", "temperature": "default" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "GLM-5.2 official release source: Claude Opus 4.8 / SWE-Marathon = 26.0.", "candidates": [ { "score": 26.0, "reference_url": "https://x.ai/news/grok-4-5", "source_type": "official_blog", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "agentic code execution", "sampling": "pass@1", "judge": "multi-layer resolution verification", "harness": "SWE-Marathon v1.0 harness", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown" }, "notes": "release45 SWE-Marathon v1.0 Resolution rate (%); source effort=max; benchmark protocol matches canonical=True. Unknown source fields remain unknown." } ] }, { "model_id": "claude-opus-4.8", "benchmark_id": "toolathlon", "score": 59.9, "reference_url": "https://www.anthropic.com/claude-sonnet-5-system-card", "reported_setting": { "effort": "max", "tools": "agentic benchmark harness", "sampling": "3 trials/task", "harness": "toolathlon" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Anthropic Claude Sonnet 5 System Card Table 8.11.5.A, page 135: Claude Opus 4.8; Toolathlon [Pass@1]=59.9. Source setting: effort=max; tools=agentic benchmark harness; sampling=3 trials/task; harness=toolathlon.", "candidates": [ { "score": 59.9, "reference_url": "https://z.ai/blog/glm-5.2", "source_type": "official_blog", "reported_setting": { "mode": "thinking", "effort": "max (source; model default is high)", "sampling": "pass@1", "prompt_style": "default", "tools": "agentic tool use", "context": "default", "max_output_tokens": "128000", "judge": "official evaluation service", "harness": "original Toolathlon public evaluation service (pre-Verified)", "temperature": "default" }, "notes": "GLM-5.2 official release source: Claude Opus 4.8 / Tool-Decathlon = 59.9." } ] }, { "model_id": "gpt-5.5", "benchmark_id": "aime_2026", "score": 98.3, "reference_url": "https://z.ai/blog/glm-5.2", "reported_setting": { "mode": "thinking", "effort": "xhigh", "sampling": "pass@1", "prompt_style": "default", "tools": "none", "temperature": "1.0", "top_p": "0.95", "context": "default", "max_output_tokens": "163840", "judge": "GPT-5.5 medium for AIME/HMMT/IMO; benchmark-specified otherwise", "harness": "official reasoning evaluation" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "GLM-5.2 official release source: GPT-5.5 / AIME 2026 = 98.3.", "candidates": [] }, { "model_id": "gpt-5.5", "benchmark_id": "hmmt_nov_2025", "score": 96.5, "reference_url": "https://z.ai/blog/glm-5.2", "reported_setting": { "mode": "thinking", "effort": "xhigh", "sampling": "pass@1", "prompt_style": "default", "tools": "none", "temperature": "1.0", "top_p": "0.95", "context": "default", "max_output_tokens": "163840", "judge": "GPT-5.5 medium for AIME/HMMT/IMO; benchmark-specified otherwise", "harness": "official reasoning evaluation" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "GLM-5.2 official release source: GPT-5.5 / HMMT Nov. 2025 = 96.5.", "candidates": [] }, { "model_id": "gpt-5.5", "benchmark_id": "hmmt_feb_2026", "score": 96.7, "reference_url": "https://z.ai/blog/glm-5.2", "reported_setting": { "mode": "thinking", "effort": "xhigh", "sampling": "pass@1", "prompt_style": "default", "tools": "none", "temperature": "1.0", "top_p": "0.95", "context": "default", "max_output_tokens": "163840", "judge": "GPT-5.5 medium for AIME/HMMT/IMO; benchmark-specified otherwise", "harness": "official reasoning evaluation" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "GLM-5.2 official release source: GPT-5.5 / HMMT Feb. 2026 = 96.7.", "candidates": [ { "score": 97.06, "reference_url": "https://huggingface.co/internlm/Intern-S2-Preview-397B", "source_type": "model_card", "reported_setting": { "mode": "source does not state", "effort": "source does not state", "tools": "none", "sampling": "source does not state", "judge": "benchmark-specific; exact version not stated", "harness": "OpenCompass or AgentCompass; exact adapter not stated", "prompt_style": "source does not state", "temperature": "source does not state", "context": "source does not state", "notes": "Shanghai AI Lab evaluated all model columns using OpenCompass, VLMEvalKit, and AgentCompass. Per-model mode/effort/sampling are not reported; only Intern-S2 maximum evaluation lengths are disclosed." }, "notes": "Intern-S2-Preview-397B official model card: HMMT-2026 = 97.06. This is a Shanghai AI Lab cross-model evaluation; the source does not disclose this model's exact mode, effort, or sampling." } ] }, { "model_id": "gpt-5.5", "benchmark_id": "nl2repo_bench", "score": 50.7, "reference_url": "https://z.ai/blog/glm-5.2", "reported_setting": { "mode": "thinking", "effort": "xhigh", "sampling": "pass@1", "prompt_style": "default", "tools": "agentic code execution", "temperature": "1.0", "top_p": "1.0", "context": "400000", "max_output_tokens": "48000", "judge": "rule-based + LLM anti-hack judgement", "harness": "official NL2Repo evaluation" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "GLM-5.2 official release source: GPT-5.5 / NL2Repo = 50.7.", "candidates": [ { "score": 45.1, "reference_url": "https://seed.bytedance.com/en/blog/seed2-1-officially-released-advancing-ai-productivity", "source_type": "official_blog", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "Trae or benchmark-specific coding-agent environment", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official", "prompt_style": "official default", "temperature": "unknown" }, "notes": "Release blog EN: Coding chart (https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/user-upload/4xfa4mqq4wlbt.png)." } ] }, { "model_id": "gpt-5.5", "benchmark_id": "swe_marathon_v1_0", "score": 12.0, "reference_url": "https://z.ai/blog/glm-5.2", "reported_setting": { "mode": "thinking", "effort": "xhigh", "sampling": "pass@1", "prompt_style": "default", "tools": "agentic code execution", "context": "1000000", "max_output_tokens": "128000", "judge": "official task verifiers", "harness": "Abundant AI SWE-Marathon v1.0; max effort", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "GLM-5.2 official release source: GPT-5.5 / SWE-Marathon = 12.0.", "candidates": [ { "score": 12.0, "reference_url": "https://media.x.ai/v1/website/4p5-5184fdf9.pdf", "source_type": "model_card", "reported_setting": { "mode": "thinking", "effort": "xhigh", "tools": "agentic code execution", "sampling": "pass@1", "judge": "multi-layer resolution verification", "harness": "SWE-Marathon v1.0 harness", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown" }, "notes": "card45 SWE-Marathon v1.0 Resolution rate (%); source effort=xhigh; benchmark protocol matches canonical=True. Unknown source fields remain unknown." } ] }, { "model_id": "gemini-3.1-pro", "benchmark_id": "hle_tools", "score": 51.4, "reference_url": "https://z.ai/blog/glm-5.2", "reported_setting": { "mode": "thinking", "effort": "high", "sampling": "pass@1", "prompt_style": "default", "tools": "search+code+web", "temperature": "1.0", "top_p": "0.95", "context": "300000", "max_output_tokens": "163840", "judge": "answer-key scoring", "harness": "official; no context-management strategy" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "GLM-5.2 official release source: Gemini 3.1 Pro / HLE (w/ Tools) = 51.4.", "candidates": [ { "score": 51.4, "reference_url": "https://deepmind.google/models/gemini/pro/", "source_type": "official_blog", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "source-reported search/code/web tools", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official Google Gemini evaluation", "prompt_style": "provider official evaluation prompt", "temperature": "source/provider setting", "context": "source/provider setting", "input_modalities": "text", "trials": "source-reported" }, "notes": "Exact provider-official score and reported setting." } ] }, { "model_id": "gemini-3.1-pro", "benchmark_id": "deep_swe_v1_1", "score": 10.0, "reference_url": "https://z.ai/blog/glm-5.2", "reported_setting": { "mode": "thinking", "effort": "high", "sampling": "pass@1", "prompt_style": "default", "tools": "agentic code execution", "temperature": "1.0", "top_p": "1.0", "context": "400000", "judge": "isolated-container verification", "harness": "official pier + mini-swe-agent; timeout=2h; 2 CPU/8GB; no internet" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "GLM-5.2 official release source: Gemini 3.1 Pro / DeepSWE = 10.0.", "candidates": [ { "score": 0.9, "reference_url": "https://huggingface.co/tencent/Hy3", "source_type": "model_card", "reported_setting": { "mode": "thinking", "effort": "high / highest available", "tools": "agentic code execution", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "mini-swe-agent; 2h/task; 2 CPU/8GB; no internet", "prompt_style": "default", "temperature": "0.9 for Hy3 family; otherwise source/provider setting", "context": "256000 for Hy3 family; otherwise source/provider setting", "notes": "Tencent own test" }, "notes": "Hy3 official model-card appendix: DeepSWE = 0.9*. Tencent own testing." }, { "score": 11.8, "reference_url": "https://openai.com/index/gpt-5-6/", "source_type": "official_blog", "reported_setting": { "effort": "high", "eval_variant": null, "code_mode": "false" }, "notes": "OpenAI GPT-5.6 release table (table 1): Gemini 3.1 Pro Preview; DeepSWE v1.1=11.8. Exact effort/variant resolved from the rendered chart point." }, { "score": 12.0, "reference_url": "https://ai.meta.com/static-resource/muse-spark-1.1-evaluation-report", "source_type": "tech_report", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "agentic coding", "sampling": "pass@1 unless source row states otherwise", "harness": "Meta MSL general-capability evaluation", "temperature": "1.0 for Muse; provider/source setting for comparisons", "context": "benchmark-specific", "configuration": "default" }, "notes": "Figure 44 · DeepSWE 1.1; metric=headline_metric. Exact printed value in the general-capability summary." }, { "score": 10.0, "reference_url": "https://seed.bytedance.com/en/blog/seed2-1-officially-released-advancing-ai-productivity", "source_type": "official_blog", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "Trae or benchmark-specific coding-agent environment", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official", "prompt_style": "official default", "temperature": "unknown" }, "notes": "Release blog EN: Coding chart (https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/user-upload/4xfa4mqq4wlbt.png)." } ] }, { "model_id": "gemini-3.1-pro", "benchmark_id": "program_bench", "score": 39.5, "reference_url": "https://z.ai/blog/glm-5.2", "reported_setting": { "mode": "thinking", "effort": "high", "sampling": "pass@1", "prompt_style": "default", "tools": "agentic code execution", "temperature": "1.0", "top_p": "1.0", "context": "400000", "max_output_tokens": "64000", "judge": "248,000+ fuzz-generated behavioral tests", "harness": "Claude Code 2.1.156; max_turns=2000; timeout=6h; 4 CPU/8GB; no internet" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "GLM-5.2 official release source: Gemini 3.1 Pro / ProgramBench = 39.5.", "candidates": [ { "score": 40.7, "reference_url": "https://seed.bytedance.com/en/blog/seed2-1-officially-released-advancing-ai-productivity", "source_type": "official_blog", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "Trae or benchmark-specific coding-agent environment", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official", "prompt_style": "official default", "temperature": "unknown" }, "notes": "Release blog EN: Coding chart (https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/user-upload/4xfa4mqq4wlbt.png)." } ] }, { "model_id": "gemini-3.1-pro", "benchmark_id": "terminal_bench_2_1", "score": 74.0, "reference_url": "https://z.ai/blog/glm-5.2", "reported_setting": { "mode": "thinking", "effort": "high", "sampling": "pass@1", "prompt_style": "default", "tools": "terminal agent", "temperature": "1.0", "top_p": "1.0", "context": "256000", "max_output_tokens": "48000", "judge": "official task verifier", "harness": "Terminus-2; parser=json; timeout=4h; max_episodes=500; 4 CPU/8GB" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "GLM-5.2 official release source: Gemini 3.1 Pro / Terminal Bench 2.1 (Terminus-2) = 74.0.", "candidates": [ { "score": 70.7, "reference_url": "https://z.ai/blog/glm-5.2", "source_type": "official_blog", "reported_setting": { "mode": "thinking", "effort": "high", "sampling": "pass@1", "prompt_style": "default", "tools": "terminal agent", "temperature": "1.0 for Claude Code; unknown otherwise", "top_p": "0.95 for Claude Code; unknown otherwise", "context": "source-reported", "max_output_tokens": "131072 for Claude Code; source-reported otherwise", "judge": "official task verifier", "harness": "Gemini CLI; Claude Code values averaged over 5 runs" }, "notes": "Z.ai Gemini CLI alternative." }, { "score": 73.8, "reference_url": "https://huggingface.co/internlm/Intern-S2-Preview-397B", "source_type": "model_card", "reported_setting": { "mode": "source does not state", "effort": "source does not state", "tools": "terminal/code execution", "sampling": "source does not state", "judge": "benchmark-specific; exact version not stated", "harness": "Terminus 2", "prompt_style": "source does not state", "temperature": "source does not state", "context": "source does not state", "notes": "Shanghai AI Lab evaluated all model columns using OpenCompass, VLMEvalKit, and AgentCompass. Per-model mode/effort/sampling are not reported; only Intern-S2 maximum evaluation lengths are disclosed." }, "notes": "Intern-S2-Preview-397B official model card: TerminalBench 2.1 = 73.8. This is a Shanghai AI Lab cross-model evaluation; the source does not disclose this model's exact mode, effort, or sampling." }, { "score": 70.3, "reference_url": "https://ai.meta.com/static-resource/muse-spark-1.1-evaluation-report", "source_type": "tech_report", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "terminal", "sampling": "pass@1 unless source row states otherwise", "harness": "Meta MSL general-capability evaluation", "temperature": "1.0 for Muse; provider/source setting for comparisons", "context": "benchmark-specific", "configuration": "default" }, "notes": "Figure 44 · Terminal-Bench 2.1; metric=headline_metric. Exact printed value in the general-capability summary." }, { "score": 70.7, "reference_url": "https://seed.bytedance.com/en/blog/seed2-1-officially-released-advancing-ai-productivity", "source_type": "official_blog", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "Trae or benchmark-specific coding-agent environment", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official", "prompt_style": "official default", "temperature": "unknown" }, "notes": "Release blog EN: Coding chart (https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/user-upload/4xfa4mqq4wlbt.png)." } ] }, { "model_id": "gemini-3.1-pro", "benchmark_id": "frontier_swe", "score": 39.6, "reference_url": "https://z.ai/blog/glm-5.2", "reported_setting": { "mode": "thinking", "effort": "high", "sampling": "pass@1", "prompt_style": "default", "tools": "agentic code execution", "context": "1000000", "max_output_tokens": "128000", "judge": "FrontierSWE dominance", "harness": "Proximal evaluation; max effort; as of 2026-06-16", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "GLM-5.2 official release source: Gemini 3.1 Pro / FrontierSWE (Dominance) = 39.6.", "candidates": [] }, { "model_id": "gemini-3.1-pro", "benchmark_id": "posttrain_bench", "score": 21.6, "reference_url": "https://z.ai/blog/glm-5.2", "reported_setting": { "mode": "thinking", "effort": "high", "sampling": "pass@1", "prompt_style": "default", "tools": "agentic post-training", "context": "1000000", "max_output_tokens": "128000", "judge": "official weighted benchmark score", "harness": "PostTrainBench evaluation; max effort", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "GLM-5.2 official release source: Gemini 3.1 Pro / PostTrainBench = 21.6.", "candidates": [] }, { "model_id": "gemini-3.1-pro", "benchmark_id": "swe_marathon_v1_0", "score": 4.0, "reference_url": "https://z.ai/blog/glm-5.2", "reported_setting": { "mode": "thinking", "effort": "high", "sampling": "pass@1", "prompt_style": "default", "tools": "agentic code execution", "context": "1000000", "max_output_tokens": "128000", "judge": "official task verifiers", "harness": "Abundant AI SWE-Marathon v1.0; max effort", "temperature": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "GLM-5.2 official release source: Gemini 3.1 Pro / SWE-Marathon = 4.0.", "candidates": [] }, { "model_id": "claude-opus-4.7", "benchmark_id": "frontier_swe", "score": 63.0, "reference_url": "https://z.ai/blog/glm-5.2", "reported_setting": { "mode": "thinking", "effort": "max (source; model default differs)", "sampling": "pass@1", "prompt_style": "default", "tools": "agentic code execution", "context": "1000000", "max_output_tokens": "128000", "judge": "FrontierSWE dominance", "harness": "Proximal evaluation; max effort; as of 2026-06-16", "temperature": "default" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "GLM-5.2 official release source: Claude Opus 4.7 / FrontierSWE (Dominance) = 63.0.", "candidates": [] }, { "model_id": "claude-opus-4.7", "benchmark_id": "posttrain_bench", "score": 28.6, "reference_url": "https://z.ai/blog/glm-5.2", "reported_setting": { "mode": "thinking", "effort": "max (source; model default differs)", "sampling": "pass@1", "prompt_style": "default", "tools": "agentic post-training", "context": "1000000", "max_output_tokens": "128000", "judge": "official weighted benchmark score", "harness": "PostTrainBench evaluation; max effort", "temperature": "default" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "GLM-5.2 official release source: Claude Opus 4.7 / PostTrainBench = 28.6.", "candidates": [ { "score": 27.4, "reference_url": "https://seed.bytedance.com/en/blog/seed2-1-officially-released-advancing-ai-productivity", "source_type": "official_blog", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "research-oriented agent environment", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official", "prompt_style": "official default", "temperature": "unknown" }, "notes": "Release blog EN: Frontier-research chart (https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/user-upload/4xfa4mqq52cdr.png). Source dashes are not zero." } ] }, { "model_id": "claude-opus-4.7", "benchmark_id": "swe_marathon_v1_0", "score": 16.0, "reference_url": "https://z.ai/blog/glm-5.2", "reported_setting": { "mode": "thinking", "effort": "max (source; model default differs)", "sampling": "pass@1", "prompt_style": "default", "tools": "agentic code execution", "context": "1000000", "max_output_tokens": "128000", "judge": "official task verifiers", "harness": "Abundant AI SWE-Marathon v1.0; max effort", "temperature": "default" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "GLM-5.2 official release source: Claude Opus 4.7 / SWE-Marathon = 16.0.", "candidates": [ { "score": 16.0, "reference_url": "https://x.ai/news/grok-4-5", "source_type": "official_blog", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "agentic code execution", "sampling": "pass@1", "judge": "multi-layer resolution verification", "harness": "SWE-Marathon v1.0 harness", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown" }, "notes": "release45 SWE-Marathon v1.0 Resolution rate (%); source effort=max; benchmark protocol matches canonical=True. Unknown source fields remain unknown." } ] }, { "model_id": "hy3-preview", "benchmark_id": "swe_bench_multilingual", "score": 68.3, "reference_url": "https://huggingface.co/tencent/Hy3", "reported_setting": { "mode": "thinking", "effort": "high / highest available", "tools": "agentic code execution", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "SWE-agent scaffold", "prompt_style": "default", "temperature": "0.9 for Hy3 family; otherwise source/provider setting", "context": "256000 for Hy3 family; otherwise source/provider setting", "notes": "Comparator value not marked as Tencent own test" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Hy3 official model-card appendix: SWE-bench Multilingual = 68.3. Comparator-reported value." }, { "model_id": "hy3", "benchmark_id": "swe_bench_multilingual", "score": 75.8, "reference_url": "https://huggingface.co/tencent/Hy3", "reported_setting": { "mode": "thinking", "effort": "high / highest available", "tools": "agentic code execution", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "SWE-agent scaffold", "prompt_style": "default", "temperature": "0.9 for Hy3 family; otherwise source/provider setting", "context": "256000 for Hy3 family; otherwise source/provider setting", "notes": "Comparator value not marked as Tencent own test" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Hy3 official model-card appendix: SWE-bench Multilingual = 75.8. Comparator-reported value." }, { "model_id": "glm-5.2", "benchmark_id": "swe_bench_multilingual", "score": 83.0, "reference_url": "https://huggingface.co/tencent/Hy3", "reported_setting": { "mode": "thinking", "effort": "high / highest available", "tools": "agentic code execution", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "SWE-agent scaffold", "prompt_style": "default", "temperature": "0.9 for Hy3 family; otherwise source/provider setting", "context": "256000 for Hy3 family; otherwise source/provider setting", "notes": "Tencent own test" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Hy3 official model-card appendix: SWE-bench Multilingual = 83.0*. Tencent own testing.", "candidates": [ { "score": 82.0, "reference_url": "https://huggingface.co/internlm/Intern-S2-Preview-397B", "source_type": "model_card", "reported_setting": { "mode": "source does not state", "effort": "source does not state", "tools": "agentic code execution", "sampling": "source does not state", "judge": "benchmark-specific; exact version not stated", "harness": "Mini-SWE-Agent", "prompt_style": "source does not state", "temperature": "source does not state", "context": "source does not state", "notes": "Shanghai AI Lab evaluated all model columns using OpenCompass, VLMEvalKit, and AgentCompass. Per-model mode/effort/sampling are not reported; only Intern-S2 maximum evaluation lengths are disclosed." }, "notes": "Intern-S2-Preview-397B official model card: SWE-Bench-Multilingual = 82.0. This is a Shanghai AI Lab cross-model evaluation; the source does not disclose this model's exact mode, effort, or sampling." } ] }, { "model_id": "doubao-seed-2.1-pro", "benchmark_id": "swe_bench_multilingual", "score": 75.1, "reference_url": "https://huggingface.co/tencent/Hy3", "reported_setting": { "mode": "thinking", "effort": "high / highest available", "tools": "agentic code execution", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "SWE-agent scaffold", "prompt_style": "default", "temperature": "0.9 for Hy3 family; otherwise source/provider setting", "context": "256000 for Hy3 family; otherwise source/provider setting", "notes": "Tencent own test" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Hy3 official model-card appendix: SWE-bench Multilingual = 75.1*. Tencent own testing." }, { "model_id": "qwen3.7-max", "benchmark_id": "swe_bench_multilingual", "score": 78.3, "reference_url": "https://huggingface.co/tencent/Hy3", "reported_setting": { "mode": "thinking", "effort": "high / highest available", "tools": "agentic code execution", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "SWE-agent scaffold", "prompt_style": "default", "temperature": "0.9 for Hy3 family; otherwise source/provider setting", "context": "256000 for Hy3 family; otherwise source/provider setting", "notes": "Comparator value not marked as Tencent own test" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Hy3 official model-card appendix: SWE-bench Multilingual = 78.3. Comparator-reported value." }, { "model_id": "claude-opus-4.8", "benchmark_id": "swe_bench_multilingual", "score": 84.8, "reference_url": "https://longcat.chat/blog/longcat-2.0/", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "agentic code execution", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "provider official report; details not reproduced by LongCat", "prompt_style": "default", "temperature": "provider official setting", "context": "default", "notes": "Asterisk: LongCat cites this value from the model provider's official report rather than measuring it in-house." }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "LongCat-2.0 official tech blog: SWE-bench Multilingual = 84.8*. Asterisked value is cited by LongCat from the model provider's official report.", "candidates": [ { "score": 84.4, "reference_url": "https://huggingface.co/tencent/Hy3", "source_type": "model_card", "reported_setting": { "mode": "thinking", "effort": "high / highest available", "tools": "agentic code execution", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "SWE-agent scaffold", "prompt_style": "default", "temperature": "0.9 for Hy3 family; otherwise source/provider setting", "context": "256000 for Hy3 family; otherwise source/provider setting", "notes": "Comparator value not marked as Tencent own test" }, "notes": "Hy3 official model-card appendix: SWE-bench Multilingual = 84.4. Comparator-reported value. [Displaced by approved LongCat-2.0 official-source audit.]" }, { "score": 77.0, "reference_url": "https://huggingface.co/internlm/Intern-S2-Preview-397B", "source_type": "model_card", "reported_setting": { "mode": "source does not state", "effort": "source does not state", "tools": "agentic code execution", "sampling": "source does not state", "judge": "benchmark-specific; exact version not stated", "harness": "Mini-SWE-Agent", "prompt_style": "source does not state", "temperature": "source does not state", "context": "source does not state", "notes": "Shanghai AI Lab evaluated all model columns using OpenCompass, VLMEvalKit, and AgentCompass. Per-model mode/effort/sampling are not reported; only Intern-S2 maximum evaluation lengths are disclosed." }, "notes": "Intern-S2-Preview-397B official model card: SWE-Bench-Multilingual = 77.0. This is a Shanghai AI Lab cross-model evaluation; the source does not disclose this model's exact mode, effort, or sampling." }, { "score": 84.4, "reference_url": "https://www.anthropic.com/claude-opus-5-system-card", "source_type": "model_card", "reported_setting": { "effort": "max", "harness": "official or cited leaderboard" }, "notes": "Table 8.1.A / SWE-bench Multilingual: Unless noted, Anthropic values use adaptive thinking at max effort, default sampling, mean over five trials. Table 8.1.A card summary." }, { "score": 84.4, "reference_url": "https://media.x.ai/v1/website/4p5-5184fdf9.pdf", "source_type": "model_card", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "agentic repository shell/editor", "sampling": "pass@1", "judge": "language-appropriate isolated tests", "harness": "fixed SWE-bench Multilingual agent scaffold", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown" }, "notes": "card45 SWE-Bench Multilingual Resolve rate (%); source effort=max; benchmark protocol matches canonical=True. Unknown source fields remain unknown." } ] }, { "model_id": "gpt-5.5", "benchmark_id": "swe_bench_multilingual", "score": 77.8, "reference_url": "https://media.x.ai/v1/website/4p5-5184fdf9.pdf", "reported_setting": { "mode": "thinking", "effort": "xhigh", "tools": "agentic repository shell/editor", "sampling": "pass@1", "judge": "language-appropriate isolated tests", "harness": "fixed SWE-bench Multilingual agent scaffold", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "card45 SWE-Bench Multilingual Resolve rate (%); source effort=xhigh; benchmark protocol matches canonical=True. Unknown source fields remain unknown.", "candidates": [ { "score": 73.33, "reference_url": "https://huggingface.co/internlm/Intern-S2-Preview-397B", "source_type": "model_card", "reported_setting": { "mode": "source does not state", "effort": "source does not state", "tools": "agentic code execution", "sampling": "source does not state", "judge": "benchmark-specific; exact version not stated", "harness": "Mini-SWE-Agent", "prompt_style": "source does not state", "temperature": "source does not state", "context": "source does not state", "notes": "Shanghai AI Lab evaluated all model columns using OpenCompass, VLMEvalKit, and AgentCompass. Per-model mode/effort/sampling are not reported; only Intern-S2 maximum evaluation lengths are disclosed." }, "notes": "Intern-S2-Preview-397B official model card: SWE-Bench-Multilingual = 73.33. This is a Shanghai AI Lab cross-model evaluation; the source does not disclose this model's exact mode, effort, or sampling." }, { "score": 77.8, "reference_url": "https://cursor.com/blog/composer-2-5", "source_type": "official_blog", "reported_setting": { "effort": "source does not state", "tools": "agentic", "sampling": "pass@1", "harness": "Cursor" }, "notes": "Composer 2.5 release benchmark image / SWE-Bench Multilingual / GPT-5.5: Cursor release benchmark image. Opus 4.7 and GPT-5.5 public evaluation values are self-reported." }, { "score": 77.3, "reference_url": "https://huggingface.co/tencent/Hy3", "source_type": "model_card", "reported_setting": { "mode": "thinking", "effort": "high / highest available", "tools": "agentic code execution", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "SWE-agent scaffold", "prompt_style": "default", "temperature": "0.9 for Hy3 family; otherwise source/provider setting", "context": "256000 for Hy3 family; otherwise source/provider setting", "notes": "Tencent own test" }, "notes": "Hy3 official model-card appendix: SWE-bench Multilingual = 77.3*. Tencent own testing." } ] }, { "model_id": "hy3-preview", "benchmark_id": "swe_bench_verified", "score": 74.4, "reference_url": "https://huggingface.co/tencent/Hy3", "reported_setting": { "mode": "thinking", "effort": "high / highest available", "tools": "agentic code execution", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "SWE-agent scaffold", "prompt_style": "default", "temperature": "0.9 for Hy3 family; otherwise source/provider setting", "context": "256000 for Hy3 family; otherwise source/provider setting", "notes": "Comparator value not marked as Tencent own test" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Hy3 official model-card appendix: SWE-bench Verified = 74.4. Comparator-reported value." }, { "model_id": "hy3", "benchmark_id": "swe_bench_verified", "score": 78.0, "reference_url": "https://huggingface.co/tencent/Hy3", "reported_setting": { "mode": "thinking", "effort": "high / highest available", "tools": "agentic code execution", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "SWE-agent scaffold", "prompt_style": "default", "temperature": "0.9 for Hy3 family; otherwise source/provider setting", "context": "256000 for Hy3 family; otherwise source/provider setting", "notes": "Comparator value not marked as Tencent own test" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Hy3 official model-card appendix: SWE-bench Verified = 78.0. Comparator-reported value." }, { "model_id": "glm-5.1", "benchmark_id": "swe_bench_verified", "score": 75.0, "reference_url": "https://huggingface.co/tencent/Hy3", "reported_setting": { "mode": "thinking", "effort": "high / highest available", "tools": "agentic code execution", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "SWE-agent scaffold", "prompt_style": "default", "temperature": "0.9 for Hy3 family; otherwise source/provider setting", "context": "256000 for Hy3 family; otherwise source/provider setting", "notes": "Tencent own test" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Hy3 official model-card appendix: SWE-bench Verified = 75.0*. Tencent own testing.", "candidates": [ { "score": 76.2, "reference_url": "https://research.nvidia.com/labs/nemotron/files/NVIDIA-Nemotron-3-Ultra-Technical-Report.pdf", "source_type": "tech_report", "reported_setting": { "version_metric": "Verified / % resolved", "tools": "agentic repository editing", "harness": "Harbor / OpenHands / AWS ECS", "sampling": "3-run average" }, "notes": "NVIDIA Nemotron 3 Ultra technical report, Final Table 10: GLM-5.1: swe_bench_verified=76.2." }, { "score": 71.3, "reference_url": "https://research.nvidia.com/labs/nemotron/files/NVIDIA-Nemotron-3-Ultra-Technical-Report.pdf", "source_type": "tech_report", "reported_setting": { "version_metric": "Mini SWE Agent scaffold / %", "tools": "agent-specific software-engineering environment", "harness": "Mini SWE Agent", "sampling": "single agent result", "variant": "Mini SWE Agent", "agent_scaffold": "Mini SWE Agent" }, "notes": "NVIDIA Nemotron 3 Ultra technical report, Figure 17: GLM-5.1: swe_bench_verified=71.3, variant=Mini SWE Agent." }, { "score": 74.8, "reference_url": "https://research.nvidia.com/labs/nemotron/files/NVIDIA-Nemotron-3-Ultra-Technical-Report.pdf", "source_type": "tech_report", "reported_setting": { "version_metric": "OpenCode scaffold / %", "tools": "agent-specific software-engineering environment", "harness": "OpenCode", "sampling": "single agent result", "variant": "OpenCode", "agent_scaffold": "OpenCode" }, "notes": "NVIDIA Nemotron 3 Ultra technical report, Figure 17: GLM-5.1: swe_bench_verified=74.8, variant=OpenCode." }, { "score": 74.4, "reference_url": "https://research.nvidia.com/labs/nemotron/files/NVIDIA-Nemotron-3-Ultra-Technical-Report.pdf", "source_type": "tech_report", "reported_setting": { "version_metric": "Pi scaffold / %", "tools": "agent-specific software-engineering environment", "harness": "Pi", "sampling": "single agent result", "variant": "Pi", "agent_scaffold": "Pi" }, "notes": "NVIDIA Nemotron 3 Ultra technical report, Figure 17: GLM-5.1: swe_bench_verified=74.4, variant=Pi." }, { "score": 74.5, "reference_url": "https://research.nvidia.com/labs/nemotron/files/NVIDIA-Nemotron-3-Ultra-Technical-Report.pdf", "source_type": "tech_report", "reported_setting": { "version_metric": "Hermes scaffold / %", "tools": "agent-specific software-engineering environment", "harness": "Hermes", "sampling": "single agent result", "variant": "Hermes", "agent_scaffold": "Hermes" }, "notes": "NVIDIA Nemotron 3 Ultra technical report, Figure 17: GLM-5.1: swe_bench_verified=74.5, variant=Hermes." }, { "score": 73.7, "reference_url": "https://research.nvidia.com/labs/nemotron/files/NVIDIA-Nemotron-3-Ultra-Technical-Report.pdf", "source_type": "tech_report", "reported_setting": { "version_metric": "OpenHands scaffold / %", "tools": "agent-specific software-engineering environment", "harness": "OpenHands", "sampling": "single agent result", "variant": "OpenHands", "agent_scaffold": "OpenHands" }, "notes": "NVIDIA Nemotron 3 Ultra technical report, Figure 17: GLM-5.1: swe_bench_verified=73.7, variant=OpenHands." }, { "score": 73.6, "reference_url": "https://research.nvidia.com/labs/nemotron/files/NVIDIA-Nemotron-3-Ultra-Technical-Report.pdf", "source_type": "tech_report", "reported_setting": { "version_metric": "Codex scaffold / %", "tools": "agent-specific software-engineering environment", "harness": "Codex", "sampling": "single agent result", "variant": "Codex", "agent_scaffold": "Codex" }, "notes": "NVIDIA Nemotron 3 Ultra technical report, Figure 17: GLM-5.1: swe_bench_verified=73.6, variant=Codex." }, { "score": 73.8, "reference_url": "https://research.nvidia.com/labs/nemotron/files/NVIDIA-Nemotron-3-Ultra-Technical-Report.pdf", "source_type": "tech_report", "reported_setting": { "version_metric": "Average scaffold / %", "tools": "agent-specific software-engineering environment", "harness": "Average", "sampling": "single agent result", "variant": "Average", "agent_scaffold": "Average" }, "notes": "NVIDIA Nemotron 3 Ultra technical report, Figure 17: GLM-5.1: swe_bench_verified=73.8, variant=Average." }, { "score": 80.2, "reference_url": "https://huggingface.co/mistralai/Mistral-Medium-3.5-128B/resolve/22b2b868a15677cfa6061277ed2f653d1349a9ab/images/image4.png?download=true", "source_type": "official_model_card", "reported_setting": { "mode": "thinking", "effort": "default (maximum reasoning setting)", "tools": "agentic coding scaffold and SWE-bench verifier", "sampling": "source does not state trial count", "judge": "benchmark-specified", "harness": "Mistral self-reported SWE-bench Verified evaluation", "prompt_style": "default", "temperature": "default", "context": "default" }, "notes": "Displayed exactly: '80.2'. Research observation: medium-image4.png:1:5:score:reported. Chart footnotes: *All model evaluations were run with maximum reasoning settings. | **Sonnet 4.6 encountered higher rates of reasoning truncation due to external API restrictions than other models, which affected its performance. | *** Self-reported | **** Self-reported, Mistral is using context management and a discard-all strategy at 100k tokens. | τ³ scores as reported by Sierra for Claude Sonnet 4.5 and Qwen3.5. Others with user simulator: gpt-5.2 with reasoning_effort: low. 4 trials. Banking domain evaluated with terminal- or embedding-based agentic search retrieval, only highest score is reported." } ] }, { "model_id": "glm-5.2", "benchmark_id": "swe_bench_verified", "score": 84.2, "reference_url": "https://huggingface.co/tencent/Hy3", "reported_setting": { "mode": "thinking", "effort": "high / highest available", "tools": "agentic code execution", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "SWE-agent scaffold", "prompt_style": "default", "temperature": "0.9 for Hy3 family; otherwise source/provider setting", "context": "256000 for Hy3 family; otherwise source/provider setting", "notes": "Tencent own test" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Hy3 official model-card appendix: SWE-bench Verified = 84.2*. Tencent own testing." }, { "model_id": "doubao-seed-2.1-pro", "benchmark_id": "swe_bench_verified", "score": 76.0, "reference_url": "https://huggingface.co/tencent/Hy3", "reported_setting": { "mode": "thinking", "effort": "high / highest available", "tools": "agentic code execution", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "SWE-agent scaffold", "prompt_style": "default", "temperature": "0.9 for Hy3 family; otherwise source/provider setting", "context": "256000 for Hy3 family; otherwise source/provider setting", "notes": "Tencent own test" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Hy3 official model-card appendix: SWE-bench Verified = 76.0*. Tencent own testing." }, { "model_id": "qwen3.7-max", "benchmark_id": "swe_bench_verified", "score": 80.4, "reference_url": "https://huggingface.co/tencent/Hy3", "reported_setting": { "mode": "thinking", "effort": "high / highest available", "tools": "agentic code execution", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "SWE-agent scaffold", "prompt_style": "default", "temperature": "0.9 for Hy3 family; otherwise source/provider setting", "context": "256000 for Hy3 family; otherwise source/provider setting", "notes": "Comparator value not marked as Tencent own test" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Hy3 official model-card appendix: SWE-bench Verified = 80.4. Comparator-reported value." }, { "model_id": "claude-opus-4.8", "benchmark_id": "swe_bench_verified", "score": 88.6, "reference_url": "https://huggingface.co/tencent/Hy3", "reported_setting": { "mode": "thinking", "effort": "high / highest available", "tools": "agentic code execution", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "SWE-agent scaffold", "prompt_style": "default", "temperature": "0.9 for Hy3 family; otherwise source/provider setting", "context": "256000 for Hy3 family; otherwise source/provider setting", "notes": "Comparator value not marked as Tencent own test" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Hy3 official model-card appendix: SWE-bench Verified = 88.6. Comparator-reported value." }, { "model_id": "gpt-5.5", "benchmark_id": "swe_bench_verified", "score": 84.4, "reference_url": "https://huggingface.co/tencent/Hy3", "reported_setting": { "mode": "thinking", "effort": "high / highest available", "tools": "agentic code execution", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "SWE-agent scaffold", "prompt_style": "default", "temperature": "0.9 for Hy3 family; otherwise source/provider setting", "context": "256000 for Hy3 family; otherwise source/provider setting", "notes": "Tencent own test" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Hy3 official model-card appendix: SWE-bench Verified = 84.4*. Tencent own testing." }, { "model_id": "hy3-preview", "benchmark_id": "swe_bench_pro", "score": 46.0, "reference_url": "https://huggingface.co/tencent/Hy3", "reported_setting": { "mode": "thinking", "effort": "high / highest available", "tools": "agentic code execution", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "SWE-agent scaffold", "prompt_style": "default", "temperature": "0.9 for Hy3 family; otherwise source/provider setting", "context": "256000 for Hy3 family; otherwise source/provider setting", "notes": "Comparator value not marked as Tencent own test" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Hy3 official model-card appendix: SWE-bench Pro = 46.0. Comparator-reported value." }, { "model_id": "hy3", "benchmark_id": "swe_bench_pro", "score": 57.9, "reference_url": "https://huggingface.co/tencent/Hy3", "reported_setting": { "mode": "thinking", "effort": "high / highest available", "tools": "agentic code execution", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "SWE-agent scaffold", "prompt_style": "default", "temperature": "0.9 for Hy3 family; otherwise source/provider setting", "context": "256000 for Hy3 family; otherwise source/provider setting", "notes": "Comparator value not marked as Tencent own test" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Hy3 official model-card appendix: SWE-bench Pro = 57.9. Comparator-reported value." }, { "model_id": "doubao-seed-2.1-pro", "benchmark_id": "swe_bench_pro", "score": 57.5, "reference_url": "https://seed.bytedance.com/en/blog/seed2-1-officially-released-advancing-ai-productivity", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "Trae or benchmark-specific coding-agent environment", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official", "prompt_style": "official default", "temperature": "unknown" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "Release blog EN: Coding chart (https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/user-upload/4xfa4mqq4wlbt.png).", "candidates": [ { "score": 51.6, "reference_url": "https://huggingface.co/tencent/Hy3", "source_type": "model_card", "reported_setting": { "mode": "thinking", "effort": "high / highest available", "tools": "agentic code execution", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "SWE-agent scaffold", "prompt_style": "default", "temperature": "0.9 for Hy3 family; otherwise source/provider setting", "context": "256000 for Hy3 family; otherwise source/provider setting", "notes": "Tencent own test" }, "notes": "Hy3 official model-card appendix: SWE-bench Pro = 51.6*. Tencent own testing." }, { "score": 57.5, "reference_url": "https://huggingface.co/tencent/Hy3", "source_type": "model_card", "reported_setting": { "mode": "thinking", "effort": "high / highest available", "tools": "agentic code execution", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "SWE-agent scaffold", "prompt_style": "default", "temperature": "0.9 for Hy3 family; otherwise source/provider setting", "context": "256000 for Hy3 family; otherwise source/provider setting", "notes": "Comparator value not marked as Tencent own test" }, "notes": "Hy3 official model-card appendix: SWE-bench Pro = 57.5. Comparator-reported value." } ] }, { "model_id": "hy3-preview", "benchmark_id": "terminal_bench_2_1", "score": 58.0, "reference_url": "https://huggingface.co/tencent/Hy3", "reported_setting": { "mode": "thinking", "effort": "high / highest available", "tools": "terminal/code execution", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "Terminus-2; parser=xml; timeout=4h; CPU=16 cores; memory=32GB; max episodes=500", "prompt_style": "default", "temperature": "0.9 for Hy3 family; otherwise source/provider setting", "context": "256000 for Hy3 family; otherwise source/provider setting", "notes": "Comparator value not marked as Tencent own test" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Hy3 official model-card appendix: Terminal-Bench 2.1 = 58.0. Comparator-reported value." }, { "model_id": "hy3", "benchmark_id": "terminal_bench_2_1", "score": 71.7, "reference_url": "https://huggingface.co/tencent/Hy3", "reported_setting": { "mode": "thinking", "effort": "high / highest available", "tools": "terminal/code execution", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "Terminus-2; parser=xml; timeout=4h; CPU=16 cores; memory=32GB; max episodes=500", "prompt_style": "default", "temperature": "0.9 for Hy3 family; otherwise source/provider setting", "context": "256000 for Hy3 family; otherwise source/provider setting", "notes": "Comparator value not marked as Tencent own test" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Hy3 official model-card appendix: Terminal-Bench 2.1 = 71.7. Comparator-reported value." }, { "model_id": "doubao-seed-2.1-pro", "benchmark_id": "terminal_bench_2_1", "score": 71.0, "reference_url": "https://seed.bytedance.com/en/blog/seed2-1-officially-released-advancing-ai-productivity", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "Trae or benchmark-specific coding-agent environment", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official", "prompt_style": "official default", "temperature": "unknown" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "Release blog EN: Coding chart (https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/user-upload/4xfa4mqq4wlbt.png).", "candidates": [ { "score": 71.0, "reference_url": "https://huggingface.co/tencent/Hy3", "source_type": "model_card", "reported_setting": { "mode": "thinking", "effort": "high / highest available", "tools": "terminal/code execution", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "Terminus-2; parser=xml; timeout=4h; CPU=16 cores; memory=32GB; max episodes=500", "prompt_style": "default", "temperature": "0.9 for Hy3 family; otherwise source/provider setting", "context": "256000 for Hy3 family; otherwise source/provider setting", "notes": "Comparator value not marked as Tencent own test" }, "notes": "Hy3 official model-card appendix: Terminal-Bench 2.1 = 71.0. Comparator-reported value." } ] }, { "model_id": "hy3-preview", "benchmark_id": "nl2repo_bench", "score": 35.3, "reference_url": "https://huggingface.co/tencent/Hy3", "reported_setting": { "mode": "thinking", "effort": "high / highest available", "tools": "Claude Code agentic code execution", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "Claude Code; 250-turn budget; timeout=12000s; 4 CPU/32GB; anti-hacking/tool-call monitoring", "prompt_style": "default", "temperature": "0.9 for Hy3 family; otherwise source/provider setting", "context": "256000 for Hy3 family; otherwise source/provider setting", "notes": "Comparator value not marked as Tencent own test" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Hy3 official model-card appendix: NL2Repo = 35.3. Comparator-reported value." }, { "model_id": "hy3", "benchmark_id": "nl2repo_bench", "score": 45.6, "reference_url": "https://huggingface.co/tencent/Hy3", "reported_setting": { "mode": "thinking", "effort": "high / highest available", "tools": "Claude Code agentic code execution", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "Claude Code; 250-turn budget; timeout=12000s; 4 CPU/32GB; anti-hacking/tool-call monitoring", "prompt_style": "default", "temperature": "0.9 for Hy3 family; otherwise source/provider setting", "context": "256000 for Hy3 family; otherwise source/provider setting", "notes": "Comparator value not marked as Tencent own test" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Hy3 official model-card appendix: NL2Repo = 45.6. Comparator-reported value." }, { "model_id": "doubao-seed-2.1-pro", "benchmark_id": "nl2repo_bench", "score": 47.0, "reference_url": "https://seed.bytedance.com/en/blog/seed2-1-officially-released-advancing-ai-productivity", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "Trae or benchmark-specific coding-agent environment", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official", "prompt_style": "official default", "temperature": "unknown" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "Release blog EN: Coding chart (https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/user-upload/4xfa4mqq4wlbt.png).", "candidates": [ { "score": 47.0, "reference_url": "https://huggingface.co/tencent/Hy3", "source_type": "model_card", "reported_setting": { "mode": "thinking", "effort": "high / highest available", "tools": "Claude Code agentic code execution", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "Claude Code; 250-turn budget; timeout=12000s; 4 CPU/32GB; anti-hacking/tool-call monitoring", "prompt_style": "default", "temperature": "0.9 for Hy3 family; otherwise source/provider setting", "context": "256000 for Hy3 family; otherwise source/provider setting", "notes": "Comparator value not marked as Tencent own test" }, "notes": "Hy3 official model-card appendix: NL2Repo = 47.0. Comparator-reported value." } ] }, { "model_id": "hy3-preview", "benchmark_id": "deep_swe_v1_1", "score": 0.9, "reference_url": "https://huggingface.co/tencent/Hy3", "reported_setting": { "mode": "thinking", "effort": "high / highest available", "tools": "agentic code execution", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "mini-swe-agent; 2h/task; 2 CPU/8GB; no internet", "prompt_style": "default", "temperature": "0.9 for Hy3 family; otherwise source/provider setting", "context": "256000 for Hy3 family; otherwise source/provider setting", "notes": "Comparator value not marked as Tencent own test" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Hy3 official model-card appendix: DeepSWE = 0.9. Comparator-reported value." }, { "model_id": "hy3", "benchmark_id": "deep_swe_v1_1", "score": 28.0, "reference_url": "https://huggingface.co/tencent/Hy3", "reported_setting": { "mode": "thinking", "effort": "high / highest available", "tools": "agentic code execution", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "mini-swe-agent; 2h/task; 2 CPU/8GB; no internet", "prompt_style": "default", "temperature": "0.9 for Hy3 family; otherwise source/provider setting", "context": "256000 for Hy3 family; otherwise source/provider setting", "notes": "Comparator value not marked as Tencent own test" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Hy3 official model-card appendix: DeepSWE = 28.0. Comparator-reported value." }, { "model_id": "doubao-seed-2.1-pro", "benchmark_id": "deep_swe_v1_1", "score": 32.7, "reference_url": "https://seed.bytedance.com/en/blog/seed2-1-officially-released-advancing-ai-productivity", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "Trae or benchmark-specific coding-agent environment", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official", "prompt_style": "official default", "temperature": "unknown" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "Release blog EN: Coding chart (https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/user-upload/4xfa4mqq4wlbt.png).", "candidates": [ { "score": 15.0, "reference_url": "https://huggingface.co/tencent/Hy3", "source_type": "model_card", "reported_setting": { "mode": "thinking", "effort": "high / highest available", "tools": "agentic code execution", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "mini-swe-agent; 2h/task; 2 CPU/8GB; no internet", "prompt_style": "default", "temperature": "0.9 for Hy3 family; otherwise source/provider setting", "context": "256000 for Hy3 family; otherwise source/provider setting", "notes": "Tencent own test" }, "notes": "Hy3 official model-card appendix: DeepSWE = 15.0*. Tencent own testing." }, { "score": 32.7, "reference_url": "https://huggingface.co/tencent/Hy3", "source_type": "model_card", "reported_setting": { "mode": "thinking", "effort": "high / highest available", "tools": "agentic code execution", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "mini-swe-agent; 2h/task; 2 CPU/8GB; no internet", "prompt_style": "default", "temperature": "0.9 for Hy3 family; otherwise source/provider setting", "context": "256000 for Hy3 family; otherwise source/provider setting", "notes": "Comparator value not marked as Tencent own test" }, "notes": "Hy3 official model-card appendix: DeepSWE = 32.7. Comparator-reported value." } ] }, { "model_id": "hy3-preview", "benchmark_id": "hy_backend_2_0", "score": 12.2, "reference_url": "https://huggingface.co/tencent/Hy3", "reported_setting": { "mode": "thinking", "effort": "high / highest available", "tools": "agentic code execution", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "Claude Code scaffold", "prompt_style": "default", "temperature": "0.9 for Hy3 family; otherwise source/provider setting", "context": "256000 for Hy3 family; otherwise source/provider setting", "notes": "Comparator value not marked as Tencent own test" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Hy3 official model-card appendix: Hy-Backend 2.0 (internal) = 12.2. Comparator-reported value." }, { "model_id": "hy3", "benchmark_id": "hy_backend_2_0", "score": 25.0, "reference_url": "https://huggingface.co/tencent/Hy3", "reported_setting": { "mode": "thinking", "effort": "high / highest available", "tools": "agentic code execution", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "Claude Code scaffold", "prompt_style": "default", "temperature": "0.9 for Hy3 family; otherwise source/provider setting", "context": "256000 for Hy3 family; otherwise source/provider setting", "notes": "Comparator value not marked as Tencent own test" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Hy3 official model-card appendix: Hy-Backend 2.0 (internal) = 25.0. Comparator-reported value." }, { "model_id": "glm-5.1", "benchmark_id": "hy_backend_2_0", "score": 24.7, "reference_url": "https://huggingface.co/tencent/Hy3", "reported_setting": { "mode": "thinking", "effort": "high / highest available", "tools": "agentic code execution", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "Claude Code scaffold", "prompt_style": "default", "temperature": "0.9 for Hy3 family; otherwise source/provider setting", "context": "256000 for Hy3 family; otherwise source/provider setting", "notes": "Tencent own test" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Hy3 official model-card appendix: Hy-Backend 2.0 (internal) = 24.7*. Tencent own testing." }, { "model_id": "glm-5.2", "benchmark_id": "hy_backend_2_0", "score": 30.4, "reference_url": "https://huggingface.co/tencent/Hy3", "reported_setting": { "mode": "thinking", "effort": "high / highest available", "tools": "agentic code execution", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "Claude Code scaffold", "prompt_style": "default", "temperature": "0.9 for Hy3 family; otherwise source/provider setting", "context": "256000 for Hy3 family; otherwise source/provider setting", "notes": "Tencent own test" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Hy3 official model-card appendix: Hy-Backend 2.0 (internal) = 30.4*. Tencent own testing." }, { "model_id": "deepseek-v4-flash", "benchmark_id": "hy_backend_2_0", "score": 16.0, "reference_url": "https://huggingface.co/tencent/Hy3", "reported_setting": { "mode": "thinking", "effort": "high / highest available", "tools": "agentic code execution", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "Claude Code scaffold", "prompt_style": "default", "temperature": "0.9 for Hy3 family; otherwise source/provider setting", "context": "256000 for Hy3 family; otherwise source/provider setting", "notes": "Tencent own test" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Hy3 official model-card appendix: Hy-Backend 2.0 (internal) = 16.0*. Tencent own testing." }, { "model_id": "deepseek-v4-pro", "benchmark_id": "hy_backend_2_0", "score": 17.5, "reference_url": "https://huggingface.co/tencent/Hy3", "reported_setting": { "mode": "thinking", "effort": "high / highest available", "tools": "agentic code execution", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "Claude Code scaffold", "prompt_style": "default", "temperature": "0.9 for Hy3 family; otherwise source/provider setting", "context": "256000 for Hy3 family; otherwise source/provider setting", "notes": "Tencent own test" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Hy3 official model-card appendix: Hy-Backend 2.0 (internal) = 17.5*. Tencent own testing." }, { "model_id": "doubao-seed-2.1-pro", "benchmark_id": "hy_backend_2_0", "score": 23.7, "reference_url": "https://huggingface.co/tencent/Hy3", "reported_setting": { "mode": "thinking", "effort": "high / highest available", "tools": "agentic code execution", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "Claude Code scaffold", "prompt_style": "default", "temperature": "0.9 for Hy3 family; otherwise source/provider setting", "context": "256000 for Hy3 family; otherwise source/provider setting", "notes": "Tencent own test" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Hy3 official model-card appendix: Hy-Backend 2.0 (internal) = 23.7*. Tencent own testing." }, { "model_id": "qwen3.7-max", "benchmark_id": "hy_backend_2_0", "score": 24.7, "reference_url": "https://huggingface.co/tencent/Hy3", "reported_setting": { "mode": "thinking", "effort": "high / highest available", "tools": "agentic code execution", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "Claude Code scaffold", "prompt_style": "default", "temperature": "0.9 for Hy3 family; otherwise source/provider setting", "context": "256000 for Hy3 family; otherwise source/provider setting", "notes": "Tencent own test" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Hy3 official model-card appendix: Hy-Backend 2.0 (internal) = 24.7*. Tencent own testing." }, { "model_id": "claude-opus-4.8", "benchmark_id": "hy_backend_2_0", "score": 38.9, "reference_url": "https://huggingface.co/tencent/Hy3", "reported_setting": { "mode": "thinking", "effort": "high / highest available", "tools": "agentic code execution", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "Claude Code scaffold", "prompt_style": "default", "temperature": "0.9 for Hy3 family; otherwise source/provider setting", "context": "256000 for Hy3 family; otherwise source/provider setting", "notes": "Tencent own test" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Hy3 official model-card appendix: Hy-Backend 2.0 (internal) = 38.9*. Tencent own testing." }, { "model_id": "gpt-5.5", "benchmark_id": "hy_backend_2_0", "score": 38.7, "reference_url": "https://huggingface.co/tencent/Hy3", "reported_setting": { "mode": "thinking", "effort": "high / highest available", "tools": "agentic code execution", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "CodeX scaffold", "prompt_style": "default", "temperature": "0.9 for Hy3 family; otherwise source/provider setting", "context": "256000 for Hy3 family; otherwise source/provider setting", "notes": "Tencent own test" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Hy3 official model-card appendix: Hy-Backend 2.0 (internal) = 38.7*. Tencent own testing." }, { "model_id": "hy3-preview", "benchmark_id": "hy_swe_max", "score": 30.0, "reference_url": "https://huggingface.co/tencent/Hy3", "reported_setting": { "mode": "thinking", "effort": "high / highest available", "tools": "agentic code execution", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "Claude Code scaffold", "prompt_style": "default", "temperature": "0.9 for Hy3 family; otherwise source/provider setting", "context": "256000 for Hy3 family; otherwise source/provider setting", "notes": "Comparator value not marked as Tencent own test" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Hy3 official model-card appendix: Hy-SWE Max (internal) = 30.0. Comparator-reported value." }, { "model_id": "hy3", "benchmark_id": "hy_swe_max", "score": 49.0, "reference_url": "https://huggingface.co/tencent/Hy3", "reported_setting": { "mode": "thinking", "effort": "high / highest available", "tools": "agentic code execution", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "Claude Code scaffold", "prompt_style": "default", "temperature": "0.9 for Hy3 family; otherwise source/provider setting", "context": "256000 for Hy3 family; otherwise source/provider setting", "notes": "Comparator value not marked as Tencent own test" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Hy3 official model-card appendix: Hy-SWE Max (internal) = 49.0. Comparator-reported value." }, { "model_id": "glm-5.1", "benchmark_id": "hy_swe_max", "score": 46.1, "reference_url": "https://huggingface.co/tencent/Hy3", "reported_setting": { "mode": "thinking", "effort": "high / highest available", "tools": "agentic code execution", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "Claude Code scaffold", "prompt_style": "default", "temperature": "0.9 for Hy3 family; otherwise source/provider setting", "context": "256000 for Hy3 family; otherwise source/provider setting", "notes": "Tencent own test" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Hy3 official model-card appendix: Hy-SWE Max (internal) = 46.1*. Tencent own testing." }, { "model_id": "glm-5.2", "benchmark_id": "hy_swe_max", "score": 63.1, "reference_url": "https://huggingface.co/tencent/Hy3", "reported_setting": { "mode": "thinking", "effort": "high / highest available", "tools": "agentic code execution", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "Claude Code scaffold", "prompt_style": "default", "temperature": "0.9 for Hy3 family; otherwise source/provider setting", "context": "256000 for Hy3 family; otherwise source/provider setting", "notes": "Tencent own test" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Hy3 official model-card appendix: Hy-SWE Max (internal) = 63.1*. Tencent own testing." }, { "model_id": "deepseek-v4-flash", "benchmark_id": "hy_swe_max", "score": 40.0, "reference_url": "https://huggingface.co/tencent/Hy3", "reported_setting": { "mode": "thinking", "effort": "high / highest available", "tools": "agentic code execution", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "Claude Code scaffold", "prompt_style": "default", "temperature": "0.9 for Hy3 family; otherwise source/provider setting", "context": "256000 for Hy3 family; otherwise source/provider setting", "notes": "Tencent own test" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Hy3 official model-card appendix: Hy-SWE Max (internal) = 40.0*. Tencent own testing." }, { "model_id": "deepseek-v4-pro", "benchmark_id": "hy_swe_max", "score": 43.1, "reference_url": "https://huggingface.co/tencent/Hy3", "reported_setting": { "mode": "thinking", "effort": "high / highest available", "tools": "agentic code execution", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "Claude Code scaffold", "prompt_style": "default", "temperature": "0.9 for Hy3 family; otherwise source/provider setting", "context": "256000 for Hy3 family; otherwise source/provider setting", "notes": "Tencent own test" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Hy3 official model-card appendix: Hy-SWE Max (internal) = 43.1*. Tencent own testing." }, { "model_id": "doubao-seed-2.1-pro", "benchmark_id": "hy_swe_max", "score": 55.1, "reference_url": "https://huggingface.co/tencent/Hy3", "reported_setting": { "mode": "thinking", "effort": "high / highest available", "tools": "agentic code execution", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "Claude Code scaffold", "prompt_style": "default", "temperature": "0.9 for Hy3 family; otherwise source/provider setting", "context": "256000 for Hy3 family; otherwise source/provider setting", "notes": "Tencent own test" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Hy3 official model-card appendix: Hy-SWE Max (internal) = 55.1*. Tencent own testing." }, { "model_id": "qwen3.7-max", "benchmark_id": "hy_swe_max", "score": 48.8, "reference_url": "https://huggingface.co/tencent/Hy3", "reported_setting": { "mode": "thinking", "effort": "high / highest available", "tools": "agentic code execution", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "Claude Code scaffold", "prompt_style": "default", "temperature": "0.9 for Hy3 family; otherwise source/provider setting", "context": "256000 for Hy3 family; otherwise source/provider setting", "notes": "Tencent own test" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Hy3 official model-card appendix: Hy-SWE Max (internal) = 48.8*. Tencent own testing." }, { "model_id": "claude-opus-4.8", "benchmark_id": "hy_swe_max", "score": 63.2, "reference_url": "https://huggingface.co/tencent/Hy3", "reported_setting": { "mode": "thinking", "effort": "high / highest available", "tools": "agentic code execution", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "Claude Code scaffold", "prompt_style": "default", "temperature": "0.9 for Hy3 family; otherwise source/provider setting", "context": "256000 for Hy3 family; otherwise source/provider setting", "notes": "Tencent own test" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Hy3 official model-card appendix: Hy-SWE Max (internal) = 63.2*. Tencent own testing." }, { "model_id": "gpt-5.5", "benchmark_id": "hy_swe_max", "score": 65.4, "reference_url": "https://huggingface.co/tencent/Hy3", "reported_setting": { "mode": "thinking", "effort": "high / highest available", "tools": "agentic code execution", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "CodeX scaffold", "prompt_style": "default", "temperature": "0.9 for Hy3 family; otherwise source/provider setting", "context": "256000 for Hy3 family; otherwise source/provider setting", "notes": "Tencent own test" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Hy3 official model-card appendix: Hy-SWE Max (internal) = 65.4*. Tencent own testing." }, { "model_id": "hy3-preview", "benchmark_id": "hy_company_bench", "score": 8.3, "reference_url": "https://huggingface.co/tencent/Hy3", "reported_setting": { "mode": "thinking", "effort": "high / highest available", "tools": "agentic code execution", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "Claude Code scaffold", "prompt_style": "default", "temperature": "0.9 for Hy3 family; otherwise source/provider setting", "context": "256000 for Hy3 family; otherwise source/provider setting", "notes": "Comparator value not marked as Tencent own test" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Hy3 official model-card appendix: Hy-CompanyBench (internal) = 8.3. Comparator-reported value." }, { "model_id": "hy3", "benchmark_id": "hy_company_bench", "score": 41.7, "reference_url": "https://huggingface.co/tencent/Hy3", "reported_setting": { "mode": "thinking", "effort": "high / highest available", "tools": "agentic code execution", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "Claude Code scaffold", "prompt_style": "default", "temperature": "0.9 for Hy3 family; otherwise source/provider setting", "context": "256000 for Hy3 family; otherwise source/provider setting", "notes": "Comparator value not marked as Tencent own test" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Hy3 official model-card appendix: Hy-CompanyBench (internal) = 41.7. Comparator-reported value." }, { "model_id": "glm-5.1", "benchmark_id": "hy_company_bench", "score": 28.3, "reference_url": "https://huggingface.co/tencent/Hy3", "reported_setting": { "mode": "thinking", "effort": "high / highest available", "tools": "agentic code execution", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "Claude Code scaffold", "prompt_style": "default", "temperature": "0.9 for Hy3 family; otherwise source/provider setting", "context": "256000 for Hy3 family; otherwise source/provider setting", "notes": "Tencent own test" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Hy3 official model-card appendix: Hy-CompanyBench (internal) = 28.3*. Tencent own testing." }, { "model_id": "glm-5.2", "benchmark_id": "hy_company_bench", "score": 55.0, "reference_url": "https://huggingface.co/tencent/Hy3", "reported_setting": { "mode": "thinking", "effort": "high / highest available", "tools": "agentic code execution", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "Claude Code scaffold", "prompt_style": "default", "temperature": "0.9 for Hy3 family; otherwise source/provider setting", "context": "256000 for Hy3 family; otherwise source/provider setting", "notes": "Tencent own test" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Hy3 official model-card appendix: Hy-CompanyBench (internal) = 55.0*. Tencent own testing." }, { "model_id": "deepseek-v4-flash", "benchmark_id": "hy_company_bench", "score": 28.3, "reference_url": "https://huggingface.co/tencent/Hy3", "reported_setting": { "mode": "thinking", "effort": "high / highest available", "tools": "agentic code execution", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "Claude Code scaffold", "prompt_style": "default", "temperature": "0.9 for Hy3 family; otherwise source/provider setting", "context": "256000 for Hy3 family; otherwise source/provider setting", "notes": "Tencent own test" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Hy3 official model-card appendix: Hy-CompanyBench (internal) = 28.3*. Tencent own testing." }, { "model_id": "deepseek-v4-pro", "benchmark_id": "hy_company_bench", "score": 18.3, "reference_url": "https://huggingface.co/tencent/Hy3", "reported_setting": { "mode": "thinking", "effort": "high / highest available", "tools": "agentic code execution", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "Claude Code scaffold", "prompt_style": "default", "temperature": "0.9 for Hy3 family; otherwise source/provider setting", "context": "256000 for Hy3 family; otherwise source/provider setting", "notes": "Tencent own test" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Hy3 official model-card appendix: Hy-CompanyBench (internal) = 18.3*. Tencent own testing." }, { "model_id": "doubao-seed-2.1-pro", "benchmark_id": "hy_company_bench", "score": 20.0, "reference_url": "https://huggingface.co/tencent/Hy3", "reported_setting": { "mode": "thinking", "effort": "high / highest available", "tools": "agentic code execution", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "Claude Code scaffold", "prompt_style": "default", "temperature": "0.9 for Hy3 family; otherwise source/provider setting", "context": "256000 for Hy3 family; otherwise source/provider setting", "notes": "Tencent own test" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Hy3 official model-card appendix: Hy-CompanyBench (internal) = 20.0*. Tencent own testing." }, { "model_id": "qwen3.7-max", "benchmark_id": "hy_company_bench", "score": 43.3, "reference_url": "https://huggingface.co/tencent/Hy3", "reported_setting": { "mode": "thinking", "effort": "high / highest available", "tools": "agentic code execution", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "Claude Code scaffold", "prompt_style": "default", "temperature": "0.9 for Hy3 family; otherwise source/provider setting", "context": "256000 for Hy3 family; otherwise source/provider setting", "notes": "Tencent own test" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Hy3 official model-card appendix: Hy-CompanyBench (internal) = 43.3*. Tencent own testing." }, { "model_id": "claude-opus-4.8", "benchmark_id": "hy_company_bench", "score": 65.0, "reference_url": "https://huggingface.co/tencent/Hy3", "reported_setting": { "mode": "thinking", "effort": "high / highest available", "tools": "agentic code execution", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "Claude Code scaffold", "prompt_style": "default", "temperature": "0.9 for Hy3 family; otherwise source/provider setting", "context": "256000 for Hy3 family; otherwise source/provider setting", "notes": "Tencent own test" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Hy3 official model-card appendix: Hy-CompanyBench (internal) = 65.0*. Tencent own testing." }, { "model_id": "gpt-5.5", "benchmark_id": "hy_company_bench", "score": 51.7, "reference_url": "https://huggingface.co/tencent/Hy3", "reported_setting": { "mode": "thinking", "effort": "high / highest available", "tools": "agentic code execution", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "CodeX scaffold", "prompt_style": "default", "temperature": "0.9 for Hy3 family; otherwise source/provider setting", "context": "256000 for Hy3 family; otherwise source/provider setting", "notes": "Tencent own test" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Hy3 official model-card appendix: Hy-CompanyBench (internal) = 51.7*. Tencent own testing." }, { "model_id": "hy3-preview", "benchmark_id": "browsecomp", "score": 67.1, "reference_url": "https://huggingface.co/tencent/Hy3", "reported_setting": { "mode": "thinking", "effort": "high / highest available", "tools": "web search/browser", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "Tencent internal search harness; BrowseComp uses self-summary context management", "prompt_style": "default", "temperature": "0.9 for Hy3 family; otherwise source/provider setting", "context": "256000 for Hy3 family; otherwise source/provider setting", "notes": "Comparator value not marked as Tencent own test" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Hy3 official model-card appendix: BrowseComp = 67.1. Comparator-reported value." }, { "model_id": "hy3", "benchmark_id": "browsecomp", "score": 84.2, "reference_url": "https://huggingface.co/tencent/Hy3", "reported_setting": { "mode": "thinking", "effort": "high / highest available", "tools": "web search/browser", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "Tencent internal search harness; BrowseComp uses self-summary context management", "prompt_style": "default", "temperature": "0.9 for Hy3 family; otherwise source/provider setting", "context": "256000 for Hy3 family; otherwise source/provider setting", "notes": "Comparator value not marked as Tencent own test" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Hy3 official model-card appendix: BrowseComp = 84.2. Comparator-reported value." }, { "model_id": "doubao-seed-2.1-pro", "benchmark_id": "browsecomp", "score": 86.2, "reference_url": "https://seed.bytedance.com/en/blog/seed2-1-officially-released-advancing-ai-productivity", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "search", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official", "prompt_style": "official default", "temperature": "unknown" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "Release blog EN: Language/search chart (https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/user-upload/4xfa4mqq510uc.png).", "candidates": [ { "score": 86.2, "reference_url": "https://huggingface.co/tencent/Hy3", "source_type": "model_card", "reported_setting": { "mode": "thinking", "effort": "high / highest available", "tools": "web search/browser", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "Tencent internal search harness; BrowseComp uses self-summary context management", "prompt_style": "default", "temperature": "0.9 for Hy3 family; otherwise source/provider setting", "context": "256000 for Hy3 family; otherwise source/provider setting", "notes": "Comparator value not marked as Tencent own test" }, "notes": "Hy3 official model-card appendix: BrowseComp = 86.2. Comparator-reported value." } ] }, { "model_id": "hy3-preview", "benchmark_id": "widesearch", "score": 67.8, "reference_url": "https://huggingface.co/tencent/Hy3", "reported_setting": { "mode": "thinking", "effort": "high / highest available", "tools": "web search/browser", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "Tencent internal search harness; BrowseComp uses self-summary context management", "prompt_style": "default", "temperature": "0.9 for Hy3 family; otherwise source/provider setting", "context": "256000 for Hy3 family; otherwise source/provider setting", "notes": "Comparator value not marked as Tencent own test" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Hy3 official model-card appendix: WideSearch = 67.8. Comparator-reported value." }, { "model_id": "hy3", "benchmark_id": "widesearch", "score": 76.4, "reference_url": "https://huggingface.co/tencent/Hy3", "reported_setting": { "mode": "thinking", "effort": "high / highest available", "tools": "web search/browser", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "Tencent internal search harness; BrowseComp uses self-summary context management", "prompt_style": "default", "temperature": "0.9 for Hy3 family; otherwise source/provider setting", "context": "256000 for Hy3 family; otherwise source/provider setting", "notes": "Comparator value not marked as Tencent own test" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Hy3 official model-card appendix: WideSearch = 76.4. Comparator-reported value." }, { "model_id": "glm-5.1", "benchmark_id": "widesearch", "score": 76.1, "reference_url": "https://huggingface.co/tencent/Hy3", "reported_setting": { "mode": "thinking", "effort": "high / highest available", "tools": "web search/browser", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "Tencent internal search harness; BrowseComp uses self-summary context management", "prompt_style": "default", "temperature": "0.9 for Hy3 family; otherwise source/provider setting", "context": "256000 for Hy3 family; otherwise source/provider setting", "notes": "Tencent own test" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Hy3 official model-card appendix: WideSearch = 76.1*. Tencent own testing." }, { "model_id": "deepseek-v4-flash", "benchmark_id": "widesearch", "score": 74.3, "reference_url": "https://huggingface.co/tencent/Hy3", "reported_setting": { "mode": "thinking", "effort": "high / highest available", "tools": "web search/browser", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "Tencent internal search harness; BrowseComp uses self-summary context management", "prompt_style": "default", "temperature": "0.9 for Hy3 family; otherwise source/provider setting", "context": "256000 for Hy3 family; otherwise source/provider setting", "notes": "Tencent own test" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Hy3 official model-card appendix: WideSearch = 74.3*. Tencent own testing." }, { "model_id": "deepseek-v4-pro", "benchmark_id": "widesearch", "score": 75.7, "reference_url": "https://huggingface.co/tencent/Hy3", "reported_setting": { "mode": "thinking", "effort": "high / highest available", "tools": "web search/browser", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "Tencent internal search harness; BrowseComp uses self-summary context management", "prompt_style": "default", "temperature": "0.9 for Hy3 family; otherwise source/provider setting", "context": "256000 for Hy3 family; otherwise source/provider setting", "notes": "Tencent own test" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Hy3 official model-card appendix: WideSearch = 75.7*. Tencent own testing." }, { "model_id": "doubao-seed-2.1-pro", "benchmark_id": "widesearch", "score": 76.4, "reference_url": "https://huggingface.co/tencent/Hy3", "reported_setting": { "mode": "thinking", "effort": "high / highest available", "tools": "web search/browser", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "Tencent internal search harness; BrowseComp uses self-summary context management", "prompt_style": "default", "temperature": "0.9 for Hy3 family; otherwise source/provider setting", "context": "256000 for Hy3 family; otherwise source/provider setting", "notes": "Tencent own test" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Hy3 official model-card appendix: WideSearch = 76.4*. Tencent own testing." }, { "model_id": "qwen3.7-max", "benchmark_id": "widesearch", "score": 72.3, "reference_url": "https://huggingface.co/tencent/Hy3", "reported_setting": { "mode": "thinking", "effort": "high / highest available", "tools": "web search/browser", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "Tencent internal search harness; BrowseComp uses self-summary context management", "prompt_style": "default", "temperature": "0.9 for Hy3 family; otherwise source/provider setting", "context": "256000 for Hy3 family; otherwise source/provider setting", "notes": "Tencent own test" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Hy3 official model-card appendix: WideSearch = 72.3*. Tencent own testing." }, { "model_id": "gemini-3.1-pro", "benchmark_id": "widesearch", "score": 71.7, "reference_url": "https://huggingface.co/tencent/Hy3", "reported_setting": { "mode": "thinking", "effort": "high / highest available", "tools": "web search/browser", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "Tencent internal search harness; BrowseComp uses self-summary context management", "prompt_style": "default", "temperature": "0.9 for Hy3 family; otherwise source/provider setting", "context": "256000 for Hy3 family; otherwise source/provider setting", "notes": "Tencent own test" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Hy3 official model-card appendix: WideSearch = 71.7*. Tencent own testing." }, { "model_id": "claude-opus-4.8", "benchmark_id": "widesearch", "score": 72.9, "reference_url": "https://huggingface.co/tencent/Hy3", "reported_setting": { "mode": "thinking", "effort": "high / highest available", "tools": "web search/browser", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "Tencent internal search harness; BrowseComp uses self-summary context management", "prompt_style": "default", "temperature": "0.9 for Hy3 family; otherwise source/provider setting", "context": "256000 for Hy3 family; otherwise source/provider setting", "notes": "Tencent own test" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Hy3 official model-card appendix: WideSearch = 72.9*. Tencent own testing." }, { "model_id": "gpt-5.5", "benchmark_id": "widesearch", "score": 80.0, "reference_url": "https://huggingface.co/tencent/Hy3", "reported_setting": { "mode": "thinking", "effort": "high / highest available", "tools": "web search/browser", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "Tencent internal search harness; BrowseComp uses self-summary context management", "prompt_style": "default", "temperature": "0.9 for Hy3 family; otherwise source/provider setting", "context": "256000 for Hy3 family; otherwise source/provider setting", "notes": "Tencent own test" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Hy3 official model-card appendix: WideSearch = 80.0*. Tencent own testing." }, { "model_id": "hy3-preview", "benchmark_id": "deepsearchqa_f1", "score": 82.8, "reference_url": "https://huggingface.co/tencent/Hy3", "reported_setting": { "mode": "thinking", "effort": "high / highest available", "tools": "web search/browser", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "Tencent internal search harness; BrowseComp uses self-summary context management", "prompt_style": "default", "temperature": "0.9 for Hy3 family; otherwise source/provider setting", "context": "256000 for Hy3 family; otherwise source/provider setting", "notes": "Comparator value not marked as Tencent own test" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Hy3 official model-card appendix: DeepSearchQA = 82.8. Comparator-reported value." }, { "model_id": "hy3", "benchmark_id": "deepsearchqa_f1", "score": 91.0, "reference_url": "https://huggingface.co/tencent/Hy3", "reported_setting": { "mode": "thinking", "effort": "high / highest available", "tools": "web search/browser", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "Tencent internal search harness; BrowseComp uses self-summary context management", "prompt_style": "default", "temperature": "0.9 for Hy3 family; otherwise source/provider setting", "context": "256000 for Hy3 family; otherwise source/provider setting", "notes": "Comparator value not marked as Tencent own test" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Hy3 official model-card appendix: DeepSearchQA = 91.0. Comparator-reported value." }, { "model_id": "glm-5.1", "benchmark_id": "deepsearchqa_f1", "score": 89.1, "reference_url": "https://huggingface.co/tencent/Hy3", "reported_setting": { "mode": "thinking", "effort": "high / highest available", "tools": "web search/browser", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "Tencent internal search harness; BrowseComp uses self-summary context management", "prompt_style": "default", "temperature": "0.9 for Hy3 family; otherwise source/provider setting", "context": "256000 for Hy3 family; otherwise source/provider setting", "notes": "Tencent own test" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Hy3 official model-card appendix: DeepSearchQA = 89.1*. Tencent own testing.", "candidates": [ { "score": 91.2, "reference_url": "https://static.stepfun.com/blog/step-3.7-flash/", "source_type": "official_blog", "reported_setting": { "quantization_config": "common release evaluation; artifact precision not established", "configuration": "base Step evaluation configuration", "mode": "source does not state", "effort": "source does not state", "tools_search": "web search/research agent", "shots": "source does not state", "trials_samples": "source does not state", "aggregation_pass_k": "source-reported score", "context": "source does not state", "output_cap": "source does not state", "temperature": "source does not state", "top_p": "source does not state", "judge": "Gemini 2.5 Flash with official Kaggle grading prompt", "harness_agent": "official StepFun evaluation; exact harness not stated", "dataset_version_split": "DeepSearchQA 900-prompt eval split", "multimodal_input": false }, "notes": "Official StepFun launch comparison table." } ] }, { "model_id": "deepseek-v4-flash", "benchmark_id": "deepsearchqa_f1", "score": 87.7, "reference_url": "https://huggingface.co/tencent/Hy3", "reported_setting": { "mode": "thinking", "effort": "high / highest available", "tools": "web search/browser", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "Tencent internal search harness; BrowseComp uses self-summary context management", "prompt_style": "default", "temperature": "0.9 for Hy3 family; otherwise source/provider setting", "context": "256000 for Hy3 family; otherwise source/provider setting", "notes": "Tencent own test" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Hy3 official model-card appendix: DeepSearchQA = 87.7*. Tencent own testing.", "candidates": [ { "score": 90.6, "reference_url": "https://static.stepfun.com/blog/step-3.7-flash/", "source_type": "official_blog", "reported_setting": { "quantization_config": "common release evaluation; artifact precision not established", "configuration": "base Step evaluation configuration", "mode": "source does not state", "effort": "source does not state", "tools_search": "web search/research agent", "shots": "source does not state", "trials_samples": "source does not state", "aggregation_pass_k": "source-reported score", "context": "source does not state", "output_cap": "source does not state", "temperature": "source does not state", "top_p": "source does not state", "judge": "Gemini 2.5 Flash with official Kaggle grading prompt", "harness_agent": "official StepFun evaluation; exact harness not stated", "dataset_version_split": "DeepSearchQA 900-prompt eval split", "multimodal_input": false }, "notes": "Official StepFun launch comparison table." } ] }, { "model_id": "deepseek-v4-pro", "benchmark_id": "deepsearchqa_f1", "score": 90.4, "reference_url": "https://huggingface.co/tencent/Hy3", "reported_setting": { "mode": "thinking", "effort": "high / highest available", "tools": "web search/browser", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "Tencent internal search harness; BrowseComp uses self-summary context management", "prompt_style": "default", "temperature": "0.9 for Hy3 family; otherwise source/provider setting", "context": "256000 for Hy3 family; otherwise source/provider setting", "notes": "Tencent own test" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Hy3 official model-card appendix: DeepSearchQA = 90.4*. Tencent own testing.", "candidates": [ { "score": 86.7, "reference_url": "https://ernie.baidu.com/blog/posts/ernie-5.1-0508-release/benchmark.png", "source_type": "official_blog", "reported_setting": { "tools": "search/agentic", "metric": "F1" }, "notes": "F1 search-agent row matches the campaign benchmark identity. Research observation obs-010." } ] }, { "model_id": "doubao-seed-2.1-pro", "benchmark_id": "deepsearchqa_f1", "score": 90.8, "reference_url": "https://huggingface.co/tencent/Hy3", "reported_setting": { "mode": "thinking", "effort": "high / highest available", "tools": "web search/browser", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "Tencent internal search harness; BrowseComp uses self-summary context management", "prompt_style": "default", "temperature": "0.9 for Hy3 family; otherwise source/provider setting", "context": "256000 for Hy3 family; otherwise source/provider setting", "notes": "Tencent own test" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Hy3 official model-card appendix: DeepSearchQA = 90.8*. Tencent own testing." }, { "model_id": "qwen3.7-max", "benchmark_id": "deepsearchqa_f1", "score": 88.4, "reference_url": "https://huggingface.co/tencent/Hy3", "reported_setting": { "mode": "thinking", "effort": "high / highest available", "tools": "web search/browser", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "Tencent internal search harness; BrowseComp uses self-summary context management", "prompt_style": "default", "temperature": "0.9 for Hy3 family; otherwise source/provider setting", "context": "256000 for Hy3 family; otherwise source/provider setting", "notes": "Tencent own test" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Hy3 official model-card appendix: DeepSearchQA = 88.4*. Tencent own testing." }, { "model_id": "gpt-5.5", "benchmark_id": "deepsearchqa_f1", "score": 95.5, "reference_url": "https://huggingface.co/tencent/Hy3", "reported_setting": { "mode": "thinking", "effort": "high / highest available", "tools": "web search/browser", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "Tencent internal search harness; BrowseComp uses self-summary context management", "prompt_style": "default", "temperature": "0.9 for Hy3 family; otherwise source/provider setting", "context": "256000 for Hy3 family; otherwise source/provider setting", "notes": "Tencent own test" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Hy3 official model-card appendix: DeepSearchQA = 95.5*. Tencent own testing.", "candidates": [ { "score": 87.8, "reference_url": "https://ai.meta.com/static-resource/muse-spark-1.1-evaluation-report", "source_type": "tech_report", "reported_setting": { "mode": "thinking", "effort": "xhigh", "tools": "search/browser", "sampling": "pass@1 unless source row states otherwise", "harness": "Meta MSL general-capability evaluation", "temperature": "1.0 for Muse; provider/source setting for comparisons", "context": "benchmark-specific", "configuration": "default" }, "notes": "Figure 44 · DeepSearchQA F1; metric=headline_metric. Exact printed value in the general-capability summary." }, { "score": 94.0, "reference_url": "https://static.stepfun.com/blog/step-3.7-flash/", "source_type": "official_blog", "reported_setting": { "quantization_config": "common release evaluation; artifact precision not established", "configuration": "base Step evaluation configuration", "mode": "source does not state", "effort": "source does not state", "tools_search": "web search/research agent", "shots": "source does not state", "trials_samples": "source does not state", "aggregation_pass_k": "source-reported score", "context": "source does not state", "output_cap": "source does not state", "temperature": "source does not state", "top_p": "source does not state", "judge": "Gemini 2.5 Flash with official Kaggle grading prompt", "harness_agent": "official StepFun evaluation; exact harness not stated", "dataset_version_split": "DeepSearchQA 900-prompt eval split", "multimodal_input": false }, "notes": "Official StepFun launch comparison table." } ] }, { "model_id": "hy3-preview", "benchmark_id": "mcpatlas", "score": 66.1, "reference_url": "https://huggingface.co/tencent/Hy3", "reported_setting": { "mode": "thinking", "effort": "high / highest available", "tools": "MCP tool servers", "sampling": "pass@1", "judge": "Gemini 2.5 Pro", "harness": "Scale April 2026 methodology; public 500; 100 tool-call budget", "prompt_style": "default", "temperature": "0.9 for Hy3 family; otherwise source/provider setting", "context": "256000 for Hy3 family; otherwise source/provider setting", "notes": "Comparator value not marked as Tencent own test" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Hy3 official model-card appendix: MCP atlas (public) = 66.1. Comparator-reported value." }, { "model_id": "hy3", "benchmark_id": "mcpatlas", "score": 79.1, "reference_url": "https://huggingface.co/tencent/Hy3", "reported_setting": { "mode": "thinking", "effort": "high / highest available", "tools": "MCP tool servers", "sampling": "pass@1", "judge": "Gemini 2.5 Pro", "harness": "Scale April 2026 methodology; public 500; 100 tool-call budget", "prompt_style": "default", "temperature": "0.9 for Hy3 family; otherwise source/provider setting", "context": "256000 for Hy3 family; otherwise source/provider setting", "notes": "Comparator value not marked as Tencent own test" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Hy3 official model-card appendix: MCP atlas (public) = 79.1. Comparator-reported value." }, { "model_id": "doubao-seed-2.1-pro", "benchmark_id": "mcpatlas", "score": 83.8, "reference_url": "https://seed.bytedance.com/en/blog/seed2-1-officially-released-advancing-ai-productivity", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "MCP servers in an agent harness", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official", "prompt_style": "official default", "temperature": "unknown" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "Release blog EN: Daily-life/tool-use chart (https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/user-upload/4xfa4mqq4ujsp.png).", "candidates": [ { "score": 80.6, "reference_url": "https://huggingface.co/tencent/Hy3", "source_type": "model_card", "reported_setting": { "mode": "thinking", "effort": "high / highest available", "tools": "MCP tool servers", "sampling": "pass@1", "judge": "Gemini 2.5 Pro", "harness": "Scale April 2026 methodology; public 500; 100 tool-call budget", "prompt_style": "default", "temperature": "0.9 for Hy3 family; otherwise source/provider setting", "context": "256000 for Hy3 family; otherwise source/provider setting", "notes": "Tencent own test" }, "notes": "Hy3 official model-card appendix: MCP atlas (public) = 80.6*. Tencent own testing." }, { "score": 83.8, "reference_url": "https://huggingface.co/tencent/Hy3", "source_type": "model_card", "reported_setting": { "mode": "thinking", "effort": "high / highest available", "tools": "MCP tool servers", "sampling": "pass@1", "judge": "Gemini 2.5 Pro", "harness": "Scale April 2026 methodology; public 500; 100 tool-call budget", "prompt_style": "default", "temperature": "0.9 for Hy3 family; otherwise source/provider setting", "context": "256000 for Hy3 family; otherwise source/provider setting", "notes": "Comparator value not marked as Tencent own test" }, "notes": "Hy3 official model-card appendix: MCP atlas (public) = 83.8. Comparator-reported value." } ] }, { "model_id": "hy3-preview", "benchmark_id": "toolathlon", "score": 32.1, "reference_url": "https://huggingface.co/tencent/Hy3", "reported_setting": { "mode": "thinking", "effort": "high / highest available", "tools": "agentic tool use", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "original Toolathlon official evaluation", "prompt_style": "default", "temperature": "0.9 for Hy3 family; otherwise source/provider setting", "context": "256000 for Hy3 family; otherwise source/provider setting", "notes": "Comparator value not marked as Tencent own test" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Hy3 official model-card appendix: Toolathlon = 32.1. Comparator-reported value." }, { "model_id": "hy3", "benchmark_id": "toolathlon", "score": 48.5, "reference_url": "https://huggingface.co/tencent/Hy3", "reported_setting": { "mode": "thinking", "effort": "high / highest available", "tools": "agentic tool use", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "original Toolathlon official evaluation", "prompt_style": "default", "temperature": "0.9 for Hy3 family; otherwise source/provider setting", "context": "256000 for Hy3 family; otherwise source/provider setting", "notes": "Comparator value not marked as Tencent own test" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Hy3 official model-card appendix: Toolathlon = 48.5. Comparator-reported value." }, { "model_id": "doubao-seed-2.1-pro", "benchmark_id": "toolathlon", "score": 50.6, "reference_url": "https://seed.bytedance.com/en/blog/seed2-1-officially-released-advancing-ai-productivity", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "agentic tool use", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official", "prompt_style": "official default", "temperature": "unknown" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "Release blog EN: Daily-life/tool-use chart (https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/user-upload/4xfa4mqq4ujsp.png).", "candidates": [ { "score": 44.8, "reference_url": "https://huggingface.co/tencent/Hy3", "source_type": "model_card", "reported_setting": { "mode": "thinking", "effort": "high / highest available", "tools": "agentic tool use", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "original Toolathlon official evaluation", "prompt_style": "default", "temperature": "0.9 for Hy3 family; otherwise source/provider setting", "context": "256000 for Hy3 family; otherwise source/provider setting", "notes": "Tencent own test" }, "notes": "Hy3 official model-card appendix: Toolathlon = 44.8*. Tencent own testing." }, { "score": 50.6, "reference_url": "https://huggingface.co/tencent/Hy3", "source_type": "model_card", "reported_setting": { "mode": "thinking", "effort": "high / highest available", "tools": "agentic tool use", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "original Toolathlon official evaluation", "prompt_style": "default", "temperature": "0.9 for Hy3 family; otherwise source/provider setting", "context": "256000 for Hy3 family; otherwise source/provider setting", "notes": "Comparator value not marked as Tencent own test" }, "notes": "Hy3 official model-card appendix: Toolathlon = 50.6. Comparator-reported value." } ] }, { "model_id": "qwen3.7-max", "benchmark_id": "toolathlon", "score": 45.1, "reference_url": "https://huggingface.co/tencent/Hy3", "reported_setting": { "mode": "thinking", "effort": "high / highest available", "tools": "agentic tool use", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "original Toolathlon official evaluation", "prompt_style": "default", "temperature": "0.9 for Hy3 family; otherwise source/provider setting", "context": "256000 for Hy3 family; otherwise source/provider setting", "notes": "Tencent own test" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Hy3 official model-card appendix: Toolathlon = 45.1*. Tencent own testing." }, { "model_id": "hy3-preview", "benchmark_id": "apex_agents", "score": 12.4, "reference_url": "https://huggingface.co/tencent/Hy3", "reported_setting": { "mode": "thinking", "effort": "high / highest available", "tools": "multiple application/software tools", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "APEX-Agents long-horizon professional task harness", "prompt_style": "default", "temperature": "0.9 for Hy3 family; otherwise source/provider setting", "context": "256000 for Hy3 family; otherwise source/provider setting", "notes": "Comparator value not marked as Tencent own test" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Hy3 official model-card appendix: Apex-Agent (pass@1) = 12.4. Comparator-reported value." }, { "model_id": "hy3", "benchmark_id": "apex_agents", "score": 25.6, "reference_url": "https://huggingface.co/tencent/Hy3", "reported_setting": { "mode": "thinking", "effort": "high / highest available", "tools": "multiple application/software tools", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "APEX-Agents long-horizon professional task harness", "prompt_style": "default", "temperature": "0.9 for Hy3 family; otherwise source/provider setting", "context": "256000 for Hy3 family; otherwise source/provider setting", "notes": "Comparator value not marked as Tencent own test" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Hy3 official model-card appendix: Apex-Agent (pass@1) = 25.6. Comparator-reported value." }, { "model_id": "glm-5.1", "benchmark_id": "apex_agents", "score": 24.3, "reference_url": "https://huggingface.co/tencent/Hy3", "reported_setting": { "mode": "thinking", "effort": "high / highest available", "tools": "multiple application/software tools", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "APEX-Agents long-horizon professional task harness", "prompt_style": "default", "temperature": "0.9 for Hy3 family; otherwise source/provider setting", "context": "256000 for Hy3 family; otherwise source/provider setting", "notes": "Tencent own test" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Hy3 official model-card appendix: Apex-Agent (pass@1) = 24.3*. Tencent own testing." }, { "model_id": "deepseek-v4-flash", "benchmark_id": "apex_agents", "score": 18.3, "reference_url": "https://huggingface.co/tencent/Hy3", "reported_setting": { "mode": "thinking", "effort": "high / highest available", "tools": "multiple application/software tools", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "APEX-Agents long-horizon professional task harness", "prompt_style": "default", "temperature": "0.9 for Hy3 family; otherwise source/provider setting", "context": "256000 for Hy3 family; otherwise source/provider setting", "notes": "Tencent own test" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Hy3 official model-card appendix: Apex-Agent (pass@1) = 18.3*. Tencent own testing." }, { "model_id": "deepseek-v4-pro", "benchmark_id": "apex_agents", "score": 20.4, "reference_url": "https://huggingface.co/tencent/Hy3", "reported_setting": { "mode": "thinking", "effort": "high / highest available", "tools": "multiple application/software tools", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "APEX-Agents long-horizon professional task harness", "prompt_style": "default", "temperature": "0.9 for Hy3 family; otherwise source/provider setting", "context": "256000 for Hy3 family; otherwise source/provider setting", "notes": "Tencent own test" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Hy3 official model-card appendix: Apex-Agent (pass@1) = 20.4*. Tencent own testing." }, { "model_id": "doubao-seed-2.1-pro", "benchmark_id": "apex_agents", "score": 33.8, "reference_url": "https://seed.bytedance.com/en/blog/seed2-1-officially-released-advancing-ai-productivity", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "long-horizon professional-task environment", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official", "prompt_style": "official default", "temperature": "unknown" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "Release blog EN: General-agent chart (https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/user-upload/4xfa4mqq4u89c.png).", "candidates": [ { "score": 33.8, "reference_url": "https://huggingface.co/tencent/Hy3", "source_type": "model_card", "reported_setting": { "mode": "thinking", "effort": "high / highest available", "tools": "multiple application/software tools", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "APEX-Agents long-horizon professional task harness", "prompt_style": "default", "temperature": "0.9 for Hy3 family; otherwise source/provider setting", "context": "256000 for Hy3 family; otherwise source/provider setting", "notes": "Comparator value not marked as Tencent own test" }, "notes": "Hy3 official model-card appendix: Apex-Agent (pass@1) = 33.8. Comparator-reported value." } ] }, { "model_id": "qwen3.7-max", "benchmark_id": "apex_agents", "score": 22.2, "reference_url": "https://huggingface.co/tencent/Hy3", "reported_setting": { "mode": "thinking", "effort": "high / highest available", "tools": "multiple application/software tools", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "APEX-Agents long-horizon professional task harness", "prompt_style": "default", "temperature": "0.9 for Hy3 family; otherwise source/provider setting", "context": "256000 for Hy3 family; otherwise source/provider setting", "notes": "Tencent own test" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Hy3 official model-card appendix: Apex-Agent (pass@1) = 22.2*. Tencent own testing." }, { "model_id": "hy3-preview", "benchmark_id": "claw_eval_pass3", "score": 55.0, "reference_url": "https://huggingface.co/tencent/Hy3", "reported_setting": { "mode": "thinking", "effort": "high / highest available", "tools": "agentic/OpenClaw-style", "sampling": "pass^3", "judge": "Gemini 3.5 Flash", "harness": "Tencent internal harness; 20260325 version; 105 queries", "prompt_style": "default", "temperature": "0.9 for Hy3 family; otherwise source/provider setting", "context": "256000 for Hy3 family; otherwise source/provider setting", "notes": "Comparator value not marked as Tencent own test" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Hy3 official model-card appendix: ClawEval (pass^3) = 55.0. Comparator-reported value." }, { "model_id": "hy3", "benchmark_id": "claw_eval_pass3", "score": 68.5, "reference_url": "https://huggingface.co/tencent/Hy3", "reported_setting": { "mode": "thinking", "effort": "high / highest available", "tools": "agentic/OpenClaw-style", "sampling": "pass^3", "judge": "Gemini 3.5 Flash", "harness": "Tencent internal harness; 20260325 version; 105 queries", "prompt_style": "default", "temperature": "0.9 for Hy3 family; otherwise source/provider setting", "context": "256000 for Hy3 family; otherwise source/provider setting", "notes": "Comparator value not marked as Tencent own test" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Hy3 official model-card appendix: ClawEval (pass^3) = 68.5. Comparator-reported value." }, { "model_id": "glm-5.1", "benchmark_id": "claw_eval_pass3", "score": 62.3, "reference_url": "https://raw.githubusercontent.com/claw-eval/claw-eval.github.io/e10516b5c2a6dfc139be35dacb54cdf5e034801d/assets/benchmark-BgnHRA8t.js", "reported_setting": { "quantization_config": "common release evaluation; artifact precision not established", "configuration": "base Step evaluation configuration", "mode": "source does not state", "effort": "source does not state", "tools_search": "official Claw-Eval agent environment", "shots": "source does not state", "trials_samples": 3, "aggregation_pass_k": "Pass^3: all three trials must pass", "context": "source does not state", "output_cap": "source does not state", "temperature": "source does not state", "top_p": "source does not state", "judge": "full-trajectory completion/safety/robustness grading", "harness_agent": "official Claw-Eval v1.1 non-multimodal leaderboard", "dataset_version_split": "Claw-Eval v1.1 non-multimodal: 161 general + 38 multi-turn tasks", "multimodal_input": false }, "matches_canonical": true, "source_type": "official_repository", "audit_status": "verified", "notes": "Official StepFun launch comparison table. Direct locked official score authority reports Claweval-v1.1=62.3; the StepFun surface is retained as physical corroboration.", "candidates": [ { "score": 62.7, "reference_url": "https://huggingface.co/tencent/Hy3", "source_type": "model_card", "reported_setting": { "mode": "thinking", "effort": "high / highest available", "tools": "agentic/OpenClaw-style", "sampling": "pass^3", "judge": "Gemini 3.5 Flash", "harness": "Tencent internal harness; 20260325 version; 105 queries", "prompt_style": "default", "temperature": "0.9 for Hy3 family; otherwise source/provider setting", "context": "256000 for Hy3 family; otherwise source/provider setting", "notes": "Comparator value not marked as Tencent own test" }, "notes": "Hy3 official model-card appendix: ClawEval (pass^3) = 62.7. Comparator-reported value." } ] }, { "model_id": "glm-5.2", "benchmark_id": "claw_eval_pass3", "score": 62.4, "reference_url": "https://huggingface.co/tencent/Hy3", "reported_setting": { "mode": "thinking", "effort": "high / highest available", "tools": "agentic/OpenClaw-style", "sampling": "pass^3", "judge": "Gemini 3.5 Flash", "harness": "Tencent internal harness; 20260325 version; 105 queries", "prompt_style": "default", "temperature": "0.9 for Hy3 family; otherwise source/provider setting", "context": "256000 for Hy3 family; otherwise source/provider setting", "notes": "Tencent own test" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Hy3 official model-card appendix: ClawEval (pass^3) = 62.4*. Tencent own testing." }, { "model_id": "deepseek-v4-flash", "benchmark_id": "claw_eval_pass3", "score": 61.8, "reference_url": "https://huggingface.co/tencent/Hy3", "reported_setting": { "mode": "thinking", "effort": "high / highest available", "tools": "agentic/OpenClaw-style", "sampling": "pass^3", "judge": "Gemini 3.5 Flash", "harness": "Tencent internal harness; 20260325 version; 105 queries", "prompt_style": "default", "temperature": "0.9 for Hy3 family; otherwise source/provider setting", "context": "256000 for Hy3 family; otherwise source/provider setting", "notes": "Tencent own test" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Hy3 official model-card appendix: ClawEval (pass^3) = 61.8*. Tencent own testing.", "candidates": [ { "score": 57.8, "reference_url": "https://raw.githubusercontent.com/claw-eval/claw-eval.github.io/19d378a646d892ca6c8148691aea02da1b184eb8/assets/benchmark-0fIVfCD7.js", "source_type": "official_model_card", "reported_setting": { "mode": "source does not state", "effort": "source does not state", "tools": "official Claw-Eval agent environment", "sampling": "Pass^3; all three trajectories must pass", "judge": "full-trajectory benchmark grading", "harness": "official Claw-Eval v1.1 leaderboard", "prompt_style": "official Claw-Eval task prompts", "temperature": "source does not state", "context": "source does not state", "input_modalities": "text", "trials": "3 per task; 597 generations", "task_subset": "non-multimodal 199-task subset", "task_count": 199 }, "notes": "Exact official leaderboard field for the non-multimodal 199-task subset. Model mode/effort is not inferred." } ] }, { "model_id": "deepseek-v4-pro", "benchmark_id": "claw_eval_pass3", "score": 58.4, "reference_url": "https://huggingface.co/tencent/Hy3", "reported_setting": { "mode": "thinking", "effort": "high / highest available", "tools": "agentic/OpenClaw-style", "sampling": "pass^3", "judge": "Gemini 3.5 Flash", "harness": "Tencent internal harness; 20260325 version; 105 queries", "prompt_style": "default", "temperature": "0.9 for Hy3 family; otherwise source/provider setting", "context": "256000 for Hy3 family; otherwise source/provider setting", "notes": "Comparator value not marked as Tencent own test" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Hy3 official model-card appendix: ClawEval (pass^3) = 58.4. Comparator-reported value.", "candidates": [ { "score": 62.1, "reference_url": "https://huggingface.co/tencent/Hy3", "source_type": "model_card", "reported_setting": { "mode": "thinking", "effort": "high / highest available", "tools": "agentic/OpenClaw-style", "sampling": "pass^3", "judge": "Gemini 3.5 Flash", "harness": "Tencent internal harness; 20260325 version; 105 queries", "prompt_style": "default", "temperature": "0.9 for Hy3 family; otherwise source/provider setting", "context": "256000 for Hy3 family; otherwise source/provider setting", "notes": "Tencent own test" }, "notes": "Hy3 official model-card appendix: ClawEval (pass^3) = 62.1*. Tencent own testing." }, { "score": 59.8, "reference_url": "https://raw.githubusercontent.com/claw-eval/claw-eval.github.io/19d378a646d892ca6c8148691aea02da1b184eb8/assets/benchmark-0fIVfCD7.js", "source_type": "official_blog", "reported_setting": { "mode": "source does not state", "effort": "source does not state", "tools": "official Claw-Eval agent environment", "sampling": "Pass^3; all three trajectories must pass", "judge": "full-trajectory benchmark grading", "harness": "official Claw-Eval v1.1 leaderboard", "prompt_style": "official Claw-Eval task prompts", "temperature": "source does not state", "context": "source does not state", "input_modalities": "text", "trials": "3 per task; 597 generations", "task_subset": "non-multimodal 199-task subset", "task_count": 199 }, "notes": "DeepSeek V4 Pro is explicitly reported at max effort. Exact official leaderboard field for the non-multimodal 199-task subset. Model mode/effort is not inferred." } ] }, { "model_id": "doubao-seed-2.1-pro", "benchmark_id": "claw_eval_pass3", "score": 62.1, "reference_url": "https://huggingface.co/tencent/Hy3", "reported_setting": { "mode": "thinking", "effort": "high / highest available", "tools": "agentic/OpenClaw-style", "sampling": "pass^3", "judge": "Gemini 3.5 Flash", "harness": "Tencent internal harness; 20260325 version; 105 queries", "prompt_style": "default", "temperature": "0.9 for Hy3 family; otherwise source/provider setting", "context": "256000 for Hy3 family; otherwise source/provider setting", "notes": "Tencent own test" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Hy3 official model-card appendix: ClawEval (pass^3) = 62.1*. Tencent own testing." }, { "model_id": "qwen3.7-max", "benchmark_id": "claw_eval_pass3", "score": 65.2, "reference_url": "https://huggingface.co/tencent/Hy3", "reported_setting": { "mode": "thinking", "effort": "high / highest available", "tools": "agentic/OpenClaw-style", "sampling": "pass^3", "judge": "Gemini 3.5 Flash", "harness": "Tencent internal harness; 20260325 version; 105 queries", "prompt_style": "default", "temperature": "0.9 for Hy3 family; otherwise source/provider setting", "context": "256000 for Hy3 family; otherwise source/provider setting", "notes": "Comparator value not marked as Tencent own test" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Hy3 official model-card appendix: ClawEval (pass^3) = 65.2. Comparator-reported value." }, { "model_id": "gemini-3.1-pro", "benchmark_id": "claw_eval_pass3", "score": 50.0, "reference_url": "https://huggingface.co/tencent/Hy3", "reported_setting": { "mode": "thinking", "effort": "high / highest available", "tools": "agentic/OpenClaw-style", "sampling": "pass^3", "judge": "Gemini 3.5 Flash", "harness": "Tencent internal harness; 20260325 version; 105 queries", "prompt_style": "default", "temperature": "0.9 for Hy3 family; otherwise source/provider setting", "context": "256000 for Hy3 family; otherwise source/provider setting", "notes": "Tencent own test" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Hy3 official model-card appendix: ClawEval (pass^3) = 50.0*. Tencent own testing.", "candidates": [ { "score": 57.8, "reference_url": "https://raw.githubusercontent.com/claw-eval/claw-eval.github.io/19d378a646d892ca6c8148691aea02da1b184eb8/assets/benchmark-0fIVfCD7.js", "source_type": "official_blog", "reported_setting": { "mode": "source does not state", "effort": "source does not state", "tools": "official Claw-Eval agent environment", "sampling": "Pass^3; all three trajectories must pass", "judge": "full-trajectory benchmark grading", "harness": "official Claw-Eval v1.1 leaderboard", "prompt_style": "official Claw-Eval task prompts", "temperature": "source does not state", "context": "source does not state", "input_modalities": "text", "trials": "3 per task; 597 generations", "task_subset": "non-multimodal 199-task subset", "task_count": 199 }, "notes": "Exact official leaderboard field for the non-multimodal 199-task subset. Model mode/effort is not inferred." }, { "score": 55.9, "reference_url": "https://raw.githubusercontent.com/claw-eval/claw-eval.github.io/19d378a646d892ca6c8148691aea02da1b184eb8/assets/benchmark-0fIVfCD7.js", "source_type": "official_blog", "reported_setting": { "mode": "source does not state", "effort": "source does not state", "tools": "official Claw-Eval agent environment", "sampling": "Pass^3; all three trajectories must pass", "judge": "full-trajectory benchmark grading", "harness": "official Claw-Eval v1.1 leaderboard", "prompt_style": "official Claw-Eval task prompts", "temperature": "source does not state", "context": "source does not state", "input_modalities": "text", "trials": "3 per task; 483 generations", "task_subset": "core-general 161-task subset", "task_count": 161 }, "notes": "Average tokens per trajectory=226975 is a physical efficiency annotation, not a benchmark score or score setting. Exact official leaderboard field for the core-general 161-task subset. Model mode/effort is not inferred." } ] }, { "model_id": "claude-opus-4.8", "benchmark_id": "claw_eval_pass3", "score": 72.1, "reference_url": "https://huggingface.co/tencent/Hy3", "reported_setting": { "mode": "thinking", "effort": "high / highest available", "tools": "agentic/OpenClaw-style", "sampling": "pass^3", "judge": "Gemini 3.5 Flash", "harness": "Tencent internal harness; 20260325 version; 105 queries", "prompt_style": "default", "temperature": "0.9 for Hy3 family; otherwise source/provider setting", "context": "256000 for Hy3 family; otherwise source/provider setting", "notes": "Tencent own test" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Hy3 official model-card appendix: ClawEval (pass^3) = 72.1*. Tencent own testing." }, { "model_id": "gpt-5.5", "benchmark_id": "claw_eval_pass3", "score": 67.8, "reference_url": "https://huggingface.co/tencent/Hy3", "reported_setting": { "mode": "thinking", "effort": "high / highest available", "tools": "agentic/OpenClaw-style", "sampling": "pass^3", "judge": "Gemini 3.5 Flash", "harness": "Tencent internal harness; 20260325 version; 105 queries", "prompt_style": "default", "temperature": "0.9 for Hy3 family; otherwise source/provider setting", "context": "256000 for Hy3 family; otherwise source/provider setting", "notes": "Tencent own test" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Hy3 official model-card appendix: ClawEval (pass^3) = 67.8*. Tencent own testing." }, { "model_id": "hy3-preview", "benchmark_id": "wildclaw_bench_35_text", "score": 45.3, "reference_url": "https://huggingface.co/tencent/Hy3", "reported_setting": { "mode": "thinking", "effort": "high / highest available", "tools": "OpenClaw harness", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "text-only 35-query subset", "prompt_style": "default", "temperature": "0.9 for Hy3 family; otherwise source/provider setting", "context": "256000 for Hy3 family; otherwise source/provider setting", "notes": "Comparator value not marked as Tencent own test" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Hy3 official model-card appendix: WildClawBench (35, text-only) = 45.3. Comparator-reported value." }, { "model_id": "hy3", "benchmark_id": "wildclaw_bench_35_text", "score": 53.6, "reference_url": "https://huggingface.co/tencent/Hy3", "reported_setting": { "mode": "thinking", "effort": "high / highest available", "tools": "OpenClaw harness", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "text-only 35-query subset", "prompt_style": "default", "temperature": "0.9 for Hy3 family; otherwise source/provider setting", "context": "256000 for Hy3 family; otherwise source/provider setting", "notes": "Comparator value not marked as Tencent own test" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Hy3 official model-card appendix: WildClawBench (35, text-only) = 53.6. Comparator-reported value." }, { "model_id": "glm-5.1", "benchmark_id": "wildclaw_bench_35_text", "score": 48.7, "reference_url": "https://huggingface.co/tencent/Hy3", "reported_setting": { "mode": "thinking", "effort": "high / highest available", "tools": "OpenClaw harness", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "text-only 35-query subset", "prompt_style": "default", "temperature": "0.9 for Hy3 family; otherwise source/provider setting", "context": "256000 for Hy3 family; otherwise source/provider setting", "notes": "Tencent own test" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Hy3 official model-card appendix: WildClawBench (35, text-only) = 48.7*. Tencent own testing." }, { "model_id": "glm-5.2", "benchmark_id": "wildclaw_bench_35_text", "score": 59.2, "reference_url": "https://huggingface.co/tencent/Hy3", "reported_setting": { "mode": "thinking", "effort": "high / highest available", "tools": "OpenClaw harness", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "text-only 35-query subset", "prompt_style": "default", "temperature": "0.9 for Hy3 family; otherwise source/provider setting", "context": "256000 for Hy3 family; otherwise source/provider setting", "notes": "Tencent own test" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Hy3 official model-card appendix: WildClawBench (35, text-only) = 59.2*. Tencent own testing." }, { "model_id": "deepseek-v4-flash", "benchmark_id": "wildclaw_bench_35_text", "score": 43.5, "reference_url": "https://huggingface.co/tencent/Hy3", "reported_setting": { "mode": "thinking", "effort": "high / highest available", "tools": "OpenClaw harness", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "text-only 35-query subset", "prompt_style": "default", "temperature": "0.9 for Hy3 family; otherwise source/provider setting", "context": "256000 for Hy3 family; otherwise source/provider setting", "notes": "Tencent own test" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Hy3 official model-card appendix: WildClawBench (35, text-only) = 43.5*. Tencent own testing." }, { "model_id": "deepseek-v4-pro", "benchmark_id": "wildclaw_bench_35_text", "score": 50.5, "reference_url": "https://huggingface.co/tencent/Hy3", "reported_setting": { "mode": "thinking", "effort": "high / highest available", "tools": "OpenClaw harness", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "text-only 35-query subset", "prompt_style": "default", "temperature": "0.9 for Hy3 family; otherwise source/provider setting", "context": "256000 for Hy3 family; otherwise source/provider setting", "notes": "Tencent own test" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Hy3 official model-card appendix: WildClawBench (35, text-only) = 50.5*. Tencent own testing." }, { "model_id": "doubao-seed-2.1-pro", "benchmark_id": "wildclaw_bench_35_text", "score": 61.7, "reference_url": "https://huggingface.co/tencent/Hy3", "reported_setting": { "mode": "thinking", "effort": "high / highest available", "tools": "OpenClaw harness", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "text-only 35-query subset", "prompt_style": "default", "temperature": "0.9 for Hy3 family; otherwise source/provider setting", "context": "256000 for Hy3 family; otherwise source/provider setting", "notes": "Comparator value not marked as Tencent own test" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Hy3 official model-card appendix: WildClawBench (35, text-only) = 61.7. Comparator-reported value.", "candidates": [ { "score": 47.1, "reference_url": "https://huggingface.co/tencent/Hy3", "source_type": "model_card", "reported_setting": { "mode": "thinking", "effort": "high / highest available", "tools": "OpenClaw harness", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "text-only 35-query subset", "prompt_style": "default", "temperature": "0.9 for Hy3 family; otherwise source/provider setting", "context": "256000 for Hy3 family; otherwise source/provider setting", "notes": "Tencent own test" }, "notes": "Hy3 official model-card appendix: WildClawBench (35, text-only) = 47.1*. Tencent own testing." } ] }, { "model_id": "qwen3.7-max", "benchmark_id": "wildclaw_bench_35_text", "score": 41.5, "reference_url": "https://huggingface.co/tencent/Hy3", "reported_setting": { "mode": "thinking", "effort": "high / highest available", "tools": "OpenClaw harness", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "text-only 35-query subset", "prompt_style": "default", "temperature": "0.9 for Hy3 family; otherwise source/provider setting", "context": "256000 for Hy3 family; otherwise source/provider setting", "notes": "Tencent own test" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Hy3 official model-card appendix: WildClawBench (35, text-only) = 41.5*. Tencent own testing." }, { "model_id": "gemini-3.1-pro", "benchmark_id": "wildclaw_bench_35_text", "score": 46.2, "reference_url": "https://huggingface.co/tencent/Hy3", "reported_setting": { "mode": "thinking", "effort": "high / highest available", "tools": "OpenClaw harness", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "text-only 35-query subset", "prompt_style": "default", "temperature": "0.9 for Hy3 family; otherwise source/provider setting", "context": "256000 for Hy3 family; otherwise source/provider setting", "notes": "Tencent own test" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Hy3 official model-card appendix: WildClawBench (35, text-only) = 46.2*. Tencent own testing." }, { "model_id": "claude-opus-4.8", "benchmark_id": "wildclaw_bench_35_text", "score": 58.4, "reference_url": "https://huggingface.co/tencent/Hy3", "reported_setting": { "mode": "thinking", "effort": "high / highest available", "tools": "OpenClaw harness", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "text-only 35-query subset", "prompt_style": "default", "temperature": "0.9 for Hy3 family; otherwise source/provider setting", "context": "256000 for Hy3 family; otherwise source/provider setting", "notes": "Tencent own test" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Hy3 official model-card appendix: WildClawBench (35, text-only) = 58.4*. Tencent own testing." }, { "model_id": "gpt-5.5", "benchmark_id": "wildclaw_bench_35_text", "score": 63.2, "reference_url": "https://huggingface.co/tencent/Hy3", "reported_setting": { "mode": "thinking", "effort": "high / highest available", "tools": "OpenClaw harness", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "text-only 35-query subset", "prompt_style": "default", "temperature": "0.9 for Hy3 family; otherwise source/provider setting", "context": "256000 for Hy3 family; otherwise source/provider setting", "notes": "Tencent own test" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Hy3 official model-card appendix: WildClawBench (35, text-only) = 63.2*. Tencent own testing." }, { "model_id": "hy3-preview", "benchmark_id": "skills_bench_text_79", "score": 29.1, "reference_url": "https://huggingface.co/tencent/Hy3", "reported_setting": { "mode": "thinking", "effort": "high / highest available", "tools": "Claude Code", "sampling": "average over 3 runs", "judge": "benchmark-specified", "harness": "79-task self-contained text-only subset; multimodal excluded", "prompt_style": "default", "temperature": "0.9 for Hy3 family; otherwise source/provider setting", "context": "256000 for Hy3 family; otherwise source/provider setting", "notes": "Comparator value not marked as Tencent own test" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Hy3 official model-card appendix: SkillsBench (79, text-only) = 29.1. Comparator-reported value." }, { "model_id": "hy3", "benchmark_id": "skills_bench_text_79", "score": 55.3, "reference_url": "https://huggingface.co/tencent/Hy3", "reported_setting": { "mode": "thinking", "effort": "high / highest available", "tools": "Claude Code", "sampling": "average over 3 runs", "judge": "benchmark-specified", "harness": "79-task self-contained text-only subset; multimodal excluded", "prompt_style": "default", "temperature": "0.9 for Hy3 family; otherwise source/provider setting", "context": "256000 for Hy3 family; otherwise source/provider setting", "notes": "Comparator value not marked as Tencent own test" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Hy3 official model-card appendix: SkillsBench (79, text-only) = 55.3. Comparator-reported value." }, { "model_id": "glm-5.1", "benchmark_id": "skills_bench_text_79", "score": 43.0, "reference_url": "https://huggingface.co/tencent/Hy3", "reported_setting": { "mode": "thinking", "effort": "high / highest available", "tools": "Claude Code", "sampling": "average over 3 runs", "judge": "benchmark-specified", "harness": "79-task self-contained text-only subset; multimodal excluded", "prompt_style": "default", "temperature": "0.9 for Hy3 family; otherwise source/provider setting", "context": "256000 for Hy3 family; otherwise source/provider setting", "notes": "Tencent own test" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Hy3 official model-card appendix: SkillsBench (79, text-only) = 43.0*. Tencent own testing." }, { "model_id": "glm-5.2", "benchmark_id": "skills_bench_text_79", "score": 51.9, "reference_url": "https://huggingface.co/tencent/Hy3", "reported_setting": { "mode": "thinking", "effort": "high / highest available", "tools": "Claude Code", "sampling": "average over 3 runs", "judge": "benchmark-specified", "harness": "79-task self-contained text-only subset; multimodal excluded", "prompt_style": "default", "temperature": "0.9 for Hy3 family; otherwise source/provider setting", "context": "256000 for Hy3 family; otherwise source/provider setting", "notes": "Tencent own test" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Hy3 official model-card appendix: SkillsBench (79, text-only) = 51.9*. Tencent own testing." }, { "model_id": "deepseek-v4-flash", "benchmark_id": "skills_bench_text_79", "score": 35.4, "reference_url": "https://huggingface.co/tencent/Hy3", "reported_setting": { "mode": "thinking", "effort": "high / highest available", "tools": "Claude Code", "sampling": "average over 3 runs", "judge": "benchmark-specified", "harness": "79-task self-contained text-only subset; multimodal excluded", "prompt_style": "default", "temperature": "0.9 for Hy3 family; otherwise source/provider setting", "context": "256000 for Hy3 family; otherwise source/provider setting", "notes": "Tencent own test" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Hy3 official model-card appendix: SkillsBench (79, text-only) = 35.4*. Tencent own testing." }, { "model_id": "deepseek-v4-pro", "benchmark_id": "skills_bench_text_79", "score": 40.5, "reference_url": "https://huggingface.co/tencent/Hy3", "reported_setting": { "mode": "thinking", "effort": "high / highest available", "tools": "Claude Code", "sampling": "average over 3 runs", "judge": "benchmark-specified", "harness": "79-task self-contained text-only subset; multimodal excluded", "prompt_style": "default", "temperature": "0.9 for Hy3 family; otherwise source/provider setting", "context": "256000 for Hy3 family; otherwise source/provider setting", "notes": "Tencent own test" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Hy3 official model-card appendix: SkillsBench (79, text-only) = 40.5*. Tencent own testing." }, { "model_id": "doubao-seed-2.1-pro", "benchmark_id": "skills_bench_text_79", "score": 53.2, "reference_url": "https://huggingface.co/tencent/Hy3", "reported_setting": { "mode": "thinking", "effort": "high / highest available", "tools": "Claude Code", "sampling": "average over 3 runs", "judge": "benchmark-specified", "harness": "79-task self-contained text-only subset; multimodal excluded", "prompt_style": "default", "temperature": "0.9 for Hy3 family; otherwise source/provider setting", "context": "256000 for Hy3 family; otherwise source/provider setting", "notes": "Tencent own test" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Hy3 official model-card appendix: SkillsBench (79, text-only) = 53.2*. Tencent own testing." }, { "model_id": "qwen3.7-max", "benchmark_id": "skills_bench_text_79", "score": 59.2, "reference_url": "https://huggingface.co/tencent/Hy3", "reported_setting": { "mode": "thinking", "effort": "high / highest available", "tools": "Claude Code", "sampling": "average over 3 runs", "judge": "benchmark-specified", "harness": "79-task self-contained text-only subset; multimodal excluded", "prompt_style": "default", "temperature": "0.9 for Hy3 family; otherwise source/provider setting", "context": "256000 for Hy3 family; otherwise source/provider setting", "notes": "Comparator value not marked as Tencent own test" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Hy3 official model-card appendix: SkillsBench (79, text-only) = 59.2. Comparator-reported value.", "candidates": [ { "score": 46.8, "reference_url": "https://huggingface.co/tencent/Hy3", "source_type": "model_card", "reported_setting": { "mode": "thinking", "effort": "high / highest available", "tools": "Claude Code", "sampling": "average over 3 runs", "judge": "benchmark-specified", "harness": "79-task self-contained text-only subset; multimodal excluded", "prompt_style": "default", "temperature": "0.9 for Hy3 family; otherwise source/provider setting", "context": "256000 for Hy3 family; otherwise source/provider setting", "notes": "Tencent own test" }, "notes": "Hy3 official model-card appendix: SkillsBench (79, text-only) = 46.8*. Tencent own testing." } ] }, { "model_id": "claude-opus-4.8", "benchmark_id": "skills_bench_text_79", "score": 64.6, "reference_url": "https://huggingface.co/tencent/Hy3", "reported_setting": { "mode": "thinking", "effort": "high / highest available", "tools": "Claude Code", "sampling": "average over 3 runs", "judge": "benchmark-specified", "harness": "79-task self-contained text-only subset; multimodal excluded", "prompt_style": "default", "temperature": "0.9 for Hy3 family; otherwise source/provider setting", "context": "256000 for Hy3 family; otherwise source/provider setting", "notes": "Tencent own test" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Hy3 official model-card appendix: SkillsBench (79, text-only) = 64.6*. Tencent own testing." }, { "model_id": "gpt-5.5", "benchmark_id": "skills_bench_text_79", "score": 61.6, "reference_url": "https://huggingface.co/tencent/Hy3", "reported_setting": { "mode": "thinking", "effort": "high / highest available", "tools": "Claude Code", "sampling": "average over 3 runs", "judge": "benchmark-specified", "harness": "79-task self-contained text-only subset; multimodal excluded", "prompt_style": "default", "temperature": "0.9 for Hy3 family; otherwise source/provider setting", "context": "256000 for Hy3 family; otherwise source/provider setting", "notes": "Tencent own test" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Hy3 official model-card appendix: SkillsBench (79, text-only) = 61.6*. Tencent own testing." }, { "model_id": "hy3-preview", "benchmark_id": "e_bench_internal", "score": 10.9, "reference_url": "https://huggingface.co/tencent/Hy3", "reported_setting": { "mode": "thinking", "effort": "high / highest available", "tools": "benchmark-specific agent tools", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "Tencent internal working-agent benchmark; public protocol not published", "prompt_style": "default", "temperature": "0.9 for Hy3 family; otherwise source/provider setting", "context": "256000 for Hy3 family; otherwise source/provider setting", "notes": "Comparator value not marked as Tencent own test" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Hy3 official model-card appendix: e-bench (internal) = 10.9. Comparator-reported value." }, { "model_id": "hy3", "benchmark_id": "e_bench_internal", "score": 50.2, "reference_url": "https://huggingface.co/tencent/Hy3", "reported_setting": { "mode": "thinking", "effort": "high / highest available", "tools": "benchmark-specific agent tools", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "Tencent internal working-agent benchmark; public protocol not published", "prompt_style": "default", "temperature": "0.9 for Hy3 family; otherwise source/provider setting", "context": "256000 for Hy3 family; otherwise source/provider setting", "notes": "Comparator value not marked as Tencent own test" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Hy3 official model-card appendix: e-bench (internal) = 50.2. Comparator-reported value." }, { "model_id": "glm-5.1", "benchmark_id": "e_bench_internal", "score": 41.0, "reference_url": "https://huggingface.co/tencent/Hy3", "reported_setting": { "mode": "thinking", "effort": "high / highest available", "tools": "benchmark-specific agent tools", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "Tencent internal working-agent benchmark; public protocol not published", "prompt_style": "default", "temperature": "0.9 for Hy3 family; otherwise source/provider setting", "context": "256000 for Hy3 family; otherwise source/provider setting", "notes": "Tencent own test" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Hy3 official model-card appendix: e-bench (internal) = 41.0*. Tencent own testing." }, { "model_id": "glm-5.2", "benchmark_id": "e_bench_internal", "score": 52.3, "reference_url": "https://huggingface.co/tencent/Hy3", "reported_setting": { "mode": "thinking", "effort": "high / highest available", "tools": "benchmark-specific agent tools", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "Tencent internal working-agent benchmark; public protocol not published", "prompt_style": "default", "temperature": "0.9 for Hy3 family; otherwise source/provider setting", "context": "256000 for Hy3 family; otherwise source/provider setting", "notes": "Tencent own test" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Hy3 official model-card appendix: e-bench (internal) = 52.3*. Tencent own testing." }, { "model_id": "deepseek-v4-flash", "benchmark_id": "e_bench_internal", "score": 31.5, "reference_url": "https://huggingface.co/tencent/Hy3", "reported_setting": { "mode": "thinking", "effort": "high / highest available", "tools": "benchmark-specific agent tools", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "Tencent internal working-agent benchmark; public protocol not published", "prompt_style": "default", "temperature": "0.9 for Hy3 family; otherwise source/provider setting", "context": "256000 for Hy3 family; otherwise source/provider setting", "notes": "Tencent own test" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Hy3 official model-card appendix: e-bench (internal) = 31.5*. Tencent own testing." }, { "model_id": "deepseek-v4-pro", "benchmark_id": "e_bench_internal", "score": 34.5, "reference_url": "https://huggingface.co/tencent/Hy3", "reported_setting": { "mode": "thinking", "effort": "high / highest available", "tools": "benchmark-specific agent tools", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "Tencent internal working-agent benchmark; public protocol not published", "prompt_style": "default", "temperature": "0.9 for Hy3 family; otherwise source/provider setting", "context": "256000 for Hy3 family; otherwise source/provider setting", "notes": "Tencent own test" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Hy3 official model-card appendix: e-bench (internal) = 34.5*. Tencent own testing." }, { "model_id": "doubao-seed-2.1-pro", "benchmark_id": "e_bench_internal", "score": 47.9, "reference_url": "https://huggingface.co/tencent/Hy3", "reported_setting": { "mode": "thinking", "effort": "high / highest available", "tools": "benchmark-specific agent tools", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "Tencent internal working-agent benchmark; public protocol not published", "prompt_style": "default", "temperature": "0.9 for Hy3 family; otherwise source/provider setting", "context": "256000 for Hy3 family; otherwise source/provider setting", "notes": "Tencent own test" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Hy3 official model-card appendix: e-bench (internal) = 47.9*. Tencent own testing." }, { "model_id": "qwen3.7-max", "benchmark_id": "e_bench_internal", "score": 50.9, "reference_url": "https://huggingface.co/tencent/Hy3", "reported_setting": { "mode": "thinking", "effort": "high / highest available", "tools": "benchmark-specific agent tools", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "Tencent internal working-agent benchmark; public protocol not published", "prompt_style": "default", "temperature": "0.9 for Hy3 family; otherwise source/provider setting", "context": "256000 for Hy3 family; otherwise source/provider setting", "notes": "Tencent own test" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Hy3 official model-card appendix: e-bench (internal) = 50.9*. Tencent own testing." }, { "model_id": "gemini-3.1-pro", "benchmark_id": "e_bench_internal", "score": 42.6, "reference_url": "https://huggingface.co/tencent/Hy3", "reported_setting": { "mode": "thinking", "effort": "high / highest available", "tools": "benchmark-specific agent tools", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "Tencent internal working-agent benchmark; public protocol not published", "prompt_style": "default", "temperature": "0.9 for Hy3 family; otherwise source/provider setting", "context": "256000 for Hy3 family; otherwise source/provider setting", "notes": "Tencent own test" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Hy3 official model-card appendix: e-bench (internal) = 42.6*. Tencent own testing." }, { "model_id": "claude-opus-4.8", "benchmark_id": "e_bench_internal", "score": 68.8, "reference_url": "https://huggingface.co/tencent/Hy3", "reported_setting": { "mode": "thinking", "effort": "high / highest available", "tools": "benchmark-specific agent tools", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "Tencent internal working-agent benchmark; public protocol not published", "prompt_style": "default", "temperature": "0.9 for Hy3 family; otherwise source/provider setting", "context": "256000 for Hy3 family; otherwise source/provider setting", "notes": "Tencent own test" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Hy3 official model-card appendix: e-bench (internal) = 68.8*. Tencent own testing." }, { "model_id": "gpt-5.5", "benchmark_id": "e_bench_internal", "score": 72.0, "reference_url": "https://huggingface.co/tencent/Hy3", "reported_setting": { "mode": "thinking", "effort": "high / highest available", "tools": "benchmark-specific agent tools", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "Tencent internal working-agent benchmark; public protocol not published", "prompt_style": "default", "temperature": "0.9 for Hy3 family; otherwise source/provider setting", "context": "256000 for Hy3 family; otherwise source/provider setting", "notes": "Tencent own test" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Hy3 official model-card appendix: e-bench (internal) = 72.0*. Tencent own testing." }, { "model_id": "hy3-preview", "benchmark_id": "hy_finmodel_bench", "score": 26.6, "reference_url": "https://huggingface.co/tencent/Hy3", "reported_setting": { "mode": "thinking", "effort": "high / highest available", "tools": "benchmark-specific agent tools", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "Tencent internal working-agent benchmark; public protocol not published", "prompt_style": "default", "temperature": "0.9 for Hy3 family; otherwise source/provider setting", "context": "256000 for Hy3 family; otherwise source/provider setting", "notes": "Comparator value not marked as Tencent own test" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Hy3 official model-card appendix: Hy-FinModelBench (internal) = 26.6. Comparator-reported value." }, { "model_id": "hy3", "benchmark_id": "hy_finmodel_bench", "score": 69.0, "reference_url": "https://huggingface.co/tencent/Hy3", "reported_setting": { "mode": "thinking", "effort": "high / highest available", "tools": "benchmark-specific agent tools", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "Tencent internal working-agent benchmark; public protocol not published", "prompt_style": "default", "temperature": "0.9 for Hy3 family; otherwise source/provider setting", "context": "256000 for Hy3 family; otherwise source/provider setting", "notes": "Comparator value not marked as Tencent own test" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Hy3 official model-card appendix: Hy-FinModelBench (internal) = 69.0. Comparator-reported value." }, { "model_id": "glm-5.1", "benchmark_id": "hy_finmodel_bench", "score": 56.3, "reference_url": "https://huggingface.co/tencent/Hy3", "reported_setting": { "mode": "thinking", "effort": "high / highest available", "tools": "benchmark-specific agent tools", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "Tencent internal working-agent benchmark; public protocol not published", "prompt_style": "default", "temperature": "0.9 for Hy3 family; otherwise source/provider setting", "context": "256000 for Hy3 family; otherwise source/provider setting", "notes": "Tencent own test" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Hy3 official model-card appendix: Hy-FinModelBench (internal) = 56.3*. Tencent own testing." }, { "model_id": "glm-5.2", "benchmark_id": "hy_finmodel_bench", "score": 77.2, "reference_url": "https://huggingface.co/tencent/Hy3", "reported_setting": { "mode": "thinking", "effort": "high / highest available", "tools": "benchmark-specific agent tools", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "Tencent internal working-agent benchmark; public protocol not published", "prompt_style": "default", "temperature": "0.9 for Hy3 family; otherwise source/provider setting", "context": "256000 for Hy3 family; otherwise source/provider setting", "notes": "Tencent own test" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Hy3 official model-card appendix: Hy-FinModelBench (internal) = 77.2*. Tencent own testing." }, { "model_id": "deepseek-v4-flash", "benchmark_id": "hy_finmodel_bench", "score": 54.5, "reference_url": "https://huggingface.co/tencent/Hy3", "reported_setting": { "mode": "thinking", "effort": "high / highest available", "tools": "benchmark-specific agent tools", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "Tencent internal working-agent benchmark; public protocol not published", "prompt_style": "default", "temperature": "0.9 for Hy3 family; otherwise source/provider setting", "context": "256000 for Hy3 family; otherwise source/provider setting", "notes": "Tencent own test" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Hy3 official model-card appendix: Hy-FinModelBench (internal) = 54.5*. Tencent own testing." }, { "model_id": "deepseek-v4-pro", "benchmark_id": "hy_finmodel_bench", "score": 57.6, "reference_url": "https://huggingface.co/tencent/Hy3", "reported_setting": { "mode": "thinking", "effort": "high / highest available", "tools": "benchmark-specific agent tools", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "Tencent internal working-agent benchmark; public protocol not published", "prompt_style": "default", "temperature": "0.9 for Hy3 family; otherwise source/provider setting", "context": "256000 for Hy3 family; otherwise source/provider setting", "notes": "Tencent own test" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Hy3 official model-card appendix: Hy-FinModelBench (internal) = 57.6*. Tencent own testing." }, { "model_id": "doubao-seed-2.1-pro", "benchmark_id": "hy_finmodel_bench", "score": 52.2, "reference_url": "https://huggingface.co/tencent/Hy3", "reported_setting": { "mode": "thinking", "effort": "high / highest available", "tools": "benchmark-specific agent tools", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "Tencent internal working-agent benchmark; public protocol not published", "prompt_style": "default", "temperature": "0.9 for Hy3 family; otherwise source/provider setting", "context": "256000 for Hy3 family; otherwise source/provider setting", "notes": "Tencent own test" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Hy3 official model-card appendix: Hy-FinModelBench (internal) = 52.2*. Tencent own testing." }, { "model_id": "qwen3.7-max", "benchmark_id": "hy_finmodel_bench", "score": 57.3, "reference_url": "https://huggingface.co/tencent/Hy3", "reported_setting": { "mode": "thinking", "effort": "high / highest available", "tools": "benchmark-specific agent tools", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "Tencent internal working-agent benchmark; public protocol not published", "prompt_style": "default", "temperature": "0.9 for Hy3 family; otherwise source/provider setting", "context": "256000 for Hy3 family; otherwise source/provider setting", "notes": "Tencent own test" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Hy3 official model-card appendix: Hy-FinModelBench (internal) = 57.3*. Tencent own testing." }, { "model_id": "gemini-3.1-pro", "benchmark_id": "hy_finmodel_bench", "score": 54.6, "reference_url": "https://huggingface.co/tencent/Hy3", "reported_setting": { "mode": "thinking", "effort": "high / highest available", "tools": "benchmark-specific agent tools", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "Tencent internal working-agent benchmark; public protocol not published", "prompt_style": "default", "temperature": "0.9 for Hy3 family; otherwise source/provider setting", "context": "256000 for Hy3 family; otherwise source/provider setting", "notes": "Tencent own test" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Hy3 official model-card appendix: Hy-FinModelBench (internal) = 54.6*. Tencent own testing." }, { "model_id": "claude-opus-4.8", "benchmark_id": "hy_finmodel_bench", "score": 74.3, "reference_url": "https://huggingface.co/tencent/Hy3", "reported_setting": { "mode": "thinking", "effort": "high / highest available", "tools": "benchmark-specific agent tools", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "Tencent internal working-agent benchmark; public protocol not published", "prompt_style": "default", "temperature": "0.9 for Hy3 family; otherwise source/provider setting", "context": "256000 for Hy3 family; otherwise source/provider setting", "notes": "Tencent own test" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Hy3 official model-card appendix: Hy-FinModelBench (internal) = 74.3*. Tencent own testing." }, { "model_id": "gpt-5.5", "benchmark_id": "hy_finmodel_bench", "score": 85.1, "reference_url": "https://huggingface.co/tencent/Hy3", "reported_setting": { "mode": "thinking", "effort": "high / highest available", "tools": "benchmark-specific agent tools", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "Tencent internal working-agent benchmark; public protocol not published", "prompt_style": "default", "temperature": "0.9 for Hy3 family; otherwise source/provider setting", "context": "256000 for Hy3 family; otherwise source/provider setting", "notes": "Tencent own test" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Hy3 official model-card appendix: Hy-FinModelBench (internal) = 85.1*. Tencent own testing." }, { "model_id": "hy3-preview", "benchmark_id": "prod_bench_internal", "score": 10.3, "reference_url": "https://huggingface.co/tencent/Hy3", "reported_setting": { "mode": "thinking", "effort": "high / highest available", "tools": "OpenClaw harness", "sampling": "pass^3", "judge": "benchmark-specified", "harness": "OpenClaw harness", "prompt_style": "default", "temperature": "0.9 for Hy3 family; otherwise source/provider setting", "context": "256000 for Hy3 family; otherwise source/provider setting", "notes": "Comparator value not marked as Tencent own test" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Hy3 official model-card appendix: ProdBench (internal, pass^3) = 10.3. Comparator-reported value." }, { "model_id": "hy3", "benchmark_id": "prod_bench_internal", "score": 23.0, "reference_url": "https://huggingface.co/tencent/Hy3", "reported_setting": { "mode": "thinking", "effort": "high / highest available", "tools": "OpenClaw harness", "sampling": "pass^3", "judge": "benchmark-specified", "harness": "OpenClaw harness", "prompt_style": "default", "temperature": "0.9 for Hy3 family; otherwise source/provider setting", "context": "256000 for Hy3 family; otherwise source/provider setting", "notes": "Comparator value not marked as Tencent own test" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Hy3 official model-card appendix: ProdBench (internal, pass^3) = 23.0. Comparator-reported value." }, { "model_id": "glm-5.1", "benchmark_id": "prod_bench_internal", "score": 17.5, "reference_url": "https://huggingface.co/tencent/Hy3", "reported_setting": { "mode": "thinking", "effort": "high / highest available", "tools": "OpenClaw harness", "sampling": "pass^3", "judge": "benchmark-specified", "harness": "OpenClaw harness", "prompt_style": "default", "temperature": "0.9 for Hy3 family; otherwise source/provider setting", "context": "256000 for Hy3 family; otherwise source/provider setting", "notes": "Tencent own test" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Hy3 official model-card appendix: ProdBench (internal, pass^3) = 17.5*. Tencent own testing." }, { "model_id": "glm-5.2", "benchmark_id": "prod_bench_internal", "score": 21.4, "reference_url": "https://huggingface.co/tencent/Hy3", "reported_setting": { "mode": "thinking", "effort": "high / highest available", "tools": "OpenClaw harness", "sampling": "pass^3", "judge": "benchmark-specified", "harness": "OpenClaw harness", "prompt_style": "default", "temperature": "0.9 for Hy3 family; otherwise source/provider setting", "context": "256000 for Hy3 family; otherwise source/provider setting", "notes": "Tencent own test" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Hy3 official model-card appendix: ProdBench (internal, pass^3) = 21.4*. Tencent own testing." }, { "model_id": "deepseek-v4-flash", "benchmark_id": "prod_bench_internal", "score": 10.3, "reference_url": "https://huggingface.co/tencent/Hy3", "reported_setting": { "mode": "thinking", "effort": "high / highest available", "tools": "OpenClaw harness", "sampling": "pass^3", "judge": "benchmark-specified", "harness": "OpenClaw harness", "prompt_style": "default", "temperature": "0.9 for Hy3 family; otherwise source/provider setting", "context": "256000 for Hy3 family; otherwise source/provider setting", "notes": "Tencent own test" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Hy3 official model-card appendix: ProdBench (internal, pass^3) = 10.3*. Tencent own testing." }, { "model_id": "deepseek-v4-pro", "benchmark_id": "prod_bench_internal", "score": 15.1, "reference_url": "https://huggingface.co/tencent/Hy3", "reported_setting": { "mode": "thinking", "effort": "high / highest available", "tools": "OpenClaw harness", "sampling": "pass^3", "judge": "benchmark-specified", "harness": "OpenClaw harness", "prompt_style": "default", "temperature": "0.9 for Hy3 family; otherwise source/provider setting", "context": "256000 for Hy3 family; otherwise source/provider setting", "notes": "Tencent own test" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Hy3 official model-card appendix: ProdBench (internal, pass^3) = 15.1*. Tencent own testing." }, { "model_id": "doubao-seed-2.1-pro", "benchmark_id": "prod_bench_internal", "score": 18.3, "reference_url": "https://huggingface.co/tencent/Hy3", "reported_setting": { "mode": "thinking", "effort": "high / highest available", "tools": "OpenClaw harness", "sampling": "pass^3", "judge": "benchmark-specified", "harness": "OpenClaw harness", "prompt_style": "default", "temperature": "0.9 for Hy3 family; otherwise source/provider setting", "context": "256000 for Hy3 family; otherwise source/provider setting", "notes": "Tencent own test" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Hy3 official model-card appendix: ProdBench (internal, pass^3) = 18.3*. Tencent own testing." }, { "model_id": "qwen3.7-max", "benchmark_id": "prod_bench_internal", "score": 15.9, "reference_url": "https://huggingface.co/tencent/Hy3", "reported_setting": { "mode": "thinking", "effort": "high / highest available", "tools": "OpenClaw harness", "sampling": "pass^3", "judge": "benchmark-specified", "harness": "OpenClaw harness", "prompt_style": "default", "temperature": "0.9 for Hy3 family; otherwise source/provider setting", "context": "256000 for Hy3 family; otherwise source/provider setting", "notes": "Tencent own test" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Hy3 official model-card appendix: ProdBench (internal, pass^3) = 15.9*. Tencent own testing." }, { "model_id": "claude-opus-4.8", "benchmark_id": "prod_bench_internal", "score": 27.8, "reference_url": "https://huggingface.co/tencent/Hy3", "reported_setting": { "mode": "thinking", "effort": "high / highest available", "tools": "OpenClaw harness", "sampling": "pass^3", "judge": "benchmark-specified", "harness": "OpenClaw harness", "prompt_style": "default", "temperature": "0.9 for Hy3 family; otherwise source/provider setting", "context": "256000 for Hy3 family; otherwise source/provider setting", "notes": "Tencent own test" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Hy3 official model-card appendix: ProdBench (internal, pass^3) = 27.8*. Tencent own testing." }, { "model_id": "gpt-5.5", "benchmark_id": "prod_bench_internal", "score": 21.4, "reference_url": "https://huggingface.co/tencent/Hy3", "reported_setting": { "mode": "thinking", "effort": "high / highest available", "tools": "OpenClaw harness", "sampling": "pass^3", "judge": "benchmark-specified", "harness": "OpenClaw harness", "prompt_style": "default", "temperature": "0.9 for Hy3 family; otherwise source/provider setting", "context": "256000 for Hy3 family; otherwise source/provider setting", "notes": "Tencent own test" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Hy3 official model-card appendix: ProdBench (internal, pass^3) = 21.4*. Tencent own testing." }, { "model_id": "hy3-preview", "benchmark_id": "hy_skillsworld", "score": 26.4, "reference_url": "https://huggingface.co/tencent/Hy3", "reported_setting": { "mode": "thinking", "effort": "high / highest available", "tools": "benchmark-specific agent tools", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "Tencent internal working-agent benchmark; public protocol not published", "prompt_style": "default", "temperature": "0.9 for Hy3 family; otherwise source/provider setting", "context": "256000 for Hy3 family; otherwise source/provider setting", "notes": "Comparator value not marked as Tencent own test" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Hy3 official model-card appendix: Hy-SkillsWorld (internal) = 26.4. Comparator-reported value." }, { "model_id": "hy3", "benchmark_id": "hy_skillsworld", "score": 45.8, "reference_url": "https://huggingface.co/tencent/Hy3", "reported_setting": { "mode": "thinking", "effort": "high / highest available", "tools": "benchmark-specific agent tools", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "Tencent internal working-agent benchmark; public protocol not published", "prompt_style": "default", "temperature": "0.9 for Hy3 family; otherwise source/provider setting", "context": "256000 for Hy3 family; otherwise source/provider setting", "notes": "Comparator value not marked as Tencent own test" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Hy3 official model-card appendix: Hy-SkillsWorld (internal) = 45.8. Comparator-reported value." }, { "model_id": "glm-5.1", "benchmark_id": "hy_skillsworld", "score": 37.5, "reference_url": "https://huggingface.co/tencent/Hy3", "reported_setting": { "mode": "thinking", "effort": "high / highest available", "tools": "benchmark-specific agent tools", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "Tencent internal working-agent benchmark; public protocol not published", "prompt_style": "default", "temperature": "0.9 for Hy3 family; otherwise source/provider setting", "context": "256000 for Hy3 family; otherwise source/provider setting", "notes": "Tencent own test" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Hy3 official model-card appendix: Hy-SkillsWorld (internal) = 37.5*. Tencent own testing." }, { "model_id": "glm-5.2", "benchmark_id": "hy_skillsworld", "score": 56.9, "reference_url": "https://huggingface.co/tencent/Hy3", "reported_setting": { "mode": "thinking", "effort": "high / highest available", "tools": "benchmark-specific agent tools", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "Tencent internal working-agent benchmark; public protocol not published", "prompt_style": "default", "temperature": "0.9 for Hy3 family; otherwise source/provider setting", "context": "256000 for Hy3 family; otherwise source/provider setting", "notes": "Tencent own test" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Hy3 official model-card appendix: Hy-SkillsWorld (internal) = 56.9*. Tencent own testing." }, { "model_id": "deepseek-v4-flash", "benchmark_id": "hy_skillsworld", "score": 38.9, "reference_url": "https://huggingface.co/tencent/Hy3", "reported_setting": { "mode": "thinking", "effort": "high / highest available", "tools": "benchmark-specific agent tools", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "Tencent internal working-agent benchmark; public protocol not published", "prompt_style": "default", "temperature": "0.9 for Hy3 family; otherwise source/provider setting", "context": "256000 for Hy3 family; otherwise source/provider setting", "notes": "Tencent own test" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Hy3 official model-card appendix: Hy-SkillsWorld (internal) = 38.9*. Tencent own testing." }, { "model_id": "deepseek-v4-pro", "benchmark_id": "hy_skillsworld", "score": 41.7, "reference_url": "https://huggingface.co/tencent/Hy3", "reported_setting": { "mode": "thinking", "effort": "high / highest available", "tools": "benchmark-specific agent tools", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "Tencent internal working-agent benchmark; public protocol not published", "prompt_style": "default", "temperature": "0.9 for Hy3 family; otherwise source/provider setting", "context": "256000 for Hy3 family; otherwise source/provider setting", "notes": "Tencent own test" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Hy3 official model-card appendix: Hy-SkillsWorld (internal) = 41.7*. Tencent own testing." }, { "model_id": "doubao-seed-2.1-pro", "benchmark_id": "hy_skillsworld", "score": 48.6, "reference_url": "https://huggingface.co/tencent/Hy3", "reported_setting": { "mode": "thinking", "effort": "high / highest available", "tools": "benchmark-specific agent tools", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "Tencent internal working-agent benchmark; public protocol not published", "prompt_style": "default", "temperature": "0.9 for Hy3 family; otherwise source/provider setting", "context": "256000 for Hy3 family; otherwise source/provider setting", "notes": "Tencent own test" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Hy3 official model-card appendix: Hy-SkillsWorld (internal) = 48.6*. Tencent own testing." }, { "model_id": "qwen3.7-max", "benchmark_id": "hy_skillsworld", "score": 42.4, "reference_url": "https://huggingface.co/tencent/Hy3", "reported_setting": { "mode": "thinking", "effort": "high / highest available", "tools": "benchmark-specific agent tools", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "Tencent internal working-agent benchmark; public protocol not published", "prompt_style": "default", "temperature": "0.9 for Hy3 family; otherwise source/provider setting", "context": "256000 for Hy3 family; otherwise source/provider setting", "notes": "Tencent own test" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Hy3 official model-card appendix: Hy-SkillsWorld (internal) = 42.4*. Tencent own testing." }, { "model_id": "claude-opus-4.8", "benchmark_id": "hy_skillsworld", "score": 59.7, "reference_url": "https://huggingface.co/tencent/Hy3", "reported_setting": { "mode": "thinking", "effort": "high / highest available", "tools": "benchmark-specific agent tools", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "Tencent internal working-agent benchmark; public protocol not published", "prompt_style": "default", "temperature": "0.9 for Hy3 family; otherwise source/provider setting", "context": "256000 for Hy3 family; otherwise source/provider setting", "notes": "Tencent own test" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Hy3 official model-card appendix: Hy-SkillsWorld (internal) = 59.7*. Tencent own testing." }, { "model_id": "gpt-5.5", "benchmark_id": "hy_skillsworld", "score": 56.9, "reference_url": "https://huggingface.co/tencent/Hy3", "reported_setting": { "mode": "thinking", "effort": "high / highest available", "tools": "benchmark-specific agent tools", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "Tencent internal working-agent benchmark; public protocol not published", "prompt_style": "default", "temperature": "0.9 for Hy3 family; otherwise source/provider setting", "context": "256000 for Hy3 family; otherwise source/provider setting", "notes": "Tencent own test" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Hy3 official model-card appendix: Hy-SkillsWorld (internal) = 56.9*. Tencent own testing." }, { "model_id": "hy3-preview", "benchmark_id": "hle_tools_text", "score": 35.4, "reference_url": "https://huggingface.co/tencent/Hy3", "reported_setting": { "mode": "thinking", "effort": "high / highest available", "tools": "search/code/web or benchmark-specific tools", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official benchmark or reported provider harness", "prompt_style": "default", "temperature": "0.9 for Hy3 family; otherwise source/provider setting", "context": "256000 for Hy3 family; otherwise source/provider setting", "notes": "Comparator value not marked as Tencent own test" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Hy3 official model-card appendix: HLE (with tools, text-only) = 35.4. Comparator-reported value." }, { "model_id": "hy3", "benchmark_id": "hle_tools_text", "score": 53.2, "reference_url": "https://huggingface.co/tencent/Hy3", "reported_setting": { "mode": "thinking", "effort": "high / highest available", "tools": "search/code/web or benchmark-specific tools", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official benchmark or reported provider harness", "prompt_style": "default", "temperature": "0.9 for Hy3 family; otherwise source/provider setting", "context": "256000 for Hy3 family; otherwise source/provider setting", "notes": "Comparator value not marked as Tencent own test" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Hy3 official model-card appendix: HLE (with tools, text-only) = 53.2. Comparator-reported value." }, { "model_id": "deepseek-v4-flash", "benchmark_id": "hle_tools_text", "score": 45.1, "reference_url": "https://huggingface.co/tencent/Hy3", "reported_setting": { "mode": "thinking", "effort": "high / highest available", "tools": "search/code/web or benchmark-specific tools", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official benchmark or reported provider harness", "prompt_style": "default", "temperature": "0.9 for Hy3 family; otherwise source/provider setting", "context": "256000 for Hy3 family; otherwise source/provider setting", "notes": "Comparator value not marked as Tencent own test" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Hy3 official model-card appendix: HLE (with tools, text-only) = 45.1. Comparator-reported value." }, { "model_id": "doubao-seed-2.1-pro", "benchmark_id": "hle_tools_text", "score": 55.7, "reference_url": "https://seed.bytedance.com/en/blog/seed2-1-officially-released-advancing-ai-productivity", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "search", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official", "prompt_style": "official default", "temperature": "unknown" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "Release blog EN: Language/search chart (https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/user-upload/4xfa4mqq510uc.png).", "candidates": [ { "score": 55.7, "reference_url": "https://huggingface.co/tencent/Hy3", "source_type": "model_card", "reported_setting": { "mode": "thinking", "effort": "high / highest available", "tools": "search/code/web or benchmark-specific tools", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official benchmark or reported provider harness", "prompt_style": "default", "temperature": "0.9 for Hy3 family; otherwise source/provider setting", "context": "256000 for Hy3 family; otherwise source/provider setting", "notes": "Comparator value not marked as Tencent own test" }, "notes": "Hy3 official model-card appendix: HLE (with tools, text-only) = 55.7. Comparator-reported value." } ] }, { "model_id": "gemini-3.1-pro", "benchmark_id": "hle_tools_text", "score": 51.4, "reference_url": "https://huggingface.co/tencent/Hy3", "reported_setting": { "mode": "thinking", "effort": "high / highest available", "tools": "search/code/web or benchmark-specific tools", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official benchmark or reported provider harness", "prompt_style": "default", "temperature": "0.9 for Hy3 family; otherwise source/provider setting", "context": "256000 for Hy3 family; otherwise source/provider setting", "notes": "Comparator value not marked as Tencent own test" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Hy3 official model-card appendix: HLE (with tools, text-only) = 51.4. Comparator-reported value.", "candidates": [ { "score": 51.4, "reference_url": "https://seed.bytedance.com/en/blog/seed2-1-officially-released-advancing-ai-productivity", "source_type": "official_blog", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "search", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official", "prompt_style": "official default", "temperature": "unknown" }, "notes": "Release blog EN: Language/search chart (https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/user-upload/4xfa4mqq510uc.png)." } ] }, { "model_id": "claude-opus-4.8", "benchmark_id": "hle_tools_text", "score": 57.9, "reference_url": "https://huggingface.co/tencent/Hy3", "reported_setting": { "mode": "thinking", "effort": "high / highest available", "tools": "search/code/web or benchmark-specific tools", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official benchmark or reported provider harness", "prompt_style": "default", "temperature": "0.9 for Hy3 family; otherwise source/provider setting", "context": "256000 for Hy3 family; otherwise source/provider setting", "notes": "Comparator value not marked as Tencent own test" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Hy3 official model-card appendix: HLE (with tools, text-only) = 57.9. Comparator-reported value." }, { "model_id": "gpt-5.5", "benchmark_id": "hle_tools_text", "score": 52.2, "reference_url": "https://huggingface.co/tencent/Hy3", "reported_setting": { "mode": "thinking", "effort": "high / highest available", "tools": "search/code/web or benchmark-specific tools", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official benchmark or reported provider harness", "prompt_style": "default", "temperature": "0.9 for Hy3 family; otherwise source/provider setting", "context": "256000 for Hy3 family; otherwise source/provider setting", "notes": "Comparator value not marked as Tencent own test" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Hy3 official model-card appendix: HLE (with tools, text-only) = 52.2. Comparator-reported value.", "candidates": [ { "score": 52.2, "reference_url": "https://seed.bytedance.com/en/blog/seed2-1-officially-released-advancing-ai-productivity", "source_type": "official_blog", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "search", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official", "prompt_style": "official default", "temperature": "unknown" }, "notes": "Release blog EN: Language/search chart (https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/user-upload/4xfa4mqq510uc.png)." } ] }, { "model_id": "hy3-preview", "benchmark_id": "hy_euler_pro", "score": 2.3, "reference_url": "https://huggingface.co/tencent/Hy3", "reported_setting": { "mode": "thinking", "effort": "high / highest available", "tools": "search/code/web or benchmark-specific tools", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official benchmark or reported provider harness", "prompt_style": "default", "temperature": "0.9 for Hy3 family; otherwise source/provider setting", "context": "256000 for Hy3 family; otherwise source/provider setting", "notes": "Comparator value not marked as Tencent own test" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Hy3 official model-card appendix: Hy-Euler pro (with tools, internal) = 2.3. Comparator-reported value." }, { "model_id": "hy3", "benchmark_id": "hy_euler_pro", "score": 24.2, "reference_url": "https://huggingface.co/tencent/Hy3", "reported_setting": { "mode": "thinking", "effort": "high / highest available", "tools": "search/code/web or benchmark-specific tools", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official benchmark or reported provider harness", "prompt_style": "default", "temperature": "0.9 for Hy3 family; otherwise source/provider setting", "context": "256000 for Hy3 family; otherwise source/provider setting", "notes": "Comparator value not marked as Tencent own test" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Hy3 official model-card appendix: Hy-Euler pro (with tools, internal) = 24.2. Comparator-reported value." }, { "model_id": "glm-5.1", "benchmark_id": "hy_euler_pro", "score": 15.9, "reference_url": "https://huggingface.co/tencent/Hy3", "reported_setting": { "mode": "thinking", "effort": "high / highest available", "tools": "search/code/web or benchmark-specific tools", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official benchmark or reported provider harness", "prompt_style": "default", "temperature": "0.9 for Hy3 family; otherwise source/provider setting", "context": "256000 for Hy3 family; otherwise source/provider setting", "notes": "Tencent own test" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Hy3 official model-card appendix: Hy-Euler pro (with tools, internal) = 15.9*. Tencent own testing." }, { "model_id": "glm-5.2", "benchmark_id": "hy_euler_pro", "score": 42.4, "reference_url": "https://huggingface.co/tencent/Hy3", "reported_setting": { "mode": "thinking", "effort": "high / highest available", "tools": "search/code/web or benchmark-specific tools", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official benchmark or reported provider harness", "prompt_style": "default", "temperature": "0.9 for Hy3 family; otherwise source/provider setting", "context": "256000 for Hy3 family; otherwise source/provider setting", "notes": "Tencent own test" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Hy3 official model-card appendix: Hy-Euler pro (with tools, internal) = 42.4*. Tencent own testing." }, { "model_id": "deepseek-v4-flash", "benchmark_id": "hy_euler_pro", "score": 9.1, "reference_url": "https://huggingface.co/tencent/Hy3", "reported_setting": { "mode": "thinking", "effort": "high / highest available", "tools": "search/code/web or benchmark-specific tools", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official benchmark or reported provider harness", "prompt_style": "default", "temperature": "0.9 for Hy3 family; otherwise source/provider setting", "context": "256000 for Hy3 family; otherwise source/provider setting", "notes": "Tencent own test" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Hy3 official model-card appendix: Hy-Euler pro (with tools, internal) = 9.1*. Tencent own testing." }, { "model_id": "deepseek-v4-pro", "benchmark_id": "hy_euler_pro", "score": 25.0, "reference_url": "https://huggingface.co/tencent/Hy3", "reported_setting": { "mode": "thinking", "effort": "high / highest available", "tools": "search/code/web or benchmark-specific tools", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official benchmark or reported provider harness", "prompt_style": "default", "temperature": "0.9 for Hy3 family; otherwise source/provider setting", "context": "256000 for Hy3 family; otherwise source/provider setting", "notes": "Tencent own test" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Hy3 official model-card appendix: Hy-Euler pro (with tools, internal) = 25.0*. Tencent own testing." }, { "model_id": "doubao-seed-2.1-pro", "benchmark_id": "hy_euler_pro", "score": 15.2, "reference_url": "https://huggingface.co/tencent/Hy3", "reported_setting": { "mode": "thinking", "effort": "high / highest available", "tools": "search/code/web or benchmark-specific tools", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official benchmark or reported provider harness", "prompt_style": "default", "temperature": "0.9 for Hy3 family; otherwise source/provider setting", "context": "256000 for Hy3 family; otherwise source/provider setting", "notes": "Tencent own test" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Hy3 official model-card appendix: Hy-Euler pro (with tools, internal) = 15.2*. Tencent own testing." }, { "model_id": "qwen3.7-max", "benchmark_id": "hy_euler_pro", "score": 50.0, "reference_url": "https://huggingface.co/tencent/Hy3", "reported_setting": { "mode": "thinking", "effort": "high / highest available", "tools": "search/code/web or benchmark-specific tools", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official benchmark or reported provider harness", "prompt_style": "default", "temperature": "0.9 for Hy3 family; otherwise source/provider setting", "context": "256000 for Hy3 family; otherwise source/provider setting", "notes": "Tencent own test" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Hy3 official model-card appendix: Hy-Euler pro (with tools, internal) = 50.0*. Tencent own testing." }, { "model_id": "gemini-3.1-pro", "benchmark_id": "hy_euler_pro", "score": 54.6, "reference_url": "https://huggingface.co/tencent/Hy3", "reported_setting": { "mode": "thinking", "effort": "high / highest available", "tools": "search/code/web or benchmark-specific tools", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official benchmark or reported provider harness", "prompt_style": "default", "temperature": "0.9 for Hy3 family; otherwise source/provider setting", "context": "256000 for Hy3 family; otherwise source/provider setting", "notes": "Tencent own test" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Hy3 official model-card appendix: Hy-Euler pro (with tools, internal) = 54.6*. Tencent own testing." }, { "model_id": "claude-opus-4.8", "benchmark_id": "hy_euler_pro", "score": 64.4, "reference_url": "https://huggingface.co/tencent/Hy3", "reported_setting": { "mode": "thinking", "effort": "high / highest available", "tools": "search/code/web or benchmark-specific tools", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official benchmark or reported provider harness", "prompt_style": "default", "temperature": "0.9 for Hy3 family; otherwise source/provider setting", "context": "256000 for Hy3 family; otherwise source/provider setting", "notes": "Tencent own test" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Hy3 official model-card appendix: Hy-Euler pro (with tools, internal) = 64.4*. Tencent own testing." }, { "model_id": "gpt-5.5", "benchmark_id": "hy_euler_pro", "score": 77.3, "reference_url": "https://huggingface.co/tencent/Hy3", "reported_setting": { "mode": "thinking", "effort": "high / highest available", "tools": "search/code/web or benchmark-specific tools", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official benchmark or reported provider harness", "prompt_style": "default", "temperature": "0.9 for Hy3 family; otherwise source/provider setting", "context": "256000 for Hy3 family; otherwise source/provider setting", "notes": "Tencent own test" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Hy3 official model-card appendix: Hy-Euler pro (with tools, internal) = 77.3*. Tencent own testing." }, { "model_id": "hy3-preview", "benchmark_id": "gpqa_diamond", "score": 87.2, "reference_url": "https://huggingface.co/tencent/Hy3", "reported_setting": { "mode": "thinking", "effort": "high / highest available", "tools": "none", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official benchmark or reported provider harness", "prompt_style": "default", "temperature": "0.9 for Hy3 family; otherwise source/provider setting", "context": "256000 for Hy3 family; otherwise source/provider setting", "notes": "Comparator value not marked as Tencent own test" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Hy3 official model-card appendix: GPQA Diamond = 87.2. Comparator-reported value." }, { "model_id": "hy3", "benchmark_id": "gpqa_diamond", "score": 90.4, "reference_url": "https://huggingface.co/tencent/Hy3", "reported_setting": { "mode": "thinking", "effort": "high / highest available", "tools": "none", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official benchmark or reported provider harness", "prompt_style": "default", "temperature": "0.9 for Hy3 family; otherwise source/provider setting", "context": "256000 for Hy3 family; otherwise source/provider setting", "notes": "Comparator value not marked as Tencent own test" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Hy3 official model-card appendix: GPQA Diamond = 90.4. Comparator-reported value." }, { "model_id": "doubao-seed-2.1-pro", "benchmark_id": "gpqa_diamond", "score": 89.8, "reference_url": "https://huggingface.co/tencent/Hy3", "reported_setting": { "mode": "thinking", "effort": "high / highest available", "tools": "none", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official benchmark or reported provider harness", "prompt_style": "default", "temperature": "0.9 for Hy3 family; otherwise source/provider setting", "context": "256000 for Hy3 family; otherwise source/provider setting", "notes": "Tencent own test" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Hy3 official model-card appendix: GPQA Diamond = 89.8*. Tencent own testing." }, { "model_id": "hy3-preview", "benchmark_id": "hle_text", "score": 30.0, "reference_url": "https://huggingface.co/tencent/Hy3", "reported_setting": { "mode": "thinking", "effort": "high / highest available", "tools": "none", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official benchmark or reported provider harness", "prompt_style": "default", "temperature": "0.9 for Hy3 family; otherwise source/provider setting", "context": "256000 for Hy3 family; otherwise source/provider setting", "notes": "Comparator value not marked as Tencent own test" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Hy3 official model-card appendix: HLE (no tools, text-only) = 30.0. Comparator-reported value." }, { "model_id": "hy3", "benchmark_id": "hle_text", "score": 37.0, "reference_url": "https://huggingface.co/tencent/Hy3", "reported_setting": { "mode": "thinking", "effort": "high / highest available", "tools": "none", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official benchmark or reported provider harness", "prompt_style": "default", "temperature": "0.9 for Hy3 family; otherwise source/provider setting", "context": "256000 for Hy3 family; otherwise source/provider setting", "notes": "Comparator value not marked as Tencent own test" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Hy3 official model-card appendix: HLE (no tools, text-only) = 37.0. Comparator-reported value." }, { "model_id": "doubao-seed-2.1-pro", "benchmark_id": "hle_text", "score": 35.9, "reference_url": "https://huggingface.co/tencent/Hy3", "reported_setting": { "mode": "thinking", "effort": "high / highest available", "tools": "none", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official benchmark or reported provider harness", "prompt_style": "default", "temperature": "0.9 for Hy3 family; otherwise source/provider setting", "context": "256000 for Hy3 family; otherwise source/provider setting", "notes": "Tencent own test" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Hy3 official model-card appendix: HLE (no tools, text-only) = 35.9*. Tencent own testing." }, { "model_id": "claude-opus-4.8", "benchmark_id": "hle_text", "score": 49.8, "reference_url": "https://huggingface.co/tencent/Hy3", "reported_setting": { "mode": "thinking", "effort": "high / highest available", "tools": "none", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official benchmark or reported provider harness", "prompt_style": "default", "temperature": "0.9 for Hy3 family; otherwise source/provider setting", "context": "256000 for Hy3 family; otherwise source/provider setting", "notes": "Comparator value not marked as Tencent own test" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Hy3 official model-card appendix: HLE (no tools, text-only) = 49.8. Comparator-reported value." }, { "model_id": "hy3-preview", "benchmark_id": "frontier_science_research", "score": 19.0, "reference_url": "https://huggingface.co/tencent/Hy3", "reported_setting": { "mode": "thinking", "effort": "high / highest available", "tools": "none", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official benchmark or reported provider harness", "prompt_style": "default", "temperature": "0.9 for Hy3 family; otherwise source/provider setting", "context": "256000 for Hy3 family; otherwise source/provider setting", "notes": "Comparator value not marked as Tencent own test" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Hy3 official model-card appendix: FrontierScience-Research = 19.0. Comparator-reported value." }, { "model_id": "hy3", "benchmark_id": "frontier_science_research", "score": 21.3, "reference_url": "https://huggingface.co/tencent/Hy3", "reported_setting": { "mode": "thinking", "effort": "high / highest available", "tools": "none", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official benchmark or reported provider harness", "prompt_style": "default", "temperature": "0.9 for Hy3 family; otherwise source/provider setting", "context": "256000 for Hy3 family; otherwise source/provider setting", "notes": "Comparator value not marked as Tencent own test" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Hy3 official model-card appendix: FrontierScience-Research = 21.3. Comparator-reported value." }, { "model_id": "glm-5.1", "benchmark_id": "frontier_science_research", "score": 13.5, "reference_url": "https://huggingface.co/tencent/Hy3", "reported_setting": { "mode": "thinking", "effort": "high / highest available", "tools": "none", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official benchmark or reported provider harness", "prompt_style": "default", "temperature": "0.9 for Hy3 family; otherwise source/provider setting", "context": "256000 for Hy3 family; otherwise source/provider setting", "notes": "Tencent own test" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Hy3 official model-card appendix: FrontierScience-Research = 13.5*. Tencent own testing." }, { "model_id": "glm-5.2", "benchmark_id": "frontier_science_research", "score": 19.8, "reference_url": "https://huggingface.co/tencent/Hy3", "reported_setting": { "mode": "thinking", "effort": "high / highest available", "tools": "none", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official benchmark or reported provider harness", "prompt_style": "default", "temperature": "0.9 for Hy3 family; otherwise source/provider setting", "context": "256000 for Hy3 family; otherwise source/provider setting", "notes": "Tencent own test" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Hy3 official model-card appendix: FrontierScience-Research = 19.8*. Tencent own testing." }, { "model_id": "deepseek-v4-flash", "benchmark_id": "frontier_science_research", "score": 16.7, "reference_url": "https://huggingface.co/tencent/Hy3", "reported_setting": { "mode": "thinking", "effort": "high / highest available", "tools": "none", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official benchmark or reported provider harness", "prompt_style": "default", "temperature": "0.9 for Hy3 family; otherwise source/provider setting", "context": "256000 for Hy3 family; otherwise source/provider setting", "notes": "Tencent own test" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Hy3 official model-card appendix: FrontierScience-Research = 16.7*. Tencent own testing." }, { "model_id": "deepseek-v4-pro", "benchmark_id": "frontier_science_research", "score": 21.3, "reference_url": "https://huggingface.co/tencent/Hy3", "reported_setting": { "mode": "thinking", "effort": "high / highest available", "tools": "none", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official benchmark or reported provider harness", "prompt_style": "default", "temperature": "0.9 for Hy3 family; otherwise source/provider setting", "context": "256000 for Hy3 family; otherwise source/provider setting", "notes": "Tencent own test" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Hy3 official model-card appendix: FrontierScience-Research = 21.3*. Tencent own testing." }, { "model_id": "doubao-seed-2.1-pro", "benchmark_id": "frontier_science_research", "score": 28.3, "reference_url": "https://seed.bytedance.com/en/blog/seed2-1-officially-released-advancing-ai-productivity", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "research-oriented agent environment", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official", "prompt_style": "official default", "temperature": "unknown" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "Release blog EN: Frontier-research chart (https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/user-upload/4xfa4mqq52cdr.png).", "candidates": [ { "score": 25.6, "reference_url": "https://huggingface.co/tencent/Hy3", "source_type": "model_card", "reported_setting": { "mode": "thinking", "effort": "high / highest available", "tools": "none", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official benchmark or reported provider harness", "prompt_style": "default", "temperature": "0.9 for Hy3 family; otherwise source/provider setting", "context": "256000 for Hy3 family; otherwise source/provider setting", "notes": "Tencent own test" }, "notes": "Hy3 official model-card appendix: FrontierScience-Research = 25.6*. Tencent own testing." }, { "score": 28.3, "reference_url": "https://huggingface.co/tencent/Hy3", "source_type": "model_card", "reported_setting": { "mode": "thinking", "effort": "high / highest available", "tools": "none", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official benchmark or reported provider harness", "prompt_style": "default", "temperature": "0.9 for Hy3 family; otherwise source/provider setting", "context": "256000 for Hy3 family; otherwise source/provider setting", "notes": "Comparator value not marked as Tencent own test" }, "notes": "Hy3 official model-card appendix: FrontierScience-Research = 28.3. Comparator-reported value." }, { "score": 33.3, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2.1/Seed2_1_Model_Card.pdf", "source_type": "tech_report", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "Deep Think section baseline; exact tools undisclosed", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "Deep Think comparison baseline", "prompt_style": "official default", "temperature": "unknown" }, "notes": "Model card Table 7 (p24). Table 7 baseline. FrontierScience value 33.3 conflicts with the corroborated Pro=28.3 in release/Table 6/Table 11." } ] }, { "model_id": "qwen3.7-max", "benchmark_id": "frontier_science_research", "score": 24.8, "reference_url": "https://huggingface.co/tencent/Hy3", "reported_setting": { "mode": "thinking", "effort": "high / highest available", "tools": "none", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official benchmark or reported provider harness", "prompt_style": "default", "temperature": "0.9 for Hy3 family; otherwise source/provider setting", "context": "256000 for Hy3 family; otherwise source/provider setting", "notes": "Tencent own test" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Hy3 official model-card appendix: FrontierScience-Research = 24.8*. Tencent own testing." }, { "model_id": "gemini-3.1-pro", "benchmark_id": "frontier_science_research", "score": 16.7, "reference_url": "https://huggingface.co/tencent/Hy3", "reported_setting": { "mode": "thinking", "effort": "high / highest available", "tools": "none", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official benchmark or reported provider harness", "prompt_style": "default", "temperature": "0.9 for Hy3 family; otherwise source/provider setting", "context": "256000 for Hy3 family; otherwise source/provider setting", "notes": "Comparator value not marked as Tencent own test" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Hy3 official model-card appendix: FrontierScience-Research = 16.7. Comparator-reported value.", "candidates": [ { "score": 16.7, "reference_url": "https://seed.bytedance.com/en/blog/seed2-1-officially-released-advancing-ai-productivity", "source_type": "official_blog", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "research-oriented agent environment", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official", "prompt_style": "official default", "temperature": "unknown" }, "notes": "Release blog EN: Frontier-research chart (https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/user-upload/4xfa4mqq52cdr.png)." }, { "score": 16.7, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2.1/Seed2_1_Model_Card.pdf", "source_type": "tech_report", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "benchmark-specified", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official", "prompt_style": "official default", "temperature": "unknown" }, "notes": "Model card Table 7 (p24)." } ] }, { "model_id": "claude-opus-4.8", "benchmark_id": "frontier_science_research", "score": 32.9, "reference_url": "https://huggingface.co/tencent/Hy3", "reported_setting": { "mode": "thinking", "effort": "high / highest available", "tools": "none", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official benchmark or reported provider harness", "prompt_style": "default", "temperature": "0.9 for Hy3 family; otherwise source/provider setting", "context": "256000 for Hy3 family; otherwise source/provider setting", "notes": "Tencent own test" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Hy3 official model-card appendix: FrontierScience-Research = 32.9*. Tencent own testing." }, { "model_id": "gpt-5.5", "benchmark_id": "frontier_science_research", "score": 33.9, "reference_url": "https://huggingface.co/tencent/Hy3", "reported_setting": { "mode": "thinking", "effort": "high / highest available", "tools": "none", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official benchmark or reported provider harness", "prompt_style": "default", "temperature": "0.9 for Hy3 family; otherwise source/provider setting", "context": "256000 for Hy3 family; otherwise source/provider setting", "notes": "Comparator value not marked as Tencent own test" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Hy3 official model-card appendix: FrontierScience-Research = 33.9. Comparator-reported value.", "candidates": [ { "score": 33.9, "reference_url": "https://seed.bytedance.com/en/blog/seed2-1-officially-released-advancing-ai-productivity", "source_type": "official_blog", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "research-oriented agent environment", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official", "prompt_style": "official default", "temperature": "unknown" }, "notes": "Release blog EN: Frontier-research chart (https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/user-upload/4xfa4mqq52cdr.png)." } ] }, { "model_id": "hy3-preview", "benchmark_id": "frontiersci_olympiad", "score": 70.0, "reference_url": "https://huggingface.co/tencent/Hy3", "reported_setting": { "mode": "thinking", "effort": "high / highest available", "tools": "none", "sampling": "pass@1", "judge": "gpt-oss-120b, high reasoning", "harness": "official paper judge prompts; self-evaluated", "prompt_style": "default", "temperature": "0.9 for Hy3 family; otherwise source/provider setting", "context": "256000 for Hy3 family; otherwise source/provider setting", "notes": "Comparator value not marked as Tencent own test" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Hy3 official model-card appendix: FrontierScience-Olympiad = 70.0. Comparator-reported value." }, { "model_id": "hy3", "benchmark_id": "frontiersci_olympiad", "score": 74.8, "reference_url": "https://huggingface.co/tencent/Hy3", "reported_setting": { "mode": "thinking", "effort": "high / highest available", "tools": "none", "sampling": "pass@1", "judge": "gpt-oss-120b, high reasoning", "harness": "official paper judge prompts; self-evaluated", "prompt_style": "default", "temperature": "0.9 for Hy3 family; otherwise source/provider setting", "context": "256000 for Hy3 family; otherwise source/provider setting", "notes": "Comparator value not marked as Tencent own test" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Hy3 official model-card appendix: FrontierScience-Olympiad = 74.8. Comparator-reported value." }, { "model_id": "glm-5.1", "benchmark_id": "frontiersci_olympiad", "score": 65.0, "reference_url": "https://huggingface.co/tencent/Hy3", "reported_setting": { "mode": "thinking", "effort": "high / highest available", "tools": "none", "sampling": "pass@1", "judge": "gpt-oss-120b, high reasoning", "harness": "official paper judge prompts; self-evaluated", "prompt_style": "default", "temperature": "0.9 for Hy3 family; otherwise source/provider setting", "context": "256000 for Hy3 family; otherwise source/provider setting", "notes": "Tencent own test" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Hy3 official model-card appendix: FrontierScience-Olympiad = 65.0*. Tencent own testing." }, { "model_id": "glm-5.2", "benchmark_id": "frontiersci_olympiad", "score": 72.5, "reference_url": "https://huggingface.co/tencent/Hy3", "reported_setting": { "mode": "thinking", "effort": "high / highest available", "tools": "none", "sampling": "pass@1", "judge": "gpt-oss-120b, high reasoning", "harness": "official paper judge prompts; self-evaluated", "prompt_style": "default", "temperature": "0.9 for Hy3 family; otherwise source/provider setting", "context": "256000 for Hy3 family; otherwise source/provider setting", "notes": "Tencent own test" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Hy3 official model-card appendix: FrontierScience-Olympiad = 72.5*. Tencent own testing." }, { "model_id": "deepseek-v4-flash", "benchmark_id": "frontiersci_olympiad", "score": 73.0, "reference_url": "https://huggingface.co/tencent/Hy3", "reported_setting": { "mode": "thinking", "effort": "high / highest available", "tools": "none", "sampling": "pass@1", "judge": "gpt-oss-120b, high reasoning", "harness": "official paper judge prompts; self-evaluated", "prompt_style": "default", "temperature": "0.9 for Hy3 family; otherwise source/provider setting", "context": "256000 for Hy3 family; otherwise source/provider setting", "notes": "Tencent own test" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Hy3 official model-card appendix: FrontierScience-Olympiad = 73.0*. Tencent own testing." }, { "model_id": "deepseek-v4-pro", "benchmark_id": "frontiersci_olympiad", "score": 70.0, "reference_url": "https://huggingface.co/tencent/Hy3", "reported_setting": { "mode": "thinking", "effort": "high / highest available", "tools": "none", "sampling": "pass@1", "judge": "gpt-oss-120b, high reasoning", "harness": "official paper judge prompts; self-evaluated", "prompt_style": "default", "temperature": "0.9 for Hy3 family; otherwise source/provider setting", "context": "256000 for Hy3 family; otherwise source/provider setting", "notes": "Tencent own test" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Hy3 official model-card appendix: FrontierScience-Olympiad = 70.0*. Tencent own testing." }, { "model_id": "doubao-seed-2.1-pro", "benchmark_id": "frontiersci_olympiad", "score": 75.0, "reference_url": "https://seed.bytedance.com/en/blog/seed2-1-officially-released-advancing-ai-productivity", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "none", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official", "prompt_style": "official default", "temperature": "unknown" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "Release blog EN: Language/search chart (https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/user-upload/4xfa4mqq510uc.png).", "candidates": [ { "score": 70.1, "reference_url": "https://huggingface.co/tencent/Hy3", "source_type": "model_card", "reported_setting": { "mode": "thinking", "effort": "high / highest available", "tools": "none", "sampling": "pass@1", "judge": "gpt-oss-120b, high reasoning", "harness": "official paper judge prompts; self-evaluated", "prompt_style": "default", "temperature": "0.9 for Hy3 family; otherwise source/provider setting", "context": "256000 for Hy3 family; otherwise source/provider setting", "notes": "Tencent own test" }, "notes": "Hy3 official model-card appendix: FrontierScience-Olympiad = 70.1*. Tencent own testing." }, { "score": 75.0, "reference_url": "https://huggingface.co/tencent/Hy3", "source_type": "model_card", "reported_setting": { "mode": "thinking", "effort": "high / highest available", "tools": "none", "sampling": "pass@1", "judge": "gpt-oss-120b, high reasoning", "harness": "official paper judge prompts; self-evaluated", "prompt_style": "default", "temperature": "0.9 for Hy3 family; otherwise source/provider setting", "context": "256000 for Hy3 family; otherwise source/provider setting", "notes": "Comparator value not marked as Tencent own test" }, "notes": "Hy3 official model-card appendix: FrontierScience-Olympiad = 75. Comparator-reported value." } ] }, { "model_id": "qwen3.7-max", "benchmark_id": "frontiersci_olympiad", "score": 74.3, "reference_url": "https://huggingface.co/tencent/Hy3", "reported_setting": { "mode": "thinking", "effort": "high / highest available", "tools": "none", "sampling": "pass@1", "judge": "gpt-oss-120b, high reasoning", "harness": "official paper judge prompts; self-evaluated", "prompt_style": "default", "temperature": "0.9 for Hy3 family; otherwise source/provider setting", "context": "256000 for Hy3 family; otherwise source/provider setting", "notes": "Tencent own test" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Hy3 official model-card appendix: FrontierScience-Olympiad = 74.3*. Tencent own testing." }, { "model_id": "gemini-3.1-pro", "benchmark_id": "frontiersci_olympiad", "score": 79.0, "reference_url": "https://huggingface.co/tencent/Hy3", "reported_setting": { "mode": "thinking", "effort": "high / highest available", "tools": "none", "sampling": "pass@1", "judge": "gpt-oss-120b, high reasoning", "harness": "official paper judge prompts; self-evaluated", "prompt_style": "default", "temperature": "0.9 for Hy3 family; otherwise source/provider setting", "context": "256000 for Hy3 family; otherwise source/provider setting", "notes": "Comparator value not marked as Tencent own test" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Hy3 official model-card appendix: FrontierScience-Olympiad = 79.0. Comparator-reported value.", "candidates": [ { "score": 79.0, "reference_url": "https://seed.bytedance.com/en/blog/seed2-1-officially-released-advancing-ai-productivity", "source_type": "official_blog", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "none", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official", "prompt_style": "official default", "temperature": "unknown" }, "notes": "Release blog EN: Language/search chart (https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/user-upload/4xfa4mqq510uc.png)." } ] }, { "model_id": "claude-opus-4.8", "benchmark_id": "frontiersci_olympiad", "score": 74.3, "reference_url": "https://huggingface.co/tencent/Hy3", "reported_setting": { "mode": "thinking", "effort": "high / highest available", "tools": "none", "sampling": "pass@1", "judge": "gpt-oss-120b, high reasoning", "harness": "official paper judge prompts; self-evaluated", "prompt_style": "default", "temperature": "0.9 for Hy3 family; otherwise source/provider setting", "context": "256000 for Hy3 family; otherwise source/provider setting", "notes": "Tencent own test" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Hy3 official model-card appendix: FrontierScience-Olympiad = 74.3*. Tencent own testing." }, { "model_id": "gpt-5.5", "benchmark_id": "frontiersci_olympiad", "score": 73.8, "reference_url": "https://huggingface.co/tencent/Hy3", "reported_setting": { "mode": "thinking", "effort": "high / highest available", "tools": "none", "sampling": "pass@1", "judge": "gpt-oss-120b, high reasoning", "harness": "official paper judge prompts; self-evaluated", "prompt_style": "default", "temperature": "0.9 for Hy3 family; otherwise source/provider setting", "context": "256000 for Hy3 family; otherwise source/provider setting", "notes": "Tencent own test" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Hy3 official model-card appendix: FrontierScience-Olympiad = 73.8*. Tencent own testing.", "candidates": [ { "score": 69.0, "reference_url": "https://seed.bytedance.com/en/blog/seed2-1-officially-released-advancing-ai-productivity", "source_type": "official_blog", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "none", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official", "prompt_style": "official default", "temperature": "unknown" }, "notes": "Release blog EN: Language/search chart (https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/user-upload/4xfa4mqq510uc.png)." } ] }, { "model_id": "hy3-preview", "benchmark_id": "usamo_2026", "score": 37.3, "reference_url": "https://huggingface.co/tencent/Hy3", "reported_setting": { "mode": "thinking", "effort": "high / highest available", "tools": "none", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official benchmark or reported provider harness", "prompt_style": "default", "temperature": "0.9 for Hy3 family; otherwise source/provider setting", "context": "256000 for Hy3 family; otherwise source/provider setting", "notes": "Comparator value not marked as Tencent own test" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Hy3 official model-card appendix: USAMO 2026 = 37.3. Comparator-reported value." }, { "model_id": "hy3", "benchmark_id": "usamo_2026", "score": 72.0, "reference_url": "https://huggingface.co/tencent/Hy3", "reported_setting": { "mode": "thinking", "effort": "high / highest available", "tools": "none", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official benchmark or reported provider harness", "prompt_style": "default", "temperature": "0.9 for Hy3 family; otherwise source/provider setting", "context": "256000 for Hy3 family; otherwise source/provider setting", "notes": "Comparator value not marked as Tencent own test" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Hy3 official model-card appendix: USAMO 2026 = 72.0. Comparator-reported value." }, { "model_id": "glm-5.1", "benchmark_id": "usamo_2026", "score": 37.3, "reference_url": "https://huggingface.co/tencent/Hy3", "reported_setting": { "mode": "thinking", "effort": "high / highest available", "tools": "none", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official benchmark or reported provider harness", "prompt_style": "default", "temperature": "0.9 for Hy3 family; otherwise source/provider setting", "context": "256000 for Hy3 family; otherwise source/provider setting", "notes": "Tencent own test" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Hy3 official model-card appendix: USAMO 2026 = 37.3*. Tencent own testing." }, { "model_id": "glm-5.2", "benchmark_id": "usamo_2026", "score": 41.3, "reference_url": "https://huggingface.co/tencent/Hy3", "reported_setting": { "mode": "thinking", "effort": "high / highest available", "tools": "none", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official benchmark or reported provider harness", "prompt_style": "default", "temperature": "0.9 for Hy3 family; otherwise source/provider setting", "context": "256000 for Hy3 family; otherwise source/provider setting", "notes": "Tencent own test" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Hy3 official model-card appendix: USAMO 2026 = 41.3*. Tencent own testing." }, { "model_id": "deepseek-v4-flash", "benchmark_id": "usamo_2026", "score": 57.9, "reference_url": "https://huggingface.co/tencent/Hy3", "reported_setting": { "mode": "thinking", "effort": "high / highest available", "tools": "none", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official benchmark or reported provider harness", "prompt_style": "default", "temperature": "0.9 for Hy3 family; otherwise source/provider setting", "context": "256000 for Hy3 family; otherwise source/provider setting", "notes": "Tencent own test" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Hy3 official model-card appendix: USAMO 2026 = 57.9*. Tencent own testing." }, { "model_id": "deepseek-v4-pro", "benchmark_id": "usamo_2026", "score": 59.2, "reference_url": "https://huggingface.co/tencent/Hy3", "reported_setting": { "mode": "thinking", "effort": "high / highest available", "tools": "none", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official benchmark or reported provider harness", "prompt_style": "default", "temperature": "0.9 for Hy3 family; otherwise source/provider setting", "context": "256000 for Hy3 family; otherwise source/provider setting", "notes": "Tencent own test" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Hy3 official model-card appendix: USAMO 2026 = 59.2*. Tencent own testing." }, { "model_id": "doubao-seed-2.1-pro", "benchmark_id": "usamo_2026", "score": 60.3, "reference_url": "https://huggingface.co/tencent/Hy3", "reported_setting": { "mode": "thinking", "effort": "high / highest available", "tools": "none", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official benchmark or reported provider harness", "prompt_style": "default", "temperature": "0.9 for Hy3 family; otherwise source/provider setting", "context": "256000 for Hy3 family; otherwise source/provider setting", "notes": "Tencent own test" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Hy3 official model-card appendix: USAMO 2026 = 60.3*. Tencent own testing." }, { "model_id": "qwen3.7-max", "benchmark_id": "usamo_2026", "score": 57.1, "reference_url": "https://huggingface.co/tencent/Hy3", "reported_setting": { "mode": "thinking", "effort": "high / highest available", "tools": "none", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official benchmark or reported provider harness", "prompt_style": "default", "temperature": "0.9 for Hy3 family; otherwise source/provider setting", "context": "256000 for Hy3 family; otherwise source/provider setting", "notes": "Tencent own test" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Hy3 official model-card appendix: USAMO 2026 = 57.1*. Tencent own testing." }, { "model_id": "claude-opus-4.8", "benchmark_id": "usamo_2026", "score": 96.7, "reference_url": "https://www.anthropic.com/claude-sonnet-5-system-card", "reported_setting": { "effort": "high", "tools": "none", "sampling": "10 attempts/problem", "harness": "usamo" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Anthropic Claude Sonnet 5 System Card Section 8.6, page 120: Claude Opus 4.8; USAMO 2026 [summary]=96.7%. Source setting: effort=high; tools=none; sampling=10 attempts/problem; harness=usamo.", "candidates": [ { "score": 92.9, "reference_url": "https://huggingface.co/tencent/Hy3", "source_type": "model_card", "reported_setting": { "mode": "thinking", "effort": "high / highest available", "tools": "none", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official benchmark or reported provider harness", "prompt_style": "default", "temperature": "0.9 for Hy3 family; otherwise source/provider setting", "context": "256000 for Hy3 family; otherwise source/provider setting", "notes": "Tencent own test" }, "notes": "Hy3 official model-card appendix: USAMO 2026 = 92.9*. Tencent own testing." } ] }, { "model_id": "gpt-5.5", "benchmark_id": "usamo_2026", "score": 98.4, "reference_url": "https://huggingface.co/tencent/Hy3", "reported_setting": { "mode": "thinking", "effort": "high / highest available", "tools": "none", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official benchmark or reported provider harness", "prompt_style": "default", "temperature": "0.9 for Hy3 family; otherwise source/provider setting", "context": "256000 for Hy3 family; otherwise source/provider setting", "notes": "Tencent own test" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Hy3 official model-card appendix: USAMO 2026 = 98.4*. Tencent own testing." }, { "model_id": "hy3-preview", "benchmark_id": "matharena_apex_2025", "score": 12.6, "reference_url": "https://huggingface.co/tencent/Hy3", "reported_setting": { "mode": "thinking", "effort": "high / highest available", "tools": "none", "sampling": "samples=16 independent runs per problem", "judge": "benchmark-specified", "harness": "MathArena Apex 2025, 12-problem set", "prompt_style": "default", "temperature": "0.9 for Hy3 family; otherwise source/provider setting", "context": "256000 for Hy3 family; otherwise source/provider setting", "notes": "Comparator value not marked as Tencent own test" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Hy3 official model-card appendix: MathArena Apex = 12.6. Comparator-reported value." }, { "model_id": "hy3", "benchmark_id": "matharena_apex_2025", "score": 38.7, "reference_url": "https://huggingface.co/tencent/Hy3", "reported_setting": { "mode": "thinking", "effort": "high / highest available", "tools": "none", "sampling": "samples=16 independent runs per problem", "judge": "benchmark-specified", "harness": "MathArena Apex 2025, 12-problem set", "prompt_style": "default", "temperature": "0.9 for Hy3 family; otherwise source/provider setting", "context": "256000 for Hy3 family; otherwise source/provider setting", "notes": "Comparator value not marked as Tencent own test" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Hy3 official model-card appendix: MathArena Apex = 38.7. Comparator-reported value." }, { "model_id": "glm-5.2", "benchmark_id": "matharena_apex_2025", "score": 16.8, "reference_url": "https://huggingface.co/tencent/Hy3", "reported_setting": { "mode": "thinking", "effort": "high / highest available", "tools": "none", "sampling": "samples=16 independent runs per problem", "judge": "benchmark-specified", "harness": "MathArena Apex 2025, 12-problem set", "prompt_style": "default", "temperature": "0.9 for Hy3 family; otherwise source/provider setting", "context": "256000 for Hy3 family; otherwise source/provider setting", "notes": "Tencent own test" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Hy3 official model-card appendix: MathArena Apex = 16.8*. Tencent own testing." }, { "model_id": "doubao-seed-2.1-pro", "benchmark_id": "matharena_apex_2025", "score": 31.3, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2.1/Seed2_1_Model_Card.pdf", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "none", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official", "prompt_style": "official default", "temperature": "unknown" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Model card Table 11 (p53). Source dash is not zero.", "candidates": [ { "score": 31.3, "reference_url": "https://huggingface.co/tencent/Hy3", "source_type": "model_card", "reported_setting": { "mode": "thinking", "effort": "high / highest available", "tools": "none", "sampling": "samples=16 independent runs per problem", "judge": "benchmark-specified", "harness": "MathArena Apex 2025, 12-problem set", "prompt_style": "default", "temperature": "0.9 for Hy3 family; otherwise source/provider setting", "context": "256000 for Hy3 family; otherwise source/provider setting", "notes": "Comparator value not marked as Tencent own test" }, "notes": "Hy3 official model-card appendix: MathArena Apex = 31.3. Comparator-reported value." } ] }, { "model_id": "qwen3.7-max", "benchmark_id": "matharena_apex_2025", "score": 44.5, "reference_url": "https://huggingface.co/tencent/Hy3", "reported_setting": { "mode": "thinking", "effort": "high / highest available", "tools": "none", "sampling": "samples=16 independent runs per problem", "judge": "benchmark-specified", "harness": "MathArena Apex 2025, 12-problem set", "prompt_style": "default", "temperature": "0.9 for Hy3 family; otherwise source/provider setting", "context": "256000 for Hy3 family; otherwise source/provider setting", "notes": "Comparator value not marked as Tencent own test" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Hy3 official model-card appendix: MathArena Apex = 44.5. Comparator-reported value." }, { "model_id": "gpt-5.5", "benchmark_id": "matharena_apex_2025", "score": 85.4, "reference_url": "https://huggingface.co/tencent/Hy3", "reported_setting": { "mode": "thinking", "effort": "high / highest available", "tools": "none", "sampling": "samples=16 independent runs per problem", "judge": "benchmark-specified", "harness": "MathArena Apex 2025, 12-problem set", "prompt_style": "default", "temperature": "0.9 for Hy3 family; otherwise source/provider setting", "context": "256000 for Hy3 family; otherwise source/provider setting", "notes": "Tencent own test" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Hy3 official model-card appendix: MathArena Apex = 85.4*. Tencent own testing.", "candidates": [ { "score": 69.8, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2.1/Seed2_1_Model_Card.pdf", "source_type": "tech_report", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "none", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official", "prompt_style": "official default", "temperature": "unknown" }, "notes": "Model card Table 11 (p53). Source dash is not zero." } ] }, { "model_id": "hy3-preview", "benchmark_id": "horizon_math_pass12", "score": 1.8, "reference_url": "https://huggingface.co/tencent/Hy3", "reported_setting": { "mode": "thinking", "effort": "high / highest available", "tools": "none", "sampling": "pass@12", "judge": "benchmark-specified", "harness": "official benchmark or reported provider harness", "prompt_style": "default", "temperature": "0.9 for Hy3 family; otherwise source/provider setting", "context": "256000 for Hy3 family; otherwise source/provider setting", "notes": "Comparator value not marked as Tencent own test" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Hy3 official model-card appendix: HorizonMath (pass@12) = 1.8. Comparator-reported value." }, { "model_id": "hy3", "benchmark_id": "horizon_math_pass12", "score": 7.1, "reference_url": "https://huggingface.co/tencent/Hy3", "reported_setting": { "mode": "thinking", "effort": "high / highest available", "tools": "none", "sampling": "pass@12", "judge": "benchmark-specified", "harness": "official benchmark or reported provider harness", "prompt_style": "default", "temperature": "0.9 for Hy3 family; otherwise source/provider setting", "context": "256000 for Hy3 family; otherwise source/provider setting", "notes": "Comparator value not marked as Tencent own test" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Hy3 official model-card appendix: HorizonMath (pass@12) = 7.1. Comparator-reported value." }, { "model_id": "glm-5.2", "benchmark_id": "horizon_math_pass12", "score": 7.1, "reference_url": "https://huggingface.co/tencent/Hy3", "reported_setting": { "mode": "thinking", "effort": "high / highest available", "tools": "none", "sampling": "pass@12", "judge": "benchmark-specified", "harness": "official benchmark or reported provider harness", "prompt_style": "default", "temperature": "0.9 for Hy3 family; otherwise source/provider setting", "context": "256000 for Hy3 family; otherwise source/provider setting", "notes": "Tencent own test" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Hy3 official model-card appendix: HorizonMath (pass@12) = 7.1*. Tencent own testing." }, { "model_id": "deepseek-v4-flash", "benchmark_id": "horizon_math_pass12", "score": 4.4, "reference_url": "https://huggingface.co/tencent/Hy3", "reported_setting": { "mode": "thinking", "effort": "high / highest available", "tools": "none", "sampling": "pass@12", "judge": "benchmark-specified", "harness": "official benchmark or reported provider harness", "prompt_style": "default", "temperature": "0.9 for Hy3 family; otherwise source/provider setting", "context": "256000 for Hy3 family; otherwise source/provider setting", "notes": "Tencent own test" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Hy3 official model-card appendix: HorizonMath (pass@12) = 4.4*. Tencent own testing." }, { "model_id": "deepseek-v4-pro", "benchmark_id": "horizon_math_pass12", "score": 5.3, "reference_url": "https://huggingface.co/tencent/Hy3", "reported_setting": { "mode": "thinking", "effort": "high / highest available", "tools": "none", "sampling": "pass@12", "judge": "benchmark-specified", "harness": "official benchmark or reported provider harness", "prompt_style": "default", "temperature": "0.9 for Hy3 family; otherwise source/provider setting", "context": "256000 for Hy3 family; otherwise source/provider setting", "notes": "Tencent own test" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Hy3 official model-card appendix: HorizonMath (pass@12) = 5.3*. Tencent own testing." }, { "model_id": "doubao-seed-2.1-pro", "benchmark_id": "horizon_math_pass12", "score": 2.0, "reference_url": "https://seed.bytedance.com/en/blog/seed2-1-officially-released-advancing-ai-productivity", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "research-oriented agent environment", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official", "prompt_style": "official default", "temperature": "unknown" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "Release blog EN: Frontier-research chart (https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/user-upload/4xfa4mqq52cdr.png).", "candidates": [ { "score": 6.3, "reference_url": "https://huggingface.co/tencent/Hy3", "source_type": "model_card", "reported_setting": { "mode": "thinking", "effort": "high / highest available", "tools": "none", "sampling": "pass@12", "judge": "benchmark-specified", "harness": "official benchmark or reported provider harness", "prompt_style": "default", "temperature": "0.9 for Hy3 family; otherwise source/provider setting", "context": "256000 for Hy3 family; otherwise source/provider setting", "notes": "Tencent own test" }, "notes": "Hy3 official model-card appendix: HorizonMath (pass@12) = 6.3*. Tencent own testing." } ] }, { "model_id": "qwen3.7-max", "benchmark_id": "horizon_math_pass12", "score": 6.2, "reference_url": "https://huggingface.co/tencent/Hy3", "reported_setting": { "mode": "thinking", "effort": "high / highest available", "tools": "none", "sampling": "pass@12", "judge": "benchmark-specified", "harness": "official benchmark or reported provider harness", "prompt_style": "default", "temperature": "0.9 for Hy3 family; otherwise source/provider setting", "context": "256000 for Hy3 family; otherwise source/provider setting", "notes": "Tencent own test" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Hy3 official model-card appendix: HorizonMath (pass@12) = 6.2*. Tencent own testing." }, { "model_id": "gemini-3.1-pro", "benchmark_id": "horizon_math_pass12", "score": 7.1, "reference_url": "https://huggingface.co/tencent/Hy3", "reported_setting": { "mode": "thinking", "effort": "high / highest available", "tools": "none", "sampling": "pass@12", "judge": "benchmark-specified", "harness": "official benchmark or reported provider harness", "prompt_style": "default", "temperature": "0.9 for Hy3 family; otherwise source/provider setting", "context": "256000 for Hy3 family; otherwise source/provider setting", "notes": "Tencent own test" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Hy3 official model-card appendix: HorizonMath (pass@12) = 7.1*. Tencent own testing.", "candidates": [ { "score": 4.0, "reference_url": "https://seed.bytedance.com/en/blog/seed2-1-officially-released-advancing-ai-productivity", "source_type": "official_blog", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "research-oriented agent environment", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official", "prompt_style": "official default", "temperature": "unknown" }, "notes": "Release blog EN: Frontier-research chart (https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/user-upload/4xfa4mqq52cdr.png)." } ] }, { "model_id": "gpt-5.5", "benchmark_id": "horizon_math_pass12", "score": 11.6, "reference_url": "https://huggingface.co/tencent/Hy3", "reported_setting": { "mode": "thinking", "effort": "high / highest available", "tools": "none", "sampling": "pass@12", "judge": "benchmark-specified", "harness": "official benchmark or reported provider harness", "prompt_style": "default", "temperature": "0.9 for Hy3 family; otherwise source/provider setting", "context": "256000 for Hy3 family; otherwise source/provider setting", "notes": "Tencent own test" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Hy3 official model-card appendix: HorizonMath (pass@12) = 11.6*. Tencent own testing.", "candidates": [ { "score": 7.1, "reference_url": "https://seed.bytedance.com/en/blog/seed2-1-officially-released-advancing-ai-productivity", "source_type": "official_blog", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "research-oriented agent environment", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official", "prompt_style": "official default", "temperature": "unknown" }, "notes": "Release blog EN: Frontier-research chart (https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/user-upload/4xfa4mqq52cdr.png)." } ] }, { "model_id": "hy3-preview", "benchmark_id": "hy_math_internal", "score": 26.1, "reference_url": "https://huggingface.co/tencent/Hy3", "reported_setting": { "mode": "thinking", "effort": "high / highest available", "tools": "none", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official benchmark or reported provider harness", "prompt_style": "default", "temperature": "0.9 for Hy3 family; otherwise source/provider setting", "context": "256000 for Hy3 family; otherwise source/provider setting", "notes": "Comparator value not marked as Tencent own test" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Hy3 official model-card appendix: Hy-Math (internal) = 26.1. Comparator-reported value." }, { "model_id": "hy3", "benchmark_id": "hy_math_internal", "score": 60.9, "reference_url": "https://huggingface.co/tencent/Hy3", "reported_setting": { "mode": "thinking", "effort": "high / highest available", "tools": "none", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official benchmark or reported provider harness", "prompt_style": "default", "temperature": "0.9 for Hy3 family; otherwise source/provider setting", "context": "256000 for Hy3 family; otherwise source/provider setting", "notes": "Comparator value not marked as Tencent own test" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Hy3 official model-card appendix: Hy-Math (internal) = 60.9. Comparator-reported value." }, { "model_id": "glm-5.1", "benchmark_id": "hy_math_internal", "score": 6.7, "reference_url": "https://huggingface.co/tencent/Hy3", "reported_setting": { "mode": "thinking", "effort": "high / highest available", "tools": "none", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official benchmark or reported provider harness", "prompt_style": "default", "temperature": "0.9 for Hy3 family; otherwise source/provider setting", "context": "256000 for Hy3 family; otherwise source/provider setting", "notes": "Tencent own test" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Hy3 official model-card appendix: Hy-Math (internal) = 6.7*. Tencent own testing." }, { "model_id": "glm-5.2", "benchmark_id": "hy_math_internal", "score": 16.4, "reference_url": "https://huggingface.co/tencent/Hy3", "reported_setting": { "mode": "thinking", "effort": "high / highest available", "tools": "none", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official benchmark or reported provider harness", "prompt_style": "default", "temperature": "0.9 for Hy3 family; otherwise source/provider setting", "context": "256000 for Hy3 family; otherwise source/provider setting", "notes": "Tencent own test" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Hy3 official model-card appendix: Hy-Math (internal) = 16.4*. Tencent own testing." }, { "model_id": "deepseek-v4-flash", "benchmark_id": "hy_math_internal", "score": 66.0, "reference_url": "https://huggingface.co/tencent/Hy3", "reported_setting": { "mode": "thinking", "effort": "high / highest available", "tools": "none", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official benchmark or reported provider harness", "prompt_style": "default", "temperature": "0.9 for Hy3 family; otherwise source/provider setting", "context": "256000 for Hy3 family; otherwise source/provider setting", "notes": "Tencent own test" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Hy3 official model-card appendix: Hy-Math (internal) = 66.0*. Tencent own testing." }, { "model_id": "deepseek-v4-pro", "benchmark_id": "hy_math_internal", "score": 63.0, "reference_url": "https://huggingface.co/tencent/Hy3", "reported_setting": { "mode": "thinking", "effort": "high / highest available", "tools": "none", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official benchmark or reported provider harness", "prompt_style": "default", "temperature": "0.9 for Hy3 family; otherwise source/provider setting", "context": "256000 for Hy3 family; otherwise source/provider setting", "notes": "Tencent own test" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Hy3 official model-card appendix: Hy-Math (internal) = 63.0*. Tencent own testing." }, { "model_id": "doubao-seed-2.1-pro", "benchmark_id": "hy_math_internal", "score": 65.3, "reference_url": "https://huggingface.co/tencent/Hy3", "reported_setting": { "mode": "thinking", "effort": "high / highest available", "tools": "none", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official benchmark or reported provider harness", "prompt_style": "default", "temperature": "0.9 for Hy3 family; otherwise source/provider setting", "context": "256000 for Hy3 family; otherwise source/provider setting", "notes": "Tencent own test" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Hy3 official model-card appendix: Hy-Math (internal) = 65.3*. Tencent own testing." }, { "model_id": "qwen3.7-max", "benchmark_id": "hy_math_internal", "score": 66.1, "reference_url": "https://huggingface.co/tencent/Hy3", "reported_setting": { "mode": "thinking", "effort": "high / highest available", "tools": "none", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official benchmark or reported provider harness", "prompt_style": "default", "temperature": "0.9 for Hy3 family; otherwise source/provider setting", "context": "256000 for Hy3 family; otherwise source/provider setting", "notes": "Tencent own test" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Hy3 official model-card appendix: Hy-Math (internal) = 66.1*. Tencent own testing." }, { "model_id": "gemini-3.1-pro", "benchmark_id": "hy_math_internal", "score": 63.3, "reference_url": "https://huggingface.co/tencent/Hy3", "reported_setting": { "mode": "thinking", "effort": "high / highest available", "tools": "none", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official benchmark or reported provider harness", "prompt_style": "default", "temperature": "0.9 for Hy3 family; otherwise source/provider setting", "context": "256000 for Hy3 family; otherwise source/provider setting", "notes": "Tencent own test" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Hy3 official model-card appendix: Hy-Math (internal) = 63.3*. Tencent own testing." }, { "model_id": "gpt-5.5", "benchmark_id": "hy_math_internal", "score": 91.4, "reference_url": "https://huggingface.co/tencent/Hy3", "reported_setting": { "mode": "thinking", "effort": "high / highest available", "tools": "none", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official benchmark or reported provider harness", "prompt_style": "default", "temperature": "0.9 for Hy3 family; otherwise source/provider setting", "context": "256000 for Hy3 family; otherwise source/provider setting", "notes": "Tencent own test" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Hy3 official model-card appendix: Hy-Math (internal) = 91.4*. Tencent own testing." }, { "model_id": "hy3-preview", "benchmark_id": "phybench", "score": 71.4, "reference_url": "https://huggingface.co/tencent/Hy3", "reported_setting": { "mode": "thinking", "effort": "high / highest available", "tools": "none", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official benchmark or reported provider harness", "prompt_style": "default", "temperature": "0.9 for Hy3 family; otherwise source/provider setting", "context": "256000 for Hy3 family; otherwise source/provider setting", "notes": "Comparator value not marked as Tencent own test" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Hy3 official model-card appendix: PHYBench = 71.4. Comparator-reported value." }, { "model_id": "hy3", "benchmark_id": "phybench", "score": 77.4, "reference_url": "https://huggingface.co/tencent/Hy3", "reported_setting": { "mode": "thinking", "effort": "high / highest available", "tools": "none", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official benchmark or reported provider harness", "prompt_style": "default", "temperature": "0.9 for Hy3 family; otherwise source/provider setting", "context": "256000 for Hy3 family; otherwise source/provider setting", "notes": "Comparator value not marked as Tencent own test" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Hy3 official model-card appendix: PHYBench = 77.4. Comparator-reported value." }, { "model_id": "glm-5.1", "benchmark_id": "phybench", "score": 70.9, "reference_url": "https://huggingface.co/tencent/Hy3", "reported_setting": { "mode": "thinking", "effort": "high / highest available", "tools": "none", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official benchmark or reported provider harness", "prompt_style": "default", "temperature": "0.9 for Hy3 family; otherwise source/provider setting", "context": "256000 for Hy3 family; otherwise source/provider setting", "notes": "Tencent own test" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Hy3 official model-card appendix: PHYBench = 70.9*. Tencent own testing." }, { "model_id": "glm-5.2", "benchmark_id": "phybench", "score": 71.5, "reference_url": "https://huggingface.co/tencent/Hy3", "reported_setting": { "mode": "thinking", "effort": "high / highest available", "tools": "none", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official benchmark or reported provider harness", "prompt_style": "default", "temperature": "0.9 for Hy3 family; otherwise source/provider setting", "context": "256000 for Hy3 family; otherwise source/provider setting", "notes": "Tencent own test" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Hy3 official model-card appendix: PHYBench = 71.5*. Tencent own testing." }, { "model_id": "deepseek-v4-flash", "benchmark_id": "phybench", "score": 76.6, "reference_url": "https://huggingface.co/tencent/Hy3", "reported_setting": { "mode": "thinking", "effort": "high / highest available", "tools": "none", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official benchmark or reported provider harness", "prompt_style": "default", "temperature": "0.9 for Hy3 family; otherwise source/provider setting", "context": "256000 for Hy3 family; otherwise source/provider setting", "notes": "Tencent own test" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Hy3 official model-card appendix: PHYBench = 76.6*. Tencent own testing." }, { "model_id": "deepseek-v4-pro", "benchmark_id": "phybench", "score": 75.4, "reference_url": "https://huggingface.co/tencent/Hy3", "reported_setting": { "mode": "thinking", "effort": "high / highest available", "tools": "none", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official benchmark or reported provider harness", "prompt_style": "default", "temperature": "0.9 for Hy3 family; otherwise source/provider setting", "context": "256000 for Hy3 family; otherwise source/provider setting", "notes": "Tencent own test" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Hy3 official model-card appendix: PHYBench = 75.4*. Tencent own testing." }, { "model_id": "doubao-seed-2.1-pro", "benchmark_id": "phybench", "score": 77.5, "reference_url": "https://huggingface.co/tencent/Hy3", "reported_setting": { "mode": "thinking", "effort": "high / highest available", "tools": "none", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official benchmark or reported provider harness", "prompt_style": "default", "temperature": "0.9 for Hy3 family; otherwise source/provider setting", "context": "256000 for Hy3 family; otherwise source/provider setting", "notes": "Tencent own test" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Hy3 official model-card appendix: PHYBench = 77.5*. Tencent own testing." }, { "model_id": "qwen3.7-max", "benchmark_id": "phybench", "score": 76.5, "reference_url": "https://huggingface.co/tencent/Hy3", "reported_setting": { "mode": "thinking", "effort": "high / highest available", "tools": "none", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official benchmark or reported provider harness", "prompt_style": "default", "temperature": "0.9 for Hy3 family; otherwise source/provider setting", "context": "256000 for Hy3 family; otherwise source/provider setting", "notes": "Tencent own test" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Hy3 official model-card appendix: PHYBench = 76.5*. Tencent own testing." }, { "model_id": "gemini-3.1-pro", "benchmark_id": "phybench", "score": 82.0, "reference_url": "https://huggingface.co/tencent/Hy3", "reported_setting": { "mode": "thinking", "effort": "high / highest available", "tools": "none", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official benchmark or reported provider harness", "prompt_style": "default", "temperature": "0.9 for Hy3 family; otherwise source/provider setting", "context": "256000 for Hy3 family; otherwise source/provider setting", "notes": "Tencent own test" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Hy3 official model-card appendix: PHYBench = 82.0*. Tencent own testing." }, { "model_id": "claude-opus-4.8", "benchmark_id": "phybench", "score": 79.6, "reference_url": "https://huggingface.co/tencent/Hy3", "reported_setting": { "mode": "thinking", "effort": "high / highest available", "tools": "none", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official benchmark or reported provider harness", "prompt_style": "default", "temperature": "0.9 for Hy3 family; otherwise source/provider setting", "context": "256000 for Hy3 family; otherwise source/provider setting", "notes": "Tencent own test" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Hy3 official model-card appendix: PHYBench = 79.6*. Tencent own testing." }, { "model_id": "gpt-5.5", "benchmark_id": "phybench", "score": 77.5, "reference_url": "https://huggingface.co/tencent/Hy3", "reported_setting": { "mode": "thinking", "effort": "high / highest available", "tools": "none", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official benchmark or reported provider harness", "prompt_style": "default", "temperature": "0.9 for Hy3 family; otherwise source/provider setting", "context": "256000 for Hy3 family; otherwise source/provider setting", "notes": "Tencent own test" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Hy3 official model-card appendix: PHYBench = 77.5*. Tencent own testing." }, { "model_id": "hy3-preview", "benchmark_id": "cmt_benchmark", "score": 19.4, "reference_url": "https://huggingface.co/tencent/Hy3", "reported_setting": { "mode": "thinking", "effort": "high / highest available", "tools": "none", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official benchmark or reported provider harness", "prompt_style": "default", "temperature": "0.9 for Hy3 family; otherwise source/provider setting", "context": "256000 for Hy3 family; otherwise source/provider setting", "notes": "Comparator value not marked as Tencent own test" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Hy3 official model-card appendix: CMT-Benchmark = 19.4. Comparator-reported value." }, { "model_id": "hy3", "benchmark_id": "cmt_benchmark", "score": 37.8, "reference_url": "https://huggingface.co/tencent/Hy3", "reported_setting": { "mode": "thinking", "effort": "high / highest available", "tools": "none", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official benchmark or reported provider harness", "prompt_style": "default", "temperature": "0.9 for Hy3 family; otherwise source/provider setting", "context": "256000 for Hy3 family; otherwise source/provider setting", "notes": "Comparator value not marked as Tencent own test" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Hy3 official model-card appendix: CMT-Benchmark = 37.8. Comparator-reported value." }, { "model_id": "glm-5.1", "benchmark_id": "cmt_benchmark", "score": 25.3, "reference_url": "https://huggingface.co/tencent/Hy3", "reported_setting": { "mode": "thinking", "effort": "high / highest available", "tools": "none", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official benchmark or reported provider harness", "prompt_style": "default", "temperature": "0.9 for Hy3 family; otherwise source/provider setting", "context": "256000 for Hy3 family; otherwise source/provider setting", "notes": "Tencent own test" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Hy3 official model-card appendix: CMT-Benchmark = 25.3*. Tencent own testing." }, { "model_id": "glm-5.2", "benchmark_id": "cmt_benchmark", "score": 34.1, "reference_url": "https://huggingface.co/tencent/Hy3", "reported_setting": { "mode": "thinking", "effort": "high / highest available", "tools": "none", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official benchmark or reported provider harness", "prompt_style": "default", "temperature": "0.9 for Hy3 family; otherwise source/provider setting", "context": "256000 for Hy3 family; otherwise source/provider setting", "notes": "Tencent own test" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Hy3 official model-card appendix: CMT-Benchmark = 34.1*. Tencent own testing." }, { "model_id": "deepseek-v4-flash", "benchmark_id": "cmt_benchmark", "score": 26.8, "reference_url": "https://huggingface.co/tencent/Hy3", "reported_setting": { "mode": "thinking", "effort": "high / highest available", "tools": "none", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official benchmark or reported provider harness", "prompt_style": "default", "temperature": "0.9 for Hy3 family; otherwise source/provider setting", "context": "256000 for Hy3 family; otherwise source/provider setting", "notes": "Tencent own test" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Hy3 official model-card appendix: CMT-Benchmark = 26.8*. Tencent own testing." }, { "model_id": "deepseek-v4-pro", "benchmark_id": "cmt_benchmark", "score": 35.0, "reference_url": "https://huggingface.co/tencent/Hy3", "reported_setting": { "mode": "thinking", "effort": "high / highest available", "tools": "none", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official benchmark or reported provider harness", "prompt_style": "default", "temperature": "0.9 for Hy3 family; otherwise source/provider setting", "context": "256000 for Hy3 family; otherwise source/provider setting", "notes": "Tencent own test" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Hy3 official model-card appendix: CMT-Benchmark = 35.0*. Tencent own testing." }, { "model_id": "doubao-seed-2.1-pro", "benchmark_id": "cmt_benchmark", "score": 31.0, "reference_url": "https://huggingface.co/tencent/Hy3", "reported_setting": { "mode": "thinking", "effort": "high / highest available", "tools": "none", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official benchmark or reported provider harness", "prompt_style": "default", "temperature": "0.9 for Hy3 family; otherwise source/provider setting", "context": "256000 for Hy3 family; otherwise source/provider setting", "notes": "Tencent own test" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Hy3 official model-card appendix: CMT-Benchmark = 31.0*. Tencent own testing." }, { "model_id": "qwen3.7-max", "benchmark_id": "cmt_benchmark", "score": 40.0, "reference_url": "https://huggingface.co/tencent/Hy3", "reported_setting": { "mode": "thinking", "effort": "high / highest available", "tools": "none", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official benchmark or reported provider harness", "prompt_style": "default", "temperature": "0.9 for Hy3 family; otherwise source/provider setting", "context": "256000 for Hy3 family; otherwise source/provider setting", "notes": "Tencent own test" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Hy3 official model-card appendix: CMT-Benchmark = 40.0*. Tencent own testing." }, { "model_id": "gemini-3.1-pro", "benchmark_id": "cmt_benchmark", "score": 47.0, "reference_url": "https://huggingface.co/tencent/Hy3", "reported_setting": { "mode": "thinking", "effort": "high / highest available", "tools": "none", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official benchmark or reported provider harness", "prompt_style": "default", "temperature": "0.9 for Hy3 family; otherwise source/provider setting", "context": "256000 for Hy3 family; otherwise source/provider setting", "notes": "Tencent own test" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Hy3 official model-card appendix: CMT-Benchmark = 47.0*. Tencent own testing." }, { "model_id": "claude-opus-4.8", "benchmark_id": "cmt_benchmark", "score": 43.8, "reference_url": "https://huggingface.co/tencent/Hy3", "reported_setting": { "mode": "thinking", "effort": "high / highest available", "tools": "none", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official benchmark or reported provider harness", "prompt_style": "default", "temperature": "0.9 for Hy3 family; otherwise source/provider setting", "context": "256000 for Hy3 family; otherwise source/provider setting", "notes": "Tencent own test" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Hy3 official model-card appendix: CMT-Benchmark = 43.8*. Tencent own testing." }, { "model_id": "gpt-5.5", "benchmark_id": "cmt_benchmark", "score": 43.6, "reference_url": "https://huggingface.co/tencent/Hy3", "reported_setting": { "mode": "thinking", "effort": "high / highest available", "tools": "none", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official benchmark or reported provider harness", "prompt_style": "default", "temperature": "0.9 for Hy3 family; otherwise source/provider setting", "context": "256000 for Hy3 family; otherwise source/provider setting", "notes": "Tencent own test" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Hy3 official model-card appendix: CMT-Benchmark = 43.6*. Tencent own testing." }, { "model_id": "hy3-preview", "benchmark_id": "imo_answerbench", "score": 84.3, "reference_url": "https://huggingface.co/tencent/Hy3", "reported_setting": { "mode": "thinking", "effort": "high / highest available", "tools": "none", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official benchmark or reported provider harness", "prompt_style": "default", "temperature": "0.9 for Hy3 family; otherwise source/provider setting", "context": "256000 for Hy3 family; otherwise source/provider setting", "notes": "Comparator value not marked as Tencent own test" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Hy3 official model-card appendix: IMOAnswerBench = 84.3. Comparator-reported value." }, { "model_id": "hy3", "benchmark_id": "imo_answerbench", "score": 90.0, "reference_url": "https://huggingface.co/tencent/Hy3", "reported_setting": { "mode": "thinking", "effort": "high / highest available", "tools": "none", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official benchmark or reported provider harness", "prompt_style": "default", "temperature": "0.9 for Hy3 family; otherwise source/provider setting", "context": "256000 for Hy3 family; otherwise source/provider setting", "notes": "Comparator value not marked as Tencent own test" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Hy3 official model-card appendix: IMOAnswerBench = 90.0. Comparator-reported value." }, { "model_id": "doubao-seed-2.1-pro", "benchmark_id": "imo_answerbench", "score": 88.5, "reference_url": "https://huggingface.co/tencent/Hy3", "reported_setting": { "mode": "thinking", "effort": "high / highest available", "tools": "none", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official benchmark or reported provider harness", "prompt_style": "default", "temperature": "0.9 for Hy3 family; otherwise source/provider setting", "context": "256000 for Hy3 family; otherwise source/provider setting", "notes": "Tencent own test" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Hy3 official model-card appendix: IMOAnswerBench = 88.5*. Tencent own testing." }, { "model_id": "gpt-5.5", "benchmark_id": "imo_answerbench", "score": 92.1, "reference_url": "https://huggingface.co/tencent/Hy3", "reported_setting": { "mode": "thinking", "effort": "high / highest available", "tools": "none", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official benchmark or reported provider harness", "prompt_style": "default", "temperature": "0.9 for Hy3 family; otherwise source/provider setting", "context": "256000 for Hy3 family; otherwise source/provider setting", "notes": "Tencent own test" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Hy3 official model-card appendix: IMOAnswerBench = 92.1*. Tencent own testing.", "candidates": [ { "score": 79.5, "reference_url": "https://longcat.chat/blog/longcat-2.0/", "source_type": "official_blog", "reported_setting": { "mode": "thinking", "effort": "source does not state", "tools": "none", "sampling": "source does not state", "judge": "IMO-AnswerBench answer autograder", "harness": "Meituan unified in-house harness", "prompt_style": "default", "temperature": "1.0; top_k=-1; top_p=0.95", "context": "source does not state", "notes": "Measured in-house by Meituan unless otherwise noted." }, "notes": "LongCat-2.0 official tech blog: IMO-AnswerBench = 79.5. Measured in-house by Meituan under the reported unified harness." } ] }, { "model_id": "hy3-preview", "benchmark_id": "superchem", "score": 47.0, "reference_url": "https://huggingface.co/tencent/Hy3", "reported_setting": { "mode": "thinking", "effort": "high / highest available", "tools": "none", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official benchmark or reported provider harness", "prompt_style": "default", "temperature": "0.9 for Hy3 family; otherwise source/provider setting", "context": "256000 for Hy3 family; otherwise source/provider setting", "notes": "Comparator value not marked as Tencent own test" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Hy3 official model-card appendix: SuperChem = 47.0. Comparator-reported value." }, { "model_id": "hy3", "benchmark_id": "superchem", "score": 54.9, "reference_url": "https://huggingface.co/tencent/Hy3", "reported_setting": { "mode": "thinking", "effort": "high / highest available", "tools": "none", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official benchmark or reported provider harness", "prompt_style": "default", "temperature": "0.9 for Hy3 family; otherwise source/provider setting", "context": "256000 for Hy3 family; otherwise source/provider setting", "notes": "Comparator value not marked as Tencent own test" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Hy3 official model-card appendix: SuperChem = 54.9. Comparator-reported value." }, { "model_id": "glm-5.1", "benchmark_id": "superchem", "score": 51.4, "reference_url": "https://huggingface.co/tencent/Hy3", "reported_setting": { "mode": "thinking", "effort": "high / highest available", "tools": "none", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official benchmark or reported provider harness", "prompt_style": "default", "temperature": "0.9 for Hy3 family; otherwise source/provider setting", "context": "256000 for Hy3 family; otherwise source/provider setting", "notes": "Tencent own test" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Hy3 official model-card appendix: SuperChem = 51.4*. Tencent own testing." }, { "model_id": "glm-5.2", "benchmark_id": "superchem", "score": 60.2, "reference_url": "https://huggingface.co/tencent/Hy3", "reported_setting": { "mode": "thinking", "effort": "high / highest available", "tools": "none", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official benchmark or reported provider harness", "prompt_style": "default", "temperature": "0.9 for Hy3 family; otherwise source/provider setting", "context": "256000 for Hy3 family; otherwise source/provider setting", "notes": "Tencent own test" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Hy3 official model-card appendix: SuperChem = 60.2*. Tencent own testing." }, { "model_id": "deepseek-v4-flash", "benchmark_id": "superchem", "score": 51.8, "reference_url": "https://huggingface.co/tencent/Hy3", "reported_setting": { "mode": "thinking", "effort": "high / highest available", "tools": "none", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official benchmark or reported provider harness", "prompt_style": "default", "temperature": "0.9 for Hy3 family; otherwise source/provider setting", "context": "256000 for Hy3 family; otherwise source/provider setting", "notes": "Tencent own test" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Hy3 official model-card appendix: SuperChem = 51.8*. Tencent own testing." }, { "model_id": "deepseek-v4-pro", "benchmark_id": "superchem", "score": 61.2, "reference_url": "https://huggingface.co/tencent/Hy3", "reported_setting": { "mode": "thinking", "effort": "high / highest available", "tools": "none", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official benchmark or reported provider harness", "prompt_style": "default", "temperature": "0.9 for Hy3 family; otherwise source/provider setting", "context": "256000 for Hy3 family; otherwise source/provider setting", "notes": "Tencent own test" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Hy3 official model-card appendix: SuperChem = 61.2*. Tencent own testing." }, { "model_id": "doubao-seed-2.1-pro", "benchmark_id": "superchem", "score": 59.8, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2.1/Seed2_1_Model_Card.pdf", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "none", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official", "prompt_style": "official default", "temperature": "unknown" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Model card Table 11 (p53).", "candidates": [ { "score": 59.8, "reference_url": "https://huggingface.co/tencent/Hy3", "source_type": "model_card", "reported_setting": { "mode": "thinking", "effort": "high / highest available", "tools": "none", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official benchmark or reported provider harness", "prompt_style": "default", "temperature": "0.9 for Hy3 family; otherwise source/provider setting", "context": "256000 for Hy3 family; otherwise source/provider setting", "notes": "Comparator value not marked as Tencent own test" }, "notes": "Hy3 official model-card appendix: SuperChem = 59.8. Comparator-reported value." } ] }, { "model_id": "qwen3.7-max", "benchmark_id": "superchem", "score": 60.5, "reference_url": "https://huggingface.co/tencent/Hy3", "reported_setting": { "mode": "thinking", "effort": "high / highest available", "tools": "none", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official benchmark or reported provider harness", "prompt_style": "default", "temperature": "0.9 for Hy3 family; otherwise source/provider setting", "context": "256000 for Hy3 family; otherwise source/provider setting", "notes": "Tencent own test" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Hy3 official model-card appendix: SuperChem = 60.5*. Tencent own testing." }, { "model_id": "gemini-3.1-pro", "benchmark_id": "superchem", "score": 69.2, "reference_url": "https://huggingface.co/tencent/Hy3", "reported_setting": { "mode": "thinking", "effort": "high / highest available", "tools": "none", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official benchmark or reported provider harness", "prompt_style": "default", "temperature": "0.9 for Hy3 family; otherwise source/provider setting", "context": "256000 for Hy3 family; otherwise source/provider setting", "notes": "Tencent own test" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Hy3 official model-card appendix: SuperChem = 69.2*. Tencent own testing.", "candidates": [ { "score": 71.4, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2.1/Seed2_1_Model_Card.pdf", "source_type": "tech_report", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "none", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official", "prompt_style": "official default", "temperature": "unknown" }, "notes": "Model card Table 11 (p53)." } ] }, { "model_id": "claude-opus-4.8", "benchmark_id": "superchem", "score": 66.9, "reference_url": "https://huggingface.co/tencent/Hy3", "reported_setting": { "mode": "thinking", "effort": "high / highest available", "tools": "none", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official benchmark or reported provider harness", "prompt_style": "default", "temperature": "0.9 for Hy3 family; otherwise source/provider setting", "context": "256000 for Hy3 family; otherwise source/provider setting", "notes": "Tencent own test" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Hy3 official model-card appendix: SuperChem = 66.9*. Tencent own testing." }, { "model_id": "gpt-5.5", "benchmark_id": "superchem", "score": 66.2, "reference_url": "https://huggingface.co/tencent/Hy3", "reported_setting": { "mode": "thinking", "effort": "high / highest available", "tools": "none", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official benchmark or reported provider harness", "prompt_style": "default", "temperature": "0.9 for Hy3 family; otherwise source/provider setting", "context": "256000 for Hy3 family; otherwise source/provider setting", "notes": "Tencent own test" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Hy3 official model-card appendix: SuperChem = 66.2*. Tencent own testing.", "candidates": [ { "score": 61.1, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2.1/Seed2_1_Model_Card.pdf", "source_type": "tech_report", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "none", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official", "prompt_style": "official default", "temperature": "unknown" }, "notes": "Model card Table 11 (p53)." } ] }, { "model_id": "hy3-preview", "benchmark_id": "cl_bench", "score": 22.8, "reference_url": "https://huggingface.co/tencent/Hy3", "reported_setting": { "mode": "thinking", "effort": "high / highest available", "tools": "none", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official benchmark or reported provider harness", "prompt_style": "default", "temperature": "0.9 for Hy3 family; otherwise source/provider setting", "context": "256000 for Hy3 family; otherwise source/provider setting", "notes": "Comparator value not marked as Tencent own test" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Hy3 official model-card appendix: CL-bench = 22.8. Comparator-reported value." }, { "model_id": "hy3", "benchmark_id": "cl_bench", "score": 23.8, "reference_url": "https://huggingface.co/tencent/Hy3", "reported_setting": { "mode": "thinking", "effort": "high / highest available", "tools": "none", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official benchmark or reported provider harness", "prompt_style": "default", "temperature": "0.9 for Hy3 family; otherwise source/provider setting", "context": "256000 for Hy3 family; otherwise source/provider setting", "notes": "Comparator value not marked as Tencent own test" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Hy3 official model-card appendix: CL-bench = 23.8. Comparator-reported value." }, { "model_id": "glm-5.1", "benchmark_id": "cl_bench", "score": 21.0, "reference_url": "https://huggingface.co/tencent/Hy3", "reported_setting": { "mode": "thinking", "effort": "high / highest available", "tools": "none", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official benchmark or reported provider harness", "prompt_style": "default", "temperature": "0.9 for Hy3 family; otherwise source/provider setting", "context": "256000 for Hy3 family; otherwise source/provider setting", "notes": "Tencent own test" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Hy3 official model-card appendix: CL-bench = 21.0*. Tencent own testing." }, { "model_id": "glm-5.2", "benchmark_id": "cl_bench", "score": 23.1, "reference_url": "https://huggingface.co/tencent/Hy3", "reported_setting": { "mode": "thinking", "effort": "high / highest available", "tools": "none", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official benchmark or reported provider harness", "prompt_style": "default", "temperature": "0.9 for Hy3 family; otherwise source/provider setting", "context": "256000 for Hy3 family; otherwise source/provider setting", "notes": "Tencent own test" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Hy3 official model-card appendix: CL-bench = 23.1*. Tencent own testing." }, { "model_id": "deepseek-v4-flash", "benchmark_id": "cl_bench", "score": 16.4, "reference_url": "https://huggingface.co/tencent/Hy3", "reported_setting": { "mode": "thinking", "effort": "high / highest available", "tools": "none", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official benchmark or reported provider harness", "prompt_style": "default", "temperature": "0.9 for Hy3 family; otherwise source/provider setting", "context": "256000 for Hy3 family; otherwise source/provider setting", "notes": "Tencent own test" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Hy3 official model-card appendix: CL-bench = 16.4*. Tencent own testing." }, { "model_id": "deepseek-v4-pro", "benchmark_id": "cl_bench", "score": 18.1, "reference_url": "https://huggingface.co/tencent/Hy3", "reported_setting": { "mode": "thinking", "effort": "high / highest available", "tools": "none", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official benchmark or reported provider harness", "prompt_style": "default", "temperature": "0.9 for Hy3 family; otherwise source/provider setting", "context": "256000 for Hy3 family; otherwise source/provider setting", "notes": "Tencent own test" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Hy3 official model-card appendix: CL-bench = 18.1*. Tencent own testing." }, { "model_id": "doubao-seed-2.1-pro", "benchmark_id": "cl_bench", "score": 23.5, "reference_url": "https://huggingface.co/tencent/Hy3", "reported_setting": { "mode": "thinking", "effort": "high / highest available", "tools": "none", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official benchmark or reported provider harness", "prompt_style": "default", "temperature": "0.9 for Hy3 family; otherwise source/provider setting", "context": "256000 for Hy3 family; otherwise source/provider setting", "notes": "Tencent own test" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Hy3 official model-card appendix: CL-bench = 23.5*. Tencent own testing." }, { "model_id": "qwen3.7-max", "benchmark_id": "cl_bench", "score": 18.4, "reference_url": "https://huggingface.co/tencent/Hy3", "reported_setting": { "mode": "thinking", "effort": "high / highest available", "tools": "none", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official benchmark or reported provider harness", "prompt_style": "default", "temperature": "0.9 for Hy3 family; otherwise source/provider setting", "context": "256000 for Hy3 family; otherwise source/provider setting", "notes": "Tencent own test" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Hy3 official model-card appendix: CL-bench = 18.4*. Tencent own testing." }, { "model_id": "gemini-3.1-pro", "benchmark_id": "cl_bench", "score": 21.3, "reference_url": "https://huggingface.co/tencent/Hy3", "reported_setting": { "mode": "thinking", "effort": "high / highest available", "tools": "none", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official benchmark or reported provider harness", "prompt_style": "default", "temperature": "0.9 for Hy3 family; otherwise source/provider setting", "context": "256000 for Hy3 family; otherwise source/provider setting", "notes": "Tencent own test" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Hy3 official model-card appendix: CL-bench = 21.3*. Tencent own testing." }, { "model_id": "claude-opus-4.8", "benchmark_id": "cl_bench", "score": 24.8, "reference_url": "https://huggingface.co/tencent/Hy3", "reported_setting": { "mode": "thinking", "effort": "high / highest available", "tools": "none", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official benchmark or reported provider harness", "prompt_style": "default", "temperature": "0.9 for Hy3 family; otherwise source/provider setting", "context": "256000 for Hy3 family; otherwise source/provider setting", "notes": "Tencent own test" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Hy3 official model-card appendix: CL-bench = 24.8*. Tencent own testing." }, { "model_id": "gpt-5.5", "benchmark_id": "cl_bench", "score": 27.8, "reference_url": "https://huggingface.co/tencent/Hy3", "reported_setting": { "mode": "thinking", "effort": "high / highest available", "tools": "none", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official benchmark or reported provider harness", "prompt_style": "default", "temperature": "0.9 for Hy3 family; otherwise source/provider setting", "context": "256000 for Hy3 family; otherwise source/provider setting", "notes": "Tencent own test" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Hy3 official model-card appendix: CL-bench = 27.8*. Tencent own testing." }, { "model_id": "hy3-preview", "benchmark_id": "cl_bench_life", "score": 15.7, "reference_url": "https://huggingface.co/tencent/Hy3", "reported_setting": { "mode": "thinking", "effort": "high / highest available", "tools": "none", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official benchmark or reported provider harness", "prompt_style": "default", "temperature": "0.9 for Hy3 family; otherwise source/provider setting", "context": "256000 for Hy3 family; otherwise source/provider setting", "notes": "Comparator value not marked as Tencent own test" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Hy3 official model-card appendix: CL-bench life = 15.7. Comparator-reported value." }, { "model_id": "hy3", "benchmark_id": "cl_bench_life", "score": 17.0, "reference_url": "https://huggingface.co/tencent/Hy3", "reported_setting": { "mode": "thinking", "effort": "high / highest available", "tools": "none", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official benchmark or reported provider harness", "prompt_style": "default", "temperature": "0.9 for Hy3 family; otherwise source/provider setting", "context": "256000 for Hy3 family; otherwise source/provider setting", "notes": "Comparator value not marked as Tencent own test" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Hy3 official model-card appendix: CL-bench life = 17.0. Comparator-reported value." }, { "model_id": "glm-5.1", "benchmark_id": "cl_bench_life", "score": 14.0, "reference_url": "https://huggingface.co/tencent/Hy3", "reported_setting": { "mode": "thinking", "effort": "high / highest available", "tools": "none", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official benchmark or reported provider harness", "prompt_style": "default", "temperature": "0.9 for Hy3 family; otherwise source/provider setting", "context": "256000 for Hy3 family; otherwise source/provider setting", "notes": "Tencent own test" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Hy3 official model-card appendix: CL-bench life = 14.0*. Tencent own testing." }, { "model_id": "glm-5.2", "benchmark_id": "cl_bench_life", "score": 21.0, "reference_url": "https://huggingface.co/tencent/Hy3", "reported_setting": { "mode": "thinking", "effort": "high / highest available", "tools": "none", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official benchmark or reported provider harness", "prompt_style": "default", "temperature": "0.9 for Hy3 family; otherwise source/provider setting", "context": "256000 for Hy3 family; otherwise source/provider setting", "notes": "Tencent own test" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Hy3 official model-card appendix: CL-bench life = 21.0*. Tencent own testing." }, { "model_id": "deepseek-v4-flash", "benchmark_id": "cl_bench_life", "score": 11.4, "reference_url": "https://huggingface.co/tencent/Hy3", "reported_setting": { "mode": "thinking", "effort": "high / highest available", "tools": "none", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official benchmark or reported provider harness", "prompt_style": "default", "temperature": "0.9 for Hy3 family; otherwise source/provider setting", "context": "256000 for Hy3 family; otherwise source/provider setting", "notes": "Tencent own test" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Hy3 official model-card appendix: CL-bench life = 11.4*. Tencent own testing." }, { "model_id": "deepseek-v4-pro", "benchmark_id": "cl_bench_life", "score": 14.3, "reference_url": "https://huggingface.co/tencent/Hy3", "reported_setting": { "mode": "thinking", "effort": "high / highest available", "tools": "none", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official benchmark or reported provider harness", "prompt_style": "default", "temperature": "0.9 for Hy3 family; otherwise source/provider setting", "context": "256000 for Hy3 family; otherwise source/provider setting", "notes": "Tencent own test" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Hy3 official model-card appendix: CL-bench life = 14.3*. Tencent own testing." }, { "model_id": "doubao-seed-2.1-pro", "benchmark_id": "cl_bench_life", "score": 13.3, "reference_url": "https://huggingface.co/tencent/Hy3", "reported_setting": { "mode": "thinking", "effort": "high / highest available", "tools": "none", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official benchmark or reported provider harness", "prompt_style": "default", "temperature": "0.9 for Hy3 family; otherwise source/provider setting", "context": "256000 for Hy3 family; otherwise source/provider setting", "notes": "Tencent own test" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Hy3 official model-card appendix: CL-bench life = 13.3*. Tencent own testing." }, { "model_id": "qwen3.7-max", "benchmark_id": "cl_bench_life", "score": 13.0, "reference_url": "https://huggingface.co/tencent/Hy3", "reported_setting": { "mode": "thinking", "effort": "high / highest available", "tools": "none", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official benchmark or reported provider harness", "prompt_style": "default", "temperature": "0.9 for Hy3 family; otherwise source/provider setting", "context": "256000 for Hy3 family; otherwise source/provider setting", "notes": "Tencent own test" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Hy3 official model-card appendix: CL-bench life = 13.0*. Tencent own testing." }, { "model_id": "gemini-3.1-pro", "benchmark_id": "cl_bench_life", "score": 16.5, "reference_url": "https://huggingface.co/tencent/Hy3", "reported_setting": { "mode": "thinking", "effort": "high / highest available", "tools": "none", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official benchmark or reported provider harness", "prompt_style": "default", "temperature": "0.9 for Hy3 family; otherwise source/provider setting", "context": "256000 for Hy3 family; otherwise source/provider setting", "notes": "Tencent own test" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Hy3 official model-card appendix: CL-bench life = 16.5*. Tencent own testing." }, { "model_id": "claude-opus-4.8", "benchmark_id": "cl_bench_life", "score": 16.9, "reference_url": "https://huggingface.co/tencent/Hy3", "reported_setting": { "mode": "thinking", "effort": "high / highest available", "tools": "none", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official benchmark or reported provider harness", "prompt_style": "default", "temperature": "0.9 for Hy3 family; otherwise source/provider setting", "context": "256000 for Hy3 family; otherwise source/provider setting", "notes": "Tencent own test" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Hy3 official model-card appendix: CL-bench life = 16.9*. Tencent own testing." }, { "model_id": "gpt-5.5", "benchmark_id": "cl_bench_life", "score": 21.1, "reference_url": "https://huggingface.co/tencent/Hy3", "reported_setting": { "mode": "thinking", "effort": "high / highest available", "tools": "none", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official benchmark or reported provider harness", "prompt_style": "default", "temperature": "0.9 for Hy3 family; otherwise source/provider setting", "context": "256000 for Hy3 family; otherwise source/provider setting", "notes": "Tencent own test" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Hy3 official model-card appendix: CL-bench life = 21.1*. Tencent own testing." }, { "model_id": "hy3-preview", "benchmark_id": "aa_lcr", "score": 66.3, "reference_url": "https://huggingface.co/tencent/Hy3", "reported_setting": { "mode": "thinking", "effort": "high / highest available", "tools": "none", "sampling": "repeats=3 per question", "judge": "Equality Checker LLM", "harness": "Artificial Analysis AA-LCR, 100 questions", "prompt_style": "default", "temperature": "0.9 for Hy3 family; otherwise source/provider setting", "context": "256000 for Hy3 family; otherwise source/provider setting", "notes": "Comparator value not marked as Tencent own test" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Hy3 official model-card appendix: AA-LCR = 66.3. Comparator-reported value." }, { "model_id": "hy3", "benchmark_id": "aa_lcr", "score": 73.4, "reference_url": "https://huggingface.co/tencent/Hy3", "reported_setting": { "mode": "thinking", "effort": "high / highest available", "tools": "none", "sampling": "repeats=3 per question", "judge": "Equality Checker LLM", "harness": "Artificial Analysis AA-LCR, 100 questions", "prompt_style": "default", "temperature": "0.9 for Hy3 family; otherwise source/provider setting", "context": "256000 for Hy3 family; otherwise source/provider setting", "notes": "Comparator value not marked as Tencent own test" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Hy3 official model-card appendix: AA-LCR = 73.4. Comparator-reported value." }, { "model_id": "glm-5.1", "benchmark_id": "aa_lcr", "score": 71.0, "reference_url": "https://huggingface.co/tencent/Hy3", "reported_setting": { "mode": "thinking", "effort": "high / highest available", "tools": "none", "sampling": "repeats=3 per question", "judge": "Equality Checker LLM", "harness": "Artificial Analysis AA-LCR, 100 questions", "prompt_style": "default", "temperature": "0.9 for Hy3 family; otherwise source/provider setting", "context": "256000 for Hy3 family; otherwise source/provider setting", "notes": "Tencent own test" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Hy3 official model-card appendix: AA-LCR = 71.0*. Tencent own testing.", "candidates": [ { "score": 66.9, "reference_url": "https://research.nvidia.com/labs/nemotron/files/NVIDIA-Nemotron-3-Ultra-Technical-Report.pdf", "source_type": "tech_report", "reported_setting": { "version_metric": "AA-LCR / %", "tools": "none", "harness": "NeMo Evaluator", "sampling": "16-run average" }, "notes": "NVIDIA Nemotron 3 Ultra technical report, Final Table 10: GLM-5.1: aa_lcr=66.9." } ] }, { "model_id": "deepseek-v4-flash", "benchmark_id": "aa_lcr", "score": 64.8, "reference_url": "https://huggingface.co/tencent/Hy3", "reported_setting": { "mode": "thinking", "effort": "high / highest available", "tools": "none", "sampling": "repeats=3 per question", "judge": "Equality Checker LLM", "harness": "Artificial Analysis AA-LCR, 100 questions", "prompt_style": "default", "temperature": "0.9 for Hy3 family; otherwise source/provider setting", "context": "256000 for Hy3 family; otherwise source/provider setting", "notes": "Tencent own test" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Hy3 official model-card appendix: AA-LCR = 64.8*. Tencent own testing.", "candidates": [ { "score": 62.7, "reference_url": "https://research.nvidia.com/labs/nemotron/files/NVIDIA-Nemotron-3-Ultra-Technical-Report.pdf", "source_type": "tech_report", "reported_setting": { "version_metric": "AA-LCR / %", "tools": "none", "harness": "NeMo Evaluator", "sampling": "16-run average" }, "notes": "NVIDIA Nemotron 3 Ultra technical report, Final Table 10: DeepSeek-V4-Flash Preview: aa_lcr=62.7." } ] }, { "model_id": "deepseek-v4-pro", "benchmark_id": "aa_lcr", "score": 71.3, "reference_url": "https://huggingface.co/tencent/Hy3", "reported_setting": { "mode": "thinking", "effort": "high / highest available", "tools": "none", "sampling": "repeats=3 per question", "judge": "Equality Checker LLM", "harness": "Artificial Analysis AA-LCR, 100 questions", "prompt_style": "default", "temperature": "0.9 for Hy3 family; otherwise source/provider setting", "context": "256000 for Hy3 family; otherwise source/provider setting", "notes": "Tencent own test" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Hy3 official model-card appendix: AA-LCR = 71.3*. Tencent own testing.", "candidates": [ { "score": 67.3, "reference_url": "https://research.nvidia.com/labs/nemotron/files/NVIDIA-Nemotron-3-Ultra-Technical-Report.pdf", "source_type": "tech_report", "reported_setting": { "version_metric": "AA-LCR / %", "tools": "none", "harness": "NeMo Evaluator", "sampling": "16-run average" }, "notes": "NVIDIA Nemotron 3 Ultra technical report, Final Table 10: DeepSeek-V4-Pro: aa_lcr=67.3." } ] }, { "model_id": "doubao-seed-2.1-pro", "benchmark_id": "aa_lcr", "score": 70.0, "reference_url": "https://huggingface.co/tencent/Hy3", "reported_setting": { "mode": "thinking", "effort": "high / highest available", "tools": "none", "sampling": "repeats=3 per question", "judge": "Equality Checker LLM", "harness": "Artificial Analysis AA-LCR, 100 questions", "prompt_style": "default", "temperature": "0.9 for Hy3 family; otherwise source/provider setting", "context": "256000 for Hy3 family; otherwise source/provider setting", "notes": "Tencent own test" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Hy3 official model-card appendix: AA-LCR = 70.0*. Tencent own testing." }, { "model_id": "qwen3.7-max", "benchmark_id": "aa_lcr", "score": 70.2, "reference_url": "https://huggingface.co/tencent/Hy3", "reported_setting": { "mode": "thinking", "effort": "high / highest available", "tools": "none", "sampling": "repeats=3 per question", "judge": "Equality Checker LLM", "harness": "Artificial Analysis AA-LCR, 100 questions", "prompt_style": "default", "temperature": "0.9 for Hy3 family; otherwise source/provider setting", "context": "256000 for Hy3 family; otherwise source/provider setting", "notes": "Tencent own test" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Hy3 official model-card appendix: AA-LCR = 70.2*. Tencent own testing." }, { "model_id": "gemini-3.1-pro", "benchmark_id": "aa_lcr", "score": 74.6, "reference_url": "https://huggingface.co/tencent/Hy3", "reported_setting": { "mode": "thinking", "effort": "high / highest available", "tools": "none", "sampling": "repeats=3 per question", "judge": "Equality Checker LLM", "harness": "Artificial Analysis AA-LCR, 100 questions", "prompt_style": "default", "temperature": "0.9 for Hy3 family; otherwise source/provider setting", "context": "256000 for Hy3 family; otherwise source/provider setting", "notes": "Tencent own test" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Hy3 official model-card appendix: AA-LCR = 74.6*. Tencent own testing." }, { "model_id": "longcat-2.0", "benchmark_id": "terminal_bench_2_1", "score": 70.8, "reference_url": "https://longcat.chat/blog/longcat-2.0/", "reported_setting": { "mode": "thinking", "effort": "source does not state", "tools": "terminal/code execution", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "Claude Code; 8 CPU/16GB; timeout=6h", "prompt_style": "default", "temperature": "1.0; top_k=-1; top_p=0.95", "context": "source does not state", "notes": "Measured in-house by Meituan unless otherwise noted." }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "LongCat-2.0 official tech blog: Terminal-Bench 2.1 = 70.8. Measured in-house by Meituan under the reported unified harness." }, { "model_id": "claude-opus-4.7", "benchmark_id": "terminal_bench_2_1", "score": 71.7, "reference_url": "https://longcat.chat/blog/longcat-2.0/", "reported_setting": { "mode": "non-thinking", "effort": "default", "tools": "terminal agent", "sampling": "pass@1 (avg 5 attempts)", "judge": "benchmark-specified", "harness": "Terminus-2", "prompt_style": "default", "temperature": "provider official setting", "context": "default", "notes": "Asterisked provider-reported value. Opus 4.7 canonical Terminal-Bench setting uses Terminus-2 with thinking disabled." }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "LongCat-2.0 official tech blog: Terminal-Bench 2.1 = 71.7*. Asterisked value is cited by LongCat from the model provider's official report.", "candidates": [ { "score": 66.1, "reference_url": "https://deepmind.google/models/model-cards/gemini-3-5-flash/", "source_type": "official_model_card", "reported_setting": { "mode": "thinking/reasoning", "effort": "max/best available", "tools": "terminal agent", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "Terminus-2; Gemini self-computed, others public leaderboard", "temperature": "default", "snapshot": "May 2026" }, "notes": "Google Gemini 3.5 Flash May 2026 table: Terminal-Bench 2.1; Terminus-2 harness. Settings and provenance are preserved per cell." }, { "score": 78.9, "reference_url": "https://x.ai/news/grok-4-5", "source_type": "official_blog", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "container terminal", "sampling": "pass@1", "judge": "verified task-success evaluator", "harness": "Grok Build for Grok; source-reported peer harnesses", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown" }, "notes": "release45 Terminal-Bench 2.1 Task success rate (%); source effort=max; benchmark protocol matches canonical=True. Unknown source fields remain unknown." }, { "score": 71.7, "reference_url": "https://seed.bytedance.com/en/blog/seed2-1-officially-released-advancing-ai-productivity", "source_type": "official_blog", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "Trae or benchmark-specific coding-agent environment", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official", "prompt_style": "official default", "temperature": "unknown" }, "notes": "Release blog EN: Coding chart (https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/user-upload/4xfa4mqq4wlbt.png)." } ] }, { "model_id": "longcat-2.0", "benchmark_id": "swe_bench_pro", "score": 59.51, "reference_url": "https://longcat.chat/blog/longcat-2.0/", "reported_setting": { "mode": "thinking", "effort": "source does not state", "tools": "agentic code execution", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "Claude Code; 4 CPU/8GB; problematic tasks corrected", "prompt_style": "default", "temperature": "1.0; top_k=-1; top_p=1.0", "context": "source does not state", "notes": "Measured in-house by Meituan unless otherwise noted." }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "LongCat-2.0 official tech blog: SWE-bench Pro = 59.5 (chart data 59.51). Measured in-house by Meituan under the reported unified harness." }, { "model_id": "longcat-2.0", "benchmark_id": "swe_bench_multilingual", "score": 77.33, "reference_url": "https://longcat.chat/blog/longcat-2.0/", "reported_setting": { "mode": "thinking", "effort": "source does not state", "tools": "agentic code execution", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "Claude Code; 4 CPU/8GB; problematic tasks corrected", "prompt_style": "default", "temperature": "1.0; top_k=-1; top_p=1.0", "context": "source does not state", "notes": "Measured in-house by Meituan unless otherwise noted." }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "LongCat-2.0 official tech blog: SWE-bench Multilingual = 77.3 (chart data 77.33). Measured in-house by Meituan under the reported unified harness." }, { "model_id": "claude-opus-4.7", "benchmark_id": "swe_bench_multilingual", "score": 80.5, "reference_url": "https://longcat.chat/blog/longcat-2.0/", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "agentic code execution", "sampling": "pass@1 (avg 5 trials)", "judge": "benchmark-specified", "harness": "provider official report; details not reproduced by LongCat", "prompt_style": "default", "temperature": "provider official setting", "context": "default", "notes": "Asterisk: LongCat cites this value from the model provider's official report rather than measuring it in-house." }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "LongCat-2.0 official tech blog: SWE-bench Multilingual = 80.5*. Asterisked value is cited by LongCat from the model provider's official report." }, { "model_id": "longcat-2.0", "benchmark_id": "forte_avg3", "score": 73.22, "reference_url": "https://longcat.chat/blog/longcat-2.0/", "reported_setting": { "mode": "thinking", "effort": "source does not state", "tools": "office-computing environment", "sampling": "trials=3 (Avg@3)", "judge": "LLM-as-judge; all-or-nothing rubric score", "harness": "OpenClaw in Docker; timeout=45m; 2 CPU/4GB; single API call=500s; max retries=10", "prompt_style": "default", "temperature": "source does not state", "context": "source does not state", "notes": "Measured in-house by Meituan unless otherwise noted." }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "LongCat-2.0 official tech blog: FORTE = 73.2 (chart data 73.22). Measured in-house by Meituan under the reported unified harness." }, { "model_id": "gemini-3.1-pro", "benchmark_id": "forte_avg3", "score": 70.31, "reference_url": "https://longcat.chat/blog/longcat-2.0/", "reported_setting": { "mode": "thinking", "effort": "source does not state", "tools": "office-computing environment", "sampling": "trials=3 (Avg@3)", "judge": "LLM-as-judge; all-or-nothing rubric score", "harness": "OpenClaw in Docker; timeout=45m; 2 CPU/4GB; single API call=500s; max retries=10", "prompt_style": "default", "temperature": "source does not state", "context": "source does not state", "notes": "Measured in-house by Meituan unless otherwise noted." }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "LongCat-2.0 official tech blog: FORTE = 70.3 (chart data 70.31). Measured in-house by Meituan under the reported unified harness." }, { "model_id": "gpt-5.5", "benchmark_id": "forte_avg3", "score": 77.78, "reference_url": "https://longcat.chat/blog/longcat-2.0/", "reported_setting": { "mode": "thinking", "effort": "source does not state", "tools": "office-computing environment", "sampling": "trials=3 (Avg@3)", "judge": "LLM-as-judge; all-or-nothing rubric score", "harness": "OpenClaw in Docker; timeout=45m; 2 CPU/4GB; single API call=500s; max retries=10", "prompt_style": "default", "temperature": "source does not state", "context": "source does not state", "notes": "Measured in-house by Meituan unless otherwise noted." }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "LongCat-2.0 official tech blog: FORTE = 77.8 (chart data 77.78). Measured in-house by Meituan under the reported unified harness." }, { "model_id": "claude-opus-4.6", "benchmark_id": "forte_avg3", "score": 73.22, "reference_url": "https://longcat.chat/blog/longcat-2.0/", "reported_setting": { "mode": "thinking", "effort": "source does not state", "tools": "office-computing environment", "sampling": "trials=3 (Avg@3)", "judge": "LLM-as-judge; all-or-nothing rubric score", "harness": "OpenClaw in Docker; timeout=45m; 2 CPU/4GB; single API call=500s; max retries=10", "prompt_style": "default", "temperature": "source does not state", "context": "source does not state", "notes": "Measured in-house by Meituan unless otherwise noted." }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "LongCat-2.0 official tech blog: FORTE = 73.2 (chart data 73.22). Measured in-house by Meituan under the reported unified harness." }, { "model_id": "claude-opus-4.7", "benchmark_id": "forte_avg3", "score": 77.6, "reference_url": "https://longcat.chat/blog/longcat-2.0/", "reported_setting": { "mode": "thinking", "effort": "source does not state", "tools": "office-computing environment", "sampling": "trials=3 (Avg@3)", "judge": "LLM-as-judge; all-or-nothing rubric score", "harness": "OpenClaw in Docker; timeout=45m; 2 CPU/4GB; single API call=500s; max retries=10", "prompt_style": "default", "temperature": "source does not state", "context": "source does not state", "notes": "Measured in-house by Meituan unless otherwise noted." }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "LongCat-2.0 official tech blog: FORTE = 77.6. Measured in-house by Meituan under the reported unified harness." }, { "model_id": "claude-opus-4.8", "benchmark_id": "forte_avg3", "score": 77.23, "reference_url": "https://longcat.chat/blog/longcat-2.0/", "reported_setting": { "mode": "thinking", "effort": "source does not state", "tools": "office-computing environment", "sampling": "trials=3 (Avg@3)", "judge": "LLM-as-judge; all-or-nothing rubric score", "harness": "OpenClaw in Docker; timeout=45m; 2 CPU/4GB; single API call=500s; max retries=10", "prompt_style": "default", "temperature": "source does not state", "context": "source does not state", "notes": "Measured in-house by Meituan unless otherwise noted." }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "LongCat-2.0 official tech blog: FORTE = 77.2 (chart data 77.23). Measured in-house by Meituan under the reported unified harness." }, { "model_id": "longcat-2.0", "benchmark_id": "browsecomp", "score": 79.86, "reference_url": "https://longcat.chat/blog/longcat-2.0/", "reported_setting": { "mode": "thinking", "effort": "source does not state", "tools": "Search and Browse", "sampling": "source does not state", "judge": "benchmark-specified", "harness": "Meituan unified in-house search harness", "prompt_style": "default", "temperature": "source does not state", "context": "source does not state", "notes": "Measured in-house by Meituan unless otherwise noted." }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "LongCat-2.0 official tech blog: BrowseComp = 79.9 (chart data 79.86). Measured in-house by Meituan under the reported unified harness." }, { "model_id": "longcat-2.0", "benchmark_id": "rwsearch", "score": 78.75, "reference_url": "https://longcat.chat/blog/longcat-2.0/", "reported_setting": { "mode": "thinking", "effort": "source does not state", "tools": "basic Search and Browse", "sampling": "source does not state", "judge": "publisher objective scorer; details undisclosed", "harness": "bare-model evaluation; no context management", "prompt_style": "default", "temperature": "source does not state", "context": "source does not state", "notes": "Measured in-house by Meituan unless otherwise noted." }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "LongCat-2.0 official tech blog: RWSearch = 78.8 (chart data 78.75). Measured in-house by Meituan under the reported unified harness." }, { "model_id": "gemini-3.1-pro", "benchmark_id": "rwsearch", "score": 76.25, "reference_url": "https://longcat.chat/blog/longcat-2.0/", "reported_setting": { "mode": "thinking", "effort": "source does not state", "tools": "basic Search and Browse", "sampling": "source does not state", "judge": "publisher objective scorer; details undisclosed", "harness": "bare-model evaluation; no context management", "prompt_style": "default", "temperature": "source does not state", "context": "source does not state", "notes": "Measured in-house by Meituan unless otherwise noted." }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "LongCat-2.0 official tech blog: RWSearch = 76.3 (chart data 76.25). Measured in-house by Meituan under the reported unified harness." }, { "model_id": "gpt-5.5", "benchmark_id": "rwsearch", "score": 85.25, "reference_url": "https://longcat.chat/blog/longcat-2.0/", "reported_setting": { "mode": "thinking", "effort": "source does not state", "tools": "basic Search and Browse", "sampling": "source does not state", "judge": "publisher objective scorer; details undisclosed", "harness": "bare-model evaluation; no context management", "prompt_style": "default", "temperature": "source does not state", "context": "source does not state", "notes": "Measured in-house by Meituan unless otherwise noted." }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "LongCat-2.0 official tech blog: RWSearch = 85.3 (chart data 85.25). Measured in-house by Meituan under the reported unified harness." }, { "model_id": "claude-opus-4.6", "benchmark_id": "rwsearch", "score": 81.25, "reference_url": "https://longcat.chat/blog/longcat-2.0/", "reported_setting": { "mode": "thinking", "effort": "source does not state", "tools": "basic Search and Browse", "sampling": "source does not state", "judge": "publisher objective scorer; details undisclosed", "harness": "bare-model evaluation; no context management", "prompt_style": "default", "temperature": "source does not state", "context": "source does not state", "notes": "Measured in-house by Meituan unless otherwise noted." }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "LongCat-2.0 official tech blog: RWSearch = 81.3 (chart data 81.25). Measured in-house by Meituan under the reported unified harness." }, { "model_id": "claude-opus-4.7", "benchmark_id": "rwsearch", "score": 79.25, "reference_url": "https://longcat.chat/blog/longcat-2.0/", "reported_setting": { "mode": "thinking", "effort": "source does not state", "tools": "basic Search and Browse", "sampling": "source does not state", "judge": "publisher objective scorer; details undisclosed", "harness": "bare-model evaluation; no context management", "prompt_style": "default", "temperature": "source does not state", "context": "source does not state", "notes": "Measured in-house by Meituan unless otherwise noted." }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "LongCat-2.0 official tech blog: RWSearch = 79.3 (chart data 79.25). Measured in-house by Meituan under the reported unified harness." }, { "model_id": "claude-opus-4.8", "benchmark_id": "rwsearch", "score": 77.25, "reference_url": "https://longcat.chat/blog/longcat-2.0/", "reported_setting": { "mode": "thinking", "effort": "source does not state", "tools": "basic Search and Browse", "sampling": "source does not state", "judge": "publisher objective scorer; details undisclosed", "harness": "bare-model evaluation; no context management", "prompt_style": "default", "temperature": "source does not state", "context": "source does not state", "notes": "Measured in-house by Meituan unless otherwise noted." }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "LongCat-2.0 official tech blog: RWSearch = 77.3 (chart data 77.25). Measured in-house by Meituan under the reported unified harness." }, { "model_id": "longcat-2.0", "benchmark_id": "ifeval", "score": 90.0, "reference_url": "https://longcat.chat/blog/longcat-2.0/", "reported_setting": { "mode": "thinking", "effort": "source does not state", "tools": "none", "sampling": "source does not state", "judge": "benchmark-specified", "harness": "Meituan unified in-house harness", "prompt_style": "default", "temperature": "0.7; top_k=-1; top_p=0.95", "context": "source does not state", "notes": "Measured in-house by Meituan unless otherwise noted." }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "LongCat-2.0 official tech blog: IFEval = 90.0. Measured in-house by Meituan under the reported unified harness." }, { "model_id": "gpt-5.5", "benchmark_id": "ifeval", "score": 95.0, "reference_url": "https://longcat.chat/blog/longcat-2.0/", "reported_setting": { "mode": "thinking", "effort": "source does not state", "tools": "none", "sampling": "source does not state", "judge": "benchmark-specified", "harness": "Meituan unified in-house harness", "prompt_style": "default", "temperature": "0.7; top_k=-1; top_p=0.95", "context": "source does not state", "notes": "Measured in-house by Meituan unless otherwise noted." }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "LongCat-2.0 official tech blog: IFEval = 95.0. Measured in-house by Meituan under the reported unified harness." }, { "model_id": "claude-opus-4.7", "benchmark_id": "ifeval", "score": 88.7, "reference_url": "https://longcat.chat/blog/longcat-2.0/", "reported_setting": { "mode": "thinking", "effort": "source does not state", "tools": "none", "sampling": "source does not state", "judge": "benchmark-specified", "harness": "Meituan unified in-house harness", "prompt_style": "default", "temperature": "0.7; top_k=-1; top_p=0.95", "context": "source does not state", "notes": "Measured in-house by Meituan unless otherwise noted." }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "LongCat-2.0 official tech blog: IFEval = 88.7. Measured in-house by Meituan under the reported unified harness." }, { "model_id": "claude-opus-4.8", "benchmark_id": "ifeval", "score": 86.0, "reference_url": "https://longcat.chat/blog/longcat-2.0/", "reported_setting": { "mode": "thinking", "effort": "source does not state", "tools": "none", "sampling": "source does not state", "judge": "benchmark-specified", "harness": "Meituan unified in-house harness", "prompt_style": "default", "temperature": "0.7; top_k=-1; top_p=0.95", "context": "source does not state", "notes": "Measured in-house by Meituan unless otherwise noted." }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "LongCat-2.0 official tech blog: IFEval = 86.0. Measured in-house by Meituan under the reported unified harness." }, { "model_id": "longcat-2.0", "benchmark_id": "imo_answerbench", "score": 81.8, "reference_url": "https://longcat.chat/blog/longcat-2.0/", "reported_setting": { "mode": "thinking", "effort": "source does not state", "tools": "none", "sampling": "source does not state", "judge": "IMO-AnswerBench answer autograder", "harness": "Meituan unified in-house harness", "prompt_style": "default", "temperature": "1.0; top_k=-1; top_p=0.95", "context": "source does not state", "notes": "Measured in-house by Meituan unless otherwise noted." }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "LongCat-2.0 official tech blog: IMO-AnswerBench = 81.8. Measured in-house by Meituan under the reported unified harness." }, { "model_id": "claude-opus-4.7", "benchmark_id": "imo_answerbench", "score": 81.8, "reference_url": "https://longcat.chat/blog/longcat-2.0/", "reported_setting": { "mode": "thinking", "effort": "source does not state", "tools": "none", "sampling": "source does not state", "judge": "IMO-AnswerBench answer autograder", "harness": "Meituan unified in-house harness", "prompt_style": "default", "temperature": "1.0; top_k=-1; top_p=0.95", "context": "source does not state", "notes": "Measured in-house by Meituan unless otherwise noted." }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "LongCat-2.0 official tech blog: IMO-AnswerBench = 81.8. Measured in-house by Meituan under the reported unified harness." }, { "model_id": "longcat-2.0", "benchmark_id": "gpqa_diamond", "score": 88.9, "reference_url": "https://longcat.chat/blog/longcat-2.0/", "reported_setting": { "mode": "thinking", "effort": "source does not state", "tools": "none", "sampling": "source does not state", "judge": "benchmark-specified", "harness": "Meituan unified in-house harness", "prompt_style": "default", "temperature": "0.7; top_k=-1; top_p=0.95", "context": "source does not state", "notes": "Measured in-house by Meituan unless otherwise noted." }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "LongCat-2.0 official tech blog: GPQA-diamond = 88.9. Measured in-house by Meituan under the reported unified harness." }, { "model_id": "intern-s2-preview-397b", "benchmark_id": "mmlu_pro", "score": 89.75, "reference_url": "https://huggingface.co/internlm/Intern-S2-Preview-397B", "reported_setting": { "mode": "thinking (default)", "effort": "source does not state", "tools": "none", "sampling": "source does not state", "judge": "benchmark-specific; exact version not stated", "harness": "OpenCompass or AgentCompass; exact adapter not stated", "prompt_style": "source does not state", "temperature": "source does not state", "context": "max 256K text", "notes": "Shanghai AI Lab evaluated all model columns using OpenCompass, VLMEvalKit, and AgentCompass. Per-model mode/effort/sampling are not reported; only Intern-S2 maximum evaluation lengths are disclosed." }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Intern-S2-Preview-397B official model card: MMLU Pro = 89.75. This is the release model's own reported evaluation." }, { "model_id": "kimi-k2.7-code", "benchmark_id": "mmlu_pro", "score": 87.1, "reference_url": "https://huggingface.co/internlm/Intern-S2-Preview-397B", "reported_setting": { "mode": "source does not state", "effort": "source does not state", "tools": "none", "sampling": "source does not state", "judge": "benchmark-specific; exact version not stated", "harness": "OpenCompass or AgentCompass; exact adapter not stated", "prompt_style": "source does not state", "temperature": "source does not state", "context": "source does not state", "notes": "Shanghai AI Lab evaluated all model columns using OpenCompass, VLMEvalKit, and AgentCompass. Per-model mode/effort/sampling are not reported; only Intern-S2 maximum evaluation lengths are disclosed." }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Intern-S2-Preview-397B official model card: MMLU Pro = 87.1. This is a Shanghai AI Lab cross-model evaluation; the source does not disclose this model's exact mode, effort, or sampling." }, { "model_id": "glm-5.2", "benchmark_id": "mmlu_pro", "score": 87.22, "reference_url": "https://huggingface.co/internlm/Intern-S2-Preview-397B", "reported_setting": { "mode": "source does not state", "effort": "source does not state", "tools": "none", "sampling": "source does not state", "judge": "benchmark-specific; exact version not stated", "harness": "OpenCompass or AgentCompass; exact adapter not stated", "prompt_style": "source does not state", "temperature": "source does not state", "context": "source does not state", "notes": "Shanghai AI Lab evaluated all model columns using OpenCompass, VLMEvalKit, and AgentCompass. Per-model mode/effort/sampling are not reported; only Intern-S2 maximum evaluation lengths are disclosed." }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Intern-S2-Preview-397B official model card: MMLU Pro = 87.22. This is a Shanghai AI Lab cross-model evaluation; the source does not disclose this model's exact mode, effort, or sampling." }, { "model_id": "gpt-5.5", "benchmark_id": "mmlu_pro", "score": 88.2, "reference_url": "https://huggingface.co/internlm/Intern-S2-Preview-397B", "reported_setting": { "mode": "source does not state", "effort": "source does not state", "tools": "none", "sampling": "source does not state", "judge": "benchmark-specific; exact version not stated", "harness": "OpenCompass or AgentCompass; exact adapter not stated", "prompt_style": "source does not state", "temperature": "source does not state", "context": "source does not state", "notes": "Shanghai AI Lab evaluated all model columns using OpenCompass, VLMEvalKit, and AgentCompass. Per-model mode/effort/sampling are not reported; only Intern-S2 maximum evaluation lengths are disclosed." }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Intern-S2-Preview-397B official model card: MMLU Pro = 88.2. This is a Shanghai AI Lab cross-model evaluation; the source does not disclose this model's exact mode, effort, or sampling." }, { "model_id": "claude-opus-4.8", "benchmark_id": "mmlu_pro", "score": 90.12, "reference_url": "https://huggingface.co/internlm/Intern-S2-Preview-397B", "reported_setting": { "mode": "source does not state", "effort": "source does not state", "tools": "none", "sampling": "source does not state", "judge": "benchmark-specific; exact version not stated", "harness": "OpenCompass or AgentCompass; exact adapter not stated", "prompt_style": "source does not state", "temperature": "source does not state", "context": "source does not state", "notes": "Shanghai AI Lab evaluated all model columns using OpenCompass, VLMEvalKit, and AgentCompass. Per-model mode/effort/sampling are not reported; only Intern-S2 maximum evaluation lengths are disclosed." }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Intern-S2-Preview-397B official model card: MMLU Pro = 90.12. This is a Shanghai AI Lab cross-model evaluation; the source does not disclose this model's exact mode, effort, or sampling." }, { "model_id": "intern-s2-preview-397b", "benchmark_id": "simpleqa_verified", "score": 69.9, "reference_url": "https://huggingface.co/internlm/Intern-S2-Preview-397B", "reported_setting": { "mode": "thinking (default)", "effort": "source does not state", "tools": "none", "sampling": "source does not state", "judge": "benchmark-specific; exact version not stated", "harness": "OpenCompass or AgentCompass; exact adapter not stated", "prompt_style": "source does not state", "temperature": "source does not state", "context": "max 256K text", "notes": "Shanghai AI Lab evaluated all model columns using OpenCompass, VLMEvalKit, and AgentCompass. Per-model mode/effort/sampling are not reported; only Intern-S2 maximum evaluation lengths are disclosed." }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Intern-S2-Preview-397B official model card: SimpleQA-Verified = 69.9. This is the release model's own reported evaluation." }, { "model_id": "qwen3.5-397b", "benchmark_id": "simpleqa_verified", "score": 54.8, "reference_url": "https://huggingface.co/internlm/Intern-S2-Preview-397B", "reported_setting": { "mode": "source does not state", "effort": "source does not state", "tools": "none", "sampling": "source does not state", "judge": "benchmark-specific; exact version not stated", "harness": "OpenCompass or AgentCompass; exact adapter not stated", "prompt_style": "source does not state", "temperature": "source does not state", "context": "source does not state", "notes": "Shanghai AI Lab evaluated all model columns using OpenCompass, VLMEvalKit, and AgentCompass. Per-model mode/effort/sampling are not reported; only Intern-S2 maximum evaluation lengths are disclosed." }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Intern-S2-Preview-397B official model card: SimpleQA-Verified = 54.8. This is a Shanghai AI Lab cross-model evaluation; the source does not disclose this model's exact mode, effort, or sampling." }, { "model_id": "kimi-k2.7-code", "benchmark_id": "simpleqa_verified", "score": 38.6, "reference_url": "https://huggingface.co/internlm/Intern-S2-Preview-397B", "reported_setting": { "mode": "source does not state", "effort": "source does not state", "tools": "none", "sampling": "source does not state", "judge": "benchmark-specific; exact version not stated", "harness": "OpenCompass or AgentCompass; exact adapter not stated", "prompt_style": "source does not state", "temperature": "source does not state", "context": "source does not state", "notes": "Shanghai AI Lab evaluated all model columns using OpenCompass, VLMEvalKit, and AgentCompass. Per-model mode/effort/sampling are not reported; only Intern-S2 maximum evaluation lengths are disclosed." }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Intern-S2-Preview-397B official model card: SimpleQA-Verified = 38.6. This is a Shanghai AI Lab cross-model evaluation; the source does not disclose this model's exact mode, effort, or sampling." }, { "model_id": "glm-5.2", "benchmark_id": "simpleqa_verified", "score": 37.9, "reference_url": "https://huggingface.co/internlm/Intern-S2-Preview-397B", "reported_setting": { "mode": "source does not state", "effort": "source does not state", "tools": "none", "sampling": "source does not state", "judge": "benchmark-specific; exact version not stated", "harness": "OpenCompass or AgentCompass; exact adapter not stated", "prompt_style": "source does not state", "temperature": "source does not state", "context": "source does not state", "notes": "Shanghai AI Lab evaluated all model columns using OpenCompass, VLMEvalKit, and AgentCompass. Per-model mode/effort/sampling are not reported; only Intern-S2 maximum evaluation lengths are disclosed." }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Intern-S2-Preview-397B official model card: SimpleQA-Verified = 37.9. This is a Shanghai AI Lab cross-model evaluation; the source does not disclose this model's exact mode, effort, or sampling." }, { "model_id": "gpt-5.5", "benchmark_id": "simpleqa_verified", "score": 64.3, "reference_url": "https://huggingface.co/internlm/Intern-S2-Preview-397B", "reported_setting": { "mode": "source does not state", "effort": "source does not state", "tools": "none", "sampling": "source does not state", "judge": "benchmark-specific; exact version not stated", "harness": "OpenCompass or AgentCompass; exact adapter not stated", "prompt_style": "source does not state", "temperature": "source does not state", "context": "source does not state", "notes": "Shanghai AI Lab evaluated all model columns using OpenCompass, VLMEvalKit, and AgentCompass. Per-model mode/effort/sampling are not reported; only Intern-S2 maximum evaluation lengths are disclosed." }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Intern-S2-Preview-397B official model card: SimpleQA-Verified = 64.3. This is a Shanghai AI Lab cross-model evaluation; the source does not disclose this model's exact mode, effort, or sampling." }, { "model_id": "claude-opus-4.8", "benchmark_id": "simpleqa_verified", "score": 43.3, "reference_url": "https://huggingface.co/internlm/Intern-S2-Preview-397B", "reported_setting": { "mode": "source does not state", "effort": "source does not state", "tools": "none", "sampling": "source does not state", "judge": "benchmark-specific; exact version not stated", "harness": "OpenCompass or AgentCompass; exact adapter not stated", "prompt_style": "source does not state", "temperature": "source does not state", "context": "source does not state", "notes": "Shanghai AI Lab evaluated all model columns using OpenCompass, VLMEvalKit, and AgentCompass. Per-model mode/effort/sampling are not reported; only Intern-S2 maximum evaluation lengths are disclosed." }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Intern-S2-Preview-397B official model card: SimpleQA-Verified = 43.3. This is a Shanghai AI Lab cross-model evaluation; the source does not disclose this model's exact mode, effort, or sampling." }, { "model_id": "intern-s2-preview-397b", "benchmark_id": "advancedif", "score": 74.44, "reference_url": "https://huggingface.co/internlm/Intern-S2-Preview-397B", "reported_setting": { "mode": "thinking (default)", "effort": "source does not state", "tools": "none", "sampling": "source does not state", "judge": "benchmark-specific; exact version not stated", "harness": "OpenCompass or AgentCompass; exact adapter not stated", "prompt_style": "source does not state", "temperature": "source does not state", "context": "max 256K text", "notes": "Shanghai AI Lab evaluated all model columns using OpenCompass, VLMEvalKit, and AgentCompass. Per-model mode/effort/sampling are not reported; only Intern-S2 maximum evaluation lengths are disclosed." }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Intern-S2-Preview-397B official model card: AdvancedIF = 74.44. This is the release model's own reported evaluation." }, { "model_id": "qwen3.5-397b", "benchmark_id": "advancedif", "score": 75.49, "reference_url": "https://huggingface.co/internlm/Intern-S2-Preview-397B", "reported_setting": { "mode": "source does not state", "effort": "source does not state", "tools": "none", "sampling": "source does not state", "judge": "benchmark-specific; exact version not stated", "harness": "OpenCompass or AgentCompass; exact adapter not stated", "prompt_style": "source does not state", "temperature": "source does not state", "context": "source does not state", "notes": "Shanghai AI Lab evaluated all model columns using OpenCompass, VLMEvalKit, and AgentCompass. Per-model mode/effort/sampling are not reported; only Intern-S2 maximum evaluation lengths are disclosed." }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Intern-S2-Preview-397B official model card: AdvancedIF = 75.49. This is a Shanghai AI Lab cross-model evaluation; the source does not disclose this model's exact mode, effort, or sampling." }, { "model_id": "deepseek-v4-pro", "benchmark_id": "advancedif", "score": 73.83, "reference_url": "https://huggingface.co/internlm/Intern-S2-Preview-397B", "reported_setting": { "mode": "source does not state", "effort": "source does not state", "tools": "none", "sampling": "source does not state", "judge": "benchmark-specific; exact version not stated", "harness": "OpenCompass or AgentCompass; exact adapter not stated", "prompt_style": "source does not state", "temperature": "source does not state", "context": "source does not state", "notes": "Shanghai AI Lab evaluated all model columns using OpenCompass, VLMEvalKit, and AgentCompass. Per-model mode/effort/sampling are not reported; only Intern-S2 maximum evaluation lengths are disclosed." }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Intern-S2-Preview-397B official model card: AdvancedIF = 73.83. This is a Shanghai AI Lab cross-model evaluation; the source does not disclose this model's exact mode, effort, or sampling." }, { "model_id": "kimi-k2.7-code", "benchmark_id": "advancedif", "score": 76.17, "reference_url": "https://huggingface.co/internlm/Intern-S2-Preview-397B", "reported_setting": { "mode": "source does not state", "effort": "source does not state", "tools": "none", "sampling": "source does not state", "judge": "benchmark-specific; exact version not stated", "harness": "OpenCompass or AgentCompass; exact adapter not stated", "prompt_style": "source does not state", "temperature": "source does not state", "context": "source does not state", "notes": "Shanghai AI Lab evaluated all model columns using OpenCompass, VLMEvalKit, and AgentCompass. Per-model mode/effort/sampling are not reported; only Intern-S2 maximum evaluation lengths are disclosed." }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Intern-S2-Preview-397B official model card: AdvancedIF = 76.17. This is a Shanghai AI Lab cross-model evaluation; the source does not disclose this model's exact mode, effort, or sampling." }, { "model_id": "glm-5.2", "benchmark_id": "advancedif", "score": 75.76, "reference_url": "https://huggingface.co/internlm/Intern-S2-Preview-397B", "reported_setting": { "mode": "source does not state", "effort": "source does not state", "tools": "none", "sampling": "source does not state", "judge": "benchmark-specific; exact version not stated", "harness": "OpenCompass or AgentCompass; exact adapter not stated", "prompt_style": "source does not state", "temperature": "source does not state", "context": "source does not state", "notes": "Shanghai AI Lab evaluated all model columns using OpenCompass, VLMEvalKit, and AgentCompass. Per-model mode/effort/sampling are not reported; only Intern-S2 maximum evaluation lengths are disclosed." }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Intern-S2-Preview-397B official model card: AdvancedIF = 75.76. This is a Shanghai AI Lab cross-model evaluation; the source does not disclose this model's exact mode, effort, or sampling." }, { "model_id": "gpt-5.5", "benchmark_id": "advancedif", "score": 76.2, "reference_url": "https://huggingface.co/internlm/Intern-S2-Preview-397B", "reported_setting": { "mode": "source does not state", "effort": "source does not state", "tools": "none", "sampling": "source does not state", "judge": "benchmark-specific; exact version not stated", "harness": "OpenCompass or AgentCompass; exact adapter not stated", "prompt_style": "source does not state", "temperature": "source does not state", "context": "source does not state", "notes": "Shanghai AI Lab evaluated all model columns using OpenCompass, VLMEvalKit, and AgentCompass. Per-model mode/effort/sampling are not reported; only Intern-S2 maximum evaluation lengths are disclosed." }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Intern-S2-Preview-397B official model card: AdvancedIF = 76.2. This is a Shanghai AI Lab cross-model evaluation; the source does not disclose this model's exact mode, effort, or sampling." }, { "model_id": "gemini-3.1-pro", "benchmark_id": "advancedif", "score": 79.78, "reference_url": "https://huggingface.co/internlm/Intern-S2-Preview-397B", "reported_setting": { "mode": "source does not state", "effort": "source does not state", "tools": "none", "sampling": "source does not state", "judge": "benchmark-specific; exact version not stated", "harness": "OpenCompass or AgentCompass; exact adapter not stated", "prompt_style": "source does not state", "temperature": "source does not state", "context": "source does not state", "notes": "Shanghai AI Lab evaluated all model columns using OpenCompass, VLMEvalKit, and AgentCompass. Per-model mode/effort/sampling are not reported; only Intern-S2 maximum evaluation lengths are disclosed." }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Intern-S2-Preview-397B official model card: AdvancedIF = 79.78. This is a Shanghai AI Lab cross-model evaluation; the source does not disclose this model's exact mode, effort, or sampling." }, { "model_id": "claude-opus-4.8", "benchmark_id": "advancedif", "score": 72.88, "reference_url": "https://huggingface.co/internlm/Intern-S2-Preview-397B", "reported_setting": { "mode": "source does not state", "effort": "source does not state", "tools": "none", "sampling": "source does not state", "judge": "benchmark-specific; exact version not stated", "harness": "OpenCompass or AgentCompass; exact adapter not stated", "prompt_style": "source does not state", "temperature": "source does not state", "context": "source does not state", "notes": "Shanghai AI Lab evaluated all model columns using OpenCompass, VLMEvalKit, and AgentCompass. Per-model mode/effort/sampling are not reported; only Intern-S2 maximum evaluation lengths are disclosed." }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Intern-S2-Preview-397B official model card: AdvancedIF = 72.88. This is a Shanghai AI Lab cross-model evaluation; the source does not disclose this model's exact mode, effort, or sampling." }, { "model_id": "intern-s2-preview-397b", "benchmark_id": "hmmt_feb_2026", "score": 91.57, "reference_url": "https://huggingface.co/internlm/Intern-S2-Preview-397B", "reported_setting": { "mode": "thinking (default)", "effort": "source does not state", "tools": "none", "sampling": "source does not state", "judge": "benchmark-specific; exact version not stated", "harness": "OpenCompass or AgentCompass; exact adapter not stated", "prompt_style": "source does not state", "temperature": "source does not state", "context": "max 256K text", "notes": "Shanghai AI Lab evaluated all model columns using OpenCompass, VLMEvalKit, and AgentCompass. Per-model mode/effort/sampling are not reported; only Intern-S2 maximum evaluation lengths are disclosed." }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Intern-S2-Preview-397B official model card: HMMT-2026 = 91.57. This is the release model's own reported evaluation." }, { "model_id": "kimi-k2.7-code", "benchmark_id": "hmmt_feb_2026", "score": 90.34, "reference_url": "https://huggingface.co/internlm/Intern-S2-Preview-397B", "reported_setting": { "mode": "source does not state", "effort": "source does not state", "tools": "none", "sampling": "source does not state", "judge": "benchmark-specific; exact version not stated", "harness": "OpenCompass or AgentCompass; exact adapter not stated", "prompt_style": "source does not state", "temperature": "source does not state", "context": "source does not state", "notes": "Shanghai AI Lab evaluated all model columns using OpenCompass, VLMEvalKit, and AgentCompass. Per-model mode/effort/sampling are not reported; only Intern-S2 maximum evaluation lengths are disclosed." }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Intern-S2-Preview-397B official model card: HMMT-2026 = 90.34. This is a Shanghai AI Lab cross-model evaluation; the source does not disclose this model's exact mode, effort, or sampling." }, { "model_id": "intern-s2-preview-397b", "benchmark_id": "mmmu_pro", "score": 80.46, "reference_url": "https://huggingface.co/internlm/Intern-S2-Preview-397B", "reported_setting": { "mode": "thinking (default)", "effort": "source does not state", "tools": "none", "sampling": "source does not state", "judge": "benchmark-specific; exact version not stated", "harness": "VLMEvalKit or AgentCompass; exact adapter not stated", "prompt_style": "source does not state", "temperature": "source does not state", "context": "max 64K multimodal", "notes": "Shanghai AI Lab evaluated all model columns using OpenCompass, VLMEvalKit, and AgentCompass. Per-model mode/effort/sampling are not reported; only Intern-S2 maximum evaluation lengths are disclosed." }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Intern-S2-Preview-397B official model card: MMMU Pro = 80.46. This is the release model's own reported evaluation." }, { "model_id": "kimi-k2.7-code", "benchmark_id": "mmmu_pro", "score": 77.92, "reference_url": "https://huggingface.co/internlm/Intern-S2-Preview-397B", "reported_setting": { "mode": "source does not state", "effort": "source does not state", "tools": "none", "sampling": "source does not state", "judge": "benchmark-specific; exact version not stated", "harness": "VLMEvalKit or AgentCompass; exact adapter not stated", "prompt_style": "source does not state", "temperature": "source does not state", "context": "source does not state", "notes": "Shanghai AI Lab evaluated all model columns using OpenCompass, VLMEvalKit, and AgentCompass. Per-model mode/effort/sampling are not reported; only Intern-S2 maximum evaluation lengths are disclosed." }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Intern-S2-Preview-397B official model card: MMMU Pro = 77.92. This is a Shanghai AI Lab cross-model evaluation; the source does not disclose this model's exact mode, effort, or sampling." }, { "model_id": "intern-s2-preview-397b", "benchmark_id": "chartqapro", "score": 69.65, "reference_url": "https://huggingface.co/internlm/Intern-S2-Preview-397B", "reported_setting": { "mode": "thinking (default)", "effort": "source does not state", "tools": "none", "sampling": "source does not state", "judge": "benchmark-specific; exact version not stated", "harness": "VLMEvalKit or AgentCompass; exact adapter not stated", "prompt_style": "source does not state", "temperature": "source does not state", "context": "max 64K multimodal", "notes": "Shanghai AI Lab evaluated all model columns using OpenCompass, VLMEvalKit, and AgentCompass. Per-model mode/effort/sampling are not reported; only Intern-S2 maximum evaluation lengths are disclosed." }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Intern-S2-Preview-397B official model card: ChartQAPro = 69.65. This is the release model's own reported evaluation." }, { "model_id": "qwen3.5-397b", "benchmark_id": "chartqapro", "score": 68.61, "reference_url": "https://huggingface.co/internlm/Intern-S2-Preview-397B", "reported_setting": { "mode": "source does not state", "effort": "source does not state", "tools": "none", "sampling": "source does not state", "judge": "benchmark-specific; exact version not stated", "harness": "VLMEvalKit or AgentCompass; exact adapter not stated", "prompt_style": "source does not state", "temperature": "source does not state", "context": "source does not state", "notes": "Shanghai AI Lab evaluated all model columns using OpenCompass, VLMEvalKit, and AgentCompass. Per-model mode/effort/sampling are not reported; only Intern-S2 maximum evaluation lengths are disclosed." }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Intern-S2-Preview-397B official model card: ChartQAPro = 68.61. This is a Shanghai AI Lab cross-model evaluation; the source does not disclose this model's exact mode, effort, or sampling." }, { "model_id": "kimi-k2.7-code", "benchmark_id": "chartqapro", "score": 54.86, "reference_url": "https://huggingface.co/internlm/Intern-S2-Preview-397B", "reported_setting": { "mode": "source does not state", "effort": "source does not state", "tools": "none", "sampling": "source does not state", "judge": "benchmark-specific; exact version not stated", "harness": "VLMEvalKit or AgentCompass; exact adapter not stated", "prompt_style": "source does not state", "temperature": "source does not state", "context": "source does not state", "notes": "Shanghai AI Lab evaluated all model columns using OpenCompass, VLMEvalKit, and AgentCompass. Per-model mode/effort/sampling are not reported; only Intern-S2 maximum evaluation lengths are disclosed." }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Intern-S2-Preview-397B official model card: ChartQAPro = 54.86. This is a Shanghai AI Lab cross-model evaluation; the source does not disclose this model's exact mode, effort, or sampling." }, { "model_id": "gpt-5.5", "benchmark_id": "chartqapro", "score": 69.23, "reference_url": "https://huggingface.co/internlm/Intern-S2-Preview-397B", "reported_setting": { "mode": "source does not state", "effort": "source does not state", "tools": "none", "sampling": "source does not state", "judge": "benchmark-specific; exact version not stated", "harness": "VLMEvalKit or AgentCompass; exact adapter not stated", "prompt_style": "source does not state", "temperature": "source does not state", "context": "source does not state", "notes": "Shanghai AI Lab evaluated all model columns using OpenCompass, VLMEvalKit, and AgentCompass. Per-model mode/effort/sampling are not reported; only Intern-S2 maximum evaluation lengths are disclosed." }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Intern-S2-Preview-397B official model card: ChartQAPro = 69.23. This is a Shanghai AI Lab cross-model evaluation; the source does not disclose this model's exact mode, effort, or sampling.", "candidates": [ { "score": 69.4, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2.1/Seed2_1_Model_Card.pdf", "source_type": "tech_report", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "none", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official", "prompt_style": "official default", "temperature": "unknown" }, "notes": "Model card Table 9 (p51)." } ] }, { "model_id": "gemini-3.1-pro", "benchmark_id": "chartqapro", "score": 71.18, "reference_url": "https://huggingface.co/internlm/Intern-S2-Preview-397B", "reported_setting": { "mode": "source does not state", "effort": "source does not state", "tools": "none", "sampling": "source does not state", "judge": "benchmark-specific; exact version not stated", "harness": "VLMEvalKit or AgentCompass; exact adapter not stated", "prompt_style": "source does not state", "temperature": "source does not state", "context": "source does not state", "notes": "Shanghai AI Lab evaluated all model columns using OpenCompass, VLMEvalKit, and AgentCompass. Per-model mode/effort/sampling are not reported; only Intern-S2 maximum evaluation lengths are disclosed." }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Intern-S2-Preview-397B official model card: ChartQAPro = 71.18. This is a Shanghai AI Lab cross-model evaluation; the source does not disclose this model's exact mode, effort, or sampling.", "candidates": [ { "score": 70.2, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2.1/Seed2_1_Model_Card.pdf", "source_type": "tech_report", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "none", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official", "prompt_style": "official default", "temperature": "unknown" }, "notes": "Model card Table 9 (p51)." } ] }, { "model_id": "claude-opus-4.8", "benchmark_id": "chartqapro", "score": 58.65, "reference_url": "https://huggingface.co/internlm/Intern-S2-Preview-397B", "reported_setting": { "mode": "source does not state", "effort": "source does not state", "tools": "none", "sampling": "source does not state", "judge": "benchmark-specific; exact version not stated", "harness": "VLMEvalKit or AgentCompass; exact adapter not stated", "prompt_style": "source does not state", "temperature": "source does not state", "context": "source does not state", "notes": "Shanghai AI Lab evaluated all model columns using OpenCompass, VLMEvalKit, and AgentCompass. Per-model mode/effort/sampling are not reported; only Intern-S2 maximum evaluation lengths are disclosed." }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Intern-S2-Preview-397B official model card: ChartQAPro = 58.65. This is a Shanghai AI Lab cross-model evaluation; the source does not disclose this model's exact mode, effort, or sampling." }, { "model_id": "intern-s2-preview-397b", "benchmark_id": "terminal_bench_2_1", "score": 67.42, "reference_url": "https://huggingface.co/internlm/Intern-S2-Preview-397B", "reported_setting": { "mode": "thinking (default)", "effort": "source does not state", "tools": "terminal/code execution", "sampling": "source does not state", "judge": "benchmark-specific; exact version not stated", "harness": "Terminus 2", "prompt_style": "source does not state", "temperature": "source does not state", "context": "max 256K text", "notes": "Shanghai AI Lab evaluated all model columns using OpenCompass, VLMEvalKit, and AgentCompass. Per-model mode/effort/sampling are not reported; only Intern-S2 maximum evaluation lengths are disclosed." }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Intern-S2-Preview-397B official model card: TerminalBench 2.1 = 67.42. This is the release model's own reported evaluation." }, { "model_id": "qwen3.5-397b", "benchmark_id": "terminal_bench_2_1", "score": 51.3, "reference_url": "https://huggingface.co/internlm/Intern-S2-Preview-397B", "reported_setting": { "mode": "source does not state", "effort": "source does not state", "tools": "terminal/code execution", "sampling": "source does not state", "judge": "benchmark-specific; exact version not stated", "harness": "Terminus 2", "prompt_style": "source does not state", "temperature": "source does not state", "context": "source does not state", "notes": "Shanghai AI Lab evaluated all model columns using OpenCompass, VLMEvalKit, and AgentCompass. Per-model mode/effort/sampling are not reported; only Intern-S2 maximum evaluation lengths are disclosed." }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Intern-S2-Preview-397B official model card: TerminalBench 2.1 = 51.3. This is a Shanghai AI Lab cross-model evaluation; the source does not disclose this model's exact mode, effort, or sampling.", "candidates": [ { "score": 49.9, "reference_url": "https://research.nvidia.com/labs/nemotron/files/NVIDIA-Nemotron-3-Ultra-Technical-Report.pdf", "source_type": "tech_report", "reported_setting": { "version_metric": "v2.1 / % resolved", "tools": "terminal and code execution", "harness": "Harbor / Terminus-2", "sampling": "8-run average" }, "notes": "NVIDIA Nemotron 3 Ultra technical report, Final Table 10: Qwen3.5-397B-A17B: terminal_bench_2_1=49.9." } ] }, { "model_id": "kimi-k2.7-code", "benchmark_id": "terminal_bench_2_1", "score": 66.29, "reference_url": "https://huggingface.co/internlm/Intern-S2-Preview-397B", "reported_setting": { "mode": "source does not state", "effort": "source does not state", "tools": "terminal/code execution", "sampling": "source does not state", "judge": "benchmark-specific; exact version not stated", "harness": "Terminus 2", "prompt_style": "source does not state", "temperature": "source does not state", "context": "source does not state", "notes": "Shanghai AI Lab evaluated all model columns using OpenCompass, VLMEvalKit, and AgentCompass. Per-model mode/effort/sampling are not reported; only Intern-S2 maximum evaluation lengths are disclosed." }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Intern-S2-Preview-397B official model card: TerminalBench 2.1 = 66.29. This is a Shanghai AI Lab cross-model evaluation; the source does not disclose this model's exact mode, effort, or sampling." }, { "model_id": "intern-s2-preview-397b", "benchmark_id": "swe_bench_pro", "score": 61.56, "reference_url": "https://huggingface.co/internlm/Intern-S2-Preview-397B", "reported_setting": { "mode": "thinking (default)", "effort": "source does not state", "tools": "agentic code execution", "sampling": "source does not state", "judge": "benchmark-specific; exact version not stated", "harness": "Mini-SWE-Agent", "prompt_style": "source does not state", "temperature": "source does not state", "context": "max 256K text", "notes": "Shanghai AI Lab evaluated all model columns using OpenCompass, VLMEvalKit, and AgentCompass. Per-model mode/effort/sampling are not reported; only Intern-S2 maximum evaluation lengths are disclosed." }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Intern-S2-Preview-397B official model card: SWE-Bench-Pro = 61.56. This is the release model's own reported evaluation." }, { "model_id": "qwen3.5-397b", "benchmark_id": "swe_bench_pro", "score": 43.55, "reference_url": "https://huggingface.co/internlm/Intern-S2-Preview-397B", "reported_setting": { "mode": "source does not state", "effort": "source does not state", "tools": "agentic code execution", "sampling": "source does not state", "judge": "benchmark-specific; exact version not stated", "harness": "Mini-SWE-Agent", "prompt_style": "source does not state", "temperature": "source does not state", "context": "source does not state", "notes": "Shanghai AI Lab evaluated all model columns using OpenCompass, VLMEvalKit, and AgentCompass. Per-model mode/effort/sampling are not reported; only Intern-S2 maximum evaluation lengths are disclosed." }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Intern-S2-Preview-397B official model card: SWE-Bench-Pro = 43.55. This is a Shanghai AI Lab cross-model evaluation; the source does not disclose this model's exact mode, effort, or sampling." }, { "model_id": "kimi-k2.7-code", "benchmark_id": "swe_bench_pro", "score": 57.59, "reference_url": "https://huggingface.co/internlm/Intern-S2-Preview-397B", "reported_setting": { "mode": "source does not state", "effort": "source does not state", "tools": "agentic code execution", "sampling": "source does not state", "judge": "benchmark-specific; exact version not stated", "harness": "Mini-SWE-Agent", "prompt_style": "source does not state", "temperature": "source does not state", "context": "source does not state", "notes": "Shanghai AI Lab evaluated all model columns using OpenCompass, VLMEvalKit, and AgentCompass. Per-model mode/effort/sampling are not reported; only Intern-S2 maximum evaluation lengths are disclosed." }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Intern-S2-Preview-397B official model card: SWE-Bench-Pro = 57.59. This is a Shanghai AI Lab cross-model evaluation; the source does not disclose this model's exact mode, effort, or sampling." }, { "model_id": "intern-s2-preview-397b", "benchmark_id": "swe_bench_multilingual", "score": 81.67, "reference_url": "https://huggingface.co/internlm/Intern-S2-Preview-397B", "reported_setting": { "mode": "thinking (default)", "effort": "source does not state", "tools": "agentic code execution", "sampling": "source does not state", "judge": "benchmark-specific; exact version not stated", "harness": "Mini-SWE-Agent", "prompt_style": "source does not state", "temperature": "source does not state", "context": "max 256K text", "notes": "Shanghai AI Lab evaluated all model columns using OpenCompass, VLMEvalKit, and AgentCompass. Per-model mode/effort/sampling are not reported; only Intern-S2 maximum evaluation lengths are disclosed." }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Intern-S2-Preview-397B official model card: SWE-Bench-Multilingual = 81.67. This is the release model's own reported evaluation." }, { "model_id": "kimi-k2.7-code", "benchmark_id": "swe_bench_multilingual", "score": 78.56, "reference_url": "https://huggingface.co/internlm/Intern-S2-Preview-397B", "reported_setting": { "mode": "source does not state", "effort": "source does not state", "tools": "agentic code execution", "sampling": "source does not state", "judge": "benchmark-specific; exact version not stated", "harness": "Mini-SWE-Agent", "prompt_style": "source does not state", "temperature": "source does not state", "context": "source does not state", "notes": "Shanghai AI Lab evaluated all model columns using OpenCompass, VLMEvalKit, and AgentCompass. Per-model mode/effort/sampling are not reported; only Intern-S2 maximum evaluation lengths are disclosed." }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Intern-S2-Preview-397B official model card: SWE-Bench-Multilingual = 78.56. This is a Shanghai AI Lab cross-model evaluation; the source does not disclose this model's exact mode, effort, or sampling." }, { "model_id": "intern-s2-preview-397b", "benchmark_id": "xlrs_bench_micro", "score": 51.97, "reference_url": "https://huggingface.co/internlm/Intern-S2-Preview-397B", "reported_setting": { "mode": "thinking (default)", "effort": "source does not state", "tools": "none", "sampling": "source does not state", "judge": "benchmark-specific; exact version not stated", "harness": "VLMEvalKit or AgentCompass; exact adapter not stated", "prompt_style": "source does not state", "temperature": "source does not state", "context": "max 64K multimodal", "notes": "Shanghai AI Lab evaluated all model columns using OpenCompass, VLMEvalKit, and AgentCompass. Per-model mode/effort/sampling are not reported; only Intern-S2 maximum evaluation lengths are disclosed." }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Intern-S2-Preview-397B official model card: XLRS-Bench = 51.97. This is the release model's own reported evaluation." }, { "model_id": "qwen3.5-397b", "benchmark_id": "xlrs_bench_micro", "score": 50.11, "reference_url": "https://huggingface.co/internlm/Intern-S2-Preview-397B", "reported_setting": { "mode": "source does not state", "effort": "source does not state", "tools": "none", "sampling": "source does not state", "judge": "benchmark-specific; exact version not stated", "harness": "VLMEvalKit or AgentCompass; exact adapter not stated", "prompt_style": "source does not state", "temperature": "source does not state", "context": "source does not state", "notes": "Shanghai AI Lab evaluated all model columns using OpenCompass, VLMEvalKit, and AgentCompass. Per-model mode/effort/sampling are not reported; only Intern-S2 maximum evaluation lengths are disclosed." }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Intern-S2-Preview-397B official model card: XLRS-Bench = 50.11. This is a Shanghai AI Lab cross-model evaluation; the source does not disclose this model's exact mode, effort, or sampling." }, { "model_id": "kimi-k2.7-code", "benchmark_id": "xlrs_bench_micro", "score": 49.9, "reference_url": "https://huggingface.co/internlm/Intern-S2-Preview-397B", "reported_setting": { "mode": "source does not state", "effort": "source does not state", "tools": "none", "sampling": "source does not state", "judge": "benchmark-specific; exact version not stated", "harness": "VLMEvalKit or AgentCompass; exact adapter not stated", "prompt_style": "source does not state", "temperature": "source does not state", "context": "source does not state", "notes": "Shanghai AI Lab evaluated all model columns using OpenCompass, VLMEvalKit, and AgentCompass. Per-model mode/effort/sampling are not reported; only Intern-S2 maximum evaluation lengths are disclosed." }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Intern-S2-Preview-397B official model card: XLRS-Bench = 49.9. This is a Shanghai AI Lab cross-model evaluation; the source does not disclose this model's exact mode, effort, or sampling." }, { "model_id": "gpt-5.5", "benchmark_id": "xlrs_bench_micro", "score": 50.96, "reference_url": "https://huggingface.co/internlm/Intern-S2-Preview-397B", "reported_setting": { "mode": "source does not state", "effort": "source does not state", "tools": "none", "sampling": "source does not state", "judge": "benchmark-specific; exact version not stated", "harness": "VLMEvalKit or AgentCompass; exact adapter not stated", "prompt_style": "source does not state", "temperature": "source does not state", "context": "source does not state", "notes": "Shanghai AI Lab evaluated all model columns using OpenCompass, VLMEvalKit, and AgentCompass. Per-model mode/effort/sampling are not reported; only Intern-S2 maximum evaluation lengths are disclosed." }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Intern-S2-Preview-397B official model card: XLRS-Bench = 50.96. This is a Shanghai AI Lab cross-model evaluation; the source does not disclose this model's exact mode, effort, or sampling." }, { "model_id": "gemini-3.1-pro", "benchmark_id": "xlrs_bench_micro", "score": 54.27, "reference_url": "https://huggingface.co/internlm/Intern-S2-Preview-397B", "reported_setting": { "mode": "source does not state", "effort": "source does not state", "tools": "none", "sampling": "source does not state", "judge": "benchmark-specific; exact version not stated", "harness": "VLMEvalKit or AgentCompass; exact adapter not stated", "prompt_style": "source does not state", "temperature": "source does not state", "context": "source does not state", "notes": "Shanghai AI Lab evaluated all model columns using OpenCompass, VLMEvalKit, and AgentCompass. Per-model mode/effort/sampling are not reported; only Intern-S2 maximum evaluation lengths are disclosed." }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Intern-S2-Preview-397B official model card: XLRS-Bench = 54.27. This is a Shanghai AI Lab cross-model evaluation; the source does not disclose this model's exact mode, effort, or sampling." }, { "model_id": "claude-opus-4.8", "benchmark_id": "xlrs_bench_micro", "score": 51.84, "reference_url": "https://huggingface.co/internlm/Intern-S2-Preview-397B", "reported_setting": { "mode": "source does not state", "effort": "source does not state", "tools": "none", "sampling": "source does not state", "judge": "benchmark-specific; exact version not stated", "harness": "VLMEvalKit or AgentCompass; exact adapter not stated", "prompt_style": "source does not state", "temperature": "source does not state", "context": "source does not state", "notes": "Shanghai AI Lab evaluated all model columns using OpenCompass, VLMEvalKit, and AgentCompass. Per-model mode/effort/sampling are not reported; only Intern-S2 maximum evaluation lengths are disclosed." }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Intern-S2-Preview-397B official model card: XLRS-Bench = 51.84. This is a Shanghai AI Lab cross-model evaluation; the source does not disclose this model's exact mode, effort, or sampling." }, { "model_id": "intern-s2-preview-397b", "benchmark_id": "microvqa", "score": 68.81, "reference_url": "https://huggingface.co/internlm/Intern-S2-Preview-397B", "reported_setting": { "mode": "thinking (default)", "effort": "source does not state", "tools": "none", "sampling": "source does not state", "judge": "benchmark-specific; exact version not stated", "harness": "VLMEvalKit or AgentCompass; exact adapter not stated", "prompt_style": "source does not state", "temperature": "source does not state", "context": "max 64K multimodal", "notes": "Shanghai AI Lab evaluated all model columns using OpenCompass, VLMEvalKit, and AgentCompass. Per-model mode/effort/sampling are not reported; only Intern-S2 maximum evaluation lengths are disclosed." }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Intern-S2-Preview-397B official model card: MicroVQA = 68.81. This is the release model's own reported evaluation." }, { "model_id": "qwen3.5-397b", "benchmark_id": "microvqa", "score": 68.71, "reference_url": "https://huggingface.co/internlm/Intern-S2-Preview-397B", "reported_setting": { "mode": "source does not state", "effort": "source does not state", "tools": "none", "sampling": "source does not state", "judge": "benchmark-specific; exact version not stated", "harness": "VLMEvalKit or AgentCompass; exact adapter not stated", "prompt_style": "source does not state", "temperature": "source does not state", "context": "source does not state", "notes": "Shanghai AI Lab evaluated all model columns using OpenCompass, VLMEvalKit, and AgentCompass. Per-model mode/effort/sampling are not reported; only Intern-S2 maximum evaluation lengths are disclosed." }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Intern-S2-Preview-397B official model card: MicroVQA = 68.71. This is a Shanghai AI Lab cross-model evaluation; the source does not disclose this model's exact mode, effort, or sampling." }, { "model_id": "kimi-k2.7-code", "benchmark_id": "microvqa", "score": 61.04, "reference_url": "https://huggingface.co/internlm/Intern-S2-Preview-397B", "reported_setting": { "mode": "source does not state", "effort": "source does not state", "tools": "none", "sampling": "source does not state", "judge": "benchmark-specific; exact version not stated", "harness": "VLMEvalKit or AgentCompass; exact adapter not stated", "prompt_style": "source does not state", "temperature": "source does not state", "context": "source does not state", "notes": "Shanghai AI Lab evaluated all model columns using OpenCompass, VLMEvalKit, and AgentCompass. Per-model mode/effort/sampling are not reported; only Intern-S2 maximum evaluation lengths are disclosed." }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Intern-S2-Preview-397B official model card: MicroVQA = 61.04. This is a Shanghai AI Lab cross-model evaluation; the source does not disclose this model's exact mode, effort, or sampling." }, { "model_id": "gpt-5.5", "benchmark_id": "microvqa", "score": 63.63, "reference_url": "https://huggingface.co/internlm/Intern-S2-Preview-397B", "reported_setting": { "mode": "source does not state", "effort": "source does not state", "tools": "none", "sampling": "source does not state", "judge": "benchmark-specific; exact version not stated", "harness": "VLMEvalKit or AgentCompass; exact adapter not stated", "prompt_style": "source does not state", "temperature": "source does not state", "context": "source does not state", "notes": "Shanghai AI Lab evaluated all model columns using OpenCompass, VLMEvalKit, and AgentCompass. Per-model mode/effort/sampling are not reported; only Intern-S2 maximum evaluation lengths are disclosed." }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Intern-S2-Preview-397B official model card: MicroVQA = 63.63. This is a Shanghai AI Lab cross-model evaluation; the source does not disclose this model's exact mode, effort, or sampling." }, { "model_id": "gemini-3.1-pro", "benchmark_id": "microvqa", "score": 71.02, "reference_url": "https://huggingface.co/internlm/Intern-S2-Preview-397B", "reported_setting": { "mode": "source does not state", "effort": "source does not state", "tools": "none", "sampling": "source does not state", "judge": "benchmark-specific; exact version not stated", "harness": "VLMEvalKit or AgentCompass; exact adapter not stated", "prompt_style": "source does not state", "temperature": "source does not state", "context": "source does not state", "notes": "Shanghai AI Lab evaluated all model columns using OpenCompass, VLMEvalKit, and AgentCompass. Per-model mode/effort/sampling are not reported; only Intern-S2 maximum evaluation lengths are disclosed." }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Intern-S2-Preview-397B official model card: MicroVQA = 71.02. This is a Shanghai AI Lab cross-model evaluation; the source does not disclose this model's exact mode, effort, or sampling." }, { "model_id": "claude-opus-4.8", "benchmark_id": "microvqa", "score": 61.8, "reference_url": "https://huggingface.co/internlm/Intern-S2-Preview-397B", "reported_setting": { "mode": "source does not state", "effort": "source does not state", "tools": "none", "sampling": "source does not state", "judge": "benchmark-specific; exact version not stated", "harness": "VLMEvalKit or AgentCompass; exact adapter not stated", "prompt_style": "source does not state", "temperature": "source does not state", "context": "source does not state", "notes": "Shanghai AI Lab evaluated all model columns using OpenCompass, VLMEvalKit, and AgentCompass. Per-model mode/effort/sampling are not reported; only Intern-S2 maximum evaluation lengths are disclosed." }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Intern-S2-Preview-397B official model card: MicroVQA = 61.8. This is a Shanghai AI Lab cross-model evaluation; the source does not disclose this model's exact mode, effort, or sampling." }, { "model_id": "intern-s2-preview-397b", "benchmark_id": "sfe", "score": 61.67, "reference_url": "https://huggingface.co/internlm/Intern-S2-Preview-397B", "reported_setting": { "mode": "thinking (default)", "effort": "source does not state", "tools": "none", "sampling": "source does not state", "judge": "benchmark-specific; exact version not stated", "harness": "VLMEvalKit or AgentCompass; exact adapter not stated", "prompt_style": "source does not state", "temperature": "source does not state", "context": "max 64K multimodal", "notes": "Shanghai AI Lab evaluated all model columns using OpenCompass, VLMEvalKit, and AgentCompass. Per-model mode/effort/sampling are not reported; only Intern-S2 maximum evaluation lengths are disclosed." }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Intern-S2-Preview-397B official model card: SFE = 61.67. This is the release model's own reported evaluation." }, { "model_id": "qwen3.5-397b", "benchmark_id": "sfe", "score": 62.97, "reference_url": "https://huggingface.co/internlm/Intern-S2-Preview-397B", "reported_setting": { "mode": "source does not state", "effort": "source does not state", "tools": "none", "sampling": "source does not state", "judge": "benchmark-specific; exact version not stated", "harness": "VLMEvalKit or AgentCompass; exact adapter not stated", "prompt_style": "source does not state", "temperature": "source does not state", "context": "source does not state", "notes": "Shanghai AI Lab evaluated all model columns using OpenCompass, VLMEvalKit, and AgentCompass. Per-model mode/effort/sampling are not reported; only Intern-S2 maximum evaluation lengths are disclosed." }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Intern-S2-Preview-397B official model card: SFE = 62.97. This is a Shanghai AI Lab cross-model evaluation; the source does not disclose this model's exact mode, effort, or sampling." }, { "model_id": "kimi-k2.7-code", "benchmark_id": "sfe", "score": 50.76, "reference_url": "https://huggingface.co/internlm/Intern-S2-Preview-397B", "reported_setting": { "mode": "source does not state", "effort": "source does not state", "tools": "none", "sampling": "source does not state", "judge": "benchmark-specific; exact version not stated", "harness": "VLMEvalKit or AgentCompass; exact adapter not stated", "prompt_style": "source does not state", "temperature": "source does not state", "context": "source does not state", "notes": "Shanghai AI Lab evaluated all model columns using OpenCompass, VLMEvalKit, and AgentCompass. Per-model mode/effort/sampling are not reported; only Intern-S2 maximum evaluation lengths are disclosed." }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Intern-S2-Preview-397B official model card: SFE = 50.76. This is a Shanghai AI Lab cross-model evaluation; the source does not disclose this model's exact mode, effort, or sampling." }, { "model_id": "gpt-5.5", "benchmark_id": "sfe", "score": 52.09, "reference_url": "https://huggingface.co/internlm/Intern-S2-Preview-397B", "reported_setting": { "mode": "source does not state", "effort": "source does not state", "tools": "none", "sampling": "source does not state", "judge": "benchmark-specific; exact version not stated", "harness": "VLMEvalKit or AgentCompass; exact adapter not stated", "prompt_style": "source does not state", "temperature": "source does not state", "context": "source does not state", "notes": "Shanghai AI Lab evaluated all model columns using OpenCompass, VLMEvalKit, and AgentCompass. Per-model mode/effort/sampling are not reported; only Intern-S2 maximum evaluation lengths are disclosed." }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Intern-S2-Preview-397B official model card: SFE = 52.09. This is a Shanghai AI Lab cross-model evaluation; the source does not disclose this model's exact mode, effort, or sampling." }, { "model_id": "gemini-3.1-pro", "benchmark_id": "sfe", "score": 59.57, "reference_url": "https://huggingface.co/internlm/Intern-S2-Preview-397B", "reported_setting": { "mode": "source does not state", "effort": "source does not state", "tools": "none", "sampling": "source does not state", "judge": "benchmark-specific; exact version not stated", "harness": "VLMEvalKit or AgentCompass; exact adapter not stated", "prompt_style": "source does not state", "temperature": "source does not state", "context": "source does not state", "notes": "Shanghai AI Lab evaluated all model columns using OpenCompass, VLMEvalKit, and AgentCompass. Per-model mode/effort/sampling are not reported; only Intern-S2 maximum evaluation lengths are disclosed." }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Intern-S2-Preview-397B official model card: SFE = 59.57. This is a Shanghai AI Lab cross-model evaluation; the source does not disclose this model's exact mode, effort, or sampling." }, { "model_id": "claude-opus-4.8", "benchmark_id": "sfe", "score": 59.08, "reference_url": "https://huggingface.co/internlm/Intern-S2-Preview-397B", "reported_setting": { "mode": "source does not state", "effort": "source does not state", "tools": "none", "sampling": "source does not state", "judge": "benchmark-specific; exact version not stated", "harness": "VLMEvalKit or AgentCompass; exact adapter not stated", "prompt_style": "source does not state", "temperature": "source does not state", "context": "source does not state", "notes": "Shanghai AI Lab evaluated all model columns using OpenCompass, VLMEvalKit, and AgentCompass. Per-model mode/effort/sampling are not reported; only Intern-S2 maximum evaluation lengths are disclosed." }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Intern-S2-Preview-397B official model card: SFE = 59.08. This is a Shanghai AI Lab cross-model evaluation; the source does not disclose this model's exact mode, effort, or sampling." }, { "model_id": "intern-s2-preview-397b", "benchmark_id": "scicode", "score": 49.11, "reference_url": "https://huggingface.co/internlm/Intern-S2-Preview-397B", "reported_setting": { "mode": "thinking (default)", "effort": "source does not state", "tools": "code execution", "sampling": "source does not state", "judge": "benchmark-specific; exact version not stated", "harness": "AgentCompass / benchmark evaluator", "prompt_style": "source does not state", "temperature": "source does not state", "context": "max 256K text", "notes": "Shanghai AI Lab evaluated all model columns using OpenCompass, VLMEvalKit, and AgentCompass. Per-model mode/effort/sampling are not reported; only Intern-S2 maximum evaluation lengths are disclosed." }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Intern-S2-Preview-397B official model card: SciCode = 49.11. This is the release model's own reported evaluation." }, { "model_id": "qwen3.5-397b", "benchmark_id": "scicode", "score": 46.35, "reference_url": "https://huggingface.co/internlm/Intern-S2-Preview-397B", "reported_setting": { "mode": "source does not state", "effort": "source does not state", "tools": "code execution", "sampling": "source does not state", "judge": "benchmark-specific; exact version not stated", "harness": "AgentCompass / benchmark evaluator", "prompt_style": "source does not state", "temperature": "source does not state", "context": "source does not state", "notes": "Shanghai AI Lab evaluated all model columns using OpenCompass, VLMEvalKit, and AgentCompass. Per-model mode/effort/sampling are not reported; only Intern-S2 maximum evaluation lengths are disclosed." }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Intern-S2-Preview-397B official model card: SciCode = 46.35. This is a Shanghai AI Lab cross-model evaluation; the source does not disclose this model's exact mode, effort, or sampling.", "candidates": [ { "score": 48.0, "reference_url": "https://research.nvidia.com/labs/nemotron/files/NVIDIA-Nemotron-3-Ultra-Technical-Report.pdf", "source_type": "tech_report", "reported_setting": { "version_metric": "subtask / %", "tools": "none", "harness": "NeMo Evaluator", "sampling": "8-run average" }, "notes": "NVIDIA Nemotron 3 Ultra technical report, Final Table 10: Qwen3.5-397B-A17B: scicode=48." } ] }, { "model_id": "deepseek-v4-pro", "benchmark_id": "scicode", "score": 47.53, "reference_url": "https://huggingface.co/internlm/Intern-S2-Preview-397B", "reported_setting": { "mode": "source does not state", "effort": "source does not state", "tools": "code execution", "sampling": "source does not state", "judge": "benchmark-specific; exact version not stated", "harness": "AgentCompass / benchmark evaluator", "prompt_style": "source does not state", "temperature": "source does not state", "context": "source does not state", "notes": "Shanghai AI Lab evaluated all model columns using OpenCompass, VLMEvalKit, and AgentCompass. Per-model mode/effort/sampling are not reported; only Intern-S2 maximum evaluation lengths are disclosed." }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Intern-S2-Preview-397B official model card: SciCode = 47.53. This is a Shanghai AI Lab cross-model evaluation; the source does not disclose this model's exact mode, effort, or sampling.", "candidates": [ { "score": 50.5, "reference_url": "https://research.nvidia.com/labs/nemotron/files/NVIDIA-Nemotron-3-Ultra-Technical-Report.pdf", "source_type": "tech_report", "reported_setting": { "version_metric": "subtask / %", "tools": "none", "harness": "NeMo Evaluator", "sampling": "8-run average" }, "notes": "NVIDIA Nemotron 3 Ultra technical report, Final Table 10: DeepSeek-V4-Pro: scicode=50.5." } ] }, { "model_id": "kimi-k2.7-code", "benchmark_id": "scicode", "score": 43.49, "reference_url": "https://huggingface.co/internlm/Intern-S2-Preview-397B", "reported_setting": { "mode": "source does not state", "effort": "source does not state", "tools": "code execution", "sampling": "source does not state", "judge": "benchmark-specific; exact version not stated", "harness": "AgentCompass / benchmark evaluator", "prompt_style": "source does not state", "temperature": "source does not state", "context": "source does not state", "notes": "Shanghai AI Lab evaluated all model columns using OpenCompass, VLMEvalKit, and AgentCompass. Per-model mode/effort/sampling are not reported; only Intern-S2 maximum evaluation lengths are disclosed." }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Intern-S2-Preview-397B official model card: SciCode = 43.49. This is a Shanghai AI Lab cross-model evaluation; the source does not disclose this model's exact mode, effort, or sampling." }, { "model_id": "intern-s2-preview-397b", "benchmark_id": "sgi_bench", "score": 49.37, "reference_url": "https://huggingface.co/internlm/Intern-S2-Preview-397B", "reported_setting": { "mode": "thinking (default)", "effort": "source does not state", "tools": "web search, PDF parser, Python, file reader", "sampling": "source does not state", "judge": "benchmark-specific; exact version not stated", "harness": "official SGI-Bench agentic evaluation", "prompt_style": "source does not state", "temperature": "source does not state", "context": "max 64K multimodal", "notes": "Shanghai AI Lab evaluated all model columns using OpenCompass, VLMEvalKit, and AgentCompass. Per-model mode/effort/sampling are not reported; only Intern-S2 maximum evaluation lengths are disclosed." }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Intern-S2-Preview-397B official model card: SGI-Bench = 49.37. This is the release model's own reported evaluation." }, { "model_id": "qwen3.5-397b", "benchmark_id": "sgi_bench", "score": 44.44, "reference_url": "https://huggingface.co/internlm/Intern-S2-Preview-397B", "reported_setting": { "mode": "source does not state", "effort": "source does not state", "tools": "web search, PDF parser, Python, file reader", "sampling": "source does not state", "judge": "benchmark-specific; exact version not stated", "harness": "official SGI-Bench agentic evaluation", "prompt_style": "source does not state", "temperature": "source does not state", "context": "source does not state", "notes": "Shanghai AI Lab evaluated all model columns using OpenCompass, VLMEvalKit, and AgentCompass. Per-model mode/effort/sampling are not reported; only Intern-S2 maximum evaluation lengths are disclosed." }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Intern-S2-Preview-397B official model card: SGI-Bench = 44.44. This is a Shanghai AI Lab cross-model evaluation; the source does not disclose this model's exact mode, effort, or sampling." }, { "model_id": "deepseek-v4-pro", "benchmark_id": "sgi_bench", "score": 45.7, "reference_url": "https://huggingface.co/internlm/Intern-S2-Preview-397B", "reported_setting": { "mode": "source does not state", "effort": "source does not state", "tools": "web search, PDF parser, Python, file reader", "sampling": "source does not state", "judge": "benchmark-specific; exact version not stated", "harness": "official SGI-Bench agentic evaluation", "prompt_style": "source does not state", "temperature": "source does not state", "context": "source does not state", "notes": "Shanghai AI Lab evaluated all model columns using OpenCompass, VLMEvalKit, and AgentCompass. Per-model mode/effort/sampling are not reported; only Intern-S2 maximum evaluation lengths are disclosed." }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Intern-S2-Preview-397B official model card: SGI-Bench = 45.7. This is a Shanghai AI Lab cross-model evaluation; the source does not disclose this model's exact mode, effort, or sampling." }, { "model_id": "kimi-k2.7-code", "benchmark_id": "sgi_bench", "score": 50.63, "reference_url": "https://huggingface.co/internlm/Intern-S2-Preview-397B", "reported_setting": { "mode": "source does not state", "effort": "source does not state", "tools": "web search, PDF parser, Python, file reader", "sampling": "source does not state", "judge": "benchmark-specific; exact version not stated", "harness": "official SGI-Bench agentic evaluation", "prompt_style": "source does not state", "temperature": "source does not state", "context": "source does not state", "notes": "Shanghai AI Lab evaluated all model columns using OpenCompass, VLMEvalKit, and AgentCompass. Per-model mode/effort/sampling are not reported; only Intern-S2 maximum evaluation lengths are disclosed." }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Intern-S2-Preview-397B official model card: SGI-Bench = 50.63. This is a Shanghai AI Lab cross-model evaluation; the source does not disclose this model's exact mode, effort, or sampling." }, { "model_id": "glm-5.2", "benchmark_id": "sgi_bench", "score": 52.41, "reference_url": "https://huggingface.co/internlm/Intern-S2-Preview-397B", "reported_setting": { "mode": "source does not state", "effort": "source does not state", "tools": "web search, PDF parser, Python, file reader", "sampling": "source does not state", "judge": "benchmark-specific; exact version not stated", "harness": "official SGI-Bench agentic evaluation", "prompt_style": "source does not state", "temperature": "source does not state", "context": "source does not state", "notes": "Shanghai AI Lab evaluated all model columns using OpenCompass, VLMEvalKit, and AgentCompass. Per-model mode/effort/sampling are not reported; only Intern-S2 maximum evaluation lengths are disclosed." }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Intern-S2-Preview-397B official model card: SGI-Bench = 52.41. This is a Shanghai AI Lab cross-model evaluation; the source does not disclose this model's exact mode, effort, or sampling." }, { "model_id": "gpt-5.5", "benchmark_id": "sgi_bench", "score": 42.77, "reference_url": "https://huggingface.co/internlm/Intern-S2-Preview-397B", "reported_setting": { "mode": "source does not state", "effort": "source does not state", "tools": "web search, PDF parser, Python, file reader", "sampling": "source does not state", "judge": "benchmark-specific; exact version not stated", "harness": "official SGI-Bench agentic evaluation", "prompt_style": "source does not state", "temperature": "source does not state", "context": "source does not state", "notes": "Shanghai AI Lab evaluated all model columns using OpenCompass, VLMEvalKit, and AgentCompass. Per-model mode/effort/sampling are not reported; only Intern-S2 maximum evaluation lengths are disclosed." }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Intern-S2-Preview-397B official model card: SGI-Bench = 42.77. This is a Shanghai AI Lab cross-model evaluation; the source does not disclose this model's exact mode, effort, or sampling." }, { "model_id": "gemini-3.1-pro", "benchmark_id": "sgi_bench", "score": 45.28, "reference_url": "https://huggingface.co/internlm/Intern-S2-Preview-397B", "reported_setting": { "mode": "source does not state", "effort": "source does not state", "tools": "web search, PDF parser, Python, file reader", "sampling": "source does not state", "judge": "benchmark-specific; exact version not stated", "harness": "official SGI-Bench agentic evaluation", "prompt_style": "source does not state", "temperature": "source does not state", "context": "source does not state", "notes": "Shanghai AI Lab evaluated all model columns using OpenCompass, VLMEvalKit, and AgentCompass. Per-model mode/effort/sampling are not reported; only Intern-S2 maximum evaluation lengths are disclosed." }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Intern-S2-Preview-397B official model card: SGI-Bench = 45.28. This is a Shanghai AI Lab cross-model evaluation; the source does not disclose this model's exact mode, effort, or sampling." }, { "model_id": "claude-opus-4.8", "benchmark_id": "sgi_bench", "score": 49.06, "reference_url": "https://huggingface.co/internlm/Intern-S2-Preview-397B", "reported_setting": { "mode": "source does not state", "effort": "source does not state", "tools": "web search, PDF parser, Python, file reader", "sampling": "source does not state", "judge": "benchmark-specific; exact version not stated", "harness": "official SGI-Bench agentic evaluation", "prompt_style": "source does not state", "temperature": "source does not state", "context": "source does not state", "notes": "Shanghai AI Lab evaluated all model columns using OpenCompass, VLMEvalKit, and AgentCompass. Per-model mode/effort/sampling are not reported; only Intern-S2 maximum evaluation lengths are disclosed." }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Intern-S2-Preview-397B official model card: SGI-Bench = 49.06. This is a Shanghai AI Lab cross-model evaluation; the source does not disclose this model's exact mode, effort, or sampling." }, { "model_id": "intern-s2-preview-397b", "benchmark_id": "researchclawbench", "score": 18.44, "reference_url": "https://huggingface.co/internlm/Intern-S2-Preview-397B", "reported_setting": { "mode": "thinking (default)", "effort": "source does not state", "tools": "scientific research environment", "sampling": "source does not state", "judge": "benchmark-specific; exact version not stated", "harness": "ResearchHarness", "prompt_style": "source does not state", "temperature": "source does not state", "context": "max 64K multimodal", "notes": "Shanghai AI Lab evaluated all model columns using OpenCompass, VLMEvalKit, and AgentCompass. Per-model mode/effort/sampling are not reported; only Intern-S2 maximum evaluation lengths are disclosed." }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Intern-S2-Preview-397B official model card: ResearchClawBench = 18.44. This is the release model's own reported evaluation." }, { "model_id": "qwen3.5-397b", "benchmark_id": "researchclawbench", "score": 15.86, "reference_url": "https://huggingface.co/internlm/Intern-S2-Preview-397B", "reported_setting": { "mode": "source does not state", "effort": "source does not state", "tools": "scientific research environment", "sampling": "source does not state", "judge": "benchmark-specific; exact version not stated", "harness": "ResearchHarness", "prompt_style": "source does not state", "temperature": "source does not state", "context": "source does not state", "notes": "Shanghai AI Lab evaluated all model columns using OpenCompass, VLMEvalKit, and AgentCompass. Per-model mode/effort/sampling are not reported; only Intern-S2 maximum evaluation lengths are disclosed." }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Intern-S2-Preview-397B official model card: ResearchClawBench = 15.86. This is a Shanghai AI Lab cross-model evaluation; the source does not disclose this model's exact mode, effort, or sampling." }, { "model_id": "deepseek-v4-pro", "benchmark_id": "researchclawbench", "score": 13.69, "reference_url": "https://huggingface.co/internlm/Intern-S2-Preview-397B", "reported_setting": { "mode": "source does not state", "effort": "source does not state", "tools": "scientific research environment", "sampling": "source does not state", "judge": "benchmark-specific; exact version not stated", "harness": "ResearchHarness", "prompt_style": "source does not state", "temperature": "source does not state", "context": "source does not state", "notes": "Shanghai AI Lab evaluated all model columns using OpenCompass, VLMEvalKit, and AgentCompass. Per-model mode/effort/sampling are not reported; only Intern-S2 maximum evaluation lengths are disclosed." }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Intern-S2-Preview-397B official model card: ResearchClawBench = 13.69. This is a Shanghai AI Lab cross-model evaluation; the source does not disclose this model's exact mode, effort, or sampling." }, { "model_id": "kimi-k2.7-code", "benchmark_id": "researchclawbench", "score": 15.4, "reference_url": "https://huggingface.co/internlm/Intern-S2-Preview-397B", "reported_setting": { "mode": "source does not state", "effort": "source does not state", "tools": "scientific research environment", "sampling": "source does not state", "judge": "benchmark-specific; exact version not stated", "harness": "ResearchHarness", "prompt_style": "source does not state", "temperature": "source does not state", "context": "source does not state", "notes": "Shanghai AI Lab evaluated all model columns using OpenCompass, VLMEvalKit, and AgentCompass. Per-model mode/effort/sampling are not reported; only Intern-S2 maximum evaluation lengths are disclosed." }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Intern-S2-Preview-397B official model card: ResearchClawBench = 15.4. This is a Shanghai AI Lab cross-model evaluation; the source does not disclose this model's exact mode, effort, or sampling." }, { "model_id": "glm-5.2", "benchmark_id": "researchclawbench", "score": 23.35, "reference_url": "https://huggingface.co/internlm/Intern-S2-Preview-397B", "reported_setting": { "mode": "source does not state", "effort": "source does not state", "tools": "scientific research environment", "sampling": "source does not state", "judge": "benchmark-specific; exact version not stated", "harness": "ResearchHarness", "prompt_style": "source does not state", "temperature": "source does not state", "context": "source does not state", "notes": "Shanghai AI Lab evaluated all model columns using OpenCompass, VLMEvalKit, and AgentCompass. Per-model mode/effort/sampling are not reported; only Intern-S2 maximum evaluation lengths are disclosed." }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Intern-S2-Preview-397B official model card: ResearchClawBench = 23.35. This is a Shanghai AI Lab cross-model evaluation; the source does not disclose this model's exact mode, effort, or sampling." }, { "model_id": "gpt-5.5", "benchmark_id": "researchclawbench", "score": 17.0, "reference_url": "https://huggingface.co/internlm/Intern-S2-Preview-397B", "reported_setting": { "mode": "source does not state", "effort": "source does not state", "tools": "scientific research environment", "sampling": "source does not state", "judge": "benchmark-specific; exact version not stated", "harness": "ResearchHarness", "prompt_style": "source does not state", "temperature": "source does not state", "context": "source does not state", "notes": "Shanghai AI Lab evaluated all model columns using OpenCompass, VLMEvalKit, and AgentCompass. Per-model mode/effort/sampling are not reported; only Intern-S2 maximum evaluation lengths are disclosed." }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Intern-S2-Preview-397B official model card: ResearchClawBench = 17.0. This is a Shanghai AI Lab cross-model evaluation; the source does not disclose this model's exact mode, effort, or sampling." }, { "model_id": "gemini-3.1-pro", "benchmark_id": "researchclawbench", "score": 14.54, "reference_url": "https://huggingface.co/internlm/Intern-S2-Preview-397B", "reported_setting": { "mode": "source does not state", "effort": "source does not state", "tools": "scientific research environment", "sampling": "source does not state", "judge": "benchmark-specific; exact version not stated", "harness": "ResearchHarness", "prompt_style": "source does not state", "temperature": "source does not state", "context": "source does not state", "notes": "Shanghai AI Lab evaluated all model columns using OpenCompass, VLMEvalKit, and AgentCompass. Per-model mode/effort/sampling are not reported; only Intern-S2 maximum evaluation lengths are disclosed." }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Intern-S2-Preview-397B official model card: ResearchClawBench = 14.54. This is a Shanghai AI Lab cross-model evaluation; the source does not disclose this model's exact mode, effort, or sampling." }, { "model_id": "claude-opus-4.8", "benchmark_id": "researchclawbench", "score": 21.74, "reference_url": "https://huggingface.co/internlm/Intern-S2-Preview-397B", "reported_setting": { "mode": "source does not state", "effort": "source does not state", "tools": "scientific research environment", "sampling": "source does not state", "judge": "benchmark-specific; exact version not stated", "harness": "ResearchHarness", "prompt_style": "source does not state", "temperature": "source does not state", "context": "source does not state", "notes": "Shanghai AI Lab evaluated all model columns using OpenCompass, VLMEvalKit, and AgentCompass. Per-model mode/effort/sampling are not reported; only Intern-S2 maximum evaluation lengths are disclosed." }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Intern-S2-Preview-397B official model card: ResearchClawBench = 21.74. This is a Shanghai AI Lab cross-model evaluation; the source does not disclose this model's exact mode, effort, or sampling." }, { "model_id": "nemotron-3-ultra-550b-a55b", "benchmark_id": "terminal_bench_2_1", "score": 56.4, "reference_url": "https://research.nvidia.com/labs/nemotron/files/NVIDIA-Nemotron-3-Ultra-Technical-Report.pdf", "reported_setting": { "version_metric": "v2.1 / % resolved", "tools": "terminal and code execution", "harness": "Harbor / Terminus-2", "sampling": "8-run average" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "NVIDIA Nemotron 3 Ultra technical report, Final Table 10: Nemotron 3 Ultra 550B-A55B: terminal_bench_2_1=56.4.", "candidates": [ { "score": 53.9, "reference_url": "https://research.nvidia.com/labs/nemotron/files/NVIDIA-Nemotron-3-Ultra-Technical-Report.pdf", "source_type": "tech_report", "reported_setting": { "version_metric": "v2.1 / % resolved", "tools": "terminal and code execution", "harness": "Harbor / Terminus-2", "sampling": "8-run average", "variant": "NVFP4 checkpoint", "precision": "NVFP4" }, "notes": "NVIDIA Nemotron 3 Ultra technical report, Quantization Table 18: Nemotron 3 Ultra NVFP4: terminal_bench_2_1=53.9, variant=NVFP4 checkpoint." }, { "score": 55.1, "reference_url": "https://research.nvidia.com/labs/nemotron/files/NVIDIA-Nemotron-3-Ultra-Technical-Report.pdf", "source_type": "tech_report", "reported_setting": { "version_metric": "Terminus scaffold / %", "tools": "agent-specific software-engineering environment", "harness": "Terminus", "sampling": "single agent result", "variant": "Terminus", "agent_scaffold": "Terminus" }, "notes": "NVIDIA Nemotron 3 Ultra technical report, Figure 17: Nemotron 3 Ultra: terminal_bench_2_1=55.1, variant=Terminus." }, { "score": 46.7, "reference_url": "https://research.nvidia.com/labs/nemotron/files/NVIDIA-Nemotron-3-Ultra-Technical-Report.pdf", "source_type": "tech_report", "reported_setting": { "version_metric": "OpenCode scaffold / %", "tools": "agent-specific software-engineering environment", "harness": "OpenCode", "sampling": "single agent result", "variant": "OpenCode", "agent_scaffold": "OpenCode" }, "notes": "NVIDIA Nemotron 3 Ultra technical report, Figure 17: Nemotron 3 Ultra: terminal_bench_2_1=46.7, variant=OpenCode." }, { "score": 52.1, "reference_url": "https://research.nvidia.com/labs/nemotron/files/NVIDIA-Nemotron-3-Ultra-Technical-Report.pdf", "source_type": "tech_report", "reported_setting": { "version_metric": "Pi scaffold / %", "tools": "agent-specific software-engineering environment", "harness": "Pi", "sampling": "single agent result", "variant": "Pi", "agent_scaffold": "Pi" }, "notes": "NVIDIA Nemotron 3 Ultra technical report, Figure 17: Nemotron 3 Ultra: terminal_bench_2_1=52.1, variant=Pi." }, { "score": 47.2, "reference_url": "https://research.nvidia.com/labs/nemotron/files/NVIDIA-Nemotron-3-Ultra-Technical-Report.pdf", "source_type": "tech_report", "reported_setting": { "version_metric": "Claude scaffold / %", "tools": "agent-specific software-engineering environment", "harness": "Claude", "sampling": "single agent result", "variant": "Claude", "agent_scaffold": "Claude" }, "notes": "NVIDIA Nemotron 3 Ultra technical report, Figure 17: Nemotron 3 Ultra: terminal_bench_2_1=47.2, variant=Claude." }, { "score": 52.8, "reference_url": "https://research.nvidia.com/labs/nemotron/files/NVIDIA-Nemotron-3-Ultra-Technical-Report.pdf", "source_type": "tech_report", "reported_setting": { "version_metric": "Hermes scaffold / %", "tools": "agent-specific software-engineering environment", "harness": "Hermes", "sampling": "single agent result", "variant": "Hermes", "agent_scaffold": "Hermes" }, "notes": "NVIDIA Nemotron 3 Ultra technical report, Figure 17: Nemotron 3 Ultra: terminal_bench_2_1=52.8, variant=Hermes." }, { "score": 52.6, "reference_url": "https://research.nvidia.com/labs/nemotron/files/NVIDIA-Nemotron-3-Ultra-Technical-Report.pdf", "source_type": "tech_report", "reported_setting": { "version_metric": "OpenHands scaffold / %", "tools": "agent-specific software-engineering environment", "harness": "OpenHands", "sampling": "single agent result", "variant": "OpenHands", "agent_scaffold": "OpenHands" }, "notes": "NVIDIA Nemotron 3 Ultra technical report, Figure 17: Nemotron 3 Ultra: terminal_bench_2_1=52.6, variant=OpenHands." }, { "score": 35.5, "reference_url": "https://research.nvidia.com/labs/nemotron/files/NVIDIA-Nemotron-3-Ultra-Technical-Report.pdf", "source_type": "tech_report", "reported_setting": { "version_metric": "Codex scaffold / %", "tools": "agent-specific software-engineering environment", "harness": "Codex", "sampling": "single agent result", "variant": "Codex", "agent_scaffold": "Codex" }, "notes": "NVIDIA Nemotron 3 Ultra technical report, Figure 17: Nemotron 3 Ultra: terminal_bench_2_1=35.5, variant=Codex." }, { "score": 48.9, "reference_url": "https://research.nvidia.com/labs/nemotron/files/NVIDIA-Nemotron-3-Ultra-Technical-Report.pdf", "source_type": "tech_report", "reported_setting": { "version_metric": "Average scaffold / %", "tools": "agent-specific software-engineering environment", "harness": "Average", "sampling": "single agent result", "variant": "Average", "agent_scaffold": "Average" }, "notes": "NVIDIA Nemotron 3 Ultra technical report, Figure 17: Nemotron 3 Ultra: terminal_bench_2_1=48.9, variant=Average." } ] }, { "model_id": "nemotron-3-ultra-550b-a55b", "benchmark_id": "gdpval_normalized_elo", "score": 46.7, "reference_url": "https://research.nvidia.com/labs/nemotron/files/NVIDIA-Nemotron-3-Ultra-Technical-Report.pdf", "reported_setting": { "version_metric": "normalized Elo / 0-100", "tools": "office workflow, web search, sandboxed code", "harness": "NeMo Gym / Stirrup", "sampling": "pass@1" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "NVIDIA Nemotron 3 Ultra technical report, Final Table 10: Nemotron 3 Ultra 550B-A55B: gdpval_normalized_elo=46.7.", "candidates": [ { "score": 47.9, "reference_url": "https://research.nvidia.com/labs/nemotron/files/NVIDIA-Nemotron-3-Ultra-Technical-Report.pdf", "source_type": "tech_report", "reported_setting": { "version_metric": "normalized Elo / 0-100", "tools": "office workflow, web search, sandboxed code", "harness": "NeMo Gym / Stirrup", "sampling": "pass@1", "variant": "NVFP4 checkpoint", "precision": "NVFP4" }, "notes": "NVIDIA Nemotron 3 Ultra technical report, Quantization Table 18: Nemotron 3 Ultra NVFP4: gdpval_normalized_elo=47.9, variant=NVFP4 checkpoint." } ] }, { "model_id": "nemotron-3-ultra-550b-a55b", "benchmark_id": "swe_bench_verified", "score": 70.7, "reference_url": "https://research.nvidia.com/labs/nemotron/files/NVIDIA-Nemotron-3-Ultra-Technical-Report.pdf", "reported_setting": { "version_metric": "Verified / % resolved", "tools": "agentic repository editing", "harness": "Harbor / OpenHands / AWS ECS", "sampling": "3-run average" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "NVIDIA Nemotron 3 Ultra technical report, Final Table 10: Nemotron 3 Ultra 550B-A55B: swe_bench_verified=70.7.", "candidates": [ { "score": 69.5, "reference_url": "https://research.nvidia.com/labs/nemotron/files/NVIDIA-Nemotron-3-Ultra-Technical-Report.pdf", "source_type": "tech_report", "reported_setting": { "version_metric": "Verified / % resolved", "tools": "agentic repository editing", "harness": "Harbor / OpenHands / AWS ECS", "sampling": "3-run average", "variant": "NVFP4 checkpoint", "precision": "NVFP4" }, "notes": "NVIDIA Nemotron 3 Ultra technical report, Quantization Table 18: Nemotron 3 Ultra NVFP4: swe_bench_verified=69.5, variant=NVFP4 checkpoint." }, { "score": 65.0, "reference_url": "https://research.nvidia.com/labs/nemotron/files/NVIDIA-Nemotron-3-Ultra-Technical-Report.pdf", "source_type": "tech_report", "reported_setting": { "version_metric": "Mini SWE Agent scaffold / %", "tools": "agent-specific software-engineering environment", "harness": "Mini SWE Agent", "sampling": "single agent result", "variant": "Mini SWE Agent", "agent_scaffold": "Mini SWE Agent" }, "notes": "NVIDIA Nemotron 3 Ultra technical report, Figure 17: Nemotron 3 Ultra: swe_bench_verified=65, variant=Mini SWE Agent." }, { "score": 67.3, "reference_url": "https://research.nvidia.com/labs/nemotron/files/NVIDIA-Nemotron-3-Ultra-Technical-Report.pdf", "source_type": "tech_report", "reported_setting": { "version_metric": "OpenCode scaffold / %", "tools": "agent-specific software-engineering environment", "harness": "OpenCode", "sampling": "single agent result", "variant": "OpenCode", "agent_scaffold": "OpenCode" }, "notes": "NVIDIA Nemotron 3 Ultra technical report, Figure 17: Nemotron 3 Ultra: swe_bench_verified=67.3, variant=OpenCode." }, { "score": 70.4, "reference_url": "https://research.nvidia.com/labs/nemotron/files/NVIDIA-Nemotron-3-Ultra-Technical-Report.pdf", "source_type": "tech_report", "reported_setting": { "version_metric": "Pi scaffold / %", "tools": "agent-specific software-engineering environment", "harness": "Pi", "sampling": "single agent result", "variant": "Pi", "agent_scaffold": "Pi" }, "notes": "NVIDIA Nemotron 3 Ultra technical report, Figure 17: Nemotron 3 Ultra: swe_bench_verified=70.4, variant=Pi." }, { "score": 60.3, "reference_url": "https://research.nvidia.com/labs/nemotron/files/NVIDIA-Nemotron-3-Ultra-Technical-Report.pdf", "source_type": "tech_report", "reported_setting": { "version_metric": "Claude scaffold / %", "tools": "agent-specific software-engineering environment", "harness": "Claude", "sampling": "single agent result", "variant": "Claude", "agent_scaffold": "Claude" }, "notes": "NVIDIA Nemotron 3 Ultra technical report, Figure 17: Nemotron 3 Ultra: swe_bench_verified=60.3, variant=Claude." }, { "score": 69.9, "reference_url": "https://research.nvidia.com/labs/nemotron/files/NVIDIA-Nemotron-3-Ultra-Technical-Report.pdf", "source_type": "tech_report", "reported_setting": { "version_metric": "Hermes scaffold / %", "tools": "agent-specific software-engineering environment", "harness": "Hermes", "sampling": "single agent result", "variant": "Hermes", "agent_scaffold": "Hermes" }, "notes": "NVIDIA Nemotron 3 Ultra technical report, Figure 17: Nemotron 3 Ultra: swe_bench_verified=69.9, variant=Hermes." }, { "score": 70.3, "reference_url": "https://research.nvidia.com/labs/nemotron/files/NVIDIA-Nemotron-3-Ultra-Technical-Report.pdf", "source_type": "tech_report", "reported_setting": { "version_metric": "OpenHands scaffold / %", "tools": "agent-specific software-engineering environment", "harness": "OpenHands", "sampling": "single agent result", "variant": "OpenHands", "agent_scaffold": "OpenHands" }, "notes": "NVIDIA Nemotron 3 Ultra technical report, Figure 17: Nemotron 3 Ultra: swe_bench_verified=70.3, variant=OpenHands." }, { "score": 21.1, "reference_url": "https://research.nvidia.com/labs/nemotron/files/NVIDIA-Nemotron-3-Ultra-Technical-Report.pdf", "source_type": "tech_report", "reported_setting": { "version_metric": "Codex scaffold / %", "tools": "agent-specific software-engineering environment", "harness": "Codex", "sampling": "single agent result", "variant": "Codex", "agent_scaffold": "Codex" }, "notes": "NVIDIA Nemotron 3 Ultra technical report, Figure 17: Nemotron 3 Ultra: swe_bench_verified=21.1, variant=Codex." }, { "score": 60.6, "reference_url": "https://research.nvidia.com/labs/nemotron/files/NVIDIA-Nemotron-3-Ultra-Technical-Report.pdf", "source_type": "tech_report", "reported_setting": { "version_metric": "Average scaffold / %", "tools": "agent-specific software-engineering environment", "harness": "Average", "sampling": "single agent result", "variant": "Average", "agent_scaffold": "Average" }, "notes": "NVIDIA Nemotron 3 Ultra technical report, Figure 17: Nemotron 3 Ultra: swe_bench_verified=60.6, variant=Average." } ] }, { "model_id": "nemotron-3-ultra-550b-a55b", "benchmark_id": "swe_bench_multilingual", "score": 67.7, "reference_url": "https://research.nvidia.com/labs/nemotron/files/NVIDIA-Nemotron-3-Ultra-Technical-Report.pdf", "reported_setting": { "version_metric": "Multilingual / % resolved", "tools": "agentic repository editing", "harness": "Harbor / OpenHands / AWS ECS", "sampling": "3-run average" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "NVIDIA Nemotron 3 Ultra technical report, Final Table 10: Nemotron 3 Ultra 550B-A55B: swe_bench_multilingual=67.7.", "candidates": [ { "score": 69.1, "reference_url": "https://research.nvidia.com/labs/nemotron/files/NVIDIA-Nemotron-3-Ultra-Technical-Report.pdf", "source_type": "tech_report", "reported_setting": { "version_metric": "Multilingual / % resolved", "tools": "agentic repository editing", "harness": "Harbor / OpenHands / AWS ECS", "sampling": "3-run average", "variant": "NVFP4 checkpoint", "precision": "NVFP4" }, "notes": "NVIDIA Nemotron 3 Ultra technical report, Quantization Table 18: Nemotron 3 Ultra NVFP4: swe_bench_multilingual=69.1, variant=NVFP4 checkpoint." } ] }, { "model_id": "nemotron-3-ultra-550b-a55b", "benchmark_id": "profbench", "score": 56.0, "reference_url": "https://research.nvidia.com/labs/nemotron/files/NVIDIA-Nemotron-3-Ultra-Technical-Report.pdf", "reported_setting": { "version_metric": "full / search score", "tools": "web search and browsing", "harness": "official/internal ProfBench scaffold", "sampling": "16-run average" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "NVIDIA Nemotron 3 Ultra technical report, Final Table 10: Nemotron 3 Ultra 550B-A55B: profbench=56.", "candidates": [ { "score": 56.4, "reference_url": "https://research.nvidia.com/labs/nemotron/files/NVIDIA-Nemotron-3-Ultra-Technical-Report.pdf", "source_type": "tech_report", "reported_setting": { "version_metric": "full / search score", "tools": "web search and browsing", "harness": "official/internal ProfBench scaffold", "sampling": "16-run average", "variant": "NVFP4 checkpoint", "precision": "NVFP4" }, "notes": "NVIDIA Nemotron 3 Ultra technical report, Quantization Table 18: Nemotron 3 Ultra NVFP4: profbench=56.4, variant=NVFP4 checkpoint." } ] }, { "model_id": "nemotron-3-ultra-550b-a55b", "benchmark_id": "pinchbench", "score": 90.0, "reference_url": "https://research.nvidia.com/labs/nemotron/files/NVIDIA-Nemotron-3-Ultra-Technical-Report.pdf", "reported_setting": { "version_metric": "current 53-task release / %", "tools": "OpenClaw coding environment", "harness": "official PinchBench scaffold", "sampling": "pass@1" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "NVIDIA Nemotron 3 Ultra technical report, Final Table 10: Nemotron 3 Ultra 550B-A55B: pinchbench=90.", "candidates": [ { "score": 89.8, "reference_url": "https://research.nvidia.com/labs/nemotron/files/NVIDIA-Nemotron-3-Ultra-Technical-Report.pdf", "source_type": "tech_report", "reported_setting": { "version_metric": "current 53-task release / %", "tools": "OpenClaw coding environment", "harness": "official PinchBench scaffold", "sampling": "pass@1", "variant": "NVFP4 checkpoint", "precision": "NVFP4" }, "notes": "NVIDIA Nemotron 3 Ultra technical report, Quantization Table 18: Nemotron 3 Ultra NVFP4: pinchbench=89.8, variant=NVFP4 checkpoint." } ] }, { "model_id": "nemotron-3-ultra-550b-a55b", "benchmark_id": "tau3_airline", "score": 81.5, "reference_url": "https://research.nvidia.com/labs/nemotron/files/NVIDIA-Nemotron-3-Ultra-Technical-Report.pdf", "reported_setting": { "version_metric": "airline / 8-trial average", "tools": "domain customer-service tools", "harness": "NeMo Gym / tau3", "sampling": "50 tasks x 8 trials" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "NVIDIA Nemotron 3 Ultra technical report, Final Table 10: Nemotron 3 Ultra 550B-A55B: tau3_airline=81.5.", "candidates": [ { "score": 80.0, "reference_url": "https://research.nvidia.com/labs/nemotron/files/NVIDIA-Nemotron-3-Ultra-Technical-Report.pdf", "source_type": "tech_report", "reported_setting": { "version_metric": "airline / 8-trial average", "tools": "domain customer-service tools", "harness": "NeMo Gym / tau3", "sampling": "50 tasks x 8 trials", "variant": "NVFP4 checkpoint", "precision": "NVFP4" }, "notes": "NVIDIA Nemotron 3 Ultra technical report, Quantization Table 18: Nemotron 3 Ultra NVFP4: tau3_airline=80, variant=NVFP4 checkpoint." } ] }, { "model_id": "nemotron-3-ultra-550b-a55b", "benchmark_id": "tau3_retail", "score": 86.4, "reference_url": "https://research.nvidia.com/labs/nemotron/files/NVIDIA-Nemotron-3-Ultra-Technical-Report.pdf", "reported_setting": { "version_metric": "retail / 8-trial average", "tools": "domain customer-service tools", "harness": "NeMo Gym / tau3", "sampling": "114 tasks x 8 trials" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "NVIDIA Nemotron 3 Ultra technical report, Final Table 10: Nemotron 3 Ultra 550B-A55B: tau3_retail=86.4.", "candidates": [ { "score": 88.4, "reference_url": "https://research.nvidia.com/labs/nemotron/files/NVIDIA-Nemotron-3-Ultra-Technical-Report.pdf", "source_type": "tech_report", "reported_setting": { "version_metric": "retail / 8-trial average", "tools": "domain customer-service tools", "harness": "NeMo Gym / tau3", "sampling": "114 tasks x 8 trials", "variant": "NVFP4 checkpoint", "precision": "NVFP4" }, "notes": "NVIDIA Nemotron 3 Ultra technical report, Quantization Table 18: Nemotron 3 Ultra NVFP4: tau3_retail=88.4, variant=NVFP4 checkpoint." } ] }, { "model_id": "nemotron-3-ultra-550b-a55b", "benchmark_id": "tau3_telecom", "score": 92.9, "reference_url": "https://research.nvidia.com/labs/nemotron/files/NVIDIA-Nemotron-3-Ultra-Technical-Report.pdf", "reported_setting": { "version_metric": "telecom / 8-trial average", "tools": "domain customer-service tools", "harness": "NeMo Gym / tau3", "sampling": "114 tasks x 8 trials" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "NVIDIA Nemotron 3 Ultra technical report, Final Table 10: Nemotron 3 Ultra 550B-A55B: tau3_telecom=92.9.", "candidates": [ { "score": 93.6, "reference_url": "https://research.nvidia.com/labs/nemotron/files/NVIDIA-Nemotron-3-Ultra-Technical-Report.pdf", "source_type": "tech_report", "reported_setting": { "version_metric": "telecom / 8-trial average", "tools": "domain customer-service tools", "harness": "NeMo Gym / tau3", "sampling": "114 tasks x 8 trials", "variant": "NVFP4 checkpoint", "precision": "NVFP4" }, "notes": "NVIDIA Nemotron 3 Ultra technical report, Quantization Table 18: Nemotron 3 Ultra NVFP4: tau3_telecom=93.6, variant=NVFP4 checkpoint." } ] }, { "model_id": "nemotron-3-ultra-550b-a55b", "benchmark_id": "tau3_banking", "score": 22.6, "reference_url": "https://research.nvidia.com/labs/nemotron/files/NVIDIA-Nemotron-3-Ultra-Technical-Report.pdf", "reported_setting": { "version_metric": "banking_knowledge / 8-trial average", "tools": "domain tools plus terminal knowledge-base search", "harness": "NeMo Gym / tau3", "sampling": "97 tasks x 8 trials" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "NVIDIA Nemotron 3 Ultra technical report, Final Table 10: Nemotron 3 Ultra 550B-A55B: tau3_banking=22.6.", "candidates": [ { "score": 19.2, "reference_url": "https://research.nvidia.com/labs/nemotron/files/NVIDIA-Nemotron-3-Ultra-Technical-Report.pdf", "source_type": "tech_report", "reported_setting": { "version_metric": "banking_knowledge / 8-trial average", "tools": "domain tools plus terminal knowledge-base search", "harness": "NeMo Gym / tau3", "sampling": "97 tasks x 8 trials", "variant": "NVFP4 checkpoint", "precision": "NVFP4" }, "notes": "NVIDIA Nemotron 3 Ultra technical report, Quantization Table 18: Nemotron 3 Ultra NVFP4: tau3_banking=19.2, variant=NVFP4 checkpoint." } ] }, { "model_id": "nemotron-3-ultra-550b-a55b", "benchmark_id": "tau3_bench", "score": 70.9, "reference_url": "https://research.nvidia.com/labs/nemotron/files/NVIDIA-Nemotron-3-Ultra-Technical-Report.pdf", "reported_setting": { "version_metric": "four-domain / 8-trial average", "tools": "domain customer-service tools", "harness": "NeMo Gym / tau3", "sampling": "375 tasks x 8 trials" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "NVIDIA Nemotron 3 Ultra technical report, Final Table 10: Nemotron 3 Ultra 550B-A55B: tau3_bench=70.9.", "candidates": [ { "score": 70.3, "reference_url": "https://research.nvidia.com/labs/nemotron/files/NVIDIA-Nemotron-3-Ultra-Technical-Report.pdf", "source_type": "tech_report", "reported_setting": { "version_metric": "four-domain / 8-trial average", "tools": "domain customer-service tools", "harness": "NeMo Gym / tau3", "sampling": "375 tasks x 8 trials", "variant": "NVFP4 checkpoint", "precision": "NVFP4" }, "notes": "NVIDIA Nemotron 3 Ultra technical report, Quantization Table 18: Nemotron 3 Ultra NVFP4: tau3_bench=70.3, variant=NVFP4 checkpoint." } ] }, { "model_id": "nemotron-3-ultra-550b-a55b", "benchmark_id": "browsecomp", "score": 44.4, "reference_url": "https://research.nvidia.com/labs/nemotron/files/NVIDIA-Nemotron-3-Ultra-Technical-Report.pdf", "reported_setting": { "version_metric": "BrowseComp / %", "tools": "Tavily web search and terminal workspace", "harness": "NVIDIA custom BrowseComp scaffold", "sampling": "pass@1" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "NVIDIA Nemotron 3 Ultra technical report, Final Table 10: Nemotron 3 Ultra 550B-A55B: browsecomp=44.4.", "candidates": [ { "score": 41.4, "reference_url": "https://research.nvidia.com/labs/nemotron/files/NVIDIA-Nemotron-3-Ultra-Technical-Report.pdf", "source_type": "tech_report", "reported_setting": { "version_metric": "BrowseComp / %", "tools": "Tavily web search and terminal workspace", "harness": "NVIDIA custom BrowseComp scaffold", "sampling": "pass@1", "variant": "NVFP4 checkpoint", "precision": "NVFP4" }, "notes": "NVIDIA Nemotron 3 Ultra technical report, Quantization Table 18: Nemotron 3 Ultra NVFP4: browsecomp=41.4, variant=NVFP4 checkpoint." } ] }, { "model_id": "nemotron-3-ultra-550b-a55b", "benchmark_id": "ioi_2025", "score": 570.0, "reference_url": "https://research.nvidia.com/labs/nemotron/files/NVIDIA-Nemotron-3-Ultra-Technical-Report.pdf", "reported_setting": { "version_metric": "contest points / 600", "tools": "code execution", "harness": "NVIDIA contest evaluation", "sampling": "pass@1" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "NVIDIA Nemotron 3 Ultra technical report, Final Table 10: Nemotron 3 Ultra 550B-A55B: ioi_2025=570.", "candidates": [ { "score": 564.7, "reference_url": "https://research.nvidia.com/labs/nemotron/files/NVIDIA-Nemotron-3-Ultra-Technical-Report.pdf", "source_type": "tech_report", "reported_setting": { "version_metric": "contest points / 600", "tools": "code execution", "harness": "NVIDIA contest evaluation", "sampling": "pass@1", "variant": "NVFP4 checkpoint", "precision": "NVFP4" }, "notes": "NVIDIA Nemotron 3 Ultra technical report, Quantization Table 18: Nemotron 3 Ultra NVFP4: ioi_2025=564.7, variant=NVFP4 checkpoint." } ] }, { "model_id": "nemotron-3-ultra-550b-a55b", "benchmark_id": "livecodebench_v6", "score": 89.0, "reference_url": "https://research.nvidia.com/labs/nemotron/files/NVIDIA-Nemotron-3-Ultra-Technical-Report.pdf", "reported_setting": { "version_metric": "v6 / pass@1", "tools": "code execution", "harness": "NeMo Gym", "sampling": "pass@1" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "NVIDIA Nemotron 3 Ultra technical report, Final Table 10: Nemotron 3 Ultra 550B-A55B: livecodebench_v6=89." }, { "model_id": "nemotron-3-ultra-550b-a55b", "benchmark_id": "imo_answerbench", "score": 88.6, "reference_url": "https://research.nvidia.com/labs/nemotron/files/NVIDIA-Nemotron-3-Ultra-Technical-Report.pdf", "reported_setting": { "version_metric": "no tools / %", "tools": "none", "harness": "NeMo Evaluator", "sampling": "5-run average", "variant": "no tools" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "NVIDIA Nemotron 3 Ultra technical report, Final Table 10: Nemotron 3 Ultra 550B-A55B: imo_answerbench=88.6, variant=no tools.", "candidates": [ { "score": 92.3, "reference_url": "https://research.nvidia.com/labs/nemotron/files/NVIDIA-Nemotron-3-Ultra-Technical-Report.pdf", "source_type": "tech_report", "reported_setting": { "version_metric": "with tools / %", "tools": "benchmark tool environment", "harness": "NeMo Evaluator", "sampling": "pass@1", "variant": "with tools" }, "notes": "NVIDIA Nemotron 3 Ultra technical report, Final Table 10: Nemotron 3 Ultra 550B-A55B: imo_answerbench=92.3, variant=with tools." } ] }, { "model_id": "nemotron-3-ultra-550b-a55b", "benchmark_id": "apex_shortlist", "score": 74.9, "reference_url": "https://research.nvidia.com/labs/nemotron/files/NVIDIA-Nemotron-3-Ultra-Technical-Report.pdf", "reported_setting": { "version_metric": "no tools / %", "tools": "none", "harness": "NeMo Evaluator", "sampling": "pass@1", "variant": "no tools" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "NVIDIA Nemotron 3 Ultra technical report, Final Table 10: Nemotron 3 Ultra 550B-A55B: apex_shortlist=74.9, variant=no tools.", "candidates": [ { "score": 84.8, "reference_url": "https://research.nvidia.com/labs/nemotron/files/NVIDIA-Nemotron-3-Ultra-Technical-Report.pdf", "source_type": "tech_report", "reported_setting": { "version_metric": "with tools / %", "tools": "benchmark tool environment", "harness": "NeMo Evaluator", "sampling": "pass@1", "variant": "with tools" }, "notes": "NVIDIA Nemotron 3 Ultra technical report, Final Table 10: Nemotron 3 Ultra 550B-A55B: apex_shortlist=84.8, variant=with tools." } ] }, { "model_id": "nemotron-3-ultra-550b-a55b", "benchmark_id": "gpqa_diamond", "score": 87.0, "reference_url": "https://research.nvidia.com/labs/nemotron/files/NVIDIA-Nemotron-3-Ultra-Technical-Report.pdf", "reported_setting": { "version_metric": "no tools / %", "tools": "none", "harness": "NeMo Evaluator", "sampling": "8-run average" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "NVIDIA Nemotron 3 Ultra technical report, Final Table 10: Nemotron 3 Ultra 550B-A55B: gpqa_diamond=87.", "candidates": [ { "score": 50.0, "reference_url": "https://research.nvidia.com/labs/nemotron/files/NVIDIA-Nemotron-3-Ultra-Technical-Report.pdf", "source_type": "tech_report", "reported_setting": { "version_metric": "5-shot CoT exact match", "tools": "none", "harness": "NeMo Evaluator", "sampling": "pass@1", "checkpoint": "base", "mode": "pretrained" }, "notes": "NVIDIA Nemotron 3 Ultra technical report, Base Table 2: Nemotron 3 Ultra Base: gpqa_diamond=50. Base checkpoint score mapped to the canonical post-trained model row as checkpoint=base." }, { "score": 87.9, "reference_url": "https://research.nvidia.com/labs/nemotron/files/NVIDIA-Nemotron-3-Ultra-Technical-Report.pdf", "source_type": "tech_report", "reported_setting": { "version_metric": "no tools / %", "tools": "none", "harness": "NeMo Evaluator", "sampling": "8-run average", "variant": "NVFP4 checkpoint", "precision": "NVFP4" }, "notes": "NVIDIA Nemotron 3 Ultra technical report, Quantization Table 18: Nemotron 3 Ultra NVFP4: gpqa_diamond=87.9, variant=NVFP4 checkpoint." }, { "score": 86.67, "reference_url": "https://research.nvidia.com/labs/nemotron/files/NVIDIA-Nemotron-3-Ultra-Technical-Report.pdf", "source_type": "tech_report", "reported_setting": { "version_metric": "averaged pass@1 plus completion tokens", "tools": "none", "harness": "NeMo Evaluator", "sampling": "pass@1", "variant": "BF16 reference", "precision": "BF16" }, "notes": "NVIDIA Nemotron 3 Ultra technical report, Quantization Table 17: Nemotron 3 Ultra BF16: gpqa_diamond=86.67, variant=BF16 reference." }, { "score": 86.36, "reference_url": "https://research.nvidia.com/labs/nemotron/files/NVIDIA-Nemotron-3-Ultra-Technical-Report.pdf", "source_type": "tech_report", "reported_setting": { "version_metric": "averaged pass@1 plus completion tokens", "tools": "none", "harness": "NeMo Evaluator", "sampling": "pass@1", "variant": "native FP4 W4A4 on Blackwell", "precision": "native FP4 W4A4 on Blackwell" }, "notes": "NVIDIA Nemotron 3 Ultra technical report, Quantization Table 17: Nemotron 3 Ultra NVFP4: gpqa_diamond=86.36, variant=native FP4 W4A4 on Blackwell." } ] }, { "model_id": "nemotron-3-ultra-550b-a55b", "benchmark_id": "scicode", "score": 44.6, "reference_url": "https://research.nvidia.com/labs/nemotron/files/NVIDIA-Nemotron-3-Ultra-Technical-Report.pdf", "reported_setting": { "version_metric": "subtask / %", "tools": "none", "harness": "NeMo Evaluator", "sampling": "8-run average" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "NVIDIA Nemotron 3 Ultra technical report, Final Table 10: Nemotron 3 Ultra 550B-A55B: scicode=44.6.", "candidates": [ { "score": 43.5, "reference_url": "https://research.nvidia.com/labs/nemotron/files/NVIDIA-Nemotron-3-Ultra-Technical-Report.pdf", "source_type": "tech_report", "reported_setting": { "version_metric": "subtask / %", "tools": "none", "harness": "NeMo Evaluator", "sampling": "8-run average", "variant": "NVFP4 checkpoint", "precision": "NVFP4" }, "notes": "NVIDIA Nemotron 3 Ultra technical report, Quantization Table 18: Nemotron 3 Ultra NVFP4: scicode=43.5, variant=NVFP4 checkpoint." } ] }, { "model_id": "nemotron-3-ultra-550b-a55b", "benchmark_id": "hle_text", "score": 26.7, "reference_url": "https://research.nvidia.com/labs/nemotron/files/NVIDIA-Nemotron-3-Ultra-Technical-Report.pdf", "reported_setting": { "version_metric": "text-only / no tools / %", "tools": "none", "harness": "NeMo Evaluator", "sampling": "pass@1", "variant": "no tools" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "NVIDIA Nemotron 3 Ultra technical report, Final Table 10: Nemotron 3 Ultra 550B-A55B: hle_text=26.7, variant=no tools.", "candidates": [ { "score": 37.4, "reference_url": "https://research.nvidia.com/labs/nemotron/files/NVIDIA-Nemotron-3-Ultra-Technical-Report.pdf", "source_type": "tech_report", "reported_setting": { "version_metric": "text-only / with tools / %", "tools": "benchmark tool environment", "harness": "NeMo Evaluator", "sampling": "pass@1", "variant": "with tools" }, "notes": "NVIDIA Nemotron 3 Ultra technical report, Final Table 10: Nemotron 3 Ultra 550B-A55B: hle_text=37.4, variant=with tools." }, { "score": 26.1, "reference_url": "https://research.nvidia.com/labs/nemotron/files/NVIDIA-Nemotron-3-Ultra-Technical-Report.pdf", "source_type": "tech_report", "reported_setting": { "version_metric": "text-only / no tools / %", "tools": "none", "harness": "NeMo Evaluator", "sampling": "pass@1", "variant": "NVFP4 checkpoint", "precision": "NVFP4" }, "notes": "NVIDIA Nemotron 3 Ultra technical report, Quantization Table 18: Nemotron 3 Ultra NVFP4: hle_text=26.1, variant=NVFP4 checkpoint." }, { "score": 26.92, "reference_url": "https://research.nvidia.com/labs/nemotron/files/NVIDIA-Nemotron-3-Ultra-Technical-Report.pdf", "source_type": "tech_report", "reported_setting": { "version_metric": "averaged pass@1 plus completion tokens", "tools": "none", "harness": "NeMo Evaluator", "sampling": "pass@1", "variant": "BF16 reference", "precision": "BF16" }, "notes": "NVIDIA Nemotron 3 Ultra technical report, Quantization Table 17: Nemotron 3 Ultra BF16: hle_text=26.92, variant=BF16 reference." }, { "score": 25.12, "reference_url": "https://research.nvidia.com/labs/nemotron/files/NVIDIA-Nemotron-3-Ultra-Technical-Report.pdf", "source_type": "tech_report", "reported_setting": { "version_metric": "averaged pass@1 plus completion tokens", "tools": "none", "harness": "NeMo Evaluator", "sampling": "pass@1", "variant": "W4A16 on Hopper", "precision": "W4A16 on Hopper" }, "notes": "NVIDIA Nemotron 3 Ultra technical report, Quantization Table 17: Nemotron 3 Ultra NVFP4: hle_text=25.12, variant=W4A16 on Hopper." }, { "score": 25.67, "reference_url": "https://research.nvidia.com/labs/nemotron/files/NVIDIA-Nemotron-3-Ultra-Technical-Report.pdf", "source_type": "tech_report", "reported_setting": { "version_metric": "averaged pass@1 plus completion tokens", "tools": "none", "harness": "NeMo Evaluator", "sampling": "pass@1", "variant": "native FP4 W4A4 on Blackwell", "precision": "native FP4 W4A4 on Blackwell" }, "notes": "NVIDIA Nemotron 3 Ultra technical report, Quantization Table 17: Nemotron 3 Ultra NVFP4: hle_text=25.67, variant=native FP4 W4A4 on Blackwell." } ] }, { "model_id": "nemotron-3-ultra-550b-a55b", "benchmark_id": "critpt", "score": 3.1, "reference_url": "https://research.nvidia.com/labs/nemotron/files/NVIDIA-Nemotron-3-Ultra-Technical-Report.pdf", "reported_setting": { "version_metric": "no tools / %", "tools": "none", "harness": "NeMo Evaluator", "sampling": "5-run average" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "NVIDIA Nemotron 3 Ultra technical report, Final Table 10: Nemotron 3 Ultra 550B-A55B: critpt=3.1.", "candidates": [ { "score": 3.4, "reference_url": "https://research.nvidia.com/labs/nemotron/files/NVIDIA-Nemotron-3-Ultra-Technical-Report.pdf", "source_type": "tech_report", "reported_setting": { "version_metric": "no tools / %", "tools": "none", "harness": "NeMo Evaluator", "sampling": "5-run average", "variant": "NVFP4 checkpoint", "precision": "NVFP4" }, "notes": "NVIDIA Nemotron 3 Ultra technical report, Quantization Table 18: Nemotron 3 Ultra NVFP4: critpt=3.4, variant=NVFP4 checkpoint." } ] }, { "model_id": "nemotron-3-ultra-550b-a55b", "benchmark_id": "mmlu_pro", "score": 86.8, "reference_url": "https://research.nvidia.com/labs/nemotron/files/NVIDIA-Nemotron-3-Ultra-Technical-Report.pdf", "reported_setting": { "version_metric": "MMLU-Pro / %", "tools": "none", "harness": "NeMo Evaluator", "sampling": "pass@1" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "NVIDIA Nemotron 3 Ultra technical report, Final Table 10: Nemotron 3 Ultra 550B-A55B: mmlu_pro=86.8.", "candidates": [ { "score": 79.07, "reference_url": "https://research.nvidia.com/labs/nemotron/files/NVIDIA-Nemotron-3-Ultra-Technical-Report.pdf", "source_type": "tech_report", "reported_setting": { "version_metric": "5-shot CoT exact match", "tools": "none", "harness": "NeMo Evaluator", "sampling": "pass@1", "checkpoint": "base", "mode": "pretrained" }, "notes": "NVIDIA Nemotron 3 Ultra technical report, Base Table 2: Nemotron 3 Ultra Base: mmlu_pro=79.07. Base checkpoint score mapped to the canonical post-trained model row as checkpoint=base." } ] }, { "model_id": "nemotron-3-ultra-550b-a55b", "benchmark_id": "aa_omniscience_accuracy", "score": 24.1, "reference_url": "https://research.nvidia.com/labs/nemotron/files/NVIDIA-Nemotron-3-Ultra-Technical-Report.pdf", "reported_setting": { "version_metric": "accuracy / %", "tools": "none", "harness": "NeMo Evaluator", "sampling": "10-run average" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "NVIDIA Nemotron 3 Ultra technical report, Final Table 10: Nemotron 3 Ultra 550B-A55B: aa_omniscience_accuracy=24.1.", "candidates": [ { "score": 24.6, "reference_url": "https://research.nvidia.com/labs/nemotron/files/NVIDIA-Nemotron-3-Ultra-Technical-Report.pdf", "source_type": "tech_report", "reported_setting": { "version_metric": "accuracy / %", "tools": "none", "harness": "NeMo Evaluator", "sampling": "10-run average", "variant": "NVFP4 checkpoint", "precision": "NVFP4" }, "notes": "NVIDIA Nemotron 3 Ultra technical report, Quantization Table 18: Nemotron 3 Ultra NVFP4: aa_omniscience_accuracy=24.6, variant=NVFP4 checkpoint." }, { "score": 24.38, "reference_url": "https://research.nvidia.com/labs/nemotron/files/NVIDIA-Nemotron-3-Ultra-Technical-Report.pdf", "source_type": "tech_report", "reported_setting": { "version_metric": "averaged pass@1 plus completion tokens", "tools": "none", "harness": "NeMo Evaluator", "sampling": "pass@1", "variant": "BF16 reference", "precision": "BF16" }, "notes": "NVIDIA Nemotron 3 Ultra technical report, Quantization Table 17: Nemotron 3 Ultra BF16: aa_omniscience_accuracy=24.38, variant=BF16 reference." }, { "score": 25.5, "reference_url": "https://research.nvidia.com/labs/nemotron/files/NVIDIA-Nemotron-3-Ultra-Technical-Report.pdf", "source_type": "tech_report", "reported_setting": { "version_metric": "averaged pass@1 plus completion tokens", "tools": "none", "harness": "NeMo Evaluator", "sampling": "pass@1", "variant": "W4A16 on Hopper", "precision": "W4A16 on Hopper" }, "notes": "NVIDIA Nemotron 3 Ultra technical report, Quantization Table 17: Nemotron 3 Ultra NVFP4: aa_omniscience_accuracy=25.5, variant=W4A16 on Hopper." }, { "score": 24.55, "reference_url": "https://research.nvidia.com/labs/nemotron/files/NVIDIA-Nemotron-3-Ultra-Technical-Report.pdf", "source_type": "tech_report", "reported_setting": { "version_metric": "averaged pass@1 plus completion tokens", "tools": "none", "harness": "NeMo Evaluator", "sampling": "pass@1", "variant": "native FP4 W4A4 on Blackwell", "precision": "native FP4 W4A4 on Blackwell" }, "notes": "NVIDIA Nemotron 3 Ultra technical report, Quantization Table 17: Nemotron 3 Ultra NVFP4: aa_omniscience_accuracy=24.55, variant=native FP4 W4A4 on Blackwell." } ] }, { "model_id": "nemotron-3-ultra-550b-a55b", "benchmark_id": "aa_omniscience_non_hallucination", "score": 78.7, "reference_url": "https://research.nvidia.com/labs/nemotron/files/NVIDIA-Nemotron-3-Ultra-Technical-Report.pdf", "reported_setting": { "version_metric": "non-hallucination / %", "tools": "none", "harness": "NeMo Evaluator", "sampling": "10-run average" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "NVIDIA Nemotron 3 Ultra technical report, Final Table 10: Nemotron 3 Ultra 550B-A55B: aa_omniscience_non_hallucination=78.7.", "candidates": [ { "score": 75.5, "reference_url": "https://research.nvidia.com/labs/nemotron/files/NVIDIA-Nemotron-3-Ultra-Technical-Report.pdf", "source_type": "tech_report", "reported_setting": { "version_metric": "non-hallucination / %", "tools": "none", "harness": "NeMo Evaluator", "sampling": "10-run average", "variant": "NVFP4 checkpoint", "precision": "NVFP4" }, "notes": "NVIDIA Nemotron 3 Ultra technical report, Quantization Table 18: Nemotron 3 Ultra NVFP4: aa_omniscience_non_hallucination=75.5, variant=NVFP4 checkpoint." } ] }, { "model_id": "nemotron-3-ultra-550b-a55b", "benchmark_id": "ifbench", "score": 81.7, "reference_url": "https://research.nvidia.com/labs/nemotron/files/NVIDIA-Nemotron-3-Ultra-Technical-Report.pdf", "reported_setting": { "version_metric": "prompt-level loose accuracy / %", "tools": "none", "harness": "NeMo Evaluator", "sampling": "8-run average" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "NVIDIA Nemotron 3 Ultra technical report, Final Table 10: Nemotron 3 Ultra 550B-A55B: ifbench=81.7.", "candidates": [ { "score": 82.3, "reference_url": "https://research.nvidia.com/labs/nemotron/files/NVIDIA-Nemotron-3-Ultra-Technical-Report.pdf", "source_type": "tech_report", "reported_setting": { "version_metric": "prompt-level loose accuracy / %", "tools": "none", "harness": "NeMo Evaluator", "sampling": "8-run average", "variant": "NVFP4 checkpoint", "precision": "NVFP4" }, "notes": "NVIDIA Nemotron 3 Ultra technical report, Quantization Table 18: Nemotron 3 Ultra NVFP4: ifbench=82.3, variant=NVFP4 checkpoint." }, { "score": 82.12, "reference_url": "https://research.nvidia.com/labs/nemotron/files/NVIDIA-Nemotron-3-Ultra-Technical-Report.pdf", "source_type": "tech_report", "reported_setting": { "version_metric": "averaged pass@1 plus completion tokens", "tools": "none", "harness": "NeMo Evaluator", "sampling": "pass@1", "variant": "BF16 reference", "precision": "BF16" }, "notes": "NVIDIA Nemotron 3 Ultra technical report, Quantization Table 17: Nemotron 3 Ultra BF16: ifbench=82.12, variant=BF16 reference." }, { "score": 82.75, "reference_url": "https://research.nvidia.com/labs/nemotron/files/NVIDIA-Nemotron-3-Ultra-Technical-Report.pdf", "source_type": "tech_report", "reported_setting": { "version_metric": "averaged pass@1 plus completion tokens", "tools": "none", "harness": "NeMo Evaluator", "sampling": "pass@1", "variant": "W4A16 on Hopper", "precision": "W4A16 on Hopper" }, "notes": "NVIDIA Nemotron 3 Ultra technical report, Quantization Table 17: Nemotron 3 Ultra NVFP4: ifbench=82.75, variant=W4A16 on Hopper." }, { "score": 82.42, "reference_url": "https://research.nvidia.com/labs/nemotron/files/NVIDIA-Nemotron-3-Ultra-Technical-Report.pdf", "source_type": "tech_report", "reported_setting": { "version_metric": "averaged pass@1 plus completion tokens", "tools": "none", "harness": "NeMo Evaluator", "sampling": "pass@1", "variant": "native FP4 W4A4 on Blackwell", "precision": "native FP4 W4A4 on Blackwell" }, "notes": "NVIDIA Nemotron 3 Ultra technical report, Quantization Table 17: Nemotron 3 Ultra NVFP4: ifbench=82.42, variant=native FP4 W4A4 on Blackwell." } ] }, { "model_id": "nemotron-3-ultra-550b-a55b", "benchmark_id": "multichallenge", "score": 63.8, "reference_url": "https://research.nvidia.com/labs/nemotron/files/NVIDIA-Nemotron-3-Ultra-Technical-Report.pdf", "reported_setting": { "version_metric": "multi-turn score / %", "tools": "none", "harness": "NeMo Evaluator", "sampling": "8-run average" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "NVIDIA Nemotron 3 Ultra technical report, Final Table 10: Nemotron 3 Ultra 550B-A55B: multichallenge=63.8." }, { "model_id": "nemotron-3-ultra-550b-a55b", "benchmark_id": "aa_lcr", "score": 65.4, "reference_url": "https://research.nvidia.com/labs/nemotron/files/NVIDIA-Nemotron-3-Ultra-Technical-Report.pdf", "reported_setting": { "version_metric": "AA-LCR / %", "tools": "none", "harness": "NeMo Evaluator", "sampling": "16-run average" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "NVIDIA Nemotron 3 Ultra technical report, Final Table 10: Nemotron 3 Ultra 550B-A55B: aa_lcr=65.4.", "candidates": [ { "score": 65.5, "reference_url": "https://research.nvidia.com/labs/nemotron/files/NVIDIA-Nemotron-3-Ultra-Technical-Report.pdf", "source_type": "tech_report", "reported_setting": { "version_metric": "AA-LCR / %", "tools": "none", "harness": "NeMo Evaluator", "sampling": "16-run average", "variant": "NVFP4 checkpoint", "precision": "NVFP4" }, "notes": "NVIDIA Nemotron 3 Ultra technical report, Quantization Table 18: Nemotron 3 Ultra NVFP4: aa_lcr=65.5, variant=NVFP4 checkpoint." }, { "score": 63.67, "reference_url": "https://research.nvidia.com/labs/nemotron/files/NVIDIA-Nemotron-3-Ultra-Technical-Report.pdf", "source_type": "tech_report", "reported_setting": { "version_metric": "averaged pass@1 plus completion tokens", "tools": "none", "harness": "NeMo Evaluator", "sampling": "pass@1", "variant": "BF16 reference", "precision": "BF16" }, "notes": "NVIDIA Nemotron 3 Ultra technical report, Quantization Table 17: Nemotron 3 Ultra BF16: aa_lcr=63.67, variant=BF16 reference." }, { "score": 65.33, "reference_url": "https://research.nvidia.com/labs/nemotron/files/NVIDIA-Nemotron-3-Ultra-Technical-Report.pdf", "source_type": "tech_report", "reported_setting": { "version_metric": "averaged pass@1 plus completion tokens", "tools": "none", "harness": "NeMo Evaluator", "sampling": "pass@1", "variant": "W4A16 on Hopper", "precision": "W4A16 on Hopper" }, "notes": "NVIDIA Nemotron 3 Ultra technical report, Quantization Table 17: Nemotron 3 Ultra NVFP4: aa_lcr=65.33, variant=W4A16 on Hopper." }, { "score": 64.0, "reference_url": "https://research.nvidia.com/labs/nemotron/files/NVIDIA-Nemotron-3-Ultra-Technical-Report.pdf", "source_type": "tech_report", "reported_setting": { "version_metric": "averaged pass@1 plus completion tokens", "tools": "none", "harness": "NeMo Evaluator", "sampling": "pass@1", "variant": "native FP4 W4A4 on Blackwell", "precision": "native FP4 W4A4 on Blackwell" }, "notes": "NVIDIA Nemotron 3 Ultra technical report, Quantization Table 17: Nemotron 3 Ultra NVFP4: aa_lcr=64, variant=native FP4 W4A4 on Blackwell." } ] }, { "model_id": "nemotron-3-ultra-550b-a55b", "benchmark_id": "ruler_1m", "score": 94.7, "reference_url": "https://research.nvidia.com/labs/nemotron/files/NVIDIA-Nemotron-3-Ultra-Technical-Report.pdf", "reported_setting": { "version_metric": "1M / %", "tools": "none", "harness": "NeMo Skills / RULER", "sampling": "pass@1" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "NVIDIA Nemotron 3 Ultra technical report, Final Table 10: Nemotron 3 Ultra 550B-A55B: ruler_1m=94.7.", "candidates": [ { "score": 76.83, "reference_url": "https://research.nvidia.com/labs/nemotron/files/NVIDIA-Nemotron-3-Ultra-Technical-Report.pdf", "source_type": "tech_report", "reported_setting": { "version_metric": "0-shot", "tools": "none", "harness": "NeMo Skills / RULER", "sampling": "pass@1", "checkpoint": "base", "mode": "pretrained" }, "notes": "NVIDIA Nemotron 3 Ultra technical report, Base Table 2: Nemotron 3 Ultra Base: ruler_1m=76.83. Base checkpoint score mapped to the canonical post-trained model row as checkpoint=base." }, { "score": 94.0, "reference_url": "https://research.nvidia.com/labs/nemotron/files/NVIDIA-Nemotron-3-Ultra-Technical-Report.pdf", "source_type": "tech_report", "reported_setting": { "version_metric": "1M / %", "tools": "none", "harness": "NeMo Skills / RULER", "sampling": "pass@1", "variant": "NVFP4 checkpoint", "precision": "NVFP4" }, "notes": "NVIDIA Nemotron 3 Ultra technical report, Quantization Table 18: Nemotron 3 Ultra NVFP4: ruler_1m=94, variant=NVFP4 checkpoint." } ] }, { "model_id": "nemotron-3-ultra-550b-a55b", "benchmark_id": "longbench_v2", "score": 61.9, "reference_url": "https://research.nvidia.com/labs/nemotron/files/NVIDIA-Nemotron-3-Ultra-Technical-Report.pdf", "reported_setting": { "version_metric": "<=1M / %", "tools": "none", "harness": "NeMo Evaluator", "sampling": "pass@1" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "NVIDIA Nemotron 3 Ultra technical report, Final Table 10: Nemotron 3 Ultra 550B-A55B: longbench_v2=61.9." }, { "model_id": "nemotron-3-ultra-550b-a55b", "benchmark_id": "wmt24pp", "score": 83.7, "reference_url": "https://research.nvidia.com/labs/nemotron/files/NVIDIA-Nemotron-3-Ultra-Technical-Report.pdf", "reported_setting": { "version_metric": "en-to-55 languages / XCOMET-XXL", "tools": "none", "harness": "NeMo Evaluator", "sampling": "54,890 translations" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "NVIDIA Nemotron 3 Ultra technical report, Final Table 10: Nemotron 3 Ultra 550B-A55B: wmt24pp=83.7." }, { "model_id": "minimax-m2.7", "benchmark_id": "terminal_bench_2_1", "score": 55.5, "reference_url": "https://research.nvidia.com/labs/nemotron/files/NVIDIA-Nemotron-3-Ultra-Technical-Report.pdf", "reported_setting": { "version_metric": "v2.1 / % resolved", "tools": "terminal and code execution", "harness": "Harbor / Terminus-2", "sampling": "8-run average" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "NVIDIA Nemotron 3 Ultra technical report, Final Table 10: MiniMax M2.7: terminal_bench_2_1=55.5." }, { "model_id": "minimax-m2.7", "benchmark_id": "gdpval_normalized_elo", "score": 47.6, "reference_url": "https://research.nvidia.com/labs/nemotron/files/NVIDIA-Nemotron-3-Ultra-Technical-Report.pdf", "reported_setting": { "version_metric": "normalized Elo / 0-100", "tools": "office workflow, web search, sandboxed code", "harness": "NeMo Gym / Stirrup", "sampling": "pass@1" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "NVIDIA Nemotron 3 Ultra technical report, Final Table 10: MiniMax M2.7: gdpval_normalized_elo=47.6." }, { "model_id": "minimax-m2.7", "benchmark_id": "swe_bench_verified", "score": 75.3, "reference_url": "https://research.nvidia.com/labs/nemotron/files/NVIDIA-Nemotron-3-Ultra-Technical-Report.pdf", "reported_setting": { "version_metric": "Verified / % resolved", "tools": "agentic repository editing", "harness": "Harbor / OpenHands / AWS ECS", "sampling": "3-run average" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "NVIDIA Nemotron 3 Ultra technical report, Final Table 10: MiniMax M2.7: swe_bench_verified=75.3." }, { "model_id": "minimax-m2.7", "benchmark_id": "profbench", "score": 52.0, "reference_url": "https://research.nvidia.com/labs/nemotron/files/NVIDIA-Nemotron-3-Ultra-Technical-Report.pdf", "reported_setting": { "version_metric": "full / search score", "tools": "web search and browsing", "harness": "official/internal ProfBench scaffold", "sampling": "16-run average" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "NVIDIA Nemotron 3 Ultra technical report, Final Table 10: MiniMax M2.7: profbench=52." }, { "model_id": "minimax-m2.7", "benchmark_id": "pinchbench", "score": 77.6, "reference_url": "https://research.nvidia.com/labs/nemotron/files/NVIDIA-Nemotron-3-Ultra-Technical-Report.pdf", "reported_setting": { "version_metric": "current 53-task release / %", "tools": "OpenClaw coding environment", "harness": "official PinchBench scaffold", "sampling": "pass@1" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "NVIDIA Nemotron 3 Ultra technical report, Final Table 10: MiniMax M2.7: pinchbench=77.6." }, { "model_id": "minimax-m2.7", "benchmark_id": "tau3_airline", "score": 75.3, "reference_url": "https://research.nvidia.com/labs/nemotron/files/NVIDIA-Nemotron-3-Ultra-Technical-Report.pdf", "reported_setting": { "version_metric": "airline / 8-trial average", "tools": "domain customer-service tools", "harness": "NeMo Gym / tau3", "sampling": "50 tasks x 8 trials" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "NVIDIA Nemotron 3 Ultra technical report, Final Table 10: MiniMax M2.7: tau3_airline=75.3." }, { "model_id": "minimax-m2.7", "benchmark_id": "tau3_retail", "score": 84.9, "reference_url": "https://research.nvidia.com/labs/nemotron/files/NVIDIA-Nemotron-3-Ultra-Technical-Report.pdf", "reported_setting": { "version_metric": "retail / 8-trial average", "tools": "domain customer-service tools", "harness": "NeMo Gym / tau3", "sampling": "114 tasks x 8 trials" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "NVIDIA Nemotron 3 Ultra technical report, Final Table 10: MiniMax M2.7: tau3_retail=84.9." }, { "model_id": "minimax-m2.7", "benchmark_id": "tau3_telecom", "score": 89.6, "reference_url": "https://research.nvidia.com/labs/nemotron/files/NVIDIA-Nemotron-3-Ultra-Technical-Report.pdf", "reported_setting": { "version_metric": "telecom / 8-trial average", "tools": "domain customer-service tools", "harness": "NeMo Gym / tau3", "sampling": "114 tasks x 8 trials" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "NVIDIA Nemotron 3 Ultra technical report, Final Table 10: MiniMax M2.7: tau3_telecom=89.6." }, { "model_id": "minimax-m2.7", "benchmark_id": "tau3_banking", "score": 14.6, "reference_url": "https://research.nvidia.com/labs/nemotron/files/NVIDIA-Nemotron-3-Ultra-Technical-Report.pdf", "reported_setting": { "version_metric": "banking_knowledge / 8-trial average", "tools": "domain tools plus terminal knowledge-base search", "harness": "NeMo Gym / tau3", "sampling": "97 tasks x 8 trials" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "NVIDIA Nemotron 3 Ultra technical report, Final Table 10: MiniMax M2.7: tau3_banking=14.6." }, { "model_id": "minimax-m2.7", "benchmark_id": "browsecomp", "score": 54.1, "reference_url": "https://research.nvidia.com/labs/nemotron/files/NVIDIA-Nemotron-3-Ultra-Technical-Report.pdf", "reported_setting": { "version_metric": "BrowseComp / %", "tools": "Tavily web search and terminal workspace", "harness": "NVIDIA custom BrowseComp scaffold", "sampling": "pass@1" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "NVIDIA Nemotron 3 Ultra technical report, Final Table 10: MiniMax M2.7: browsecomp=54.1." }, { "model_id": "minimax-m2.7", "benchmark_id": "livecodebench_v6", "score": 77.2, "reference_url": "https://research.nvidia.com/labs/nemotron/files/NVIDIA-Nemotron-3-Ultra-Technical-Report.pdf", "reported_setting": { "version_metric": "v6 / pass@1", "tools": "code execution", "harness": "NeMo Gym", "sampling": "pass@1" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "NVIDIA Nemotron 3 Ultra technical report, Final Table 10: MiniMax M2.7: livecodebench_v6=77.2." }, { "model_id": "minimax-m2.7", "benchmark_id": "apex_shortlist", "score": 28.9, "reference_url": "https://research.nvidia.com/labs/nemotron/files/NVIDIA-Nemotron-3-Ultra-Technical-Report.pdf", "reported_setting": { "version_metric": "no tools / %", "tools": "none", "harness": "NeMo Evaluator", "sampling": "pass@1", "variant": "no tools" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "NVIDIA Nemotron 3 Ultra technical report, Final Table 10: MiniMax M2.7: apex_shortlist=28.9, variant=no tools.", "candidates": [ { "score": 51.9, "reference_url": "https://research.nvidia.com/labs/nemotron/files/NVIDIA-Nemotron-3-Ultra-Technical-Report.pdf", "source_type": "tech_report", "reported_setting": { "version_metric": "with tools / %", "tools": "benchmark tool environment", "harness": "NeMo Evaluator", "sampling": "pass@1", "variant": "with tools" }, "notes": "NVIDIA Nemotron 3 Ultra technical report, Final Table 10: MiniMax M2.7: apex_shortlist=51.9, variant=with tools." } ] }, { "model_id": "minimax-m2.7", "benchmark_id": "scicode", "score": 38.3, "reference_url": "https://research.nvidia.com/labs/nemotron/files/NVIDIA-Nemotron-3-Ultra-Technical-Report.pdf", "reported_setting": { "version_metric": "subtask / %", "tools": "none", "harness": "NeMo Evaluator", "sampling": "8-run average" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "NVIDIA Nemotron 3 Ultra technical report, Final Table 10: MiniMax M2.7: scicode=38.3." }, { "model_id": "minimax-m2.7", "benchmark_id": "hle_text", "score": 23.1, "reference_url": "https://research.nvidia.com/labs/nemotron/files/NVIDIA-Nemotron-3-Ultra-Technical-Report.pdf", "reported_setting": { "version_metric": "text-only / no tools / %", "tools": "none", "harness": "NeMo Evaluator", "sampling": "pass@1", "variant": "no tools" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "NVIDIA Nemotron 3 Ultra technical report, Final Table 10: MiniMax M2.7: hle_text=23.1, variant=no tools." }, { "model_id": "minimax-m2.7", "benchmark_id": "critpt", "score": 0.6, "reference_url": "https://research.nvidia.com/labs/nemotron/files/NVIDIA-Nemotron-3-Ultra-Technical-Report.pdf", "reported_setting": { "version_metric": "no tools / %", "tools": "none", "harness": "NeMo Evaluator", "sampling": "5-run average" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "NVIDIA Nemotron 3 Ultra technical report, Final Table 10: MiniMax M2.7: critpt=0.6." }, { "model_id": "minimax-m2.7", "benchmark_id": "mmlu_pro", "score": 81.9, "reference_url": "https://research.nvidia.com/labs/nemotron/files/NVIDIA-Nemotron-3-Ultra-Technical-Report.pdf", "reported_setting": { "version_metric": "MMLU-Pro / %", "tools": "none", "harness": "NeMo Evaluator", "sampling": "pass@1" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "NVIDIA Nemotron 3 Ultra technical report, Final Table 10: MiniMax M2.7: mmlu_pro=81.9." }, { "model_id": "minimax-m2.7", "benchmark_id": "aa_omniscience_accuracy", "score": 20.5, "reference_url": "https://research.nvidia.com/labs/nemotron/files/NVIDIA-Nemotron-3-Ultra-Technical-Report.pdf", "reported_setting": { "version_metric": "accuracy / %", "tools": "none", "harness": "NeMo Evaluator", "sampling": "10-run average" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "NVIDIA Nemotron 3 Ultra technical report, Final Table 10: MiniMax M2.7: aa_omniscience_accuracy=20.5." }, { "model_id": "minimax-m2.7", "benchmark_id": "aa_omniscience_non_hallucination", "score": 74.4, "reference_url": "https://research.nvidia.com/labs/nemotron/files/NVIDIA-Nemotron-3-Ultra-Technical-Report.pdf", "reported_setting": { "version_metric": "non-hallucination / %", "tools": "none", "harness": "NeMo Evaluator", "sampling": "10-run average" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "NVIDIA Nemotron 3 Ultra technical report, Final Table 10: MiniMax M2.7: aa_omniscience_non_hallucination=74.4." }, { "model_id": "minimax-m2.7", "benchmark_id": "ifbench", "score": 74.6, "reference_url": "https://research.nvidia.com/labs/nemotron/files/NVIDIA-Nemotron-3-Ultra-Technical-Report.pdf", "reported_setting": { "version_metric": "prompt-level loose accuracy / %", "tools": "none", "harness": "NeMo Evaluator", "sampling": "8-run average" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "NVIDIA Nemotron 3 Ultra technical report, Final Table 10: MiniMax M2.7: ifbench=74.6." }, { "model_id": "minimax-m2.7", "benchmark_id": "multichallenge", "score": 42.5, "reference_url": "https://research.nvidia.com/labs/nemotron/files/NVIDIA-Nemotron-3-Ultra-Technical-Report.pdf", "reported_setting": { "version_metric": "multi-turn score / %", "tools": "none", "harness": "NeMo Evaluator", "sampling": "8-run average" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "NVIDIA Nemotron 3 Ultra technical report, Final Table 10: MiniMax M2.7: multichallenge=42.5." }, { "model_id": "minimax-m2.7", "benchmark_id": "aa_lcr", "score": 69.8, "reference_url": "https://research.nvidia.com/labs/nemotron/files/NVIDIA-Nemotron-3-Ultra-Technical-Report.pdf", "reported_setting": { "version_metric": "AA-LCR / %", "tools": "none", "harness": "NeMo Evaluator", "sampling": "16-run average" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "NVIDIA Nemotron 3 Ultra technical report, Final Table 10: MiniMax M2.7: aa_lcr=69.8." }, { "model_id": "minimax-m2.7", "benchmark_id": "wmt24pp", "score": 82.8, "reference_url": "https://research.nvidia.com/labs/nemotron/files/NVIDIA-Nemotron-3-Ultra-Technical-Report.pdf", "reported_setting": { "version_metric": "en-to-55 languages / XCOMET-XXL", "tools": "none", "harness": "NeMo Evaluator", "sampling": "54,890 translations" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "NVIDIA Nemotron 3 Ultra technical report, Final Table 10: MiniMax M2.7: wmt24pp=82.8." }, { "model_id": "glm-5.1", "benchmark_id": "gdpval_normalized_elo", "score": 54.7, "reference_url": "https://research.nvidia.com/labs/nemotron/files/NVIDIA-Nemotron-3-Ultra-Technical-Report.pdf", "reported_setting": { "version_metric": "normalized Elo / 0-100", "tools": "office workflow, web search, sandboxed code", "harness": "NeMo Gym / Stirrup", "sampling": "pass@1" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "NVIDIA Nemotron 3 Ultra technical report, Final Table 10: GLM-5.1: gdpval_normalized_elo=54.7." }, { "model_id": "glm-5.1", "benchmark_id": "profbench", "score": 46.0, "reference_url": "https://research.nvidia.com/labs/nemotron/files/NVIDIA-Nemotron-3-Ultra-Technical-Report.pdf", "reported_setting": { "version_metric": "full / search score", "tools": "web search and browsing", "harness": "official/internal ProfBench scaffold", "sampling": "16-run average" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "NVIDIA Nemotron 3 Ultra technical report, Final Table 10: GLM-5.1: profbench=46." }, { "model_id": "glm-5.1", "benchmark_id": "pinchbench", "score": 81.2, "reference_url": "https://research.nvidia.com/labs/nemotron/files/NVIDIA-Nemotron-3-Ultra-Technical-Report.pdf", "reported_setting": { "version_metric": "current 53-task release / %", "tools": "OpenClaw coding environment", "harness": "official PinchBench scaffold", "sampling": "pass@1" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "NVIDIA Nemotron 3 Ultra technical report, Final Table 10: GLM-5.1: pinchbench=81.2." }, { "model_id": "glm-5.1", "benchmark_id": "tau3_airline", "score": 85.0, "reference_url": "https://research.nvidia.com/labs/nemotron/files/NVIDIA-Nemotron-3-Ultra-Technical-Report.pdf", "reported_setting": { "version_metric": "airline / 8-trial average", "tools": "domain customer-service tools", "harness": "NeMo Gym / tau3", "sampling": "50 tasks x 8 trials" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "NVIDIA Nemotron 3 Ultra technical report, Final Table 10: GLM-5.1: tau3_airline=85.", "candidates": [ { "score": 79.5, "reference_url": "https://huggingface.co/mistralai/Mistral-Medium-3.5-128B/resolve/22b2b868a15677cfa6061277ed2f653d1349a9ab/images/image4.png?download=true", "source_type": "official_model_card", "reported_setting": { "mode": "thinking", "effort": "default (maximum reasoning setting)", "tools": "τ³ domain tools and retrieval environment", "sampling": "trials=4", "judge": "benchmark-specified", "harness": "Mistral τ³ evaluation with GPT-5.2 low-effort user simulator", "prompt_style": "default", "temperature": "default", "context": "default" }, "notes": "Displayed exactly: '79.5'. Research observation: medium-image4.png:3:5:score:reported. Chart footnotes: *All model evaluations were run with maximum reasoning settings. | **Sonnet 4.6 encountered higher rates of reasoning truncation due to external API restrictions than other models, which affected its performance. | *** Self-reported | **** Self-reported, Mistral is using context management and a discard-all strategy at 100k tokens. | τ³ scores as reported by Sierra for Claude Sonnet 4.5 and Qwen3.5. Others with user simulator: gpt-5.2 with reasoning_effort: low. 4 trials. Banking domain evaluated with terminal- or embedding-based agentic search retrieval, only highest score is reported." } ] }, { "model_id": "glm-5.1", "benchmark_id": "tau3_retail", "score": 84.1, "reference_url": "https://research.nvidia.com/labs/nemotron/files/NVIDIA-Nemotron-3-Ultra-Technical-Report.pdf", "reported_setting": { "version_metric": "retail / 8-trial average", "tools": "domain customer-service tools", "harness": "NeMo Gym / tau3", "sampling": "114 tasks x 8 trials" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "NVIDIA Nemotron 3 Ultra technical report, Final Table 10: GLM-5.1: tau3_retail=84.1.", "candidates": [ { "score": 76.3, "reference_url": "https://huggingface.co/mistralai/Mistral-Medium-3.5-128B/resolve/22b2b868a15677cfa6061277ed2f653d1349a9ab/images/image4.png?download=true", "source_type": "official_model_card", "reported_setting": { "mode": "thinking", "effort": "default (maximum reasoning setting)", "tools": "τ³ domain tools and retrieval environment", "sampling": "trials=4", "judge": "benchmark-specified", "harness": "Mistral τ³ evaluation with GPT-5.2 low-effort user simulator", "prompt_style": "default", "temperature": "default", "context": "default" }, "notes": "Displayed exactly: '76.3'. Research observation: medium-image4.png:4:5:score:reported. Chart footnotes: *All model evaluations were run with maximum reasoning settings. | **Sonnet 4.6 encountered higher rates of reasoning truncation due to external API restrictions than other models, which affected its performance. | *** Self-reported | **** Self-reported, Mistral is using context management and a discard-all strategy at 100k tokens. | τ³ scores as reported by Sierra for Claude Sonnet 4.5 and Qwen3.5. Others with user simulator: gpt-5.2 with reasoning_effort: low. 4 trials. Banking domain evaluated with terminal- or embedding-based agentic search retrieval, only highest score is reported." } ] }, { "model_id": "glm-5.1", "benchmark_id": "tau3_telecom", "score": 96.9, "reference_url": "https://research.nvidia.com/labs/nemotron/files/NVIDIA-Nemotron-3-Ultra-Technical-Report.pdf", "reported_setting": { "version_metric": "telecom / 8-trial average", "tools": "domain customer-service tools", "harness": "NeMo Gym / tau3", "sampling": "114 tasks x 8 trials" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "NVIDIA Nemotron 3 Ultra technical report, Final Table 10: GLM-5.1: tau3_telecom=96.9.", "candidates": [ { "score": 98.7, "reference_url": "https://huggingface.co/mistralai/Mistral-Medium-3.5-128B/resolve/22b2b868a15677cfa6061277ed2f653d1349a9ab/images/image4.png?download=true", "source_type": "official_model_card", "reported_setting": { "mode": "thinking", "effort": "default (maximum reasoning setting)", "tools": "τ³ domain tools and retrieval environment", "sampling": "trials=4", "judge": "benchmark-specified", "harness": "Mistral τ³ evaluation with GPT-5.2 low-effort user simulator", "prompt_style": "default", "temperature": "default", "context": "default" }, "notes": "Displayed exactly: '98.7'. Research observation: medium-image4.png:2:5:score:reported. Chart footnotes: *All model evaluations were run with maximum reasoning settings. | **Sonnet 4.6 encountered higher rates of reasoning truncation due to external API restrictions than other models, which affected its performance. | *** Self-reported | **** Self-reported, Mistral is using context management and a discard-all strategy at 100k tokens. | τ³ scores as reported by Sierra for Claude Sonnet 4.5 and Qwen3.5. Others with user simulator: gpt-5.2 with reasoning_effort: low. 4 trials. Banking domain evaluated with terminal- or embedding-based agentic search retrieval, only highest score is reported." } ] }, { "model_id": "glm-5.1", "benchmark_id": "tau3_banking", "score": 12.8, "reference_url": "https://research.nvidia.com/labs/nemotron/files/NVIDIA-Nemotron-3-Ultra-Technical-Report.pdf", "reported_setting": { "version_metric": "banking_knowledge / 8-trial average", "tools": "domain tools plus terminal knowledge-base search", "harness": "NeMo Gym / tau3", "sampling": "97 tasks x 8 trials" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "NVIDIA Nemotron 3 Ultra technical report, Final Table 10: GLM-5.1: tau3_banking=12.8.", "candidates": [ { "score": 16.2, "reference_url": "https://huggingface.co/mistralai/Mistral-Medium-3.5-128B/resolve/22b2b868a15677cfa6061277ed2f653d1349a9ab/images/image4.png?download=true", "source_type": "official_model_card", "reported_setting": { "mode": "thinking", "effort": "default (maximum reasoning setting)", "tools": "τ³ Banking terminal or embedding retrieval; higher of the two reported", "sampling": "trials=4", "judge": "benchmark-specified", "harness": "Mistral τ³ evaluation with GPT-5.2 low-effort user simulator", "prompt_style": "default", "temperature": "default", "context": "default" }, "notes": "Displayed exactly: '16.2'. Research observation: medium-image4.png:5:5:score:reported. Chart footnotes: *All model evaluations were run with maximum reasoning settings. | **Sonnet 4.6 encountered higher rates of reasoning truncation due to external API restrictions than other models, which affected its performance. | *** Self-reported | **** Self-reported, Mistral is using context management and a discard-all strategy at 100k tokens. | τ³ scores as reported by Sierra for Claude Sonnet 4.5 and Qwen3.5. Others with user simulator: gpt-5.2 with reasoning_effort: low. 4 trials. Banking domain evaluated with terminal- or embedding-based agentic search retrieval, only highest score is reported." } ] }, { "model_id": "glm-5.1", "benchmark_id": "ioi_2025", "score": 456.5, "reference_url": "https://research.nvidia.com/labs/nemotron/files/NVIDIA-Nemotron-3-Ultra-Technical-Report.pdf", "reported_setting": { "version_metric": "contest points / 600", "tools": "code execution", "harness": "NVIDIA contest evaluation", "sampling": "pass@1" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "NVIDIA Nemotron 3 Ultra technical report, Final Table 10: GLM-5.1: ioi_2025=456.5." }, { "model_id": "glm-5.1", "benchmark_id": "livecodebench_v6", "score": 85.7, "reference_url": "https://research.nvidia.com/labs/nemotron/files/NVIDIA-Nemotron-3-Ultra-Technical-Report.pdf", "reported_setting": { "version_metric": "v6 / pass@1", "tools": "code execution", "harness": "NeMo Gym", "sampling": "pass@1" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "NVIDIA Nemotron 3 Ultra technical report, Final Table 10: GLM-5.1: livecodebench_v6=85.7." }, { "model_id": "glm-5.1", "benchmark_id": "scicode", "score": 47.7, "reference_url": "https://research.nvidia.com/labs/nemotron/files/NVIDIA-Nemotron-3-Ultra-Technical-Report.pdf", "reported_setting": { "version_metric": "subtask / %", "tools": "none", "harness": "NeMo Evaluator", "sampling": "8-run average" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "NVIDIA Nemotron 3 Ultra technical report, Final Table 10: GLM-5.1: scicode=47.7." }, { "model_id": "glm-5.1", "benchmark_id": "aa_omniscience_accuracy", "score": 31.3, "reference_url": "https://research.nvidia.com/labs/nemotron/files/NVIDIA-Nemotron-3-Ultra-Technical-Report.pdf", "reported_setting": { "version_metric": "accuracy / %", "tools": "none", "harness": "NeMo Evaluator", "sampling": "10-run average" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "NVIDIA Nemotron 3 Ultra technical report, Final Table 10: GLM-5.1: aa_omniscience_accuracy=31.3." }, { "model_id": "glm-5.1", "benchmark_id": "aa_omniscience_non_hallucination", "score": 66.8, "reference_url": "https://research.nvidia.com/labs/nemotron/files/NVIDIA-Nemotron-3-Ultra-Technical-Report.pdf", "reported_setting": { "version_metric": "non-hallucination / %", "tools": "none", "harness": "NeMo Evaluator", "sampling": "10-run average" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "NVIDIA Nemotron 3 Ultra technical report, Final Table 10: GLM-5.1: aa_omniscience_non_hallucination=66.8." }, { "model_id": "glm-5.1", "benchmark_id": "ifbench", "score": 76.6, "reference_url": "https://research.nvidia.com/labs/nemotron/files/NVIDIA-Nemotron-3-Ultra-Technical-Report.pdf", "reported_setting": { "version_metric": "prompt-level loose accuracy / %", "tools": "none", "harness": "NeMo Evaluator", "sampling": "8-run average" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "NVIDIA Nemotron 3 Ultra technical report, Final Table 10: GLM-5.1: ifbench=76.6." }, { "model_id": "glm-5.1", "benchmark_id": "multichallenge", "score": 63.0, "reference_url": "https://research.nvidia.com/labs/nemotron/files/NVIDIA-Nemotron-3-Ultra-Technical-Report.pdf", "reported_setting": { "version_metric": "multi-turn score / %", "tools": "none", "harness": "NeMo Evaluator", "sampling": "8-run average" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "NVIDIA Nemotron 3 Ultra technical report, Final Table 10: GLM-5.1: multichallenge=63." }, { "model_id": "glm-5.1", "benchmark_id": "wmt24pp", "score": 84.4, "reference_url": "https://research.nvidia.com/labs/nemotron/files/NVIDIA-Nemotron-3-Ultra-Technical-Report.pdf", "reported_setting": { "version_metric": "en-to-55 languages / XCOMET-XXL", "tools": "none", "harness": "NeMo Evaluator", "sampling": "54,890 translations" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "NVIDIA Nemotron 3 Ultra technical report, Final Table 10: GLM-5.1: wmt24pp=84.4." }, { "model_id": "kimi-k2.6", "benchmark_id": "terminal_bench_2_1", "score": 67.2, "reference_url": "https://research.nvidia.com/labs/nemotron/files/NVIDIA-Nemotron-3-Ultra-Technical-Report.pdf", "reported_setting": { "version_metric": "v2.1 / % resolved", "tools": "terminal and code execution", "harness": "Harbor / Terminus-2", "sampling": "8-run average" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "NVIDIA Nemotron 3 Ultra technical report, Final Table 10: Kimi K2.6: terminal_bench_2_1=67.2.", "candidates": [ { "score": 51.7, "reference_url": "https://research.nvidia.com/labs/nemotron/files/NVIDIA-Nemotron-3-Ultra-Technical-Report.pdf", "source_type": "tech_report", "reported_setting": { "version_metric": "OpenCode scaffold / %", "tools": "agent-specific software-engineering environment", "harness": "OpenCode", "sampling": "single agent result", "variant": "OpenCode", "agent_scaffold": "OpenCode" }, "notes": "NVIDIA Nemotron 3 Ultra technical report, Figure 17: Kimi K2.6: terminal_bench_2_1=51.7, variant=OpenCode." }, { "score": 58.0, "reference_url": "https://research.nvidia.com/labs/nemotron/files/NVIDIA-Nemotron-3-Ultra-Technical-Report.pdf", "source_type": "tech_report", "reported_setting": { "version_metric": "Pi scaffold / %", "tools": "agent-specific software-engineering environment", "harness": "Pi", "sampling": "single agent result", "variant": "Pi", "agent_scaffold": "Pi" }, "notes": "NVIDIA Nemotron 3 Ultra technical report, Figure 17: Kimi K2.6: terminal_bench_2_1=58, variant=Pi." }, { "score": 57.5, "reference_url": "https://research.nvidia.com/labs/nemotron/files/NVIDIA-Nemotron-3-Ultra-Technical-Report.pdf", "source_type": "tech_report", "reported_setting": { "version_metric": "Claude scaffold / %", "tools": "agent-specific software-engineering environment", "harness": "Claude", "sampling": "single agent result", "variant": "Claude", "agent_scaffold": "Claude" }, "notes": "NVIDIA Nemotron 3 Ultra technical report, Figure 17: Kimi K2.6: terminal_bench_2_1=57.5, variant=Claude." }, { "score": 61.4, "reference_url": "https://research.nvidia.com/labs/nemotron/files/NVIDIA-Nemotron-3-Ultra-Technical-Report.pdf", "source_type": "tech_report", "reported_setting": { "version_metric": "Hermes scaffold / %", "tools": "agent-specific software-engineering environment", "harness": "Hermes", "sampling": "single agent result", "variant": "Hermes", "agent_scaffold": "Hermes" }, "notes": "NVIDIA Nemotron 3 Ultra technical report, Figure 17: Kimi K2.6: terminal_bench_2_1=61.4, variant=Hermes." }, { "score": 63.8, "reference_url": "https://research.nvidia.com/labs/nemotron/files/NVIDIA-Nemotron-3-Ultra-Technical-Report.pdf", "source_type": "tech_report", "reported_setting": { "version_metric": "OpenHands scaffold / %", "tools": "agent-specific software-engineering environment", "harness": "OpenHands", "sampling": "single agent result", "variant": "OpenHands", "agent_scaffold": "OpenHands" }, "notes": "NVIDIA Nemotron 3 Ultra technical report, Figure 17: Kimi K2.6: terminal_bench_2_1=63.8, variant=OpenHands." }, { "score": 48.3, "reference_url": "https://research.nvidia.com/labs/nemotron/files/NVIDIA-Nemotron-3-Ultra-Technical-Report.pdf", "source_type": "tech_report", "reported_setting": { "version_metric": "Codex scaffold / %", "tools": "agent-specific software-engineering environment", "harness": "Codex", "sampling": "single agent result", "variant": "Codex", "agent_scaffold": "Codex" }, "notes": "NVIDIA Nemotron 3 Ultra technical report, Figure 17: Kimi K2.6: terminal_bench_2_1=48.3, variant=Codex." }, { "score": 58.3, "reference_url": "https://research.nvidia.com/labs/nemotron/files/NVIDIA-Nemotron-3-Ultra-Technical-Report.pdf", "source_type": "tech_report", "reported_setting": { "version_metric": "Average scaffold / %", "tools": "agent-specific software-engineering environment", "harness": "Average", "sampling": "single agent result", "variant": "Average", "agent_scaffold": "Average" }, "notes": "NVIDIA Nemotron 3 Ultra technical report, Figure 17: Kimi K2.6: terminal_bench_2_1=58.3, variant=Average." } ] }, { "model_id": "kimi-k2.6", "benchmark_id": "gdpval_normalized_elo", "score": 50.4, "reference_url": "https://research.nvidia.com/labs/nemotron/files/NVIDIA-Nemotron-3-Ultra-Technical-Report.pdf", "reported_setting": { "version_metric": "normalized Elo / 0-100", "tools": "office workflow, web search, sandboxed code", "harness": "NeMo Gym / Stirrup", "sampling": "pass@1" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "NVIDIA Nemotron 3 Ultra technical report, Final Table 10: Kimi K2.6: gdpval_normalized_elo=50.4." }, { "model_id": "kimi-k2.6", "benchmark_id": "profbench", "score": 56.0, "reference_url": "https://research.nvidia.com/labs/nemotron/files/NVIDIA-Nemotron-3-Ultra-Technical-Report.pdf", "reported_setting": { "version_metric": "full / search score", "tools": "web search and browsing", "harness": "official/internal ProfBench scaffold", "sampling": "16-run average" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "NVIDIA Nemotron 3 Ultra technical report, Final Table 10: Kimi K2.6: profbench=56." }, { "model_id": "kimi-k2.6", "benchmark_id": "pinchbench", "score": 90.2, "reference_url": "https://research.nvidia.com/labs/nemotron/files/NVIDIA-Nemotron-3-Ultra-Technical-Report.pdf", "reported_setting": { "version_metric": "current 53-task release / %", "tools": "OpenClaw coding environment", "harness": "official PinchBench scaffold", "sampling": "pass@1" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "NVIDIA Nemotron 3 Ultra technical report, Final Table 10: Kimi K2.6: pinchbench=90.2." }, { "model_id": "kimi-k2.6", "benchmark_id": "tau3_airline", "score": 85.8, "reference_url": "https://research.nvidia.com/labs/nemotron/files/NVIDIA-Nemotron-3-Ultra-Technical-Report.pdf", "reported_setting": { "version_metric": "airline / 8-trial average", "tools": "domain customer-service tools", "harness": "NeMo Gym / tau3", "sampling": "50 tasks x 8 trials" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "NVIDIA Nemotron 3 Ultra technical report, Final Table 10: Kimi K2.6: tau3_airline=85.8." }, { "model_id": "kimi-k2.6", "benchmark_id": "tau3_retail", "score": 82.9, "reference_url": "https://research.nvidia.com/labs/nemotron/files/NVIDIA-Nemotron-3-Ultra-Technical-Report.pdf", "reported_setting": { "version_metric": "retail / 8-trial average", "tools": "domain customer-service tools", "harness": "NeMo Gym / tau3", "sampling": "114 tasks x 8 trials" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "NVIDIA Nemotron 3 Ultra technical report, Final Table 10: Kimi K2.6: tau3_retail=82.9." }, { "model_id": "kimi-k2.6", "benchmark_id": "tau3_telecom", "score": 97.8, "reference_url": "https://research.nvidia.com/labs/nemotron/files/NVIDIA-Nemotron-3-Ultra-Technical-Report.pdf", "reported_setting": { "version_metric": "telecom / 8-trial average", "tools": "domain customer-service tools", "harness": "NeMo Gym / tau3", "sampling": "114 tasks x 8 trials" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "NVIDIA Nemotron 3 Ultra technical report, Final Table 10: Kimi K2.6: tau3_telecom=97.8." }, { "model_id": "kimi-k2.6", "benchmark_id": "tau3_banking", "score": 23.1, "reference_url": "https://research.nvidia.com/labs/nemotron/files/NVIDIA-Nemotron-3-Ultra-Technical-Report.pdf", "reported_setting": { "version_metric": "banking_knowledge / 8-trial average", "tools": "domain tools plus terminal knowledge-base search", "harness": "NeMo Gym / tau3", "sampling": "97 tasks x 8 trials" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "NVIDIA Nemotron 3 Ultra technical report, Final Table 10: Kimi K2.6: tau3_banking=23.1." }, { "model_id": "kimi-k2.6", "benchmark_id": "tau3_bench", "score": 72.4, "reference_url": "https://research.nvidia.com/labs/nemotron/files/NVIDIA-Nemotron-3-Ultra-Technical-Report.pdf", "reported_setting": { "version_metric": "four-domain / 8-trial average", "tools": "domain customer-service tools", "harness": "NeMo Gym / tau3", "sampling": "375 tasks x 8 trials" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "NVIDIA Nemotron 3 Ultra technical report, Final Table 10: Kimi K2.6: tau3_bench=72.4." }, { "model_id": "kimi-k2.6", "benchmark_id": "ioi_2025", "score": 585.0, "reference_url": "https://research.nvidia.com/labs/nemotron/files/NVIDIA-Nemotron-3-Ultra-Technical-Report.pdf", "reported_setting": { "version_metric": "contest points / 600", "tools": "code execution", "harness": "NVIDIA contest evaluation", "sampling": "pass@1" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "NVIDIA Nemotron 3 Ultra technical report, Final Table 10: Kimi K2.6: ioi_2025=585." }, { "model_id": "kimi-k2.6", "benchmark_id": "critpt", "score": 9.1, "reference_url": "https://research.nvidia.com/labs/nemotron/files/NVIDIA-Nemotron-3-Ultra-Technical-Report.pdf", "reported_setting": { "version_metric": "no tools / %", "tools": "none", "harness": "NeMo Evaluator", "sampling": "5-run average" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "NVIDIA Nemotron 3 Ultra technical report, Final Table 10: Kimi K2.6: critpt=9.1." }, { "model_id": "kimi-k2.6", "benchmark_id": "aa_omniscience_accuracy", "score": 35.5, "reference_url": "https://research.nvidia.com/labs/nemotron/files/NVIDIA-Nemotron-3-Ultra-Technical-Report.pdf", "reported_setting": { "version_metric": "accuracy / %", "tools": "none", "harness": "NeMo Evaluator", "sampling": "10-run average" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "NVIDIA Nemotron 3 Ultra technical report, Final Table 10: Kimi K2.6: aa_omniscience_accuracy=35.5." }, { "model_id": "kimi-k2.6", "benchmark_id": "aa_omniscience_non_hallucination", "score": 67.1, "reference_url": "https://research.nvidia.com/labs/nemotron/files/NVIDIA-Nemotron-3-Ultra-Technical-Report.pdf", "reported_setting": { "version_metric": "non-hallucination / %", "tools": "none", "harness": "NeMo Evaluator", "sampling": "10-run average" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "NVIDIA Nemotron 3 Ultra technical report, Final Table 10: Kimi K2.6: aa_omniscience_non_hallucination=67.1." }, { "model_id": "kimi-k2.6", "benchmark_id": "ifbench", "score": 73.7, "reference_url": "https://research.nvidia.com/labs/nemotron/files/NVIDIA-Nemotron-3-Ultra-Technical-Report.pdf", "reported_setting": { "version_metric": "prompt-level loose accuracy / %", "tools": "none", "harness": "NeMo Evaluator", "sampling": "8-run average" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "NVIDIA Nemotron 3 Ultra technical report, Final Table 10: Kimi K2.6: ifbench=73.7." }, { "model_id": "kimi-k2.6", "benchmark_id": "multichallenge", "score": 63.1, "reference_url": "https://research.nvidia.com/labs/nemotron/files/NVIDIA-Nemotron-3-Ultra-Technical-Report.pdf", "reported_setting": { "version_metric": "multi-turn score / %", "tools": "none", "harness": "NeMo Evaluator", "sampling": "8-run average" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "NVIDIA Nemotron 3 Ultra technical report, Final Table 10: Kimi K2.6: multichallenge=63.1." }, { "model_id": "kimi-k2.6", "benchmark_id": "aa_lcr", "score": 70.2, "reference_url": "https://research.nvidia.com/labs/nemotron/files/NVIDIA-Nemotron-3-Ultra-Technical-Report.pdf", "reported_setting": { "version_metric": "AA-LCR / %", "tools": "none", "harness": "NeMo Evaluator", "sampling": "16-run average" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "NVIDIA Nemotron 3 Ultra technical report, Final Table 10: Kimi K2.6: aa_lcr=70.2." }, { "model_id": "kimi-k2.6", "benchmark_id": "wmt24pp", "score": 84.5, "reference_url": "https://research.nvidia.com/labs/nemotron/files/NVIDIA-Nemotron-3-Ultra-Technical-Report.pdf", "reported_setting": { "version_metric": "en-to-55 languages / XCOMET-XXL", "tools": "none", "harness": "NeMo Evaluator", "sampling": "54,890 translations" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "NVIDIA Nemotron 3 Ultra technical report, Final Table 10: Kimi K2.6: wmt24pp=84.5." }, { "model_id": "qwen3.5-397b", "benchmark_id": "gdpval_normalized_elo", "score": 34.6, "reference_url": "https://research.nvidia.com/labs/nemotron/files/NVIDIA-Nemotron-3-Ultra-Technical-Report.pdf", "reported_setting": { "version_metric": "normalized Elo / 0-100", "tools": "office workflow, web search, sandboxed code", "harness": "NeMo Gym / Stirrup", "sampling": "pass@1" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "NVIDIA Nemotron 3 Ultra technical report, Final Table 10: Qwen3.5-397B-A17B: gdpval_normalized_elo=34.6." }, { "model_id": "qwen3.5-397b", "benchmark_id": "profbench", "score": 53.0, "reference_url": "https://research.nvidia.com/labs/nemotron/files/NVIDIA-Nemotron-3-Ultra-Technical-Report.pdf", "reported_setting": { "version_metric": "full / search score", "tools": "web search and browsing", "harness": "official/internal ProfBench scaffold", "sampling": "16-run average" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "NVIDIA Nemotron 3 Ultra technical report, Final Table 10: Qwen3.5-397B-A17B: profbench=53." }, { "model_id": "qwen3.5-397b", "benchmark_id": "pinchbench", "score": 86.6, "reference_url": "https://research.nvidia.com/labs/nemotron/files/NVIDIA-Nemotron-3-Ultra-Technical-Report.pdf", "reported_setting": { "version_metric": "current 53-task release / %", "tools": "OpenClaw coding environment", "harness": "official PinchBench scaffold", "sampling": "pass@1" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "NVIDIA Nemotron 3 Ultra technical report, Final Table 10: Qwen3.5-397B-A17B: pinchbench=86.6." }, { "model_id": "qwen3.5-397b", "benchmark_id": "tau3_airline", "score": 76.5, "reference_url": "https://research.nvidia.com/labs/nemotron/files/NVIDIA-Nemotron-3-Ultra-Technical-Report.pdf", "reported_setting": { "version_metric": "airline / 8-trial average", "tools": "domain customer-service tools", "harness": "NeMo Gym / tau3", "sampling": "50 tasks x 8 trials" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "NVIDIA Nemotron 3 Ultra technical report, Final Table 10: Qwen3.5-397B-A17B: tau3_airline=76.5.", "candidates": [ { "score": 81.5, "reference_url": "https://huggingface.co/mistralai/Mistral-Medium-3.5-128B/resolve/22b2b868a15677cfa6061277ed2f653d1349a9ab/images/image4.png?download=true", "source_type": "official_model_card", "reported_setting": { "mode": "thinking", "effort": "default (maximum reasoning setting)", "tools": "τ³ domain tools and retrieval environment", "sampling": "Sierra-reported; trial count not stated here", "judge": "benchmark-specified", "harness": "Sierra-reported τ³ result", "prompt_style": "default (thinking by default)", "temperature": "default", "context": "default (262K)" }, "notes": "Displayed exactly: '81.5'. Research observation: medium-image4.png:3:6:score:reported. Chart footnotes: *All model evaluations were run with maximum reasoning settings. | **Sonnet 4.6 encountered higher rates of reasoning truncation due to external API restrictions than other models, which affected its performance. | *** Self-reported | **** Self-reported, Mistral is using context management and a discard-all strategy at 100k tokens. | τ³ scores as reported by Sierra for Claude Sonnet 4.5 and Qwen3.5. Others with user simulator: gpt-5.2 with reasoning_effort: low. 4 trials. Banking domain evaluated with terminal- or embedding-based agentic search retrieval, only highest score is reported." } ] }, { "model_id": "qwen3.5-397b", "benchmark_id": "tau3_retail", "score": 88.5, "reference_url": "https://research.nvidia.com/labs/nemotron/files/NVIDIA-Nemotron-3-Ultra-Technical-Report.pdf", "reported_setting": { "version_metric": "retail / 8-trial average", "tools": "domain customer-service tools", "harness": "NeMo Gym / tau3", "sampling": "114 tasks x 8 trials" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "NVIDIA Nemotron 3 Ultra technical report, Final Table 10: Qwen3.5-397B-A17B: tau3_retail=88.5.", "candidates": [ { "score": 84.4, "reference_url": "https://huggingface.co/mistralai/Mistral-Medium-3.5-128B/resolve/22b2b868a15677cfa6061277ed2f653d1349a9ab/images/image4.png?download=true", "source_type": "official_model_card", "reported_setting": { "mode": "thinking", "effort": "default (maximum reasoning setting)", "tools": "τ³ domain tools and retrieval environment", "sampling": "Sierra-reported; trial count not stated here", "judge": "benchmark-specified", "harness": "Sierra-reported τ³ result", "prompt_style": "default (thinking by default)", "temperature": "default", "context": "default (262K)" }, "notes": "Displayed exactly: '84.4'. Research observation: medium-image4.png:4:6:score:reported. Chart footnotes: *All model evaluations were run with maximum reasoning settings. | **Sonnet 4.6 encountered higher rates of reasoning truncation due to external API restrictions than other models, which affected its performance. | *** Self-reported | **** Self-reported, Mistral is using context management and a discard-all strategy at 100k tokens. | τ³ scores as reported by Sierra for Claude Sonnet 4.5 and Qwen3.5. Others with user simulator: gpt-5.2 with reasoning_effort: low. 4 trials. Banking domain evaluated with terminal- or embedding-based agentic search retrieval, only highest score is reported." } ] }, { "model_id": "qwen3.5-397b", "benchmark_id": "tau3_telecom", "score": 98.0, "reference_url": "https://research.nvidia.com/labs/nemotron/files/NVIDIA-Nemotron-3-Ultra-Technical-Report.pdf", "reported_setting": { "version_metric": "telecom / 8-trial average", "tools": "domain customer-service tools", "harness": "NeMo Gym / tau3", "sampling": "114 tasks x 8 trials" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "NVIDIA Nemotron 3 Ultra technical report, Final Table 10: Qwen3.5-397B-A17B: tau3_telecom=98.", "candidates": [ { "score": 97.8, "reference_url": "https://huggingface.co/mistralai/Mistral-Medium-3.5-128B/resolve/22b2b868a15677cfa6061277ed2f653d1349a9ab/images/image4.png?download=true", "source_type": "official_model_card", "reported_setting": { "mode": "thinking", "effort": "default (maximum reasoning setting)", "tools": "τ³ domain tools and retrieval environment", "sampling": "Sierra-reported; trial count not stated here", "judge": "benchmark-specified", "harness": "Sierra-reported τ³ result", "prompt_style": "default (thinking by default)", "temperature": "default", "context": "default (262K)" }, "notes": "Displayed exactly: '97.8'. Research observation: medium-image4.png:2:6:score:reported. Chart footnotes: *All model evaluations were run with maximum reasoning settings. | **Sonnet 4.6 encountered higher rates of reasoning truncation due to external API restrictions than other models, which affected its performance. | *** Self-reported | **** Self-reported, Mistral is using context management and a discard-all strategy at 100k tokens. | τ³ scores as reported by Sierra for Claude Sonnet 4.5 and Qwen3.5. Others with user simulator: gpt-5.2 with reasoning_effort: low. 4 trials. Banking domain evaluated with terminal- or embedding-based agentic search retrieval, only highest score is reported." } ] }, { "model_id": "qwen3.5-397b", "benchmark_id": "tau3_banking", "score": 20.9, "reference_url": "https://research.nvidia.com/labs/nemotron/files/NVIDIA-Nemotron-3-Ultra-Technical-Report.pdf", "reported_setting": { "version_metric": "banking_knowledge / 8-trial average", "tools": "domain tools plus terminal knowledge-base search", "harness": "NeMo Gym / tau3", "sampling": "97 tasks x 8 trials" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "NVIDIA Nemotron 3 Ultra technical report, Final Table 10: Qwen3.5-397B-A17B: tau3_banking=20.9.", "candidates": [ { "score": 9.8, "reference_url": "https://huggingface.co/mistralai/Mistral-Medium-3.5-128B/resolve/22b2b868a15677cfa6061277ed2f653d1349a9ab/images/image4.png?download=true", "source_type": "official_model_card", "reported_setting": { "mode": "thinking", "effort": "default (maximum reasoning setting)", "tools": "τ³ Banking terminal or embedding retrieval; higher of the two reported", "sampling": "Sierra-reported; trial count not stated here", "judge": "benchmark-specified", "harness": "Sierra-reported τ³ result", "prompt_style": "default (thinking by default)", "temperature": "default", "context": "default (262K)" }, "notes": "Displayed exactly: '9.8'. Research observation: medium-image4.png:5:6:score:reported. Chart footnotes: *All model evaluations were run with maximum reasoning settings. | **Sonnet 4.6 encountered higher rates of reasoning truncation due to external API restrictions than other models, which affected its performance. | *** Self-reported | **** Self-reported, Mistral is using context management and a discard-all strategy at 100k tokens. | τ³ scores as reported by Sierra for Claude Sonnet 4.5 and Qwen3.5. Others with user simulator: gpt-5.2 with reasoning_effort: low. 4 trials. Banking domain evaluated with terminal- or embedding-based agentic search retrieval, only highest score is reported." } ] }, { "model_id": "qwen3.5-397b", "benchmark_id": "tau3_bench", "score": 71.0, "reference_url": "https://research.nvidia.com/labs/nemotron/files/NVIDIA-Nemotron-3-Ultra-Technical-Report.pdf", "reported_setting": { "version_metric": "four-domain / 8-trial average", "tools": "domain customer-service tools", "harness": "NeMo Gym / tau3", "sampling": "375 tasks x 8 trials" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "NVIDIA Nemotron 3 Ultra technical report, Final Table 10: Qwen3.5-397B-A17B: tau3_bench=71." }, { "model_id": "qwen3.5-397b", "benchmark_id": "ioi_2025", "score": 441.3, "reference_url": "https://research.nvidia.com/labs/nemotron/files/NVIDIA-Nemotron-3-Ultra-Technical-Report.pdf", "reported_setting": { "version_metric": "contest points / 600", "tools": "code execution", "harness": "NVIDIA contest evaluation", "sampling": "pass@1" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "NVIDIA Nemotron 3 Ultra technical report, Final Table 10: Qwen3.5-397B-A17B: ioi_2025=441.3." }, { "model_id": "qwen3.5-397b", "benchmark_id": "livecodebench_v6", "score": 79.3, "reference_url": "https://research.nvidia.com/labs/nemotron/files/NVIDIA-Nemotron-3-Ultra-Technical-Report.pdf", "reported_setting": { "version_metric": "v6 / pass@1", "tools": "code execution", "harness": "NeMo Gym", "sampling": "pass@1" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "NVIDIA Nemotron 3 Ultra technical report, Final Table 10: Qwen3.5-397B-A17B: livecodebench_v6=79.3." }, { "model_id": "qwen3.5-397b", "benchmark_id": "apex_shortlist", "score": 61.4, "reference_url": "https://research.nvidia.com/labs/nemotron/files/NVIDIA-Nemotron-3-Ultra-Technical-Report.pdf", "reported_setting": { "version_metric": "no tools / %", "tools": "none", "harness": "NeMo Evaluator", "sampling": "pass@1", "variant": "no tools" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "NVIDIA Nemotron 3 Ultra technical report, Final Table 10: Qwen3.5-397B-A17B: apex_shortlist=61.4, variant=no tools.", "candidates": [ { "score": 60.4, "reference_url": "https://research.nvidia.com/labs/nemotron/files/NVIDIA-Nemotron-3-Ultra-Technical-Report.pdf", "source_type": "tech_report", "reported_setting": { "version_metric": "with tools / %", "tools": "benchmark tool environment", "harness": "NeMo Evaluator", "sampling": "pass@1", "variant": "with tools" }, "notes": "NVIDIA Nemotron 3 Ultra technical report, Final Table 10: Qwen3.5-397B-A17B: apex_shortlist=60.4, variant=with tools." } ] }, { "model_id": "qwen3.5-397b", "benchmark_id": "hle_text", "score": 28.5, "reference_url": "https://research.nvidia.com/labs/nemotron/files/NVIDIA-Nemotron-3-Ultra-Technical-Report.pdf", "reported_setting": { "version_metric": "text-only / no tools / %", "tools": "none", "harness": "NeMo Evaluator", "sampling": "pass@1", "variant": "no tools" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "NVIDIA Nemotron 3 Ultra technical report, Final Table 10: Qwen3.5-397B-A17B: hle_text=28.5, variant=no tools.", "candidates": [ { "score": 48.3, "reference_url": "https://research.nvidia.com/labs/nemotron/files/NVIDIA-Nemotron-3-Ultra-Technical-Report.pdf", "source_type": "tech_report", "reported_setting": { "version_metric": "text-only / with tools / %", "tools": "benchmark tool environment", "harness": "NeMo Evaluator", "sampling": "pass@1", "variant": "with tools" }, "notes": "NVIDIA Nemotron 3 Ultra technical report, Final Table 10: Qwen3.5-397B-A17B: hle_text=48.3, variant=with tools." } ] }, { "model_id": "qwen3.5-397b", "benchmark_id": "critpt", "score": 2.4, "reference_url": "https://research.nvidia.com/labs/nemotron/files/NVIDIA-Nemotron-3-Ultra-Technical-Report.pdf", "reported_setting": { "version_metric": "no tools / %", "tools": "none", "harness": "NeMo Evaluator", "sampling": "5-run average" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "NVIDIA Nemotron 3 Ultra technical report, Final Table 10: Qwen3.5-397B-A17B: critpt=2.4." }, { "model_id": "qwen3.5-397b", "benchmark_id": "aa_omniscience_accuracy", "score": 35.9, "reference_url": "https://research.nvidia.com/labs/nemotron/files/NVIDIA-Nemotron-3-Ultra-Technical-Report.pdf", "reported_setting": { "version_metric": "accuracy / %", "tools": "none", "harness": "NeMo Evaluator", "sampling": "10-run average" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "NVIDIA Nemotron 3 Ultra technical report, Final Table 10: Qwen3.5-397B-A17B: aa_omniscience_accuracy=35.9." }, { "model_id": "qwen3.5-397b", "benchmark_id": "aa_omniscience_non_hallucination", "score": 7.4, "reference_url": "https://research.nvidia.com/labs/nemotron/files/NVIDIA-Nemotron-3-Ultra-Technical-Report.pdf", "reported_setting": { "version_metric": "non-hallucination / %", "tools": "none", "harness": "NeMo Evaluator", "sampling": "10-run average" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "NVIDIA Nemotron 3 Ultra technical report, Final Table 10: Qwen3.5-397B-A17B: aa_omniscience_non_hallucination=7.4." }, { "model_id": "qwen3.5-397b", "benchmark_id": "ruler_1m", "score": 90.1, "reference_url": "https://research.nvidia.com/labs/nemotron/files/NVIDIA-Nemotron-3-Ultra-Technical-Report.pdf", "reported_setting": { "version_metric": "1M / %", "tools": "none", "harness": "NeMo Skills / RULER", "sampling": "pass@1" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "NVIDIA Nemotron 3 Ultra technical report, Final Table 10: Qwen3.5-397B-A17B: ruler_1m=90.1." }, { "model_id": "qwen3.5-397b", "benchmark_id": "wmt24pp", "score": 86.8, "reference_url": "https://research.nvidia.com/labs/nemotron/files/NVIDIA-Nemotron-3-Ultra-Technical-Report.pdf", "reported_setting": { "version_metric": "en-to-55 languages / XCOMET-XXL", "tools": "none", "harness": "NeMo Evaluator", "sampling": "54,890 translations" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "NVIDIA Nemotron 3 Ultra technical report, Final Table 10: Qwen3.5-397B-A17B: wmt24pp=86.8." }, { "model_id": "deepseek-v4-pro", "benchmark_id": "gdpval_normalized_elo", "score": 54.6, "reference_url": "https://research.nvidia.com/labs/nemotron/files/NVIDIA-Nemotron-3-Ultra-Technical-Report.pdf", "reported_setting": { "version_metric": "normalized Elo / 0-100", "tools": "office workflow, web search, sandboxed code", "harness": "NeMo Gym / Stirrup", "sampling": "pass@1" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "NVIDIA Nemotron 3 Ultra technical report, Final Table 10: DeepSeek-V4-Pro: gdpval_normalized_elo=54.6." }, { "model_id": "deepseek-v4-pro", "benchmark_id": "profbench", "score": 59.9, "reference_url": "https://research.nvidia.com/labs/nemotron/files/NVIDIA-Nemotron-3-Ultra-Technical-Report.pdf", "reported_setting": { "version_metric": "full / search score", "tools": "web search and browsing", "harness": "official/internal ProfBench scaffold", "sampling": "16-run average" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "NVIDIA Nemotron 3 Ultra technical report, Final Table 10: DeepSeek-V4-Pro: profbench=59.9." }, { "model_id": "deepseek-v4-pro", "benchmark_id": "pinchbench", "score": 88.6, "reference_url": "https://research.nvidia.com/labs/nemotron/files/NVIDIA-Nemotron-3-Ultra-Technical-Report.pdf", "reported_setting": { "version_metric": "current 53-task release / %", "tools": "OpenClaw coding environment", "harness": "official PinchBench scaffold", "sampling": "pass@1" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "NVIDIA Nemotron 3 Ultra technical report, Final Table 10: DeepSeek-V4-Pro: pinchbench=88.6." }, { "model_id": "deepseek-v4-pro", "benchmark_id": "tau3_airline", "score": 80.8, "reference_url": "https://research.nvidia.com/labs/nemotron/files/NVIDIA-Nemotron-3-Ultra-Technical-Report.pdf", "reported_setting": { "version_metric": "airline / 8-trial average", "tools": "domain customer-service tools", "harness": "NeMo Gym / tau3", "sampling": "50 tasks x 8 trials" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "NVIDIA Nemotron 3 Ultra technical report, Final Table 10: DeepSeek-V4-Pro: tau3_airline=80.8." }, { "model_id": "deepseek-v4-pro", "benchmark_id": "tau3_retail", "score": 88.9, "reference_url": "https://research.nvidia.com/labs/nemotron/files/NVIDIA-Nemotron-3-Ultra-Technical-Report.pdf", "reported_setting": { "version_metric": "retail / 8-trial average", "tools": "domain customer-service tools", "harness": "NeMo Gym / tau3", "sampling": "114 tasks x 8 trials" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "NVIDIA Nemotron 3 Ultra technical report, Final Table 10: DeepSeek-V4-Pro: tau3_retail=88.9." }, { "model_id": "deepseek-v4-pro", "benchmark_id": "tau3_telecom", "score": 96.3, "reference_url": "https://research.nvidia.com/labs/nemotron/files/NVIDIA-Nemotron-3-Ultra-Technical-Report.pdf", "reported_setting": { "version_metric": "telecom / 8-trial average", "tools": "domain customer-service tools", "harness": "NeMo Gym / tau3", "sampling": "114 tasks x 8 trials" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "NVIDIA Nemotron 3 Ultra technical report, Final Table 10: DeepSeek-V4-Pro: tau3_telecom=96.3." }, { "model_id": "deepseek-v4-pro", "benchmark_id": "tau3_banking", "score": 25.9, "reference_url": "https://research.nvidia.com/labs/nemotron/files/NVIDIA-Nemotron-3-Ultra-Technical-Report.pdf", "reported_setting": { "version_metric": "banking_knowledge / 8-trial average", "tools": "domain tools plus terminal knowledge-base search", "harness": "NeMo Gym / tau3", "sampling": "97 tasks x 8 trials" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "NVIDIA Nemotron 3 Ultra technical report, Final Table 10: DeepSeek-V4-Pro: tau3_banking=25.9." }, { "model_id": "deepseek-v4-pro", "benchmark_id": "tau3_bench", "score": 73.2, "reference_url": "https://research.nvidia.com/labs/nemotron/files/NVIDIA-Nemotron-3-Ultra-Technical-Report.pdf", "reported_setting": { "version_metric": "four-domain / 8-trial average", "tools": "domain customer-service tools", "harness": "NeMo Gym / tau3", "sampling": "375 tasks x 8 trials" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "NVIDIA Nemotron 3 Ultra technical report, Final Table 10: DeepSeek-V4-Pro: tau3_bench=73.2.", "candidates": [ { "score": 67.5, "reference_url": "https://ernie.baidu.com/blog/posts/ernie-5.1-0508-release/benchmark.png", "source_type": "official_blog", "reported_setting": { "tools": "agentic", "sampling": "avg@4" }, "notes": "The source's avg@4 agentic setting matches the campaign row. Research observation obs-006." } ] }, { "model_id": "deepseek-v4-pro", "benchmark_id": "ioi_2025", "score": 580.1, "reference_url": "https://research.nvidia.com/labs/nemotron/files/NVIDIA-Nemotron-3-Ultra-Technical-Report.pdf", "reported_setting": { "version_metric": "contest points / 600", "tools": "code execution", "harness": "NVIDIA contest evaluation", "sampling": "pass@1" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "NVIDIA Nemotron 3 Ultra technical report, Final Table 10: DeepSeek-V4-Pro: ioi_2025=580.1." }, { "model_id": "deepseek-v4-pro", "benchmark_id": "livecodebench_v6", "score": 93.5, "reference_url": "https://microsoft.ai/wp-content/uploads/2026/06/main_20260602_2.pdf", "reported_setting": { "mode": "thinking", "effort": "Max", "tools": "none (per-bench override)", "sampling": "pass@1", "judge": "rule-based", "harness": "official", "prompt_style": "default", "temperature": "default", "context": "default", "notes": "Per DeepSeek V4-Pro model card. Three reasoning modes available (Non-Think/High/Max); BP canonical = Max mode (most powerful). Other modes' values stored as candidates if needed." }, "matches_canonical": true, "source_type": "official_paper", "audit_status": "verified", "notes": "Displayed exactly as 93.5. Official Microsoft-reported result. Comparator value is quoted from that provider's official model card/release; use the campaign canonical setting.", "candidates": [ { "score": 89.8, "reference_url": "https://research.nvidia.com/labs/nemotron/files/NVIDIA-Nemotron-3-Ultra-Technical-Report.pdf", "source_type": "tech_report", "reported_setting": { "version_metric": "v6 / %", "tools": "code execution", "harness": "NeMo Evaluator", "sampling": "pass@1", "variant": "High", "effort": "High" }, "notes": "NVIDIA Nemotron 3 Ultra technical report, Internal Table 3: DeepSeek V4 Pro (High): livecodebench_v6=89.8, variant=High." }, { "score": 92.5, "reference_url": "https://research.nvidia.com/labs/nemotron/files/NVIDIA-Nemotron-3-Ultra-Technical-Report.pdf", "source_type": "tech_report", "reported_setting": { "version_metric": "v6 / pass@1", "tools": "code execution", "harness": "NeMo Gym", "sampling": "pass@1" }, "notes": "NVIDIA Nemotron 3 Ultra technical report, Final Table 10: DeepSeek-V4-Pro: livecodebench_v6=92.5." } ] }, { "model_id": "deepseek-v4-pro", "benchmark_id": "aa_omniscience_accuracy", "score": 46.8, "reference_url": "https://research.nvidia.com/labs/nemotron/files/NVIDIA-Nemotron-3-Ultra-Technical-Report.pdf", "reported_setting": { "version_metric": "accuracy / %", "tools": "none", "harness": "NeMo Evaluator", "sampling": "10-run average" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "NVIDIA Nemotron 3 Ultra technical report, Final Table 10: DeepSeek-V4-Pro: aa_omniscience_accuracy=46.8." }, { "model_id": "deepseek-v4-pro", "benchmark_id": "aa_omniscience_non_hallucination", "score": 5.7, "reference_url": "https://research.nvidia.com/labs/nemotron/files/NVIDIA-Nemotron-3-Ultra-Technical-Report.pdf", "reported_setting": { "version_metric": "non-hallucination / %", "tools": "none", "harness": "NeMo Evaluator", "sampling": "10-run average" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "NVIDIA Nemotron 3 Ultra technical report, Final Table 10: DeepSeek-V4-Pro: aa_omniscience_non_hallucination=5.7." }, { "model_id": "deepseek-v4-pro", "benchmark_id": "ifbench", "score": 79.1, "reference_url": "https://research.nvidia.com/labs/nemotron/files/NVIDIA-Nemotron-3-Ultra-Technical-Report.pdf", "reported_setting": { "version_metric": "prompt-level loose accuracy / %", "tools": "none", "harness": "NeMo Evaluator", "sampling": "8-run average" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "NVIDIA Nemotron 3 Ultra technical report, Final Table 10: DeepSeek-V4-Pro: ifbench=79.1." }, { "model_id": "deepseek-v4-pro", "benchmark_id": "multichallenge", "score": 64.1, "reference_url": "https://research.nvidia.com/labs/nemotron/files/NVIDIA-Nemotron-3-Ultra-Technical-Report.pdf", "reported_setting": { "version_metric": "multi-turn score / %", "tools": "none", "harness": "NeMo Evaluator", "sampling": "8-run average" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "NVIDIA Nemotron 3 Ultra technical report, Final Table 10: DeepSeek-V4-Pro: multichallenge=64.1." }, { "model_id": "deepseek-v4-pro", "benchmark_id": "ruler_1m", "score": 94.2, "reference_url": "https://research.nvidia.com/labs/nemotron/files/NVIDIA-Nemotron-3-Ultra-Technical-Report.pdf", "reported_setting": { "version_metric": "1M / %", "tools": "none", "harness": "NeMo Skills / RULER", "sampling": "pass@1" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "NVIDIA Nemotron 3 Ultra technical report, Final Table 10: DeepSeek-V4-Pro: ruler_1m=94.2." }, { "model_id": "deepseek-v4-pro", "benchmark_id": "longbench_v2", "score": 62.1, "reference_url": "https://research.nvidia.com/labs/nemotron/files/NVIDIA-Nemotron-3-Ultra-Technical-Report.pdf", "reported_setting": { "version_metric": "<=1M / %", "tools": "none", "harness": "NeMo Evaluator", "sampling": "pass@1" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "NVIDIA Nemotron 3 Ultra technical report, Final Table 10: DeepSeek-V4-Pro: longbench_v2=62.1." }, { "model_id": "deepseek-v4-pro", "benchmark_id": "wmt24pp", "score": 85.9, "reference_url": "https://research.nvidia.com/labs/nemotron/files/NVIDIA-Nemotron-3-Ultra-Technical-Report.pdf", "reported_setting": { "version_metric": "en-to-55 languages / XCOMET-XXL", "tools": "none", "harness": "NeMo Evaluator", "sampling": "54,890 translations" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "NVIDIA Nemotron 3 Ultra technical report, Final Table 10: DeepSeek-V4-Pro: wmt24pp=85.9." }, { "model_id": "deepseek-v4-flash", "benchmark_id": "gdpval_normalized_elo", "score": 50.2, "reference_url": "https://research.nvidia.com/labs/nemotron/files/NVIDIA-Nemotron-3-Ultra-Technical-Report.pdf", "reported_setting": { "version_metric": "normalized Elo / 0-100", "tools": "office workflow, web search, sandboxed code", "harness": "NeMo Gym / Stirrup", "sampling": "pass@1" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "NVIDIA Nemotron 3 Ultra technical report, Final Table 10: DeepSeek-V4-Flash Preview: gdpval_normalized_elo=50.2." }, { "model_id": "deepseek-v4-flash", "benchmark_id": "profbench", "score": 57.0, "reference_url": "https://research.nvidia.com/labs/nemotron/files/NVIDIA-Nemotron-3-Ultra-Technical-Report.pdf", "reported_setting": { "version_metric": "full / search score", "tools": "web search and browsing", "harness": "official/internal ProfBench scaffold", "sampling": "16-run average" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "NVIDIA Nemotron 3 Ultra technical report, Final Table 10: DeepSeek-V4-Flash Preview: profbench=57." }, { "model_id": "deepseek-v4-flash", "benchmark_id": "pinchbench", "score": 91.3, "reference_url": "https://research.nvidia.com/labs/nemotron/files/NVIDIA-Nemotron-3-Ultra-Technical-Report.pdf", "reported_setting": { "version_metric": "current 53-task release / %", "tools": "OpenClaw coding environment", "harness": "official PinchBench scaffold", "sampling": "pass@1" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "NVIDIA Nemotron 3 Ultra technical report, Final Table 10: DeepSeek-V4-Flash Preview: pinchbench=91.3." }, { "model_id": "deepseek-v4-flash", "benchmark_id": "tau3_airline", "score": 80.8, "reference_url": "https://research.nvidia.com/labs/nemotron/files/NVIDIA-Nemotron-3-Ultra-Technical-Report.pdf", "reported_setting": { "version_metric": "airline / 8-trial average", "tools": "domain customer-service tools", "harness": "NeMo Gym / tau3", "sampling": "50 tasks x 8 trials" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "NVIDIA Nemotron 3 Ultra technical report, Final Table 10: DeepSeek-V4-Flash Preview: tau3_airline=80.8." }, { "model_id": "deepseek-v4-flash", "benchmark_id": "tau3_retail", "score": 89.1, "reference_url": "https://research.nvidia.com/labs/nemotron/files/NVIDIA-Nemotron-3-Ultra-Technical-Report.pdf", "reported_setting": { "version_metric": "retail / 8-trial average", "tools": "domain customer-service tools", "harness": "NeMo Gym / tau3", "sampling": "114 tasks x 8 trials" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "NVIDIA Nemotron 3 Ultra technical report, Final Table 10: DeepSeek-V4-Flash Preview: tau3_retail=89.1." }, { "model_id": "deepseek-v4-flash", "benchmark_id": "tau3_telecom", "score": 98.3, "reference_url": "https://research.nvidia.com/labs/nemotron/files/NVIDIA-Nemotron-3-Ultra-Technical-Report.pdf", "reported_setting": { "version_metric": "telecom / 8-trial average", "tools": "domain customer-service tools", "harness": "NeMo Gym / tau3", "sampling": "114 tasks x 8 trials" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "NVIDIA Nemotron 3 Ultra technical report, Final Table 10: DeepSeek-V4-Flash Preview: tau3_telecom=98.3." }, { "model_id": "deepseek-v4-flash", "benchmark_id": "tau3_banking", "score": 26.7, "reference_url": "https://research.nvidia.com/labs/nemotron/files/NVIDIA-Nemotron-3-Ultra-Technical-Report.pdf", "reported_setting": { "version_metric": "banking_knowledge / 8-trial average", "tools": "domain tools plus terminal knowledge-base search", "harness": "NeMo Gym / tau3", "sampling": "97 tasks x 8 trials" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "NVIDIA Nemotron 3 Ultra technical report, Final Table 10: DeepSeek-V4-Flash Preview: tau3_banking=26.7." }, { "model_id": "deepseek-v4-flash", "benchmark_id": "tau3_bench", "score": 73.7, "reference_url": "https://research.nvidia.com/labs/nemotron/files/NVIDIA-Nemotron-3-Ultra-Technical-Report.pdf", "reported_setting": { "version_metric": "four-domain / 8-trial average", "tools": "domain customer-service tools", "harness": "NeMo Gym / tau3", "sampling": "375 tasks x 8 trials" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "NVIDIA Nemotron 3 Ultra technical report, Final Table 10: DeepSeek-V4-Flash Preview: tau3_bench=73.7." }, { "model_id": "deepseek-v4-flash", "benchmark_id": "livecodebench_v6", "score": 90.9, "reference_url": "https://research.nvidia.com/labs/nemotron/files/NVIDIA-Nemotron-3-Ultra-Technical-Report.pdf", "reported_setting": { "version_metric": "v6 / pass@1", "tools": "code execution", "harness": "NeMo Gym", "sampling": "pass@1" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "NVIDIA Nemotron 3 Ultra technical report, Final Table 10: DeepSeek-V4-Flash Preview: livecodebench_v6=90.9." }, { "model_id": "deepseek-v4-flash", "benchmark_id": "scicode", "score": 48.2, "reference_url": "https://research.nvidia.com/labs/nemotron/files/NVIDIA-Nemotron-3-Ultra-Technical-Report.pdf", "reported_setting": { "version_metric": "subtask / %", "tools": "none", "harness": "NeMo Evaluator", "sampling": "8-run average" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "NVIDIA Nemotron 3 Ultra technical report, Final Table 10: DeepSeek-V4-Flash Preview: scicode=48.2." }, { "model_id": "deepseek-v4-flash", "benchmark_id": "critpt", "score": 10.6, "reference_url": "https://research.nvidia.com/labs/nemotron/files/NVIDIA-Nemotron-3-Ultra-Technical-Report.pdf", "reported_setting": { "version_metric": "no tools / %", "tools": "none", "harness": "NeMo Evaluator", "sampling": "5-run average" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "NVIDIA Nemotron 3 Ultra technical report, Final Table 10: DeepSeek-V4-Flash Preview: critpt=10.6." }, { "model_id": "deepseek-v4-flash", "benchmark_id": "aa_omniscience_accuracy", "score": 39.9, "reference_url": "https://research.nvidia.com/labs/nemotron/files/NVIDIA-Nemotron-3-Ultra-Technical-Report.pdf", "reported_setting": { "version_metric": "accuracy / %", "tools": "none", "harness": "NeMo Evaluator", "sampling": "10-run average" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "NVIDIA Nemotron 3 Ultra technical report, Final Table 10: DeepSeek-V4-Flash Preview: aa_omniscience_accuracy=39.9." }, { "model_id": "deepseek-v4-flash", "benchmark_id": "aa_omniscience_non_hallucination", "score": 2.8, "reference_url": "https://research.nvidia.com/labs/nemotron/files/NVIDIA-Nemotron-3-Ultra-Technical-Report.pdf", "reported_setting": { "version_metric": "non-hallucination / %", "tools": "none", "harness": "NeMo Evaluator", "sampling": "10-run average" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "NVIDIA Nemotron 3 Ultra technical report, Final Table 10: DeepSeek-V4-Flash Preview: aa_omniscience_non_hallucination=2.8." }, { "model_id": "deepseek-v4-flash", "benchmark_id": "ifbench", "score": 82.0, "reference_url": "https://research.nvidia.com/labs/nemotron/files/NVIDIA-Nemotron-3-Ultra-Technical-Report.pdf", "reported_setting": { "version_metric": "prompt-level loose accuracy / %", "tools": "none", "harness": "NeMo Evaluator", "sampling": "8-run average" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "NVIDIA Nemotron 3 Ultra technical report, Final Table 10: DeepSeek-V4-Flash Preview: ifbench=82." }, { "model_id": "deepseek-v4-flash", "benchmark_id": "multichallenge", "score": 63.5, "reference_url": "https://research.nvidia.com/labs/nemotron/files/NVIDIA-Nemotron-3-Ultra-Technical-Report.pdf", "reported_setting": { "version_metric": "multi-turn score / %", "tools": "none", "harness": "NeMo Evaluator", "sampling": "8-run average" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "NVIDIA Nemotron 3 Ultra technical report, Final Table 10: DeepSeek-V4-Flash Preview: multichallenge=63.5." }, { "model_id": "deepseek-v4-flash", "benchmark_id": "ruler_1m", "score": 87.7, "reference_url": "https://research.nvidia.com/labs/nemotron/files/NVIDIA-Nemotron-3-Ultra-Technical-Report.pdf", "reported_setting": { "version_metric": "1M / %", "tools": "none", "harness": "NeMo Skills / RULER", "sampling": "pass@1" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "NVIDIA Nemotron 3 Ultra technical report, Final Table 10: DeepSeek-V4-Flash Preview: ruler_1m=87.7." }, { "model_id": "deepseek-v4-flash", "benchmark_id": "longbench_v2", "score": 57.0, "reference_url": "https://research.nvidia.com/labs/nemotron/files/NVIDIA-Nemotron-3-Ultra-Technical-Report.pdf", "reported_setting": { "version_metric": "<=1M / %", "tools": "none", "harness": "NeMo Evaluator", "sampling": "pass@1" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "NVIDIA Nemotron 3 Ultra technical report, Final Table 10: DeepSeek-V4-Flash Preview: longbench_v2=57." }, { "model_id": "deepseek-v4-flash", "benchmark_id": "wmt24pp", "score": 85.9, "reference_url": "https://research.nvidia.com/labs/nemotron/files/NVIDIA-Nemotron-3-Ultra-Technical-Report.pdf", "reported_setting": { "version_metric": "en-to-55 languages / XCOMET-XXL", "tools": "none", "harness": "NeMo Evaluator", "sampling": "54,890 translations" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "NVIDIA Nemotron 3 Ultra technical report, Final Table 10: DeepSeek-V4-Flash Preview: wmt24pp=85.9." }, { "model_id": "nemotron-3-ultra-550b-a55b", "benchmark_id": "mmlu", "score": 89.08, "reference_url": "https://research.nvidia.com/labs/nemotron/files/NVIDIA-Nemotron-3-Ultra-Technical-Report.pdf", "reported_setting": { "version_metric": "5-shot accuracy", "tools": "none", "harness": "NeMo Evaluator", "sampling": "pass@1", "checkpoint": "base", "mode": "pretrained" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "NVIDIA Nemotron 3 Ultra technical report, Base Table 2: Nemotron 3 Ultra Base: mmlu=89.08. Base checkpoint score mapped to the canonical post-trained model row as checkpoint=base." }, { "model_id": "nemotron-3-ultra-550b-a55b", "benchmark_id": "agieval_en", "score": 78.73, "reference_url": "https://research.nvidia.com/labs/nemotron/files/NVIDIA-Nemotron-3-Ultra-Technical-Report.pdf", "reported_setting": { "version_metric": "3/5-shot CoT exact match", "tools": "none", "harness": "NeMo Evaluator", "sampling": "pass@1", "checkpoint": "base", "mode": "pretrained" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "NVIDIA Nemotron 3 Ultra technical report, Base Table 2: Nemotron 3 Ultra Base: agieval_en=78.73. Base checkpoint score mapped to the canonical post-trained model row as checkpoint=base." }, { "model_id": "nemotron-3-ultra-550b-a55b", "benchmark_id": "gsm8k", "score": 88.1, "reference_url": "https://research.nvidia.com/labs/nemotron/files/NVIDIA-Nemotron-3-Ultra-Technical-Report.pdf", "reported_setting": { "version_metric": "8-shot CoT exact match", "tools": "none", "harness": "NeMo Evaluator", "sampling": "pass@1", "checkpoint": "base", "mode": "pretrained" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "NVIDIA Nemotron 3 Ultra technical report, Base Table 2: Nemotron 3 Ultra Base: gsm8k=88.1. Base checkpoint score mapped to the canonical post-trained model row as checkpoint=base." }, { "model_id": "nemotron-3-ultra-550b-a55b", "benchmark_id": "math_test", "score": 82.0, "reference_url": "https://research.nvidia.com/labs/nemotron/files/NVIDIA-Nemotron-3-Ultra-Technical-Report.pdf", "reported_setting": { "version_metric": "Minerva 4-shot exact match", "tools": "none", "harness": "NeMo Evaluator", "sampling": "pass@1", "checkpoint": "base", "mode": "pretrained" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "NVIDIA Nemotron 3 Ultra technical report, Base Table 2: Nemotron 3 Ultra Base: math_test=82. Base checkpoint score mapped to the canonical post-trained model row as checkpoint=base." }, { "model_id": "nemotron-3-ultra-550b-a55b", "benchmark_id": "humaneval", "score": 83.84, "reference_url": "https://research.nvidia.com/labs/nemotron/files/NVIDIA-Nemotron-3-Ultra-Technical-Report.pdf", "reported_setting": { "version_metric": "EvalPlus sanitized sampled pass@1 n=32", "tools": "none", "harness": "NeMo Evaluator", "sampling": "164 problems x 32 samples", "checkpoint": "base", "mode": "pretrained" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "NVIDIA Nemotron 3 Ultra technical report, Base Table 2: Nemotron 3 Ultra Base: humaneval=83.84. Base checkpoint score mapped to the canonical post-trained model row as checkpoint=base." }, { "model_id": "nemotron-3-ultra-550b-a55b", "benchmark_id": "mbpp_sanitized", "score": 85.97, "reference_url": "https://research.nvidia.com/labs/nemotron/files/NVIDIA-Nemotron-3-Ultra-Technical-Report.pdf", "reported_setting": { "version_metric": "3-shot sampled pass@1 n=32", "tools": "none", "harness": "NeMo Evaluator", "sampling": "427 problems x 32 samples", "checkpoint": "base", "mode": "pretrained" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "NVIDIA Nemotron 3 Ultra technical report, Base Table 2: Nemotron 3 Ultra Base: mbpp_sanitized=85.97. Base checkpoint score mapped to the canonical post-trained model row as checkpoint=base." }, { "model_id": "nemotron-3-ultra-550b-a55b", "benchmark_id": "arc_challenge", "score": 97.35, "reference_url": "https://research.nvidia.com/labs/nemotron/files/NVIDIA-Nemotron-3-Ultra-Technical-Report.pdf", "reported_setting": { "version_metric": "25-shot normalized accuracy", "tools": "none", "harness": "NeMo Evaluator", "sampling": "pass@1", "checkpoint": "base", "mode": "pretrained" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "NVIDIA Nemotron 3 Ultra technical report, Base Table 2: Nemotron 3 Ultra Base: arc_challenge=97.35. Base checkpoint score mapped to the canonical post-trained model row as checkpoint=base." }, { "model_id": "nemotron-3-ultra-550b-a55b", "benchmark_id": "hellaswag", "score": 90.51, "reference_url": "https://research.nvidia.com/labs/nemotron/files/NVIDIA-Nemotron-3-Ultra-Technical-Report.pdf", "reported_setting": { "version_metric": "10-shot normalized accuracy", "tools": "none", "harness": "NeMo Evaluator", "sampling": "pass@1", "checkpoint": "base", "mode": "pretrained" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "NVIDIA Nemotron 3 Ultra technical report, Base Table 2: Nemotron 3 Ultra Base: hellaswag=90.51. Base checkpoint score mapped to the canonical post-trained model row as checkpoint=base." }, { "model_id": "nemotron-3-ultra-550b-a55b", "benchmark_id": "openbookqa", "score": 48.6, "reference_url": "https://research.nvidia.com/labs/nemotron/files/NVIDIA-Nemotron-3-Ultra-Technical-Report.pdf", "reported_setting": { "version_metric": "0-shot normalized accuracy", "tools": "none", "harness": "NeMo Evaluator", "sampling": "pass@1", "checkpoint": "base", "mode": "pretrained" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "NVIDIA Nemotron 3 Ultra technical report, Base Table 2: Nemotron 3 Ultra Base: openbookqa=48.6. Base checkpoint score mapped to the canonical post-trained model row as checkpoint=base." }, { "model_id": "nemotron-3-ultra-550b-a55b", "benchmark_id": "piqa", "score": 83.79, "reference_url": "https://research.nvidia.com/labs/nemotron/files/NVIDIA-Nemotron-3-Ultra-Technical-Report.pdf", "reported_setting": { "version_metric": "0-shot normalized accuracy", "tools": "none", "harness": "NeMo Evaluator", "sampling": "pass@1", "checkpoint": "base", "mode": "pretrained" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "NVIDIA Nemotron 3 Ultra technical report, Base Table 2: Nemotron 3 Ultra Base: piqa=83.79. Base checkpoint score mapped to the canonical post-trained model row as checkpoint=base." }, { "model_id": "nemotron-3-ultra-550b-a55b", "benchmark_id": "winogrande", "score": 79.32, "reference_url": "https://research.nvidia.com/labs/nemotron/files/NVIDIA-Nemotron-3-Ultra-Technical-Report.pdf", "reported_setting": { "version_metric": "5-shot accuracy", "tools": "none", "harness": "NeMo Evaluator", "sampling": "pass@1", "checkpoint": "base", "mode": "pretrained" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "NVIDIA Nemotron 3 Ultra technical report, Base Table 2: Nemotron 3 Ultra Base: winogrande=79.32. Base checkpoint score mapped to the canonical post-trained model row as checkpoint=base." }, { "model_id": "nemotron-3-ultra-550b-a55b", "benchmark_id": "race", "score": 92.15, "reference_url": "https://research.nvidia.com/labs/nemotron/files/NVIDIA-Nemotron-3-Ultra-Technical-Report.pdf", "reported_setting": { "version_metric": "0-shot accuracy", "tools": "none", "harness": "NeMo Evaluator", "sampling": "pass@1", "checkpoint": "base", "mode": "pretrained" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "NVIDIA Nemotron 3 Ultra technical report, Base Table 2: Nemotron 3 Ultra Base: race=92.15. Base checkpoint score mapped to the canonical post-trained model row as checkpoint=base." }, { "model_id": "nemotron-3-ultra-550b-a55b", "benchmark_id": "global_mmlu_lite", "score": 90.13, "reference_url": "https://research.nvidia.com/labs/nemotron/files/NVIDIA-Nemotron-3-Ultra-Technical-Report.pdf", "reported_setting": { "version_metric": "5-shot average", "tools": "none", "harness": "NeMo Evaluator", "sampling": "pass@1", "checkpoint": "base", "mode": "pretrained" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "NVIDIA Nemotron 3 Ultra technical report, Base Table 2: Nemotron 3 Ultra Base: global_mmlu_lite=90.13. Base checkpoint score mapped to the canonical post-trained model row as checkpoint=base." }, { "model_id": "nemotron-3-ultra-550b-a55b", "benchmark_id": "mgsm", "score": 87.73, "reference_url": "https://research.nvidia.com/labs/nemotron/files/NVIDIA-Nemotron-3-Ultra-Technical-Report.pdf", "reported_setting": { "version_metric": "8-shot native CoT average", "tools": "none", "harness": "NeMo Evaluator", "sampling": "pass@1", "checkpoint": "base", "mode": "pretrained" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "NVIDIA Nemotron 3 Ultra technical report, Base Table 2: Nemotron 3 Ultra Base: mgsm=87.73. Base checkpoint score mapped to the canonical post-trained model row as checkpoint=base." }, { "model_id": "nemotron-3-ultra-550b-a55b", "benchmark_id": "ruler_64k", "score": 95.3, "reference_url": "https://research.nvidia.com/labs/nemotron/files/NVIDIA-Nemotron-3-Ultra-Technical-Report.pdf", "reported_setting": { "version_metric": "0-shot", "tools": "none", "harness": "NeMo Skills / RULER", "sampling": "pass@1", "checkpoint": "base", "mode": "pretrained" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "NVIDIA Nemotron 3 Ultra technical report, Base Table 2: Nemotron 3 Ultra Base: ruler_64k=95.3. Base checkpoint score mapped to the canonical post-trained model row as checkpoint=base." }, { "model_id": "nemotron-3-ultra-550b-a55b", "benchmark_id": "ruler_128k", "score": 92.49, "reference_url": "https://research.nvidia.com/labs/nemotron/files/NVIDIA-Nemotron-3-Ultra-Technical-Report.pdf", "reported_setting": { "version_metric": "0-shot", "tools": "none", "harness": "NeMo Skills / RULER", "sampling": "pass@1", "checkpoint": "base", "mode": "pretrained" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "NVIDIA Nemotron 3 Ultra technical report, Base Table 2: Nemotron 3 Ultra Base: ruler_128k=92.49. Base checkpoint score mapped to the canonical post-trained model row as checkpoint=base." }, { "model_id": "nemotron-3-ultra-550b-a55b", "benchmark_id": "ruler_256k", "score": 86.22, "reference_url": "https://research.nvidia.com/labs/nemotron/files/NVIDIA-Nemotron-3-Ultra-Technical-Report.pdf", "reported_setting": { "version_metric": "0-shot", "tools": "none", "harness": "NeMo Skills / RULER", "sampling": "pass@1", "checkpoint": "base", "mode": "pretrained" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "NVIDIA Nemotron 3 Ultra technical report, Base Table 2: Nemotron 3 Ultra Base: ruler_256k=86.22. Base checkpoint score mapped to the canonical post-trained model row as checkpoint=base." }, { "model_id": "nemotron-3-ultra-550b-a55b", "benchmark_id": "ruler_512k", "score": 84.54, "reference_url": "https://research.nvidia.com/labs/nemotron/files/NVIDIA-Nemotron-3-Ultra-Technical-Report.pdf", "reported_setting": { "version_metric": "0-shot", "tools": "none", "harness": "NeMo Skills / RULER", "sampling": "pass@1", "checkpoint": "base", "mode": "pretrained" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "NVIDIA Nemotron 3 Ultra technical report, Base Table 2: Nemotron 3 Ultra Base: ruler_512k=84.54. Base checkpoint score mapped to the canonical post-trained model row as checkpoint=base." }, { "model_id": "deepseek-v3.2", "benchmark_id": "agieval_en", "score": 70.13, "reference_url": "https://research.nvidia.com/labs/nemotron/files/NVIDIA-Nemotron-3-Ultra-Technical-Report.pdf", "reported_setting": { "version_metric": "3/5-shot CoT exact match", "tools": "none", "harness": "NeMo Evaluator", "sampling": "pass@1", "checkpoint": "base", "mode": "pretrained" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "NVIDIA Nemotron 3 Ultra technical report, Base Table 2: DeepSeek-V3.2-Exp-Base: agieval_en=70.13. Base checkpoint score mapped to the canonical post-trained model row as checkpoint=base." }, { "model_id": "deepseek-v3.2", "benchmark_id": "gsm8k", "score": 84.38, "reference_url": "https://research.nvidia.com/labs/nemotron/files/NVIDIA-Nemotron-3-Ultra-Technical-Report.pdf", "reported_setting": { "version_metric": "8-shot CoT exact match", "tools": "none", "harness": "NeMo Evaluator", "sampling": "pass@1", "checkpoint": "base", "mode": "pretrained" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "NVIDIA Nemotron 3 Ultra technical report, Base Table 2: DeepSeek-V3.2-Exp-Base: gsm8k=84.38. Base checkpoint score mapped to the canonical post-trained model row as checkpoint=base." }, { "model_id": "deepseek-v3.2", "benchmark_id": "math_test", "score": 60.12, "reference_url": "https://research.nvidia.com/labs/nemotron/files/NVIDIA-Nemotron-3-Ultra-Technical-Report.pdf", "reported_setting": { "version_metric": "Minerva 4-shot exact match", "tools": "none", "harness": "NeMo Evaluator", "sampling": "pass@1", "checkpoint": "base", "mode": "pretrained" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "NVIDIA Nemotron 3 Ultra technical report, Base Table 2: DeepSeek-V3.2-Exp-Base: math_test=60.12. Base checkpoint score mapped to the canonical post-trained model row as checkpoint=base." }, { "model_id": "deepseek-v3.2", "benchmark_id": "mbpp_sanitized", "score": 58.66, "reference_url": "https://research.nvidia.com/labs/nemotron/files/NVIDIA-Nemotron-3-Ultra-Technical-Report.pdf", "reported_setting": { "version_metric": "3-shot sampled pass@1 n=32", "tools": "none", "harness": "NeMo Evaluator", "sampling": "427 problems x 32 samples", "checkpoint": "base", "mode": "pretrained" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "NVIDIA Nemotron 3 Ultra technical report, Base Table 2: DeepSeek-V3.2-Exp-Base: mbpp_sanitized=58.66. Base checkpoint score mapped to the canonical post-trained model row as checkpoint=base." }, { "model_id": "deepseek-v3.2", "benchmark_id": "arc_challenge", "score": 95.22, "reference_url": "https://research.nvidia.com/labs/nemotron/files/NVIDIA-Nemotron-3-Ultra-Technical-Report.pdf", "reported_setting": { "version_metric": "25-shot normalized accuracy", "tools": "none", "harness": "NeMo Evaluator", "sampling": "pass@1", "checkpoint": "base", "mode": "pretrained" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "NVIDIA Nemotron 3 Ultra technical report, Base Table 2: DeepSeek-V3.2-Exp-Base: arc_challenge=95.22. Base checkpoint score mapped to the canonical post-trained model row as checkpoint=base." }, { "model_id": "deepseek-v3.2", "benchmark_id": "hellaswag", "score": 89.44, "reference_url": "https://research.nvidia.com/labs/nemotron/files/NVIDIA-Nemotron-3-Ultra-Technical-Report.pdf", "reported_setting": { "version_metric": "10-shot normalized accuracy", "tools": "none", "harness": "NeMo Evaluator", "sampling": "pass@1", "checkpoint": "base", "mode": "pretrained" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "NVIDIA Nemotron 3 Ultra technical report, Base Table 2: DeepSeek-V3.2-Exp-Base: hellaswag=89.44. Base checkpoint score mapped to the canonical post-trained model row as checkpoint=base." }, { "model_id": "deepseek-v3.2", "benchmark_id": "openbookqa", "score": 48.2, "reference_url": "https://research.nvidia.com/labs/nemotron/files/NVIDIA-Nemotron-3-Ultra-Technical-Report.pdf", "reported_setting": { "version_metric": "0-shot normalized accuracy", "tools": "none", "harness": "NeMo Evaluator", "sampling": "pass@1", "checkpoint": "base", "mode": "pretrained" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "NVIDIA Nemotron 3 Ultra technical report, Base Table 2: DeepSeek-V3.2-Exp-Base: openbookqa=48.2. Base checkpoint score mapped to the canonical post-trained model row as checkpoint=base." }, { "model_id": "deepseek-v3.2", "benchmark_id": "piqa", "score": 85.09, "reference_url": "https://research.nvidia.com/labs/nemotron/files/NVIDIA-Nemotron-3-Ultra-Technical-Report.pdf", "reported_setting": { "version_metric": "0-shot normalized accuracy", "tools": "none", "harness": "NeMo Evaluator", "sampling": "pass@1", "checkpoint": "base", "mode": "pretrained" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "NVIDIA Nemotron 3 Ultra technical report, Base Table 2: DeepSeek-V3.2-Exp-Base: piqa=85.09. Base checkpoint score mapped to the canonical post-trained model row as checkpoint=base." }, { "model_id": "deepseek-v3.2", "benchmark_id": "winogrande", "score": 83.43, "reference_url": "https://research.nvidia.com/labs/nemotron/files/NVIDIA-Nemotron-3-Ultra-Technical-Report.pdf", "reported_setting": { "version_metric": "5-shot accuracy", "tools": "none", "harness": "NeMo Evaluator", "sampling": "pass@1", "checkpoint": "base", "mode": "pretrained" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "NVIDIA Nemotron 3 Ultra technical report, Base Table 2: DeepSeek-V3.2-Exp-Base: winogrande=83.43. Base checkpoint score mapped to the canonical post-trained model row as checkpoint=base." }, { "model_id": "deepseek-v3.2", "benchmark_id": "race", "score": 93.21, "reference_url": "https://research.nvidia.com/labs/nemotron/files/NVIDIA-Nemotron-3-Ultra-Technical-Report.pdf", "reported_setting": { "version_metric": "0-shot accuracy", "tools": "none", "harness": "NeMo Evaluator", "sampling": "pass@1", "checkpoint": "base", "mode": "pretrained" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "NVIDIA Nemotron 3 Ultra technical report, Base Table 2: DeepSeek-V3.2-Exp-Base: race=93.21. Base checkpoint score mapped to the canonical post-trained model row as checkpoint=base." }, { "model_id": "deepseek-v3.2", "benchmark_id": "global_mmlu_lite", "score": 85.59, "reference_url": "https://research.nvidia.com/labs/nemotron/files/NVIDIA-Nemotron-3-Ultra-Technical-Report.pdf", "reported_setting": { "version_metric": "5-shot average", "tools": "none", "harness": "NeMo Evaluator", "sampling": "pass@1", "checkpoint": "base", "mode": "pretrained" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "NVIDIA Nemotron 3 Ultra technical report, Base Table 2: DeepSeek-V3.2-Exp-Base: global_mmlu_lite=85.59. Base checkpoint score mapped to the canonical post-trained model row as checkpoint=base." }, { "model_id": "deepseek-v3.2", "benchmark_id": "mgsm", "score": 82.33, "reference_url": "https://research.nvidia.com/labs/nemotron/files/NVIDIA-Nemotron-3-Ultra-Technical-Report.pdf", "reported_setting": { "version_metric": "8-shot native CoT average", "tools": "none", "harness": "NeMo Evaluator", "sampling": "pass@1", "checkpoint": "base", "mode": "pretrained" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "NVIDIA Nemotron 3 Ultra technical report, Base Table 2: DeepSeek-V3.2-Exp-Base: mgsm=82.33. Base checkpoint score mapped to the canonical post-trained model row as checkpoint=base." }, { "model_id": "deepseek-v3.2", "benchmark_id": "ruler_64k", "score": 93.3, "reference_url": "https://research.nvidia.com/labs/nemotron/files/NVIDIA-Nemotron-3-Ultra-Technical-Report.pdf", "reported_setting": { "version_metric": "0-shot", "tools": "none", "harness": "NeMo Skills / RULER", "sampling": "pass@1", "checkpoint": "base", "mode": "pretrained" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "NVIDIA Nemotron 3 Ultra technical report, Base Table 2: DeepSeek-V3.2-Exp-Base: ruler_64k=93.3. Base checkpoint score mapped to the canonical post-trained model row as checkpoint=base." }, { "model_id": "deepseek-v3.2", "benchmark_id": "ruler_128k", "score": 91.88, "reference_url": "https://research.nvidia.com/labs/nemotron/files/NVIDIA-Nemotron-3-Ultra-Technical-Report.pdf", "reported_setting": { "version_metric": "0-shot", "tools": "none", "harness": "NeMo Skills / RULER", "sampling": "pass@1", "checkpoint": "base", "mode": "pretrained" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "NVIDIA Nemotron 3 Ultra technical report, Base Table 2: DeepSeek-V3.2-Exp-Base: ruler_128k=91.88. Base checkpoint score mapped to the canonical post-trained model row as checkpoint=base." }, { "model_id": "mistral-large-3", "benchmark_id": "agieval_en", "score": 69.3, "reference_url": "https://research.nvidia.com/labs/nemotron/files/NVIDIA-Nemotron-3-Ultra-Technical-Report.pdf", "reported_setting": { "version_metric": "3/5-shot CoT exact match", "tools": "none", "harness": "NeMo Evaluator", "sampling": "pass@1", "checkpoint": "base", "mode": "pretrained" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "NVIDIA Nemotron 3 Ultra technical report, Base Table 2: Mistral Large 3 675B Base 2512: agieval_en=69.3. Base checkpoint score mapped to the canonical post-trained model row as checkpoint=base." }, { "model_id": "mistral-large-3", "benchmark_id": "gsm8k", "score": 91.21, "reference_url": "https://research.nvidia.com/labs/nemotron/files/NVIDIA-Nemotron-3-Ultra-Technical-Report.pdf", "reported_setting": { "version_metric": "8-shot CoT exact match", "tools": "none", "harness": "NeMo Evaluator", "sampling": "pass@1", "checkpoint": "base", "mode": "pretrained" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "NVIDIA Nemotron 3 Ultra technical report, Base Table 2: Mistral Large 3 675B Base 2512: gsm8k=91.21. Base checkpoint score mapped to the canonical post-trained model row as checkpoint=base." }, { "model_id": "mistral-large-3", "benchmark_id": "math_test", "score": 62.88, "reference_url": "https://research.nvidia.com/labs/nemotron/files/NVIDIA-Nemotron-3-Ultra-Technical-Report.pdf", "reported_setting": { "version_metric": "Minerva 4-shot exact match", "tools": "none", "harness": "NeMo Evaluator", "sampling": "pass@1", "checkpoint": "base", "mode": "pretrained" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "NVIDIA Nemotron 3 Ultra technical report, Base Table 2: Mistral Large 3 675B Base 2512: math_test=62.88. Base checkpoint score mapped to the canonical post-trained model row as checkpoint=base." }, { "model_id": "mistral-large-3", "benchmark_id": "mbpp_sanitized", "score": 84.08, "reference_url": "https://research.nvidia.com/labs/nemotron/files/NVIDIA-Nemotron-3-Ultra-Technical-Report.pdf", "reported_setting": { "version_metric": "3-shot sampled pass@1 n=32", "tools": "none", "harness": "NeMo Evaluator", "sampling": "427 problems x 32 samples", "checkpoint": "base", "mode": "pretrained" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "NVIDIA Nemotron 3 Ultra technical report, Base Table 2: Mistral Large 3 675B Base 2512: mbpp_sanitized=84.08. Base checkpoint score mapped to the canonical post-trained model row as checkpoint=base." }, { "model_id": "mistral-large-3", "benchmark_id": "arc_challenge", "score": 97.27, "reference_url": "https://research.nvidia.com/labs/nemotron/files/NVIDIA-Nemotron-3-Ultra-Technical-Report.pdf", "reported_setting": { "version_metric": "25-shot normalized accuracy", "tools": "none", "harness": "NeMo Evaluator", "sampling": "pass@1", "checkpoint": "base", "mode": "pretrained" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "NVIDIA Nemotron 3 Ultra technical report, Base Table 2: Mistral Large 3 675B Base 2512: arc_challenge=97.27. Base checkpoint score mapped to the canonical post-trained model row as checkpoint=base." }, { "model_id": "mistral-large-3", "benchmark_id": "hellaswag", "score": 88.88, "reference_url": "https://research.nvidia.com/labs/nemotron/files/NVIDIA-Nemotron-3-Ultra-Technical-Report.pdf", "reported_setting": { "version_metric": "10-shot normalized accuracy", "tools": "none", "harness": "NeMo Evaluator", "sampling": "pass@1", "checkpoint": "base", "mode": "pretrained" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "NVIDIA Nemotron 3 Ultra technical report, Base Table 2: Mistral Large 3 675B Base 2512: hellaswag=88.88. Base checkpoint score mapped to the canonical post-trained model row as checkpoint=base." }, { "model_id": "mistral-large-3", "benchmark_id": "openbookqa", "score": 51.4, "reference_url": "https://research.nvidia.com/labs/nemotron/files/NVIDIA-Nemotron-3-Ultra-Technical-Report.pdf", "reported_setting": { "version_metric": "0-shot normalized accuracy", "tools": "none", "harness": "NeMo Evaluator", "sampling": "pass@1", "checkpoint": "base", "mode": "pretrained" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "NVIDIA Nemotron 3 Ultra technical report, Base Table 2: Mistral Large 3 675B Base 2512: openbookqa=51.4. Base checkpoint score mapped to the canonical post-trained model row as checkpoint=base." }, { "model_id": "mistral-large-3", "benchmark_id": "piqa", "score": 84.82, "reference_url": "https://research.nvidia.com/labs/nemotron/files/NVIDIA-Nemotron-3-Ultra-Technical-Report.pdf", "reported_setting": { "version_metric": "0-shot normalized accuracy", "tools": "none", "harness": "NeMo Evaluator", "sampling": "pass@1", "checkpoint": "base", "mode": "pretrained" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "NVIDIA Nemotron 3 Ultra technical report, Base Table 2: Mistral Large 3 675B Base 2512: piqa=84.82. Base checkpoint score mapped to the canonical post-trained model row as checkpoint=base." }, { "model_id": "mistral-large-3", "benchmark_id": "winogrande", "score": 82.08, "reference_url": "https://research.nvidia.com/labs/nemotron/files/NVIDIA-Nemotron-3-Ultra-Technical-Report.pdf", "reported_setting": { "version_metric": "5-shot accuracy", "tools": "none", "harness": "NeMo Evaluator", "sampling": "pass@1", "checkpoint": "base", "mode": "pretrained" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "NVIDIA Nemotron 3 Ultra technical report, Base Table 2: Mistral Large 3 675B Base 2512: winogrande=82.08. Base checkpoint score mapped to the canonical post-trained model row as checkpoint=base." }, { "model_id": "mistral-large-3", "benchmark_id": "race", "score": 93.3, "reference_url": "https://research.nvidia.com/labs/nemotron/files/NVIDIA-Nemotron-3-Ultra-Technical-Report.pdf", "reported_setting": { "version_metric": "0-shot accuracy", "tools": "none", "harness": "NeMo Evaluator", "sampling": "pass@1", "checkpoint": "base", "mode": "pretrained" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "NVIDIA Nemotron 3 Ultra technical report, Base Table 2: Mistral Large 3 675B Base 2512: race=93.3. Base checkpoint score mapped to the canonical post-trained model row as checkpoint=base." }, { "model_id": "mistral-large-3", "benchmark_id": "global_mmlu_lite", "score": 87.34, "reference_url": "https://research.nvidia.com/labs/nemotron/files/NVIDIA-Nemotron-3-Ultra-Technical-Report.pdf", "reported_setting": { "version_metric": "5-shot average", "tools": "none", "harness": "NeMo Evaluator", "sampling": "pass@1", "checkpoint": "base", "mode": "pretrained" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "NVIDIA Nemotron 3 Ultra technical report, Base Table 2: Mistral Large 3 675B Base 2512: global_mmlu_lite=87.34. Base checkpoint score mapped to the canonical post-trained model row as checkpoint=base." }, { "model_id": "mistral-large-3", "benchmark_id": "mgsm", "score": 82.93, "reference_url": "https://research.nvidia.com/labs/nemotron/files/NVIDIA-Nemotron-3-Ultra-Technical-Report.pdf", "reported_setting": { "version_metric": "8-shot native CoT average", "tools": "none", "harness": "NeMo Evaluator", "sampling": "pass@1", "checkpoint": "base", "mode": "pretrained" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "NVIDIA Nemotron 3 Ultra technical report, Base Table 2: Mistral Large 3 675B Base 2512: mgsm=82.93. Base checkpoint score mapped to the canonical post-trained model row as checkpoint=base." }, { "model_id": "mistral-large-3", "benchmark_id": "ruler_64k", "score": 90.11, "reference_url": "https://research.nvidia.com/labs/nemotron/files/NVIDIA-Nemotron-3-Ultra-Technical-Report.pdf", "reported_setting": { "version_metric": "0-shot", "tools": "none", "harness": "NeMo Skills / RULER", "sampling": "pass@1", "checkpoint": "base", "mode": "pretrained" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "NVIDIA Nemotron 3 Ultra technical report, Base Table 2: Mistral Large 3 675B Base 2512: ruler_64k=90.11. Base checkpoint score mapped to the canonical post-trained model row as checkpoint=base." }, { "model_id": "mistral-large-3", "benchmark_id": "ruler_128k", "score": 55.77, "reference_url": "https://research.nvidia.com/labs/nemotron/files/NVIDIA-Nemotron-3-Ultra-Technical-Report.pdf", "reported_setting": { "version_metric": "0-shot", "tools": "none", "harness": "NeMo Skills / RULER", "sampling": "pass@1", "checkpoint": "base", "mode": "pretrained" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "NVIDIA Nemotron 3 Ultra technical report, Base Table 2: Mistral Large 3 675B Base 2512: ruler_128k=55.77. Base checkpoint score mapped to the canonical post-trained model row as checkpoint=base." }, { "model_id": "mistral-large-3", "benchmark_id": "ruler_256k", "score": 35.5, "reference_url": "https://research.nvidia.com/labs/nemotron/files/NVIDIA-Nemotron-3-Ultra-Technical-Report.pdf", "reported_setting": { "version_metric": "0-shot", "tools": "none", "harness": "NeMo Skills / RULER", "sampling": "pass@1", "checkpoint": "base", "mode": "pretrained" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "NVIDIA Nemotron 3 Ultra technical report, Base Table 2: Mistral Large 3 675B Base 2512: ruler_256k=35.5. Base checkpoint score mapped to the canonical post-trained model row as checkpoint=base." }, { "model_id": "kimi-k2", "benchmark_id": "agieval_en", "score": 72.55, "reference_url": "https://research.nvidia.com/labs/nemotron/files/NVIDIA-Nemotron-3-Ultra-Technical-Report.pdf", "reported_setting": { "version_metric": "3/5-shot CoT exact match", "tools": "none", "harness": "NeMo Evaluator", "sampling": "pass@1", "checkpoint": "base", "mode": "pretrained" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "NVIDIA Nemotron 3 Ultra technical report, Base Table 2: Kimi K2 Base: agieval_en=72.55. Base checkpoint score mapped to the canonical post-trained model row as checkpoint=base." }, { "model_id": "kimi-k2", "benchmark_id": "math_test", "score": 68.4, "reference_url": "https://research.nvidia.com/labs/nemotron/files/NVIDIA-Nemotron-3-Ultra-Technical-Report.pdf", "reported_setting": { "version_metric": "Minerva 4-shot exact match", "tools": "none", "harness": "NeMo Evaluator", "sampling": "pass@1", "checkpoint": "base", "mode": "pretrained" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "NVIDIA Nemotron 3 Ultra technical report, Base Table 2: Kimi K2 Base: math_test=68.4. Base checkpoint score mapped to the canonical post-trained model row as checkpoint=base." }, { "model_id": "kimi-k2", "benchmark_id": "mbpp_sanitized", "score": 72.14, "reference_url": "https://research.nvidia.com/labs/nemotron/files/NVIDIA-Nemotron-3-Ultra-Technical-Report.pdf", "reported_setting": { "version_metric": "3-shot sampled pass@1 n=32", "tools": "none", "harness": "NeMo Evaluator", "sampling": "427 problems x 32 samples", "checkpoint": "base", "mode": "pretrained" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "NVIDIA Nemotron 3 Ultra technical report, Base Table 2: Kimi K2 Base: mbpp_sanitized=72.14. Base checkpoint score mapped to the canonical post-trained model row as checkpoint=base." }, { "model_id": "kimi-k2", "benchmark_id": "arc_challenge", "score": 95.82, "reference_url": "https://research.nvidia.com/labs/nemotron/files/NVIDIA-Nemotron-3-Ultra-Technical-Report.pdf", "reported_setting": { "version_metric": "25-shot normalized accuracy", "tools": "none", "harness": "NeMo Evaluator", "sampling": "pass@1", "checkpoint": "base", "mode": "pretrained" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "NVIDIA Nemotron 3 Ultra technical report, Base Table 2: Kimi K2 Base: arc_challenge=95.82. Base checkpoint score mapped to the canonical post-trained model row as checkpoint=base.", "candidates": [ { "score": 96.2, "reference_url": "https://huggingface.co/XiaomiMiMo/MiMo-V2.5-Pro/resolve/21d1ecfecd7bd70f31be25ca49d7edd21f003659/README.md", "source_type": "official_model_card_base_checkpoint", "reported_setting": { "mode": "base pretrained checkpoint", "effort": "not applicable", "tools": "none", "sampling": "25-shot; single completion per prompt", "judge": "benchmark-native rule-based evaluator", "harness": "Xiaomi MiMo base-checkpoint evaluation", "prompt_style": "25-shot few-shot continuation", "temperature": "source does not state", "context": "source does not state", "input_modalities": "text", "trials": "one completion per item", "checkpoint": "base" }, "notes": "Base checkpoint observation attached to the canonical post-trained model ID. Base result is attached to the canonical post-trained model ID. The Xiaomi table does not state this comparator Base checkpoint's context length, and no comparator Base config is locked; no Pro context value or conflict is copied onto this observation." } ] }, { "model_id": "kimi-k2", "benchmark_id": "hellaswag", "score": 90.92, "reference_url": "https://research.nvidia.com/labs/nemotron/files/NVIDIA-Nemotron-3-Ultra-Technical-Report.pdf", "reported_setting": { "version_metric": "10-shot normalized accuracy", "tools": "none", "harness": "NeMo Evaluator", "sampling": "pass@1", "checkpoint": "base", "mode": "pretrained" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "NVIDIA Nemotron 3 Ultra technical report, Base Table 2: Kimi K2 Base: hellaswag=90.92. Base checkpoint score mapped to the canonical post-trained model row as checkpoint=base.", "candidates": [ { "score": 94.6, "reference_url": "https://huggingface.co/XiaomiMiMo/MiMo-V2.5-Pro/resolve/21d1ecfecd7bd70f31be25ca49d7edd21f003659/README.md", "source_type": "official_model_card_base_checkpoint", "reported_setting": { "mode": "base pretrained checkpoint", "effort": "not applicable", "tools": "none", "sampling": "10-shot; single completion per prompt", "judge": "benchmark-native rule-based evaluator", "harness": "Xiaomi MiMo base-checkpoint evaluation", "prompt_style": "10-shot few-shot continuation", "temperature": "source does not state", "context": "source does not state", "input_modalities": "text", "trials": "one completion per item", "checkpoint": "base" }, "notes": "Base checkpoint observation attached to the canonical post-trained model ID. Base result is attached to the canonical post-trained model ID. The Xiaomi table does not state this comparator Base checkpoint's context length, and no comparator Base config is locked; no Pro context value or conflict is copied onto this observation." } ] }, { "model_id": "kimi-k2", "benchmark_id": "openbookqa", "score": 50.8, "reference_url": "https://research.nvidia.com/labs/nemotron/files/NVIDIA-Nemotron-3-Ultra-Technical-Report.pdf", "reported_setting": { "version_metric": "0-shot normalized accuracy", "tools": "none", "harness": "NeMo Evaluator", "sampling": "pass@1", "checkpoint": "base", "mode": "pretrained" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "NVIDIA Nemotron 3 Ultra technical report, Base Table 2: Kimi K2 Base: openbookqa=50.8. Base checkpoint score mapped to the canonical post-trained model row as checkpoint=base." }, { "model_id": "kimi-k2", "benchmark_id": "piqa", "score": 85.47, "reference_url": "https://research.nvidia.com/labs/nemotron/files/NVIDIA-Nemotron-3-Ultra-Technical-Report.pdf", "reported_setting": { "version_metric": "0-shot normalized accuracy", "tools": "none", "harness": "NeMo Evaluator", "sampling": "pass@1", "checkpoint": "base", "mode": "pretrained" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "NVIDIA Nemotron 3 Ultra technical report, Base Table 2: Kimi K2 Base: piqa=85.47. Base checkpoint score mapped to the canonical post-trained model row as checkpoint=base." }, { "model_id": "kimi-k2", "benchmark_id": "winogrande", "score": 84.21, "reference_url": "https://research.nvidia.com/labs/nemotron/files/NVIDIA-Nemotron-3-Ultra-Technical-Report.pdf", "reported_setting": { "version_metric": "5-shot accuracy", "tools": "none", "harness": "NeMo Evaluator", "sampling": "pass@1", "checkpoint": "base", "mode": "pretrained" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "NVIDIA Nemotron 3 Ultra technical report, Base Table 2: Kimi K2 Base: winogrande=84.21. Base checkpoint score mapped to the canonical post-trained model row as checkpoint=base." }, { "model_id": "kimi-k2", "benchmark_id": "race", "score": 91.96, "reference_url": "https://research.nvidia.com/labs/nemotron/files/NVIDIA-Nemotron-3-Ultra-Technical-Report.pdf", "reported_setting": { "version_metric": "0-shot accuracy", "tools": "none", "harness": "NeMo Evaluator", "sampling": "pass@1", "checkpoint": "base", "mode": "pretrained" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "NVIDIA Nemotron 3 Ultra technical report, Base Table 2: Kimi K2 Base: race=91.96. Base checkpoint score mapped to the canonical post-trained model row as checkpoint=base." }, { "model_id": "kimi-k2", "benchmark_id": "global_mmlu_lite", "score": 85.63, "reference_url": "https://research.nvidia.com/labs/nemotron/files/NVIDIA-Nemotron-3-Ultra-Technical-Report.pdf", "reported_setting": { "version_metric": "5-shot average", "tools": "none", "harness": "NeMo Evaluator", "sampling": "pass@1", "checkpoint": "base", "mode": "pretrained" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "NVIDIA Nemotron 3 Ultra technical report, Base Table 2: Kimi K2 Base: global_mmlu_lite=85.63. Base checkpoint score mapped to the canonical post-trained model row as checkpoint=base." }, { "model_id": "kimi-k2", "benchmark_id": "mgsm", "score": 85.2, "reference_url": "https://research.nvidia.com/labs/nemotron/files/NVIDIA-Nemotron-3-Ultra-Technical-Report.pdf", "reported_setting": { "version_metric": "8-shot native CoT average", "tools": "none", "harness": "NeMo Evaluator", "sampling": "pass@1", "checkpoint": "base", "mode": "pretrained" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "NVIDIA Nemotron 3 Ultra technical report, Base Table 2: Kimi K2 Base: mgsm=85.2. Base checkpoint score mapped to the canonical post-trained model row as checkpoint=base." }, { "model_id": "kimi-k2", "benchmark_id": "ruler_64k", "score": 93.79, "reference_url": "https://research.nvidia.com/labs/nemotron/files/NVIDIA-Nemotron-3-Ultra-Technical-Report.pdf", "reported_setting": { "version_metric": "0-shot", "tools": "none", "harness": "NeMo Skills / RULER", "sampling": "pass@1", "checkpoint": "base", "mode": "pretrained" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "NVIDIA Nemotron 3 Ultra technical report, Base Table 2: Kimi K2 Base: ruler_64k=93.79. Base checkpoint score mapped to the canonical post-trained model row as checkpoint=base." }, { "model_id": "kimi-k2", "benchmark_id": "ruler_128k", "score": 88.61, "reference_url": "https://research.nvidia.com/labs/nemotron/files/NVIDIA-Nemotron-3-Ultra-Technical-Report.pdf", "reported_setting": { "version_metric": "0-shot", "tools": "none", "harness": "NeMo Skills / RULER", "sampling": "pass@1", "checkpoint": "base", "mode": "pretrained" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "NVIDIA Nemotron 3 Ultra technical report, Base Table 2: Kimi K2 Base: ruler_128k=88.61. Base checkpoint score mapped to the canonical post-trained model row as checkpoint=base." }, { "model_id": "glm-4.5", "benchmark_id": "mmlu", "score": 86.5, "reference_url": "https://research.nvidia.com/labs/nemotron/files/NVIDIA-Nemotron-3-Ultra-Technical-Report.pdf", "reported_setting": { "version_metric": "5-shot accuracy", "tools": "none", "harness": "NeMo Evaluator", "sampling": "pass@1", "checkpoint": "base", "mode": "pretrained" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "NVIDIA Nemotron 3 Ultra technical report, Base Table 2: GLM-4.5 Base: mmlu=86.5. Base checkpoint score mapped to the canonical post-trained model row as checkpoint=base." }, { "model_id": "glm-4.5", "benchmark_id": "mmlu_pro", "score": 65.78, "reference_url": "https://research.nvidia.com/labs/nemotron/files/NVIDIA-Nemotron-3-Ultra-Technical-Report.pdf", "reported_setting": { "version_metric": "5-shot CoT exact match", "tools": "none", "harness": "NeMo Evaluator", "sampling": "pass@1", "checkpoint": "base", "mode": "pretrained" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "NVIDIA Nemotron 3 Ultra technical report, Base Table 2: GLM-4.5 Base: mmlu_pro=65.78. Base checkpoint score mapped to the canonical post-trained model row as checkpoint=base." }, { "model_id": "glm-4.5", "benchmark_id": "agieval_en", "score": 70.06, "reference_url": "https://research.nvidia.com/labs/nemotron/files/NVIDIA-Nemotron-3-Ultra-Technical-Report.pdf", "reported_setting": { "version_metric": "3/5-shot CoT exact match", "tools": "none", "harness": "NeMo Evaluator", "sampling": "pass@1", "checkpoint": "base", "mode": "pretrained" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "NVIDIA Nemotron 3 Ultra technical report, Base Table 2: GLM-4.5 Base: agieval_en=70.06. Base checkpoint score mapped to the canonical post-trained model row as checkpoint=base." }, { "model_id": "glm-4.5", "benchmark_id": "gpqa_diamond", "score": 34.85, "reference_url": "https://research.nvidia.com/labs/nemotron/files/NVIDIA-Nemotron-3-Ultra-Technical-Report.pdf", "reported_setting": { "version_metric": "5-shot CoT exact match", "tools": "none", "harness": "NeMo Evaluator", "sampling": "pass@1", "checkpoint": "base", "mode": "pretrained" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "NVIDIA Nemotron 3 Ultra technical report, Base Table 2: GLM-4.5 Base: gpqa_diamond=34.85. Base checkpoint score mapped to the canonical post-trained model row as checkpoint=base." }, { "model_id": "glm-4.5", "benchmark_id": "gsm8k", "score": 85.37, "reference_url": "https://research.nvidia.com/labs/nemotron/files/NVIDIA-Nemotron-3-Ultra-Technical-Report.pdf", "reported_setting": { "version_metric": "8-shot CoT exact match", "tools": "none", "harness": "NeMo Evaluator", "sampling": "pass@1", "checkpoint": "base", "mode": "pretrained" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "NVIDIA Nemotron 3 Ultra technical report, Base Table 2: GLM-4.5 Base: gsm8k=85.37. Base checkpoint score mapped to the canonical post-trained model row as checkpoint=base." }, { "model_id": "glm-4.5", "benchmark_id": "math_test", "score": 57.58, "reference_url": "https://research.nvidia.com/labs/nemotron/files/NVIDIA-Nemotron-3-Ultra-Technical-Report.pdf", "reported_setting": { "version_metric": "Minerva 4-shot exact match", "tools": "none", "harness": "NeMo Evaluator", "sampling": "pass@1", "checkpoint": "base", "mode": "pretrained" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "NVIDIA Nemotron 3 Ultra technical report, Base Table 2: GLM-4.5 Base: math_test=57.58. Base checkpoint score mapped to the canonical post-trained model row as checkpoint=base." }, { "model_id": "glm-4.5", "benchmark_id": "humaneval", "score": 78.16, "reference_url": "https://research.nvidia.com/labs/nemotron/files/NVIDIA-Nemotron-3-Ultra-Technical-Report.pdf", "reported_setting": { "version_metric": "EvalPlus sanitized sampled pass@1 n=32", "tools": "none", "harness": "NeMo Evaluator", "sampling": "164 problems x 32 samples", "checkpoint": "base", "mode": "pretrained" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "NVIDIA Nemotron 3 Ultra technical report, Base Table 2: GLM-4.5 Base: humaneval=78.16. Base checkpoint score mapped to the canonical post-trained model row as checkpoint=base." }, { "model_id": "glm-4.5", "benchmark_id": "mbpp_sanitized", "score": 76.69, "reference_url": "https://research.nvidia.com/labs/nemotron/files/NVIDIA-Nemotron-3-Ultra-Technical-Report.pdf", "reported_setting": { "version_metric": "3-shot sampled pass@1 n=32", "tools": "none", "harness": "NeMo Evaluator", "sampling": "427 problems x 32 samples", "checkpoint": "base", "mode": "pretrained" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "NVIDIA Nemotron 3 Ultra technical report, Base Table 2: GLM-4.5 Base: mbpp_sanitized=76.69. Base checkpoint score mapped to the canonical post-trained model row as checkpoint=base." }, { "model_id": "glm-4.5", "benchmark_id": "arc_challenge", "score": 96.59, "reference_url": "https://research.nvidia.com/labs/nemotron/files/NVIDIA-Nemotron-3-Ultra-Technical-Report.pdf", "reported_setting": { "version_metric": "25-shot normalized accuracy", "tools": "none", "harness": "NeMo Evaluator", "sampling": "pass@1", "checkpoint": "base", "mode": "pretrained" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "NVIDIA Nemotron 3 Ultra technical report, Base Table 2: GLM-4.5 Base: arc_challenge=96.59. Base checkpoint score mapped to the canonical post-trained model row as checkpoint=base." }, { "model_id": "glm-4.5", "benchmark_id": "hellaswag", "score": 90.17, "reference_url": "https://research.nvidia.com/labs/nemotron/files/NVIDIA-Nemotron-3-Ultra-Technical-Report.pdf", "reported_setting": { "version_metric": "10-shot normalized accuracy", "tools": "none", "harness": "NeMo Evaluator", "sampling": "pass@1", "checkpoint": "base", "mode": "pretrained" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "NVIDIA Nemotron 3 Ultra technical report, Base Table 2: GLM-4.5 Base: hellaswag=90.17. Base checkpoint score mapped to the canonical post-trained model row as checkpoint=base." }, { "model_id": "glm-4.5", "benchmark_id": "openbookqa", "score": 49.6, "reference_url": "https://research.nvidia.com/labs/nemotron/files/NVIDIA-Nemotron-3-Ultra-Technical-Report.pdf", "reported_setting": { "version_metric": "0-shot normalized accuracy", "tools": "none", "harness": "NeMo Evaluator", "sampling": "pass@1", "checkpoint": "base", "mode": "pretrained" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "NVIDIA Nemotron 3 Ultra technical report, Base Table 2: GLM-4.5 Base: openbookqa=49.6. Base checkpoint score mapped to the canonical post-trained model row as checkpoint=base." }, { "model_id": "glm-4.5", "benchmark_id": "piqa", "score": 85.09, "reference_url": "https://research.nvidia.com/labs/nemotron/files/NVIDIA-Nemotron-3-Ultra-Technical-Report.pdf", "reported_setting": { "version_metric": "0-shot normalized accuracy", "tools": "none", "harness": "NeMo Evaluator", "sampling": "pass@1", "checkpoint": "base", "mode": "pretrained" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "NVIDIA Nemotron 3 Ultra technical report, Base Table 2: GLM-4.5 Base: piqa=85.09. Base checkpoint score mapped to the canonical post-trained model row as checkpoint=base." }, { "model_id": "glm-4.5", "benchmark_id": "winogrande", "score": 85.24, "reference_url": "https://research.nvidia.com/labs/nemotron/files/NVIDIA-Nemotron-3-Ultra-Technical-Report.pdf", "reported_setting": { "version_metric": "5-shot accuracy", "tools": "none", "harness": "NeMo Evaluator", "sampling": "pass@1", "checkpoint": "base", "mode": "pretrained" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "NVIDIA Nemotron 3 Ultra technical report, Base Table 2: GLM-4.5 Base: winogrande=85.24. Base checkpoint score mapped to the canonical post-trained model row as checkpoint=base." }, { "model_id": "glm-4.5", "benchmark_id": "race", "score": 92.15, "reference_url": "https://research.nvidia.com/labs/nemotron/files/NVIDIA-Nemotron-3-Ultra-Technical-Report.pdf", "reported_setting": { "version_metric": "0-shot accuracy", "tools": "none", "harness": "NeMo Evaluator", "sampling": "pass@1", "checkpoint": "base", "mode": "pretrained" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "NVIDIA Nemotron 3 Ultra technical report, Base Table 2: GLM-4.5 Base: race=92.15. Base checkpoint score mapped to the canonical post-trained model row as checkpoint=base." }, { "model_id": "glm-4.5", "benchmark_id": "global_mmlu_lite", "score": 85.81, "reference_url": "https://research.nvidia.com/labs/nemotron/files/NVIDIA-Nemotron-3-Ultra-Technical-Report.pdf", "reported_setting": { "version_metric": "5-shot average", "tools": "none", "harness": "NeMo Evaluator", "sampling": "pass@1", "checkpoint": "base", "mode": "pretrained" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "NVIDIA Nemotron 3 Ultra technical report, Base Table 2: GLM-4.5 Base: global_mmlu_lite=85.81. Base checkpoint score mapped to the canonical post-trained model row as checkpoint=base." }, { "model_id": "glm-4.5", "benchmark_id": "mgsm", "score": 81.27, "reference_url": "https://research.nvidia.com/labs/nemotron/files/NVIDIA-Nemotron-3-Ultra-Technical-Report.pdf", "reported_setting": { "version_metric": "8-shot native CoT average", "tools": "none", "harness": "NeMo Evaluator", "sampling": "pass@1", "checkpoint": "base", "mode": "pretrained" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "NVIDIA Nemotron 3 Ultra technical report, Base Table 2: GLM-4.5 Base: mgsm=81.27. Base checkpoint score mapped to the canonical post-trained model row as checkpoint=base." }, { "model_id": "glm-4.5", "benchmark_id": "ruler_64k", "score": 16.12, "reference_url": "https://research.nvidia.com/labs/nemotron/files/NVIDIA-Nemotron-3-Ultra-Technical-Report.pdf", "reported_setting": { "version_metric": "0-shot", "tools": "none", "harness": "NeMo Skills / RULER", "sampling": "pass@1", "checkpoint": "base", "mode": "pretrained" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "NVIDIA Nemotron 3 Ultra technical report, Base Table 2: GLM-4.5 Base: ruler_64k=16.12. Base checkpoint score mapped to the canonical post-trained model row as checkpoint=base." }, { "model_id": "glm-4.5", "benchmark_id": "ruler_128k", "score": 0.0, "reference_url": "https://research.nvidia.com/labs/nemotron/files/NVIDIA-Nemotron-3-Ultra-Technical-Report.pdf", "reported_setting": { "version_metric": "0-shot", "tools": "none", "harness": "NeMo Skills / RULER", "sampling": "pass@1", "checkpoint": "base", "mode": "pretrained" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "NVIDIA Nemotron 3 Ultra technical report, Base Table 2: GLM-4.5 Base: ruler_128k=0. Base checkpoint score mapped to the canonical post-trained model row as checkpoint=base." }, { "model_id": "nemotron-3-ultra-550b-a55b", "benchmark_id": "imo_proofbench_advanced", "score": 82.3, "reference_url": "https://research.nvidia.com/labs/nemotron/files/NVIDIA-Nemotron-3-Ultra-Technical-Report.pdf", "reported_setting": { "version_metric": "final five-round percentage", "tools": "search and proof-refinement pipeline", "harness": "NVIDIA test-time-scaling math pipeline", "sampling": "five refinement rounds", "variant": "R5 final" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "NVIDIA Nemotron 3 Ultra technical report, Final Table 11: Nemotron 3 Ultra: imo_proofbench_advanced=82.3, variant=R5 final.", "candidates": [ { "score": 36.7, "reference_url": "https://research.nvidia.com/labs/nemotron/files/NVIDIA-Nemotron-3-Ultra-Technical-Report.pdf", "source_type": "tech_report", "reported_setting": { "version_metric": "cumulative percentage by refinement round", "tools": "search and proof-refinement pipeline", "harness": "NVIDIA test-time-scaling math pipeline", "sampling": "cumulative by round", "variant": "R1 generate", "test_time_scaling_round": "R1 generate" }, "notes": "NVIDIA Nemotron 3 Ultra technical report, Figure 13: Nemotron 3 Ultra: imo_proofbench_advanced=36.7, variant=R1 generate." }, { "score": 59.5, "reference_url": "https://research.nvidia.com/labs/nemotron/files/NVIDIA-Nemotron-3-Ultra-Technical-Report.pdf", "source_type": "tech_report", "reported_setting": { "version_metric": "cumulative percentage by refinement round", "tools": "search and proof-refinement pipeline", "harness": "NVIDIA test-time-scaling math pipeline", "sampling": "cumulative by round", "variant": "R2 verify/refine", "test_time_scaling_round": "R2 verify/refine" }, "notes": "NVIDIA Nemotron 3 Ultra technical report, Figure 13: Nemotron 3 Ultra: imo_proofbench_advanced=59.5, variant=R2 verify/refine." }, { "score": 72.9, "reference_url": "https://research.nvidia.com/labs/nemotron/files/NVIDIA-Nemotron-3-Ultra-Technical-Report.pdf", "source_type": "tech_report", "reported_setting": { "version_metric": "cumulative percentage by refinement round", "tools": "search and proof-refinement pipeline", "harness": "NVIDIA test-time-scaling math pipeline", "sampling": "cumulative by round", "variant": "R3 verify/refine", "test_time_scaling_round": "R3 verify/refine" }, "notes": "NVIDIA Nemotron 3 Ultra technical report, Figure 13: Nemotron 3 Ultra: imo_proofbench_advanced=72.9, variant=R3 verify/refine." } ] }, { "model_id": "nemotron-3-ultra-550b-a55b", "benchmark_id": "imo_2025", "score": 83.3, "reference_url": "https://research.nvidia.com/labs/nemotron/files/NVIDIA-Nemotron-3-Ultra-Technical-Report.pdf", "reported_setting": { "version_metric": "final five-round percentage", "tools": "search and proof-refinement pipeline", "harness": "NVIDIA test-time-scaling math pipeline", "sampling": "five refinement rounds", "variant": "R5 final" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "NVIDIA Nemotron 3 Ultra technical report, Final Table 11: Nemotron 3 Ultra: imo_2025=83.3, variant=R5 final." }, { "model_id": "nemotron-3-ultra-550b-a55b", "benchmark_id": "putnam_2025", "score": 96.7, "reference_url": "https://research.nvidia.com/labs/nemotron/files/NVIDIA-Nemotron-3-Ultra-Technical-Report.pdf", "reported_setting": { "version_metric": "final five-round percentage", "tools": "search and proof-refinement pipeline", "harness": "NVIDIA test-time-scaling math pipeline", "sampling": "five refinement rounds", "variant": "R5 final" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "NVIDIA Nemotron 3 Ultra technical report, Final Table 11: Nemotron 3 Ultra: putnam_2025=96.7, variant=R5 final." }, { "model_id": "nemotron-3-ultra-550b-a55b", "benchmark_id": "usamo_2026", "score": 97.6, "reference_url": "https://research.nvidia.com/labs/nemotron/files/NVIDIA-Nemotron-3-Ultra-Technical-Report.pdf", "reported_setting": { "version_metric": "final five-round percentage", "tools": "search and proof-refinement pipeline", "harness": "NVIDIA test-time-scaling math pipeline", "sampling": "five refinement rounds", "variant": "R5 final" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "NVIDIA Nemotron 3 Ultra technical report, Final Table 11: Nemotron 3 Ultra: usamo_2026=97.6, variant=R5 final." }, { "model_id": "gpt-5.6-terra", "benchmark_id": "agents_last_exam", "score": 50.4, "reference_url": "https://openai.com/index/gpt-5-6/", "reported_setting": { "effort": "max", "eval_variant": null, "code_mode": "true" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-5.6 release table (table 0): GPT‑5.6 Terra; Agents' Last Exam=50.4. Exact effort/variant resolved from the rendered chart point.", "candidates": [ { "score": 40.3, "reference_url": "https://openai.com/index/gpt-5-6/", "source_type": "official_blog", "reported_setting": { "effort": "low", "eval_variant": null, "code_mode": "true" }, "notes": "OpenAI GPT-5.6 release figure (Agents' Last Exam): GPT-5.6 Terra; Agents' Last Exam=40.3. Figure 0: Long-horizon agentic workflows across professional domains.. Cost, latency, and output-token coordinates are retained as review metadata, not score-matrix metrics." }, { "score": 42.6, "reference_url": "https://openai.com/index/gpt-5-6/", "source_type": "official_blog", "reported_setting": { "effort": "medium", "eval_variant": null, "code_mode": "true" }, "notes": "OpenAI GPT-5.6 release figure (Agents' Last Exam): GPT-5.6 Terra; Agents' Last Exam=42.6. Figure 0: Long-horizon agentic workflows across professional domains.. Cost, latency, and output-token coordinates are retained as review metadata, not score-matrix metrics." }, { "score": 46.3, "reference_url": "https://openai.com/index/gpt-5-6/", "source_type": "official_blog", "reported_setting": { "effort": "high", "eval_variant": null, "code_mode": "true" }, "notes": "OpenAI GPT-5.6 release figure (Agents' Last Exam): GPT-5.6 Terra; Agents' Last Exam=46.3. Figure 0: Long-horizon agentic workflows across professional domains.. Cost, latency, and output-token coordinates are retained as review metadata, not score-matrix metrics." }, { "score": 48.5, "reference_url": "https://openai.com/index/gpt-5-6/", "source_type": "official_blog", "reported_setting": { "effort": "xhigh", "eval_variant": null, "code_mode": "true" }, "notes": "OpenAI GPT-5.6 release figure (Agents' Last Exam): GPT-5.6 Terra; Agents' Last Exam=48.5. Figure 0: Long-horizon agentic workflows across professional domains.. Cost, latency, and output-token coordinates are retained as review metadata, not score-matrix metrics." } ] }, { "model_id": "gpt-5.6-luna", "benchmark_id": "agents_last_exam", "score": 50.3, "reference_url": "https://openai.com/index/gpt-5-6/", "reported_setting": { "effort": "max", "eval_variant": null, "code_mode": "true" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-5.6 release table (table 0): GPT‑5.6 Luna; Agents' Last Exam=50.3. Exact effort/variant resolved from the rendered chart point.", "candidates": [ { "score": 30.7, "reference_url": "https://openai.com/index/gpt-5-6/", "source_type": "official_blog", "reported_setting": { "effort": "low", "eval_variant": null, "code_mode": "true" }, "notes": "OpenAI GPT-5.6 release figure (Agents' Last Exam): GPT-5.6 Luna; Agents' Last Exam=30.7. Figure 0: Long-horizon agentic workflows across professional domains.. Cost, latency, and output-token coordinates are retained as review metadata, not score-matrix metrics." }, { "score": 36.1, "reference_url": "https://openai.com/index/gpt-5-6/", "source_type": "official_blog", "reported_setting": { "effort": "medium", "eval_variant": null, "code_mode": "true" }, "notes": "OpenAI GPT-5.6 release figure (Agents' Last Exam): GPT-5.6 Luna; Agents' Last Exam=36.1. Figure 0: Long-horizon agentic workflows across professional domains.. Cost, latency, and output-token coordinates are retained as review metadata, not score-matrix metrics." }, { "score": 45.4, "reference_url": "https://openai.com/index/gpt-5-6/", "source_type": "official_blog", "reported_setting": { "effort": "high", "eval_variant": null, "code_mode": "true" }, "notes": "OpenAI GPT-5.6 release figure (Agents' Last Exam): GPT-5.6 Luna; Agents' Last Exam=45.4. Figure 0: Long-horizon agentic workflows across professional domains.. Cost, latency, and output-token coordinates are retained as review metadata, not score-matrix metrics." }, { "score": 48.7, "reference_url": "https://openai.com/index/gpt-5-6/", "source_type": "official_blog", "reported_setting": { "effort": "xhigh", "eval_variant": null, "code_mode": "true" }, "notes": "OpenAI GPT-5.6 release figure (Agents' Last Exam): GPT-5.6 Luna; Agents' Last Exam=48.7. Figure 0: Long-horizon agentic workflows across professional domains.. Cost, latency, and output-token coordinates are retained as review metadata, not score-matrix metrics." } ] }, { "model_id": "gemini-3.1-pro", "benchmark_id": "agents_last_exam", "score": 32.1, "reference_url": "https://openai.com/index/gpt-5-6/", "reported_setting": { "effort": "xhigh", "eval_variant": null, "code_mode": "true" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-5.6 release table (table 0): Gemini 3.1 Pro Preview; Agents' Last Exam=32.1. Exact effort/variant resolved from the rendered chart point.", "candidates": [ { "score": 15.8, "reference_url": "https://seed.bytedance.com/en/blog/seed2-1-officially-released-advancing-ai-productivity", "source_type": "official_blog", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "computer-use environment", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official", "prompt_style": "official default", "temperature": "unknown" }, "notes": "Release blog EN: General-agent chart (https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/user-upload/4xfa4mqq4u89c.png). Turbo is a source dash, not zero." } ] }, { "model_id": "gpt-5.6-terra", "benchmark_id": "gdpval_aa_elo", "score": 1593.0, "reference_url": "https://openai.com/index/gpt-5-6/", "reported_setting": { "effort": "max", "eval_variant": null, "code_mode": "false" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-5.6 release table (table 0): GPT‑5.6 Terra; GDPval-AA v2=1593.0. OpenAI row explicitly identifies GDPval-AA v2. Exact effort/variant resolved from the rendered chart point.", "candidates": [ { "score": 1239.0, "reference_url": "https://openai.com/index/gpt-5-6/", "source_type": "official_blog", "reported_setting": { "effort": "none", "eval_variant": null, "code_mode": "false" }, "notes": "OpenAI GPT-5.6 release figure (GDPval-AA v2): GPT-5.6 Terra; GDPval-AA v2=1239.0. Figure 15: GDPval-AA v2. Cost, latency, and output-token coordinates are retained as review metadata, not score-matrix metrics." }, { "score": 1249.0, "reference_url": "https://openai.com/index/gpt-5-6/", "source_type": "official_blog", "reported_setting": { "effort": "low", "eval_variant": null, "code_mode": "false" }, "notes": "OpenAI GPT-5.6 release figure (GDPval-AA v2): GPT-5.6 Terra; GDPval-AA v2=1249.0. Figure 15: GDPval-AA v2. Cost, latency, and output-token coordinates are retained as review metadata, not score-matrix metrics." }, { "score": 1404.0, "reference_url": "https://openai.com/index/gpt-5-6/", "source_type": "official_blog", "reported_setting": { "effort": "medium", "eval_variant": null, "code_mode": "false" }, "notes": "OpenAI GPT-5.6 release figure (GDPval-AA v2): GPT-5.6 Terra; GDPval-AA v2=1404.0. Figure 15: GDPval-AA v2. Cost, latency, and output-token coordinates are retained as review metadata, not score-matrix metrics." }, { "score": 1513.0, "reference_url": "https://openai.com/index/gpt-5-6/", "source_type": "official_blog", "reported_setting": { "effort": "high", "eval_variant": null, "code_mode": "false" }, "notes": "OpenAI GPT-5.6 release figure (GDPval-AA v2): GPT-5.6 Terra; GDPval-AA v2=1513.0. Figure 15: GDPval-AA v2. Cost, latency, and output-token coordinates are retained as review metadata, not score-matrix metrics." }, { "score": 1572.0, "reference_url": "https://openai.com/index/gpt-5-6/", "source_type": "official_blog", "reported_setting": { "effort": "xhigh", "eval_variant": null, "code_mode": "false" }, "notes": "OpenAI GPT-5.6 release figure (GDPval-AA v2): GPT-5.6 Terra; GDPval-AA v2=1572.0. Figure 15: GDPval-AA v2. Cost, latency, and output-token coordinates are retained as review metadata, not score-matrix metrics." } ] }, { "model_id": "gpt-5.6-luna", "benchmark_id": "gdpval_aa_elo", "score": 1591.8, "reference_url": "https://openai.com/index/gpt-5-6/", "reported_setting": { "effort": "max", "eval_variant": null, "code_mode": "false" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-5.6 release table (table 0): GPT‑5.6 Luna; GDPval-AA v2=1591.8. OpenAI row explicitly identifies GDPval-AA v2. Exact effort/variant resolved from the rendered chart point.", "candidates": [ { "score": 1068.0, "reference_url": "https://openai.com/index/gpt-5-6/", "source_type": "official_blog", "reported_setting": { "effort": "none", "eval_variant": null, "code_mode": "false" }, "notes": "OpenAI GPT-5.6 release figure (GDPval-AA v2): GPT-5.6 Luna; GDPval-AA v2=1068.0. Figure 15: GDPval-AA v2. Cost, latency, and output-token coordinates are retained as review metadata, not score-matrix metrics." }, { "score": 1146.0, "reference_url": "https://openai.com/index/gpt-5-6/", "source_type": "official_blog", "reported_setting": { "effort": "low", "eval_variant": null, "code_mode": "false" }, "notes": "OpenAI GPT-5.6 release figure (GDPval-AA v2): GPT-5.6 Luna; GDPval-AA v2=1146.0. Figure 15: GDPval-AA v2. Cost, latency, and output-token coordinates are retained as review metadata, not score-matrix metrics." }, { "score": 1273.0, "reference_url": "https://openai.com/index/gpt-5-6/", "source_type": "official_blog", "reported_setting": { "effort": "medium", "eval_variant": null, "code_mode": "false" }, "notes": "OpenAI GPT-5.6 release figure (GDPval-AA v2): GPT-5.6 Luna; GDPval-AA v2=1273.0. Figure 15: GDPval-AA v2. Cost, latency, and output-token coordinates are retained as review metadata, not score-matrix metrics." }, { "score": 1472.0, "reference_url": "https://openai.com/index/gpt-5-6/", "source_type": "official_blog", "reported_setting": { "effort": "high", "eval_variant": null, "code_mode": "false" }, "notes": "OpenAI GPT-5.6 release figure (GDPval-AA v2): GPT-5.6 Luna; GDPval-AA v2=1472.0. Figure 15: GDPval-AA v2. Cost, latency, and output-token coordinates are retained as review metadata, not score-matrix metrics." }, { "score": 1539.0, "reference_url": "https://openai.com/index/gpt-5-6/", "source_type": "official_blog", "reported_setting": { "effort": "xhigh", "eval_variant": null, "code_mode": "false" }, "notes": "OpenAI GPT-5.6 release figure (GDPval-AA v2): GPT-5.6 Luna; GDPval-AA v2=1539.0. Figure 15: GDPval-AA v2. Cost, latency, and output-token coordinates are retained as review metadata, not score-matrix metrics." } ] }, { "model_id": "gemini-3.5-flash", "benchmark_id": "gdpval_aa_elo", "score": 1348.8, "reference_url": "https://openai.com/index/gpt-5-6/", "reported_setting": { "effort": "high", "eval_variant": null, "code_mode": "false" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-5.6 release table (table 0): Gemini 3.5 Flash; GDPval-AA v2=1348.8. OpenAI row explicitly identifies GDPval-AA v2. Exact effort/variant resolved from the rendered chart point.", "candidates": [ { "score": 1357.0, "reference_url": "https://www.anthropic.com/claude-sonnet-5-system-card", "source_type": "model_card", "reported_setting": { "effort": "source does not state", "tools": "none", "sampling": "avg 5 trials unless source states otherwise", "harness": "table8" }, "notes": "Anthropic Claude Sonnet 5 System Card Table 8.1.A, page 115: Gemini 3.5 Flash; GDPval-AA v2 [summary]=1357. Source setting: effort=source does not state; tools=none; sampling=avg 5 trials unless source states otherwise; harness=table8." }, { "score": 1656.0, "reference_url": "https://static.stepfun.com/blog/step-3.7-flash/", "source_type": "official_blog", "reported_setting": { "quantization_config": "common release evaluation; artifact precision not established", "configuration": "base Step evaluation configuration", "mode": "source does not state", "effort": "source does not state", "tools_search": "official Artificial Analysis Stirrup harness", "shots": "source does not state", "trials_samples": "source does not state", "aggregation_pass_k": "source-reported score", "context": "source does not state", "output_cap": "source does not state", "temperature": "source does not state", "top_p": "source does not state", "judge": "pairwise Elo evaluation anchored to human experts", "harness_agent": "official Artificial Analysis Stirrup leaderboard", "dataset_version_split": "GDPval-AA historical 220-task Stirrup leaderboard quoted 2026-05-29", "multimodal_input": false }, "notes": "Official StepFun launch comparison table." } ] }, { "model_id": "gpt-5.6-sol", "benchmark_id": "management_consulting_tasks_internal", "score": 43.2, "reference_url": "https://openai.com/index/gpt-5-6/", "reported_setting": { "mode": "thinking", "effort": "source does not state", "tools": "none", "sampling": "source does not state", "judge": "benchmark-specified", "harness": "official release evaluation", "prompt_style": "source does not state", "temperature": "source does not state", "context": "source does not state" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-5.6 release table (table 0): GPT‑5.6 Sol; Management Consulting Tasks (Internal)=43.2. Named internal evaluation; task count not disclosed." }, { "model_id": "gpt-5.6-terra", "benchmark_id": "management_consulting_tasks_internal", "score": 37.2, "reference_url": "https://openai.com/index/gpt-5-6/", "reported_setting": { "mode": "thinking", "effort": "source does not state", "tools": "none", "sampling": "source does not state", "judge": "benchmark-specified", "harness": "official release evaluation", "prompt_style": "source does not state", "temperature": "source does not state", "context": "source does not state" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-5.6 release table (table 0): GPT‑5.6 Terra; Management Consulting Tasks (Internal)=37.2. Named internal evaluation; task count not disclosed." }, { "model_id": "gpt-5.6-luna", "benchmark_id": "management_consulting_tasks_internal", "score": 35.4, "reference_url": "https://openai.com/index/gpt-5-6/", "reported_setting": { "mode": "thinking", "effort": "source does not state", "tools": "none", "sampling": "source does not state", "judge": "benchmark-specified", "harness": "official release evaluation", "prompt_style": "source does not state", "temperature": "source does not state", "context": "source does not state" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-5.6 release table (table 0): GPT‑5.6 Luna; Management Consulting Tasks (Internal)=35.4. Named internal evaluation; task count not disclosed." }, { "model_id": "gpt-5.5", "benchmark_id": "management_consulting_tasks_internal", "score": 31.3, "reference_url": "https://openai.com/index/gpt-5-6/", "reported_setting": { "mode": "thinking", "effort": "source does not state", "tools": "none", "sampling": "source does not state", "judge": "benchmark-specified", "harness": "official release evaluation", "prompt_style": "source does not state", "temperature": "source does not state", "context": "source does not state" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-5.6 release table (table 0): GPT‑5.5; Management Consulting Tasks (Internal)=31.3. Named internal evaluation; task count not disclosed." }, { "model_id": "claude-fable-5", "benchmark_id": "management_consulting_tasks_internal", "score": 35.5, "reference_url": "https://openai.com/index/gpt-5-6/", "reported_setting": { "mode": "thinking", "effort": "source does not state", "tools": "none", "sampling": "source does not state", "judge": "benchmark-specified", "harness": "official release evaluation", "prompt_style": "source does not state", "temperature": "source does not state", "context": "source does not state" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-5.6 release table (table 0): Claude Fable 5; Management Consulting Tasks (Internal)=35.5. Named internal evaluation; task count not disclosed." }, { "model_id": "claude-opus-4.8", "benchmark_id": "management_consulting_tasks_internal", "score": 31.6, "reference_url": "https://openai.com/index/gpt-5-6/", "reported_setting": { "mode": "thinking", "effort": "source does not state", "tools": "none", "sampling": "source does not state", "judge": "benchmark-specified", "harness": "official release evaluation", "prompt_style": "source does not state", "temperature": "source does not state", "context": "source does not state" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-5.6 release table (table 0): Claude Opus 4.8; Management Consulting Tasks (Internal)=31.6. Named internal evaluation; task count not disclosed." }, { "model_id": "gemini-3.1-pro", "benchmark_id": "management_consulting_tasks_internal", "score": 13.2, "reference_url": "https://openai.com/index/gpt-5-6/", "reported_setting": { "mode": "thinking", "effort": "source does not state", "tools": "none", "sampling": "source does not state", "judge": "benchmark-specified", "harness": "official release evaluation", "prompt_style": "source does not state", "temperature": "source does not state", "context": "source does not state" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-5.6 release table (table 0): Gemini 3.1 Pro Preview; Management Consulting Tasks (Internal)=13.2. Named internal evaluation; task count not disclosed." }, { "model_id": "gpt-5.6-sol", "benchmark_id": "big_finance_bench", "score": 53.0, "reference_url": "https://openai.com/index/gpt-5-6/", "reported_setting": { "mode": "thinking", "effort": "source does not state", "tools": "none", "sampling": "source does not state", "judge": "benchmark-specified", "harness": "official release evaluation", "prompt_style": "source does not state", "temperature": "source does not state", "context": "source does not state" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-5.6 release table (table 0): GPT‑5.6 Sol; Big Finance Bench=53.0. Distinct from the existing Finance Bench row." }, { "model_id": "gpt-5.6-terra", "benchmark_id": "big_finance_bench", "score": 51.0, "reference_url": "https://openai.com/index/gpt-5-6/", "reported_setting": { "mode": "thinking", "effort": "source does not state", "tools": "none", "sampling": "source does not state", "judge": "benchmark-specified", "harness": "official release evaluation", "prompt_style": "source does not state", "temperature": "source does not state", "context": "source does not state" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-5.6 release table (table 0): GPT‑5.6 Terra; Big Finance Bench=51.0. Distinct from the existing Finance Bench row." }, { "model_id": "gpt-5.6-luna", "benchmark_id": "big_finance_bench", "score": 36.0, "reference_url": "https://openai.com/index/gpt-5-6/", "reported_setting": { "mode": "thinking", "effort": "source does not state", "tools": "none", "sampling": "source does not state", "judge": "benchmark-specified", "harness": "official release evaluation", "prompt_style": "source does not state", "temperature": "source does not state", "context": "source does not state" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-5.6 release table (table 0): GPT‑5.6 Luna; Big Finance Bench=36.0. Distinct from the existing Finance Bench row." }, { "model_id": "gpt-5.5", "benchmark_id": "big_finance_bench", "score": 49.0, "reference_url": "https://openai.com/index/gpt-5-6/", "reported_setting": { "mode": "thinking", "effort": "source does not state", "tools": "none", "sampling": "source does not state", "judge": "benchmark-specified", "harness": "official release evaluation", "prompt_style": "source does not state", "temperature": "source does not state", "context": "source does not state" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-5.6 release table (table 0): GPT‑5.5; Big Finance Bench=49.0. Distinct from the existing Finance Bench row." }, { "model_id": "claude-opus-4.8", "benchmark_id": "big_finance_bench", "score": 44.0, "reference_url": "https://openai.com/index/gpt-5-6/", "reported_setting": { "mode": "thinking", "effort": "source does not state", "tools": "none", "sampling": "source does not state", "judge": "benchmark-specified", "harness": "official release evaluation", "prompt_style": "source does not state", "temperature": "source does not state", "context": "source does not state" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-5.6 release table (table 0): Claude Opus 4.8; Big Finance Bench=44.0. Distinct from the existing Finance Bench row." }, { "model_id": "gpt-5.6-terra", "benchmark_id": "aa_intelligence_index_v4_1", "score": 55.0, "reference_url": "https://openai.com/index/gpt-5-6/", "reported_setting": { "effort": "max", "eval_variant": null, "code_mode": "false" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-5.6 release table (table 0): GPT‑5.6 Terra; Artificial Analysis Intelligence Index v4.1=55.0. Exact effort/variant resolved from the rendered chart point.", "candidates": [ { "score": 33.97, "reference_url": "https://openai.com/index/gpt-5-6/", "source_type": "official_blog", "reported_setting": { "effort": "none", "eval_variant": null, "code_mode": "false" }, "notes": "OpenAI GPT-5.6 release figure (Artificial Analysis Intelligence Index v4.1): GPT-5.6 Terra; Artificial Analysis Intelligence Index v4.1=33.97. Figure 0: Composite of nine independent evaluations spanning agentic work, coding, scientific reasoning, and general capabilities.. Cost, latency, and output-token coordinates are retained as review metadata, not score-matrix metrics." }, { "score": 40.47, "reference_url": "https://openai.com/index/gpt-5-6/", "source_type": "official_blog", "reported_setting": { "effort": "low", "eval_variant": null, "code_mode": "false" }, "notes": "OpenAI GPT-5.6 release figure (Artificial Analysis Intelligence Index v4.1): GPT-5.6 Terra; Artificial Analysis Intelligence Index v4.1=40.47. Figure 0: Composite of nine independent evaluations spanning agentic work, coding, scientific reasoning, and general capabilities.. Cost, latency, and output-token coordinates are retained as review metadata, not score-matrix metrics." }, { "score": 45.57, "reference_url": "https://openai.com/index/gpt-5-6/", "source_type": "official_blog", "reported_setting": { "effort": "medium", "eval_variant": null, "code_mode": "false" }, "notes": "OpenAI GPT-5.6 release figure (Artificial Analysis Intelligence Index v4.1): GPT-5.6 Terra; Artificial Analysis Intelligence Index v4.1=45.57. Figure 0: Composite of nine independent evaluations spanning agentic work, coding, scientific reasoning, and general capabilities.. Cost, latency, and output-token coordinates are retained as review metadata, not score-matrix metrics." }, { "score": 48.95, "reference_url": "https://openai.com/index/gpt-5-6/", "source_type": "official_blog", "reported_setting": { "effort": "high", "eval_variant": null, "code_mode": "false" }, "notes": "OpenAI GPT-5.6 release figure (Artificial Analysis Intelligence Index v4.1): GPT-5.6 Terra; Artificial Analysis Intelligence Index v4.1=48.95. Figure 0: Composite of nine independent evaluations spanning agentic work, coding, scientific reasoning, and general capabilities.. Cost, latency, and output-token coordinates are retained as review metadata, not score-matrix metrics." }, { "score": 51.6, "reference_url": "https://openai.com/index/gpt-5-6/", "source_type": "official_blog", "reported_setting": { "effort": "xhigh", "eval_variant": null, "code_mode": "false" }, "notes": "OpenAI GPT-5.6 release figure (Artificial Analysis Intelligence Index v4.1): GPT-5.6 Terra; Artificial Analysis Intelligence Index v4.1=51.6. Figure 0: Composite of nine independent evaluations spanning agentic work, coding, scientific reasoning, and general capabilities.. Cost, latency, and output-token coordinates are retained as review metadata, not score-matrix metrics." } ] }, { "model_id": "gpt-5.6-luna", "benchmark_id": "aa_intelligence_index_v4_1", "score": 51.2, "reference_url": "https://openai.com/index/gpt-5-6/", "reported_setting": { "effort": "max", "eval_variant": null, "code_mode": "false" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-5.6 release table (table 0): GPT‑5.6 Luna; Artificial Analysis Intelligence Index v4.1=51.2. Exact effort/variant resolved from the rendered chart point.", "candidates": [ { "score": 26.56, "reference_url": "https://openai.com/index/gpt-5-6/", "source_type": "official_blog", "reported_setting": { "effort": "none", "eval_variant": null, "code_mode": "false" }, "notes": "OpenAI GPT-5.6 release figure (Artificial Analysis Intelligence Index v4.1): GPT-5.6 Luna; Artificial Analysis Intelligence Index v4.1=26.56. Figure 0: Composite of nine independent evaluations spanning agentic work, coding, scientific reasoning, and general capabilities.. Cost, latency, and output-token coordinates are retained as review metadata, not score-matrix metrics." }, { "score": 33.26, "reference_url": "https://openai.com/index/gpt-5-6/", "source_type": "official_blog", "reported_setting": { "effort": "low", "eval_variant": null, "code_mode": "false" }, "notes": "OpenAI GPT-5.6 release figure (Artificial Analysis Intelligence Index v4.1): GPT-5.6 Luna; Artificial Analysis Intelligence Index v4.1=33.26. Figure 0: Composite of nine independent evaluations spanning agentic work, coding, scientific reasoning, and general capabilities.. Cost, latency, and output-token coordinates are retained as review metadata, not score-matrix metrics." }, { "score": 38.05, "reference_url": "https://openai.com/index/gpt-5-6/", "source_type": "official_blog", "reported_setting": { "effort": "medium", "eval_variant": null, "code_mode": "false" }, "notes": "OpenAI GPT-5.6 release figure (Artificial Analysis Intelligence Index v4.1): GPT-5.6 Luna; Artificial Analysis Intelligence Index v4.1=38.05. Figure 0: Composite of nine independent evaluations spanning agentic work, coding, scientific reasoning, and general capabilities.. Cost, latency, and output-token coordinates are retained as review metadata, not score-matrix metrics." }, { "score": 46.06, "reference_url": "https://openai.com/index/gpt-5-6/", "source_type": "official_blog", "reported_setting": { "effort": "high", "eval_variant": null, "code_mode": "false" }, "notes": "OpenAI GPT-5.6 release figure (Artificial Analysis Intelligence Index v4.1): GPT-5.6 Luna; Artificial Analysis Intelligence Index v4.1=46.06. Figure 0: Composite of nine independent evaluations spanning agentic work, coding, scientific reasoning, and general capabilities.. Cost, latency, and output-token coordinates are retained as review metadata, not score-matrix metrics." }, { "score": 49.07, "reference_url": "https://openai.com/index/gpt-5-6/", "source_type": "official_blog", "reported_setting": { "effort": "xhigh", "eval_variant": null, "code_mode": "false" }, "notes": "OpenAI GPT-5.6 release figure (Artificial Analysis Intelligence Index v4.1): GPT-5.6 Luna; Artificial Analysis Intelligence Index v4.1=49.07. Figure 0: Composite of nine independent evaluations spanning agentic work, coding, scientific reasoning, and general capabilities.. Cost, latency, and output-token coordinates are retained as review metadata, not score-matrix metrics." } ] }, { "model_id": "gemini-3.1-pro", "benchmark_id": "aa_intelligence_index_v4_1", "score": 46.5, "reference_url": "https://openai.com/index/gpt-5-6/", "reported_setting": { "effort": "n/a", "eval_variant": null, "code_mode": "false" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-5.6 release table (table 0): Gemini 3.1 Pro Preview; Artificial Analysis Intelligence Index v4.1=46.5. Exact effort/variant resolved from the rendered chart point." }, { "model_id": "gemini-3.5-flash", "benchmark_id": "aa_intelligence_index_v4_1", "score": 50.2, "reference_url": "https://openai.com/index/gpt-5-6/", "reported_setting": { "effort": "high", "eval_variant": null, "code_mode": "false" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-5.6 release table (table 0): Gemini 3.5 Flash; Artificial Analysis Intelligence Index v4.1=50.2. Exact effort/variant resolved from the rendered chart point." }, { "model_id": "gpt-5.6-sol", "benchmark_id": "aa_coding_agent_index_v1_1", "score": 80.0, "reference_url": "https://openai.com/index/gpt-5-6/", "reported_setting": { "effort": "max", "eval_variant": null, "code_mode": "false" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-5.6 release table (table 1): GPT‑5.6 Sol; Artificial Analysis Coding Agent Index v1.1=80.0. Exact effort/variant resolved from the rendered chart point.", "candidates": [ { "score": 58.4, "reference_url": "https://openai.com/index/gpt-5-6/", "source_type": "official_blog", "reported_setting": { "effort": "none", "eval_variant": null, "code_mode": "false" }, "notes": "OpenAI GPT-5.6 release figure (Artificial Analysis Coding Index): GPT-5.6 Sol; Artificial Analysis Coding Agent Index v1.1=58.4. Figure 1: Independent coding-agent index across implementation, terminal use, and real codebases.. Cost, latency, and output-token coordinates are retained as review metadata, not score-matrix metrics." }, { "score": 69.1, "reference_url": "https://openai.com/index/gpt-5-6/", "source_type": "official_blog", "reported_setting": { "effort": "low", "eval_variant": null, "code_mode": "false" }, "notes": "OpenAI GPT-5.6 release figure (Artificial Analysis Coding Index): GPT-5.6 Sol; Artificial Analysis Coding Agent Index v1.1=69.1. Figure 1: Independent coding-agent index across implementation, terminal use, and real codebases.. Cost, latency, and output-token coordinates are retained as review metadata, not score-matrix metrics." }, { "score": 74.6, "reference_url": "https://openai.com/index/gpt-5-6/", "source_type": "official_blog", "reported_setting": { "effort": "medium", "eval_variant": null, "code_mode": "false" }, "notes": "OpenAI GPT-5.6 release figure (Artificial Analysis Coding Index): GPT-5.6 Sol; Artificial Analysis Coding Agent Index v1.1=74.6. Figure 1: Independent coding-agent index across implementation, terminal use, and real codebases.. Cost, latency, and output-token coordinates are retained as review metadata, not score-matrix metrics." }, { "score": 77.1, "reference_url": "https://openai.com/index/gpt-5-6/", "source_type": "official_blog", "reported_setting": { "effort": "high", "eval_variant": null, "code_mode": "false" }, "notes": "OpenAI GPT-5.6 release figure (Artificial Analysis Coding Index): GPT-5.6 Sol; Artificial Analysis Coding Agent Index v1.1=77.1. Figure 1: Independent coding-agent index across implementation, terminal use, and real codebases.. Cost, latency, and output-token coordinates are retained as review metadata, not score-matrix metrics." }, { "score": 78.7, "reference_url": "https://openai.com/index/gpt-5-6/", "source_type": "official_blog", "reported_setting": { "effort": "xhigh", "eval_variant": null, "code_mode": "false" }, "notes": "OpenAI GPT-5.6 release figure (Artificial Analysis Coding Index): GPT-5.6 Sol; Artificial Analysis Coding Agent Index v1.1=78.7. Figure 1: Independent coding-agent index across implementation, terminal use, and real codebases.. Cost, latency, and output-token coordinates are retained as review metadata, not score-matrix metrics." } ] }, { "model_id": "gpt-5.6-terra", "benchmark_id": "aa_coding_agent_index_v1_1", "score": 77.4, "reference_url": "https://openai.com/index/gpt-5-6/", "reported_setting": { "effort": "max", "eval_variant": null, "code_mode": "false" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-5.6 release table (table 1): GPT‑5.6 Terra; Artificial Analysis Coding Agent Index v1.1=77.4. Exact effort/variant resolved from the rendered chart point.", "candidates": [ { "score": 40.3, "reference_url": "https://openai.com/index/gpt-5-6/", "source_type": "official_blog", "reported_setting": { "effort": "none", "eval_variant": null, "code_mode": "false" }, "notes": "OpenAI GPT-5.6 release figure (Artificial Analysis Coding Index): GPT-5.6 Terra; Artificial Analysis Coding Agent Index v1.1=40.3. Figure 1: Independent coding-agent index across implementation, terminal use, and real codebases.. Cost, latency, and output-token coordinates are retained as review metadata, not score-matrix metrics." }, { "score": 53.8, "reference_url": "https://openai.com/index/gpt-5-6/", "source_type": "official_blog", "reported_setting": { "effort": "low", "eval_variant": null, "code_mode": "false" }, "notes": "OpenAI GPT-5.6 release figure (Artificial Analysis Coding Index): GPT-5.6 Terra; Artificial Analysis Coding Agent Index v1.1=53.8. Figure 1: Independent coding-agent index across implementation, terminal use, and real codebases.. Cost, latency, and output-token coordinates are retained as review metadata, not score-matrix metrics." }, { "score": 64.2, "reference_url": "https://openai.com/index/gpt-5-6/", "source_type": "official_blog", "reported_setting": { "effort": "medium", "eval_variant": null, "code_mode": "false" }, "notes": "OpenAI GPT-5.6 release figure (Artificial Analysis Coding Index): GPT-5.6 Terra; Artificial Analysis Coding Agent Index v1.1=64.2. Figure 1: Independent coding-agent index across implementation, terminal use, and real codebases.. Cost, latency, and output-token coordinates are retained as review metadata, not score-matrix metrics." }, { "score": 71.8, "reference_url": "https://openai.com/index/gpt-5-6/", "source_type": "official_blog", "reported_setting": { "effort": "high", "eval_variant": null, "code_mode": "false" }, "notes": "OpenAI GPT-5.6 release figure (Artificial Analysis Coding Index): GPT-5.6 Terra; Artificial Analysis Coding Agent Index v1.1=71.8. Figure 1: Independent coding-agent index across implementation, terminal use, and real codebases.. Cost, latency, and output-token coordinates are retained as review metadata, not score-matrix metrics." }, { "score": 73.2, "reference_url": "https://openai.com/index/gpt-5-6/", "source_type": "official_blog", "reported_setting": { "effort": "xhigh", "eval_variant": null, "code_mode": "false" }, "notes": "OpenAI GPT-5.6 release figure (Artificial Analysis Coding Index): GPT-5.6 Terra; Artificial Analysis Coding Agent Index v1.1=73.2. Figure 1: Independent coding-agent index across implementation, terminal use, and real codebases.. Cost, latency, and output-token coordinates are retained as review metadata, not score-matrix metrics." } ] }, { "model_id": "gpt-5.6-luna", "benchmark_id": "aa_coding_agent_index_v1_1", "score": 74.6, "reference_url": "https://openai.com/index/gpt-5-6/", "reported_setting": { "effort": "max", "eval_variant": null, "code_mode": "false" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-5.6 release table (table 1): GPT‑5.6 Luna; Artificial Analysis Coding Agent Index v1.1=74.6. Exact effort/variant resolved from the rendered chart point.", "candidates": [ { "score": 37.3, "reference_url": "https://openai.com/index/gpt-5-6/", "source_type": "official_blog", "reported_setting": { "effort": "none", "eval_variant": null, "code_mode": "false" }, "notes": "OpenAI GPT-5.6 release figure (Artificial Analysis Coding Index): GPT-5.6 Luna; Artificial Analysis Coding Agent Index v1.1=37.3. Figure 1: Independent coding-agent index across implementation, terminal use, and real codebases.. Cost, latency, and output-token coordinates are retained as review metadata, not score-matrix metrics." }, { "score": 42.4, "reference_url": "https://openai.com/index/gpt-5-6/", "source_type": "official_blog", "reported_setting": { "effort": "low", "eval_variant": null, "code_mode": "false" }, "notes": "OpenAI GPT-5.6 release figure (Artificial Analysis Coding Index): GPT-5.6 Luna; Artificial Analysis Coding Agent Index v1.1=42.4. Figure 1: Independent coding-agent index across implementation, terminal use, and real codebases.. Cost, latency, and output-token coordinates are retained as review metadata, not score-matrix metrics." }, { "score": 58.7, "reference_url": "https://openai.com/index/gpt-5-6/", "source_type": "official_blog", "reported_setting": { "effort": "medium", "eval_variant": null, "code_mode": "false" }, "notes": "OpenAI GPT-5.6 release figure (Artificial Analysis Coding Index): GPT-5.6 Luna; Artificial Analysis Coding Agent Index v1.1=58.7. Figure 1: Independent coding-agent index across implementation, terminal use, and real codebases.. Cost, latency, and output-token coordinates are retained as review metadata, not score-matrix metrics." }, { "score": 67.9, "reference_url": "https://openai.com/index/gpt-5-6/", "source_type": "official_blog", "reported_setting": { "effort": "high", "eval_variant": null, "code_mode": "false" }, "notes": "OpenAI GPT-5.6 release figure (Artificial Analysis Coding Index): GPT-5.6 Luna; Artificial Analysis Coding Agent Index v1.1=67.9. Figure 1: Independent coding-agent index across implementation, terminal use, and real codebases.. Cost, latency, and output-token coordinates are retained as review metadata, not score-matrix metrics." }, { "score": 70.8, "reference_url": "https://openai.com/index/gpt-5-6/", "source_type": "official_blog", "reported_setting": { "effort": "xhigh", "eval_variant": null, "code_mode": "false" }, "notes": "OpenAI GPT-5.6 release figure (Artificial Analysis Coding Index): GPT-5.6 Luna; Artificial Analysis Coding Agent Index v1.1=70.8. Figure 1: Independent coding-agent index across implementation, terminal use, and real codebases.. Cost, latency, and output-token coordinates are retained as review metadata, not score-matrix metrics." } ] }, { "model_id": "gpt-5.5", "benchmark_id": "aa_coding_agent_index_v1_1", "score": 76.4, "reference_url": "https://openai.com/index/gpt-5-6/", "reported_setting": { "effort": "xhigh", "eval_variant": null, "code_mode": "false" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-5.6 release table (table 1): GPT‑5.5; Artificial Analysis Coding Agent Index v1.1=76.4. Exact effort/variant resolved from the rendered chart point.", "candidates": [ { "score": 45.5, "reference_url": "https://openai.com/index/gpt-5-6/", "source_type": "official_blog", "reported_setting": { "effort": "none", "eval_variant": null, "code_mode": "false" }, "notes": "OpenAI GPT-5.6 release figure (Artificial Analysis Coding Index): GPT-5.5; Artificial Analysis Coding Agent Index v1.1=45.5. Figure 1: Independent coding-agent index across implementation, terminal use, and real codebases.. Cost, latency, and output-token coordinates are retained as review metadata, not score-matrix metrics." }, { "score": 57.7, "reference_url": "https://openai.com/index/gpt-5-6/", "source_type": "official_blog", "reported_setting": { "effort": "low", "eval_variant": null, "code_mode": "false" }, "notes": "OpenAI GPT-5.6 release figure (Artificial Analysis Coding Index): GPT-5.5; Artificial Analysis Coding Agent Index v1.1=57.7. Figure 1: Independent coding-agent index across implementation, terminal use, and real codebases.. Cost, latency, and output-token coordinates are retained as review metadata, not score-matrix metrics." }, { "score": 70.5, "reference_url": "https://openai.com/index/gpt-5-6/", "source_type": "official_blog", "reported_setting": { "effort": "medium", "eval_variant": null, "code_mode": "false" }, "notes": "OpenAI GPT-5.6 release figure (Artificial Analysis Coding Index): GPT-5.5; Artificial Analysis Coding Agent Index v1.1=70.5. Figure 1: Independent coding-agent index across implementation, terminal use, and real codebases.. Cost, latency, and output-token coordinates are retained as review metadata, not score-matrix metrics." }, { "score": 72.5, "reference_url": "https://openai.com/index/gpt-5-6/", "source_type": "official_blog", "reported_setting": { "effort": "high", "eval_variant": null, "code_mode": "false" }, "notes": "OpenAI GPT-5.6 release figure (Artificial Analysis Coding Index): GPT-5.5; Artificial Analysis Coding Agent Index v1.1=72.5. Figure 1: Independent coding-agent index across implementation, terminal use, and real codebases.. Cost, latency, and output-token coordinates are retained as review metadata, not score-matrix metrics." } ] }, { "model_id": "claude-fable-5", "benchmark_id": "aa_coding_agent_index_v1_1", "score": 77.2, "reference_url": "https://openai.com/index/gpt-5-6/", "reported_setting": { "effort": "max", "eval_variant": null, "code_mode": "false" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-5.6 release table (table 1): Claude Fable 5; Artificial Analysis Coding Agent Index v1.1=77.2. Exact effort/variant resolved from the rendered chart point." }, { "model_id": "claude-opus-4.8", "benchmark_id": "aa_coding_agent_index_v1_1", "score": 72.5, "reference_url": "https://openai.com/index/gpt-5-6/", "reported_setting": { "effort": "max", "eval_variant": null, "code_mode": "false" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-5.6 release table (table 1): Claude Opus 4.8; Artificial Analysis Coding Agent Index v1.1=72.5. Exact effort/variant resolved from the rendered chart point.", "candidates": [ { "score": 67.0, "reference_url": "https://openai.com/index/gpt-5-6/", "source_type": "official_blog", "reported_setting": { "effort": "medium", "eval_variant": null, "code_mode": "false" }, "notes": "OpenAI GPT-5.6 release figure (Artificial Analysis Coding Index): Claude Opus 4.8; Artificial Analysis Coding Agent Index v1.1=67.0. Figure 1: Independent coding-agent index across implementation, terminal use, and real codebases.. Cost, latency, and output-token coordinates are retained as review metadata, not score-matrix metrics." } ] }, { "model_id": "gemini-3.1-pro", "benchmark_id": "aa_coding_agent_index_v1_1", "score": 42.7, "reference_url": "https://openai.com/index/gpt-5-6/", "reported_setting": { "effort": "n/a", "eval_variant": null, "code_mode": "false" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-5.6 release table (table 1): Gemini 3.1 Pro Preview; Artificial Analysis Coding Agent Index v1.1=42.7. Exact effort/variant resolved from the rendered chart point." }, { "model_id": "gpt-5.6-sol", "benchmark_id": "swe_bench_pro", "score": 64.6, "reference_url": "https://openai.com/index/gpt-5-6/", "reported_setting": { "mode": "thinking", "effort": "source does not state", "tools": "none", "sampling": "source does not state", "judge": "benchmark-specified", "harness": "official release evaluation", "prompt_style": "source does not state", "temperature": "source does not state", "context": "source does not state" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-5.6 release table (table 1): GPT‑5.6 Sol; SWE-Bench Pro=64.6." }, { "model_id": "gpt-5.6-terra", "benchmark_id": "swe_bench_pro", "score": 63.4, "reference_url": "https://openai.com/index/gpt-5-6/", "reported_setting": { "mode": "thinking", "effort": "source does not state", "tools": "none", "sampling": "source does not state", "judge": "benchmark-specified", "harness": "official release evaluation", "prompt_style": "source does not state", "temperature": "source does not state", "context": "source does not state" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-5.6 release table (table 1): GPT‑5.6 Terra; SWE-Bench Pro=63.4." }, { "model_id": "gpt-5.6-luna", "benchmark_id": "swe_bench_pro", "score": 62.7, "reference_url": "https://openai.com/index/gpt-5-6/", "reported_setting": { "mode": "thinking", "effort": "source does not state", "tools": "none", "sampling": "source does not state", "judge": "benchmark-specified", "harness": "official release evaluation", "prompt_style": "source does not state", "temperature": "source does not state", "context": "source does not state" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-5.6 release table (table 1): GPT‑5.6 Luna; SWE-Bench Pro=62.7." }, { "model_id": "claude-fable-5", "benchmark_id": "swe_bench_pro", "score": 80.4, "reference_url": "https://media.x.ai/v1/website/4p5-5184fdf9.pdf", "reported_setting": { "mode": "thinking", "effort": "max, with fallback", "tools": "agentic repository shell/editor", "sampling": "pass@1", "judge": "verification reward and repository tests", "harness": "fixed SWE-bench Pro agent scaffold", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "card45 SWE-Bench Pro Resolve rate (%); source effort=max, with fallback; benchmark protocol matches canonical=True. Unknown source fields remain unknown.", "candidates": [ { "score": 80.3, "reference_url": "https://openai.com/index/gpt-5-6/", "source_type": "official_blog", "reported_setting": { "mode": "thinking", "effort": "source does not state", "tools": "none", "sampling": "source does not state", "judge": "benchmark-specified", "harness": "official release evaluation", "prompt_style": "source does not state", "temperature": "source does not state", "context": "source does not state", "deployment": "claude-mythos-5", "safeguards": "lifted for Project Glasswing", "notes": "Anthropic documents Mythos 5 as the same underlying model and capabilities as Fable 5, without Fable's safety classifiers." }, "notes": "OpenAI GPT-5.6 release table (table 1): Claude Mythos 5; SWE-Bench Pro=80.3." }, { "score": 80.0, "reference_url": "https://openai.com/index/gpt-5-6/", "source_type": "official_blog", "reported_setting": { "mode": "thinking", "effort": "source does not state", "tools": "none", "sampling": "source does not state", "judge": "benchmark-specified", "harness": "official release evaluation", "prompt_style": "source does not state", "temperature": "source does not state", "context": "source does not state" }, "notes": "OpenAI GPT-5.6 release table (table 1): Claude Fable 5; SWE-Bench Pro=80.0." }, { "score": 80.4, "reference_url": "https://x.ai/news/grok-4-5", "source_type": "official_blog", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "agentic repository shell/editor", "sampling": "pass@1", "judge": "verification reward and repository tests", "harness": "fixed SWE-bench Pro agent scaffold", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown" }, "notes": "release45 SWE-Bench Pro Resolve rate (%); source effort=max; benchmark protocol matches canonical=True. Unknown source fields remain unknown." } ] }, { "model_id": "gpt-5.6-terra", "benchmark_id": "deep_swe_v1_1", "score": 69.6, "reference_url": "https://openai.com/index/gpt-5-6/", "reported_setting": { "effort": "max", "eval_variant": null, "code_mode": "false" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-5.6 release table (table 1): GPT‑5.6 Terra; DeepSWE v1.1=69.6. Exact effort/variant resolved from the rendered chart point.", "candidates": [ { "score": 14.0, "reference_url": "https://openai.com/index/gpt-5-6/", "source_type": "official_blog", "reported_setting": { "effort": "none", "eval_variant": null, "code_mode": "false" }, "notes": "OpenAI GPT-5.6 release figure (DeepSWE v1.1): GPT-5.6 Terra; DeepSWE v1.1=14.0. Figure 1: DeepSWE v1.1. Cost, latency, and output-token coordinates are retained as review metadata, not score-matrix metrics." }, { "score": 24.05, "reference_url": "https://openai.com/index/gpt-5-6/", "source_type": "official_blog", "reported_setting": { "effort": "low", "eval_variant": null, "code_mode": "false" }, "notes": "OpenAI GPT-5.6 release figure (DeepSWE v1.1): GPT-5.6 Terra; DeepSWE v1.1=24.05. Figure 1: DeepSWE v1.1. Cost, latency, and output-token coordinates are retained as review metadata, not score-matrix metrics." }, { "score": 35.11, "reference_url": "https://openai.com/index/gpt-5-6/", "source_type": "official_blog", "reported_setting": { "effort": "medium", "eval_variant": null, "code_mode": "false" }, "notes": "OpenAI GPT-5.6 release figure (DeepSWE v1.1): GPT-5.6 Terra; DeepSWE v1.1=35.11. Figure 1: DeepSWE v1.1. Cost, latency, and output-token coordinates are retained as review metadata, not score-matrix metrics." }, { "score": 53.76, "reference_url": "https://openai.com/index/gpt-5-6/", "source_type": "official_blog", "reported_setting": { "effort": "high", "eval_variant": null, "code_mode": "false" }, "notes": "OpenAI GPT-5.6 release figure (DeepSWE v1.1): GPT-5.6 Terra; DeepSWE v1.1=53.76. Figure 1: DeepSWE v1.1. Cost, latency, and output-token coordinates are retained as review metadata, not score-matrix metrics." }, { "score": 60.18, "reference_url": "https://openai.com/index/gpt-5-6/", "source_type": "official_blog", "reported_setting": { "effort": "xhigh", "eval_variant": null, "code_mode": "false" }, "notes": "OpenAI GPT-5.6 release figure (DeepSWE v1.1): GPT-5.6 Terra; DeepSWE v1.1=60.18. Figure 1: DeepSWE v1.1. Cost, latency, and output-token coordinates are retained as review metadata, not score-matrix metrics." }, { "score": 64.8, "reference_url": "https://research.meta.ai/articles/introducing-muse-code-and-muse-spark-1-2/evaluations/deepswe-1-1-v1.png", "source_type": "official_blog", "reported_setting": { "mode": "thinking", "effort": "max", "prompt_style": "not disclosed", "temperature": "not disclosed", "context": "not disclosed", "agent": "Codex", "attempts": 5, "tools": "Codex repository shell/editor toolset", "sampling": "pass@1 averaged across five attempts per task", "judge": "functional verifier plus regression checks in a pristine verifier container", "harness": "Meta internal agent evaluation framework; Codex; isolated Daytona sandbox; Harbor conversion", "internet": "disabled during rollout and grading; model endpoint only", "configuration": "five-language Harbor conversion with pristine verifier" }, "notes": "Exact printed value from the locked Meta release chart; evaluation settings come from the locked official methodology." } ] }, { "model_id": "gpt-5.6-luna", "benchmark_id": "deep_swe_v1_1", "score": 67.2, "reference_url": "https://openai.com/index/gpt-5-6/", "reported_setting": { "effort": "max", "eval_variant": null, "code_mode": "false" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-5.6 release table (table 1): GPT‑5.6 Luna; DeepSWE v1.1=67.2. Exact effort/variant resolved from the rendered chart point.", "candidates": [ { "score": 2.21, "reference_url": "https://openai.com/index/gpt-5-6/", "source_type": "official_blog", "reported_setting": { "effort": "none", "eval_variant": null, "code_mode": "false" }, "notes": "OpenAI GPT-5.6 release figure (DeepSWE v1.1): GPT-5.6 Luna; DeepSWE v1.1=2.21. Figure 1: DeepSWE v1.1. Cost, latency, and output-token coordinates are retained as review metadata, not score-matrix metrics." }, { "score": 1.55, "reference_url": "https://openai.com/index/gpt-5-6/", "source_type": "official_blog", "reported_setting": { "effort": "low", "eval_variant": null, "code_mode": "false" }, "notes": "OpenAI GPT-5.6 release figure (DeepSWE v1.1): GPT-5.6 Luna; DeepSWE v1.1=1.55. Figure 1: DeepSWE v1.1. Cost, latency, and output-token coordinates are retained as review metadata, not score-matrix metrics." }, { "score": 11.28, "reference_url": "https://openai.com/index/gpt-5-6/", "source_type": "official_blog", "reported_setting": { "effort": "medium", "eval_variant": null, "code_mode": "false" }, "notes": "OpenAI GPT-5.6 release figure (DeepSWE v1.1): GPT-5.6 Luna; DeepSWE v1.1=11.28. Figure 1: DeepSWE v1.1. Cost, latency, and output-token coordinates are retained as review metadata, not score-matrix metrics." }, { "score": 44.25, "reference_url": "https://openai.com/index/gpt-5-6/", "source_type": "official_blog", "reported_setting": { "effort": "high", "eval_variant": null, "code_mode": "false" }, "notes": "OpenAI GPT-5.6 release figure (DeepSWE v1.1): GPT-5.6 Luna; DeepSWE v1.1=44.25. Figure 1: DeepSWE v1.1. Cost, latency, and output-token coordinates are retained as review metadata, not score-matrix metrics." }, { "score": 56.86, "reference_url": "https://openai.com/index/gpt-5-6/", "source_type": "official_blog", "reported_setting": { "effort": "xhigh", "eval_variant": null, "code_mode": "false" }, "notes": "OpenAI GPT-5.6 release figure (DeepSWE v1.1): GPT-5.6 Luna; DeepSWE v1.1=56.86. Figure 1: DeepSWE v1.1. Cost, latency, and output-token coordinates are retained as review metadata, not score-matrix metrics." }, { "score": 67.0, "reference_url": "https://deepmind.google/models/model-cards/gemini-3-6-flash/", "source_type": "official_model_card", "reported_setting": { "mode": "thinking/reasoning", "effort": "maximum/best available", "tools": "code execution", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "DataCurve DeepSWE v1.1 leaderboard", "temperature": "default", "snapshot": "July 2026" }, "notes": "Google Gemini 3.6 Flash July 2026 matrix: DeepSWE v1.1. Exact effort, tools, sampling and harness are preserved." } ] }, { "model_id": "gpt-5.6-terra", "benchmark_id": "terminal_bench_2_1", "score": 87.4, "reference_url": "https://openai.com/index/gpt-5-6/", "reported_setting": { "mode": "thinking", "effort": "source does not state", "tools": "none", "sampling": "source does not state", "judge": "benchmark-specified", "harness": "official release evaluation", "prompt_style": "source does not state", "temperature": "source does not state", "context": "source does not state" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-5.6 release table (table 1): GPT‑5.6 Terra; Terminal-Bench 2.1=87.4.", "candidates": [ { "score": 55.51, "reference_url": "https://openai.com/index/gpt-5-6/", "source_type": "official_blog", "reported_setting": { "effort": "low", "eval_variant": null, "code_mode": "true" }, "notes": "OpenAI GPT-5.6 release figure (Terminal-Bench 2.1): GPT-5.6 Terra; Terminal-Bench 2.1=55.51. Figure 1: Terminal-Bench 2.1. Cost, latency, and output-token coordinates are retained as review metadata, not score-matrix metrics." }, { "score": 71.69, "reference_url": "https://openai.com/index/gpt-5-6/", "source_type": "official_blog", "reported_setting": { "effort": "medium", "eval_variant": null, "code_mode": "true" }, "notes": "OpenAI GPT-5.6 release figure (Terminal-Bench 2.1): GPT-5.6 Terra; Terminal-Bench 2.1=71.69. Figure 1: Terminal-Bench 2.1. Cost, latency, and output-token coordinates are retained as review metadata, not score-matrix metrics." }, { "score": 76.63, "reference_url": "https://openai.com/index/gpt-5-6/", "source_type": "official_blog", "reported_setting": { "effort": "high", "eval_variant": null, "code_mode": "true" }, "notes": "OpenAI GPT-5.6 release figure (Terminal-Bench 2.1): GPT-5.6 Terra; Terminal-Bench 2.1=76.63. Figure 1: Terminal-Bench 2.1. Cost, latency, and output-token coordinates are retained as review metadata, not score-matrix metrics." }, { "score": 79.55, "reference_url": "https://openai.com/index/gpt-5-6/", "source_type": "official_blog", "reported_setting": { "effort": "xhigh", "eval_variant": null, "code_mode": "true" }, "notes": "OpenAI GPT-5.6 release figure (Terminal-Bench 2.1): GPT-5.6 Terra; Terminal-Bench 2.1=79.55. Figure 1: Terminal-Bench 2.1. Cost, latency, and output-token coordinates are retained as review metadata, not score-matrix metrics." }, { "score": 84.27, "reference_url": "https://openai.com/index/gpt-5-6/", "source_type": "official_blog", "reported_setting": { "effort": "max", "eval_variant": null, "code_mode": "true" }, "notes": "OpenAI GPT-5.6 release figure (Terminal-Bench 2.1): GPT-5.6 Terra; Terminal-Bench 2.1=84.27. Figure 1: Terminal-Bench 2.1. Cost, latency, and output-token coordinates are retained as review metadata, not score-matrix metrics." }, { "score": 78.4, "reference_url": "https://www.tbench.ai/leaderboard/terminal-bench/2.1", "source_type": "leaderboard", "reported_setting": { "mode": "thinking", "effort": "max", "prompt_style": "not disclosed", "temperature": "not disclosed", "context": "not disclosed", "output_limit": "not disclosed", "trials": 5, "dataset_version": "Terminal-Bench 2.1", "dataset_split": "89 tasks", "tools": "Codex terminal toolset", "sampling": "pass@1", "aggregation": "mean across five attempts per task", "judge": "Terminal-Bench 2.1 executable verifier", "harness": "Codex official leaderboard submission" }, "notes": "Current original-leaderboard configuration differs from the provider release value and remains a separate candidate." }, { "score": 81.8, "reference_url": "https://research.meta.ai/articles/introducing-muse-code-and-muse-spark-1-2/evaluations/terminal-bench-2-1-v1.png", "source_type": "official_blog", "reported_setting": { "mode": "thinking", "effort": "max", "prompt_style": "not disclosed", "temperature": "not disclosed", "context": "not disclosed", "agent": "Codex", "attempts": 5, "tools": "Codex terminal coding toolset", "sampling": "pass@1 averaged across five attempts per task", "judge": "Terminal-Bench 2.1 executable task verifier", "harness": "Meta internal agent evaluation framework; Codex; isolated Daytona sandbox", "internet": "not disclosed", "configuration": "selected official agent product in isolated Daytona" }, "notes": "Exact printed value from the locked Meta release chart; evaluation settings come from the locked official methodology." } ] }, { "model_id": "gpt-5.6-luna", "benchmark_id": "terminal_bench_2_1", "score": 84.7, "reference_url": "https://openai.com/index/gpt-5-6/", "reported_setting": { "mode": "thinking", "effort": "source does not state", "tools": "none", "sampling": "source does not state", "judge": "benchmark-specified", "harness": "official release evaluation", "prompt_style": "source does not state", "temperature": "source does not state", "context": "source does not state" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-5.6 release table (table 1): GPT‑5.6 Luna; Terminal-Bench 2.1=84.7.", "candidates": [ { "score": 49.44, "reference_url": "https://openai.com/index/gpt-5-6/", "source_type": "official_blog", "reported_setting": { "effort": "low", "eval_variant": null, "code_mode": "true" }, "notes": "OpenAI GPT-5.6 release figure (Terminal-Bench 2.1): GPT-5.6 Luna; Terminal-Bench 2.1=49.44. Figure 1: Terminal-Bench 2.1. Cost, latency, and output-token coordinates are retained as review metadata, not score-matrix metrics." }, { "score": 58.65, "reference_url": "https://openai.com/index/gpt-5-6/", "source_type": "official_blog", "reported_setting": { "effort": "medium", "eval_variant": null, "code_mode": "true" }, "notes": "OpenAI GPT-5.6 release figure (Terminal-Bench 2.1): GPT-5.6 Luna; Terminal-Bench 2.1=58.65. Figure 1: Terminal-Bench 2.1. Cost, latency, and output-token coordinates are retained as review metadata, not score-matrix metrics." }, { "score": 73.93, "reference_url": "https://openai.com/index/gpt-5-6/", "source_type": "official_blog", "reported_setting": { "effort": "high", "eval_variant": null, "code_mode": "true" }, "notes": "OpenAI GPT-5.6 release figure (Terminal-Bench 2.1): GPT-5.6 Luna; Terminal-Bench 2.1=73.93. Figure 1: Terminal-Bench 2.1. Cost, latency, and output-token coordinates are retained as review metadata, not score-matrix metrics." }, { "score": 74.61, "reference_url": "https://openai.com/index/gpt-5-6/", "source_type": "official_blog", "reported_setting": { "effort": "xhigh", "eval_variant": null, "code_mode": "true" }, "notes": "OpenAI GPT-5.6 release figure (Terminal-Bench 2.1): GPT-5.6 Luna; Terminal-Bench 2.1=74.61. Figure 1: Terminal-Bench 2.1. Cost, latency, and output-token coordinates are retained as review metadata, not score-matrix metrics." }, { "score": 82.47, "reference_url": "https://openai.com/index/gpt-5-6/", "source_type": "official_blog", "reported_setting": { "effort": "max", "eval_variant": null, "code_mode": "true" }, "notes": "OpenAI GPT-5.6 release figure (Terminal-Bench 2.1): GPT-5.6 Luna; Terminal-Bench 2.1=82.47. Figure 1: Terminal-Bench 2.1. Cost, latency, and output-token coordinates are retained as review metadata, not score-matrix metrics." } ] }, { "model_id": "gpt-5.6-sol", "benchmark_id": "genebench_pro", "score": 28.7, "reference_url": "https://openai.com/index/gpt-5-6/", "reported_setting": { "effort": "max", "eval_variant": null, "code_mode": "false" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-5.6 release table (table 2): GPT‑5.6 Sol; GeneBench Pro=28.7. Distinct professional variant from the existing GeneBench row. Exact effort/variant resolved from the rendered chart point.", "candidates": [ { "score": 3.72, "reference_url": "https://openai.com/index/gpt-5-6/", "source_type": "official_blog", "reported_setting": { "effort": "none", "eval_variant": null, "code_mode": "false" }, "notes": "OpenAI GPT-5.6 release figure (GeneBench Pro): GPT-5.6 Sol; GeneBench Pro=3.72. Figure 28: Long-horizon genomics and quantitative-biology analyses.. Cost, latency, and output-token coordinates are retained as review metadata, not score-matrix metrics." }, { "score": 7.67, "reference_url": "https://openai.com/index/gpt-5-6/", "source_type": "official_blog", "reported_setting": { "effort": "low", "eval_variant": null, "code_mode": "false" }, "notes": "OpenAI GPT-5.6 release figure (GeneBench Pro): GPT-5.6 Sol; GeneBench Pro=7.67. Figure 28: Long-horizon genomics and quantitative-biology analyses.. Cost, latency, and output-token coordinates are retained as review metadata, not score-matrix metrics." }, { "score": 15.12, "reference_url": "https://openai.com/index/gpt-5-6/", "source_type": "official_blog", "reported_setting": { "effort": "medium", "eval_variant": null, "code_mode": "false" }, "notes": "OpenAI GPT-5.6 release figure (GeneBench Pro): GPT-5.6 Sol; GeneBench Pro=15.12. Figure 28: Long-horizon genomics and quantitative-biology analyses.. Cost, latency, and output-token coordinates are retained as review metadata, not score-matrix metrics." }, { "score": 22.64, "reference_url": "https://openai.com/index/gpt-5-6/", "source_type": "official_blog", "reported_setting": { "effort": "high", "eval_variant": null, "code_mode": "false" }, "notes": "OpenAI GPT-5.6 release figure (GeneBench Pro): GPT-5.6 Sol; GeneBench Pro=22.64. Figure 28: Long-horizon genomics and quantitative-biology analyses.. Cost, latency, and output-token coordinates are retained as review metadata, not score-matrix metrics." }, { "score": 24.41, "reference_url": "https://openai.com/index/gpt-5-6/", "source_type": "official_blog", "reported_setting": { "effort": "xhigh", "eval_variant": null, "code_mode": "false" }, "notes": "OpenAI GPT-5.6 release figure (GeneBench Pro): GPT-5.6 Sol; GeneBench Pro=24.41. Figure 28: Long-horizon genomics and quantitative-biology analyses.. Cost, latency, and output-token coordinates are retained as review metadata, not score-matrix metrics." } ] }, { "model_id": "gpt-5.6-terra", "benchmark_id": "genebench_pro", "score": 23.3, "reference_url": "https://openai.com/index/gpt-5-6/", "reported_setting": { "effort": "max", "eval_variant": null, "code_mode": "false" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-5.6 release table (table 2): GPT‑5.6 Terra; GeneBench Pro=23.3. Distinct professional variant from the existing GeneBench row. Exact effort/variant resolved from the rendered chart point.", "candidates": [ { "score": 1.01, "reference_url": "https://openai.com/index/gpt-5-6/", "source_type": "official_blog", "reported_setting": { "effort": "none", "eval_variant": null, "code_mode": "false" }, "notes": "OpenAI GPT-5.6 release figure (GeneBench Pro): GPT-5.6 Terra; GeneBench Pro=1.01. Figure 28: Long-horizon genomics and quantitative-biology analyses.. Cost, latency, and output-token coordinates are retained as review metadata, not score-matrix metrics." }, { "score": 4.21, "reference_url": "https://openai.com/index/gpt-5-6/", "source_type": "official_blog", "reported_setting": { "effort": "low", "eval_variant": null, "code_mode": "false" }, "notes": "OpenAI GPT-5.6 release figure (GeneBench Pro): GPT-5.6 Terra; GeneBench Pro=4.21. Figure 28: Long-horizon genomics and quantitative-biology analyses.. Cost, latency, and output-token coordinates are retained as review metadata, not score-matrix metrics." }, { "score": 6.28, "reference_url": "https://openai.com/index/gpt-5-6/", "source_type": "official_blog", "reported_setting": { "effort": "medium", "eval_variant": null, "code_mode": "false" }, "notes": "OpenAI GPT-5.6 release figure (GeneBench Pro): GPT-5.6 Terra; GeneBench Pro=6.28. Figure 28: Long-horizon genomics and quantitative-biology analyses.. Cost, latency, and output-token coordinates are retained as review metadata, not score-matrix metrics." }, { "score": 13.57, "reference_url": "https://openai.com/index/gpt-5-6/", "source_type": "official_blog", "reported_setting": { "effort": "high", "eval_variant": null, "code_mode": "false" }, "notes": "OpenAI GPT-5.6 release figure (GeneBench Pro): GPT-5.6 Terra; GeneBench Pro=13.57. Figure 28: Long-horizon genomics and quantitative-biology analyses.. Cost, latency, and output-token coordinates are retained as review metadata, not score-matrix metrics." }, { "score": 16.2, "reference_url": "https://openai.com/index/gpt-5-6/", "source_type": "official_blog", "reported_setting": { "effort": "xhigh", "eval_variant": null, "code_mode": "false" }, "notes": "OpenAI GPT-5.6 release figure (GeneBench Pro): GPT-5.6 Terra; GeneBench Pro=16.2. Figure 28: Long-horizon genomics and quantitative-biology analyses.. Cost, latency, and output-token coordinates are retained as review metadata, not score-matrix metrics." } ] }, { "model_id": "gpt-5.6-luna", "benchmark_id": "genebench_pro", "score": 10.8, "reference_url": "https://openai.com/index/gpt-5-6/", "reported_setting": { "effort": "max", "eval_variant": null, "code_mode": "false" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-5.6 release table (table 2): GPT‑5.6 Luna; GeneBench Pro=10.8. Distinct professional variant from the existing GeneBench row. Exact effort/variant resolved from the rendered chart point.", "candidates": [ { "score": 0.78, "reference_url": "https://openai.com/index/gpt-5-6/", "source_type": "official_blog", "reported_setting": { "effort": "none", "eval_variant": null, "code_mode": "false" }, "notes": "OpenAI GPT-5.6 release figure (GeneBench Pro): GPT-5.6 Luna; GeneBench Pro=0.78. Figure 28: Long-horizon genomics and quantitative-biology analyses.. Cost, latency, and output-token coordinates are retained as review metadata, not score-matrix metrics." }, { "score": 1.63, "reference_url": "https://openai.com/index/gpt-5-6/", "source_type": "official_blog", "reported_setting": { "effort": "low", "eval_variant": null, "code_mode": "false" }, "notes": "OpenAI GPT-5.6 release figure (GeneBench Pro): GPT-5.6 Luna; GeneBench Pro=1.63. Figure 28: Long-horizon genomics and quantitative-biology analyses.. Cost, latency, and output-token coordinates are retained as review metadata, not score-matrix metrics." }, { "score": 1.86, "reference_url": "https://openai.com/index/gpt-5-6/", "source_type": "official_blog", "reported_setting": { "effort": "medium", "eval_variant": null, "code_mode": "false" }, "notes": "OpenAI GPT-5.6 release figure (GeneBench Pro): GPT-5.6 Luna; GeneBench Pro=1.86. Figure 28: Long-horizon genomics and quantitative-biology analyses.. Cost, latency, and output-token coordinates are retained as review metadata, not score-matrix metrics." }, { "score": 4.7, "reference_url": "https://openai.com/index/gpt-5-6/", "source_type": "official_blog", "reported_setting": { "effort": "high", "eval_variant": null, "code_mode": "false" }, "notes": "OpenAI GPT-5.6 release figure (GeneBench Pro): GPT-5.6 Luna; GeneBench Pro=4.7. Figure 28: Long-horizon genomics and quantitative-biology analyses.. Cost, latency, and output-token coordinates are retained as review metadata, not score-matrix metrics." }, { "score": 8.04, "reference_url": "https://openai.com/index/gpt-5-6/", "source_type": "official_blog", "reported_setting": { "effort": "xhigh", "eval_variant": null, "code_mode": "false" }, "notes": "OpenAI GPT-5.6 release figure (GeneBench Pro): GPT-5.6 Luna; GeneBench Pro=8.04. Figure 28: Long-horizon genomics and quantitative-biology analyses.. Cost, latency, and output-token coordinates are retained as review metadata, not score-matrix metrics." } ] }, { "model_id": "gpt-5.5", "benchmark_id": "genebench_pro", "score": 12.0, "reference_url": "https://openai.com/index/gpt-5-6/", "reported_setting": { "effort": "xhigh", "eval_variant": null, "code_mode": "false" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-5.6 release table (table 2): GPT‑5.5; GeneBench Pro=12.0. Distinct professional variant from the existing GeneBench row. Exact effort/variant resolved from the rendered chart point.", "candidates": [ { "score": 0.78, "reference_url": "https://openai.com/index/gpt-5-6/", "source_type": "official_blog", "reported_setting": { "effort": "none", "eval_variant": null, "code_mode": "false" }, "notes": "OpenAI GPT-5.6 release figure (GeneBench Pro): GPT-5.5; GeneBench Pro=0.78. Figure 28: Long-horizon genomics and quantitative-biology analyses.. Cost, latency, and output-token coordinates are retained as review metadata, not score-matrix metrics." }, { "score": 2.41, "reference_url": "https://openai.com/index/gpt-5-6/", "source_type": "official_blog", "reported_setting": { "effort": "low", "eval_variant": null, "code_mode": "false" }, "notes": "OpenAI GPT-5.6 release figure (GeneBench Pro): GPT-5.5; GeneBench Pro=2.41. Figure 28: Long-horizon genomics and quantitative-biology analyses.. Cost, latency, and output-token coordinates are retained as review metadata, not score-matrix metrics." }, { "score": 5.89, "reference_url": "https://openai.com/index/gpt-5-6/", "source_type": "official_blog", "reported_setting": { "effort": "medium", "eval_variant": null, "code_mode": "false" }, "notes": "OpenAI GPT-5.6 release figure (GeneBench Pro): GPT-5.5; GeneBench Pro=5.89. Figure 28: Long-horizon genomics and quantitative-biology analyses.. Cost, latency, and output-token coordinates are retained as review metadata, not score-matrix metrics." }, { "score": 9.28, "reference_url": "https://openai.com/index/gpt-5-6/", "source_type": "official_blog", "reported_setting": { "effort": "high", "eval_variant": null, "code_mode": "false" }, "notes": "OpenAI GPT-5.6 release figure (GeneBench Pro): GPT-5.5; GeneBench Pro=9.28. Figure 28: Long-horizon genomics and quantitative-biology analyses.. Cost, latency, and output-token coordinates are retained as review metadata, not score-matrix metrics." } ] }, { "model_id": "claude-opus-4.8", "benchmark_id": "genebench_pro", "score": 16.0, "reference_url": "https://openai.com/index/gpt-5-6/", "reported_setting": { "effort": "reference", "eval_variant": null, "code_mode": null }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-5.6 release table (table 2): Claude Opus 4.8; GeneBench Pro=16.0. Distinct professional variant from the existing GeneBench row. Exact effort/variant resolved from the rendered chart point." }, { "model_id": "gemini-3.1-pro", "benchmark_id": "genebench_pro", "score": 3.1, "reference_url": "https://openai.com/index/gpt-5-6/", "reported_setting": { "effort": "reference", "eval_variant": null, "code_mode": null }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-5.6 release table (table 2): Gemini 3.1 Pro Preview; GeneBench Pro=3.1. Distinct professional variant from the existing GeneBench row. Exact effort/variant resolved from the rendered chart point." }, { "model_id": "gemini-3.5-flash", "benchmark_id": "genebench_pro", "score": 8.14, "reference_url": "https://openai.com/index/gpt-5-6/", "reported_setting": { "effort": "reference", "eval_variant": null, "code_mode": null }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-5.6 release table (table 2): Gemini 3.5 Flash; GeneBench Pro=8.14. Distinct professional variant from the existing GeneBench row. Exact effort/variant resolved from the rendered chart point." }, { "model_id": "gpt-5.6-sol", "benchmark_id": "lifescibench", "score": 59.9, "reference_url": "https://openai.com/index/gpt-5-6/", "reported_setting": { "effort": "max", "eval_variant": null, "code_mode": "false" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-5.6 release table (table 2): GPT‑5.6 Sol; LifeSciBench=59.9. Exact effort/variant resolved from the rendered chart point.", "candidates": [ { "score": 39.64, "reference_url": "https://openai.com/index/gpt-5-6/", "source_type": "official_blog", "reported_setting": { "effort": "none", "eval_variant": null, "code_mode": "false" }, "notes": "OpenAI GPT-5.6 release figure (LifeSciBench): GPT-5.6 Sol; LifeSciBench=39.64. Figure 28: LifeSciBench. Cost, latency, and output-token coordinates are retained as review metadata, not score-matrix metrics." }, { "score": 42.63, "reference_url": "https://openai.com/index/gpt-5-6/", "source_type": "official_blog", "reported_setting": { "effort": "low", "eval_variant": null, "code_mode": "false" }, "notes": "OpenAI GPT-5.6 release figure (LifeSciBench): GPT-5.6 Sol; LifeSciBench=42.63. Figure 28: LifeSciBench. Cost, latency, and output-token coordinates are retained as review metadata, not score-matrix metrics." }, { "score": 49.84, "reference_url": "https://openai.com/index/gpt-5-6/", "source_type": "official_blog", "reported_setting": { "effort": "medium", "eval_variant": null, "code_mode": "false" }, "notes": "OpenAI GPT-5.6 release figure (LifeSciBench): GPT-5.6 Sol; LifeSciBench=49.84. Figure 28: LifeSciBench. Cost, latency, and output-token coordinates are retained as review metadata, not score-matrix metrics." }, { "score": 54.61, "reference_url": "https://openai.com/index/gpt-5-6/", "source_type": "official_blog", "reported_setting": { "effort": "high", "eval_variant": null, "code_mode": "false" }, "notes": "OpenAI GPT-5.6 release figure (LifeSciBench): GPT-5.6 Sol; LifeSciBench=54.61. Figure 28: LifeSciBench. Cost, latency, and output-token coordinates are retained as review metadata, not score-matrix metrics." }, { "score": 56.29, "reference_url": "https://openai.com/index/gpt-5-6/", "source_type": "official_blog", "reported_setting": { "effort": "xhigh", "eval_variant": null, "code_mode": "false" }, "notes": "OpenAI GPT-5.6 release figure (LifeSciBench): GPT-5.6 Sol; LifeSciBench=56.29. Figure 28: LifeSciBench. Cost, latency, and output-token coordinates are retained as review metadata, not score-matrix metrics." } ] }, { "model_id": "gpt-5.6-terra", "benchmark_id": "lifescibench", "score": 56.0, "reference_url": "https://openai.com/index/gpt-5-6/", "reported_setting": { "effort": "max", "eval_variant": null, "code_mode": "false" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-5.6 release table (table 2): GPT‑5.6 Terra; LifeSciBench=56.0. Exact effort/variant resolved from the rendered chart point.", "candidates": [ { "score": 37.21, "reference_url": "https://openai.com/index/gpt-5-6/", "source_type": "official_blog", "reported_setting": { "effort": "low", "eval_variant": null, "code_mode": "false" }, "notes": "OpenAI GPT-5.6 release figure (LifeSciBench): GPT-5.6 Terra; LifeSciBench=37.21. Figure 28: LifeSciBench. Cost, latency, and output-token coordinates are retained as review metadata, not score-matrix metrics." }, { "score": 41.38, "reference_url": "https://openai.com/index/gpt-5-6/", "source_type": "official_blog", "reported_setting": { "effort": "medium", "eval_variant": null, "code_mode": "false" }, "notes": "OpenAI GPT-5.6 release figure (LifeSciBench): GPT-5.6 Terra; LifeSciBench=41.38. Figure 28: LifeSciBench. Cost, latency, and output-token coordinates are retained as review metadata, not score-matrix metrics." }, { "score": 48.09, "reference_url": "https://openai.com/index/gpt-5-6/", "source_type": "official_blog", "reported_setting": { "effort": "high", "eval_variant": null, "code_mode": "false" }, "notes": "OpenAI GPT-5.6 release figure (LifeSciBench): GPT-5.6 Terra; LifeSciBench=48.09. Figure 28: LifeSciBench. Cost, latency, and output-token coordinates are retained as review metadata, not score-matrix metrics." }, { "score": 50.82, "reference_url": "https://openai.com/index/gpt-5-6/", "source_type": "official_blog", "reported_setting": { "effort": "xhigh", "eval_variant": null, "code_mode": "false" }, "notes": "OpenAI GPT-5.6 release figure (LifeSciBench): GPT-5.6 Terra; LifeSciBench=50.82. Figure 28: LifeSciBench. Cost, latency, and output-token coordinates are retained as review metadata, not score-matrix metrics." } ] }, { "model_id": "gpt-5.6-luna", "benchmark_id": "lifescibench", "score": 51.2, "reference_url": "https://openai.com/index/gpt-5-6/", "reported_setting": { "effort": "max", "eval_variant": null, "code_mode": "false" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-5.6 release table (table 2): GPT‑5.6 Luna; LifeSciBench=51.2. Exact effort/variant resolved from the rendered chart point.", "candidates": [ { "score": 32.51, "reference_url": "https://openai.com/index/gpt-5-6/", "source_type": "official_blog", "reported_setting": { "effort": "low", "eval_variant": null, "code_mode": "false" }, "notes": "OpenAI GPT-5.6 release figure (LifeSciBench): GPT-5.6 Luna; LifeSciBench=32.51. Figure 28: LifeSciBench. Cost, latency, and output-token coordinates are retained as review metadata, not score-matrix metrics." }, { "score": 36.23, "reference_url": "https://openai.com/index/gpt-5-6/", "source_type": "official_blog", "reported_setting": { "effort": "medium", "eval_variant": null, "code_mode": "false" }, "notes": "OpenAI GPT-5.6 release figure (LifeSciBench): GPT-5.6 Luna; LifeSciBench=36.23. Figure 28: LifeSciBench. Cost, latency, and output-token coordinates are retained as review metadata, not score-matrix metrics." }, { "score": 43.49, "reference_url": "https://openai.com/index/gpt-5-6/", "source_type": "official_blog", "reported_setting": { "effort": "high", "eval_variant": null, "code_mode": "false" }, "notes": "OpenAI GPT-5.6 release figure (LifeSciBench): GPT-5.6 Luna; LifeSciBench=43.49. Figure 28: LifeSciBench. Cost, latency, and output-token coordinates are retained as review metadata, not score-matrix metrics." }, { "score": 47.99, "reference_url": "https://openai.com/index/gpt-5-6/", "source_type": "official_blog", "reported_setting": { "effort": "xhigh", "eval_variant": null, "code_mode": "false" }, "notes": "OpenAI GPT-5.6 release figure (LifeSciBench): GPT-5.6 Luna; LifeSciBench=47.99. Figure 28: LifeSciBench. Cost, latency, and output-token coordinates are retained as review metadata, not score-matrix metrics." } ] }, { "model_id": "gpt-5.5", "benchmark_id": "lifescibench", "score": 50.4, "reference_url": "https://openai.com/index/gpt-5-6/", "reported_setting": { "effort": "xhigh", "eval_variant": null, "code_mode": "false" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-5.6 release table (table 2): GPT‑5.5; LifeSciBench=50.4. Exact effort/variant resolved from the rendered chart point.", "candidates": [ { "score": 37.9, "reference_url": "https://openai.com/index/gpt-5-6/", "source_type": "official_blog", "reported_setting": { "effort": "low", "eval_variant": null, "code_mode": "false" }, "notes": "OpenAI GPT-5.6 release figure (LifeSciBench): GPT-5.5; LifeSciBench=37.9. Figure 28: LifeSciBench. Cost, latency, and output-token coordinates are retained as review metadata, not score-matrix metrics." }, { "score": 44.89, "reference_url": "https://openai.com/index/gpt-5-6/", "source_type": "official_blog", "reported_setting": { "effort": "medium", "eval_variant": null, "code_mode": "false" }, "notes": "OpenAI GPT-5.6 release figure (LifeSciBench): GPT-5.5; LifeSciBench=44.89. Figure 28: LifeSciBench. Cost, latency, and output-token coordinates are retained as review metadata, not score-matrix metrics." }, { "score": 48.39, "reference_url": "https://openai.com/index/gpt-5-6/", "source_type": "official_blog", "reported_setting": { "effort": "high", "eval_variant": null, "code_mode": "false" }, "notes": "OpenAI GPT-5.6 release figure (LifeSciBench): GPT-5.5; LifeSciBench=48.39. Figure 28: LifeSciBench. Cost, latency, and output-token coordinates are retained as review metadata, not score-matrix metrics." } ] }, { "model_id": "claude-opus-4.8", "benchmark_id": "lifescibench", "score": 53.6, "reference_url": "https://openai.com/index/gpt-5-6/", "reported_setting": { "effort": "reference", "eval_variant": null, "code_mode": null }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-5.6 release table (table 2): Claude Opus 4.8; LifeSciBench=53.6. Exact effort/variant resolved from the rendered chart point." }, { "model_id": "gpt-5.6-sol", "benchmark_id": "medchembench_internal", "score": 48.3, "reference_url": "https://openai.com/index/gpt-5-6/", "reported_setting": { "effort": "max", "eval_variant": null, "code_mode": "false" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-5.6 release table (table 2): GPT‑5.6 Sol; MedChemBench (Internal)=48.3. Named internal evaluation; task count not disclosed. Exact effort/variant resolved from the rendered chart point.", "candidates": [ { "score": 18.78, "reference_url": "https://openai.com/index/gpt-5-6/", "source_type": "official_blog", "reported_setting": { "effort": "none", "eval_variant": null, "code_mode": "false" }, "notes": "OpenAI GPT-5.6 release figure (MedChemBench): GPT-5.6 Sol; MedChemBench (Internal)=18.78. Figure 28: MedChemBench. Cost, latency, and output-token coordinates are retained as review metadata, not score-matrix metrics." }, { "score": 26.7, "reference_url": "https://openai.com/index/gpt-5-6/", "source_type": "official_blog", "reported_setting": { "effort": "low", "eval_variant": null, "code_mode": "false" }, "notes": "OpenAI GPT-5.6 release figure (MedChemBench): GPT-5.6 Sol; MedChemBench (Internal)=26.7. Figure 28: MedChemBench. Cost, latency, and output-token coordinates are retained as review metadata, not score-matrix metrics." }, { "score": 37.16, "reference_url": "https://openai.com/index/gpt-5-6/", "source_type": "official_blog", "reported_setting": { "effort": "medium", "eval_variant": null, "code_mode": "false" }, "notes": "OpenAI GPT-5.6 release figure (MedChemBench): GPT-5.6 Sol; MedChemBench (Internal)=37.16. Figure 28: MedChemBench. Cost, latency, and output-token coordinates are retained as review metadata, not score-matrix metrics." }, { "score": 43.41, "reference_url": "https://openai.com/index/gpt-5-6/", "source_type": "official_blog", "reported_setting": { "effort": "high", "eval_variant": null, "code_mode": "false" }, "notes": "OpenAI GPT-5.6 release figure (MedChemBench): GPT-5.6 Sol; MedChemBench (Internal)=43.41. Figure 28: MedChemBench. Cost, latency, and output-token coordinates are retained as review metadata, not score-matrix metrics." }, { "score": 46.05, "reference_url": "https://openai.com/index/gpt-5-6/", "source_type": "official_blog", "reported_setting": { "effort": "xhigh", "eval_variant": null, "code_mode": "false" }, "notes": "OpenAI GPT-5.6 release figure (MedChemBench): GPT-5.6 Sol; MedChemBench (Internal)=46.05. Figure 28: MedChemBench. Cost, latency, and output-token coordinates are retained as review metadata, not score-matrix metrics." } ] }, { "model_id": "gpt-5.6-terra", "benchmark_id": "medchembench_internal", "score": 35.0, "reference_url": "https://openai.com/index/gpt-5-6/", "reported_setting": { "effort": "max", "eval_variant": null, "code_mode": "false" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-5.6 release table (table 2): GPT‑5.6 Terra; MedChemBench (Internal)=35.0. Named internal evaluation; task count not disclosed. Exact effort/variant resolved from the rendered chart point.", "candidates": [ { "score": 13.79, "reference_url": "https://openai.com/index/gpt-5-6/", "source_type": "official_blog", "reported_setting": { "effort": "low", "eval_variant": null, "code_mode": "false" }, "notes": "OpenAI GPT-5.6 release figure (MedChemBench): GPT-5.6 Terra; MedChemBench (Internal)=13.79. Figure 28: MedChemBench. Cost, latency, and output-token coordinates are retained as review metadata, not score-matrix metrics." }, { "score": 17.4, "reference_url": "https://openai.com/index/gpt-5-6/", "source_type": "official_blog", "reported_setting": { "effort": "medium", "eval_variant": null, "code_mode": "false" }, "notes": "OpenAI GPT-5.6 release figure (MedChemBench): GPT-5.6 Terra; MedChemBench (Internal)=17.4. Figure 28: MedChemBench. Cost, latency, and output-token coordinates are retained as review metadata, not score-matrix metrics." }, { "score": 25.26, "reference_url": "https://openai.com/index/gpt-5-6/", "source_type": "official_blog", "reported_setting": { "effort": "high", "eval_variant": null, "code_mode": "false" }, "notes": "OpenAI GPT-5.6 release figure (MedChemBench): GPT-5.6 Terra; MedChemBench (Internal)=25.26. Figure 28: MedChemBench. Cost, latency, and output-token coordinates are retained as review metadata, not score-matrix metrics." }, { "score": 28.97, "reference_url": "https://openai.com/index/gpt-5-6/", "source_type": "official_blog", "reported_setting": { "effort": "xhigh", "eval_variant": null, "code_mode": "false" }, "notes": "OpenAI GPT-5.6 release figure (MedChemBench): GPT-5.6 Terra; MedChemBench (Internal)=28.97. Figure 28: MedChemBench. Cost, latency, and output-token coordinates are retained as review metadata, not score-matrix metrics." } ] }, { "model_id": "gpt-5.6-luna", "benchmark_id": "medchembench_internal", "score": 30.4, "reference_url": "https://openai.com/index/gpt-5-6/", "reported_setting": { "effort": "max", "eval_variant": null, "code_mode": "false" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-5.6 release table (table 2): GPT‑5.6 Luna; MedChemBench (Internal)=30.4. Named internal evaluation; task count not disclosed. Exact effort/variant resolved from the rendered chart point.", "candidates": [ { "score": 12.23, "reference_url": "https://openai.com/index/gpt-5-6/", "source_type": "official_blog", "reported_setting": { "effort": "low", "eval_variant": null, "code_mode": "false" }, "notes": "OpenAI GPT-5.6 release figure (MedChemBench): GPT-5.6 Luna; MedChemBench (Internal)=12.23. Figure 28: MedChemBench. Cost, latency, and output-token coordinates are retained as review metadata, not score-matrix metrics." }, { "score": 14.16, "reference_url": "https://openai.com/index/gpt-5-6/", "source_type": "official_blog", "reported_setting": { "effort": "medium", "eval_variant": null, "code_mode": "false" }, "notes": "OpenAI GPT-5.6 release figure (MedChemBench): GPT-5.6 Luna; MedChemBench (Internal)=14.16. Figure 28: MedChemBench. Cost, latency, and output-token coordinates are retained as review metadata, not score-matrix metrics." }, { "score": 22.86, "reference_url": "https://openai.com/index/gpt-5-6/", "source_type": "official_blog", "reported_setting": { "effort": "high", "eval_variant": null, "code_mode": "false" }, "notes": "OpenAI GPT-5.6 release figure (MedChemBench): GPT-5.6 Luna; MedChemBench (Internal)=22.86. Figure 28: MedChemBench. Cost, latency, and output-token coordinates are retained as review metadata, not score-matrix metrics." }, { "score": 27.62, "reference_url": "https://openai.com/index/gpt-5-6/", "source_type": "official_blog", "reported_setting": { "effort": "xhigh", "eval_variant": null, "code_mode": "false" }, "notes": "OpenAI GPT-5.6 release figure (MedChemBench): GPT-5.6 Luna; MedChemBench (Internal)=27.62. Figure 28: MedChemBench. Cost, latency, and output-token coordinates are retained as review metadata, not score-matrix metrics." } ] }, { "model_id": "gpt-5.5", "benchmark_id": "medchembench_internal", "score": 35.5, "reference_url": "https://openai.com/index/gpt-5-6/", "reported_setting": { "effort": "xhigh", "eval_variant": null, "code_mode": "false" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-5.6 release table (table 2): GPT‑5.5; MedChemBench (Internal)=35.5. Named internal evaluation; task count not disclosed. Exact effort/variant resolved from the rendered chart point.", "candidates": [ { "score": 19.17, "reference_url": "https://openai.com/index/gpt-5-6/", "source_type": "official_blog", "reported_setting": { "effort": "low", "eval_variant": null, "code_mode": "false" }, "notes": "OpenAI GPT-5.6 release figure (MedChemBench): GPT-5.5; MedChemBench (Internal)=19.17. Figure 28: MedChemBench. Cost, latency, and output-token coordinates are retained as review metadata, not score-matrix metrics." }, { "score": 25.6, "reference_url": "https://openai.com/index/gpt-5-6/", "source_type": "official_blog", "reported_setting": { "effort": "medium", "eval_variant": null, "code_mode": "false" }, "notes": "OpenAI GPT-5.6 release figure (MedChemBench): GPT-5.5; MedChemBench (Internal)=25.6. Figure 28: MedChemBench. Cost, latency, and output-token coordinates are retained as review metadata, not score-matrix metrics." }, { "score": 30.82, "reference_url": "https://openai.com/index/gpt-5-6/", "source_type": "official_blog", "reported_setting": { "effort": "high", "eval_variant": null, "code_mode": "false" }, "notes": "OpenAI GPT-5.6 release figure (MedChemBench): GPT-5.5; MedChemBench (Internal)=30.82. Figure 28: MedChemBench. Cost, latency, and output-token coordinates are retained as review metadata, not score-matrix metrics." } ] }, { "model_id": "gpt-5.6-sol", "benchmark_id": "healthbench_professional_length_adjusted", "score": 60.5, "reference_url": "https://openai.com/index/gpt-5-6/", "reported_setting": { "mode": "thinking", "effort": "source does not state", "tools": "none", "sampling": "source does not state", "judge": "benchmark-specified", "harness": "official release evaluation", "prompt_style": "source does not state", "temperature": "source does not state", "context": "source does not state" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-5.6 release table (table 2): GPT‑5.6 Sol; HealthBench Professional⁶=60.5. OpenAI's official length-adjusted scoring; not comparable to Anthropic system-card HealthBench Professional scores." }, { "model_id": "gpt-5.6-terra", "benchmark_id": "healthbench_professional_length_adjusted", "score": 57.7, "reference_url": "https://openai.com/index/gpt-5-6/", "reported_setting": { "mode": "thinking", "effort": "source does not state", "tools": "none", "sampling": "source does not state", "judge": "benchmark-specified", "harness": "official release evaluation", "prompt_style": "source does not state", "temperature": "source does not state", "context": "source does not state" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-5.6 release table (table 2): GPT‑5.6 Terra; HealthBench Professional⁶=57.7. OpenAI's official length-adjusted scoring; not comparable to Anthropic system-card HealthBench Professional scores." }, { "model_id": "gpt-5.6-luna", "benchmark_id": "healthbench_professional_length_adjusted", "score": 55.7, "reference_url": "https://openai.com/index/gpt-5-6/", "reported_setting": { "mode": "thinking", "effort": "source does not state", "tools": "none", "sampling": "source does not state", "judge": "benchmark-specified", "harness": "official release evaluation", "prompt_style": "source does not state", "temperature": "source does not state", "context": "source does not state" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-5.6 release table (table 2): GPT‑5.6 Luna; HealthBench Professional⁶=55.7. OpenAI's official length-adjusted scoring; not comparable to Anthropic system-card HealthBench Professional scores." }, { "model_id": "gpt-5.5", "benchmark_id": "healthbench_professional_length_adjusted", "score": 49.5, "reference_url": "https://openai.com/index/gpt-5-6/", "reported_setting": { "mode": "thinking", "effort": "source does not state", "tools": "none", "sampling": "source does not state", "judge": "benchmark-specified", "harness": "official release evaluation", "prompt_style": "source does not state", "temperature": "source does not state", "context": "source does not state" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-5.6 release table (table 2): GPT‑5.5; HealthBench Professional⁶=49.5. OpenAI's official length-adjusted scoring; not comparable to Anthropic system-card HealthBench Professional scores.", "candidates": [ { "score": 51.8, "reference_url": "https://deploymentsafety.openai.com/gpt-5-6", "source_type": "model_card", "reported_setting": { "effort": "source does not state", "tools": "none", "harness": "OpenAI official HealthBench scoring" }, "notes": "OpenAI GPT-5.6 System Card table (HealthBench length-adjusted scores): gpt-5.5; HealthBench Professional length-adjusted=51.8. System Card Table 6. Adjusted and unadjusted metrics are kept as distinct benchmark identities." } ] }, { "model_id": "claude-fable-5", "benchmark_id": "healthbench_professional_length_adjusted", "score": 60.9, "reference_url": "https://openai.com/index/gpt-5-6/", "reported_setting": { "mode": "thinking", "effort": "source does not state", "tools": "none", "sampling": "source does not state", "judge": "benchmark-specified", "harness": "official release evaluation", "prompt_style": "source does not state", "temperature": "source does not state", "context": "source does not state" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-5.6 release table (table 2): Claude Fable 5; HealthBench Professional⁶=60.9. OpenAI's official length-adjusted scoring; not comparable to Anthropic system-card HealthBench Professional scores.", "candidates": [ { "score": 66.0, "reference_url": "https://www.anthropic.com/claude-opus-5-system-card", "source_type": "model_card", "reported_setting": { "effort": "max", "harness": "official or cited leaderboard", "deployment": "claude-mythos-5" }, "notes": "Table 8.1.A / HealthBench Professional Length-Adjusted: Unless noted, Anthropic values use adaptive thinking at max effort, default sampling, mean over five trials. Table 8.1.A card summary." }, { "score": 66.0, "reference_url": "https://www.anthropic.com/claude-opus-5-system-card", "source_type": "model_card", "reported_setting": { "effort": "max", "tools": "none", "deployment": "claude-mythos-5" }, "notes": "Figure 8.15 / HealthBench Professional Length-Adjusted / Claude Mythos 5: Five trials; Opus 4.8 grader; no customized system prompt." } ] }, { "model_id": "claude-opus-4.8", "benchmark_id": "healthbench_professional_length_adjusted", "score": 53.0, "reference_url": "https://openai.com/index/gpt-5-6/", "reported_setting": { "mode": "thinking", "effort": "source does not state", "tools": "none", "sampling": "source does not state", "judge": "benchmark-specified", "harness": "official release evaluation", "prompt_style": "source does not state", "temperature": "source does not state", "context": "source does not state" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-5.6 release table (table 2): Claude Opus 4.8; HealthBench Professional⁶=53.0. OpenAI's official length-adjusted scoring; not comparable to Anthropic system-card HealthBench Professional scores.", "candidates": [ { "score": 57.4, "reference_url": "https://www.anthropic.com/claude-sonnet-5-system-card", "source_type": "model_card", "reported_setting": { "effort": "max", "tools": "none", "sampling": "avg 5 trials unless source states otherwise", "harness": "health" }, "notes": "Anthropic Claude Sonnet 5 System Card Figure 8.12.2.A, page 139: Claude Opus 4.8; HealthBench Professional [summary]=57.4%. Source setting: effort=max; tools=none; sampling=avg 5 trials unless source states otherwise; harness=health." }, { "score": 57.4, "reference_url": "https://www.anthropic.com/claude-opus-5-system-card", "source_type": "model_card", "reported_setting": { "effort": "max", "harness": "official or cited leaderboard" }, "notes": "Table 8.1.A / HealthBench Professional Length-Adjusted: Unless noted, Anthropic values use adaptive thinking at max effort, default sampling, mean over five trials. Table 8.1.A card summary." }, { "score": 57.4, "reference_url": "https://www.anthropic.com/claude-opus-5-system-card", "source_type": "model_card", "reported_setting": { "effort": "max", "tools": "none" }, "notes": "Figure 8.15 / HealthBench Professional Length-Adjusted / Claude Opus 4.8: Five trials; Opus 4.8 grader; no customized system prompt." }, { "score": 55.8, "reference_url": "https://ai.meta.com/static-resource/muse-spark-1.1-evaluation-report", "source_type": "tech_report", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "none", "sampling": "pass@1 unless source row states otherwise", "harness": "Meta MSL general-capability evaluation", "temperature": "1.0 for Muse; provider/source setting for comparisons", "context": "benchmark-specific", "configuration": "default" }, "notes": "Figure 44 · HealthBench Professional length-adjusted; metric=headline_metric. Exact printed value in the general-capability summary." } ] }, { "model_id": "gpt-5.6-terra", "benchmark_id": "osworld_2_0", "score": 50.2, "reference_url": "https://openai.com/index/gpt-5-6/", "reported_setting": { "effort": "max", "eval_variant": null, "code_mode": "false" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-5.6 release table (table 3): GPT‑5.6 Terra; OSWorld 2.0=50.2. Exact effort/variant resolved from the rendered chart point.", "candidates": [ { "score": 4.77, "reference_url": "https://openai.com/index/gpt-5-6/", "source_type": "official_blog", "reported_setting": { "effort": "none", "eval_variant": null, "code_mode": "false" }, "notes": "OpenAI GPT-5.6 release figure (OSWorld 2.0): GPT-5.6 Terra; OSWorld 2.0=4.77. Figure 15: OSWorld 2.0. Cost, latency, and output-token coordinates are retained as review metadata, not score-matrix metrics." }, { "score": 13.11, "reference_url": "https://openai.com/index/gpt-5-6/", "source_type": "official_blog", "reported_setting": { "effort": "low", "eval_variant": null, "code_mode": "false" }, "notes": "OpenAI GPT-5.6 release figure (OSWorld 2.0): GPT-5.6 Terra; OSWorld 2.0=13.11. Figure 15: OSWorld 2.0. Cost, latency, and output-token coordinates are retained as review metadata, not score-matrix metrics." }, { "score": 22.23, "reference_url": "https://openai.com/index/gpt-5-6/", "source_type": "official_blog", "reported_setting": { "effort": "medium", "eval_variant": null, "code_mode": "false" }, "notes": "OpenAI GPT-5.6 release figure (OSWorld 2.0): GPT-5.6 Terra; OSWorld 2.0=22.23. Figure 15: OSWorld 2.0. Cost, latency, and output-token coordinates are retained as review metadata, not score-matrix metrics." }, { "score": 36.61, "reference_url": "https://openai.com/index/gpt-5-6/", "source_type": "official_blog", "reported_setting": { "effort": "high", "eval_variant": null, "code_mode": "false" }, "notes": "OpenAI GPT-5.6 release figure (OSWorld 2.0): GPT-5.6 Terra; OSWorld 2.0=36.61. Figure 15: OSWorld 2.0. Cost, latency, and output-token coordinates are retained as review metadata, not score-matrix metrics." }, { "score": 46.1, "reference_url": "https://openai.com/index/gpt-5-6/", "source_type": "official_blog", "reported_setting": { "effort": "xhigh", "eval_variant": null, "code_mode": "false" }, "notes": "OpenAI GPT-5.6 release figure (OSWorld 2.0): GPT-5.6 Terra; OSWorld 2.0=46.1. Figure 15: OSWorld 2.0. Cost, latency, and output-token coordinates are retained as review metadata, not score-matrix metrics." } ] }, { "model_id": "gpt-5.6-luna", "benchmark_id": "osworld_2_0", "score": 45.6, "reference_url": "https://openai.com/index/gpt-5-6/", "reported_setting": { "effort": "max", "eval_variant": null, "code_mode": "false" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-5.6 release table (table 3): GPT‑5.6 Luna; OSWorld 2.0=45.6. Exact effort/variant resolved from the rendered chart point.", "candidates": [ { "score": 4.84, "reference_url": "https://openai.com/index/gpt-5-6/", "source_type": "official_blog", "reported_setting": { "effort": "none", "eval_variant": null, "code_mode": "false" }, "notes": "OpenAI GPT-5.6 release figure (OSWorld 2.0): GPT-5.6 Luna; OSWorld 2.0=4.84. Figure 15: OSWorld 2.0. Cost, latency, and output-token coordinates are retained as review metadata, not score-matrix metrics." }, { "score": 9.55, "reference_url": "https://openai.com/index/gpt-5-6/", "source_type": "official_blog", "reported_setting": { "effort": "low", "eval_variant": null, "code_mode": "false" }, "notes": "OpenAI GPT-5.6 release figure (OSWorld 2.0): GPT-5.6 Luna; OSWorld 2.0=9.55. Figure 15: OSWorld 2.0. Cost, latency, and output-token coordinates are retained as review metadata, not score-matrix metrics." }, { "score": 18.03, "reference_url": "https://openai.com/index/gpt-5-6/", "source_type": "official_blog", "reported_setting": { "effort": "medium", "eval_variant": null, "code_mode": "false" }, "notes": "OpenAI GPT-5.6 release figure (OSWorld 2.0): GPT-5.6 Luna; OSWorld 2.0=18.03. Figure 15: OSWorld 2.0. Cost, latency, and output-token coordinates are retained as review metadata, not score-matrix metrics." }, { "score": 33.04, "reference_url": "https://openai.com/index/gpt-5-6/", "source_type": "official_blog", "reported_setting": { "effort": "high", "eval_variant": null, "code_mode": "false" }, "notes": "OpenAI GPT-5.6 release figure (OSWorld 2.0): GPT-5.6 Luna; OSWorld 2.0=33.04. Figure 15: OSWorld 2.0. Cost, latency, and output-token coordinates are retained as review metadata, not score-matrix metrics." }, { "score": 41.83, "reference_url": "https://openai.com/index/gpt-5-6/", "source_type": "official_blog", "reported_setting": { "effort": "xhigh", "eval_variant": null, "code_mode": "false" }, "notes": "OpenAI GPT-5.6 release figure (OSWorld 2.0): GPT-5.6 Luna; OSWorld 2.0=41.83. Figure 15: OSWorld 2.0. Cost, latency, and output-token coordinates are retained as review metadata, not score-matrix metrics." } ] }, { "model_id": "gpt-5.6-terra", "benchmark_id": "browsecomp", "score": 87.5, "reference_url": "https://openai.com/index/gpt-5-6/", "reported_setting": { "effort": "max", "eval_variant": null, "code_mode": "false" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-5.6 release table (table 3): GPT‑5.6 Terra; BrowseComp=87.5. Exact effort/variant resolved from the rendered chart point.", "candidates": [ { "score": 50.47, "reference_url": "https://openai.com/index/gpt-5-6/", "source_type": "official_blog", "reported_setting": { "effort": "low", "eval_variant": null, "code_mode": "false" }, "notes": "OpenAI GPT-5.6 release figure (BrowseComp): GPT-5.6 Terra; BrowseComp=50.47. Figure 15: Agentic browsing tasks.. Cost, latency, and output-token coordinates are retained as review metadata, not score-matrix metrics." }, { "score": 67.38, "reference_url": "https://openai.com/index/gpt-5-6/", "source_type": "official_blog", "reported_setting": { "effort": "medium", "eval_variant": null, "code_mode": "false" }, "notes": "OpenAI GPT-5.6 release figure (BrowseComp): GPT-5.6 Terra; BrowseComp=67.38. Figure 15: Agentic browsing tasks.. Cost, latency, and output-token coordinates are retained as review metadata, not score-matrix metrics." }, { "score": 79.3, "reference_url": "https://openai.com/index/gpt-5-6/", "source_type": "official_blog", "reported_setting": { "effort": "high", "eval_variant": null, "code_mode": "false" }, "notes": "OpenAI GPT-5.6 release figure (BrowseComp): GPT-5.6 Terra; BrowseComp=79.3. Figure 15: Agentic browsing tasks.. Cost, latency, and output-token coordinates are retained as review metadata, not score-matrix metrics." }, { "score": 81.52, "reference_url": "https://openai.com/index/gpt-5-6/", "source_type": "official_blog", "reported_setting": { "effort": "xhigh", "eval_variant": null, "code_mode": "false" }, "notes": "OpenAI GPT-5.6 release figure (BrowseComp): GPT-5.6 Terra; BrowseComp=81.52. Figure 15: Agentic browsing tasks.. Cost, latency, and output-token coordinates are retained as review metadata, not score-matrix metrics." } ] }, { "model_id": "gpt-5.6-luna", "benchmark_id": "browsecomp", "score": 83.3, "reference_url": "https://openai.com/index/gpt-5-6/", "reported_setting": { "effort": "max", "eval_variant": null, "code_mode": "false" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-5.6 release table (table 3): GPT‑5.6 Luna; BrowseComp=83.3. Exact effort/variant resolved from the rendered chart point.", "candidates": [ { "score": 45.73, "reference_url": "https://openai.com/index/gpt-5-6/", "source_type": "official_blog", "reported_setting": { "effort": "low", "eval_variant": null, "code_mode": "false" }, "notes": "OpenAI GPT-5.6 release figure (BrowseComp): GPT-5.6 Luna; BrowseComp=45.73. Figure 15: Agentic browsing tasks.. Cost, latency, and output-token coordinates are retained as review metadata, not score-matrix metrics." }, { "score": 58.85, "reference_url": "https://openai.com/index/gpt-5-6/", "source_type": "official_blog", "reported_setting": { "effort": "medium", "eval_variant": null, "code_mode": "false" }, "notes": "OpenAI GPT-5.6 release figure (BrowseComp): GPT-5.6 Luna; BrowseComp=58.85. Figure 15: Agentic browsing tasks.. Cost, latency, and output-token coordinates are retained as review metadata, not score-matrix metrics." }, { "score": 80.02, "reference_url": "https://openai.com/index/gpt-5-6/", "source_type": "official_blog", "reported_setting": { "effort": "high", "eval_variant": null, "code_mode": "false" }, "notes": "OpenAI GPT-5.6 release figure (BrowseComp): GPT-5.6 Luna; BrowseComp=80.02. Figure 15: Agentic browsing tasks.. Cost, latency, and output-token coordinates are retained as review metadata, not score-matrix metrics." }, { "score": 84.04, "reference_url": "https://openai.com/index/gpt-5-6/", "source_type": "official_blog", "reported_setting": { "effort": "xhigh", "eval_variant": null, "code_mode": "false" }, "notes": "OpenAI GPT-5.6 release figure (BrowseComp): GPT-5.6 Luna; BrowseComp=84.04. Figure 15: Agentic browsing tasks.. Cost, latency, and output-token coordinates are retained as review metadata, not score-matrix metrics." } ] }, { "model_id": "gpt-5.6-sol", "benchmark_id": "benchcad", "score": 70.6, "reference_url": "https://openai.com/index/gpt-5-6/", "reported_setting": { "mode": "thinking", "effort": "source does not state", "tools": "none", "sampling": "source does not state", "judge": "benchmark-specified", "harness": "official release evaluation", "prompt_style": "source does not state", "temperature": "source does not state", "context": "source does not state" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-5.6 release table (table 3): GPT‑5.6 Sol; BenchCAD=70.6.", "candidates": [ { "score": 83.4, "reference_url": "https://openai.com/index/gpt-5-6/", "source_type": "official_blog", "reported_setting": { "mode": "thinking", "effort": "source does not state", "tools": "python", "sampling": "source does not state", "judge": "benchmark-specified", "harness": "official release evaluation", "prompt_style": "source does not state", "temperature": "source does not state", "context": "source does not state" }, "notes": "OpenAI GPT-5.6 release table (table 3): GPT‑5.6 Sol; BenchCAD (python tool)=83.4." } ] }, { "model_id": "gpt-5.6-terra", "benchmark_id": "benchcad", "score": 62.3, "reference_url": "https://openai.com/index/gpt-5-6/", "reported_setting": { "mode": "thinking", "effort": "source does not state", "tools": "none", "sampling": "source does not state", "judge": "benchmark-specified", "harness": "official release evaluation", "prompt_style": "source does not state", "temperature": "source does not state", "context": "source does not state" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-5.6 release table (table 3): GPT‑5.6 Terra; BenchCAD=62.3.", "candidates": [ { "score": 78.2, "reference_url": "https://openai.com/index/gpt-5-6/", "source_type": "official_blog", "reported_setting": { "mode": "thinking", "effort": "source does not state", "tools": "python", "sampling": "source does not state", "judge": "benchmark-specified", "harness": "official release evaluation", "prompt_style": "source does not state", "temperature": "source does not state", "context": "source does not state" }, "notes": "OpenAI GPT-5.6 release table (table 3): GPT‑5.6 Terra; BenchCAD (python tool)=78.2." } ] }, { "model_id": "gpt-5.6-luna", "benchmark_id": "benchcad", "score": 63.1, "reference_url": "https://openai.com/index/gpt-5-6/", "reported_setting": { "mode": "thinking", "effort": "source does not state", "tools": "none", "sampling": "source does not state", "judge": "benchmark-specified", "harness": "official release evaluation", "prompt_style": "source does not state", "temperature": "source does not state", "context": "source does not state" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-5.6 release table (table 3): GPT‑5.6 Luna; BenchCAD=63.1.", "candidates": [ { "score": 73.9, "reference_url": "https://openai.com/index/gpt-5-6/", "source_type": "official_blog", "reported_setting": { "mode": "thinking", "effort": "source does not state", "tools": "python", "sampling": "source does not state", "judge": "benchmark-specified", "harness": "official release evaluation", "prompt_style": "source does not state", "temperature": "source does not state", "context": "source does not state" }, "notes": "OpenAI GPT-5.6 release table (table 3): GPT‑5.6 Luna; BenchCAD (python tool)=73.9." } ] }, { "model_id": "gpt-5.5", "benchmark_id": "benchcad", "score": 44.4, "reference_url": "https://openai.com/index/gpt-5-6/", "reported_setting": { "mode": "thinking", "effort": "source does not state", "tools": "none", "sampling": "source does not state", "judge": "benchmark-specified", "harness": "official release evaluation", "prompt_style": "source does not state", "temperature": "source does not state", "context": "source does not state" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-5.6 release table (table 3): GPT‑5.5; BenchCAD=44.4.", "candidates": [ { "score": 55.8, "reference_url": "https://openai.com/index/gpt-5-6/", "source_type": "official_blog", "reported_setting": { "mode": "thinking", "effort": "source does not state", "tools": "python", "sampling": "source does not state", "judge": "benchmark-specified", "harness": "official release evaluation", "prompt_style": "source does not state", "temperature": "source does not state", "context": "source does not state" }, "notes": "OpenAI GPT-5.6 release table (table 3): GPT‑5.5; BenchCAD (python tool)=55.8." } ] }, { "model_id": "claude-fable-5", "benchmark_id": "benchcad", "score": 38.4, "reference_url": "https://openai.com/index/gpt-5-6/", "reported_setting": { "mode": "thinking", "effort": "source does not state", "tools": "none", "sampling": "source does not state", "judge": "benchmark-specified", "harness": "official release evaluation", "prompt_style": "source does not state", "temperature": "source does not state", "context": "source does not state", "deployment": "claude-mythos-5", "safeguards": "lifted for Project Glasswing", "notes": "Anthropic documents Mythos 5 as the same underlying model and capabilities as Fable 5, without Fable's safety classifiers." }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-5.6 release table (table 3): Claude Mythos 5; BenchCAD=38.4.", "candidates": [ { "score": 65.0, "reference_url": "https://openai.com/index/gpt-5-6/", "source_type": "official_blog", "reported_setting": { "mode": "thinking", "effort": "source does not state", "tools": "python", "sampling": "source does not state", "judge": "benchmark-specified", "harness": "official release evaluation", "prompt_style": "source does not state", "temperature": "source does not state", "context": "source does not state", "deployment": "claude-mythos-5", "safeguards": "lifted for Project Glasswing", "notes": "Anthropic documents Mythos 5 as the same underlying model and capabilities as Fable 5, without Fable's safety classifiers." }, "notes": "OpenAI GPT-5.6 release table (table 3): Claude Mythos 5; BenchCAD (python tool)=65.0." } ] }, { "model_id": "claude-mythos", "benchmark_id": "benchcad", "score": 35.5, "reference_url": "https://openai.com/index/gpt-5-6/", "reported_setting": { "mode": "thinking", "effort": "source does not state", "tools": "none", "sampling": "source does not state", "judge": "benchmark-specified", "harness": "official release evaluation", "prompt_style": "source does not state", "temperature": "source does not state", "context": "source does not state" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-5.6 release table (table 3): Claude Mythos Preview; BenchCAD=35.5.", "candidates": [ { "score": 61.0, "reference_url": "https://openai.com/index/gpt-5-6/", "source_type": "official_blog", "reported_setting": { "mode": "thinking", "effort": "source does not state", "tools": "python", "sampling": "source does not state", "judge": "benchmark-specified", "harness": "official release evaluation", "prompt_style": "source does not state", "temperature": "source does not state", "context": "source does not state" }, "notes": "OpenAI GPT-5.6 release table (table 3): Claude Mythos Preview; BenchCAD (python tool)=61.0." } ] }, { "model_id": "claude-opus-4.8", "benchmark_id": "benchcad", "score": 27.3, "reference_url": "https://openai.com/index/gpt-5-6/", "reported_setting": { "mode": "thinking", "effort": "source does not state", "tools": "none", "sampling": "source does not state", "judge": "benchmark-specified", "harness": "official release evaluation", "prompt_style": "source does not state", "temperature": "source does not state", "context": "source does not state" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-5.6 release table (table 3): Claude Opus 4.8; BenchCAD=27.3.", "candidates": [ { "score": 51.8, "reference_url": "https://openai.com/index/gpt-5-6/", "source_type": "official_blog", "reported_setting": { "mode": "thinking", "effort": "source does not state", "tools": "python", "sampling": "source does not state", "judge": "benchmark-specified", "harness": "official release evaluation", "prompt_style": "source does not state", "temperature": "source does not state", "context": "source does not state" }, "notes": "OpenAI GPT-5.6 release table (table 3): Claude Opus 4.8; BenchCAD (python tool)=51.8." } ] }, { "model_id": "gpt-5.6-sol", "benchmark_id": "ctf_internal", "score": 96.67, "reference_url": "https://openai.com/index/gpt-5-6/", "reported_setting": { "effort": "max", "eval_variant": null, "code_mode": "false" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-5.6 release table (table 4): GPT‑5.6 Sol; Capture-the-Flag Challenges=96.67. System Card identifies the release-page row as Capture the Flag (Internal). Exact effort/variant resolved from the rendered chart point. Stored at the System Card figure's exact printed precision; release table displays 96.7%.", "candidates": [ { "score": 42.78, "reference_url": "https://openai.com/index/gpt-5-6/", "source_type": "official_blog", "reported_setting": { "effort": "low", "eval_variant": null, "code_mode": "false" }, "notes": "OpenAI GPT-5.6 release figure (Capture-the-Flag Challenges): GPT-5.6 Sol; Capture-the-Flag Challenges=42.78. Figure 27: Capture-the-Flag Challenges. Cost, latency, and output-token coordinates are retained as review metadata, not score-matrix metrics." }, { "score": 72.18, "reference_url": "https://openai.com/index/gpt-5-6/", "source_type": "official_blog", "reported_setting": { "effort": "medium", "eval_variant": null, "code_mode": "false" }, "notes": "OpenAI GPT-5.6 release figure (Capture-the-Flag Challenges): GPT-5.6 Sol; Capture-the-Flag Challenges=72.18. Figure 27: Capture-the-Flag Challenges. Cost, latency, and output-token coordinates are retained as review metadata, not score-matrix metrics." }, { "score": 86.85, "reference_url": "https://openai.com/index/gpt-5-6/", "source_type": "official_blog", "reported_setting": { "effort": "high", "eval_variant": null, "code_mode": "false" }, "notes": "OpenAI GPT-5.6 release figure (Capture-the-Flag Challenges): GPT-5.6 Sol; Capture-the-Flag Challenges=86.85. Figure 27: Capture-the-Flag Challenges. Cost, latency, and output-token coordinates are retained as review metadata, not score-matrix metrics." }, { "score": 91.76, "reference_url": "https://openai.com/index/gpt-5-6/", "source_type": "official_blog", "reported_setting": { "effort": "xhigh", "eval_variant": null, "code_mode": "false" }, "notes": "OpenAI GPT-5.6 release figure (Capture-the-Flag Challenges): GPT-5.6 Sol; Capture-the-Flag Challenges=91.76. Figure 27: Capture-the-Flag Challenges. Cost, latency, and output-token coordinates are retained as review metadata, not score-matrix metrics." } ] }, { "model_id": "gpt-5.6-terra", "benchmark_id": "ctf_internal", "score": 91.84, "reference_url": "https://openai.com/index/gpt-5-6/", "reported_setting": { "effort": "max", "eval_variant": null, "code_mode": "false" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-5.6 release table (table 4): GPT‑5.6 Terra; Capture-the-Flag Challenges=91.84. System Card identifies the release-page row as Capture the Flag (Internal). Exact effort/variant resolved from the rendered chart point. Stored at the System Card figure's exact printed precision; release table displays 91.8%.", "candidates": [ { "score": 20.86, "reference_url": "https://openai.com/index/gpt-5-6/", "source_type": "official_blog", "reported_setting": { "effort": "low", "eval_variant": null, "code_mode": "false" }, "notes": "OpenAI GPT-5.6 release figure (Capture-the-Flag Challenges): GPT-5.6 Terra; Capture-the-Flag Challenges=20.86. Figure 27: Capture-the-Flag Challenges. Cost, latency, and output-token coordinates are retained as review metadata, not score-matrix metrics." }, { "score": 42.33, "reference_url": "https://openai.com/index/gpt-5-6/", "source_type": "official_blog", "reported_setting": { "effort": "medium", "eval_variant": null, "code_mode": "false" }, "notes": "OpenAI GPT-5.6 release figure (Capture-the-Flag Challenges): GPT-5.6 Terra; Capture-the-Flag Challenges=42.33. Figure 27: Capture-the-Flag Challenges. Cost, latency, and output-token coordinates are retained as review metadata, not score-matrix metrics." }, { "score": 70.45, "reference_url": "https://openai.com/index/gpt-5-6/", "source_type": "official_blog", "reported_setting": { "effort": "high", "eval_variant": null, "code_mode": "false" }, "notes": "OpenAI GPT-5.6 release figure (Capture-the-Flag Challenges): GPT-5.6 Terra; Capture-the-Flag Challenges=70.45. Figure 27: Capture-the-Flag Challenges. Cost, latency, and output-token coordinates are retained as review metadata, not score-matrix metrics." }, { "score": 80.5, "reference_url": "https://openai.com/index/gpt-5-6/", "source_type": "official_blog", "reported_setting": { "effort": "xhigh", "eval_variant": null, "code_mode": "false" }, "notes": "OpenAI GPT-5.6 release figure (Capture-the-Flag Challenges): GPT-5.6 Terra; Capture-the-Flag Challenges=80.5. Figure 27: Capture-the-Flag Challenges. Cost, latency, and output-token coordinates are retained as review metadata, not score-matrix metrics." } ] }, { "model_id": "gpt-5.6-luna", "benchmark_id": "ctf_internal", "score": 85.19, "reference_url": "https://openai.com/index/gpt-5-6/", "reported_setting": { "effort": "max", "eval_variant": null, "code_mode": "false" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-5.6 release table (table 4): GPT‑5.6 Luna; Capture-the-Flag Challenges=85.19. System Card identifies the release-page row as Capture the Flag (Internal). Exact effort/variant resolved from the rendered chart point. Stored at the System Card figure's exact printed precision; release table displays 85.2%.", "candidates": [ { "score": 10.73, "reference_url": "https://openai.com/index/gpt-5-6/", "source_type": "official_blog", "reported_setting": { "effort": "low", "eval_variant": null, "code_mode": "false" }, "notes": "OpenAI GPT-5.6 release figure (Capture-the-Flag Challenges): GPT-5.6 Luna; Capture-the-Flag Challenges=10.73. Figure 27: Capture-the-Flag Challenges. Cost, latency, and output-token coordinates are retained as review metadata, not score-matrix metrics." }, { "score": 21.32, "reference_url": "https://openai.com/index/gpt-5-6/", "source_type": "official_blog", "reported_setting": { "effort": "medium", "eval_variant": null, "code_mode": "false" }, "notes": "OpenAI GPT-5.6 release figure (Capture-the-Flag Challenges): GPT-5.6 Luna; Capture-the-Flag Challenges=21.32. Figure 27: Capture-the-Flag Challenges. Cost, latency, and output-token coordinates are retained as review metadata, not score-matrix metrics." }, { "score": 49.58, "reference_url": "https://openai.com/index/gpt-5-6/", "source_type": "official_blog", "reported_setting": { "effort": "high", "eval_variant": null, "code_mode": "false" }, "notes": "OpenAI GPT-5.6 release figure (Capture-the-Flag Challenges): GPT-5.6 Luna; Capture-the-Flag Challenges=49.58. Figure 27: Capture-the-Flag Challenges. Cost, latency, and output-token coordinates are retained as review metadata, not score-matrix metrics." }, { "score": 76.04, "reference_url": "https://openai.com/index/gpt-5-6/", "source_type": "official_blog", "reported_setting": { "effort": "xhigh", "eval_variant": null, "code_mode": "false" }, "notes": "OpenAI GPT-5.6 release figure (Capture-the-Flag Challenges): GPT-5.6 Luna; Capture-the-Flag Challenges=76.04. Figure 27: Capture-the-Flag Challenges. Cost, latency, and output-token coordinates are retained as review metadata, not score-matrix metrics." } ] }, { "model_id": "gpt-5.6-sol", "benchmark_id": "sec_bench_pro", "score": 71.2, "reference_url": "https://openai.com/index/gpt-5-6/", "reported_setting": { "effort": "max", "eval_variant": null, "code_mode": "false" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-5.6 release table (table 4): GPT‑5.6 Sol; SEC-Bench Pro=71.2. Exact effort/variant resolved from the rendered chart point.", "candidates": [ { "score": 74.3, "reference_url": "https://openai.com/index/gpt-5-6/", "source_type": "official_blog", "reported_setting": { "effort": "max", "eval_variant": "4 agents", "code_mode": "true" }, "notes": "OpenAI GPT-5.6 release table (table 4): GPT‑5.6 Sol Ultra; SEC-Bench Pro=74.3. Exact effort/variant resolved from the rendered chart point." }, { "score": 17.08, "reference_url": "https://openai.com/index/gpt-5-6/", "source_type": "official_blog", "reported_setting": { "effort": "low", "eval_variant": "1 agent", "code_mode": "true" }, "notes": "OpenAI GPT-5.6 release figure (SEC-Bench Pro (Multi-Agent)): GPT-5.6 Sol · 1 agent; SEC-Bench Pro=17.08. Figure 2: Score-latency/cost/token frontier for 1, 4, and 16 Sol agents.. Cost, latency, and output-token coordinates are retained as review metadata, not score-matrix metrics." }, { "score": 37.3, "reference_url": "https://openai.com/index/gpt-5-6/", "source_type": "official_blog", "reported_setting": { "effort": "medium", "eval_variant": "1 agent", "code_mode": "true" }, "notes": "OpenAI GPT-5.6 release figure (SEC-Bench Pro (Multi-Agent)): GPT-5.6 Sol · 1 agent; SEC-Bench Pro=37.3. Figure 2: Score-latency/cost/token frontier for 1, 4, and 16 Sol agents.. Cost, latency, and output-token coordinates are retained as review metadata, not score-matrix metrics." }, { "score": 53.14, "reference_url": "https://openai.com/index/gpt-5-6/", "source_type": "official_blog", "reported_setting": { "effort": "high", "eval_variant": "1 agent", "code_mode": "true" }, "notes": "OpenAI GPT-5.6 release figure (SEC-Bench Pro (Multi-Agent)): GPT-5.6 Sol · 1 agent; SEC-Bench Pro=53.14. Figure 2: Score-latency/cost/token frontier for 1, 4, and 16 Sol agents.. Cost, latency, and output-token coordinates are retained as review metadata, not score-matrix metrics." }, { "score": 64.07, "reference_url": "https://openai.com/index/gpt-5-6/", "source_type": "official_blog", "reported_setting": { "effort": "xhigh", "eval_variant": "1 agent", "code_mode": "true" }, "notes": "OpenAI GPT-5.6 release figure (SEC-Bench Pro (Multi-Agent)): GPT-5.6 Sol · 1 agent; SEC-Bench Pro=64.07. Figure 2: Score-latency/cost/token frontier for 1, 4, and 16 Sol agents.. Cost, latency, and output-token coordinates are retained as review metadata, not score-matrix metrics." }, { "score": 71.45, "reference_url": "https://openai.com/index/gpt-5-6/", "source_type": "official_blog", "reported_setting": { "effort": "max", "eval_variant": "1 agent", "code_mode": "true" }, "notes": "OpenAI GPT-5.6 release figure (SEC-Bench Pro (Multi-Agent)): GPT-5.6 Sol · 1 agent; SEC-Bench Pro=71.45. Figure 2: Score-latency/cost/token frontier for 1, 4, and 16 Sol agents.. Cost, latency, and output-token coordinates are retained as review metadata, not score-matrix metrics." }, { "score": 20.36, "reference_url": "https://openai.com/index/gpt-5-6/", "source_type": "official_blog", "reported_setting": { "effort": "low", "eval_variant": "4 agents", "code_mode": "true" }, "notes": "OpenAI GPT-5.6 release figure (SEC-Bench Pro (Multi-Agent)): GPT-5.6 Sol · 4 agents; SEC-Bench Pro=20.36. Figure 2: Score-latency/cost/token frontier for 1, 4, and 16 Sol agents.. Cost, latency, and output-token coordinates are retained as review metadata, not score-matrix metrics." }, { "score": 38.93, "reference_url": "https://openai.com/index/gpt-5-6/", "source_type": "official_blog", "reported_setting": { "effort": "medium", "eval_variant": "4 agents", "code_mode": "true" }, "notes": "OpenAI GPT-5.6 release figure (SEC-Bench Pro (Multi-Agent)): GPT-5.6 Sol · 4 agents; SEC-Bench Pro=38.93. Figure 2: Score-latency/cost/token frontier for 1, 4, and 16 Sol agents.. Cost, latency, and output-token coordinates are retained as review metadata, not score-matrix metrics." }, { "score": 50.14, "reference_url": "https://openai.com/index/gpt-5-6/", "source_type": "official_blog", "reported_setting": { "effort": "high", "eval_variant": "4 agents", "code_mode": "true" }, "notes": "OpenAI GPT-5.6 release figure (SEC-Bench Pro (Multi-Agent)): GPT-5.6 Sol · 4 agents; SEC-Bench Pro=50.14. Figure 2: Score-latency/cost/token frontier for 1, 4, and 16 Sol agents.. Cost, latency, and output-token coordinates are retained as review metadata, not score-matrix metrics." }, { "score": 65.3, "reference_url": "https://openai.com/index/gpt-5-6/", "source_type": "official_blog", "reported_setting": { "effort": "xhigh", "eval_variant": "4 agents", "code_mode": "true" }, "notes": "OpenAI GPT-5.6 release figure (SEC-Bench Pro (Multi-Agent)): GPT-5.6 Sol · 4 agents; SEC-Bench Pro=65.3. Figure 2: Score-latency/cost/token frontier for 1, 4, and 16 Sol agents.. Cost, latency, and output-token coordinates are retained as review metadata, not score-matrix metrics." }, { "score": 22.4, "reference_url": "https://openai.com/index/gpt-5-6/", "source_type": "official_blog", "reported_setting": { "effort": "low", "eval_variant": "16 agents", "code_mode": "true" }, "notes": "OpenAI GPT-5.6 release figure (SEC-Bench Pro (Multi-Agent)): GPT-5.6 Sol · 16 agents; SEC-Bench Pro=22.4. Figure 2: Score-latency/cost/token frontier for 1, 4, and 16 Sol agents.. Cost, latency, and output-token coordinates are retained as review metadata, not score-matrix metrics." }, { "score": 45.63, "reference_url": "https://openai.com/index/gpt-5-6/", "source_type": "official_blog", "reported_setting": { "effort": "medium", "eval_variant": "16 agents", "code_mode": "true" }, "notes": "OpenAI GPT-5.6 release figure (SEC-Bench Pro (Multi-Agent)): GPT-5.6 Sol · 16 agents; SEC-Bench Pro=45.63. Figure 2: Score-latency/cost/token frontier for 1, 4, and 16 Sol agents.. Cost, latency, and output-token coordinates are retained as review metadata, not score-matrix metrics." }, { "score": 61.48, "reference_url": "https://openai.com/index/gpt-5-6/", "source_type": "official_blog", "reported_setting": { "effort": "high", "eval_variant": "16 agents", "code_mode": "true" }, "notes": "OpenAI GPT-5.6 release figure (SEC-Bench Pro (Multi-Agent)): GPT-5.6 Sol · 16 agents; SEC-Bench Pro=61.48. Figure 2: Score-latency/cost/token frontier for 1, 4, and 16 Sol agents.. Cost, latency, and output-token coordinates are retained as review metadata, not score-matrix metrics." }, { "score": 66.67, "reference_url": "https://openai.com/index/gpt-5-6/", "source_type": "official_blog", "reported_setting": { "effort": "xhigh", "eval_variant": "16 agents", "code_mode": "true" }, "notes": "OpenAI GPT-5.6 release figure (SEC-Bench Pro (Multi-Agent)): GPT-5.6 Sol · 16 agents; SEC-Bench Pro=66.67. Figure 2: Score-latency/cost/token frontier for 1, 4, and 16 Sol agents.. Cost, latency, and output-token coordinates are retained as review metadata, not score-matrix metrics." }, { "score": 76.23, "reference_url": "https://openai.com/index/gpt-5-6/", "source_type": "official_blog", "reported_setting": { "effort": "max", "eval_variant": "16 agents", "code_mode": "true" }, "notes": "OpenAI GPT-5.6 release figure (SEC-Bench Pro (Multi-Agent)): GPT-5.6 Sol · 16 agents; SEC-Bench Pro=76.23. Figure 2: Score-latency/cost/token frontier for 1, 4, and 16 Sol agents.. Cost, latency, and output-token coordinates are retained as review metadata, not score-matrix metrics." }, { "score": 5.74, "reference_url": "https://openai.com/index/gpt-5-6/", "source_type": "official_blog", "reported_setting": { "effort": "none", "eval_variant": null, "code_mode": "false" }, "notes": "OpenAI GPT-5.6 release figure (SEC-Bench Pro): GPT-5.6 Sol; SEC-Bench Pro=5.74. Figure 27: SEC-Bench Pro. Cost, latency, and output-token coordinates are retained as review metadata, not score-matrix metrics." }, { "score": 11.61, "reference_url": "https://openai.com/index/gpt-5-6/", "source_type": "official_blog", "reported_setting": { "effort": "low", "eval_variant": null, "code_mode": "false" }, "notes": "OpenAI GPT-5.6 release figure (SEC-Bench Pro): GPT-5.6 Sol; SEC-Bench Pro=11.61. Figure 27: SEC-Bench Pro. Cost, latency, and output-token coordinates are retained as review metadata, not score-matrix metrics." }, { "score": 32.92, "reference_url": "https://openai.com/index/gpt-5-6/", "source_type": "official_blog", "reported_setting": { "effort": "medium", "eval_variant": null, "code_mode": "false" }, "notes": "OpenAI GPT-5.6 release figure (SEC-Bench Pro): GPT-5.6 Sol; SEC-Bench Pro=32.92. Figure 27: SEC-Bench Pro. Cost, latency, and output-token coordinates are retained as review metadata, not score-matrix metrics." }, { "score": 45.36, "reference_url": "https://openai.com/index/gpt-5-6/", "source_type": "official_blog", "reported_setting": { "effort": "high", "eval_variant": null, "code_mode": "false" }, "notes": "OpenAI GPT-5.6 release figure (SEC-Bench Pro): GPT-5.6 Sol; SEC-Bench Pro=45.36. Figure 27: SEC-Bench Pro. Cost, latency, and output-token coordinates are retained as review metadata, not score-matrix metrics." }, { "score": 65.44, "reference_url": "https://openai.com/index/gpt-5-6/", "source_type": "official_blog", "reported_setting": { "effort": "xhigh", "eval_variant": null, "code_mode": "false" }, "notes": "OpenAI GPT-5.6 release figure (SEC-Bench Pro): GPT-5.6 Sol; SEC-Bench Pro=65.44. Figure 27: SEC-Bench Pro. Cost, latency, and output-token coordinates are retained as review metadata, not score-matrix metrics." } ] }, { "model_id": "gpt-5.6-terra", "benchmark_id": "sec_bench_pro", "score": 57.7, "reference_url": "https://openai.com/index/gpt-5-6/", "reported_setting": { "effort": "max", "eval_variant": null, "code_mode": "false" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-5.6 release table (table 4): GPT‑5.6 Terra; SEC-Bench Pro=57.7. Exact effort/variant resolved from the rendered chart point.", "candidates": [ { "score": 1.23, "reference_url": "https://openai.com/index/gpt-5-6/", "source_type": "official_blog", "reported_setting": { "effort": "none", "eval_variant": null, "code_mode": "false" }, "notes": "OpenAI GPT-5.6 release figure (SEC-Bench Pro): GPT-5.6 Terra; SEC-Bench Pro=1.23. Figure 27: SEC-Bench Pro. Cost, latency, and output-token coordinates are retained as review metadata, not score-matrix metrics." }, { "score": 5.46, "reference_url": "https://openai.com/index/gpt-5-6/", "source_type": "official_blog", "reported_setting": { "effort": "low", "eval_variant": null, "code_mode": "false" }, "notes": "OpenAI GPT-5.6 release figure (SEC-Bench Pro): GPT-5.6 Terra; SEC-Bench Pro=5.46. Figure 27: SEC-Bench Pro. Cost, latency, and output-token coordinates are retained as review metadata, not score-matrix metrics." }, { "score": 15.85, "reference_url": "https://openai.com/index/gpt-5-6/", "source_type": "official_blog", "reported_setting": { "effort": "medium", "eval_variant": null, "code_mode": "false" }, "notes": "OpenAI GPT-5.6 release figure (SEC-Bench Pro): GPT-5.6 Terra; SEC-Bench Pro=15.85. Figure 27: SEC-Bench Pro. Cost, latency, and output-token coordinates are retained as review metadata, not score-matrix metrics." }, { "score": 30.05, "reference_url": "https://openai.com/index/gpt-5-6/", "source_type": "official_blog", "reported_setting": { "effort": "high", "eval_variant": null, "code_mode": "false" }, "notes": "OpenAI GPT-5.6 release figure (SEC-Bench Pro): GPT-5.6 Terra; SEC-Bench Pro=30.05. Figure 27: SEC-Bench Pro. Cost, latency, and output-token coordinates are retained as review metadata, not score-matrix metrics." }, { "score": 42.08, "reference_url": "https://openai.com/index/gpt-5-6/", "source_type": "official_blog", "reported_setting": { "effort": "xhigh", "eval_variant": null, "code_mode": "false" }, "notes": "OpenAI GPT-5.6 release figure (SEC-Bench Pro): GPT-5.6 Terra; SEC-Bench Pro=42.08. Figure 27: SEC-Bench Pro. Cost, latency, and output-token coordinates are retained as review metadata, not score-matrix metrics." } ] }, { "model_id": "gpt-5.6-luna", "benchmark_id": "sec_bench_pro", "score": 48.9, "reference_url": "https://openai.com/index/gpt-5-6/", "reported_setting": { "effort": "max", "eval_variant": null, "code_mode": "false" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-5.6 release table (table 4): GPT‑5.6 Luna; SEC-Bench Pro=48.9. Exact effort/variant resolved from the rendered chart point.", "candidates": [ { "score": 0.0, "reference_url": "https://openai.com/index/gpt-5-6/", "source_type": "official_blog", "reported_setting": { "effort": "none", "eval_variant": null, "code_mode": "false" }, "notes": "OpenAI GPT-5.6 release figure (SEC-Bench Pro): GPT-5.6 Luna; SEC-Bench Pro=0.0. Figure 27: SEC-Bench Pro. Cost, latency, and output-token coordinates are retained as review metadata, not score-matrix metrics." }, { "score": 1.37, "reference_url": "https://openai.com/index/gpt-5-6/", "source_type": "official_blog", "reported_setting": { "effort": "low", "eval_variant": null, "code_mode": "false" }, "notes": "OpenAI GPT-5.6 release figure (SEC-Bench Pro): GPT-5.6 Luna; SEC-Bench Pro=1.37. Figure 27: SEC-Bench Pro. Cost, latency, and output-token coordinates are retained as review metadata, not score-matrix metrics." }, { "score": 5.6, "reference_url": "https://openai.com/index/gpt-5-6/", "source_type": "official_blog", "reported_setting": { "effort": "medium", "eval_variant": null, "code_mode": "false" }, "notes": "OpenAI GPT-5.6 release figure (SEC-Bench Pro): GPT-5.6 Luna; SEC-Bench Pro=5.6. Figure 27: SEC-Bench Pro. Cost, latency, and output-token coordinates are retained as review metadata, not score-matrix metrics." }, { "score": 16.8, "reference_url": "https://openai.com/index/gpt-5-6/", "source_type": "official_blog", "reported_setting": { "effort": "high", "eval_variant": null, "code_mode": "false" }, "notes": "OpenAI GPT-5.6 release figure (SEC-Bench Pro): GPT-5.6 Luna; SEC-Bench Pro=16.8. Figure 27: SEC-Bench Pro. Cost, latency, and output-token coordinates are retained as review metadata, not score-matrix metrics." }, { "score": 34.02, "reference_url": "https://openai.com/index/gpt-5-6/", "source_type": "official_blog", "reported_setting": { "effort": "xhigh", "eval_variant": null, "code_mode": "false" }, "notes": "OpenAI GPT-5.6 release figure (SEC-Bench Pro): GPT-5.6 Luna; SEC-Bench Pro=34.02. Figure 27: SEC-Bench Pro. Cost, latency, and output-token coordinates are retained as review metadata, not score-matrix metrics." } ] }, { "model_id": "gpt-5.5", "benchmark_id": "sec_bench_pro", "score": 45.8, "reference_url": "https://openai.com/index/gpt-5-6/", "reported_setting": { "effort": "xhigh", "eval_variant": null, "code_mode": "false" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-5.6 release table (table 4): GPT‑5.5; SEC-Bench Pro=45.8. Exact effort/variant resolved from the rendered chart point.", "candidates": [ { "score": 1.5, "reference_url": "https://openai.com/index/gpt-5-6/", "source_type": "official_blog", "reported_setting": { "effort": "none", "eval_variant": null, "code_mode": "false" }, "notes": "OpenAI GPT-5.6 release figure (SEC-Bench Pro): GPT-5.5; SEC-Bench Pro=1.5. Figure 27: SEC-Bench Pro. Cost, latency, and output-token coordinates are retained as review metadata, not score-matrix metrics." }, { "score": 6.01, "reference_url": "https://openai.com/index/gpt-5-6/", "source_type": "official_blog", "reported_setting": { "effort": "low", "eval_variant": null, "code_mode": "false" }, "notes": "OpenAI GPT-5.6 release figure (SEC-Bench Pro): GPT-5.5; SEC-Bench Pro=6.01. Figure 27: SEC-Bench Pro. Cost, latency, and output-token coordinates are retained as review metadata, not score-matrix metrics." }, { "score": 27.6, "reference_url": "https://openai.com/index/gpt-5-6/", "source_type": "official_blog", "reported_setting": { "effort": "medium", "eval_variant": null, "code_mode": "false" }, "notes": "OpenAI GPT-5.6 release figure (SEC-Bench Pro): GPT-5.5; SEC-Bench Pro=27.6. Figure 27: SEC-Bench Pro. Cost, latency, and output-token coordinates are retained as review metadata, not score-matrix metrics." }, { "score": 41.8, "reference_url": "https://openai.com/index/gpt-5-6/", "source_type": "official_blog", "reported_setting": { "effort": "high", "eval_variant": null, "code_mode": "false" }, "notes": "OpenAI GPT-5.6 release figure (SEC-Bench Pro): GPT-5.5; SEC-Bench Pro=41.8. Figure 27: SEC-Bench Pro. Cost, latency, and output-token coordinates are retained as review metadata, not score-matrix metrics." } ] }, { "model_id": "gpt-5.6-sol", "benchmark_id": "exploitbench", "score": 73.5, "reference_url": "https://openai.com/index/gpt-5-6/", "reported_setting": { "effort": "max", "eval_variant": null, "code_mode": "false" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-5.6 release table (table 4): GPT‑5.6 Sol; ExploitBench=73.5. Exact effort/variant resolved from the rendered chart point.", "candidates": [ { "score": 27.9, "reference_url": "https://openai.com/index/gpt-5-6/", "source_type": "official_blog", "reported_setting": { "effort": "low", "eval_variant": null, "code_mode": "false" }, "notes": "OpenAI GPT-5.6 release figure (ExploitBench): GPT-5.6 Sol; ExploitBench=27.9. Figure 27: Progress from vulnerable code to arbitrary code execution.. Cost, latency, and output-token coordinates are retained as review metadata, not score-matrix metrics." }, { "score": 51.8, "reference_url": "https://openai.com/index/gpt-5-6/", "source_type": "official_blog", "reported_setting": { "effort": "medium", "eval_variant": null, "code_mode": "false" }, "notes": "OpenAI GPT-5.6 release figure (ExploitBench): GPT-5.6 Sol; ExploitBench=51.8. Figure 27: Progress from vulnerable code to arbitrary code execution.. Cost, latency, and output-token coordinates are retained as review metadata, not score-matrix metrics." }, { "score": 63.0, "reference_url": "https://openai.com/index/gpt-5-6/", "source_type": "official_blog", "reported_setting": { "effort": "high", "eval_variant": null, "code_mode": "false" }, "notes": "OpenAI GPT-5.6 release figure (ExploitBench): GPT-5.6 Sol; ExploitBench=63.0. Figure 27: Progress from vulnerable code to arbitrary code execution.. Cost, latency, and output-token coordinates are retained as review metadata, not score-matrix metrics." }, { "score": 70.3, "reference_url": "https://openai.com/index/gpt-5-6/", "source_type": "official_blog", "reported_setting": { "effort": "xhigh", "eval_variant": null, "code_mode": "false" }, "notes": "OpenAI GPT-5.6 release figure (ExploitBench): GPT-5.6 Sol; ExploitBench=70.3. Figure 27: Progress from vulnerable code to arbitrary code execution.. Cost, latency, and output-token coordinates are retained as review metadata, not score-matrix metrics." } ] }, { "model_id": "gpt-5.6-terra", "benchmark_id": "exploitbench", "score": 52.9, "reference_url": "https://openai.com/index/gpt-5-6/", "reported_setting": { "effort": "max", "eval_variant": null, "code_mode": "false" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-5.6 release table (table 4): GPT‑5.6 Terra; ExploitBench=52.9. Exact effort/variant resolved from the rendered chart point.", "candidates": [ { "score": 22.1, "reference_url": "https://openai.com/index/gpt-5-6/", "source_type": "official_blog", "reported_setting": { "effort": "low", "eval_variant": null, "code_mode": "false" }, "notes": "OpenAI GPT-5.6 release figure (ExploitBench): GPT-5.6 Terra; ExploitBench=22.1. Figure 27: Progress from vulnerable code to arbitrary code execution.. Cost, latency, and output-token coordinates are retained as review metadata, not score-matrix metrics." }, { "score": 28.8, "reference_url": "https://openai.com/index/gpt-5-6/", "source_type": "official_blog", "reported_setting": { "effort": "medium", "eval_variant": null, "code_mode": "false" }, "notes": "OpenAI GPT-5.6 release figure (ExploitBench): GPT-5.6 Terra; ExploitBench=28.8. Figure 27: Progress from vulnerable code to arbitrary code execution.. Cost, latency, and output-token coordinates are retained as review metadata, not score-matrix metrics." }, { "score": 41.3, "reference_url": "https://openai.com/index/gpt-5-6/", "source_type": "official_blog", "reported_setting": { "effort": "high", "eval_variant": null, "code_mode": "false" }, "notes": "OpenAI GPT-5.6 release figure (ExploitBench): GPT-5.6 Terra; ExploitBench=41.3. Figure 27: Progress from vulnerable code to arbitrary code execution.. Cost, latency, and output-token coordinates are retained as review metadata, not score-matrix metrics." }, { "score": 46.2, "reference_url": "https://openai.com/index/gpt-5-6/", "source_type": "official_blog", "reported_setting": { "effort": "xhigh", "eval_variant": null, "code_mode": "false" }, "notes": "OpenAI GPT-5.6 release figure (ExploitBench): GPT-5.6 Terra; ExploitBench=46.2. Figure 27: Progress from vulnerable code to arbitrary code execution.. Cost, latency, and output-token coordinates are retained as review metadata, not score-matrix metrics." } ] }, { "model_id": "gpt-5.6-luna", "benchmark_id": "exploitbench", "score": 33.2, "reference_url": "https://openai.com/index/gpt-5-6/", "reported_setting": { "effort": "max", "eval_variant": null, "code_mode": "false" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-5.6 release table (table 4): GPT‑5.6 Luna; ExploitBench=33.2. Exact effort/variant resolved from the rendered chart point.", "candidates": [ { "score": 16.8, "reference_url": "https://openai.com/index/gpt-5-6/", "source_type": "official_blog", "reported_setting": { "effort": "low", "eval_variant": null, "code_mode": "false" }, "notes": "OpenAI GPT-5.6 release figure (ExploitBench): GPT-5.6 Luna; ExploitBench=16.8. Figure 27: Progress from vulnerable code to arbitrary code execution.. Cost, latency, and output-token coordinates are retained as review metadata, not score-matrix metrics." }, { "score": 23.8, "reference_url": "https://openai.com/index/gpt-5-6/", "source_type": "official_blog", "reported_setting": { "effort": "medium", "eval_variant": null, "code_mode": "false" }, "notes": "OpenAI GPT-5.6 release figure (ExploitBench): GPT-5.6 Luna; ExploitBench=23.8. Figure 27: Progress from vulnerable code to arbitrary code execution.. Cost, latency, and output-token coordinates are retained as review metadata, not score-matrix metrics." }, { "score": 29.9, "reference_url": "https://openai.com/index/gpt-5-6/", "source_type": "official_blog", "reported_setting": { "effort": "high", "eval_variant": null, "code_mode": "false" }, "notes": "OpenAI GPT-5.6 release figure (ExploitBench): GPT-5.6 Luna; ExploitBench=29.9. Figure 27: Progress from vulnerable code to arbitrary code execution.. Cost, latency, and output-token coordinates are retained as review metadata, not score-matrix metrics." }, { "score": 30.9, "reference_url": "https://openai.com/index/gpt-5-6/", "source_type": "official_blog", "reported_setting": { "effort": "xhigh", "eval_variant": null, "code_mode": "false" }, "notes": "OpenAI GPT-5.6 release figure (ExploitBench): GPT-5.6 Luna; ExploitBench=30.9. Figure 27: Progress from vulnerable code to arbitrary code execution.. Cost, latency, and output-token coordinates are retained as review metadata, not score-matrix metrics." } ] }, { "model_id": "gpt-5.5", "benchmark_id": "exploitbench", "score": 47.9, "reference_url": "https://openai.com/index/gpt-5-6/", "reported_setting": { "effort": "xhigh", "eval_variant": null, "code_mode": "false" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-5.6 release table (table 4): GPT‑5.5; ExploitBench=47.9. Exact effort/variant resolved from the rendered chart point.", "candidates": [ { "score": 27.9, "reference_url": "https://openai.com/index/gpt-5-6/", "source_type": "official_blog", "reported_setting": { "effort": "low", "eval_variant": null, "code_mode": "false" }, "notes": "OpenAI GPT-5.6 release figure (ExploitBench): GPT-5.5; ExploitBench=27.9. Figure 27: Progress from vulnerable code to arbitrary code execution.. Cost, latency, and output-token coordinates are retained as review metadata, not score-matrix metrics." }, { "score": 40.1, "reference_url": "https://openai.com/index/gpt-5-6/", "source_type": "official_blog", "reported_setting": { "effort": "medium", "eval_variant": null, "code_mode": "false" }, "notes": "OpenAI GPT-5.6 release figure (ExploitBench): GPT-5.5; ExploitBench=40.1. Figure 27: Progress from vulnerable code to arbitrary code execution.. Cost, latency, and output-token coordinates are retained as review metadata, not score-matrix metrics." }, { "score": 47.7, "reference_url": "https://openai.com/index/gpt-5-6/", "source_type": "official_blog", "reported_setting": { "effort": "high", "eval_variant": null, "code_mode": "false" }, "notes": "OpenAI GPT-5.6 release figure (ExploitBench): GPT-5.5; ExploitBench=47.7. Figure 27: Progress from vulnerable code to arbitrary code execution.. Cost, latency, and output-token coordinates are retained as review metadata, not score-matrix metrics." } ] }, { "model_id": "claude-fable-5", "benchmark_id": "exploitbench", "score": 78.0, "reference_url": "https://openai.com/index/gpt-5-6/", "reported_setting": { "effort": "API, 5 seeds", "eval_variant": null, "code_mode": null, "deployment": "claude-mythos-5", "safeguards": "lifted for Project Glasswing" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-5.6 release table (table 4): Claude Mythos 5; ExploitBench=78.0. Exact effort/variant resolved from the rendered chart point.", "candidates": [ { "score": 78.0, "reference_url": "https://www.anthropic.com/claude-sonnet-5-system-card", "source_type": "model_card", "reported_setting": { "effort": "source does not state", "tools": "cyber execution environment", "sampling": "pass@1", "harness": "exploit", "deployment": "claude-mythos-5" }, "notes": "Anthropic Claude Sonnet 5 System Card Figure 3.2.1.A, page 31: Claude Mythos 5; ExploitBench [Cap%]=78. Source setting: effort=source does not state; tools=cyber execution environment; sampling=pass@1; harness=exploit; deployment=claude-mythos-5." } ] }, { "model_id": "claude-mythos", "benchmark_id": "exploitbench", "score": 74.2, "reference_url": "https://openai.com/index/gpt-5-6/", "reported_setting": { "effort": "n/a", "eval_variant": null, "code_mode": "false" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-5.6 release table (table 4): Claude Mythos Preview; ExploitBench=74.2. Exact effort/variant resolved from the rendered chart point." }, { "model_id": "claude-opus-4.8", "benchmark_id": "exploitbench", "score": 40.0, "reference_url": "https://www.anthropic.com/claude-sonnet-5-system-card", "reported_setting": { "effort": "source does not state", "tools": "cyber execution environment", "sampling": "pass@1", "harness": "exploit" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Anthropic Claude Sonnet 5 System Card Figure 3.2.1.A, page 31: Claude Opus 4.8; ExploitBench [Cap%]=40. Source setting: effort=source does not state; tools=cyber execution environment; sampling=pass@1; harness=exploit.", "candidates": [ { "score": 40.0, "reference_url": "https://openai.com/index/gpt-5-6/", "source_type": "official_blog", "reported_setting": { "effort": "API, 5 seeds", "eval_variant": null, "code_mode": null }, "notes": "OpenAI GPT-5.6 release table (table 4): Claude Opus 4.8; ExploitBench=40.0. Exact effort/variant resolved from the rendered chart point." } ] }, { "model_id": "gpt-5.6-sol", "benchmark_id": "exploitgym", "score": 33.7, "reference_url": "https://openai.com/index/gpt-5-6/", "reported_setting": { "effort": "max", "eval_variant": "6h", "code_mode": "false" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-5.6 release table (table 4): GPT‑5.6 Sol; ExploitGym=33.7. Exact effort/variant resolved from the rendered chart point.", "candidates": [ { "score": 5.6, "reference_url": "https://openai.com/index/gpt-5-6/", "source_type": "official_blog", "reported_setting": { "effort": "low", "eval_variant": "2h", "code_mode": "false" }, "notes": "OpenAI GPT-5.6 release figure (ExploitGym): GPT-5.6 Sol · 2h cap; ExploitGym=5.6. Figure 27: ExploitGym. Cost, latency, and output-token coordinates are retained as review metadata, not score-matrix metrics." }, { "score": 15.7, "reference_url": "https://openai.com/index/gpt-5-6/", "source_type": "official_blog", "reported_setting": { "effort": "medium", "eval_variant": "2h", "code_mode": "false" }, "notes": "OpenAI GPT-5.6 release figure (ExploitGym): GPT-5.6 Sol · 2h cap; ExploitGym=15.7. Figure 27: ExploitGym. Cost, latency, and output-token coordinates are retained as review metadata, not score-matrix metrics." }, { "score": 21.7, "reference_url": "https://openai.com/index/gpt-5-6/", "source_type": "official_blog", "reported_setting": { "effort": "high", "eval_variant": "2h", "code_mode": "false" }, "notes": "OpenAI GPT-5.6 release figure (ExploitGym): GPT-5.6 Sol · 2h cap; ExploitGym=21.7. Figure 27: ExploitGym. Cost, latency, and output-token coordinates are retained as review metadata, not score-matrix metrics." }, { "score": 23.1, "reference_url": "https://openai.com/index/gpt-5-6/", "source_type": "official_blog", "reported_setting": { "effort": "xhigh", "eval_variant": "2h", "code_mode": "false" }, "notes": "OpenAI GPT-5.6 release figure (ExploitGym): GPT-5.6 Sol · 2h cap; ExploitGym=23.1. Figure 27: ExploitGym. Cost, latency, and output-token coordinates are retained as review metadata, not score-matrix metrics." }, { "score": 24.9, "reference_url": "https://openai.com/index/gpt-5-6/", "source_type": "official_blog", "reported_setting": { "effort": "max", "eval_variant": "2h", "code_mode": "false" }, "notes": "OpenAI GPT-5.6 release figure (ExploitGym): GPT-5.6 Sol · 2h cap; ExploitGym=24.9. Figure 27: ExploitGym. Cost, latency, and output-token coordinates are retained as review metadata, not score-matrix metrics." }, { "score": 5.9, "reference_url": "https://openai.com/index/gpt-5-6/", "source_type": "official_blog", "reported_setting": { "effort": "low", "eval_variant": "6h", "code_mode": "false" }, "notes": "OpenAI GPT-5.6 release figure (ExploitGym): GPT-5.6 Sol · 6h cap; ExploitGym=5.9. Figure 27: ExploitGym. Cost, latency, and output-token coordinates are retained as review metadata, not score-matrix metrics." }, { "score": 18.8, "reference_url": "https://openai.com/index/gpt-5-6/", "source_type": "official_blog", "reported_setting": { "effort": "medium", "eval_variant": "6h", "code_mode": "false" }, "notes": "OpenAI GPT-5.6 release figure (ExploitGym): GPT-5.6 Sol · 6h cap; ExploitGym=18.8. Figure 27: ExploitGym. Cost, latency, and output-token coordinates are retained as review metadata, not score-matrix metrics." }, { "score": 26.4, "reference_url": "https://openai.com/index/gpt-5-6/", "source_type": "official_blog", "reported_setting": { "effort": "high", "eval_variant": "6h", "code_mode": "false" }, "notes": "OpenAI GPT-5.6 release figure (ExploitGym): GPT-5.6 Sol · 6h cap; ExploitGym=26.4. Figure 27: ExploitGym. Cost, latency, and output-token coordinates are retained as review metadata, not score-matrix metrics." }, { "score": 30.7, "reference_url": "https://openai.com/index/gpt-5-6/", "source_type": "official_blog", "reported_setting": { "effort": "xhigh", "eval_variant": "6h", "code_mode": "false" }, "notes": "OpenAI GPT-5.6 release figure (ExploitGym): GPT-5.6 Sol · 6h cap; ExploitGym=30.7. Figure 27: ExploitGym. Cost, latency, and output-token coordinates are retained as review metadata, not score-matrix metrics." } ] }, { "model_id": "gpt-5.6-terra", "benchmark_id": "exploitgym", "score": 23.2, "reference_url": "https://openai.com/index/gpt-5-6/", "reported_setting": { "effort": "max", "eval_variant": "6h", "code_mode": "false" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-5.6 release table (table 4): GPT‑5.6 Terra; ExploitGym=23.2. Exact effort/variant resolved from the rendered chart point.", "candidates": [ { "score": 0.2, "reference_url": "https://openai.com/index/gpt-5-6/", "source_type": "official_blog", "reported_setting": { "effort": "low", "eval_variant": "2h", "code_mode": "false" }, "notes": "OpenAI GPT-5.6 release figure (ExploitGym): GPT-5.6 Terra · 2h cap; ExploitGym=0.2. Figure 27: ExploitGym. Cost, latency, and output-token coordinates are retained as review metadata, not score-matrix metrics." }, { "score": 1.4, "reference_url": "https://openai.com/index/gpt-5-6/", "source_type": "official_blog", "reported_setting": { "effort": "medium", "eval_variant": "2h", "code_mode": "false" }, "notes": "OpenAI GPT-5.6 release figure (ExploitGym): GPT-5.6 Terra · 2h cap; ExploitGym=1.4. Figure 27: ExploitGym. Cost, latency, and output-token coordinates are retained as review metadata, not score-matrix metrics." }, { "score": 9.4, "reference_url": "https://openai.com/index/gpt-5-6/", "source_type": "official_blog", "reported_setting": { "effort": "high", "eval_variant": "2h", "code_mode": "false" }, "notes": "OpenAI GPT-5.6 release figure (ExploitGym): GPT-5.6 Terra · 2h cap; ExploitGym=9.4. Figure 27: ExploitGym. Cost, latency, and output-token coordinates are retained as review metadata, not score-matrix metrics." }, { "score": 14.4, "reference_url": "https://openai.com/index/gpt-5-6/", "source_type": "official_blog", "reported_setting": { "effort": "xhigh", "eval_variant": "2h", "code_mode": "false" }, "notes": "OpenAI GPT-5.6 release figure (ExploitGym): GPT-5.6 Terra · 2h cap; ExploitGym=14.4. Figure 27: ExploitGym. Cost, latency, and output-token coordinates are retained as review metadata, not score-matrix metrics." }, { "score": 17.1, "reference_url": "https://openai.com/index/gpt-5-6/", "source_type": "official_blog", "reported_setting": { "effort": "max", "eval_variant": "2h", "code_mode": "false" }, "notes": "OpenAI GPT-5.6 release figure (ExploitGym): GPT-5.6 Terra · 2h cap; ExploitGym=17.1. Figure 27: ExploitGym. Cost, latency, and output-token coordinates are retained as review metadata, not score-matrix metrics." }, { "score": 0.2, "reference_url": "https://openai.com/index/gpt-5-6/", "source_type": "official_blog", "reported_setting": { "effort": "low", "eval_variant": "6h", "code_mode": "false" }, "notes": "OpenAI GPT-5.6 release figure (ExploitGym): GPT-5.6 Terra · 6h cap; ExploitGym=0.2. Figure 27: ExploitGym. Cost, latency, and output-token coordinates are retained as review metadata, not score-matrix metrics." }, { "score": 1.4, "reference_url": "https://openai.com/index/gpt-5-6/", "source_type": "official_blog", "reported_setting": { "effort": "medium", "eval_variant": "6h", "code_mode": "false" }, "notes": "OpenAI GPT-5.6 release figure (ExploitGym): GPT-5.6 Terra · 6h cap; ExploitGym=1.4. Figure 27: ExploitGym. Cost, latency, and output-token coordinates are retained as review metadata, not score-matrix metrics." }, { "score": 10.1, "reference_url": "https://openai.com/index/gpt-5-6/", "source_type": "official_blog", "reported_setting": { "effort": "high", "eval_variant": "6h", "code_mode": "false" }, "notes": "OpenAI GPT-5.6 release figure (ExploitGym): GPT-5.6 Terra · 6h cap; ExploitGym=10.1. Figure 27: ExploitGym. Cost, latency, and output-token coordinates are retained as review metadata, not score-matrix metrics." }, { "score": 18.2, "reference_url": "https://openai.com/index/gpt-5-6/", "source_type": "official_blog", "reported_setting": { "effort": "xhigh", "eval_variant": "6h", "code_mode": "false" }, "notes": "OpenAI GPT-5.6 release figure (ExploitGym): GPT-5.6 Terra · 6h cap; ExploitGym=18.2. Figure 27: ExploitGym. Cost, latency, and output-token coordinates are retained as review metadata, not score-matrix metrics." } ] }, { "model_id": "gpt-5.6-luna", "benchmark_id": "exploitgym", "score": 12.4, "reference_url": "https://openai.com/index/gpt-5-6/", "reported_setting": { "effort": "max", "eval_variant": "6h", "code_mode": "false" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-5.6 release table (table 4): GPT‑5.6 Luna; ExploitGym=12.4. Exact effort/variant resolved from the rendered chart point.", "candidates": [ { "score": 0.0, "reference_url": "https://openai.com/index/gpt-5-6/", "source_type": "official_blog", "reported_setting": { "effort": "low", "eval_variant": "2h", "code_mode": "false" }, "notes": "OpenAI GPT-5.6 release figure (ExploitGym): GPT-5.6 Luna · 2h cap; ExploitGym=0.0. Figure 27: ExploitGym. Cost, latency, and output-token coordinates are retained as review metadata, not score-matrix metrics." }, { "score": 0.3, "reference_url": "https://openai.com/index/gpt-5-6/", "source_type": "official_blog", "reported_setting": { "effort": "medium", "eval_variant": "2h", "code_mode": "false" }, "notes": "OpenAI GPT-5.6 release figure (ExploitGym): GPT-5.6 Luna · 2h cap; ExploitGym=0.3. Figure 27: ExploitGym. Cost, latency, and output-token coordinates are retained as review metadata, not score-matrix metrics." }, { "score": 4.5, "reference_url": "https://openai.com/index/gpt-5-6/", "source_type": "official_blog", "reported_setting": { "effort": "high", "eval_variant": "2h", "code_mode": "false" }, "notes": "OpenAI GPT-5.6 release figure (ExploitGym): GPT-5.6 Luna · 2h cap; ExploitGym=4.5. Figure 27: ExploitGym. Cost, latency, and output-token coordinates are retained as review metadata, not score-matrix metrics." }, { "score": 6.6, "reference_url": "https://openai.com/index/gpt-5-6/", "source_type": "official_blog", "reported_setting": { "effort": "xhigh", "eval_variant": "2h", "code_mode": "false" }, "notes": "OpenAI GPT-5.6 release figure (ExploitGym): GPT-5.6 Luna · 2h cap; ExploitGym=6.6. Figure 27: ExploitGym. Cost, latency, and output-token coordinates are retained as review metadata, not score-matrix metrics." }, { "score": 8.4, "reference_url": "https://openai.com/index/gpt-5-6/", "source_type": "official_blog", "reported_setting": { "effort": "max", "eval_variant": "2h", "code_mode": "false" }, "notes": "OpenAI GPT-5.6 release figure (ExploitGym): GPT-5.6 Luna · 2h cap; ExploitGym=8.4. Figure 27: ExploitGym. Cost, latency, and output-token coordinates are retained as review metadata, not score-matrix metrics." }, { "score": 0.0, "reference_url": "https://openai.com/index/gpt-5-6/", "source_type": "official_blog", "reported_setting": { "effort": "low", "eval_variant": "6h", "code_mode": "false" }, "notes": "OpenAI GPT-5.6 release figure (ExploitGym): GPT-5.6 Luna · 6h cap; ExploitGym=0.0. Figure 27: ExploitGym. Cost, latency, and output-token coordinates are retained as review metadata, not score-matrix metrics." }, { "score": 0.3, "reference_url": "https://openai.com/index/gpt-5-6/", "source_type": "official_blog", "reported_setting": { "effort": "medium", "eval_variant": "6h", "code_mode": "false" }, "notes": "OpenAI GPT-5.6 release figure (ExploitGym): GPT-5.6 Luna · 6h cap; ExploitGym=0.3. Figure 27: ExploitGym. Cost, latency, and output-token coordinates are retained as review metadata, not score-matrix metrics." }, { "score": 4.6, "reference_url": "https://openai.com/index/gpt-5-6/", "source_type": "official_blog", "reported_setting": { "effort": "high", "eval_variant": "6h", "code_mode": "false" }, "notes": "OpenAI GPT-5.6 release figure (ExploitGym): GPT-5.6 Luna · 6h cap; ExploitGym=4.6. Figure 27: ExploitGym. Cost, latency, and output-token coordinates are retained as review metadata, not score-matrix metrics." }, { "score": 11.0, "reference_url": "https://openai.com/index/gpt-5-6/", "source_type": "official_blog", "reported_setting": { "effort": "xhigh", "eval_variant": "6h", "code_mode": "false" }, "notes": "OpenAI GPT-5.6 release figure (ExploitGym): GPT-5.6 Luna · 6h cap; ExploitGym=11.0. Figure 27: ExploitGym. Cost, latency, and output-token coordinates are retained as review metadata, not score-matrix metrics." } ] }, { "model_id": "gpt-5.5", "benchmark_id": "exploitgym", "score": 15.1, "reference_url": "https://openai.com/index/gpt-5-6/", "reported_setting": { "effort": "xhigh", "eval_variant": "2h", "code_mode": "false" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-5.6 release table (table 4): GPT‑5.5; ExploitGym=15.1. Exact effort/variant resolved from the rendered chart point.", "candidates": [ { "score": 14.8, "reference_url": "https://ai.meta.com/static-resource/muse-spark-1.1-evaluation-report", "source_type": "tech_report", "reported_setting": { "mode": "thinking", "effort": "xhigh", "tools": "benchmark-specified", "sampling": "pass@1", "harness": "Meta safety/preparedness evaluation", "temperature": "1.0 for Muse; provider/source setting for comparisons", "context": "source-specific", "configuration": "default", "notes": "Capability scorecard; Muse evaluated without system prompt; high reasoning." }, "notes": "Table 1 · p6; metric=pass@1_percent. Claude cyber-benchmark cells omitted where refusal compromised capability measurement." } ] }, { "model_id": "gpt-5.6-sol", "benchmark_id": "internal_research_debugging", "score": 68.3, "reference_url": "https://openai.com/index/gpt-5-6/", "reported_setting": { "effort": "max", "eval_variant": null, "code_mode": "false" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-5.6 release table (table 5): GPT‑5.6 Sol; Internal Research Debugging Evaluation=68.3. Named internal evaluation; task count not disclosed. Exact effort/variant resolved from the rendered chart point.", "candidates": [ { "score": 30.34, "reference_url": "https://openai.com/index/gpt-5-6/", "source_type": "official_blog", "reported_setting": { "effort": "low", "eval_variant": null, "code_mode": "false" }, "notes": "OpenAI GPT-5.6 release figure (Internal Research Debugging Evaluation): GPT-5.6 Sol; Internal Research Debugging Evaluation=30.34. Figure 29: Internal Research Debugging Evaluation. Cost, latency, and output-token coordinates are retained as review metadata, not score-matrix metrics." }, { "score": 48.04, "reference_url": "https://openai.com/index/gpt-5-6/", "source_type": "official_blog", "reported_setting": { "effort": "medium", "eval_variant": null, "code_mode": "false" }, "notes": "OpenAI GPT-5.6 release figure (Internal Research Debugging Evaluation): GPT-5.6 Sol; Internal Research Debugging Evaluation=48.04. Figure 29: Internal Research Debugging Evaluation. Cost, latency, and output-token coordinates are retained as review metadata, not score-matrix metrics." }, { "score": 55.32, "reference_url": "https://openai.com/index/gpt-5-6/", "source_type": "official_blog", "reported_setting": { "effort": "high", "eval_variant": null, "code_mode": "false" }, "notes": "OpenAI GPT-5.6 release figure (Internal Research Debugging Evaluation): GPT-5.6 Sol; Internal Research Debugging Evaluation=55.32. Figure 29: Internal Research Debugging Evaluation. Cost, latency, and output-token coordinates are retained as review metadata, not score-matrix metrics." }, { "score": 60.7, "reference_url": "https://openai.com/index/gpt-5-6/", "source_type": "official_blog", "reported_setting": { "effort": "xhigh", "eval_variant": null, "code_mode": "false" }, "notes": "OpenAI GPT-5.6 release figure (Internal Research Debugging Evaluation): GPT-5.6 Sol; Internal Research Debugging Evaluation=60.7. Figure 29: Internal Research Debugging Evaluation. Cost, latency, and output-token coordinates are retained as review metadata, not score-matrix metrics." } ] }, { "model_id": "gpt-5.6-terra", "benchmark_id": "internal_research_debugging", "score": 67.8, "reference_url": "https://openai.com/index/gpt-5-6/", "reported_setting": { "effort": "max", "eval_variant": null, "code_mode": "false" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-5.6 release table (table 5): GPT‑5.6 Terra; Internal Research Debugging Evaluation=67.8. Named internal evaluation; task count not disclosed. Exact effort/variant resolved from the rendered chart point.", "candidates": [ { "score": 10.09, "reference_url": "https://openai.com/index/gpt-5-6/", "source_type": "official_blog", "reported_setting": { "effort": "none", "eval_variant": null, "code_mode": "false" }, "notes": "OpenAI GPT-5.6 release figure (Internal Research Debugging Evaluation): GPT-5.6 Terra; Internal Research Debugging Evaluation=10.09. Figure 29: Internal Research Debugging Evaluation. Cost, latency, and output-token coordinates are retained as review metadata, not score-matrix metrics." }, { "score": 17.01, "reference_url": "https://openai.com/index/gpt-5-6/", "source_type": "official_blog", "reported_setting": { "effort": "low", "eval_variant": null, "code_mode": "false" }, "notes": "OpenAI GPT-5.6 release figure (Internal Research Debugging Evaluation): GPT-5.6 Terra; Internal Research Debugging Evaluation=17.01. Figure 29: Internal Research Debugging Evaluation. Cost, latency, and output-token coordinates are retained as review metadata, not score-matrix metrics." }, { "score": 40.4, "reference_url": "https://openai.com/index/gpt-5-6/", "source_type": "official_blog", "reported_setting": { "effort": "medium", "eval_variant": null, "code_mode": "false" }, "notes": "OpenAI GPT-5.6 release figure (Internal Research Debugging Evaluation): GPT-5.6 Terra; Internal Research Debugging Evaluation=40.4. Figure 29: Internal Research Debugging Evaluation. Cost, latency, and output-token coordinates are retained as review metadata, not score-matrix metrics." }, { "score": 49.15, "reference_url": "https://openai.com/index/gpt-5-6/", "source_type": "official_blog", "reported_setting": { "effort": "high", "eval_variant": null, "code_mode": "false" }, "notes": "OpenAI GPT-5.6 release figure (Internal Research Debugging Evaluation): GPT-5.6 Terra; Internal Research Debugging Evaluation=49.15. Figure 29: Internal Research Debugging Evaluation. Cost, latency, and output-token coordinates are retained as review metadata, not score-matrix metrics." }, { "score": 52.13, "reference_url": "https://openai.com/index/gpt-5-6/", "source_type": "official_blog", "reported_setting": { "effort": "xhigh", "eval_variant": null, "code_mode": "false" }, "notes": "OpenAI GPT-5.6 release figure (Internal Research Debugging Evaluation): GPT-5.6 Terra; Internal Research Debugging Evaluation=52.13. Figure 29: Internal Research Debugging Evaluation. Cost, latency, and output-token coordinates are retained as review metadata, not score-matrix metrics." } ] }, { "model_id": "gpt-5.6-luna", "benchmark_id": "internal_research_debugging", "score": 50.8, "reference_url": "https://openai.com/index/gpt-5-6/", "reported_setting": { "effort": "max", "eval_variant": null, "code_mode": "false" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-5.6 release table (table 5): GPT‑5.6 Luna; Internal Research Debugging Evaluation=50.8. Named internal evaluation; task count not disclosed. Exact effort/variant resolved from the rendered chart point.", "candidates": [ { "score": 8.02, "reference_url": "https://openai.com/index/gpt-5-6/", "source_type": "official_blog", "reported_setting": { "effort": "none", "eval_variant": null, "code_mode": "false" }, "notes": "OpenAI GPT-5.6 release figure (Internal Research Debugging Evaluation): GPT-5.6 Luna; Internal Research Debugging Evaluation=8.02. Figure 29: Internal Research Debugging Evaluation. Cost, latency, and output-token coordinates are retained as review metadata, not score-matrix metrics." }, { "score": 12.13, "reference_url": "https://openai.com/index/gpt-5-6/", "source_type": "official_blog", "reported_setting": { "effort": "low", "eval_variant": null, "code_mode": "false" }, "notes": "OpenAI GPT-5.6 release figure (Internal Research Debugging Evaluation): GPT-5.6 Luna; Internal Research Debugging Evaluation=12.13. Figure 29: Internal Research Debugging Evaluation. Cost, latency, and output-token coordinates are retained as review metadata, not score-matrix metrics." }, { "score": 22.1, "reference_url": "https://openai.com/index/gpt-5-6/", "source_type": "official_blog", "reported_setting": { "effort": "medium", "eval_variant": null, "code_mode": "false" }, "notes": "OpenAI GPT-5.6 release figure (Internal Research Debugging Evaluation): GPT-5.6 Luna; Internal Research Debugging Evaluation=22.1. Figure 29: Internal Research Debugging Evaluation. Cost, latency, and output-token coordinates are retained as review metadata, not score-matrix metrics." }, { "score": 36.03, "reference_url": "https://openai.com/index/gpt-5-6/", "source_type": "official_blog", "reported_setting": { "effort": "high", "eval_variant": null, "code_mode": "false" }, "notes": "OpenAI GPT-5.6 release figure (Internal Research Debugging Evaluation): GPT-5.6 Luna; Internal Research Debugging Evaluation=36.03. Figure 29: Internal Research Debugging Evaluation. Cost, latency, and output-token coordinates are retained as review metadata, not score-matrix metrics." }, { "score": 45.95, "reference_url": "https://openai.com/index/gpt-5-6/", "source_type": "official_blog", "reported_setting": { "effort": "xhigh", "eval_variant": null, "code_mode": "false" }, "notes": "OpenAI GPT-5.6 release figure (Internal Research Debugging Evaluation): GPT-5.6 Luna; Internal Research Debugging Evaluation=45.95. Figure 29: Internal Research Debugging Evaluation. Cost, latency, and output-token coordinates are retained as review metadata, not score-matrix metrics." } ] }, { "model_id": "gpt-5.5", "benchmark_id": "internal_research_debugging", "score": 50.0, "reference_url": "https://openai.com/index/gpt-5-6/", "reported_setting": { "effort": "xhigh", "eval_variant": null, "code_mode": "false" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-5.6 release table (table 5): GPT‑5.5; Internal Research Debugging Evaluation=50.0. Named internal evaluation; task count not disclosed. Exact effort/variant resolved from the rendered chart point.", "candidates": [ { "score": 14.57, "reference_url": "https://openai.com/index/gpt-5-6/", "source_type": "official_blog", "reported_setting": { "effort": "none", "eval_variant": null, "code_mode": "false" }, "notes": "OpenAI GPT-5.6 release figure (Internal Research Debugging Evaluation): GPT-5.5; Internal Research Debugging Evaluation=14.57. Figure 29: Internal Research Debugging Evaluation. Cost, latency, and output-token coordinates are retained as review metadata, not score-matrix metrics." }, { "score": 27.52, "reference_url": "https://openai.com/index/gpt-5-6/", "source_type": "official_blog", "reported_setting": { "effort": "low", "eval_variant": null, "code_mode": "false" }, "notes": "OpenAI GPT-5.6 release figure (Internal Research Debugging Evaluation): GPT-5.5; Internal Research Debugging Evaluation=27.52. Figure 29: Internal Research Debugging Evaluation. Cost, latency, and output-token coordinates are retained as review metadata, not score-matrix metrics." }, { "score": 42.56, "reference_url": "https://openai.com/index/gpt-5-6/", "source_type": "official_blog", "reported_setting": { "effort": "medium", "eval_variant": null, "code_mode": "false" }, "notes": "OpenAI GPT-5.6 release figure (Internal Research Debugging Evaluation): GPT-5.5; Internal Research Debugging Evaluation=42.56. Figure 29: Internal Research Debugging Evaluation. Cost, latency, and output-token coordinates are retained as review metadata, not score-matrix metrics." }, { "score": 46.38, "reference_url": "https://openai.com/index/gpt-5-6/", "source_type": "official_blog", "reported_setting": { "effort": "high", "eval_variant": null, "code_mode": "false" }, "notes": "OpenAI GPT-5.6 release figure (Internal Research Debugging Evaluation): GPT-5.5; Internal Research Debugging Evaluation=46.38. Figure 29: Internal Research Debugging Evaluation. Cost, latency, and output-token coordinates are retained as review metadata, not score-matrix metrics." } ] }, { "model_id": "gpt-5.6-sol", "benchmark_id": "kernelgen_1p", "score": 61.1, "reference_url": "https://openai.com/index/gpt-5-6/", "reported_setting": { "effort": "max", "eval_variant": null, "code_mode": "false" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-5.6 release table (table 5): GPT‑5.6 Sol; KernelGen 1P=61.1. Exact effort/variant resolved from the rendered chart point.", "candidates": [ { "score": 2.35, "reference_url": "https://openai.com/index/gpt-5-6/", "source_type": "official_blog", "reported_setting": { "effort": "low", "eval_variant": null, "code_mode": "false" }, "notes": "OpenAI GPT-5.6 release figure (KernelGen 1P): GPT-5.6 Sol; KernelGen 1P=2.35. Figure 29: KernelGen 1P. Cost, latency, and output-token coordinates are retained as review metadata, not score-matrix metrics." }, { "score": 12.66, "reference_url": "https://openai.com/index/gpt-5-6/", "source_type": "official_blog", "reported_setting": { "effort": "medium", "eval_variant": null, "code_mode": "false" }, "notes": "OpenAI GPT-5.6 release figure (KernelGen 1P): GPT-5.6 Sol; KernelGen 1P=12.66. Figure 29: KernelGen 1P. Cost, latency, and output-token coordinates are retained as review metadata, not score-matrix metrics." }, { "score": 22.66, "reference_url": "https://openai.com/index/gpt-5-6/", "source_type": "official_blog", "reported_setting": { "effort": "high", "eval_variant": null, "code_mode": "false" }, "notes": "OpenAI GPT-5.6 release figure (KernelGen 1P): GPT-5.6 Sol; KernelGen 1P=22.66. Figure 29: KernelGen 1P. Cost, latency, and output-token coordinates are retained as review metadata, not score-matrix metrics." }, { "score": 32.07, "reference_url": "https://openai.com/index/gpt-5-6/", "source_type": "official_blog", "reported_setting": { "effort": "xhigh", "eval_variant": null, "code_mode": "false" }, "notes": "OpenAI GPT-5.6 release figure (KernelGen 1P): GPT-5.6 Sol; KernelGen 1P=32.07. Figure 29: KernelGen 1P. Cost, latency, and output-token coordinates are retained as review metadata, not score-matrix metrics." } ] }, { "model_id": "gpt-5.6-terra", "benchmark_id": "kernelgen_1p", "score": 49.2, "reference_url": "https://openai.com/index/gpt-5-6/", "reported_setting": { "effort": "max", "eval_variant": null, "code_mode": "false" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-5.6 release table (table 5): GPT‑5.6 Terra; KernelGen 1P=49.2. Exact effort/variant resolved from the rendered chart point.", "candidates": [ { "score": 0.0, "reference_url": "https://openai.com/index/gpt-5-6/", "source_type": "official_blog", "reported_setting": { "effort": "none", "eval_variant": null, "code_mode": "false" }, "notes": "OpenAI GPT-5.6 release figure (KernelGen 1P): GPT-5.6 Terra; KernelGen 1P=0.0. Figure 29: KernelGen 1P. Cost, latency, and output-token coordinates are retained as review metadata, not score-matrix metrics." }, { "score": 0.06, "reference_url": "https://openai.com/index/gpt-5-6/", "source_type": "official_blog", "reported_setting": { "effort": "low", "eval_variant": null, "code_mode": "false" }, "notes": "OpenAI GPT-5.6 release figure (KernelGen 1P): GPT-5.6 Terra; KernelGen 1P=0.06. Figure 29: KernelGen 1P. Cost, latency, and output-token coordinates are retained as review metadata, not score-matrix metrics." }, { "score": 2.7, "reference_url": "https://openai.com/index/gpt-5-6/", "source_type": "official_blog", "reported_setting": { "effort": "medium", "eval_variant": null, "code_mode": "false" }, "notes": "OpenAI GPT-5.6 release figure (KernelGen 1P): GPT-5.6 Terra; KernelGen 1P=2.7. Figure 29: KernelGen 1P. Cost, latency, and output-token coordinates are retained as review metadata, not score-matrix metrics." }, { "score": 15.08, "reference_url": "https://openai.com/index/gpt-5-6/", "source_type": "official_blog", "reported_setting": { "effort": "high", "eval_variant": null, "code_mode": "false" }, "notes": "OpenAI GPT-5.6 release figure (KernelGen 1P): GPT-5.6 Terra; KernelGen 1P=15.08. Figure 29: KernelGen 1P. Cost, latency, and output-token coordinates are retained as review metadata, not score-matrix metrics." }, { "score": 22.83, "reference_url": "https://openai.com/index/gpt-5-6/", "source_type": "official_blog", "reported_setting": { "effort": "xhigh", "eval_variant": null, "code_mode": "false" }, "notes": "OpenAI GPT-5.6 release figure (KernelGen 1P): GPT-5.6 Terra; KernelGen 1P=22.83. Figure 29: KernelGen 1P. Cost, latency, and output-token coordinates are retained as review metadata, not score-matrix metrics." } ] }, { "model_id": "gpt-5.6-luna", "benchmark_id": "kernelgen_1p", "score": 22.4, "reference_url": "https://openai.com/index/gpt-5-6/", "reported_setting": { "effort": "max", "eval_variant": null, "code_mode": "false" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-5.6 release table (table 5): GPT‑5.6 Luna; KernelGen 1P=22.4. Exact effort/variant resolved from the rendered chart point.", "candidates": [ { "score": 0.0, "reference_url": "https://openai.com/index/gpt-5-6/", "source_type": "official_blog", "reported_setting": { "effort": "none", "eval_variant": null, "code_mode": "false" }, "notes": "OpenAI GPT-5.6 release figure (KernelGen 1P): GPT-5.6 Luna; KernelGen 1P=0.0. Figure 29: KernelGen 1P. Cost, latency, and output-token coordinates are retained as review metadata, not score-matrix metrics." }, { "score": 0.02, "reference_url": "https://openai.com/index/gpt-5-6/", "source_type": "official_blog", "reported_setting": { "effort": "low", "eval_variant": null, "code_mode": "false" }, "notes": "OpenAI GPT-5.6 release figure (KernelGen 1P): GPT-5.6 Luna; KernelGen 1P=0.02. Figure 29: KernelGen 1P. Cost, latency, and output-token coordinates are retained as review metadata, not score-matrix metrics." }, { "score": 0.33, "reference_url": "https://openai.com/index/gpt-5-6/", "source_type": "official_blog", "reported_setting": { "effort": "medium", "eval_variant": null, "code_mode": "false" }, "notes": "OpenAI GPT-5.6 release figure (KernelGen 1P): GPT-5.6 Luna; KernelGen 1P=0.33. Figure 29: KernelGen 1P. Cost, latency, and output-token coordinates are retained as review metadata, not score-matrix metrics." }, { "score": 7.13, "reference_url": "https://openai.com/index/gpt-5-6/", "source_type": "official_blog", "reported_setting": { "effort": "high", "eval_variant": null, "code_mode": "false" }, "notes": "OpenAI GPT-5.6 release figure (KernelGen 1P): GPT-5.6 Luna; KernelGen 1P=7.13. Figure 29: KernelGen 1P. Cost, latency, and output-token coordinates are retained as review metadata, not score-matrix metrics." }, { "score": 16.01, "reference_url": "https://openai.com/index/gpt-5-6/", "source_type": "official_blog", "reported_setting": { "effort": "xhigh", "eval_variant": null, "code_mode": "false" }, "notes": "OpenAI GPT-5.6 release figure (KernelGen 1P): GPT-5.6 Luna; KernelGen 1P=16.01. Figure 29: KernelGen 1P. Cost, latency, and output-token coordinates are retained as review metadata, not score-matrix metrics." } ] }, { "model_id": "gpt-5.5", "benchmark_id": "kernelgen_1p", "score": 29.3, "reference_url": "https://openai.com/index/gpt-5-6/", "reported_setting": { "effort": "xhigh", "eval_variant": null, "code_mode": "false" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-5.6 release table (table 5): GPT‑5.5; KernelGen 1P=29.3. Exact effort/variant resolved from the rendered chart point.", "candidates": [ { "score": 0.9, "reference_url": "https://openai.com/index/gpt-5-6/", "source_type": "official_blog", "reported_setting": { "effort": "low", "eval_variant": null, "code_mode": "false" }, "notes": "OpenAI GPT-5.6 release figure (KernelGen 1P): GPT-5.5; KernelGen 1P=0.9. Figure 29: KernelGen 1P. Cost, latency, and output-token coordinates are retained as review metadata, not score-matrix metrics." }, { "score": 9.58, "reference_url": "https://openai.com/index/gpt-5-6/", "source_type": "official_blog", "reported_setting": { "effort": "medium", "eval_variant": null, "code_mode": "false" }, "notes": "OpenAI GPT-5.6 release figure (KernelGen 1P): GPT-5.5; KernelGen 1P=9.58. Figure 29: KernelGen 1P. Cost, latency, and output-token coordinates are retained as review metadata, not score-matrix metrics." }, { "score": 18.75, "reference_url": "https://openai.com/index/gpt-5-6/", "source_type": "official_blog", "reported_setting": { "effort": "high", "eval_variant": null, "code_mode": "false" }, "notes": "OpenAI GPT-5.6 release figure (KernelGen 1P): GPT-5.5; KernelGen 1P=18.75. Figure 29: KernelGen 1P. Cost, latency, and output-token coordinates are retained as review metadata, not score-matrix metrics." } ] }, { "model_id": "gpt-5.6-sol", "benchmark_id": "nanogpt_self_improvement", "score": 9.69, "reference_url": "https://openai.com/index/gpt-5-6/", "reported_setting": { "effort": "max", "eval_variant": null, "code_mode": "false" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-5.6 release table (table 5): GPT‑5.6 Sol; NanoGPT=9.69. Exact effort/variant resolved from the rendered chart point.", "candidates": [ { "score": 0.18, "reference_url": "https://openai.com/index/gpt-5-6/", "source_type": "official_blog", "reported_setting": { "effort": "none", "eval_variant": null, "code_mode": "false" }, "notes": "OpenAI GPT-5.6 release figure (NanoGPT): GPT-5.6 Sol; NanoGPT=0.18. Figure 29: NanoGPT. Cost, latency, and output-token coordinates are retained as review metadata, not score-matrix metrics." }, { "score": 0.42, "reference_url": "https://openai.com/index/gpt-5-6/", "source_type": "official_blog", "reported_setting": { "effort": "low", "eval_variant": null, "code_mode": "false" }, "notes": "OpenAI GPT-5.6 release figure (NanoGPT): GPT-5.6 Sol; NanoGPT=0.42. Figure 29: NanoGPT. Cost, latency, and output-token coordinates are retained as review metadata, not score-matrix metrics." }, { "score": 1.05, "reference_url": "https://openai.com/index/gpt-5-6/", "source_type": "official_blog", "reported_setting": { "effort": "medium", "eval_variant": null, "code_mode": "false" }, "notes": "OpenAI GPT-5.6 release figure (NanoGPT): GPT-5.6 Sol; NanoGPT=1.05. Figure 29: NanoGPT. Cost, latency, and output-token coordinates are retained as review metadata, not score-matrix metrics." }, { "score": 1.96, "reference_url": "https://openai.com/index/gpt-5-6/", "source_type": "official_blog", "reported_setting": { "effort": "high", "eval_variant": null, "code_mode": "false" }, "notes": "OpenAI GPT-5.6 release figure (NanoGPT): GPT-5.6 Sol; NanoGPT=1.96. Figure 29: NanoGPT. Cost, latency, and output-token coordinates are retained as review metadata, not score-matrix metrics." }, { "score": 2.52, "reference_url": "https://openai.com/index/gpt-5-6/", "source_type": "official_blog", "reported_setting": { "effort": "xhigh", "eval_variant": null, "code_mode": "false" }, "notes": "OpenAI GPT-5.6 release figure (NanoGPT): GPT-5.6 Sol; NanoGPT=2.52. Figure 29: NanoGPT. Cost, latency, and output-token coordinates are retained as review metadata, not score-matrix metrics." } ] }, { "model_id": "gpt-5.6-terra", "benchmark_id": "nanogpt_self_improvement", "score": 14.5, "reference_url": "https://openai.com/index/gpt-5-6/", "reported_setting": { "effort": "max", "eval_variant": null, "code_mode": "false" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-5.6 release table (table 5): GPT‑5.6 Terra; NanoGPT=14.5. Exact effort/variant resolved from the rendered chart point.", "candidates": [ { "score": 0.03, "reference_url": "https://openai.com/index/gpt-5-6/", "source_type": "official_blog", "reported_setting": { "effort": "low", "eval_variant": null, "code_mode": "false" }, "notes": "OpenAI GPT-5.6 release figure (NanoGPT): GPT-5.6 Terra; NanoGPT=0.03. Figure 29: NanoGPT. Cost, latency, and output-token coordinates are retained as review metadata, not score-matrix metrics." }, { "score": 0.31, "reference_url": "https://openai.com/index/gpt-5-6/", "source_type": "official_blog", "reported_setting": { "effort": "medium", "eval_variant": null, "code_mode": "false" }, "notes": "OpenAI GPT-5.6 release figure (NanoGPT): GPT-5.6 Terra; NanoGPT=0.31. Figure 29: NanoGPT. Cost, latency, and output-token coordinates are retained as review metadata, not score-matrix metrics." }, { "score": 0.37, "reference_url": "https://openai.com/index/gpt-5-6/", "source_type": "official_blog", "reported_setting": { "effort": "high", "eval_variant": null, "code_mode": "false" }, "notes": "OpenAI GPT-5.6 release figure (NanoGPT): GPT-5.6 Terra; NanoGPT=0.37. Figure 29: NanoGPT. Cost, latency, and output-token coordinates are retained as review metadata, not score-matrix metrics." }, { "score": 0.77, "reference_url": "https://openai.com/index/gpt-5-6/", "source_type": "official_blog", "reported_setting": { "effort": "xhigh", "eval_variant": null, "code_mode": "false" }, "notes": "OpenAI GPT-5.6 release figure (NanoGPT): GPT-5.6 Terra; NanoGPT=0.77. Figure 29: NanoGPT. Cost, latency, and output-token coordinates are retained as review metadata, not score-matrix metrics." } ] }, { "model_id": "gpt-5.6-luna", "benchmark_id": "nanogpt_self_improvement", "score": 1.66, "reference_url": "https://openai.com/index/gpt-5-6/", "reported_setting": { "effort": "max", "eval_variant": null, "code_mode": "false" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-5.6 release table (table 5): GPT‑5.6 Luna; NanoGPT=1.66. Exact effort/variant resolved from the rendered chart point.", "candidates": [ { "score": 0.04, "reference_url": "https://openai.com/index/gpt-5-6/", "source_type": "official_blog", "reported_setting": { "effort": "low", "eval_variant": null, "code_mode": "false" }, "notes": "OpenAI GPT-5.6 release figure (NanoGPT): GPT-5.6 Luna; NanoGPT=0.04. Figure 29: NanoGPT. Cost, latency, and output-token coordinates are retained as review metadata, not score-matrix metrics." }, { "score": 0.04, "reference_url": "https://openai.com/index/gpt-5-6/", "source_type": "official_blog", "reported_setting": { "effort": "medium", "eval_variant": null, "code_mode": "false" }, "notes": "OpenAI GPT-5.6 release figure (NanoGPT): GPT-5.6 Luna; NanoGPT=0.04. Figure 29: NanoGPT. Cost, latency, and output-token coordinates are retained as review metadata, not score-matrix metrics." }, { "score": 0.2, "reference_url": "https://openai.com/index/gpt-5-6/", "source_type": "official_blog", "reported_setting": { "effort": "high", "eval_variant": null, "code_mode": "false" }, "notes": "OpenAI GPT-5.6 release figure (NanoGPT): GPT-5.6 Luna; NanoGPT=0.2. Figure 29: NanoGPT. Cost, latency, and output-token coordinates are retained as review metadata, not score-matrix metrics." }, { "score": 0.83, "reference_url": "https://openai.com/index/gpt-5-6/", "source_type": "official_blog", "reported_setting": { "effort": "xhigh", "eval_variant": null, "code_mode": "false" }, "notes": "OpenAI GPT-5.6 release figure (NanoGPT): GPT-5.6 Luna; NanoGPT=0.83. Figure 29: NanoGPT. Cost, latency, and output-token coordinates are retained as review metadata, not score-matrix metrics." } ] }, { "model_id": "gpt-5.5", "benchmark_id": "nanogpt_self_improvement", "score": 2.65, "reference_url": "https://openai.com/index/gpt-5-6/", "reported_setting": { "effort": "xhigh", "eval_variant": null, "code_mode": "false" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-5.6 release table (table 5): GPT‑5.5; NanoGPT=2.65. Exact effort/variant resolved from the rendered chart point.", "candidates": [ { "score": 0.02, "reference_url": "https://openai.com/index/gpt-5-6/", "source_type": "official_blog", "reported_setting": { "effort": "none", "eval_variant": null, "code_mode": "false" }, "notes": "OpenAI GPT-5.6 release figure (NanoGPT): GPT-5.5; NanoGPT=0.02. Figure 29: NanoGPT. Cost, latency, and output-token coordinates are retained as review metadata, not score-matrix metrics." }, { "score": 0.13, "reference_url": "https://openai.com/index/gpt-5-6/", "source_type": "official_blog", "reported_setting": { "effort": "low", "eval_variant": null, "code_mode": "false" }, "notes": "OpenAI GPT-5.6 release figure (NanoGPT): GPT-5.5; NanoGPT=0.13. Figure 29: NanoGPT. Cost, latency, and output-token coordinates are retained as review metadata, not score-matrix metrics." }, { "score": 0.38, "reference_url": "https://openai.com/index/gpt-5-6/", "source_type": "official_blog", "reported_setting": { "effort": "medium", "eval_variant": null, "code_mode": "false" }, "notes": "OpenAI GPT-5.6 release figure (NanoGPT): GPT-5.5; NanoGPT=0.38. Figure 29: NanoGPT. Cost, latency, and output-token coordinates are retained as review metadata, not score-matrix metrics." }, { "score": 1.15, "reference_url": "https://openai.com/index/gpt-5-6/", "source_type": "official_blog", "reported_setting": { "effort": "high", "eval_variant": null, "code_mode": "false" }, "notes": "OpenAI GPT-5.6 release figure (NanoGPT): GPT-5.5; NanoGPT=1.15. Figure 29: NanoGPT. Cost, latency, and output-token coordinates are retained as review metadata, not score-matrix metrics." } ] }, { "model_id": "gpt-5.6-sol", "benchmark_id": "posttrain_bench_lite", "score": 50.3, "reference_url": "https://openai.com/index/gpt-5-6/", "reported_setting": { "mode": "thinking", "effort": "source does not state", "tools": "none", "sampling": "source does not state", "judge": "benchmark-specified", "harness": "official release evaluation", "prompt_style": "source does not state", "temperature": "source does not state", "context": "source does not state" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-5.6 release table (table 5): GPT‑5.6 Sol; PostTrainBench Lite=50.3. Distinct Lite variant from the existing PostTrainBench row." }, { "model_id": "gpt-5.6-terra", "benchmark_id": "posttrain_bench_lite", "score": 51.5, "reference_url": "https://openai.com/index/gpt-5-6/", "reported_setting": { "mode": "thinking", "effort": "source does not state", "tools": "none", "sampling": "source does not state", "judge": "benchmark-specified", "harness": "official release evaluation", "prompt_style": "source does not state", "temperature": "source does not state", "context": "source does not state" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-5.6 release table (table 5): GPT‑5.6 Terra; PostTrainBench Lite=51.5. Distinct Lite variant from the existing PostTrainBench row." }, { "model_id": "gpt-5.6-luna", "benchmark_id": "posttrain_bench_lite", "score": 29.6, "reference_url": "https://openai.com/index/gpt-5-6/", "reported_setting": { "mode": "thinking", "effort": "source does not state", "tools": "none", "sampling": "source does not state", "judge": "benchmark-specified", "harness": "official release evaluation", "prompt_style": "source does not state", "temperature": "source does not state", "context": "source does not state" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-5.6 release table (table 5): GPT‑5.6 Luna; PostTrainBench Lite=29.6. Distinct Lite variant from the existing PostTrainBench row." }, { "model_id": "gpt-5.5", "benchmark_id": "posttrain_bench_lite", "score": 38.8, "reference_url": "https://openai.com/index/gpt-5-6/", "reported_setting": { "mode": "thinking", "effort": "source does not state", "tools": "none", "sampling": "source does not state", "judge": "benchmark-specified", "harness": "official release evaluation", "prompt_style": "source does not state", "temperature": "source does not state", "context": "source does not state" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-5.6 release table (table 5): GPT‑5.5; PostTrainBench Lite=38.8. Distinct Lite variant from the existing PostTrainBench row." }, { "model_id": "gpt-5.6-sol", "benchmark_id": "rsi_index", "score": 57.9, "reference_url": "https://openai.com/index/gpt-5-6/", "reported_setting": { "effort": "max", "eval_variant": null, "code_mode": "false" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-5.6 release table (table 5): GPT‑5.6 Sol; RSI Index=57.9. Exact effort/variant resolved from the rendered chart point.", "candidates": [ { "score": 28.1, "reference_url": "https://openai.com/index/gpt-5-6/", "source_type": "official_blog", "reported_setting": { "effort": "low", "eval_variant": null, "code_mode": "false" }, "notes": "OpenAI GPT-5.6 release figure (RSI Index): GPT-5.6 Sol; RSI Index=28.1. Figure 29: Aggregate capability on evaluations related to recursive self-improvement.. Cost, latency, and output-token coordinates are retained as review metadata, not score-matrix metrics." }, { "score": 41.15, "reference_url": "https://openai.com/index/gpt-5-6/", "source_type": "official_blog", "reported_setting": { "effort": "medium", "eval_variant": null, "code_mode": "false" }, "notes": "OpenAI GPT-5.6 release figure (RSI Index): GPT-5.6 Sol; RSI Index=41.15. Figure 29: Aggregate capability on evaluations related to recursive self-improvement.. Cost, latency, and output-token coordinates are retained as review metadata, not score-matrix metrics." }, { "score": 46.74, "reference_url": "https://openai.com/index/gpt-5-6/", "source_type": "official_blog", "reported_setting": { "effort": "high", "eval_variant": null, "code_mode": "false" }, "notes": "OpenAI GPT-5.6 release figure (RSI Index): GPT-5.6 Sol; RSI Index=46.74. Figure 29: Aggregate capability on evaluations related to recursive self-improvement.. Cost, latency, and output-token coordinates are retained as review metadata, not score-matrix metrics." }, { "score": 48.7, "reference_url": "https://openai.com/index/gpt-5-6/", "source_type": "official_blog", "reported_setting": { "effort": "xhigh", "eval_variant": null, "code_mode": "false" }, "notes": "OpenAI GPT-5.6 release figure (RSI Index): GPT-5.6 Sol; RSI Index=48.7. Figure 29: Aggregate capability on evaluations related to recursive self-improvement.. Cost, latency, and output-token coordinates are retained as review metadata, not score-matrix metrics." } ] }, { "model_id": "gpt-5.6-terra", "benchmark_id": "rsi_index", "score": 56.3, "reference_url": "https://openai.com/index/gpt-5-6/", "reported_setting": { "effort": "max", "eval_variant": null, "code_mode": "false" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-5.6 release table (table 5): GPT‑5.6 Terra; RSI Index=56.3. Exact effort/variant resolved from the rendered chart point.", "candidates": [ { "score": 20.39, "reference_url": "https://openai.com/index/gpt-5-6/", "source_type": "official_blog", "reported_setting": { "effort": "low", "eval_variant": null, "code_mode": "false" }, "notes": "OpenAI GPT-5.6 release figure (RSI Index): GPT-5.6 Terra; RSI Index=20.39. Figure 29: Aggregate capability on evaluations related to recursive self-improvement.. Cost, latency, and output-token coordinates are retained as review metadata, not score-matrix metrics." }, { "score": 31.24, "reference_url": "https://openai.com/index/gpt-5-6/", "source_type": "official_blog", "reported_setting": { "effort": "medium", "eval_variant": null, "code_mode": "false" }, "notes": "OpenAI GPT-5.6 release figure (RSI Index): GPT-5.6 Terra; RSI Index=31.24. Figure 29: Aggregate capability on evaluations related to recursive self-improvement.. Cost, latency, and output-token coordinates are retained as review metadata, not score-matrix metrics." }, { "score": 41.92, "reference_url": "https://openai.com/index/gpt-5-6/", "source_type": "official_blog", "reported_setting": { "effort": "high", "eval_variant": null, "code_mode": "false" }, "notes": "OpenAI GPT-5.6 release figure (RSI Index): GPT-5.6 Terra; RSI Index=41.92. Figure 29: Aggregate capability on evaluations related to recursive self-improvement.. Cost, latency, and output-token coordinates are retained as review metadata, not score-matrix metrics." }, { "score": 46.49, "reference_url": "https://openai.com/index/gpt-5-6/", "source_type": "official_blog", "reported_setting": { "effort": "xhigh", "eval_variant": null, "code_mode": "false" }, "notes": "OpenAI GPT-5.6 release figure (RSI Index): GPT-5.6 Terra; RSI Index=46.49. Figure 29: Aggregate capability on evaluations related to recursive self-improvement.. Cost, latency, and output-token coordinates are retained as review metadata, not score-matrix metrics." } ] }, { "model_id": "gpt-5.6-luna", "benchmark_id": "rsi_index", "score": 41.9, "reference_url": "https://openai.com/index/gpt-5-6/", "reported_setting": { "effort": "max", "eval_variant": null, "code_mode": "false" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-5.6 release table (table 5): GPT‑5.6 Luna; RSI Index=41.9. Exact effort/variant resolved from the rendered chart point.", "candidates": [ { "score": 10.85, "reference_url": "https://openai.com/index/gpt-5-6/", "source_type": "official_blog", "reported_setting": { "effort": "low", "eval_variant": null, "code_mode": "false" }, "notes": "OpenAI GPT-5.6 release figure (RSI Index): GPT-5.6 Luna; RSI Index=10.85. Figure 29: Aggregate capability on evaluations related to recursive self-improvement.. Cost, latency, and output-token coordinates are retained as review metadata, not score-matrix metrics." }, { "score": 16.81, "reference_url": "https://openai.com/index/gpt-5-6/", "source_type": "official_blog", "reported_setting": { "effort": "medium", "eval_variant": null, "code_mode": "false" }, "notes": "OpenAI GPT-5.6 release figure (RSI Index): GPT-5.6 Luna; RSI Index=16.81. Figure 29: Aggregate capability on evaluations related to recursive self-improvement.. Cost, latency, and output-token coordinates are retained as review metadata, not score-matrix metrics." }, { "score": 30.51, "reference_url": "https://openai.com/index/gpt-5-6/", "source_type": "official_blog", "reported_setting": { "effort": "high", "eval_variant": null, "code_mode": "false" }, "notes": "OpenAI GPT-5.6 release figure (RSI Index): GPT-5.6 Luna; RSI Index=30.51. Figure 29: Aggregate capability on evaluations related to recursive self-improvement.. Cost, latency, and output-token coordinates are retained as review metadata, not score-matrix metrics." }, { "score": 38.45, "reference_url": "https://openai.com/index/gpt-5-6/", "source_type": "official_blog", "reported_setting": { "effort": "xhigh", "eval_variant": null, "code_mode": "false" }, "notes": "OpenAI GPT-5.6 release figure (RSI Index): GPT-5.6 Luna; RSI Index=38.45. Figure 29: Aggregate capability on evaluations related to recursive self-improvement.. Cost, latency, and output-token coordinates are retained as review metadata, not score-matrix metrics." } ] }, { "model_id": "gpt-5.5", "benchmark_id": "rsi_index", "score": 41.7, "reference_url": "https://openai.com/index/gpt-5-6/", "reported_setting": { "effort": "xhigh", "eval_variant": null, "code_mode": "false" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-5.6 release table (table 5): GPT‑5.5; RSI Index=41.7. Exact effort/variant resolved from the rendered chart point.", "candidates": [ { "score": 19.54, "reference_url": "https://openai.com/index/gpt-5-6/", "source_type": "official_blog", "reported_setting": { "effort": "low", "eval_variant": null, "code_mode": "false" }, "notes": "OpenAI GPT-5.6 release figure (RSI Index): GPT-5.5; RSI Index=19.54. Figure 29: Aggregate capability on evaluations related to recursive self-improvement.. Cost, latency, and output-token coordinates are retained as review metadata, not score-matrix metrics." }, { "score": 32.57, "reference_url": "https://openai.com/index/gpt-5-6/", "source_type": "official_blog", "reported_setting": { "effort": "medium", "eval_variant": null, "code_mode": "false" }, "notes": "OpenAI GPT-5.6 release figure (RSI Index): GPT-5.5; RSI Index=32.57. Figure 29: Aggregate capability on evaluations related to recursive self-improvement.. Cost, latency, and output-token coordinates are retained as review metadata, not score-matrix metrics." }, { "score": 36.91, "reference_url": "https://openai.com/index/gpt-5-6/", "source_type": "official_blog", "reported_setting": { "effort": "high", "eval_variant": null, "code_mode": "false" }, "notes": "OpenAI GPT-5.6 release figure (RSI Index): GPT-5.5; RSI Index=36.91. Figure 29: Aggregate capability on evaluations related to recursive self-improvement.. Cost, latency, and output-token coordinates are retained as review metadata, not score-matrix metrics." } ] }, { "model_id": "gpt-5.6-terra", "benchmark_id": "mmmu_pro", "score": 80.7, "reference_url": "https://openai.com/index/gpt-5-6/", "reported_setting": { "mode": "thinking", "effort": "source does not state", "tools": "none", "sampling": "source does not state", "judge": "benchmark-specified", "harness": "official release evaluation", "prompt_style": "source does not state", "temperature": "source does not state", "context": "source does not state" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-5.6 release table (table 6): GPT‑5.6 Terra; MMMU Pro (no tools)=80.7.", "candidates": [ { "score": 82.0, "reference_url": "https://openai.com/index/gpt-5-6/", "source_type": "official_blog", "reported_setting": { "mode": "thinking", "effort": "source does not state", "tools": "tools", "sampling": "source does not state", "judge": "benchmark-specified", "harness": "official release evaluation", "prompt_style": "source does not state", "temperature": "source does not state", "context": "source does not state" }, "notes": "OpenAI GPT-5.6 release table (table 6): GPT‑5.6 Terra; MMMU Pro (with tools)=82.0." } ] }, { "model_id": "gpt-5.6-luna", "benchmark_id": "mmmu_pro", "score": 78.4, "reference_url": "https://openai.com/index/gpt-5-6/", "reported_setting": { "mode": "thinking", "effort": "source does not state", "tools": "none", "sampling": "source does not state", "judge": "benchmark-specified", "harness": "official release evaluation", "prompt_style": "source does not state", "temperature": "source does not state", "context": "source does not state" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-5.6 release table (table 6): GPT‑5.6 Luna; MMMU Pro (no tools)=78.4.", "candidates": [ { "score": 79.5, "reference_url": "https://openai.com/index/gpt-5-6/", "source_type": "official_blog", "reported_setting": { "mode": "thinking", "effort": "source does not state", "tools": "tools", "sampling": "source does not state", "judge": "benchmark-specified", "harness": "official release evaluation", "prompt_style": "source does not state", "temperature": "source does not state", "context": "source does not state" }, "notes": "OpenAI GPT-5.6 release table (table 6): GPT‑5.6 Luna; MMMU Pro (with tools)=79.5." } ] }, { "model_id": "gpt-5.6-sol", "benchmark_id": "gdp_pdf", "score": 30.7, "reference_url": "https://openai.com/index/gpt-5-6/", "reported_setting": { "mode": "thinking", "effort": "source does not state", "tools": "none", "sampling": "source does not state", "judge": "benchmark-specified", "harness": "official release evaluation", "prompt_style": "source does not state", "temperature": "source does not state", "context": "source does not state" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-5.6 release table (table 6): GPT‑5.6 Sol; gdp.pdf=30.7." }, { "model_id": "gpt-5.6-terra", "benchmark_id": "gdp_pdf", "score": 24.7, "reference_url": "https://openai.com/index/gpt-5-6/", "reported_setting": { "mode": "thinking", "effort": "source does not state", "tools": "none", "sampling": "source does not state", "judge": "benchmark-specified", "harness": "official release evaluation", "prompt_style": "source does not state", "temperature": "source does not state", "context": "source does not state" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-5.6 release table (table 6): GPT‑5.6 Terra; gdp.pdf=24.7." }, { "model_id": "gpt-5.6-luna", "benchmark_id": "gdp_pdf", "score": 22.7, "reference_url": "https://openai.com/index/gpt-5-6/", "reported_setting": { "mode": "thinking", "effort": "source does not state", "tools": "none", "sampling": "source does not state", "judge": "benchmark-specified", "harness": "official release evaluation", "prompt_style": "source does not state", "temperature": "source does not state", "context": "source does not state" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-5.6 release table (table 6): GPT‑5.6 Luna; gdp.pdf=22.7." }, { "model_id": "gpt-5.5", "benchmark_id": "gdp_pdf", "score": 26.0, "reference_url": "https://openai.com/index/gpt-5-6/", "reported_setting": { "mode": "thinking", "effort": "source does not state", "tools": "none", "sampling": "source does not state", "judge": "benchmark-specified", "harness": "official release evaluation", "prompt_style": "source does not state", "temperature": "source does not state", "context": "source does not state" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-5.6 release table (table 6): GPT‑5.5; gdp.pdf=26.0." }, { "model_id": "claude-fable-5", "benchmark_id": "gdp_pdf", "score": 29.8, "reference_url": "https://openai.com/index/gpt-5-6/", "reported_setting": { "mode": "thinking", "effort": "source does not state", "tools": "none", "sampling": "source does not state", "judge": "benchmark-specified", "harness": "official release evaluation", "prompt_style": "source does not state", "temperature": "source does not state", "context": "source does not state" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-5.6 release table (table 6): Claude Fable 5; gdp.pdf=29.8." }, { "model_id": "claude-opus-4.8", "benchmark_id": "gdp_pdf", "score": 22.5, "reference_url": "https://openai.com/index/gpt-5-6/", "reported_setting": { "mode": "thinking", "effort": "source does not state", "tools": "none", "sampling": "source does not state", "judge": "benchmark-specified", "harness": "official release evaluation", "prompt_style": "source does not state", "temperature": "source does not state", "context": "source does not state" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-5.6 release table (table 6): Claude Opus 4.8; gdp.pdf=22.5." }, { "model_id": "gemini-3.1-pro", "benchmark_id": "gdp_pdf", "score": 16.7, "reference_url": "https://openai.com/index/gpt-5-6/", "reported_setting": { "mode": "thinking", "effort": "source does not state", "tools": "none", "sampling": "source does not state", "judge": "benchmark-specified", "harness": "official release evaluation", "prompt_style": "source does not state", "temperature": "source does not state", "context": "source does not state" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-5.6 release table (table 6): Gemini 3.1 Pro Preview; gdp.pdf=16.7." }, { "model_id": "gpt-5.6-terra", "benchmark_id": "gpqa_diamond", "score": 92.9, "reference_url": "https://openai.com/index/gpt-5-6/", "reported_setting": { "mode": "thinking", "effort": "source does not state", "tools": "none", "sampling": "source does not state", "judge": "benchmark-specified", "harness": "official release evaluation", "prompt_style": "source does not state", "temperature": "source does not state", "context": "source does not state" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-5.6 release table (table 7): GPT‑5.6 Terra; GPQA Diamond=92.9." }, { "model_id": "gpt-5.6-luna", "benchmark_id": "gpqa_diamond", "score": 92.3, "reference_url": "https://openai.com/index/gpt-5-6/", "reported_setting": { "mode": "thinking", "effort": "source does not state", "tools": "none", "sampling": "source does not state", "judge": "benchmark-specified", "harness": "official release evaluation", "prompt_style": "source does not state", "temperature": "source does not state", "context": "source does not state" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-5.6 release table (table 7): GPT‑5.6 Luna; GPQA Diamond=92.3." }, { "model_id": "gpt-5.6-sol", "benchmark_id": "frontiermath_tier_1_3_v2", "score": 89.0, "reference_url": "https://openai.com/index/gpt-5-6/", "reported_setting": { "mode": "thinking", "effort": "source does not state", "tools": "none", "sampling": "source does not state", "judge": "benchmark-specified", "harness": "official release evaluation", "prompt_style": "source does not state", "temperature": "source does not state", "context": "source does not state" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-5.6 release table (table 7): GPT‑5.6 Sol; FrontierMath Tier 1-3 (v2)=89.0. Version 2 is distinct from the existing FrontierMath row." }, { "model_id": "gpt-5.6-terra", "benchmark_id": "frontiermath_tier_1_3_v2", "score": 84.9, "reference_url": "https://openai.com/index/gpt-5-6/", "reported_setting": { "mode": "thinking", "effort": "source does not state", "tools": "none", "sampling": "source does not state", "judge": "benchmark-specified", "harness": "official release evaluation", "prompt_style": "source does not state", "temperature": "source does not state", "context": "source does not state" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-5.6 release table (table 7): GPT‑5.6 Terra; FrontierMath Tier 1-3 (v2)=84.9. Version 2 is distinct from the existing FrontierMath row." }, { "model_id": "gpt-5.6-luna", "benchmark_id": "frontiermath_tier_1_3_v2", "score": 78.6, "reference_url": "https://openai.com/index/gpt-5-6/", "reported_setting": { "mode": "thinking", "effort": "source does not state", "tools": "none", "sampling": "source does not state", "judge": "benchmark-specified", "harness": "official release evaluation", "prompt_style": "source does not state", "temperature": "source does not state", "context": "source does not state" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-5.6 release table (table 7): GPT‑5.6 Luna; FrontierMath Tier 1-3 (v2)=78.6. Version 2 is distinct from the existing FrontierMath row." }, { "model_id": "gpt-5.5", "benchmark_id": "frontiermath_tier_1_3_v2", "score": 85.3, "reference_url": "https://openai.com/index/gpt-5-6/", "reported_setting": { "mode": "thinking", "effort": "source does not state", "tools": "none", "sampling": "source does not state", "judge": "benchmark-specified", "harness": "official release evaluation", "prompt_style": "source does not state", "temperature": "source does not state", "context": "source does not state" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-5.6 release table (table 7): GPT‑5.5; FrontierMath Tier 1-3 (v2)=85.3. Version 2 is distinct from the existing FrontierMath row." }, { "model_id": "claude-fable-5", "benchmark_id": "frontiermath_tier_1_3_v2", "score": 87.0, "reference_url": "https://openai.com/index/gpt-5-6/", "reported_setting": { "mode": "thinking", "effort": "source does not state", "tools": "none", "sampling": "source does not state", "judge": "benchmark-specified", "harness": "official release evaluation", "prompt_style": "source does not state", "temperature": "source does not state", "context": "source does not state" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-5.6 release table (table 7): Claude Fable 5; FrontierMath Tier 1-3 (v2)=87.0. Version 2 is distinct from the existing FrontierMath row." }, { "model_id": "claude-opus-4.8", "benchmark_id": "frontiermath_tier_1_3_v2", "score": 80.0, "reference_url": "https://openai.com/index/gpt-5-6/", "reported_setting": { "mode": "thinking", "effort": "source does not state", "tools": "none", "sampling": "source does not state", "judge": "benchmark-specified", "harness": "official release evaluation", "prompt_style": "source does not state", "temperature": "source does not state", "context": "source does not state" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-5.6 release table (table 7): Claude Opus 4.8; FrontierMath Tier 1-3 (v2)=80.0. Version 2 is distinct from the existing FrontierMath row." }, { "model_id": "gemini-3.1-pro", "benchmark_id": "frontiermath_tier_1_3_v2", "score": 59.6, "reference_url": "https://openai.com/index/gpt-5-6/", "reported_setting": { "mode": "thinking", "effort": "source does not state", "tools": "none", "sampling": "source does not state", "judge": "benchmark-specified", "harness": "official release evaluation", "prompt_style": "source does not state", "temperature": "source does not state", "context": "source does not state" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-5.6 release table (table 7): Gemini 3.1 Pro Preview; FrontierMath Tier 1-3 (v2)=59.6. Version 2 is distinct from the existing FrontierMath row." }, { "model_id": "gpt-5.6-sol", "benchmark_id": "frontiermath_tier_4_v2", "score": 83.0, "reference_url": "https://openai.com/index/gpt-5-6/", "reported_setting": { "mode": "thinking", "effort": "source does not state", "tools": "none", "sampling": "source does not state", "judge": "benchmark-specified", "harness": "official release evaluation", "prompt_style": "source does not state", "temperature": "source does not state", "context": "source does not state" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-5.6 release table (table 7): GPT‑5.6 Sol; FrontierMath Tier 4 (v2)=83.0. Version 2 is distinct from the existing FrontierMath Tier 4 row." }, { "model_id": "gpt-5.6-terra", "benchmark_id": "frontiermath_tier_4_v2", "score": 68.3, "reference_url": "https://openai.com/index/gpt-5-6/", "reported_setting": { "mode": "thinking", "effort": "source does not state", "tools": "none", "sampling": "source does not state", "judge": "benchmark-specified", "harness": "official release evaluation", "prompt_style": "source does not state", "temperature": "source does not state", "context": "source does not state" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-5.6 release table (table 7): GPT‑5.6 Terra; FrontierMath Tier 4 (v2)=68.3. Version 2 is distinct from the existing FrontierMath Tier 4 row." }, { "model_id": "gpt-5.6-luna", "benchmark_id": "frontiermath_tier_4_v2", "score": 58.5, "reference_url": "https://openai.com/index/gpt-5-6/", "reported_setting": { "mode": "thinking", "effort": "source does not state", "tools": "none", "sampling": "source does not state", "judge": "benchmark-specified", "harness": "official release evaluation", "prompt_style": "source does not state", "temperature": "source does not state", "context": "source does not state" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-5.6 release table (table 7): GPT‑5.6 Luna; FrontierMath Tier 4 (v2)=58.5. Version 2 is distinct from the existing FrontierMath Tier 4 row." }, { "model_id": "gpt-5.5", "benchmark_id": "frontiermath_tier_4_v2", "score": 72.5, "reference_url": "https://openai.com/index/gpt-5-6/", "reported_setting": { "mode": "thinking", "effort": "source does not state", "tools": "none", "sampling": "source does not state", "judge": "benchmark-specified", "harness": "official release evaluation", "prompt_style": "source does not state", "temperature": "source does not state", "context": "source does not state" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-5.6 release table (table 7): GPT‑5.5; FrontierMath Tier 4 (v2)=72.5. Version 2 is distinct from the existing FrontierMath Tier 4 row." }, { "model_id": "claude-fable-5", "benchmark_id": "frontiermath_tier_4_v2", "score": 87.8, "reference_url": "https://openai.com/index/gpt-5-6/", "reported_setting": { "mode": "thinking", "effort": "source does not state", "tools": "none", "sampling": "source does not state", "judge": "benchmark-specified", "harness": "official release evaluation", "prompt_style": "source does not state", "temperature": "source does not state", "context": "source does not state" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-5.6 release table (table 7): Claude Fable 5; FrontierMath Tier 4 (v2)=87.8. Version 2 is distinct from the existing FrontierMath Tier 4 row." }, { "model_id": "claude-opus-4.8", "benchmark_id": "frontiermath_tier_4_v2", "score": 56.1, "reference_url": "https://openai.com/index/gpt-5-6/", "reported_setting": { "mode": "thinking", "effort": "source does not state", "tools": "none", "sampling": "source does not state", "judge": "benchmark-specified", "harness": "official release evaluation", "prompt_style": "source does not state", "temperature": "source does not state", "context": "source does not state" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-5.6 release table (table 7): Claude Opus 4.8; FrontierMath Tier 4 (v2)=56.1. Version 2 is distinct from the existing FrontierMath Tier 4 row." }, { "model_id": "gpt-5.6-terra", "benchmark_id": "automation_bench", "score": 15.2, "reference_url": "https://openai.com/index/gpt-5-6/", "reported_setting": { "effort": "max", "eval_variant": null, "code_mode": "false" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-5.6 release table (table 8): GPT‑5.6 Terra; AutomationBench=15.2. Exact effort/variant resolved from the rendered chart point.", "candidates": [ { "score": 4.1, "reference_url": "https://openai.com/index/gpt-5-6/", "source_type": "official_blog", "reported_setting": { "effort": "none", "eval_variant": null, "code_mode": "false" }, "notes": "OpenAI GPT-5.6 release figure (AutomationBench): GPT-5.6 Terra; AutomationBench=4.1. Figure 15: AutomationBench. Cost, latency, and output-token coordinates are retained as review metadata, not score-matrix metrics." }, { "score": 6.2, "reference_url": "https://openai.com/index/gpt-5-6/", "source_type": "official_blog", "reported_setting": { "effort": "low", "eval_variant": null, "code_mode": "false" }, "notes": "OpenAI GPT-5.6 release figure (AutomationBench): GPT-5.6 Terra; AutomationBench=6.2. Figure 15: AutomationBench. Cost, latency, and output-token coordinates are retained as review metadata, not score-matrix metrics." }, { "score": 9.3, "reference_url": "https://openai.com/index/gpt-5-6/", "source_type": "official_blog", "reported_setting": { "effort": "medium", "eval_variant": null, "code_mode": "false" }, "notes": "OpenAI GPT-5.6 release figure (AutomationBench): GPT-5.6 Terra; AutomationBench=9.3. Figure 15: AutomationBench. Cost, latency, and output-token coordinates are retained as review metadata, not score-matrix metrics." }, { "score": 10.0, "reference_url": "https://openai.com/index/gpt-5-6/", "source_type": "official_blog", "reported_setting": { "effort": "high", "eval_variant": null, "code_mode": "false" }, "notes": "OpenAI GPT-5.6 release figure (AutomationBench): GPT-5.6 Terra; AutomationBench=10.0. Figure 15: AutomationBench. Cost, latency, and output-token coordinates are retained as review metadata, not score-matrix metrics." }, { "score": 12.3, "reference_url": "https://openai.com/index/gpt-5-6/", "source_type": "official_blog", "reported_setting": { "effort": "xhigh", "eval_variant": null, "code_mode": "false" }, "notes": "OpenAI GPT-5.6 release figure (AutomationBench): GPT-5.6 Terra; AutomationBench=12.3. Figure 15: AutomationBench. Cost, latency, and output-token coordinates are retained as review metadata, not score-matrix metrics." } ] }, { "model_id": "gpt-5.6-luna", "benchmark_id": "automation_bench", "score": 14.9, "reference_url": "https://openai.com/index/gpt-5-6/", "reported_setting": { "effort": "max", "eval_variant": null, "code_mode": "false" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-5.6 release table (table 8): GPT‑5.6 Luna; AutomationBench=14.9. Exact effort/variant resolved from the rendered chart point.", "candidates": [ { "score": 1.4, "reference_url": "https://openai.com/index/gpt-5-6/", "source_type": "official_blog", "reported_setting": { "effort": "none", "eval_variant": null, "code_mode": "false" }, "notes": "OpenAI GPT-5.6 release figure (AutomationBench): GPT-5.6 Luna; AutomationBench=1.4. Figure 15: AutomationBench. Cost, latency, and output-token coordinates are retained as review metadata, not score-matrix metrics." }, { "score": 3.7, "reference_url": "https://openai.com/index/gpt-5-6/", "source_type": "official_blog", "reported_setting": { "effort": "low", "eval_variant": null, "code_mode": "false" }, "notes": "OpenAI GPT-5.6 release figure (AutomationBench): GPT-5.6 Luna; AutomationBench=3.7. Figure 15: AutomationBench. Cost, latency, and output-token coordinates are retained as review metadata, not score-matrix metrics." }, { "score": 6.2, "reference_url": "https://openai.com/index/gpt-5-6/", "source_type": "official_blog", "reported_setting": { "effort": "medium", "eval_variant": null, "code_mode": "false" }, "notes": "OpenAI GPT-5.6 release figure (AutomationBench): GPT-5.6 Luna; AutomationBench=6.2. Figure 15: AutomationBench. Cost, latency, and output-token coordinates are retained as review metadata, not score-matrix metrics." }, { "score": 10.2, "reference_url": "https://openai.com/index/gpt-5-6/", "source_type": "official_blog", "reported_setting": { "effort": "high", "eval_variant": null, "code_mode": "false" }, "notes": "OpenAI GPT-5.6 release figure (AutomationBench): GPT-5.6 Luna; AutomationBench=10.2. Figure 15: AutomationBench. Cost, latency, and output-token coordinates are retained as review metadata, not score-matrix metrics." }, { "score": 12.9, "reference_url": "https://openai.com/index/gpt-5-6/", "source_type": "official_blog", "reported_setting": { "effort": "xhigh", "eval_variant": null, "code_mode": "false" }, "notes": "OpenAI GPT-5.6 release figure (AutomationBench): GPT-5.6 Luna; AutomationBench=12.9. Figure 15: AutomationBench. Cost, latency, and output-token coordinates are retained as review metadata, not score-matrix metrics." } ] }, { "model_id": "gemini-3.5-flash", "benchmark_id": "automation_bench", "score": 14.5, "reference_url": "https://openai.com/index/gpt-5-6/", "reported_setting": { "effort": "high", "eval_variant": null, "code_mode": "false" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-5.6 release table (table 8): Gemini 3.5 Flash; AutomationBench=14.5. Exact effort/variant resolved from the rendered chart point." }, { "model_id": "gpt-5.6-sol", "benchmark_id": "toolathlon", "score": 58.0, "reference_url": "https://openai.com/index/gpt-5-6/", "reported_setting": { "mode": "thinking", "effort": "source does not state", "tools": "none", "sampling": "source does not state", "judge": "benchmark-specified", "harness": "official release evaluation", "prompt_style": "source does not state", "temperature": "source does not state", "context": "source does not state" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-5.6 release table (table 8): GPT‑5.6 Sol; Toolathlon=58.0." }, { "model_id": "gpt-5.6-terra", "benchmark_id": "toolathlon", "score": 53.1, "reference_url": "https://openai.com/index/gpt-5-6/", "reported_setting": { "mode": "thinking", "effort": "source does not state", "tools": "none", "sampling": "source does not state", "judge": "benchmark-specified", "harness": "official release evaluation", "prompt_style": "source does not state", "temperature": "source does not state", "context": "source does not state" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-5.6 release table (table 8): GPT‑5.6 Terra; Toolathlon=53.1." }, { "model_id": "gpt-5.6-luna", "benchmark_id": "toolathlon", "score": 53.4, "reference_url": "https://openai.com/index/gpt-5-6/", "reported_setting": { "mode": "thinking", "effort": "source does not state", "tools": "none", "sampling": "source does not state", "judge": "benchmark-specified", "harness": "official release evaluation", "prompt_style": "source does not state", "temperature": "source does not state", "context": "source does not state" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-5.6 release table (table 8): GPT‑5.6 Luna; Toolathlon=53.4." }, { "model_id": "claude-mythos", "benchmark_id": "toolathlon", "score": 61.1, "reference_url": "https://www.anthropic.com/claude-sonnet-5-system-card", "reported_setting": { "effort": "max", "tools": "agentic benchmark harness", "sampling": "3 trials/task", "harness": "toolathlon" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Anthropic Claude Sonnet 5 System Card Table 8.11.5.A, page 135: Claude Mythos Preview; Toolathlon [Pass@1]=61.1. Source setting: effort=max; tools=agentic benchmark harness; sampling=3 trials/task; harness=toolathlon.", "candidates": [ { "score": 61.1, "reference_url": "https://openai.com/index/gpt-5-6/", "source_type": "official_blog", "reported_setting": { "mode": "thinking", "effort": "source does not state", "tools": "none", "sampling": "source does not state", "judge": "benchmark-specified", "harness": "official release evaluation", "prompt_style": "source does not state", "temperature": "source does not state", "context": "source does not state" }, "notes": "OpenAI GPT-5.6 release table (table 8): Claude Mythos Preview; Toolathlon=61.1." } ] }, { "model_id": "claude-fable-5", "benchmark_id": "toolathlon", "score": 61.7, "reference_url": "https://www.anthropic.com/claude-sonnet-5-system-card", "reported_setting": { "effort": "max", "tools": "agentic benchmark harness", "sampling": "3 trials/task", "harness": "toolathlon" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Anthropic Claude Sonnet 5 System Card Table 8.11.5.A, page 135: Claude Fable 5; Toolathlon [Pass@1]=61.7. Source setting: effort=max; tools=agentic benchmark harness; sampling=3 trials/task; harness=toolathlon.", "candidates": [ { "score": 61.7, "reference_url": "https://openai.com/index/gpt-5-6/", "source_type": "official_blog", "reported_setting": { "mode": "thinking", "effort": "source does not state", "tools": "none", "sampling": "source does not state", "judge": "benchmark-specified", "harness": "official release evaluation", "prompt_style": "source does not state", "temperature": "source does not state", "context": "source does not state" }, "notes": "OpenAI GPT-5.6 release table (table 8): Claude Fable 5; Toolathlon=61.7." } ] }, { "model_id": "gpt-5.6-sol", "benchmark_id": "mrcr_v2_8needle_256k_512k", "score": 91.5, "reference_url": "https://openai.com/index/gpt-5-6/", "reported_setting": { "mode": "thinking", "effort": "source does not state", "tools": "none", "sampling": "source does not state", "judge": "benchmark-specified", "harness": "official release evaluation", "prompt_style": "source does not state", "temperature": "source does not state", "context": "source does not state" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-5.6 release table (table 9): GPT‑5.6 Sol; OpenAI MRCR v2 8-needle 256K-512K=91.5." }, { "model_id": "gpt-5.6-terra", "benchmark_id": "mrcr_v2_8needle_256k_512k", "score": 89.6, "reference_url": "https://openai.com/index/gpt-5-6/", "reported_setting": { "mode": "thinking", "effort": "source does not state", "tools": "none", "sampling": "source does not state", "judge": "benchmark-specified", "harness": "official release evaluation", "prompt_style": "source does not state", "temperature": "source does not state", "context": "source does not state" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-5.6 release table (table 9): GPT‑5.6 Terra; OpenAI MRCR v2 8-needle 256K-512K=89.6." }, { "model_id": "gpt-5.6-luna", "benchmark_id": "mrcr_v2_8needle_256k_512k", "score": 41.3, "reference_url": "https://openai.com/index/gpt-5-6/", "reported_setting": { "mode": "thinking", "effort": "source does not state", "tools": "none", "sampling": "source does not state", "judge": "benchmark-specified", "harness": "official release evaluation", "prompt_style": "source does not state", "temperature": "source does not state", "context": "source does not state" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-5.6 release table (table 9): GPT‑5.6 Luna; OpenAI MRCR v2 8-needle 256K-512K=41.3." }, { "model_id": "gpt-5.5", "benchmark_id": "mrcr_v2_8needle_256k_512k", "score": 81.5, "reference_url": "https://openai.com/index/gpt-5-6/", "reported_setting": { "mode": "thinking", "effort": "source does not state", "tools": "none", "sampling": "source does not state", "judge": "benchmark-specified", "harness": "official release evaluation", "prompt_style": "source does not state", "temperature": "source does not state", "context": "source does not state" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-5.6 release table (table 9): GPT‑5.5; OpenAI MRCR v2 8-needle 256K-512K=81.5." }, { "model_id": "gpt-5.6-sol", "benchmark_id": "mrcr_v2_8needle", "score": 73.8, "reference_url": "https://openai.com/index/gpt-5-6/", "reported_setting": { "mode": "thinking", "effort": "source does not state", "tools": "none", "sampling": "source does not state", "judge": "benchmark-specified", "harness": "official release evaluation", "prompt_style": "source does not state", "temperature": "source does not state", "context": "source does not state" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-5.6 release table (table 9): GPT‑5.6 Sol; OpenAI MRCR v2 8-needle 512K-1M=73.8. The existing generic 8-needle row already contains GPT-5.5=74.0, which exactly matches this 512K-1M source row." }, { "model_id": "gpt-5.6-terra", "benchmark_id": "mrcr_v2_8needle", "score": 72.5, "reference_url": "https://openai.com/index/gpt-5-6/", "reported_setting": { "mode": "thinking", "effort": "source does not state", "tools": "none", "sampling": "source does not state", "judge": "benchmark-specified", "harness": "official release evaluation", "prompt_style": "source does not state", "temperature": "source does not state", "context": "source does not state" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-5.6 release table (table 9): GPT‑5.6 Terra; OpenAI MRCR v2 8-needle 512K-1M=72.5. The existing generic 8-needle row already contains GPT-5.5=74.0, which exactly matches this 512K-1M source row." }, { "model_id": "gpt-5.6-luna", "benchmark_id": "mrcr_v2_8needle", "score": 41.3, "reference_url": "https://openai.com/index/gpt-5-6/", "reported_setting": { "mode": "thinking", "effort": "source does not state", "tools": "none", "sampling": "source does not state", "judge": "benchmark-specified", "harness": "official release evaluation", "prompt_style": "source does not state", "temperature": "source does not state", "context": "source does not state" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-5.6 release table (table 9): GPT‑5.6 Luna; OpenAI MRCR v2 8-needle 512K-1M=41.3. The existing generic 8-needle row already contains GPT-5.5=74.0, which exactly matches this 512K-1M source row." }, { "model_id": "gpt-5.6-sol", "benchmark_id": "graphwalks_bfs_256k", "score": 90.7, "reference_url": "https://openai.com/index/gpt-5-6/", "reported_setting": { "mode": "thinking", "effort": "source does not state", "tools": "none", "sampling": "source does not state", "judge": "benchmark-specified", "harness": "official release evaluation", "prompt_style": "source does not state", "temperature": "source does not state", "context": "source does not state" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-5.6 release table (table 9): GPT‑5.6 Sol; GraphWalks BFS 256k f1=90.7. Distinct point from the existing 256K-1M averaged row." }, { "model_id": "gpt-5.6-terra", "benchmark_id": "graphwalks_bfs_256k", "score": 76.9, "reference_url": "https://openai.com/index/gpt-5-6/", "reported_setting": { "mode": "thinking", "effort": "source does not state", "tools": "none", "sampling": "source does not state", "judge": "benchmark-specified", "harness": "official release evaluation", "prompt_style": "source does not state", "temperature": "source does not state", "context": "source does not state" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-5.6 release table (table 9): GPT‑5.6 Terra; GraphWalks BFS 256k f1=76.9. Distinct point from the existing 256K-1M averaged row." }, { "model_id": "gpt-5.6-luna", "benchmark_id": "graphwalks_bfs_256k", "score": 81.3, "reference_url": "https://openai.com/index/gpt-5-6/", "reported_setting": { "mode": "thinking", "effort": "source does not state", "tools": "none", "sampling": "source does not state", "judge": "benchmark-specified", "harness": "official release evaluation", "prompt_style": "source does not state", "temperature": "source does not state", "context": "source does not state" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-5.6 release table (table 9): GPT‑5.6 Luna; GraphWalks BFS 256k f1=81.3. Distinct point from the existing 256K-1M averaged row." }, { "model_id": "gpt-5.5", "benchmark_id": "graphwalks_bfs_256k", "score": 73.7, "reference_url": "https://openai.com/index/gpt-5-6/", "reported_setting": { "mode": "thinking", "effort": "source does not state", "tools": "none", "sampling": "source does not state", "judge": "benchmark-specified", "harness": "official release evaluation", "prompt_style": "source does not state", "temperature": "source does not state", "context": "source does not state" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-5.6 release table (table 9): GPT‑5.5; GraphWalks BFS 256k f1=73.7. Distinct point from the existing 256K-1M averaged row." }, { "model_id": "claude-fable-5", "benchmark_id": "graphwalks_bfs_256k", "score": 91.1, "reference_url": "https://openai.com/index/gpt-5-6/", "reported_setting": { "mode": "thinking", "effort": "source does not state", "tools": "none", "sampling": "source does not state", "judge": "benchmark-specified", "harness": "official release evaluation", "prompt_style": "source does not state", "temperature": "source does not state", "context": "source does not state", "deployment": "claude-mythos-5", "safeguards": "lifted for Project Glasswing", "notes": "Anthropic documents Mythos 5 as the same underlying model and capabilities as Fable 5, without Fable's safety classifiers." }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-5.6 release table (table 9): Claude Mythos 5; GraphWalks BFS 256k f1=91.1. Distinct point from the existing 256K-1M averaged row." }, { "model_id": "claude-mythos", "benchmark_id": "graphwalks_bfs_256k", "score": 85.7, "reference_url": "https://openai.com/index/gpt-5-6/", "reported_setting": { "mode": "thinking", "effort": "source does not state", "tools": "none", "sampling": "source does not state", "judge": "benchmark-specified", "harness": "official release evaluation", "prompt_style": "source does not state", "temperature": "source does not state", "context": "source does not state" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-5.6 release table (table 9): Claude Mythos Preview; GraphWalks BFS 256k f1=85.7. Distinct point from the existing 256K-1M averaged row." }, { "model_id": "claude-opus-4.8", "benchmark_id": "graphwalks_bfs_256k", "score": 85.9, "reference_url": "https://openai.com/index/gpt-5-6/", "reported_setting": { "mode": "thinking", "effort": "source does not state", "tools": "none", "sampling": "source does not state", "judge": "benchmark-specified", "harness": "official release evaluation", "prompt_style": "source does not state", "temperature": "source does not state", "context": "source does not state" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-5.6 release table (table 9): Claude Opus 4.8; GraphWalks BFS 256k f1=85.9. Distinct point from the existing 256K-1M averaged row." }, { "model_id": "gpt-5.6-sol", "benchmark_id": "graphwalks_bfs_1m", "score": 77.1, "reference_url": "https://openai.com/index/gpt-5-6/", "reported_setting": { "mode": "thinking", "effort": "source does not state", "tools": "none", "sampling": "source does not state", "judge": "benchmark-specified", "harness": "official release evaluation", "prompt_style": "source does not state", "temperature": "source does not state", "context": "source does not state" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-5.6 release table (table 9): GPT‑5.6 Sol; GraphWalks BFS 1mil f1=77.1. Distinct point from the existing 256K-1M averaged row." }, { "model_id": "gpt-5.6-terra", "benchmark_id": "graphwalks_bfs_1m", "score": 71.2, "reference_url": "https://openai.com/index/gpt-5-6/", "reported_setting": { "mode": "thinking", "effort": "source does not state", "tools": "none", "sampling": "source does not state", "judge": "benchmark-specified", "harness": "official release evaluation", "prompt_style": "source does not state", "temperature": "source does not state", "context": "source does not state" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-5.6 release table (table 9): GPT‑5.6 Terra; GraphWalks BFS 1mil f1=71.2. Distinct point from the existing 256K-1M averaged row." }, { "model_id": "gpt-5.6-luna", "benchmark_id": "graphwalks_bfs_1m", "score": 51.2, "reference_url": "https://openai.com/index/gpt-5-6/", "reported_setting": { "mode": "thinking", "effort": "source does not state", "tools": "none", "sampling": "source does not state", "judge": "benchmark-specified", "harness": "official release evaluation", "prompt_style": "source does not state", "temperature": "source does not state", "context": "source does not state" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-5.6 release table (table 9): GPT‑5.6 Luna; GraphWalks BFS 1mil f1=51.2. Distinct point from the existing 256K-1M averaged row." }, { "model_id": "gpt-5.5", "benchmark_id": "graphwalks_bfs_1m", "score": 45.4, "reference_url": "https://openai.com/index/gpt-5-6/", "reported_setting": { "mode": "thinking", "effort": "source does not state", "tools": "none", "sampling": "source does not state", "judge": "benchmark-specified", "harness": "official release evaluation", "prompt_style": "source does not state", "temperature": "source does not state", "context": "source does not state" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-5.6 release table (table 9): GPT‑5.5; GraphWalks BFS 1mil f1=45.4. Distinct point from the existing 256K-1M averaged row." }, { "model_id": "claude-fable-5", "benchmark_id": "graphwalks_bfs_1m", "score": 79.4, "reference_url": "https://openai.com/index/gpt-5-6/", "reported_setting": { "mode": "thinking", "effort": "source does not state", "tools": "none", "sampling": "source does not state", "judge": "benchmark-specified", "harness": "official release evaluation", "prompt_style": "source does not state", "temperature": "source does not state", "context": "source does not state", "deployment": "claude-mythos-5", "safeguards": "lifted for Project Glasswing", "notes": "Anthropic documents Mythos 5 as the same underlying model and capabilities as Fable 5, without Fable's safety classifiers." }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-5.6 release table (table 9): Claude Mythos 5; GraphWalks BFS 1mil f1=79.4. Distinct point from the existing 256K-1M averaged row." }, { "model_id": "claude-mythos", "benchmark_id": "graphwalks_bfs_1m", "score": 74.3, "reference_url": "https://openai.com/index/gpt-5-6/", "reported_setting": { "mode": "thinking", "effort": "source does not state", "tools": "none", "sampling": "source does not state", "judge": "benchmark-specified", "harness": "official release evaluation", "prompt_style": "source does not state", "temperature": "source does not state", "context": "source does not state" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-5.6 release table (table 9): Claude Mythos Preview; GraphWalks BFS 1mil f1=74.3. Distinct point from the existing 256K-1M averaged row." }, { "model_id": "claude-opus-4.8", "benchmark_id": "graphwalks_bfs_1m", "score": 68.1, "reference_url": "https://openai.com/index/gpt-5-6/", "reported_setting": { "mode": "thinking", "effort": "source does not state", "tools": "none", "sampling": "source does not state", "judge": "benchmark-specified", "harness": "official release evaluation", "prompt_style": "source does not state", "temperature": "source does not state", "context": "source does not state" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-5.6 release table (table 9): Claude Opus 4.8; GraphWalks BFS 1mil f1=68.1. Distinct point from the existing 256K-1M averaged row." }, { "model_id": "gpt-5.6-sol", "benchmark_id": "arc_agi_3", "score": 7.78, "reference_url": "https://openai.com/index/gpt-5-6/", "reported_setting": { "mode": "thinking", "effort": "source does not state", "tools": "none", "sampling": "source does not state", "judge": "benchmark-specified", "harness": "official release evaluation", "prompt_style": "source does not state", "temperature": "source does not state", "context": "source does not state" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-5.6 release table (table 10): GPT‑5.6 Sol; ARC-AGI-3⁷=7.78.", "candidates": [ { "score": 0.33, "reference_url": "https://arcprize.org/leaderboard", "source_type": "leaderboard", "reported_setting": { "effort": "low" }, "notes": "ARC Prize leaderboard / GPT-5.6 Sol (Low): Relative Human Action Efficiency; total cost $12805.6." }, { "score": 1.07, "reference_url": "https://arcprize.org/leaderboard", "source_type": "leaderboard", "reported_setting": { "effort": "medium" }, "notes": "ARC Prize leaderboard / GPT-5.6 Sol (Medium): Relative Human Action Efficiency; total cost $12971.2." }, { "score": 2.15, "reference_url": "https://arcprize.org/leaderboard", "source_type": "leaderboard", "reported_setting": { "effort": "high" }, "notes": "ARC Prize leaderboard / GPT-5.6 Sol (High): Relative Human Action Efficiency; total cost $15176.1." }, { "score": 6.99, "reference_url": "https://arcprize.org/leaderboard", "source_type": "leaderboard", "reported_setting": { "effort": "xhigh" }, "notes": "ARC Prize leaderboard / GPT-5.6 Sol (XHigh): Relative Human Action Efficiency; total cost $19216.4." } ] }, { "model_id": "gpt-5.6-terra", "benchmark_id": "arc_agi_3", "score": 0.8, "reference_url": "https://openai.com/index/gpt-5-6/", "reported_setting": { "mode": "thinking", "effort": "source does not state", "tools": "none", "sampling": "source does not state", "judge": "benchmark-specified", "harness": "official release evaluation", "prompt_style": "source does not state", "temperature": "source does not state", "context": "source does not state" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-5.6 release table (table 10): GPT‑5.6 Terra; ARC-AGI-3⁷=0.8.", "candidates": [ { "score": 0.01, "reference_url": "https://arcprize.org/leaderboard", "source_type": "leaderboard", "reported_setting": { "effort": "low" }, "notes": "ARC Prize leaderboard / GPT-5.6 Terra (Low): Relative Human Action Efficiency; total cost $5068.48." }, { "score": 0.08, "reference_url": "https://arcprize.org/leaderboard", "source_type": "leaderboard", "reported_setting": { "effort": "medium" }, "notes": "ARC Prize leaderboard / GPT-5.6 Terra (Medium): Relative Human Action Efficiency; total cost $5895.31." }, { "score": 0.49, "reference_url": "https://arcprize.org/leaderboard", "source_type": "leaderboard", "reported_setting": { "effort": "high" }, "notes": "ARC Prize leaderboard / GPT-5.6 Terra (High): Relative Human Action Efficiency; total cost $5881.39." }, { "score": 0.65, "reference_url": "https://arcprize.org/leaderboard", "source_type": "leaderboard", "reported_setting": { "effort": "xhigh" }, "notes": "ARC Prize leaderboard / GPT-5.6 Terra (XHigh): Relative Human Action Efficiency; total cost $6804.04." } ] }, { "model_id": "gpt-5.6-luna", "benchmark_id": "arc_agi_3", "score": 0.18, "reference_url": "https://openai.com/index/gpt-5-6/", "reported_setting": { "mode": "thinking", "effort": "source does not state", "tools": "none", "sampling": "source does not state", "judge": "benchmark-specified", "harness": "official release evaluation", "prompt_style": "source does not state", "temperature": "source does not state", "context": "source does not state" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-5.6 release table (table 10): GPT‑5.6 Luna; ARC-AGI-3⁷=0.18.", "candidates": [ { "score": 0.17, "reference_url": "https://arcprize.org/leaderboard", "source_type": "leaderboard", "reported_setting": { "effort": "low" }, "notes": "ARC Prize leaderboard / GPT-5.6 Luna (Low): Relative Human Action Efficiency; total cost $2307.49." }, { "score": 0.17, "reference_url": "https://arcprize.org/leaderboard", "source_type": "leaderboard", "reported_setting": { "effort": "medium" }, "notes": "ARC Prize leaderboard / GPT-5.6 Luna (Medium): Relative Human Action Efficiency; total cost $2379.13." }, { "score": 0.1, "reference_url": "https://arcprize.org/leaderboard", "source_type": "leaderboard", "reported_setting": { "effort": "high" }, "notes": "ARC Prize leaderboard / GPT-5.6 Luna (High): Relative Human Action Efficiency; total cost $2482.44." }, { "score": 0.02, "reference_url": "https://arcprize.org/leaderboard", "source_type": "leaderboard", "reported_setting": { "effort": "xhigh" }, "notes": "ARC Prize leaderboard / GPT-5.6 Luna (XHigh): Relative Human Action Efficiency; total cost $2801.59." } ] }, { "model_id": "gpt-5.5", "benchmark_id": "arc_agi_3", "score": 0.43, "reference_url": "https://openai.com/index/gpt-5-6/", "reported_setting": { "mode": "thinking", "effort": "source does not state", "tools": "none", "sampling": "source does not state", "judge": "benchmark-specified", "harness": "official release evaluation", "prompt_style": "source does not state", "temperature": "source does not state", "context": "source does not state" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-5.6 release table (table 10): GPT‑5.5; ARC-AGI-3⁷=0.43." }, { "model_id": "claude-opus-4.8", "benchmark_id": "arc_agi_3", "score": 1.52, "reference_url": "https://arcprize.org/leaderboard", "reported_setting": { "effort": "high" }, "matches_canonical": true, "source_type": "leaderboard", "audit_status": "verified", "notes": "ARC Prize leaderboard / Claude Opus 4.8 (High): Relative Human Action Efficiency; total cost $10000.", "candidates": [ { "score": 1.5, "reference_url": "https://openai.com/index/gpt-5-6/", "source_type": "official_blog", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "source does not state", "judge": "benchmark-specified", "harness": "official release evaluation", "prompt_style": "source does not state", "temperature": "source does not state", "context": "source does not state", "notes": "Only published ARC-AGI-3 result; source explicitly says high, not max." }, "notes": "OpenAI GPT-5.6 release table (table 10): Claude Opus 4.8; ARC-AGI-3⁷=1.5." } ] }, { "model_id": "gemini-3.1-pro", "benchmark_id": "arc_agi_3", "score": 0.42, "reference_url": "https://openai.com/index/gpt-5-6/", "reported_setting": { "mode": "thinking", "effort": "source does not state", "tools": "none", "sampling": "source does not state", "judge": "benchmark-specified", "harness": "official release evaluation", "prompt_style": "source does not state", "temperature": "source does not state", "context": "source does not state" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-5.6 release table (table 10): Gemini 3.1 Pro Preview; ARC-AGI-3⁷=0.42." }, { "model_id": "gpt-5.4", "benchmark_id": "automation_bench", "score": 7.3, "reference_url": "https://openai.com/index/gpt-5-6/", "reported_setting": { "effort": "xhigh", "eval_variant": null, "code_mode": "false" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-5.6 release figure (AutomationBench): GPT-5.4; AutomationBench=7.3. Figure 15: AutomationBench. Cost, latency, and output-token coordinates are retained as review metadata, not score-matrix metrics.", "candidates": [ { "score": 1.2, "reference_url": "https://openai.com/index/gpt-5-6/", "source_type": "official_blog", "reported_setting": { "effort": "none", "eval_variant": null, "code_mode": "false" }, "notes": "OpenAI GPT-5.6 release figure (AutomationBench): GPT-5.4; AutomationBench=1.2. Figure 15: AutomationBench. Cost, latency, and output-token coordinates are retained as review metadata, not score-matrix metrics." }, { "score": 3.5, "reference_url": "https://openai.com/index/gpt-5-6/", "source_type": "official_blog", "reported_setting": { "effort": "low", "eval_variant": null, "code_mode": "false" }, "notes": "OpenAI GPT-5.6 release figure (AutomationBench): GPT-5.4; AutomationBench=3.5. Figure 15: AutomationBench. Cost, latency, and output-token coordinates are retained as review metadata, not score-matrix metrics." }, { "score": 6.1, "reference_url": "https://openai.com/index/gpt-5-6/", "source_type": "official_blog", "reported_setting": { "effort": "medium", "eval_variant": null, "code_mode": "false" }, "notes": "OpenAI GPT-5.6 release figure (AutomationBench): GPT-5.4; AutomationBench=6.1. Figure 15: AutomationBench. Cost, latency, and output-token coordinates are retained as review metadata, not score-matrix metrics." }, { "score": 7.6, "reference_url": "https://openai.com/index/gpt-5-6/", "source_type": "official_blog", "reported_setting": { "effort": "high", "eval_variant": null, "code_mode": "false" }, "notes": "OpenAI GPT-5.6 release figure (AutomationBench): GPT-5.4; AutomationBench=7.6. Figure 15: AutomationBench. Cost, latency, and output-token coordinates are retained as review metadata, not score-matrix metrics." } ] }, { "model_id": "gpt-5.4", "benchmark_id": "exploitbench", "score": 38.0, "reference_url": "https://openai.com/index/gpt-5-6/", "reported_setting": { "effort": "xhigh", "eval_variant": null, "code_mode": "false" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-5.6 release figure (ExploitBench): GPT-5.4; ExploitBench=38.0. Figure 27: Progress from vulnerable code to arbitrary code execution.. Cost, latency, and output-token coordinates are retained as review metadata, not score-matrix metrics.", "candidates": [ { "score": 26.4, "reference_url": "https://openai.com/index/gpt-5-6/", "source_type": "official_blog", "reported_setting": { "effort": "low", "eval_variant": null, "code_mode": "false" }, "notes": "OpenAI GPT-5.6 release figure (ExploitBench): GPT-5.4; ExploitBench=26.4. Figure 27: Progress from vulnerable code to arbitrary code execution.. Cost, latency, and output-token coordinates are retained as review metadata, not score-matrix metrics." }, { "score": 30.9, "reference_url": "https://openai.com/index/gpt-5-6/", "source_type": "official_blog", "reported_setting": { "effort": "medium", "eval_variant": null, "code_mode": "false" }, "notes": "OpenAI GPT-5.6 release figure (ExploitBench): GPT-5.4; ExploitBench=30.9. Figure 27: Progress from vulnerable code to arbitrary code execution.. Cost, latency, and output-token coordinates are retained as review metadata, not score-matrix metrics." }, { "score": 33.4, "reference_url": "https://openai.com/index/gpt-5-6/", "source_type": "official_blog", "reported_setting": { "effort": "high", "eval_variant": null, "code_mode": "false" }, "notes": "OpenAI GPT-5.6 release figure (ExploitBench): GPT-5.4; ExploitBench=33.4. Figure 27: Progress from vulnerable code to arbitrary code execution.. Cost, latency, and output-token coordinates are retained as review metadata, not score-matrix metrics." } ] }, { "model_id": "gpt-5.4", "benchmark_id": "exploitgym", "score": 7.0, "reference_url": "https://openai.com/index/gpt-5-6/", "reported_setting": { "effort": "xhigh", "eval_variant": "2h", "code_mode": "false" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-5.6 release figure (ExploitGym): GPT-5.4 · 2h cap; ExploitGym=7.0. Figure 27: ExploitGym. Cost, latency, and output-token coordinates are retained as review metadata, not score-matrix metrics." }, { "model_id": "gpt-5.4", "benchmark_id": "sec_bench_pro", "score": 35.93, "reference_url": "https://openai.com/index/gpt-5-6/", "reported_setting": { "effort": "xhigh", "eval_variant": null, "code_mode": "false" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-5.6 release figure (SEC-Bench Pro): GPT-5.4; SEC-Bench Pro=35.93. Figure 27: SEC-Bench Pro. Cost, latency, and output-token coordinates are retained as review metadata, not score-matrix metrics.", "candidates": [ { "score": 12.57, "reference_url": "https://openai.com/index/gpt-5-6/", "source_type": "official_blog", "reported_setting": { "effort": "low", "eval_variant": null, "code_mode": "false" }, "notes": "OpenAI GPT-5.6 release figure (SEC-Bench Pro): GPT-5.4; SEC-Bench Pro=12.57. Figure 27: SEC-Bench Pro. Cost, latency, and output-token coordinates are retained as review metadata, not score-matrix metrics." }, { "score": 24.59, "reference_url": "https://openai.com/index/gpt-5-6/", "source_type": "official_blog", "reported_setting": { "effort": "medium", "eval_variant": null, "code_mode": "false" }, "notes": "OpenAI GPT-5.6 release figure (SEC-Bench Pro): GPT-5.4; SEC-Bench Pro=24.59. Figure 27: SEC-Bench Pro. Cost, latency, and output-token coordinates are retained as review metadata, not score-matrix metrics." }, { "score": 34.43, "reference_url": "https://openai.com/index/gpt-5-6/", "source_type": "official_blog", "reported_setting": { "effort": "high", "eval_variant": null, "code_mode": "false" }, "notes": "OpenAI GPT-5.6 release figure (SEC-Bench Pro): GPT-5.4; SEC-Bench Pro=34.43. Figure 27: SEC-Bench Pro. Cost, latency, and output-token coordinates are retained as review metadata, not score-matrix metrics." } ] }, { "model_id": "gpt-5.4", "benchmark_id": "genebench_pro", "score": 8.87, "reference_url": "https://openai.com/index/gpt-5-6/", "reported_setting": { "effort": "xhigh", "eval_variant": null, "code_mode": "false" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-5.6 release figure (GeneBench Pro): GPT-5.4; GeneBench Pro=8.87. Figure 28: Long-horizon genomics and quantitative-biology analyses.. Cost, latency, and output-token coordinates are retained as review metadata, not score-matrix metrics.", "candidates": [ { "score": 0.85, "reference_url": "https://openai.com/index/gpt-5-6/", "source_type": "official_blog", "reported_setting": { "effort": "none", "eval_variant": null, "code_mode": "false" }, "notes": "OpenAI GPT-5.6 release figure (GeneBench Pro): GPT-5.4; GeneBench Pro=0.85. Figure 28: Long-horizon genomics and quantitative-biology analyses.. Cost, latency, and output-token coordinates are retained as review metadata, not score-matrix metrics." }, { "score": 3.04, "reference_url": "https://openai.com/index/gpt-5-6/", "source_type": "official_blog", "reported_setting": { "effort": "low", "eval_variant": null, "code_mode": "false" }, "notes": "OpenAI GPT-5.6 release figure (GeneBench Pro): GPT-5.4; GeneBench Pro=3.04. Figure 28: Long-horizon genomics and quantitative-biology analyses.. Cost, latency, and output-token coordinates are retained as review metadata, not score-matrix metrics." }, { "score": 4.96, "reference_url": "https://openai.com/index/gpt-5-6/", "source_type": "official_blog", "reported_setting": { "effort": "medium", "eval_variant": null, "code_mode": "false" }, "notes": "OpenAI GPT-5.6 release figure (GeneBench Pro): GPT-5.4; GeneBench Pro=4.96. Figure 28: Long-horizon genomics and quantitative-biology analyses.. Cost, latency, and output-token coordinates are retained as review metadata, not score-matrix metrics." }, { "score": 6.99, "reference_url": "https://openai.com/index/gpt-5-6/", "source_type": "official_blog", "reported_setting": { "effort": "high", "eval_variant": null, "code_mode": "false" }, "notes": "OpenAI GPT-5.6 release figure (GeneBench Pro): GPT-5.4; GeneBench Pro=6.99. Figure 28: Long-horizon genomics and quantitative-biology analyses.. Cost, latency, and output-token coordinates are retained as review metadata, not score-matrix metrics." } ] }, { "model_id": "gpt-5.4", "benchmark_id": "lifescibench", "score": 44.17, "reference_url": "https://openai.com/index/gpt-5-6/", "reported_setting": { "effort": "xhigh", "eval_variant": null, "code_mode": "false" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-5.6 release figure (LifeSciBench): GPT-5.4; LifeSciBench=44.17. Figure 28: LifeSciBench. Cost, latency, and output-token coordinates are retained as review metadata, not score-matrix metrics.", "candidates": [ { "score": 36.06, "reference_url": "https://openai.com/index/gpt-5-6/", "source_type": "official_blog", "reported_setting": { "effort": "low", "eval_variant": null, "code_mode": "false" }, "notes": "OpenAI GPT-5.6 release figure (LifeSciBench): GPT-5.4; LifeSciBench=36.06. Figure 28: LifeSciBench. Cost, latency, and output-token coordinates are retained as review metadata, not score-matrix metrics." }, { "score": 39.33, "reference_url": "https://openai.com/index/gpt-5-6/", "source_type": "official_blog", "reported_setting": { "effort": "medium", "eval_variant": null, "code_mode": "false" }, "notes": "OpenAI GPT-5.6 release figure (LifeSciBench): GPT-5.4; LifeSciBench=39.33. Figure 28: LifeSciBench. Cost, latency, and output-token coordinates are retained as review metadata, not score-matrix metrics." }, { "score": 41.54, "reference_url": "https://openai.com/index/gpt-5-6/", "source_type": "official_blog", "reported_setting": { "effort": "high", "eval_variant": null, "code_mode": "false" }, "notes": "OpenAI GPT-5.6 release figure (LifeSciBench): GPT-5.4; LifeSciBench=41.54. Figure 28: LifeSciBench. Cost, latency, and output-token coordinates are retained as review metadata, not score-matrix metrics." } ] }, { "model_id": "gpt-5.4", "benchmark_id": "medchembench_internal", "score": 23.98, "reference_url": "https://openai.com/index/gpt-5-6/", "reported_setting": { "effort": "xhigh", "eval_variant": null, "code_mode": "false" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-5.6 release figure (MedChemBench): GPT-5.4; MedChemBench (Internal)=23.98. Figure 28: MedChemBench. Cost, latency, and output-token coordinates are retained as review metadata, not score-matrix metrics.", "candidates": [ { "score": 16.2, "reference_url": "https://openai.com/index/gpt-5-6/", "source_type": "official_blog", "reported_setting": { "effort": "low", "eval_variant": null, "code_mode": "false" }, "notes": "OpenAI GPT-5.6 release figure (MedChemBench): GPT-5.4; MedChemBench (Internal)=16.2. Figure 28: MedChemBench. Cost, latency, and output-token coordinates are retained as review metadata, not score-matrix metrics." }, { "score": 18.75, "reference_url": "https://openai.com/index/gpt-5-6/", "source_type": "official_blog", "reported_setting": { "effort": "medium", "eval_variant": null, "code_mode": "false" }, "notes": "OpenAI GPT-5.6 release figure (MedChemBench): GPT-5.4; MedChemBench (Internal)=18.75. Figure 28: MedChemBench. Cost, latency, and output-token coordinates are retained as review metadata, not score-matrix metrics." }, { "score": 20.95, "reference_url": "https://openai.com/index/gpt-5-6/", "source_type": "official_blog", "reported_setting": { "effort": "high", "eval_variant": null, "code_mode": "false" }, "notes": "OpenAI GPT-5.6 release figure (MedChemBench): GPT-5.4; MedChemBench (Internal)=20.95. Figure 28: MedChemBench. Cost, latency, and output-token coordinates are retained as review metadata, not score-matrix metrics." } ] }, { "model_id": "gpt-5.4", "benchmark_id": "rsi_index", "score": 36.36, "reference_url": "https://openai.com/index/gpt-5-6/", "reported_setting": { "effort": "xhigh", "eval_variant": null, "code_mode": "false" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-5.6 release figure (RSI Index): GPT-5.4; RSI Index=36.36. Figure 29: Aggregate capability on evaluations related to recursive self-improvement.. Cost, latency, and output-token coordinates are retained as review metadata, not score-matrix metrics.", "candidates": [ { "score": 31.33, "reference_url": "https://openai.com/index/gpt-5-6/", "source_type": "official_blog", "reported_setting": { "effort": "medium", "eval_variant": null, "code_mode": "false" }, "notes": "OpenAI GPT-5.6 release figure (RSI Index): GPT-5.4; RSI Index=31.33. Figure 29: Aggregate capability on evaluations related to recursive self-improvement.. Cost, latency, and output-token coordinates are retained as review metadata, not score-matrix metrics." }, { "score": 32.89, "reference_url": "https://openai.com/index/gpt-5-6/", "source_type": "official_blog", "reported_setting": { "effort": "high", "eval_variant": null, "code_mode": "false" }, "notes": "OpenAI GPT-5.6 release figure (RSI Index): GPT-5.4; RSI Index=32.89. Figure 29: Aggregate capability on evaluations related to recursive self-improvement.. Cost, latency, and output-token coordinates are retained as review metadata, not score-matrix metrics." } ] }, { "model_id": "gpt-5.4", "benchmark_id": "internal_research_debugging", "score": 47.71, "reference_url": "https://openai.com/index/gpt-5-6/", "reported_setting": { "effort": "xhigh", "eval_variant": null, "code_mode": "false" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-5.6 release figure (Internal Research Debugging Evaluation): GPT-5.4; Internal Research Debugging Evaluation=47.71. Figure 29: Internal Research Debugging Evaluation. Cost, latency, and output-token coordinates are retained as review metadata, not score-matrix metrics.", "candidates": [ { "score": 26.12, "reference_url": "https://openai.com/index/gpt-5-6/", "source_type": "official_blog", "reported_setting": { "effort": "low", "eval_variant": null, "code_mode": "false" }, "notes": "OpenAI GPT-5.6 release figure (Internal Research Debugging Evaluation): GPT-5.4; Internal Research Debugging Evaluation=26.12. Figure 29: Internal Research Debugging Evaluation. Cost, latency, and output-token coordinates are retained as review metadata, not score-matrix metrics." }, { "score": 38.64, "reference_url": "https://openai.com/index/gpt-5-6/", "source_type": "official_blog", "reported_setting": { "effort": "medium", "eval_variant": null, "code_mode": "false" }, "notes": "OpenAI GPT-5.6 release figure (Internal Research Debugging Evaluation): GPT-5.4; Internal Research Debugging Evaluation=38.64. Figure 29: Internal Research Debugging Evaluation. Cost, latency, and output-token coordinates are retained as review metadata, not score-matrix metrics." }, { "score": 44.02, "reference_url": "https://openai.com/index/gpt-5-6/", "source_type": "official_blog", "reported_setting": { "effort": "high", "eval_variant": null, "code_mode": "false" }, "notes": "OpenAI GPT-5.6 release figure (Internal Research Debugging Evaluation): GPT-5.4; Internal Research Debugging Evaluation=44.02. Figure 29: Internal Research Debugging Evaluation. Cost, latency, and output-token coordinates are retained as review metadata, not score-matrix metrics." } ] }, { "model_id": "gpt-5.4", "benchmark_id": "kernelgen_1p", "score": 17.96, "reference_url": "https://openai.com/index/gpt-5-6/", "reported_setting": { "effort": "xhigh", "eval_variant": null, "code_mode": "false" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-5.6 release figure (KernelGen 1P): GPT-5.4; KernelGen 1P=17.96. Figure 29: KernelGen 1P. Cost, latency, and output-token coordinates are retained as review metadata, not score-matrix metrics.", "candidates": [ { "score": 1.13, "reference_url": "https://openai.com/index/gpt-5-6/", "source_type": "official_blog", "reported_setting": { "effort": "low", "eval_variant": null, "code_mode": "false" }, "notes": "OpenAI GPT-5.6 release figure (KernelGen 1P): GPT-5.4; KernelGen 1P=1.13. Figure 29: KernelGen 1P. Cost, latency, and output-token coordinates are retained as review metadata, not score-matrix metrics." }, { "score": 6.33, "reference_url": "https://openai.com/index/gpt-5-6/", "source_type": "official_blog", "reported_setting": { "effort": "medium", "eval_variant": null, "code_mode": "false" }, "notes": "OpenAI GPT-5.6 release figure (KernelGen 1P): GPT-5.4; KernelGen 1P=6.33. Figure 29: KernelGen 1P. Cost, latency, and output-token coordinates are retained as review metadata, not score-matrix metrics." }, { "score": 11.29, "reference_url": "https://openai.com/index/gpt-5-6/", "source_type": "official_blog", "reported_setting": { "effort": "high", "eval_variant": null, "code_mode": "false" }, "notes": "OpenAI GPT-5.6 release figure (KernelGen 1P): GPT-5.4; KernelGen 1P=11.29. Figure 29: KernelGen 1P. Cost, latency, and output-token coordinates are retained as review metadata, not score-matrix metrics." } ] }, { "model_id": "gpt-5.4", "benchmark_id": "nanogpt_self_improvement", "score": 2.3, "reference_url": "https://openai.com/index/gpt-5-6/", "reported_setting": { "effort": "xhigh", "eval_variant": null, "code_mode": "false" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "OpenAI GPT-5.6 release figure (NanoGPT): GPT-5.4; NanoGPT=2.3. Figure 29: NanoGPT. Cost, latency, and output-token coordinates are retained as review metadata, not score-matrix metrics.", "candidates": [ { "score": 0.3, "reference_url": "https://openai.com/index/gpt-5-6/", "source_type": "official_blog", "reported_setting": { "effort": "low", "eval_variant": null, "code_mode": "false" }, "notes": "OpenAI GPT-5.6 release figure (NanoGPT): GPT-5.4; NanoGPT=0.3. Figure 29: NanoGPT. Cost, latency, and output-token coordinates are retained as review metadata, not score-matrix metrics." }, { "score": 0.53, "reference_url": "https://openai.com/index/gpt-5-6/", "source_type": "official_blog", "reported_setting": { "effort": "medium", "eval_variant": null, "code_mode": "false" }, "notes": "OpenAI GPT-5.6 release figure (NanoGPT): GPT-5.4; NanoGPT=0.53. Figure 29: NanoGPT. Cost, latency, and output-token coordinates are retained as review metadata, not score-matrix metrics." }, { "score": 0.87, "reference_url": "https://openai.com/index/gpt-5-6/", "source_type": "official_blog", "reported_setting": { "effort": "high", "eval_variant": null, "code_mode": "false" }, "notes": "OpenAI GPT-5.6 release figure (NanoGPT): GPT-5.4; NanoGPT=0.87. Figure 29: NanoGPT. Cost, latency, and output-token coordinates are retained as review metadata, not score-matrix metrics." } ] }, { "model_id": "gpt-5", "benchmark_id": "healthbench_professional_length_adjusted", "score": 46.2, "reference_url": "https://deploymentsafety.openai.com/gpt-5-6", "reported_setting": { "effort": "source does not state", "tools": "none", "harness": "OpenAI official HealthBench scoring" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "OpenAI GPT-5.6 System Card table (HealthBench length-adjusted scores): gpt-5; HealthBench Professional length-adjusted=46.2. System Card Table 6. Adjusted and unadjusted metrics are kept as distinct benchmark identities." }, { "model_id": "gpt-5", "benchmark_id": "healthbench_professional", "score": 51.0, "reference_url": "https://deploymentsafety.openai.com/gpt-5-6", "reported_setting": { "effort": "source does not state", "tools": "none", "harness": "OpenAI official HealthBench scoring" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "OpenAI GPT-5.6 System Card table (HealthBench length-adjusted scores): gpt-5; HealthBench Professional unadjusted=51.0. System Card Table 6. Adjusted and unadjusted metrics are kept as distinct benchmark identities." }, { "model_id": "gpt-5.1", "benchmark_id": "healthbench_professional_length_adjusted", "score": 39.6, "reference_url": "https://deploymentsafety.openai.com/gpt-5-6", "reported_setting": { "effort": "source does not state", "tools": "none", "harness": "OpenAI official HealthBench scoring" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "OpenAI GPT-5.6 System Card table (HealthBench length-adjusted scores): gpt-5.1; HealthBench Professional length-adjusted=39.6. System Card Table 6. Adjusted and unadjusted metrics are kept as distinct benchmark identities." }, { "model_id": "gpt-5.1", "benchmark_id": "healthbench_professional", "score": 48.0, "reference_url": "https://deploymentsafety.openai.com/gpt-5-6", "reported_setting": { "effort": "source does not state", "tools": "none", "harness": "OpenAI official HealthBench scoring" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "OpenAI GPT-5.6 System Card table (HealthBench length-adjusted scores): gpt-5.1; HealthBench Professional unadjusted=48.0. System Card Table 6. Adjusted and unadjusted metrics are kept as distinct benchmark identities." }, { "model_id": "gpt-5.2", "benchmark_id": "healthbench_professional_length_adjusted", "score": 45.9, "reference_url": "https://deploymentsafety.openai.com/gpt-5-6", "reported_setting": { "effort": "source does not state", "tools": "none", "harness": "OpenAI official HealthBench scoring" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "OpenAI GPT-5.6 System Card table (HealthBench length-adjusted scores): gpt-5.2; HealthBench Professional length-adjusted=45.9. System Card Table 6. Adjusted and unadjusted metrics are kept as distinct benchmark identities." }, { "model_id": "gpt-5.2", "benchmark_id": "healthbench_professional", "score": 50.0, "reference_url": "https://deploymentsafety.openai.com/gpt-5-6", "reported_setting": { "effort": "source does not state", "tools": "none", "harness": "OpenAI official HealthBench scoring" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "OpenAI GPT-5.6 System Card table (HealthBench length-adjusted scores): gpt-5.2; HealthBench Professional unadjusted=50.0. System Card Table 6. Adjusted and unadjusted metrics are kept as distinct benchmark identities." }, { "model_id": "gpt-5.4", "benchmark_id": "healthbench_professional_length_adjusted", "score": 48.1, "reference_url": "https://deploymentsafety.openai.com/gpt-5-6", "reported_setting": { "effort": "source does not state", "tools": "none", "harness": "OpenAI official HealthBench scoring" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "OpenAI GPT-5.6 System Card table (HealthBench length-adjusted scores): gpt-5.4; HealthBench Professional length-adjusted=48.1. System Card Table 6. Adjusted and unadjusted metrics are kept as distinct benchmark identities." }, { "model_id": "gpt-5.4", "benchmark_id": "healthbench_professional", "score": 51.9, "reference_url": "https://deploymentsafety.openai.com/gpt-5-6", "reported_setting": { "effort": "source does not state", "tools": "none", "harness": "OpenAI official HealthBench scoring" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "OpenAI GPT-5.6 System Card table (HealthBench length-adjusted scores): gpt-5.4; HealthBench Professional unadjusted=51.9. System Card Table 6. Adjusted and unadjusted metrics are kept as distinct benchmark identities." }, { "model_id": "gpt-5.5", "benchmark_id": "healthbench_professional", "score": 57.2, "reference_url": "https://deploymentsafety.openai.com/gpt-5-6", "reported_setting": { "effort": "source does not state", "tools": "none", "harness": "OpenAI official HealthBench scoring" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "OpenAI GPT-5.6 System Card table (HealthBench length-adjusted scores): gpt-5.5; HealthBench Professional unadjusted=57.2. System Card Table 6. Adjusted and unadjusted metrics are kept as distinct benchmark identities." }, { "model_id": "gpt-5.6-sol", "benchmark_id": "healthbench_professional", "score": 64.1, "reference_url": "https://deploymentsafety.openai.com/gpt-5-6", "reported_setting": { "effort": "source does not state", "tools": "none", "harness": "OpenAI official HealthBench scoring" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "OpenAI GPT-5.6 System Card table (HealthBench length-adjusted scores): gpt-5.6-sol; HealthBench Professional unadjusted=64.1. System Card Table 6. Adjusted and unadjusted metrics are kept as distinct benchmark identities." }, { "model_id": "gpt-5.6-terra", "benchmark_id": "healthbench_professional", "score": 62.4, "reference_url": "https://deploymentsafety.openai.com/gpt-5-6", "reported_setting": { "effort": "source does not state", "tools": "none", "harness": "OpenAI official HealthBench scoring" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "OpenAI GPT-5.6 System Card table (HealthBench length-adjusted scores): gpt-5.6-terra; HealthBench Professional unadjusted=62.4. System Card Table 6. Adjusted and unadjusted metrics are kept as distinct benchmark identities." }, { "model_id": "gpt-5.6-luna", "benchmark_id": "healthbench_professional", "score": 59.8, "reference_url": "https://deploymentsafety.openai.com/gpt-5-6", "reported_setting": { "effort": "source does not state", "tools": "none", "harness": "OpenAI official HealthBench scoring" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "OpenAI GPT-5.6 System Card table (HealthBench length-adjusted scores): gpt-5.6-luna; HealthBench Professional unadjusted=59.8. System Card Table 6. Adjusted and unadjusted metrics are kept as distinct benchmark identities." }, { "model_id": "gpt-5", "benchmark_id": "healthbench_length_adjusted", "score": 57.7, "reference_url": "https://deploymentsafety.openai.com/gpt-5-6", "reported_setting": { "effort": "source does not state", "tools": "none", "harness": "OpenAI official HealthBench scoring" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "OpenAI GPT-5.6 System Card table (HealthBench length-adjusted scores): gpt-5; HealthBench length-adjusted=57.7. System Card Table 6. Adjusted and unadjusted metrics are kept as distinct benchmark identities." }, { "model_id": "gpt-5.1", "benchmark_id": "healthbench_length_adjusted", "score": 50.9, "reference_url": "https://deploymentsafety.openai.com/gpt-5-6", "reported_setting": { "effort": "source does not state", "tools": "none", "harness": "OpenAI official HealthBench scoring" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "OpenAI GPT-5.6 System Card table (HealthBench length-adjusted scores): gpt-5.1; HealthBench length-adjusted=50.9. System Card Table 6. Adjusted and unadjusted metrics are kept as distinct benchmark identities." }, { "model_id": "gpt-5.1", "benchmark_id": "healthbench", "score": 64.2, "reference_url": "https://deploymentsafety.openai.com/gpt-5-6", "reported_setting": { "effort": "source does not state", "tools": "none", "harness": "OpenAI official HealthBench scoring" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "OpenAI GPT-5.6 System Card table (HealthBench length-adjusted scores): gpt-5.1; HealthBench unadjusted=64.2. System Card Table 6. Adjusted and unadjusted metrics are kept as distinct benchmark identities." }, { "model_id": "gpt-5.2", "benchmark_id": "healthbench_length_adjusted", "score": 56.8, "reference_url": "https://deploymentsafety.openai.com/gpt-5-6", "reported_setting": { "effort": "source does not state", "tools": "none", "harness": "OpenAI official HealthBench scoring" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "OpenAI GPT-5.6 System Card table (HealthBench length-adjusted scores): gpt-5.2; HealthBench length-adjusted=56.8. System Card Table 6. Adjusted and unadjusted metrics are kept as distinct benchmark identities." }, { "model_id": "gpt-5.4", "benchmark_id": "healthbench_length_adjusted", "score": 54.0, "reference_url": "https://deploymentsafety.openai.com/gpt-5-6", "reported_setting": { "effort": "source does not state", "tools": "none", "harness": "OpenAI official HealthBench scoring" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "OpenAI GPT-5.6 System Card table (HealthBench length-adjusted scores): gpt-5.4; HealthBench length-adjusted=54.0. System Card Table 6. Adjusted and unadjusted metrics are kept as distinct benchmark identities." }, { "model_id": "gpt-5.4", "benchmark_id": "healthbench", "score": 55.7, "reference_url": "https://deploymentsafety.openai.com/gpt-5-6", "reported_setting": { "effort": "source does not state", "tools": "none", "harness": "OpenAI official HealthBench scoring" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "OpenAI GPT-5.6 System Card table (HealthBench length-adjusted scores): gpt-5.4; HealthBench unadjusted=55.7. System Card Table 6. Adjusted and unadjusted metrics are kept as distinct benchmark identities." }, { "model_id": "gpt-5.5", "benchmark_id": "healthbench_length_adjusted", "score": 56.5, "reference_url": "https://deploymentsafety.openai.com/gpt-5-6", "reported_setting": { "effort": "source does not state", "tools": "none", "harness": "OpenAI official HealthBench scoring" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "OpenAI GPT-5.6 System Card table (HealthBench length-adjusted scores): gpt-5.5; HealthBench length-adjusted=56.5. System Card Table 6. Adjusted and unadjusted metrics are kept as distinct benchmark identities." }, { "model_id": "gpt-5.5", "benchmark_id": "healthbench", "score": 58.4, "reference_url": "https://deploymentsafety.openai.com/gpt-5-6", "reported_setting": { "effort": "source does not state", "tools": "none", "harness": "OpenAI official HealthBench scoring" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "OpenAI GPT-5.6 System Card table (HealthBench length-adjusted scores): gpt-5.5; HealthBench unadjusted=58.4. System Card Table 6. Adjusted and unadjusted metrics are kept as distinct benchmark identities." }, { "model_id": "gpt-5.6-sol", "benchmark_id": "healthbench_length_adjusted", "score": 57.0, "reference_url": "https://deploymentsafety.openai.com/gpt-5-6", "reported_setting": { "effort": "source does not state", "tools": "none", "harness": "OpenAI official HealthBench scoring" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "OpenAI GPT-5.6 System Card table (HealthBench length-adjusted scores): gpt-5.6-sol; HealthBench length-adjusted=57.0. System Card Table 6. Adjusted and unadjusted metrics are kept as distinct benchmark identities." }, { "model_id": "gpt-5.6-sol", "benchmark_id": "healthbench", "score": 55.6, "reference_url": "https://deploymentsafety.openai.com/gpt-5-6", "reported_setting": { "effort": "source does not state", "tools": "none", "harness": "OpenAI official HealthBench scoring" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "OpenAI GPT-5.6 System Card table (HealthBench length-adjusted scores): gpt-5.6-sol; HealthBench unadjusted=55.6. System Card Table 6. Adjusted and unadjusted metrics are kept as distinct benchmark identities." }, { "model_id": "gpt-5.6-terra", "benchmark_id": "healthbench_length_adjusted", "score": 57.0, "reference_url": "https://deploymentsafety.openai.com/gpt-5-6", "reported_setting": { "effort": "source does not state", "tools": "none", "harness": "OpenAI official HealthBench scoring" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "OpenAI GPT-5.6 System Card table (HealthBench length-adjusted scores): gpt-5.6-terra; HealthBench length-adjusted=57.0. System Card Table 6. Adjusted and unadjusted metrics are kept as distinct benchmark identities." }, { "model_id": "gpt-5.6-terra", "benchmark_id": "healthbench", "score": 58.7, "reference_url": "https://deploymentsafety.openai.com/gpt-5-6", "reported_setting": { "effort": "source does not state", "tools": "none", "harness": "OpenAI official HealthBench scoring" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "OpenAI GPT-5.6 System Card table (HealthBench length-adjusted scores): gpt-5.6-terra; HealthBench unadjusted=58.7. System Card Table 6. Adjusted and unadjusted metrics are kept as distinct benchmark identities." }, { "model_id": "gpt-5.6-luna", "benchmark_id": "healthbench_length_adjusted", "score": 55.8, "reference_url": "https://deploymentsafety.openai.com/gpt-5-6", "reported_setting": { "effort": "source does not state", "tools": "none", "harness": "OpenAI official HealthBench scoring" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "OpenAI GPT-5.6 System Card table (HealthBench length-adjusted scores): gpt-5.6-luna; HealthBench length-adjusted=55.8. System Card Table 6. Adjusted and unadjusted metrics are kept as distinct benchmark identities." }, { "model_id": "gpt-5.6-luna", "benchmark_id": "healthbench", "score": 55.4, "reference_url": "https://deploymentsafety.openai.com/gpt-5-6", "reported_setting": { "effort": "source does not state", "tools": "none", "harness": "OpenAI official HealthBench scoring" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "OpenAI GPT-5.6 System Card table (HealthBench length-adjusted scores): gpt-5.6-luna; HealthBench unadjusted=55.4. System Card Table 6. Adjusted and unadjusted metrics are kept as distinct benchmark identities." }, { "model_id": "gpt-5", "benchmark_id": "healthbench_hard_length_adjusted", "score": 34.7, "reference_url": "https://deploymentsafety.openai.com/gpt-5-6", "reported_setting": { "effort": "source does not state", "tools": "none", "harness": "OpenAI official HealthBench scoring" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "OpenAI GPT-5.6 System Card table (HealthBench length-adjusted scores): gpt-5; HealthBench Hard length-adjusted=34.7. System Card Table 6. Adjusted and unadjusted metrics are kept as distinct benchmark identities." }, { "model_id": "gpt-5.1", "benchmark_id": "healthbench_hard_length_adjusted", "score": 25.4, "reference_url": "https://deploymentsafety.openai.com/gpt-5-6", "reported_setting": { "effort": "source does not state", "tools": "none", "harness": "OpenAI official HealthBench scoring" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "OpenAI GPT-5.6 System Card table (HealthBench length-adjusted scores): gpt-5.1; HealthBench Hard length-adjusted=25.4. System Card Table 6. Adjusted and unadjusted metrics are kept as distinct benchmark identities." }, { "model_id": "gpt-5.1", "benchmark_id": "healthbench_hard", "score": 41.4, "reference_url": "https://deploymentsafety.openai.com/gpt-5-6", "reported_setting": { "effort": "source does not state", "tools": "none", "harness": "OpenAI official HealthBench scoring" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "OpenAI GPT-5.6 System Card table (HealthBench length-adjusted scores): gpt-5.1; HealthBench Hard unadjusted=41.4. System Card Table 6. Adjusted and unadjusted metrics are kept as distinct benchmark identities." }, { "model_id": "gpt-5.2", "benchmark_id": "healthbench_hard_length_adjusted", "score": 34.3, "reference_url": "https://deploymentsafety.openai.com/gpt-5-6", "reported_setting": { "effort": "source does not state", "tools": "none", "harness": "OpenAI official HealthBench scoring" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "OpenAI GPT-5.6 System Card table (HealthBench length-adjusted scores): gpt-5.2; HealthBench Hard length-adjusted=34.3. System Card Table 6. Adjusted and unadjusted metrics are kept as distinct benchmark identities." }, { "model_id": "gpt-5.4", "benchmark_id": "healthbench_hard_length_adjusted", "score": 29.1, "reference_url": "https://deploymentsafety.openai.com/gpt-5-6", "reported_setting": { "effort": "source does not state", "tools": "none", "harness": "OpenAI official HealthBench scoring" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "OpenAI GPT-5.6 System Card table (HealthBench length-adjusted scores): gpt-5.4; HealthBench Hard length-adjusted=29.1. System Card Table 6. Adjusted and unadjusted metrics are kept as distinct benchmark identities." }, { "model_id": "gpt-5.4", "benchmark_id": "healthbench_hard", "score": 30.3, "reference_url": "https://deploymentsafety.openai.com/gpt-5-6", "reported_setting": { "effort": "source does not state", "tools": "none", "harness": "OpenAI official HealthBench scoring" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "OpenAI GPT-5.6 System Card table (HealthBench length-adjusted scores): gpt-5.4; HealthBench Hard unadjusted=30.3. System Card Table 6. Adjusted and unadjusted metrics are kept as distinct benchmark identities." }, { "model_id": "gpt-5.5", "benchmark_id": "healthbench_hard_length_adjusted", "score": 31.5, "reference_url": "https://deploymentsafety.openai.com/gpt-5-6", "reported_setting": { "effort": "source does not state", "tools": "none", "harness": "OpenAI official HealthBench scoring" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "OpenAI GPT-5.6 System Card table (HealthBench length-adjusted scores): gpt-5.5; HealthBench Hard length-adjusted=31.5. System Card Table 6. Adjusted and unadjusted metrics are kept as distinct benchmark identities." }, { "model_id": "gpt-5.5", "benchmark_id": "healthbench_hard", "score": 33.8, "reference_url": "https://deploymentsafety.openai.com/gpt-5-6", "reported_setting": { "effort": "source does not state", "tools": "none", "harness": "OpenAI official HealthBench scoring" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "OpenAI GPT-5.6 System Card table (HealthBench length-adjusted scores): gpt-5.5; HealthBench Hard unadjusted=33.8. System Card Table 6. Adjusted and unadjusted metrics are kept as distinct benchmark identities." }, { "model_id": "gpt-5.6-sol", "benchmark_id": "healthbench_hard_length_adjusted", "score": 33.1, "reference_url": "https://deploymentsafety.openai.com/gpt-5-6", "reported_setting": { "effort": "source does not state", "tools": "none", "harness": "OpenAI official HealthBench scoring" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "OpenAI GPT-5.6 System Card table (HealthBench length-adjusted scores): gpt-5.6-sol; HealthBench Hard length-adjusted=33.1. System Card Table 6. Adjusted and unadjusted metrics are kept as distinct benchmark identities." }, { "model_id": "gpt-5.6-sol", "benchmark_id": "healthbench_hard", "score": 31.1, "reference_url": "https://deploymentsafety.openai.com/gpt-5-6", "reported_setting": { "effort": "source does not state", "tools": "none", "harness": "OpenAI official HealthBench scoring" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "OpenAI GPT-5.6 System Card table (HealthBench length-adjusted scores): gpt-5.6-sol; HealthBench Hard unadjusted=31.1. System Card Table 6. Adjusted and unadjusted metrics are kept as distinct benchmark identities." }, { "model_id": "gpt-5.6-terra", "benchmark_id": "healthbench_hard_length_adjusted", "score": 32.7, "reference_url": "https://deploymentsafety.openai.com/gpt-5-6", "reported_setting": { "effort": "source does not state", "tools": "none", "harness": "OpenAI official HealthBench scoring" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "OpenAI GPT-5.6 System Card table (HealthBench length-adjusted scores): gpt-5.6-terra; HealthBench Hard length-adjusted=32.7. System Card Table 6. Adjusted and unadjusted metrics are kept as distinct benchmark identities." }, { "model_id": "gpt-5.6-terra", "benchmark_id": "healthbench_hard", "score": 34.3, "reference_url": "https://deploymentsafety.openai.com/gpt-5-6", "reported_setting": { "effort": "source does not state", "tools": "none", "harness": "OpenAI official HealthBench scoring" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "OpenAI GPT-5.6 System Card table (HealthBench length-adjusted scores): gpt-5.6-terra; HealthBench Hard unadjusted=34.3. System Card Table 6. Adjusted and unadjusted metrics are kept as distinct benchmark identities." }, { "model_id": "gpt-5.6-luna", "benchmark_id": "healthbench_hard_length_adjusted", "score": 32.0, "reference_url": "https://deploymentsafety.openai.com/gpt-5-6", "reported_setting": { "effort": "source does not state", "tools": "none", "harness": "OpenAI official HealthBench scoring" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "OpenAI GPT-5.6 System Card table (HealthBench length-adjusted scores): gpt-5.6-luna; HealthBench Hard length-adjusted=32.0. System Card Table 6. Adjusted and unadjusted metrics are kept as distinct benchmark identities." }, { "model_id": "gpt-5.6-luna", "benchmark_id": "healthbench_hard", "score": 31.4, "reference_url": "https://deploymentsafety.openai.com/gpt-5-6", "reported_setting": { "effort": "source does not state", "tools": "none", "harness": "OpenAI official HealthBench scoring" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "OpenAI GPT-5.6 System Card table (HealthBench length-adjusted scores): gpt-5.6-luna; HealthBench Hard unadjusted=31.4. System Card Table 6. Adjusted and unadjusted metrics are kept as distinct benchmark identities." }, { "model_id": "gpt-5", "benchmark_id": "healthbench_consensus_length_adjusted", "score": 95.6, "reference_url": "https://deploymentsafety.openai.com/gpt-5-6", "reported_setting": { "effort": "source does not state", "tools": "none", "harness": "OpenAI official HealthBench scoring" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "OpenAI GPT-5.6 System Card table (HealthBench length-adjusted scores): gpt-5; HealthBench Consensus length-adjusted=95.6. System Card Table 6. Adjusted and unadjusted metrics are kept as distinct benchmark identities." }, { "model_id": "gpt-5", "benchmark_id": "healthbench_consensus", "score": 96.0, "reference_url": "https://deploymentsafety.openai.com/gpt-5-6", "reported_setting": { "effort": "source does not state", "tools": "none", "harness": "OpenAI official HealthBench scoring" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "OpenAI GPT-5.6 System Card table (HealthBench length-adjusted scores): gpt-5; HealthBench Consensus unadjusted=96.0. System Card Table 6. Adjusted and unadjusted metrics are kept as distinct benchmark identities." }, { "model_id": "gpt-5.1", "benchmark_id": "healthbench_consensus_length_adjusted", "score": 95.0, "reference_url": "https://deploymentsafety.openai.com/gpt-5-6", "reported_setting": { "effort": "source does not state", "tools": "none", "harness": "OpenAI official HealthBench scoring" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "OpenAI GPT-5.6 System Card table (HealthBench length-adjusted scores): gpt-5.1; HealthBench Consensus length-adjusted=95.0. System Card Table 6. Adjusted and unadjusted metrics are kept as distinct benchmark identities." }, { "model_id": "gpt-5.1", "benchmark_id": "healthbench_consensus", "score": 95.8, "reference_url": "https://deploymentsafety.openai.com/gpt-5-6", "reported_setting": { "effort": "source does not state", "tools": "none", "harness": "OpenAI official HealthBench scoring" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "OpenAI GPT-5.6 System Card table (HealthBench length-adjusted scores): gpt-5.1; HealthBench Consensus unadjusted=95.8. System Card Table 6. Adjusted and unadjusted metrics are kept as distinct benchmark identities." }, { "model_id": "gpt-5.2", "benchmark_id": "healthbench_consensus_length_adjusted", "score": 94.4, "reference_url": "https://deploymentsafety.openai.com/gpt-5-6", "reported_setting": { "effort": "source does not state", "tools": "none", "harness": "OpenAI official HealthBench scoring" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "OpenAI GPT-5.6 System Card table (HealthBench length-adjusted scores): gpt-5.2; HealthBench Consensus length-adjusted=94.4. System Card Table 6. Adjusted and unadjusted metrics are kept as distinct benchmark identities." }, { "model_id": "gpt-5.2", "benchmark_id": "healthbench_consensus", "score": 94.7, "reference_url": "https://deploymentsafety.openai.com/gpt-5-6", "reported_setting": { "effort": "source does not state", "tools": "none", "harness": "OpenAI official HealthBench scoring" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "OpenAI GPT-5.6 System Card table (HealthBench length-adjusted scores): gpt-5.2; HealthBench Consensus unadjusted=94.7. System Card Table 6. Adjusted and unadjusted metrics are kept as distinct benchmark identities." }, { "model_id": "gpt-5.4", "benchmark_id": "healthbench_consensus_length_adjusted", "score": 96.3, "reference_url": "https://deploymentsafety.openai.com/gpt-5-6", "reported_setting": { "effort": "source does not state", "tools": "none", "harness": "OpenAI official HealthBench scoring" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "OpenAI GPT-5.6 System Card table (HealthBench length-adjusted scores): gpt-5.4; HealthBench Consensus length-adjusted=96.3. System Card Table 6. Adjusted and unadjusted metrics are kept as distinct benchmark identities." }, { "model_id": "gpt-5.4", "benchmark_id": "healthbench_consensus", "score": 96.4, "reference_url": "https://deploymentsafety.openai.com/gpt-5-6", "reported_setting": { "effort": "source does not state", "tools": "none", "harness": "OpenAI official HealthBench scoring" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "OpenAI GPT-5.6 System Card table (HealthBench length-adjusted scores): gpt-5.4; HealthBench Consensus unadjusted=96.4. System Card Table 6. Adjusted and unadjusted metrics are kept as distinct benchmark identities." }, { "model_id": "gpt-5.5", "benchmark_id": "healthbench_consensus_length_adjusted", "score": 95.6, "reference_url": "https://deploymentsafety.openai.com/gpt-5-6", "reported_setting": { "effort": "source does not state", "tools": "none", "harness": "OpenAI official HealthBench scoring" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "OpenAI GPT-5.6 System Card table (HealthBench length-adjusted scores): gpt-5.5; HealthBench Consensus length-adjusted=95.6. System Card Table 6. Adjusted and unadjusted metrics are kept as distinct benchmark identities." }, { "model_id": "gpt-5.5", "benchmark_id": "healthbench_consensus", "score": 95.7, "reference_url": "https://deploymentsafety.openai.com/gpt-5-6", "reported_setting": { "effort": "source does not state", "tools": "none", "harness": "OpenAI official HealthBench scoring" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "OpenAI GPT-5.6 System Card table (HealthBench length-adjusted scores): gpt-5.5; HealthBench Consensus unadjusted=95.7. System Card Table 6. Adjusted and unadjusted metrics are kept as distinct benchmark identities." }, { "model_id": "gpt-5.6-sol", "benchmark_id": "healthbench_consensus_length_adjusted", "score": 95.5, "reference_url": "https://deploymentsafety.openai.com/gpt-5-6", "reported_setting": { "effort": "source does not state", "tools": "none", "harness": "OpenAI official HealthBench scoring" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "OpenAI GPT-5.6 System Card table (HealthBench length-adjusted scores): gpt-5.6-sol; HealthBench Consensus length-adjusted=95.5. System Card Table 6. Adjusted and unadjusted metrics are kept as distinct benchmark identities." }, { "model_id": "gpt-5.6-sol", "benchmark_id": "healthbench_consensus", "score": 95.3, "reference_url": "https://deploymentsafety.openai.com/gpt-5-6", "reported_setting": { "effort": "source does not state", "tools": "none", "harness": "OpenAI official HealthBench scoring" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "OpenAI GPT-5.6 System Card table (HealthBench length-adjusted scores): gpt-5.6-sol; HealthBench Consensus unadjusted=95.3. System Card Table 6. Adjusted and unadjusted metrics are kept as distinct benchmark identities." }, { "model_id": "gpt-5.6-terra", "benchmark_id": "healthbench_consensus_length_adjusted", "score": 95.1, "reference_url": "https://deploymentsafety.openai.com/gpt-5-6", "reported_setting": { "effort": "source does not state", "tools": "none", "harness": "OpenAI official HealthBench scoring" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "OpenAI GPT-5.6 System Card table (HealthBench length-adjusted scores): gpt-5.6-terra; HealthBench Consensus length-adjusted=95.1. System Card Table 6. Adjusted and unadjusted metrics are kept as distinct benchmark identities." }, { "model_id": "gpt-5.6-terra", "benchmark_id": "healthbench_consensus", "score": 95.2, "reference_url": "https://deploymentsafety.openai.com/gpt-5-6", "reported_setting": { "effort": "source does not state", "tools": "none", "harness": "OpenAI official HealthBench scoring" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "OpenAI GPT-5.6 System Card table (HealthBench length-adjusted scores): gpt-5.6-terra; HealthBench Consensus unadjusted=95.2. System Card Table 6. Adjusted and unadjusted metrics are kept as distinct benchmark identities." }, { "model_id": "gpt-5.6-luna", "benchmark_id": "healthbench_consensus_length_adjusted", "score": 95.1, "reference_url": "https://deploymentsafety.openai.com/gpt-5-6", "reported_setting": { "effort": "source does not state", "tools": "none", "harness": "OpenAI official HealthBench scoring" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "OpenAI GPT-5.6 System Card table (HealthBench length-adjusted scores): gpt-5.6-luna; HealthBench Consensus length-adjusted=95.1. System Card Table 6. Adjusted and unadjusted metrics are kept as distinct benchmark identities." }, { "model_id": "gpt-5.6-luna", "benchmark_id": "healthbench_consensus", "score": 95.1, "reference_url": "https://deploymentsafety.openai.com/gpt-5-6", "reported_setting": { "effort": "source does not state", "tools": "none", "harness": "OpenAI official HealthBench scoring" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "OpenAI GPT-5.6 System Card table (HealthBench length-adjusted scores): gpt-5.6-luna; HealthBench Consensus unadjusted=95.1. System Card Table 6. Adjusted and unadjusted metrics are kept as distinct benchmark identities." }, { "model_id": "gpt-5.5", "benchmark_id": "aav_capsid_packaging_prediction", "score": 0.528, "reference_url": "https://deploymentsafety.openai.com/gpt-5-6", "reported_setting": { "effort": "source does not state", "tools": "benchmark-specified", "harness": "OpenAI preparedness evaluation" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "OpenAI GPT-5.6 System Card table (AAV Capsid Packaging Prediction): gpt-5.5; AAV Capsid Packaging Prediction=0.528." }, { "model_id": "gpt-5.6-sol", "benchmark_id": "aav_capsid_packaging_prediction", "score": 0.529, "reference_url": "https://deploymentsafety.openai.com/gpt-5-6", "reported_setting": { "effort": "source does not state", "tools": "benchmark-specified", "harness": "OpenAI preparedness evaluation" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "OpenAI GPT-5.6 System Card table (AAV Capsid Packaging Prediction): gpt-5.6-sol; AAV Capsid Packaging Prediction=0.529." }, { "model_id": "gpt-5.4", "benchmark_id": "hard_negative_protein_binding", "score": 3.5, "reference_url": "https://deploymentsafety.openai.com/gpt-5-6", "reported_setting": { "effort": "source does not state", "tools": "benchmark-specified", "harness": "OpenAI preparedness evaluation" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "OpenAI GPT-5.6 System Card table (Hard-negative protein binding prediction): gpt-5.4-thinking; Hard-negative protein binding prediction=3.5." }, { "model_id": "gpt-5.5", "benchmark_id": "hard_negative_protein_binding", "score": 0.4, "reference_url": "https://deploymentsafety.openai.com/gpt-5-6", "reported_setting": { "effort": "source does not state", "tools": "benchmark-specified", "harness": "OpenAI preparedness evaluation" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "OpenAI GPT-5.6 System Card table (Hard-negative protein binding prediction): gpt-5.5; Hard-negative protein binding prediction=0.4.", "candidates": [ { "score": 0.0, "reference_url": "https://deploymentsafety.openai.com/gpt-5-6", "source_type": "model_card", "reported_setting": { "effort": "pro", "tools": "benchmark-specified", "harness": "OpenAI preparedness evaluation" }, "notes": "OpenAI GPT-5.6 System Card table (Hard-negative protein binding prediction): gpt-5.5-pro; Hard-negative protein binding prediction=0.0." } ] }, { "model_id": "gpt-5.6-sol", "benchmark_id": "hard_negative_protein_binding", "score": 7.6, "reference_url": "https://deploymentsafety.openai.com/gpt-5-6", "reported_setting": { "effort": "source does not state", "tools": "benchmark-specified", "harness": "OpenAI preparedness evaluation" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "OpenAI GPT-5.6 System Card table (Hard-negative protein binding prediction): gpt-5.6-sol; Hard-negative protein binding prediction=7.6." }, { "model_id": "gpt-5.4", "benchmark_id": "dna_tf_binding_design", "score": 12.82, "reference_url": "https://deploymentsafety.openai.com/gpt-5-6", "reported_setting": { "effort": "source does not state", "tools": "benchmark-specified", "harness": "OpenAI preparedness evaluation" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "OpenAI GPT-5.6 System Card table (DNA sequence design for transcription factor binding): gpt-5.4-thinking; DNA sequence design for transcription factor binding=12.82." }, { "model_id": "gpt-5.5", "benchmark_id": "dna_tf_binding_design", "score": 13.82, "reference_url": "https://deploymentsafety.openai.com/gpt-5-6", "reported_setting": { "effort": "source does not state", "tools": "benchmark-specified", "harness": "OpenAI preparedness evaluation" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "OpenAI GPT-5.6 System Card table (DNA sequence design for transcription factor binding): gpt-5.5; DNA sequence design for transcription factor binding=13.82.", "candidates": [ { "score": 16.5, "reference_url": "https://deploymentsafety.openai.com/gpt-5-6", "source_type": "model_card", "reported_setting": { "effort": "pro", "tools": "benchmark-specified", "harness": "OpenAI preparedness evaluation" }, "notes": "OpenAI GPT-5.6 System Card table (DNA sequence design for transcription factor binding): gpt-5.5-pro; DNA sequence design for transcription factor binding=16.5." } ] }, { "model_id": "gpt-5.6-sol", "benchmark_id": "dna_tf_binding_design", "score": 13.7, "reference_url": "https://deploymentsafety.openai.com/gpt-5-6", "reported_setting": { "effort": "source does not state", "tools": "benchmark-specified", "harness": "OpenAI preparedness evaluation" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "OpenAI GPT-5.6 System Card table (DNA sequence design for transcription factor binding): gpt-5.6-sol; DNA sequence design for transcription factor binding=13.7." }, { "model_id": "gpt-5.1", "benchmark_id": "first_person_fairness_harm_overall", "score": 1.28, "reference_url": "https://deploymentsafety.openai.com/gpt-5-6", "reported_setting": { "effort": "source does not state", "tools": "none", "harness": "OpenAI First-Person Fairness Evaluation" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "OpenAI GPT-5.6 System Card figure (First-Person Fairness Evaluation): GPT-5.1 Thinking; First-Person Fairness Evaluation=1.28. All seven exact bar values are printed. Whiskers show 95% confidence intervals but endpoints are unlabeled. The card defines harm_overall as expected male-versus-female biased-answer difference based on evaluation performance divided by 10." }, { "model_id": "gpt-5.2", "benchmark_id": "first_person_fairness_harm_overall", "score": 1.0, "reference_url": "https://deploymentsafety.openai.com/gpt-5-6", "reported_setting": { "effort": "source does not state", "tools": "none", "harness": "OpenAI First-Person Fairness Evaluation" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "OpenAI GPT-5.6 System Card figure (First-Person Fairness Evaluation): GPT-5.2 Thinking; First-Person Fairness Evaluation=1.0. All seven exact bar values are printed. Whiskers show 95% confidence intervals but endpoints are unlabeled. The card defines harm_overall as expected male-versus-female biased-answer difference based on evaluation performance divided by 10." }, { "model_id": "gpt-5.4", "benchmark_id": "first_person_fairness_harm_overall", "score": 0.88, "reference_url": "https://deploymentsafety.openai.com/gpt-5-6", "reported_setting": { "effort": "source does not state", "tools": "none", "harness": "OpenAI First-Person Fairness Evaluation" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "OpenAI GPT-5.6 System Card figure (First-Person Fairness Evaluation): GPT-5.4 Thinking; First-Person Fairness Evaluation=0.88. All seven exact bar values are printed. Whiskers show 95% confidence intervals but endpoints are unlabeled. The card defines harm_overall as expected male-versus-female biased-answer difference based on evaluation performance divided by 10." }, { "model_id": "gpt-5.5", "benchmark_id": "first_person_fairness_harm_overall", "score": 1.12, "reference_url": "https://deploymentsafety.openai.com/gpt-5-6", "reported_setting": { "effort": "source does not state", "tools": "none", "harness": "OpenAI First-Person Fairness Evaluation" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "OpenAI GPT-5.6 System Card figure (First-Person Fairness Evaluation): GPT-5.5; First-Person Fairness Evaluation=1.12. All seven exact bar values are printed. Whiskers show 95% confidence intervals but endpoints are unlabeled. The card defines harm_overall as expected male-versus-female biased-answer difference based on evaluation performance divided by 10." }, { "model_id": "gpt-5.6-sol", "benchmark_id": "first_person_fairness_harm_overall", "score": 0.98, "reference_url": "https://deploymentsafety.openai.com/gpt-5-6", "reported_setting": { "effort": "source does not state", "tools": "none", "harness": "OpenAI First-Person Fairness Evaluation" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "OpenAI GPT-5.6 System Card figure (First-Person Fairness Evaluation): GPT-5.6 Sol; First-Person Fairness Evaluation=0.98. All seven exact bar values are printed. Whiskers show 95% confidence intervals but endpoints are unlabeled. The card defines harm_overall as expected male-versus-female biased-answer difference based on evaluation performance divided by 10." }, { "model_id": "gpt-5.6-terra", "benchmark_id": "first_person_fairness_harm_overall", "score": 0.88, "reference_url": "https://deploymentsafety.openai.com/gpt-5-6", "reported_setting": { "effort": "source does not state", "tools": "none", "harness": "OpenAI First-Person Fairness Evaluation" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "OpenAI GPT-5.6 System Card figure (First-Person Fairness Evaluation): GPT-5.6 Terra; First-Person Fairness Evaluation=0.88. All seven exact bar values are printed. Whiskers show 95% confidence intervals but endpoints are unlabeled. The card defines harm_overall as expected male-versus-female biased-answer difference based on evaluation performance divided by 10." }, { "model_id": "gpt-5.6-luna", "benchmark_id": "first_person_fairness_harm_overall", "score": 0.61, "reference_url": "https://deploymentsafety.openai.com/gpt-5-6", "reported_setting": { "effort": "source does not state", "tools": "none", "harness": "OpenAI First-Person Fairness Evaluation" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "OpenAI GPT-5.6 System Card figure (First-Person Fairness Evaluation): GPT-5.6 Luna; First-Person Fairness Evaluation=0.61. All seven exact bar values are printed. Whiskers show 95% confidence intervals but endpoints are unlabeled. The card defines harm_overall as expected male-versus-female biased-answer difference based on evaluation performance divided by 10." }, { "model_id": "gpt-5.6-sol", "benchmark_id": "frontiercyber_easy", "score": 11.0, "reference_url": "https://deploymentsafety.openai.com/gpt-5-6", "reported_setting": { "effort": "source does not state", "tools": "agentic cyber environment", "harness": "Irregular agent harness", "sampling": "5/44" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "OpenAI GPT-5.6 System Card figure (FrontierCyber success rates): GPT-5.6 Sol; FrontierCyber Easy=11.0. All eight plotted percentages are exact labels; caption supplies exact numerators/denominators. Results use Irregular's agent harness. Denominators vary slightly due to device availability constraints. The benchmark tests zero-day discovery and exploitation in current off-the-shelf software and hardware." }, { "model_id": "gpt-5.5", "benchmark_id": "frontiercyber_easy", "score": 6.0, "reference_url": "https://deploymentsafety.openai.com/gpt-5-6", "reported_setting": { "effort": "source does not state", "tools": "agentic cyber environment", "harness": "Irregular agent harness", "sampling": "3/44" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "OpenAI GPT-5.6 System Card figure (FrontierCyber success rates): GPT-5.5; FrontierCyber Easy=6.0. All eight plotted percentages are exact labels; caption supplies exact numerators/denominators. Results use Irregular's agent harness. Denominators vary slightly due to device availability constraints. The benchmark tests zero-day discovery and exploitation in current off-the-shelf software and hardware." }, { "model_id": "gpt-5.6-sol", "benchmark_id": "frontiercyber_medium", "score": 12.0, "reference_url": "https://deploymentsafety.openai.com/gpt-5-6", "reported_setting": { "effort": "source does not state", "tools": "agentic cyber environment", "harness": "Irregular agent harness", "sampling": "10/77" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "OpenAI GPT-5.6 System Card figure (FrontierCyber success rates): GPT-5.6 Sol; FrontierCyber Medium=12.0. All eight plotted percentages are exact labels; caption supplies exact numerators/denominators. Results use Irregular's agent harness. Denominators vary slightly due to device availability constraints. The benchmark tests zero-day discovery and exploitation in current off-the-shelf software and hardware." }, { "model_id": "gpt-5.5", "benchmark_id": "frontiercyber_medium", "score": 6.0, "reference_url": "https://deploymentsafety.openai.com/gpt-5-6", "reported_setting": { "effort": "source does not state", "tools": "agentic cyber environment", "harness": "Irregular agent harness", "sampling": "5/80" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "OpenAI GPT-5.6 System Card figure (FrontierCyber success rates): GPT-5.5; FrontierCyber Medium=6.0. All eight plotted percentages are exact labels; caption supplies exact numerators/denominators. Results use Irregular's agent harness. Denominators vary slightly due to device availability constraints. The benchmark tests zero-day discovery and exploitation in current off-the-shelf software and hardware." }, { "model_id": "gpt-5.6-sol", "benchmark_id": "frontiercyber_hard", "score": 5.0, "reference_url": "https://deploymentsafety.openai.com/gpt-5-6", "reported_setting": { "effort": "source does not state", "tools": "agentic cyber environment", "harness": "Irregular agent harness", "sampling": "4/67" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "OpenAI GPT-5.6 System Card figure (FrontierCyber success rates): GPT-5.6 Sol; FrontierCyber Hard=5.0. All eight plotted percentages are exact labels; caption supplies exact numerators/denominators. Results use Irregular's agent harness. Denominators vary slightly due to device availability constraints. The benchmark tests zero-day discovery and exploitation in current off-the-shelf software and hardware." }, { "model_id": "gpt-5.5", "benchmark_id": "frontiercyber_hard", "score": 4.0, "reference_url": "https://deploymentsafety.openai.com/gpt-5-6", "reported_setting": { "effort": "source does not state", "tools": "agentic cyber environment", "harness": "Irregular agent harness", "sampling": "3/69" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "OpenAI GPT-5.6 System Card figure (FrontierCyber success rates): GPT-5.5; FrontierCyber Hard=4.0. All eight plotted percentages are exact labels; caption supplies exact numerators/denominators. Results use Irregular's agent harness. Denominators vary slightly due to device availability constraints. The benchmark tests zero-day discovery and exploitation in current off-the-shelf software and hardware." }, { "model_id": "gpt-5.6-sol", "benchmark_id": "frontiercyber_elite", "score": 0.0, "reference_url": "https://deploymentsafety.openai.com/gpt-5-6", "reported_setting": { "effort": "source does not state", "tools": "agentic cyber environment", "harness": "Irregular agent harness", "sampling": "0/9" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "OpenAI GPT-5.6 System Card figure (FrontierCyber success rates): GPT-5.6 Sol; FrontierCyber Elite=0.0. All eight plotted percentages are exact labels; caption supplies exact numerators/denominators. Results use Irregular's agent harness. Denominators vary slightly due to device availability constraints. The benchmark tests zero-day discovery and exploitation in current off-the-shelf software and hardware." }, { "model_id": "gpt-5.5", "benchmark_id": "frontiercyber_elite", "score": 0.0, "reference_url": "https://deploymentsafety.openai.com/gpt-5-6", "reported_setting": { "effort": "source does not state", "tools": "agentic cyber environment", "harness": "Irregular agent harness", "sampling": "0/12" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "OpenAI GPT-5.6 System Card figure (FrontierCyber success rates): GPT-5.5; FrontierCyber Elite=0.0. All eight plotted percentages are exact labels; caption supplies exact numerators/denominators. Results use Irregular's agent harness. Denominators vary slightly due to device availability constraints. The benchmark tests zero-day discovery and exploitation in current off-the-shelf software and hardware." }, { "model_id": "claude-opus-4.7", "benchmark_id": "long_form_virology_task_1_sequence_design", "score": 0.89, "reference_url": "https://www.anthropic.com/claude-sonnet-5-system-card", "reported_setting": { "effort": "source does not state", "tools": "none", "sampling": "pass@1", "harness": "biology" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Anthropic Claude Sonnet 5 System Card Figure 2.2.4.A, page 17: Claude Opus 4.7; Long-form virology tasks [Task 1 / Sequence design]=0.89. Source setting: effort=source does not state; tools=none; sampling=pass@1; harness=biology." }, { "model_id": "claude-opus-4.8", "benchmark_id": "long_form_virology_task_1_sequence_design", "score": 0.88, "reference_url": "https://www.anthropic.com/claude-sonnet-5-system-card", "reported_setting": { "effort": "source does not state", "tools": "none", "sampling": "pass@1", "harness": "biology" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Anthropic Claude Sonnet 5 System Card Figure 2.2.4.A, page 17: Claude Opus 4.8; Long-form virology tasks [Task 1 / Sequence design]=0.88. Source setting: effort=source does not state; tools=none; sampling=pass@1; harness=biology." }, { "model_id": "claude-fable-5", "benchmark_id": "long_form_virology_task_1_sequence_design", "score": 0.88, "reference_url": "https://www.anthropic.com/claude-sonnet-5-system-card", "reported_setting": { "effort": "source does not state", "tools": "none", "sampling": "pass@1", "harness": "biology", "deployment": "claude-mythos-5" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Anthropic Claude Sonnet 5 System Card Figure 2.2.4.A, page 17: Claude Mythos 5; Long-form virology tasks [Task 1 / Sequence design]=0.88. Source setting: effort=source does not state; tools=none; sampling=pass@1; harness=biology; deployment=claude-mythos-5." }, { "model_id": "claude-sonnet-5", "benchmark_id": "long_form_virology_task_1_sequence_design", "score": 0.9, "reference_url": "https://www.anthropic.com/claude-sonnet-5-system-card", "reported_setting": { "effort": "source does not state", "tools": "none", "sampling": "pass@1", "harness": "biology" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Anthropic Claude Sonnet 5 System Card Figure 2.2.4.A, page 17: Claude Sonnet 5; Long-form virology tasks [Task 1 / Sequence design]=0.90. Source setting: effort=source does not state; tools=none; sampling=pass@1; harness=biology." }, { "model_id": "claude-opus-4.7", "benchmark_id": "long_form_virology_task_1_protocol_design", "score": 0.93, "reference_url": "https://www.anthropic.com/claude-sonnet-5-system-card", "reported_setting": { "effort": "source does not state", "tools": "none", "sampling": "pass@1", "harness": "biology" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Anthropic Claude Sonnet 5 System Card Figure 2.2.4.A, page 17: Claude Opus 4.7; Long-form virology tasks [Task 1 / Protocol design]=0.93. Source setting: effort=source does not state; tools=none; sampling=pass@1; harness=biology." }, { "model_id": "claude-opus-4.8", "benchmark_id": "long_form_virology_task_1_protocol_design", "score": 0.87, "reference_url": "https://www.anthropic.com/claude-sonnet-5-system-card", "reported_setting": { "effort": "source does not state", "tools": "none", "sampling": "pass@1", "harness": "biology" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Anthropic Claude Sonnet 5 System Card Figure 2.2.4.A, page 17: Claude Opus 4.8; Long-form virology tasks [Task 1 / Protocol design]=0.87. Source setting: effort=source does not state; tools=none; sampling=pass@1; harness=biology." }, { "model_id": "claude-fable-5", "benchmark_id": "long_form_virology_task_1_protocol_design", "score": 0.87, "reference_url": "https://www.anthropic.com/claude-sonnet-5-system-card", "reported_setting": { "effort": "source does not state", "tools": "none", "sampling": "pass@1", "harness": "biology", "deployment": "claude-mythos-5" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Anthropic Claude Sonnet 5 System Card Figure 2.2.4.A, page 17: Claude Mythos 5; Long-form virology tasks [Task 1 / Protocol design]=0.87. Source setting: effort=source does not state; tools=none; sampling=pass@1; harness=biology; deployment=claude-mythos-5." }, { "model_id": "claude-sonnet-5", "benchmark_id": "long_form_virology_task_1_protocol_design", "score": 0.87, "reference_url": "https://www.anthropic.com/claude-sonnet-5-system-card", "reported_setting": { "effort": "source does not state", "tools": "none", "sampling": "pass@1", "harness": "biology" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Anthropic Claude Sonnet 5 System Card Figure 2.2.4.A, page 17: Claude Sonnet 5; Long-form virology tasks [Task 1 / Protocol design]=0.87. Source setting: effort=source does not state; tools=none; sampling=pass@1; harness=biology." }, { "model_id": "claude-opus-4.7", "benchmark_id": "long_form_virology_task_1_end_to_end", "score": 0.82, "reference_url": "https://www.anthropic.com/claude-sonnet-5-system-card", "reported_setting": { "effort": "source does not state", "tools": "none", "sampling": "pass@1", "harness": "biology" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Anthropic Claude Sonnet 5 System Card Figure 2.2.4.A, page 17: Claude Opus 4.7; Long-form virology tasks [Task 1 / End-to-end]=0.82. Source setting: effort=source does not state; tools=none; sampling=pass@1; harness=biology." }, { "model_id": "claude-opus-4.8", "benchmark_id": "long_form_virology_task_1_end_to_end", "score": 0.72, "reference_url": "https://www.anthropic.com/claude-sonnet-5-system-card", "reported_setting": { "effort": "source does not state", "tools": "none", "sampling": "pass@1", "harness": "biology" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Anthropic Claude Sonnet 5 System Card Figure 2.2.4.A, page 17: Claude Opus 4.8; Long-form virology tasks [Task 1 / End-to-end]=0.72. Source setting: effort=source does not state; tools=none; sampling=pass@1; harness=biology.", "candidates": [ { "score": 0.77, "reference_url": "https://www.anthropic.com/claude-opus-5-system-card", "source_type": "model_card", "reported_setting": { "effort": "source does not state" }, "notes": "Figure 2.2.4.A / Long-form virology / Task 1 / End-to-end / Claude Opus 4.8: Two long-form virology tasks." } ] }, { "model_id": "claude-fable-5", "benchmark_id": "long_form_virology_task_1_end_to_end", "score": 0.77, "reference_url": "https://www.anthropic.com/claude-sonnet-5-system-card", "reported_setting": { "effort": "source does not state", "tools": "none", "sampling": "pass@1", "harness": "biology", "deployment": "claude-mythos-5" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Anthropic Claude Sonnet 5 System Card Figure 2.2.4.A, page 17: Claude Mythos 5; Long-form virology tasks [Task 1 / End-to-end]=0.77. Source setting: effort=source does not state; tools=none; sampling=pass@1; harness=biology; deployment=claude-mythos-5." }, { "model_id": "claude-sonnet-5", "benchmark_id": "long_form_virology_task_1_end_to_end", "score": 0.79, "reference_url": "https://www.anthropic.com/claude-sonnet-5-system-card", "reported_setting": { "effort": "source does not state", "tools": "none", "sampling": "pass@1", "harness": "biology" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Anthropic Claude Sonnet 5 System Card Figure 2.2.4.A, page 17: Claude Sonnet 5; Long-form virology tasks [Task 1 / End-to-end]=0.79. Source setting: effort=source does not state; tools=none; sampling=pass@1; harness=biology." }, { "model_id": "claude-opus-4.7", "benchmark_id": "long_form_virology_task_2_sequence_design", "score": 1.0, "reference_url": "https://www.anthropic.com/claude-sonnet-5-system-card", "reported_setting": { "effort": "source does not state", "tools": "none", "sampling": "pass@1", "harness": "biology" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Anthropic Claude Sonnet 5 System Card Figure 2.2.4.A, page 17: Claude Opus 4.7; Long-form virology tasks [Task 2 / Sequence design]=1.00. Source setting: effort=source does not state; tools=none; sampling=pass@1; harness=biology." }, { "model_id": "claude-opus-4.8", "benchmark_id": "long_form_virology_task_2_sequence_design", "score": 1.0, "reference_url": "https://www.anthropic.com/claude-sonnet-5-system-card", "reported_setting": { "effort": "source does not state", "tools": "none", "sampling": "pass@1", "harness": "biology" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Anthropic Claude Sonnet 5 System Card Figure 2.2.4.A, page 17: Claude Opus 4.8; Long-form virology tasks [Task 2 / Sequence design]=1.00. Source setting: effort=source does not state; tools=none; sampling=pass@1; harness=biology." }, { "model_id": "claude-fable-5", "benchmark_id": "long_form_virology_task_2_sequence_design", "score": 1.0, "reference_url": "https://www.anthropic.com/claude-sonnet-5-system-card", "reported_setting": { "effort": "source does not state", "tools": "none", "sampling": "pass@1", "harness": "biology", "deployment": "claude-mythos-5" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Anthropic Claude Sonnet 5 System Card Figure 2.2.4.A, page 17: Claude Mythos 5; Long-form virology tasks [Task 2 / Sequence design]=1.00. Source setting: effort=source does not state; tools=none; sampling=pass@1; harness=biology; deployment=claude-mythos-5." }, { "model_id": "claude-sonnet-5", "benchmark_id": "long_form_virology_task_2_sequence_design", "score": 1.0, "reference_url": "https://www.anthropic.com/claude-sonnet-5-system-card", "reported_setting": { "effort": "source does not state", "tools": "none", "sampling": "pass@1", "harness": "biology" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Anthropic Claude Sonnet 5 System Card Figure 2.2.4.A, page 17: Claude Sonnet 5; Long-form virology tasks [Task 2 / Sequence design]=1.00. Source setting: effort=source does not state; tools=none; sampling=pass@1; harness=biology." }, { "model_id": "claude-opus-4.7", "benchmark_id": "long_form_virology_task_2_protocol_design", "score": 0.94, "reference_url": "https://www.anthropic.com/claude-sonnet-5-system-card", "reported_setting": { "effort": "source does not state", "tools": "none", "sampling": "pass@1", "harness": "biology" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Anthropic Claude Sonnet 5 System Card Figure 2.2.4.A, page 17: Claude Opus 4.7; Long-form virology tasks [Task 2 / Protocol design]=0.94. Source setting: effort=source does not state; tools=none; sampling=pass@1; harness=biology." }, { "model_id": "claude-opus-4.8", "benchmark_id": "long_form_virology_task_2_protocol_design", "score": 0.9, "reference_url": "https://www.anthropic.com/claude-sonnet-5-system-card", "reported_setting": { "effort": "source does not state", "tools": "none", "sampling": "pass@1", "harness": "biology" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Anthropic Claude Sonnet 5 System Card Figure 2.2.4.A, page 17: Claude Opus 4.8; Long-form virology tasks [Task 2 / Protocol design]=0.90. Source setting: effort=source does not state; tools=none; sampling=pass@1; harness=biology." }, { "model_id": "claude-fable-5", "benchmark_id": "long_form_virology_task_2_protocol_design", "score": 0.91, "reference_url": "https://www.anthropic.com/claude-sonnet-5-system-card", "reported_setting": { "effort": "source does not state", "tools": "none", "sampling": "pass@1", "harness": "biology", "deployment": "claude-mythos-5" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Anthropic Claude Sonnet 5 System Card Figure 2.2.4.A, page 17: Claude Mythos 5; Long-form virology tasks [Task 2 / Protocol design]=0.91. Source setting: effort=source does not state; tools=none; sampling=pass@1; harness=biology; deployment=claude-mythos-5." }, { "model_id": "claude-sonnet-5", "benchmark_id": "long_form_virology_task_2_protocol_design", "score": 0.9, "reference_url": "https://www.anthropic.com/claude-sonnet-5-system-card", "reported_setting": { "effort": "source does not state", "tools": "none", "sampling": "pass@1", "harness": "biology" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Anthropic Claude Sonnet 5 System Card Figure 2.2.4.A, page 17: Claude Sonnet 5; Long-form virology tasks [Task 2 / Protocol design]=0.90. Source setting: effort=source does not state; tools=none; sampling=pass@1; harness=biology." }, { "model_id": "claude-opus-4.7", "benchmark_id": "long_form_virology_task_2_end_to_end", "score": 0.94, "reference_url": "https://www.anthropic.com/claude-sonnet-5-system-card", "reported_setting": { "effort": "source does not state", "tools": "none", "sampling": "pass@1", "harness": "biology" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Anthropic Claude Sonnet 5 System Card Figure 2.2.4.A, page 17: Claude Opus 4.7; Long-form virology tasks [Task 2 / End-to-end]=0.94. Source setting: effort=source does not state; tools=none; sampling=pass@1; harness=biology." }, { "model_id": "claude-opus-4.8", "benchmark_id": "long_form_virology_task_2_end_to_end", "score": 0.9, "reference_url": "https://www.anthropic.com/claude-sonnet-5-system-card", "reported_setting": { "effort": "source does not state", "tools": "none", "sampling": "pass@1", "harness": "biology" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Anthropic Claude Sonnet 5 System Card Figure 2.2.4.A, page 17: Claude Opus 4.8; Long-form virology tasks [Task 2 / End-to-end]=0.90. Source setting: effort=source does not state; tools=none; sampling=pass@1; harness=biology." }, { "model_id": "claude-fable-5", "benchmark_id": "long_form_virology_task_2_end_to_end", "score": 0.91, "reference_url": "https://www.anthropic.com/claude-sonnet-5-system-card", "reported_setting": { "effort": "source does not state", "tools": "none", "sampling": "pass@1", "harness": "biology", "deployment": "claude-mythos-5" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Anthropic Claude Sonnet 5 System Card Figure 2.2.4.A, page 17: Claude Mythos 5; Long-form virology tasks [Task 2 / End-to-end]=0.91. Source setting: effort=source does not state; tools=none; sampling=pass@1; harness=biology; deployment=claude-mythos-5." }, { "model_id": "claude-sonnet-5", "benchmark_id": "long_form_virology_task_2_end_to_end", "score": 0.9, "reference_url": "https://www.anthropic.com/claude-sonnet-5-system-card", "reported_setting": { "effort": "source does not state", "tools": "none", "sampling": "pass@1", "harness": "biology" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Anthropic Claude Sonnet 5 System Card Figure 2.2.4.A, page 17: Claude Sonnet 5; Long-form virology tasks [Task 2 / End-to-end]=0.90. Source setting: effort=source does not state; tools=none; sampling=pass@1; harness=biology." }, { "model_id": "claude-opus-4.7", "benchmark_id": "ai_rd_kernel_best_speedup", "score": 371.75, "reference_url": "https://www.anthropic.com/claude-sonnet-5-system-card", "reported_setting": { "effort": "source does not state", "tools": "none", "sampling": "pass@1", "harness": "aird" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Anthropic Claude Sonnet 5 System Card Table 2.3.3.A, page 26: Claude Opus 4.7; AI R&D rule-out evaluations [Kernel task / best speedup]=371.75×. Source setting: effort=source does not state; tools=none; sampling=pass@1; harness=aird." }, { "model_id": "claude-fable-5", "benchmark_id": "ai_rd_kernel_best_speedup", "score": 430.93, "reference_url": "https://www.anthropic.com/claude-sonnet-5-system-card", "reported_setting": { "effort": "source does not state", "tools": "none", "sampling": "pass@1", "harness": "aird", "deployment": "claude-mythos-5" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Anthropic Claude Sonnet 5 System Card Table 2.3.3.A, page 26: Claude Mythos 5; AI R&D rule-out evaluations [Kernel task / best speedup]=430.93×. Source setting: effort=source does not state; tools=none; sampling=pass@1; harness=aird; deployment=claude-mythos-5." }, { "model_id": "claude-sonnet-5", "benchmark_id": "ai_rd_kernel_best_speedup", "score": 284.52, "reference_url": "https://www.anthropic.com/claude-sonnet-5-system-card", "reported_setting": { "effort": "source does not state", "tools": "none", "sampling": "pass@1", "harness": "aird" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Anthropic Claude Sonnet 5 System Card Table 2.3.3.A, page 26: Claude Sonnet 5; AI R&D rule-out evaluations [Kernel task / best speedup]=284.52×. Source setting: effort=source does not state; tools=none; sampling=pass@1; harness=aird." }, { "model_id": "claude-opus-4.7", "benchmark_id": "ai_rd_time_series_forecasting_mse", "score": 4.78, "reference_url": "https://www.anthropic.com/claude-sonnet-5-system-card", "reported_setting": { "effort": "source does not state", "tools": "none", "sampling": "pass@1", "harness": "aird" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Anthropic Claude Sonnet 5 System Card Table 2.3.3.A, page 26: Claude Opus 4.7; AI R&D rule-out evaluations [Time Series Forecasting / MSE, lower better]=4.78. Source setting: effort=source does not state; tools=none; sampling=pass@1; harness=aird." }, { "model_id": "claude-fable-5", "benchmark_id": "ai_rd_time_series_forecasting_mse", "score": 4.51, "reference_url": "https://www.anthropic.com/claude-sonnet-5-system-card", "reported_setting": { "effort": "source does not state", "tools": "none", "sampling": "pass@1", "harness": "aird", "deployment": "claude-mythos-5" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Anthropic Claude Sonnet 5 System Card Table 2.3.3.A, page 26: Claude Mythos 5; AI R&D rule-out evaluations [Time Series Forecasting / MSE, lower better]=4.51. Source setting: effort=source does not state; tools=none; sampling=pass@1; harness=aird; deployment=claude-mythos-5." }, { "model_id": "claude-sonnet-5", "benchmark_id": "ai_rd_time_series_forecasting_mse", "score": 5.8, "reference_url": "https://www.anthropic.com/claude-sonnet-5-system-card", "reported_setting": { "effort": "source does not state", "tools": "none", "sampling": "pass@1", "harness": "aird" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Anthropic Claude Sonnet 5 System Card Table 2.3.3.A, page 26: Claude Sonnet 5; AI R&D rule-out evaluations [Time Series Forecasting / MSE, lower better]=5.80. Source setting: effort=source does not state; tools=none; sampling=pass@1; harness=aird." }, { "model_id": "claude-opus-4.7", "benchmark_id": "ai_rd_llm_training_speedup", "score": 50.67, "reference_url": "https://www.anthropic.com/claude-sonnet-5-system-card", "reported_setting": { "effort": "source does not state", "tools": "none", "sampling": "pass@1", "harness": "aird" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Anthropic Claude Sonnet 5 System Card Table 2.3.3.A, page 26: Claude Opus 4.7; AI R&D rule-out evaluations [LLM training / average speedup]=50.67×. Source setting: effort=source does not state; tools=none; sampling=pass@1; harness=aird." }, { "model_id": "claude-fable-5", "benchmark_id": "ai_rd_llm_training_speedup", "score": 69.61, "reference_url": "https://www.anthropic.com/claude-sonnet-5-system-card", "reported_setting": { "effort": "source does not state", "tools": "none", "sampling": "pass@1", "harness": "aird", "deployment": "claude-mythos-5" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Anthropic Claude Sonnet 5 System Card Table 2.3.3.A, page 26: Claude Mythos 5; AI R&D rule-out evaluations [LLM training / average speedup]=69.61×. Source setting: effort=source does not state; tools=none; sampling=pass@1; harness=aird; deployment=claude-mythos-5." }, { "model_id": "claude-sonnet-5", "benchmark_id": "ai_rd_llm_training_speedup", "score": 26.49, "reference_url": "https://www.anthropic.com/claude-sonnet-5-system-card", "reported_setting": { "effort": "source does not state", "tools": "none", "sampling": "pass@1", "harness": "aird" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Anthropic Claude Sonnet 5 System Card Table 2.3.3.A, page 26: Claude Sonnet 5; AI R&D rule-out evaluations [LLM training / average speedup]=26.49×. Source setting: effort=source does not state; tools=none; sampling=pass@1; harness=aird." }, { "model_id": "claude-opus-4.7", "benchmark_id": "ai_rd_quadruped_rl", "score": 24.73, "reference_url": "https://www.anthropic.com/claude-sonnet-5-system-card", "reported_setting": { "effort": "source does not state", "tools": "none", "sampling": "pass@1", "harness": "aird" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Anthropic Claude Sonnet 5 System Card Table 2.3.3.A, page 26: Claude Opus 4.7; AI R&D rule-out evaluations [Quadruped RL / highest score, no hparams]=24.73. Source setting: effort=source does not state; tools=none; sampling=pass@1; harness=aird." }, { "model_id": "claude-fable-5", "benchmark_id": "ai_rd_quadruped_rl", "score": 29.55, "reference_url": "https://www.anthropic.com/claude-sonnet-5-system-card", "reported_setting": { "effort": "source does not state", "tools": "none", "sampling": "pass@1", "harness": "aird", "deployment": "claude-mythos-5" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Anthropic Claude Sonnet 5 System Card Table 2.3.3.A, page 26: Claude Mythos 5; AI R&D rule-out evaluations [Quadruped RL / highest score, no hparams]=29.55. Source setting: effort=source does not state; tools=none; sampling=pass@1; harness=aird; deployment=claude-mythos-5." }, { "model_id": "claude-sonnet-5", "benchmark_id": "ai_rd_quadruped_rl", "score": 19.94, "reference_url": "https://www.anthropic.com/claude-sonnet-5-system-card", "reported_setting": { "effort": "source does not state", "tools": "none", "sampling": "pass@1", "harness": "aird" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Anthropic Claude Sonnet 5 System Card Table 2.3.3.A, page 26: Claude Sonnet 5; AI R&D rule-out evaluations [Quadruped RL / highest score, no hparams]=19.94. Source setting: effort=source does not state; tools=none; sampling=pass@1; harness=aird." }, { "model_id": "claude-opus-4.7", "benchmark_id": "ai_rd_novel_compiler", "score": 70.4, "reference_url": "https://www.anthropic.com/claude-sonnet-5-system-card", "reported_setting": { "effort": "source does not state", "tools": "none", "sampling": "pass@1", "harness": "aird" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Anthropic Claude Sonnet 5 System Card Table 2.3.3.A, page 26: Claude Opus 4.7; AI R&D rule-out evaluations [Novel Compiler / pass rate complex tests]=70.4%. Source setting: effort=source does not state; tools=none; sampling=pass@1; harness=aird." }, { "model_id": "claude-fable-5", "benchmark_id": "ai_rd_novel_compiler", "score": 85.3, "reference_url": "https://www.anthropic.com/claude-sonnet-5-system-card", "reported_setting": { "effort": "source does not state", "tools": "none", "sampling": "pass@1", "harness": "aird", "deployment": "claude-mythos-5" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Anthropic Claude Sonnet 5 System Card Table 2.3.3.A, page 26: Claude Mythos 5; AI R&D rule-out evaluations [Novel Compiler / pass rate complex tests]=85.3%. Source setting: effort=source does not state; tools=none; sampling=pass@1; harness=aird; deployment=claude-mythos-5." }, { "model_id": "claude-sonnet-5", "benchmark_id": "ai_rd_novel_compiler", "score": 76.1, "reference_url": "https://www.anthropic.com/claude-sonnet-5-system-card", "reported_setting": { "effort": "source does not state", "tools": "none", "sampling": "pass@1", "harness": "aird" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Anthropic Claude Sonnet 5 System Card Table 2.3.3.A, page 26: Claude Sonnet 5; AI R&D rule-out evaluations [Novel Compiler / pass rate complex tests]=76.1%. Source setting: effort=source does not state; tools=none; sampling=pass@1; harness=aird." }, { "model_id": "claude-sonnet-5", "benchmark_id": "exploitbench", "score": 31.0, "reference_url": "https://www.anthropic.com/claude-sonnet-5-system-card", "reported_setting": { "effort": "source does not state", "tools": "cyber execution environment", "sampling": "pass@1", "harness": "exploit" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Anthropic Claude Sonnet 5 System Card Figure 3.2.1.A, page 31: Claude Sonnet 5; ExploitBench [Cap%]=31. Source setting: effort=source does not state; tools=cyber execution environment; sampling=pass@1; harness=exploit." }, { "model_id": "claude-sonnet-4.6", "benchmark_id": "exploitbench", "score": 24.0, "reference_url": "https://www.anthropic.com/claude-sonnet-5-system-card", "reported_setting": { "effort": "source does not state", "tools": "cyber execution environment", "sampling": "pass@1", "harness": "exploit" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Anthropic Claude Sonnet 5 System Card Figure 3.2.1.A, page 31: Sonnet 4.6; ExploitBench [Cap%]=24. Source setting: effort=source does not state; tools=cyber execution environment; sampling=pass@1; harness=exploit." }, { "model_id": "claude-sonnet-4.6", "benchmark_id": "oss_fuzz_control_flow_hijack_count", "score": 0.0, "reference_url": "https://www.anthropic.com/claude-sonnet-5-system-card", "reported_setting": { "effort": "source does not state", "tools": "cyber execution environment", "sampling": "pass@1", "harness": "oss" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Anthropic Claude Sonnet 5 System Card Figure 3.2.2.A, page 32: Claude Sonnet 4.6; OSS-Fuzz exploit-primitive discovery [grade 1.0 count]=0. Source setting: effort=source does not state; tools=cyber execution environment; sampling=pass@1; harness=oss." }, { "model_id": "claude-sonnet-4.6", "benchmark_id": "oss_fuzz_any_progress", "score": 31.6, "reference_url": "https://www.anthropic.com/claude-sonnet-5-system-card", "reported_setting": { "effort": "source does not state", "tools": "cyber execution environment", "sampling": "pass@1", "harness": "oss" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Anthropic Claude Sonnet 5 System Card Figure 3.2.2.A, page 32: Claude Sonnet 4.6; OSS-Fuzz exploit-primitive discovery [any progress]=31.6%. Source setting: effort=source does not state; tools=cyber execution environment; sampling=pass@1; harness=oss." }, { "model_id": "claude-opus-4.8", "benchmark_id": "oss_fuzz_control_flow_hijack_count", "score": 0.0, "reference_url": "https://www.anthropic.com/claude-sonnet-5-system-card", "reported_setting": { "effort": "source does not state", "tools": "cyber execution environment", "sampling": "pass@1", "harness": "oss" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Anthropic Claude Sonnet 5 System Card Figure 3.2.2.A, page 32: Claude Opus 4.8; OSS-Fuzz exploit-primitive discovery [grade 1.0 count]=0. Source setting: effort=source does not state; tools=cyber execution environment; sampling=pass@1; harness=oss." }, { "model_id": "claude-opus-4.8", "benchmark_id": "oss_fuzz_any_progress", "score": 61.5, "reference_url": "https://www.anthropic.com/claude-sonnet-5-system-card", "reported_setting": { "effort": "source does not state", "tools": "cyber execution environment", "sampling": "pass@1", "harness": "oss" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Anthropic Claude Sonnet 5 System Card Figure 3.2.2.A, page 32: Claude Opus 4.8; OSS-Fuzz exploit-primitive discovery [any progress]=61.5%. Source setting: effort=source does not state; tools=cyber execution environment; sampling=pass@1; harness=oss." }, { "model_id": "claude-fable-5", "benchmark_id": "oss_fuzz_control_flow_hijack_count", "score": 13.0, "reference_url": "https://www.anthropic.com/claude-sonnet-5-system-card", "reported_setting": { "effort": "source does not state", "tools": "cyber execution environment", "sampling": "pass@1", "harness": "oss", "deployment": "claude-mythos-5" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Anthropic Claude Sonnet 5 System Card Figure 3.2.2.A, page 32: Claude Mythos 5; OSS-Fuzz exploit-primitive discovery [grade 1.0 count]=13. Source setting: effort=source does not state; tools=cyber execution environment; sampling=pass@1; harness=oss; deployment=claude-mythos-5." }, { "model_id": "claude-fable-5", "benchmark_id": "oss_fuzz_any_progress", "score": 80.0, "reference_url": "https://www.anthropic.com/claude-sonnet-5-system-card", "reported_setting": { "effort": "source does not state", "tools": "cyber execution environment", "sampling": "pass@1", "harness": "oss", "deployment": "claude-mythos-5" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Anthropic Claude Sonnet 5 System Card Figure 3.2.2.A, page 32: Claude Mythos 5; OSS-Fuzz exploit-primitive discovery [any progress]=80.0%. Source setting: effort=source does not state; tools=cyber execution environment; sampling=pass@1; harness=oss; deployment=claude-mythos-5." }, { "model_id": "claude-sonnet-5", "benchmark_id": "oss_fuzz_control_flow_hijack_count", "score": 0.0, "reference_url": "https://www.anthropic.com/claude-sonnet-5-system-card", "reported_setting": { "effort": "source does not state", "tools": "cyber execution environment", "sampling": "pass@1", "harness": "oss" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Anthropic Claude Sonnet 5 System Card Figure 3.2.2.A, page 32: Claude Sonnet 5; OSS-Fuzz exploit-primitive discovery [grade 1.0 count]=0. Source setting: effort=source does not state; tools=cyber execution environment; sampling=pass@1; harness=oss." }, { "model_id": "claude-sonnet-5", "benchmark_id": "oss_fuzz_any_progress", "score": 54.5, "reference_url": "https://www.anthropic.com/claude-sonnet-5-system-card", "reported_setting": { "effort": "source does not state", "tools": "cyber execution environment", "sampling": "pass@1", "harness": "oss" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Anthropic Claude Sonnet 5 System Card Figure 3.2.2.A, page 32: Claude Sonnet 5; OSS-Fuzz exploit-primitive discovery [any progress]=54.5%. Source setting: effort=source does not state; tools=cyber execution environment; sampling=pass@1; harness=oss." }, { "model_id": "claude-fable-5", "benchmark_id": "cybergym", "score": 83.8, "reference_url": "https://www.anthropic.com/claude-sonnet-5-system-card", "reported_setting": { "effort": "source does not state", "tools": "cyber execution environment", "sampling": "pass@1", "harness": "cybergym", "deployment": "claude-mythos-5" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Anthropic Claude Sonnet 5 System Card Figure 3.2.3.A, page 33: Claude Mythos 5; CyberGym vulnerability discovery [targeted]=83.8%. Source setting: effort=source does not state; tools=cyber execution environment; sampling=pass@1; harness=cybergym; deployment=claude-mythos-5." }, { "model_id": "claude-sonnet-5", "benchmark_id": "cybergym", "score": 52.7, "reference_url": "https://www.anthropic.com/claude-sonnet-5-system-card", "reported_setting": { "effort": "source does not state", "tools": "cyber execution environment", "sampling": "pass@1", "harness": "cybergym" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Anthropic Claude Sonnet 5 System Card Figure 3.2.3.A, page 33: Claude Sonnet 5; CyberGym vulnerability discovery [targeted]=52.7%. Source setting: effort=source does not state; tools=cyber execution environment; sampling=pass@1; harness=cybergym." }, { "model_id": "claude-sonnet-4.6", "benchmark_id": "firefox_147_exploit_development_working_exploit", "score": 0.0, "reference_url": "https://www.anthropic.com/claude-sonnet-5-system-card", "reported_setting": { "effort": "source does not state", "tools": "cyber execution environment", "sampling": "pass@1", "harness": "firefox" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Anthropic Claude Sonnet 5 System Card Figure 3.2.4.A, page 35: Claude Sonnet 4.6; Firefox 147 exploit development [working]=0.0%. Source setting: effort=source does not state; tools=cyber execution environment; sampling=pass@1; harness=firefox." }, { "model_id": "claude-sonnet-4.6", "benchmark_id": "firefox_147_exploit_development_any_success", "score": 8.8, "reference_url": "https://www.anthropic.com/claude-sonnet-5-system-card", "reported_setting": { "effort": "source does not state", "tools": "cyber execution environment", "sampling": "pass@1", "harness": "firefox" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Anthropic Claude Sonnet 5 System Card Figure 3.2.4.A, page 35: Claude Sonnet 4.6; Firefox 147 exploit development [any success]=8.8%. Source setting: effort=source does not state; tools=cyber execution environment; sampling=pass@1; harness=firefox." }, { "model_id": "claude-opus-4.8", "benchmark_id": "firefox_147_exploit_development_working_exploit", "score": 8.8, "reference_url": "https://www.anthropic.com/claude-sonnet-5-system-card", "reported_setting": { "effort": "source does not state", "tools": "cyber execution environment", "sampling": "pass@1", "harness": "firefox" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Anthropic Claude Sonnet 5 System Card Figure 3.2.4.A, page 35: Claude Opus 4.8; Firefox 147 exploit development [working]=8.8%. Source setting: effort=source does not state; tools=cyber execution environment; sampling=pass@1; harness=firefox." }, { "model_id": "claude-opus-4.8", "benchmark_id": "firefox_147_exploit_development_any_success", "score": 68.8, "reference_url": "https://www.anthropic.com/claude-sonnet-5-system-card", "reported_setting": { "effort": "source does not state", "tools": "cyber execution environment", "sampling": "pass@1", "harness": "firefox" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Anthropic Claude Sonnet 5 System Card Figure 3.2.4.A, page 35: Claude Opus 4.8; Firefox 147 exploit development [any success]=68.8%. Source setting: effort=source does not state; tools=cyber execution environment; sampling=pass@1; harness=firefox." }, { "model_id": "claude-fable-5", "benchmark_id": "firefox_147_exploit_development_working_exploit", "score": 88.4, "reference_url": "https://www.anthropic.com/claude-sonnet-5-system-card", "reported_setting": { "effort": "source does not state", "tools": "cyber execution environment", "sampling": "pass@1", "harness": "firefox", "deployment": "claude-mythos-5" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Anthropic Claude Sonnet 5 System Card Figure 3.2.4.A, page 35: Claude Mythos 5; Firefox 147 exploit development [working]=88.4%. Source setting: effort=source does not state; tools=cyber execution environment; sampling=pass@1; harness=firefox; deployment=claude-mythos-5." }, { "model_id": "claude-fable-5", "benchmark_id": "firefox_147_exploit_development_any_success", "score": 90.0, "reference_url": "https://www.anthropic.com/claude-sonnet-5-system-card", "reported_setting": { "effort": "source does not state", "tools": "cyber execution environment", "sampling": "pass@1", "harness": "firefox", "deployment": "claude-mythos-5" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Anthropic Claude Sonnet 5 System Card Figure 3.2.4.A, page 35: Claude Mythos 5; Firefox 147 exploit development [any success]=90.0%. Source setting: effort=source does not state; tools=cyber execution environment; sampling=pass@1; harness=firefox; deployment=claude-mythos-5." }, { "model_id": "claude-sonnet-5", "benchmark_id": "firefox_147_exploit_development_working_exploit", "score": 0.0, "reference_url": "https://www.anthropic.com/claude-sonnet-5-system-card", "reported_setting": { "effort": "source does not state", "tools": "cyber execution environment", "sampling": "pass@1", "harness": "firefox" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Anthropic Claude Sonnet 5 System Card Figure 3.2.4.A, page 35: Claude Sonnet 5; Firefox 147 exploit development [working]=0.0%. Source setting: effort=source does not state; tools=cyber execution environment; sampling=pass@1; harness=firefox." }, { "model_id": "claude-sonnet-5", "benchmark_id": "firefox_147_exploit_development_any_success", "score": 13.2, "reference_url": "https://www.anthropic.com/claude-sonnet-5-system-card", "reported_setting": { "effort": "source does not state", "tools": "cyber execution environment", "sampling": "pass@1", "harness": "firefox" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Anthropic Claude Sonnet 5 System Card Figure 3.2.4.A, page 35: Claude Sonnet 5; Firefox 147 exploit development [any success]=13.2%. Source setting: effort=source does not state; tools=cyber execution environment; sampling=pass@1; harness=firefox." }, { "model_id": "claude-sonnet-5", "benchmark_id": "swe_bench_pro", "score": 63.2, "reference_url": "https://www.anthropic.com/claude-sonnet-5-system-card", "reported_setting": { "effort": "max", "tools": "none", "sampling": "avg 5 trials unless source states otherwise", "harness": "table8" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Anthropic Claude Sonnet 5 System Card Table 8.1.A, page 115: Claude Sonnet 5; SWE-bench Pro [summary]=63.2. Source setting: effort=max; tools=none; sampling=avg 5 trials unless source states otherwise; harness=table8." }, { "model_id": "gemini-3.5-flash", "benchmark_id": "swe_bench_pro", "score": 55.1, "reference_url": "https://deepmind.google/models/model-cards/gemini-3-5-flash/", "reported_setting": { "mode": "thinking/reasoning", "effort": "medium (runtime default)", "tools": "code execution", "sampling": "Gemini avg 5 runs, single attempt per run", "judge": "benchmark-specified", "harness": "internal Antigravity for Gemini; provider/public for others", "temperature": "default", "snapshot": "May 2026" }, "matches_canonical": false, "source_type": "official_model_card", "audit_status": "verified", "notes": "Replace third-party provenance with Google provider-own May evaluation for SWE-Bench Pro (Public); numeric score is unchanged." }, { "model_id": "claude-sonnet-5", "benchmark_id": "terminal_bench_2_1", "score": 80.4, "reference_url": "https://www.anthropic.com/claude-sonnet-5-system-card", "reported_setting": { "effort": "max", "tools": "none", "sampling": "avg 5 trials unless source states otherwise", "harness": "table8" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Anthropic Claude Sonnet 5 System Card Table 8.1.A, page 115: Claude Sonnet 5; Terminal-Bench 2.1 [summary]=80.4. Source setting: effort=max; tools=none; sampling=avg 5 trials unless source states otherwise; harness=table8.", "candidates": [ { "score": 74.6, "reference_url": "https://www.tbench.ai/leaderboard/terminal-bench/2.1", "source_type": "leaderboard", "reported_setting": { "mode": "thinking", "effort": "high", "prompt_style": "not disclosed", "temperature": "not disclosed", "context": "not disclosed", "output_limit": "not disclosed", "trials": 5, "dataset_version": "Terminal-Bench 2.1", "dataset_split": "89 tasks", "tools": "Claude Code terminal toolset", "sampling": "pass@1", "aggregation": "mean across five attempts per task", "judge": "Terminal-Bench 2.1 executable verifier", "harness": "Claude Code official leaderboard submission" }, "notes": "Current original-leaderboard configuration differs from the provider release value and remains a separate candidate." } ] }, { "model_id": "claude-sonnet-4.6", "benchmark_id": "terminal_bench_2_1", "score": 67.0, "reference_url": "https://www.anthropic.com/claude-sonnet-5-system-card", "reported_setting": { "effort": "max", "tools": "none", "sampling": "avg 5 trials unless source states otherwise", "harness": "table8" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Anthropic Claude Sonnet 5 System Card Table 8.1.A, page 115: Claude Sonnet 4.6; Terminal-Bench 2.1 [summary]=67.0. Source setting: effort=max; tools=none; sampling=avg 5 trials unless source states otherwise; harness=table8." }, { "model_id": "gemini-3.5-flash", "benchmark_id": "terminal_bench_2_1", "score": 76.2, "reference_url": "https://deepmind.google/models/model-cards/gemini-3-5-flash/", "reported_setting": { "mode": "thinking/reasoning", "effort": "medium (runtime default)", "tools": "terminal agent", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "Terminus-2; Gemini self-computed, others public leaderboard", "temperature": "default", "snapshot": "May 2026" }, "matches_canonical": false, "source_type": "official_model_card", "audit_status": "verified", "notes": "Replace third-party provenance with Google provider-own May evaluation for Terminal-Bench 2.1; numeric score is unchanged." }, { "model_id": "claude-sonnet-5", "benchmark_id": "browsecomp", "score": 79.4, "reference_url": "https://www.anthropic.com/claude-opus-5-system-card", "reported_setting": { "effort": "high", "tools": "benchmark-specific" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Figure 8.10.2.B / Claude Sonnet 5 / high: Fixed 10M token budget with web/programmatic/code tools and context compaction.", "candidates": [ { "score": 86.6, "reference_url": "https://www.anthropic.com/claude-sonnet-5-system-card", "source_type": "model_card", "reported_setting": { "effort": "max", "tools": "none", "sampling": "avg 5 trials unless source states otherwise", "harness": "table8" }, "notes": "Anthropic Claude Sonnet 5 System Card Table 8.1.A, page 115: Claude Sonnet 5; BrowseComp / multi agent [summary]=86.6. Source setting: effort=max; tools=none; sampling=avg 5 trials unless source states otherwise; harness=table8." }, { "score": 79.3, "reference_url": "https://www.anthropic.com/claude-sonnet-5-system-card", "source_type": "model_card", "reported_setting": { "effort": "max", "tools": "web search, fetch, programmatic tools, code execution", "sampling": "pass@1", "harness": "browse", "token_cap": "1M" }, "notes": "Anthropic Claude Sonnet 5 System Card Figure 8.9.2.A, page 124: Claude Sonnet 5; BrowseComp [1M]=79.3%. Source setting: effort=max; tools=web search, fetch, programmatic tools, code execution; sampling=pass@1; harness=browse; token_cap=1M." }, { "score": 82.9, "reference_url": "https://www.anthropic.com/claude-sonnet-5-system-card", "source_type": "model_card", "reported_setting": { "effort": "max", "tools": "web search, fetch, programmatic tools, code execution", "sampling": "pass@1", "harness": "browse", "token_cap": "3M" }, "notes": "Anthropic Claude Sonnet 5 System Card Figure 8.9.2.A, page 124: Claude Sonnet 5; BrowseComp [3M]=82.9%. Source setting: effort=max; tools=web search, fetch, programmatic tools, code execution; sampling=pass@1; harness=browse; token_cap=3M." }, { "score": 84.7, "reference_url": "https://www.anthropic.com/claude-sonnet-5-system-card", "source_type": "model_card", "reported_setting": { "effort": "max", "tools": "web search, fetch, programmatic tools, code execution", "sampling": "pass@1", "harness": "browse", "token_cap": "10M" }, "notes": "Anthropic Claude Sonnet 5 System Card Figure 8.9.2.A, page 124: Claude Sonnet 5; BrowseComp [10M]=84.7%. Source setting: effort=max; tools=web search, fetch, programmatic tools, code execution; sampling=pass@1; harness=browse; token_cap=10M." }, { "score": 84.7, "reference_url": "https://www.anthropic.com/claude-sonnet-5-system-card", "source_type": "model_card", "reported_setting": { "effort": "max", "tools": "none", "sampling": "avg 5 trials unless source states otherwise", "harness": "table8" }, "notes": "Anthropic Claude Sonnet 5 System Card Table 8.1.A, page 115: Claude Sonnet 5; BrowseComp / single agent [summary]=84.7. Source setting: effort=max; tools=none; sampling=avg 5 trials unless source states otherwise; harness=table8." }, { "score": 59.9, "reference_url": "https://www.anthropic.com/claude-opus-5-system-card", "source_type": "model_card", "reported_setting": { "effort": "low", "tools": "benchmark-specific" }, "notes": "Figure 8.10.2.B / Claude Sonnet 5 / low: Fixed 10M token budget with web/programmatic/code tools and context compaction." }, { "score": 71.4, "reference_url": "https://www.anthropic.com/claude-opus-5-system-card", "source_type": "model_card", "reported_setting": { "effort": "medium", "tools": "benchmark-specific" }, "notes": "Figure 8.10.2.B / Claude Sonnet 5 / medium: Fixed 10M token budget with web/programmatic/code tools and context compaction." }, { "score": 82.5, "reference_url": "https://www.anthropic.com/claude-opus-5-system-card", "source_type": "model_card", "reported_setting": { "effort": "xhigh", "tools": "benchmark-specific" }, "notes": "Figure 8.10.2.B / Claude Sonnet 5 / xhigh: Fixed 10M token budget with web/programmatic/code tools and context compaction." }, { "score": 79.3, "reference_url": "https://www.anthropic.com/claude-opus-5-system-card", "source_type": "model_card", "reported_setting": { "effort": "max", "tools": "web/programmatic/code", "harness": "token-budget scaling" }, "notes": "Figure 8.10.2.A / Claude Sonnet 5 / 1M: Token-budget scaling; source reports exact labels at 1M, 3M, and 10M. Budget variants are candidates; fixed 10M effort ladder supplies the preferred canonical-setting value." }, { "score": 82.9, "reference_url": "https://www.anthropic.com/claude-opus-5-system-card", "source_type": "model_card", "reported_setting": { "effort": "max", "tools": "web/programmatic/code", "harness": "token-budget scaling" }, "notes": "Figure 8.10.2.A / Claude Sonnet 5 / 3M: Token-budget scaling; source reports exact labels at 1M, 3M, and 10M. Budget variants are candidates; fixed 10M effort ladder supplies the preferred canonical-setting value." } ] }, { "model_id": "claude-sonnet-5", "benchmark_id": "hle", "score": 39.3, "reference_url": "https://www.anthropic.com/claude-opus-5-system-card", "reported_setting": { "effort": "high", "tools": "benchmark-specific" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Figure 8.10.1.B / Claude Sonnet 5 / high: 2,500 questions; 1M token cap; no tools and no context compaction.", "candidates": [ { "score": 43.2, "reference_url": "https://www.anthropic.com/claude-sonnet-5-system-card", "source_type": "model_card", "reported_setting": { "effort": "max", "tools": "none", "sampling": "avg 5 trials unless source states otherwise", "harness": "table8" }, "notes": "Anthropic Claude Sonnet 5 System Card Table 8.1.A, page 115: Claude Sonnet 5; Humanity’s Last Exam / no tools [summary]=43.2. Source setting: effort=max; tools=none; sampling=avg 5 trials unless source states otherwise; harness=table8." }, { "score": 25.8, "reference_url": "https://www.anthropic.com/claude-opus-5-system-card", "source_type": "model_card", "reported_setting": { "effort": "low", "tools": "benchmark-specific" }, "notes": "Figure 8.10.1.B / Claude Sonnet 5 / low: 2,500 questions; 1M token cap; no tools and no context compaction." }, { "score": 34.4, "reference_url": "https://www.anthropic.com/claude-opus-5-system-card", "source_type": "model_card", "reported_setting": { "effort": "medium", "tools": "benchmark-specific" }, "notes": "Figure 8.10.1.B / Claude Sonnet 5 / medium: 2,500 questions; 1M token cap; no tools and no context compaction." }, { "score": 42.0, "reference_url": "https://www.anthropic.com/claude-opus-5-system-card", "source_type": "model_card", "reported_setting": { "effort": "xhigh", "tools": "benchmark-specific" }, "notes": "Figure 8.10.1.B / Claude Sonnet 5 / xhigh: 2,500 questions; 1M token cap; no tools and no context compaction." }, { "score": 44.1, "reference_url": "https://www.anthropic.com/claude-opus-5-system-card", "source_type": "model_card", "reported_setting": { "effort": "max", "tools": "benchmark-specific" }, "notes": "Figure 8.10.1.B / Claude Sonnet 5 / max: 2,500 questions; 1M token cap; no tools and no context compaction." } ] }, { "model_id": "gemini-3.5-flash", "benchmark_id": "hle", "score": 40.2, "reference_url": "https://deepmind.google/models/model-cards/gemini-3-5-flash/", "reported_setting": { "mode": "thinking/reasoning", "effort": "medium (runtime default)", "tools": "none", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "Google self-computed for Gemini; provider for others", "temperature": "default", "snapshot": "May 2026" }, "matches_canonical": false, "source_type": "official_model_card", "audit_status": "verified", "notes": "Replace third-party provenance with Google provider-own May evaluation for Humanity's Last Exam: full text+MM; numeric score is unchanged." }, { "model_id": "claude-sonnet-4.6", "benchmark_id": "hle_tools", "score": 46.8, "reference_url": "https://www.anthropic.com/claude-sonnet-5-system-card", "reported_setting": { "effort": "max", "tools": "web search, fetch, and code execution", "sampling": "avg 5 trials unless source states otherwise", "harness": "table8" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Anthropic Claude Sonnet 5 System Card Table 8.1.A, page 115: Claude Sonnet 4.6; Humanity’s Last Exam / with tools [summary]=46.8. Source setting: effort=max; tools=web search, fetch, and code execution; sampling=avg 5 trials unless source states otherwise; harness=table8.", "candidates": [ { "score": 34.9, "reference_url": "https://www.anthropic.com/claude-sonnet-5-system-card", "source_type": "model_card", "reported_setting": { "effort": "low", "tools": "web search, fetch, and code execution", "sampling": "pass@1", "harness": "hle" }, "notes": "Anthropic Claude Sonnet 5 System Card Figure 8.9.1.B, page 123: Claude Sonnet 4.6; Humanity’s Last Exam [low]=34.9%. Source setting: effort=low; tools=web search, fetch, and code execution; sampling=pass@1; harness=hle." }, { "score": 46.5, "reference_url": "https://www.anthropic.com/claude-sonnet-5-system-card", "source_type": "model_card", "reported_setting": { "effort": "medium", "tools": "web search, fetch, and code execution", "sampling": "pass@1", "harness": "hle" }, "notes": "Anthropic Claude Sonnet 5 System Card Figure 8.9.1.B, page 123: Claude Sonnet 4.6; Humanity’s Last Exam [med]=46.5%. Source setting: effort=medium; tools=web search, fetch, and code execution; sampling=pass@1; harness=hle." }, { "score": 49.7, "reference_url": "https://www.anthropic.com/claude-sonnet-5-system-card", "source_type": "model_card", "reported_setting": { "effort": "high", "tools": "web search, fetch, and code execution", "sampling": "pass@1", "harness": "hle" }, "notes": "Anthropic Claude Sonnet 5 System Card Figure 8.9.1.B, page 123: Claude Sonnet 4.6; Humanity’s Last Exam [high]=49.7%. Source setting: effort=high; tools=web search, fetch, and code execution; sampling=pass@1; harness=hle." } ] }, { "model_id": "claude-sonnet-5", "benchmark_id": "osworld_verified", "score": 81.2, "reference_url": "https://www.anthropic.com/claude-sonnet-5-system-card", "reported_setting": { "effort": "max", "tools": "none", "sampling": "avg 5 trials unless source states otherwise", "harness": "table8" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Anthropic Claude Sonnet 5 System Card Table 8.1.A, page 115: Claude Sonnet 5; OSWorld-Verified [summary]=81.2. Source setting: effort=max; tools=none; sampling=avg 5 trials unless source states otherwise; harness=table8.", "candidates": [ { "score": 81.2, "reference_url": "https://www.anthropic.com/claude-sonnet-5-system-card", "source_type": "model_card", "reported_setting": { "effort": "max", "tools": "GUI computer-use harness", "sampling": "avg 5 trials unless source states otherwise", "harness": "osworld" }, "notes": "Anthropic Claude Sonnet 5 System Card Figure 8.10.2.A, page 126: Claude Sonnet 5; OSWorld-Verified [summary]=81.2%. Source setting: effort=max; tools=GUI computer-use harness; sampling=avg 5 trials unless source states otherwise; harness=osworld." } ] }, { "model_id": "claude-sonnet-4.6", "benchmark_id": "osworld_verified", "score": 78.5, "reference_url": "https://www.anthropic.com/claude-sonnet-5-system-card", "reported_setting": { "effort": "max", "tools": "none", "sampling": "avg 5 trials unless source states otherwise", "harness": "table8" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Anthropic Claude Sonnet 5 System Card Table 8.1.A, page 115: Claude Sonnet 4.6; OSWorld-Verified [summary]=78.5. Source setting: effort=max; tools=none; sampling=avg 5 trials unless source states otherwise; harness=table8.", "candidates": [ { "score": 78.5, "reference_url": "https://www.anthropic.com/claude-sonnet-5-system-card", "source_type": "model_card", "reported_setting": { "effort": "max", "tools": "GUI computer-use harness", "sampling": "avg 5 trials unless source states otherwise", "harness": "osworld" }, "notes": "Anthropic Claude Sonnet 5 System Card Figure 8.10.2.A, page 126: Claude Sonnet 4.6; OSWorld-Verified [summary]=78.5%. Source setting: effort=max; tools=GUI computer-use harness; sampling=avg 5 trials unless source states otherwise; harness=osworld." }, { "score": 72.5, "reference_url": "https://deepmind.google/models/model-cards/gemini-3-5-flash/", "source_type": "official_model_card", "reported_setting": { "mode": "thinking/reasoning", "effort": "max/best available", "tools": "pyautogui + UI-specific function declarations", "sampling": "Gemini avg 5 runs, single attempt per run", "judge": "benchmark-specified", "harness": "OSWorld Docker, 1080p, max 100 steps", "temperature": "default", "snapshot": "May 2026" }, "notes": "Google Gemini 3.5 Flash May 2026 table: OSWorld-Verified; Docker, 1080p, max 100 steps. Settings and provenance are preserved per cell." } ] }, { "model_id": "gemini-3.5-flash", "benchmark_id": "osworld_verified", "score": 78.4, "reference_url": "https://deepmind.google/models/model-cards/gemini-3-5-flash/", "reported_setting": { "mode": "thinking/reasoning", "effort": "medium (runtime default)", "tools": "pyautogui + UI-specific function declarations", "sampling": "Gemini avg 5 runs, single attempt per run", "judge": "benchmark-specified", "harness": "OSWorld Docker, 1080p, max 100 steps", "temperature": "default", "snapshot": "May 2026" }, "matches_canonical": false, "source_type": "official_model_card", "audit_status": "verified", "notes": "Replace third-party provenance with Google provider-own May evaluation for OSWorld-Verified; numeric score is unchanged.", "candidates": [ { "score": 78.4, "reference_url": "https://www.anthropic.com/claude-sonnet-5-system-card", "source_type": "model_card", "reported_setting": { "effort": "source does not state", "tools": "GUI computer-use harness", "sampling": "avg 5 trials unless source states otherwise", "harness": "osworld" }, "notes": "Anthropic Claude Sonnet 5 System Card Figure 8.10.2.A, page 126: Gemini 3.5 Flash; OSWorld-Verified [summary]=78.4%. Source setting: effort=source does not state; tools=GUI computer-use harness; sampling=avg 5 trials unless source states otherwise; harness=osworld." } ] }, { "model_id": "claude-sonnet-4.6", "benchmark_id": "frontiercode_main_v1", "score": 15.1, "reference_url": "https://www.anthropic.com/claude-sonnet-5-system-card", "reported_setting": { "effort": "max", "tools": "none", "sampling": "avg 5 trials unless source states otherwise", "harness": "table8" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Anthropic Claude Sonnet 5 System Card Table 8.1.A, page 115: Claude Sonnet 4.6; FrontierCode v1 [summary]=15.1. Source setting: effort=max; tools=none; sampling=avg 5 trials unless source states otherwise; harness=table8.", "candidates": [ { "score": 13.6, "reference_url": "https://www.anthropic.com/claude-sonnet-5-system-card", "source_type": "model_card", "reported_setting": { "effort": "medium", "tools": "agentic benchmark harness", "sampling": "pass@1", "harness": "frontier" }, "notes": "Anthropic Claude Sonnet 5 System Card Figure 8.4.A, page 118: Claude Sonnet 4.6; FrontierCode v1 [med]=13.6%. Source setting: effort=medium; tools=agentic benchmark harness; sampling=pass@1; harness=frontier." }, { "score": 15.1, "reference_url": "https://www.anthropic.com/claude-sonnet-5-system-card", "source_type": "model_card", "reported_setting": { "effort": "high", "tools": "agentic benchmark harness", "sampling": "pass@1", "harness": "frontier" }, "notes": "Anthropic Claude Sonnet 5 System Card Figure 8.4.A, page 118: Claude Sonnet 4.6; FrontierCode v1 [high]=15.1%. Source setting: effort=high; tools=agentic benchmark harness; sampling=pass@1; harness=frontier." }, { "score": 13.2, "reference_url": "https://www.anthropic.com/claude-sonnet-5-system-card", "source_type": "model_card", "reported_setting": { "effort": "max", "tools": "agentic benchmark harness", "sampling": "pass@1", "harness": "frontier" }, "notes": "Anthropic Claude Sonnet 5 System Card Figure 8.4.A, page 118: Claude Sonnet 4.6; FrontierCode v1 [max]=13.2%. Source setting: effort=max; tools=agentic benchmark harness; sampling=pass@1; harness=frontier." } ] }, { "model_id": "claude-sonnet-5", "benchmark_id": "automation_bench_private_heldout", "score": 13.5, "reference_url": "https://www.anthropic.com/claude-sonnet-5-system-card", "reported_setting": { "effort": "max", "tools": "none", "sampling": "avg 5 trials unless source states otherwise", "harness": "table8" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Anthropic Claude Sonnet 5 System Card Table 8.1.A, page 115: Claude Sonnet 5; AutomationBench [summary]=13.5. Source setting: effort=max; tools=none; sampling=avg 5 trials unless source states otherwise; harness=table8.", "candidates": [ { "score": 13.5, "reference_url": "https://www.anthropic.com/claude-sonnet-5-system-card", "source_type": "model_card", "reported_setting": { "effort": "max", "tools": "agentic benchmark harness", "sampling": "pass@1", "harness": "automation" }, "notes": "Anthropic Claude Sonnet 5 System Card Figure 8.11.6.A, page 136: Claude Sonnet 5; AutomationBench [max]=13.5%. Source setting: effort=max; tools=agentic benchmark harness; sampling=pass@1; harness=automation." } ] }, { "model_id": "claude-sonnet-4.6", "benchmark_id": "automation_bench_private_heldout", "score": 5.3, "reference_url": "https://www.anthropic.com/claude-sonnet-5-system-card", "reported_setting": { "effort": "max", "tools": "none", "sampling": "avg 5 trials unless source states otherwise", "harness": "table8" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Anthropic Claude Sonnet 5 System Card Table 8.1.A, page 115: Claude Sonnet 4.6; AutomationBench [summary]=5.3. Source setting: effort=max; tools=none; sampling=avg 5 trials unless source states otherwise; harness=table8.", "candidates": [ { "score": 5.3, "reference_url": "https://www.anthropic.com/claude-sonnet-5-system-card", "source_type": "model_card", "reported_setting": { "effort": "max", "tools": "agentic benchmark harness", "sampling": "pass@1", "harness": "automation" }, "notes": "Anthropic Claude Sonnet 5 System Card Figure 8.11.6.A, page 136: Claude Sonnet 4.6; AutomationBench [max]=5.3%. Source setting: effort=max; tools=agentic benchmark harness; sampling=pass@1; harness=automation." } ] }, { "model_id": "claude-sonnet-5", "benchmark_id": "legal_agent_benchmark_public", "score": 8.9, "reference_url": "https://www.anthropic.com/claude-sonnet-5-system-card", "reported_setting": { "effort": "max", "tools": "none", "sampling": "avg 5 trials unless source states otherwise", "harness": "table8" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Anthropic Claude Sonnet 5 System Card Table 8.1.A, page 115: Claude Sonnet 5; Legal Agent Benchmark / Full Public Set [summary]=8.9. Source setting: effort=max; tools=none; sampling=avg 5 trials unless source states otherwise; harness=table8.", "candidates": [ { "score": 8.92, "reference_url": "https://www.anthropic.com/claude-sonnet-5-system-card", "source_type": "model_card", "reported_setting": { "effort": "max", "tools": "agentic benchmark harness", "sampling": "pass@1", "harness": "legal" }, "notes": "Anthropic Claude Sonnet 5 System Card Section 8.11.3, page 133: Claude Sonnet 5; Legal Agent Benchmark / Full Public Set [all pass]=8.92% ±0.36 (n=5). Source setting: effort=max; tools=agentic benchmark harness; sampling=pass@1; harness=legal." } ] }, { "model_id": "claude-sonnet-4.6", "benchmark_id": "legal_agent_benchmark_public", "score": 8.0, "reference_url": "https://www.anthropic.com/claude-sonnet-5-system-card", "reported_setting": { "effort": "max", "tools": "none", "sampling": "avg 5 trials unless source states otherwise", "harness": "table8" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Anthropic Claude Sonnet 5 System Card Table 8.1.A, page 115: Claude Sonnet 4.6; Legal Agent Benchmark / Full Public Set [summary]=8.0. Source setting: effort=max; tools=none; sampling=avg 5 trials unless source states otherwise; harness=table8.", "candidates": [ { "score": 8.0, "reference_url": "https://www.anthropic.com/claude-sonnet-5-system-card", "source_type": "model_card", "reported_setting": { "effort": "max", "tools": "agentic benchmark harness", "sampling": "pass@1", "harness": "legal" }, "notes": "Anthropic Claude Sonnet 5 System Card Section 8.11.3, page 133: Claude Sonnet 4.6; Legal Agent Benchmark / Full Public Set [all pass]=8.00% ±0.19 (n=5). Source setting: effort=max; tools=agentic benchmark harness; sampling=pass@1; harness=legal." } ] }, { "model_id": "claude-sonnet-5", "benchmark_id": "legal_agent_benchmark_harvey_held_out", "score": 5.8, "reference_url": "https://www.anthropic.com/claude-sonnet-5-system-card", "reported_setting": { "effort": "max", "tools": "none", "sampling": "avg 5 trials unless source states otherwise", "harness": "table8" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Anthropic Claude Sonnet 5 System Card Table 8.1.A, page 115: Claude Sonnet 5; Legal Agent Benchmark / Harvey Held-Out Set [summary]=5.8. Source setting: effort=max; tools=none; sampling=avg 5 trials unless source states otherwise; harness=table8.", "candidates": [ { "score": 5.8, "reference_url": "https://www.anthropic.com/claude-sonnet-5-system-card", "source_type": "model_card", "reported_setting": { "effort": "max", "tools": "agentic benchmark harness", "sampling": "pass@1", "harness": "legal" }, "notes": "Anthropic Claude Sonnet 5 System Card Section 8.11.3, page 134: Claude Sonnet 5; Legal Agent Benchmark / Harvey Held-Out Set [all pass]=5.8%. Source setting: effort=max; tools=agentic benchmark harness; sampling=pass@1; harness=legal." } ] }, { "model_id": "claude-sonnet-4.6", "benchmark_id": "legal_agent_benchmark_harvey_held_out", "score": 5.4, "reference_url": "https://www.anthropic.com/claude-sonnet-5-system-card", "reported_setting": { "effort": "max", "tools": "none", "sampling": "avg 5 trials unless source states otherwise", "harness": "table8" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Anthropic Claude Sonnet 5 System Card Table 8.1.A, page 115: Claude Sonnet 4.6; Legal Agent Benchmark / Harvey Held-Out Set [summary]=5.4. Source setting: effort=max; tools=none; sampling=avg 5 trials unless source states otherwise; harness=table8." }, { "model_id": "gpt-5.5", "benchmark_id": "legal_agent_benchmark_harvey_held_out", "score": 2.1, "reference_url": "https://www.anthropic.com/claude-sonnet-5-system-card", "reported_setting": { "effort": "source does not state", "tools": "none", "sampling": "avg 5 trials unless source states otherwise", "harness": "table8" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Anthropic Claude Sonnet 5 System Card Table 8.1.A, page 115: GPT-5.5; Legal Agent Benchmark / Harvey Held-Out Set [summary]=2.1. Source setting: effort=source does not state; tools=none; sampling=avg 5 trials unless source states otherwise; harness=table8." }, { "model_id": "gemini-3.5-flash", "benchmark_id": "legal_agent_benchmark_harvey_held_out", "score": 0.8, "reference_url": "https://www.anthropic.com/claude-sonnet-5-system-card", "reported_setting": { "effort": "source does not state", "tools": "none", "sampling": "avg 5 trials unless source states otherwise", "harness": "table8" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Anthropic Claude Sonnet 5 System Card Table 8.1.A, page 115: Gemini 3.5 Flash; Legal Agent Benchmark / Harvey Held-Out Set [summary]=0.8. Source setting: effort=source does not state; tools=none; sampling=avg 5 trials unless source states otherwise; harness=table8." }, { "model_id": "claude-sonnet-5", "benchmark_id": "healthbench_professional_length_adjusted", "score": 57.8, "reference_url": "https://www.anthropic.com/claude-sonnet-5-system-card", "reported_setting": { "effort": "max", "tools": "none", "sampling": "avg 5 trials unless source states otherwise", "harness": "table8" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Anthropic Claude Sonnet 5 System Card Table 8.1.A, page 115: Claude Sonnet 5; HealthBench Professional [summary]=57.8. Source setting: effort=max; tools=none; sampling=avg 5 trials unless source states otherwise; harness=table8." }, { "model_id": "claude-sonnet-4.6", "benchmark_id": "healthbench_professional_length_adjusted", "score": 44.2, "reference_url": "https://www.anthropic.com/claude-sonnet-5-system-card", "reported_setting": { "effort": "max", "tools": "none", "sampling": "avg 5 trials unless source states otherwise", "harness": "table8" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Anthropic Claude Sonnet 5 System Card Table 8.1.A, page 115: Claude Sonnet 4.6; HealthBench Professional [summary]=44.2. Source setting: effort=max; tools=none; sampling=avg 5 trials unless source states otherwise; harness=table8.", "candidates": [ { "score": 38.0, "reference_url": "https://microsoft.ai/wp-content/uploads/2026/06/main_20260602_2.pdf", "source_type": "official_paper", "reported_setting": { "mode": "thinking", "effort": "maximum reasoning effort", "tools": "none; benchmark-provided tools where required", "sampling": "source does not state trial count", "judge": "benchmark-specific Microsoft implementation", "harness": "Microsoft independent evaluation suite", "prompt_style": "benchmark-specific", "temperature": "source does not state", "context": "maximum sequence length", "notes": "Sonnet 4.6 result generated by Microsoft's own evaluation." }, "notes": "Displayed exactly as 38. Official Microsoft-reported result." } ] }, { "model_id": "gpt-5.5", "benchmark_id": "frontiercode_main_v1", "score": 25.5, "reference_url": "https://www.anthropic.com/claude-sonnet-5-system-card", "reported_setting": { "effort": "xhigh", "tools": "agentic benchmark harness", "sampling": "pass@1", "harness": "frontier" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Anthropic Claude Sonnet 5 System Card Figure 8.4.A, page 118: GPT-5.5; FrontierCode v1 [xhigh]=25.5%. Source setting: effort=xhigh; tools=agentic benchmark harness; sampling=pass@1; harness=frontier.", "candidates": [ { "score": 25.5, "reference_url": "https://www.anthropic.com/claude-sonnet-5-system-card", "source_type": "model_card", "reported_setting": { "effort": "source does not state", "tools": "none", "sampling": "avg 5 trials unless source states otherwise", "harness": "table8" }, "notes": "Anthropic Claude Sonnet 5 System Card Table 8.1.A, page 115: GPT-5.5; FrontierCode v1 [summary]=25.5. Source setting: effort=source does not state; tools=none; sampling=avg 5 trials unless source states otherwise; harness=table8." }, { "score": 21.1, "reference_url": "https://www.anthropic.com/claude-sonnet-5-system-card", "source_type": "model_card", "reported_setting": { "effort": "low", "tools": "agentic benchmark harness", "sampling": "pass@1", "harness": "frontier" }, "notes": "Anthropic Claude Sonnet 5 System Card Figure 8.4.A, page 118: GPT-5.5; FrontierCode v1 [low]=21.1%. Source setting: effort=low; tools=agentic benchmark harness; sampling=pass@1; harness=frontier." }, { "score": 22.4, "reference_url": "https://www.anthropic.com/claude-sonnet-5-system-card", "source_type": "model_card", "reported_setting": { "effort": "medium", "tools": "agentic benchmark harness", "sampling": "pass@1", "harness": "frontier" }, "notes": "Anthropic Claude Sonnet 5 System Card Figure 8.4.A, page 118: GPT-5.5; FrontierCode v1 [med]=22.4%. Source setting: effort=medium; tools=agentic benchmark harness; sampling=pass@1; harness=frontier." }, { "score": 25.3, "reference_url": "https://www.anthropic.com/claude-sonnet-5-system-card", "source_type": "model_card", "reported_setting": { "effort": "high", "tools": "agentic benchmark harness", "sampling": "pass@1", "harness": "frontier" }, "notes": "Anthropic Claude Sonnet 5 System Card Figure 8.4.A, page 118: GPT-5.5; FrontierCode v1 [high]=25.3%. Source setting: effort=high; tools=agentic benchmark harness; sampling=pass@1; harness=frontier." } ] }, { "model_id": "claude-opus-4.8", "benchmark_id": "frontiercode_main_v1", "score": 30.3, "reference_url": "https://www.anthropic.com/claude-sonnet-5-system-card", "reported_setting": { "effort": "high", "tools": "agentic benchmark harness", "sampling": "pass@1", "harness": "frontier" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Anthropic Claude Sonnet 5 System Card Figure 8.4.A, page 118: Claude Opus 4.8; FrontierCode v1 [high]=30.3%. Source setting: effort=high; tools=agentic benchmark harness; sampling=pass@1; harness=frontier.", "candidates": [ { "score": 25.3, "reference_url": "https://www.anthropic.com/claude-sonnet-5-system-card", "source_type": "model_card", "reported_setting": { "effort": "low", "tools": "agentic benchmark harness", "sampling": "pass@1", "harness": "frontier" }, "notes": "Anthropic Claude Sonnet 5 System Card Figure 8.4.A, page 118: Claude Opus 4.8; FrontierCode v1 [low]=25.3%. Source setting: effort=low; tools=agentic benchmark harness; sampling=pass@1; harness=frontier." }, { "score": 26.9, "reference_url": "https://www.anthropic.com/claude-sonnet-5-system-card", "source_type": "model_card", "reported_setting": { "effort": "medium", "tools": "agentic benchmark harness", "sampling": "pass@1", "harness": "frontier" }, "notes": "Anthropic Claude Sonnet 5 System Card Figure 8.4.A, page 118: Claude Opus 4.8; FrontierCode v1 [med]=26.9%. Source setting: effort=medium; tools=agentic benchmark harness; sampling=pass@1; harness=frontier." }, { "score": 34.3, "reference_url": "https://www.anthropic.com/claude-sonnet-5-system-card", "source_type": "model_card", "reported_setting": { "effort": "xhigh", "tools": "agentic benchmark harness", "sampling": "pass@1", "harness": "frontier" }, "notes": "Anthropic Claude Sonnet 5 System Card Figure 8.4.A, page 118: Claude Opus 4.8; FrontierCode v1 [xhigh]=34.3%. Source setting: effort=xhigh; tools=agentic benchmark harness; sampling=pass@1; harness=frontier." }, { "score": 31.3, "reference_url": "https://www.anthropic.com/claude-sonnet-5-system-card", "source_type": "model_card", "reported_setting": { "effort": "max", "tools": "agentic benchmark harness", "sampling": "pass@1", "harness": "frontier" }, "notes": "Anthropic Claude Sonnet 5 System Card Figure 8.4.A, page 118: Claude Opus 4.8; FrontierCode v1 [max]=31.3%. Source setting: effort=max; tools=agentic benchmark harness; sampling=pass@1; harness=frontier." } ] }, { "model_id": "claude-sonnet-5", "benchmark_id": "frontiercode_main_v1", "score": 28.9, "reference_url": "https://www.anthropic.com/claude-sonnet-5-system-card", "reported_setting": { "effort": "high", "tools": "agentic benchmark harness", "sampling": "pass@1", "harness": "frontier" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Anthropic Claude Sonnet 5 System Card Figure 8.4.A, page 118: Claude Sonnet 5; FrontierCode v1 [high]=28.9%. Source setting: effort=high; tools=agentic benchmark harness; sampling=pass@1; harness=frontier.", "candidates": [ { "score": 38.8, "reference_url": "https://www.anthropic.com/claude-sonnet-5-system-card", "source_type": "model_card", "reported_setting": { "effort": "max", "tools": "none", "sampling": "avg 5 trials unless source states otherwise", "harness": "table8" }, "notes": "Anthropic Claude Sonnet 5 System Card Table 8.1.A, page 115: Claude Sonnet 5; FrontierCode v1 [summary]=38.8. Source setting: effort=max; tools=none; sampling=avg 5 trials unless source states otherwise; harness=table8." }, { "score": 18.1, "reference_url": "https://www.anthropic.com/claude-sonnet-5-system-card", "source_type": "model_card", "reported_setting": { "effort": "low", "tools": "agentic benchmark harness", "sampling": "pass@1", "harness": "frontier" }, "notes": "Anthropic Claude Sonnet 5 System Card Figure 8.4.A, page 118: Claude Sonnet 5; FrontierCode v1 [low]=18.1%. Source setting: effort=low; tools=agentic benchmark harness; sampling=pass@1; harness=frontier." }, { "score": 26.6, "reference_url": "https://www.anthropic.com/claude-sonnet-5-system-card", "source_type": "model_card", "reported_setting": { "effort": "medium", "tools": "agentic benchmark harness", "sampling": "pass@1", "harness": "frontier" }, "notes": "Anthropic Claude Sonnet 5 System Card Figure 8.4.A, page 118: Claude Sonnet 5; FrontierCode v1 [med]=26.6%. Source setting: effort=medium; tools=agentic benchmark harness; sampling=pass@1; harness=frontier." }, { "score": 34.0, "reference_url": "https://www.anthropic.com/claude-sonnet-5-system-card", "source_type": "model_card", "reported_setting": { "effort": "xhigh", "tools": "agentic benchmark harness", "sampling": "pass@1", "harness": "frontier" }, "notes": "Anthropic Claude Sonnet 5 System Card Figure 8.4.A, page 118: Claude Sonnet 5; FrontierCode v1 [xhigh]=34.0%. Source setting: effort=xhigh; tools=agentic benchmark harness; sampling=pass@1; harness=frontier." }, { "score": 38.8, "reference_url": "https://www.anthropic.com/claude-sonnet-5-system-card", "source_type": "model_card", "reported_setting": { "effort": "max", "tools": "agentic benchmark harness", "sampling": "pass@1", "harness": "frontier" }, "notes": "Anthropic Claude Sonnet 5 System Card Figure 8.4.A, page 118: Claude Sonnet 5; FrontierCode v1 [max]=38.8%. Source setting: effort=max; tools=agentic benchmark harness; sampling=pass@1; harness=frontier." } ] }, { "model_id": "claude-fable-5", "benchmark_id": "frontiercode_main_v1", "score": 44.7, "reference_url": "https://www.anthropic.com/claude-sonnet-5-system-card", "reported_setting": { "effort": "max", "tools": "agentic benchmark harness", "sampling": "pass@1", "harness": "frontier" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Anthropic Claude Sonnet 5 System Card Figure 8.4.A, page 118: Claude Fable 5; FrontierCode v1 [max]=44.7%. Source setting: effort=max; tools=agentic benchmark harness; sampling=pass@1; harness=frontier.", "candidates": [ { "score": 37.3, "reference_url": "https://www.anthropic.com/claude-sonnet-5-system-card", "source_type": "model_card", "reported_setting": { "effort": "low", "tools": "agentic benchmark harness", "sampling": "pass@1", "harness": "frontier" }, "notes": "Anthropic Claude Sonnet 5 System Card Figure 8.4.A, page 118: Claude Fable 5; FrontierCode v1 [low]=37.3%. Source setting: effort=low; tools=agentic benchmark harness; sampling=pass@1; harness=frontier." }, { "score": 41.1, "reference_url": "https://www.anthropic.com/claude-sonnet-5-system-card", "source_type": "model_card", "reported_setting": { "effort": "medium", "tools": "agentic benchmark harness", "sampling": "pass@1", "harness": "frontier" }, "notes": "Anthropic Claude Sonnet 5 System Card Figure 8.4.A, page 118: Claude Fable 5; FrontierCode v1 [med]=41.1%. Source setting: effort=medium; tools=agentic benchmark harness; sampling=pass@1; harness=frontier." }, { "score": 42.9, "reference_url": "https://www.anthropic.com/claude-sonnet-5-system-card", "source_type": "model_card", "reported_setting": { "effort": "high", "tools": "agentic benchmark harness", "sampling": "pass@1", "harness": "frontier" }, "notes": "Anthropic Claude Sonnet 5 System Card Figure 8.4.A, page 118: Claude Fable 5; FrontierCode v1 [high]=42.9%. Source setting: effort=high; tools=agentic benchmark harness; sampling=pass@1; harness=frontier." }, { "score": 46.3, "reference_url": "https://www.anthropic.com/claude-sonnet-5-system-card", "source_type": "model_card", "reported_setting": { "effort": "xhigh", "tools": "agentic benchmark harness", "sampling": "pass@1", "harness": "frontier" }, "notes": "Anthropic Claude Sonnet 5 System Card Figure 8.4.A, page 118: Claude Fable 5; FrontierCode v1 [xhigh]=46.3%. Source setting: effort=xhigh; tools=agentic benchmark harness; sampling=pass@1; harness=frontier." } ] }, { "model_id": "gpt-5.5", "benchmark_id": "cursorbench_3_1", "score": 64.3, "reference_url": "https://www.anthropic.com/claude-sonnet-5-system-card", "reported_setting": { "effort": "xhigh", "tools": "agentic benchmark harness", "sampling": "pass@1", "harness": "cursor" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Anthropic Claude Sonnet 5 System Card Figure 8.5.A, page 119: GPT-5.5; CursorBench [xhigh]=64.3%. Source setting: effort=xhigh; tools=agentic benchmark harness; sampling=pass@1; harness=cursor.", "candidates": [ { "score": 48.8, "reference_url": "https://www.anthropic.com/claude-sonnet-5-system-card", "source_type": "model_card", "reported_setting": { "effort": "low", "tools": "agentic benchmark harness", "sampling": "pass@1", "harness": "cursor" }, "notes": "Anthropic Claude Sonnet 5 System Card Figure 8.5.A, page 119: GPT-5.5; CursorBench [low]=48.8%. Source setting: effort=low; tools=agentic benchmark harness; sampling=pass@1; harness=cursor." }, { "score": 59.2, "reference_url": "https://www.anthropic.com/claude-sonnet-5-system-card", "source_type": "model_card", "reported_setting": { "effort": "medium", "tools": "agentic benchmark harness", "sampling": "pass@1", "harness": "cursor" }, "notes": "Anthropic Claude Sonnet 5 System Card Figure 8.5.A, page 119: GPT-5.5; CursorBench [med]=59.2%. Source setting: effort=medium; tools=agentic benchmark harness; sampling=pass@1; harness=cursor." }, { "score": 62.6, "reference_url": "https://www.anthropic.com/claude-sonnet-5-system-card", "source_type": "model_card", "reported_setting": { "effort": "high", "tools": "agentic benchmark harness", "sampling": "pass@1", "harness": "cursor" }, "notes": "Anthropic Claude Sonnet 5 System Card Figure 8.5.A, page 119: GPT-5.5; CursorBench [high]=62.6%. Source setting: effort=high; tools=agentic benchmark harness; sampling=pass@1; harness=cursor." } ] }, { "model_id": "claude-sonnet-4.6", "benchmark_id": "cursorbench_3_1", "score": 47.5, "reference_url": "https://www.anthropic.com/claude-sonnet-5-system-card", "reported_setting": { "effort": "max", "tools": "agentic benchmark harness", "sampling": "pass@1", "harness": "cursor" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Anthropic Claude Sonnet 5 System Card Figure 8.5.A, page 119: Claude Sonnet 4.6; CursorBench [max]=47.5%. Source setting: effort=max; tools=agentic benchmark harness; sampling=pass@1; harness=cursor.", "candidates": [ { "score": 41.6, "reference_url": "https://www.anthropic.com/claude-sonnet-5-system-card", "source_type": "model_card", "reported_setting": { "effort": "low", "tools": "agentic benchmark harness", "sampling": "pass@1", "harness": "cursor" }, "notes": "Anthropic Claude Sonnet 5 System Card Figure 8.5.A, page 119: Claude Sonnet 4.6; CursorBench [low]=41.6%. Source setting: effort=low; tools=agentic benchmark harness; sampling=pass@1; harness=cursor." }, { "score": 46.0, "reference_url": "https://www.anthropic.com/claude-sonnet-5-system-card", "source_type": "model_card", "reported_setting": { "effort": "medium", "tools": "agentic benchmark harness", "sampling": "pass@1", "harness": "cursor" }, "notes": "Anthropic Claude Sonnet 5 System Card Figure 8.5.A, page 119: Claude Sonnet 4.6; CursorBench [med]=46.0%. Source setting: effort=medium; tools=agentic benchmark harness; sampling=pass@1; harness=cursor." }, { "score": 48.2, "reference_url": "https://www.anthropic.com/claude-sonnet-5-system-card", "source_type": "model_card", "reported_setting": { "effort": "high", "tools": "agentic benchmark harness", "sampling": "pass@1", "harness": "cursor" }, "notes": "Anthropic Claude Sonnet 5 System Card Figure 8.5.A, page 119: Claude Sonnet 4.6; CursorBench [high]=48.2%. Source setting: effort=high; tools=agentic benchmark harness; sampling=pass@1; harness=cursor." } ] }, { "model_id": "claude-opus-4.7", "benchmark_id": "cursorbench_3_1", "score": 61.6, "reference_url": "https://cursor.com/blog/composer-2-5", "reported_setting": { "effort": "xhigh", "tools": "agentic", "sampling": "pass@1", "harness": "Cursor", "source_default": true }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "Composer 2.5 release benchmark image / CursorBench 3.1 / Opus 4.7 / xhigh: Cursor release benchmark image. Opus 4.7 and GPT-5.5 public evaluation values are self-reported.", "candidates": [ { "score": 52.7, "reference_url": "https://www.anthropic.com/claude-sonnet-5-system-card", "source_type": "model_card", "reported_setting": { "effort": "medium", "tools": "agentic benchmark harness", "sampling": "pass@1", "harness": "cursor" }, "notes": "Anthropic Claude Sonnet 5 System Card Figure 8.5.A, page 119: Claude Opus 4.7; CursorBench [med]=52.7%. Source setting: effort=medium; tools=agentic benchmark harness; sampling=pass@1; harness=cursor." }, { "score": 59.4, "reference_url": "https://www.anthropic.com/claude-sonnet-5-system-card", "source_type": "model_card", "reported_setting": { "effort": "high", "tools": "agentic benchmark harness", "sampling": "pass@1", "harness": "cursor" }, "notes": "Anthropic Claude Sonnet 5 System Card Figure 8.5.A, page 119: Claude Opus 4.7; CursorBench [high]=59.4%. Source setting: effort=high; tools=agentic benchmark harness; sampling=pass@1; harness=cursor." }, { "score": 64.8, "reference_url": "https://www.anthropic.com/claude-sonnet-5-system-card", "source_type": "model_card", "reported_setting": { "effort": "max", "tools": "agentic benchmark harness", "sampling": "pass@1", "harness": "cursor" }, "notes": "Anthropic Claude Sonnet 5 System Card Figure 8.5.A, page 119: Claude Opus 4.7; CursorBench [max]=64.8%. Source setting: effort=max; tools=agentic benchmark harness; sampling=pass@1; harness=cursor." }, { "score": 48.3, "reference_url": "https://www.anthropic.com/claude-sonnet-5-system-card", "source_type": "model_card", "reported_setting": { "effort": "low", "tools": "agentic benchmark harness", "sampling": "pass@1", "harness": "cursor" }, "notes": "Anthropic Claude Sonnet 5 System Card Figure 8.5.A, page 119: Claude Opus 4.7; CursorBench [low]=48.3%. Source setting: effort=low; tools=agentic benchmark harness; sampling=pass@1; harness=cursor." } ] }, { "model_id": "claude-opus-4.8", "benchmark_id": "cursorbench_3_1", "score": 58.4, "reference_url": "https://www.anthropic.com/claude-sonnet-5-system-card", "reported_setting": { "effort": "high", "tools": "agentic benchmark harness", "sampling": "pass@1", "harness": "cursor" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Anthropic Claude Sonnet 5 System Card Figure 8.5.A, page 119: Claude Opus 4.8; CursorBench [high]=58.4%. Source setting: effort=high; tools=agentic benchmark harness; sampling=pass@1; harness=cursor.", "candidates": [ { "score": 54.3, "reference_url": "https://www.anthropic.com/claude-sonnet-5-system-card", "source_type": "model_card", "reported_setting": { "effort": "low", "tools": "agentic benchmark harness", "sampling": "pass@1", "harness": "cursor" }, "notes": "Anthropic Claude Sonnet 5 System Card Figure 8.5.A, page 119: Claude Opus 4.8; CursorBench [low]=54.3%. Source setting: effort=low; tools=agentic benchmark harness; sampling=pass@1; harness=cursor." }, { "score": 56.6, "reference_url": "https://www.anthropic.com/claude-sonnet-5-system-card", "source_type": "model_card", "reported_setting": { "effort": "medium", "tools": "agentic benchmark harness", "sampling": "pass@1", "harness": "cursor" }, "notes": "Anthropic Claude Sonnet 5 System Card Figure 8.5.A, page 119: Claude Opus 4.8; CursorBench [med]=56.6%. Source setting: effort=medium; tools=agentic benchmark harness; sampling=pass@1; harness=cursor." }, { "score": 62.1, "reference_url": "https://www.anthropic.com/claude-sonnet-5-system-card", "source_type": "model_card", "reported_setting": { "effort": "xhigh", "tools": "agentic benchmark harness", "sampling": "pass@1", "harness": "cursor" }, "notes": "Anthropic Claude Sonnet 5 System Card Figure 8.5.A, page 119: Claude Opus 4.8; CursorBench [xhigh]=62.1%. Source setting: effort=xhigh; tools=agentic benchmark harness; sampling=pass@1; harness=cursor." }, { "score": 63.8, "reference_url": "https://www.anthropic.com/claude-sonnet-5-system-card", "source_type": "model_card", "reported_setting": { "effort": "max", "tools": "agentic benchmark harness", "sampling": "pass@1", "harness": "cursor" }, "notes": "Anthropic Claude Sonnet 5 System Card Figure 8.5.A, page 119: Claude Opus 4.8; CursorBench [max]=63.8%. Source setting: effort=max; tools=agentic benchmark harness; sampling=pass@1; harness=cursor." } ] }, { "model_id": "claude-sonnet-5", "benchmark_id": "cursorbench_3_1", "score": 57.0, "reference_url": "https://www.anthropic.com/claude-sonnet-5-system-card", "reported_setting": { "effort": "high", "tools": "agentic benchmark harness", "sampling": "pass@1", "harness": "cursor" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Anthropic Claude Sonnet 5 System Card Figure 8.5.A, page 119: Claude Sonnet 5; CursorBench [high]=57.0%. Source setting: effort=high; tools=agentic benchmark harness; sampling=pass@1; harness=cursor.", "candidates": [ { "score": 47.7, "reference_url": "https://www.anthropic.com/claude-sonnet-5-system-card", "source_type": "model_card", "reported_setting": { "effort": "low", "tools": "agentic benchmark harness", "sampling": "pass@1", "harness": "cursor" }, "notes": "Anthropic Claude Sonnet 5 System Card Figure 8.5.A, page 119: Claude Sonnet 5; CursorBench [low]=47.7%. Source setting: effort=low; tools=agentic benchmark harness; sampling=pass@1; harness=cursor." }, { "score": 54.9, "reference_url": "https://www.anthropic.com/claude-sonnet-5-system-card", "source_type": "model_card", "reported_setting": { "effort": "medium", "tools": "agentic benchmark harness", "sampling": "pass@1", "harness": "cursor" }, "notes": "Anthropic Claude Sonnet 5 System Card Figure 8.5.A, page 119: Claude Sonnet 5; CursorBench [med]=54.9%. Source setting: effort=medium; tools=agentic benchmark harness; sampling=pass@1; harness=cursor." }, { "score": 59.0, "reference_url": "https://www.anthropic.com/claude-sonnet-5-system-card", "source_type": "model_card", "reported_setting": { "effort": "xhigh", "tools": "agentic benchmark harness", "sampling": "pass@1", "harness": "cursor" }, "notes": "Anthropic Claude Sonnet 5 System Card Figure 8.5.A, page 119: Claude Sonnet 5; CursorBench [xhigh]=59.0%. Source setting: effort=xhigh; tools=agentic benchmark harness; sampling=pass@1; harness=cursor." }, { "score": 61.2, "reference_url": "https://www.anthropic.com/claude-sonnet-5-system-card", "source_type": "model_card", "reported_setting": { "effort": "max", "tools": "agentic benchmark harness", "sampling": "pass@1", "harness": "cursor" }, "notes": "Anthropic Claude Sonnet 5 System Card Figure 8.5.A, page 119: Claude Sonnet 5; CursorBench [max]=61.2%. Source setting: effort=max; tools=agentic benchmark harness; sampling=pass@1; harness=cursor." } ] }, { "model_id": "claude-fable-5", "benchmark_id": "cursorbench_3_1", "score": 72.9, "reference_url": "https://www.anthropic.com/claude-sonnet-5-system-card", "reported_setting": { "effort": "max", "tools": "agentic benchmark harness", "sampling": "pass@1", "harness": "cursor" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Anthropic Claude Sonnet 5 System Card Figure 8.5.A, page 119: Claude Fable 5; CursorBench [max]=72.9%. Source setting: effort=max; tools=agentic benchmark harness; sampling=pass@1; harness=cursor.", "candidates": [ { "score": 64.2, "reference_url": "https://www.anthropic.com/claude-sonnet-5-system-card", "source_type": "model_card", "reported_setting": { "effort": "low", "tools": "agentic benchmark harness", "sampling": "pass@1", "harness": "cursor" }, "notes": "Anthropic Claude Sonnet 5 System Card Figure 8.5.A, page 119: Claude Fable 5; CursorBench [low]=64.2%. Source setting: effort=low; tools=agentic benchmark harness; sampling=pass@1; harness=cursor." }, { "score": 69.8, "reference_url": "https://www.anthropic.com/claude-sonnet-5-system-card", "source_type": "model_card", "reported_setting": { "effort": "medium", "tools": "agentic benchmark harness", "sampling": "pass@1", "harness": "cursor" }, "notes": "Anthropic Claude Sonnet 5 System Card Figure 8.5.A, page 119: Claude Fable 5; CursorBench [med]=69.8%. Source setting: effort=medium; tools=agentic benchmark harness; sampling=pass@1; harness=cursor." }, { "score": 70.6, "reference_url": "https://www.anthropic.com/claude-sonnet-5-system-card", "source_type": "model_card", "reported_setting": { "effort": "high", "tools": "agentic benchmark harness", "sampling": "pass@1", "harness": "cursor" }, "notes": "Anthropic Claude Sonnet 5 System Card Figure 8.5.A, page 119: Claude Fable 5; CursorBench [high]=70.6%. Source setting: effort=high; tools=agentic benchmark harness; sampling=pass@1; harness=cursor." }, { "score": 72.0, "reference_url": "https://www.anthropic.com/claude-sonnet-5-system-card", "source_type": "model_card", "reported_setting": { "effort": "xhigh", "tools": "agentic benchmark harness", "sampling": "pass@1", "harness": "cursor" }, "notes": "Anthropic Claude Sonnet 5 System Card Figure 8.5.A, page 119: Claude Fable 5; CursorBench [xhigh]=72.0%. Source setting: effort=xhigh; tools=agentic benchmark harness; sampling=pass@1; harness=cursor." } ] }, { "model_id": "gemini-3.5-flash", "benchmark_id": "arxivmath_2026_04_05", "score": 47.7, "reference_url": "https://www.anthropic.com/claude-sonnet-5-system-card", "reported_setting": { "effort": "source does not state", "tools": "none", "sampling": "4 attempts/problem", "harness": "arxiv" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Anthropic Claude Sonnet 5 System Card Figure 8.7.A, page 121: Gemini 3.5 Flash; ArxivMath / April+May 2026 [summary]=47.7. Source setting: effort=source does not state; tools=none; sampling=4 attempts/problem; harness=arxiv." }, { "model_id": "gpt-5.5", "benchmark_id": "arxivmath_2026_04_05", "score": 72.2, "reference_url": "https://www.anthropic.com/claude-sonnet-5-system-card", "reported_setting": { "effort": "source does not state", "tools": "none", "sampling": "4 attempts/problem", "harness": "arxiv" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Anthropic Claude Sonnet 5 System Card Figure 8.7.A, page 121: GPT-5.5; ArxivMath / April+May 2026 [summary]=72.2. Source setting: effort=source does not state; tools=none; sampling=4 attempts/problem; harness=arxiv." }, { "model_id": "claude-opus-4.8", "benchmark_id": "arxivmath_2026_04_05", "score": 71.0, "reference_url": "https://www.anthropic.com/claude-sonnet-5-system-card", "reported_setting": { "effort": "max", "tools": "none", "sampling": "4 attempts/problem", "harness": "arxiv" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Anthropic Claude Sonnet 5 System Card Figure 8.7.A, page 121: Claude Opus 4.8; ArxivMath / April+May 2026 [summary]=71.0. Source setting: effort=max; tools=none; sampling=4 attempts/problem; harness=arxiv." }, { "model_id": "claude-fable-5", "benchmark_id": "arxivmath_2026_04_05", "score": 78.6, "reference_url": "https://www.anthropic.com/claude-sonnet-5-system-card", "reported_setting": { "effort": "max", "tools": "none", "sampling": "4 attempts/problem", "harness": "arxiv" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Anthropic Claude Sonnet 5 System Card Figure 8.7.A, page 121: Claude Fable 5; ArxivMath / April+May 2026 [summary]=78.6. Source setting: effort=max; tools=none; sampling=4 attempts/problem; harness=arxiv." }, { "model_id": "claude-sonnet-5", "benchmark_id": "arxivmath_2026_04_05", "score": 65.7, "reference_url": "https://www.anthropic.com/claude-sonnet-5-system-card", "reported_setting": { "effort": "max", "tools": "none", "sampling": "4 attempts/problem", "harness": "arxiv" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Anthropic Claude Sonnet 5 System Card Figure 8.7.A, page 121: Claude Sonnet 5; ArxivMath / April+May 2026 [without tools]=65.7. Source setting: effort=max; tools=none; sampling=4 attempts/problem; harness=arxiv.", "candidates": [ { "score": 72.2, "reference_url": "https://www.anthropic.com/claude-sonnet-5-system-card", "source_type": "model_card", "reported_setting": { "effort": "max", "tools": "python / benchmark tools", "sampling": "4 attempts/problem", "harness": "arxiv" }, "notes": "Anthropic Claude Sonnet 5 System Card Figure 8.7.A, page 121: Claude Sonnet 5; ArxivMath / April+May 2026 [with tools]=72.2. Source setting: effort=max; tools=python / benchmark tools; sampling=4 attempts/problem; harness=arxiv." } ] }, { "model_id": "claude-opus-4.6", "benchmark_id": "hle_tools", "score": 46.8, "reference_url": "https://www.anthropic.com/claude-sonnet-5-system-card", "reported_setting": { "effort": "max", "tools": "python / benchmark tools", "sampling": "pass@1", "harness": "hle" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Anthropic Claude Sonnet 5 System Card Figure 8.9.1.A, page 122: Claude Opus 4.6; Humanity’s Last Exam [with tools]=46.8%. Source setting: effort=max; tools=python / benchmark tools; sampling=pass@1; harness=hle.", "candidates": [ { "score": 53.1, "reference_url": "https://deepmind.google/models/gemini/deep-think/", "source_type": "official_blog", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "search + code execution", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official / provider-reported", "temperature": "default", "snapshot": "Feb 2026" }, "notes": "Google February table reports provider-sourced Opus 4.6 Max with search+code on the full HLE set. Existing Anthropic-sourced primary remains preferred; retain 53.1 as an alternate." } ] }, { "model_id": "claude-sonnet-5", "benchmark_id": "hle_tools", "score": 52.8, "reference_url": "https://www.anthropic.com/claude-sonnet-5-system-card", "reported_setting": { "effort": "high", "tools": "web search, fetch, and code execution", "sampling": "pass@1", "harness": "hle" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Anthropic Claude Sonnet 5 System Card Figure 8.9.1.B, page 123: Claude Sonnet 5; Humanity’s Last Exam [high]=52.8%. Source setting: effort=high; tools=web search, fetch, and code execution; sampling=pass@1; harness=hle.", "candidates": [ { "score": 57.4, "reference_url": "https://www.anthropic.com/claude-sonnet-5-system-card", "source_type": "model_card", "reported_setting": { "effort": "max", "tools": "web search, fetch, and code execution", "sampling": "avg 5 trials unless source states otherwise", "harness": "table8" }, "notes": "Anthropic Claude Sonnet 5 System Card Table 8.1.A, page 115: Claude Sonnet 5; Humanity’s Last Exam / with tools [summary]=57.4. Source setting: effort=max; tools=web search, fetch, and code execution; sampling=avg 5 trials unless source states otherwise; harness=table8." }, { "score": 57.4, "reference_url": "https://www.anthropic.com/claude-sonnet-5-system-card", "source_type": "model_card", "reported_setting": { "effort": "max", "tools": "python / benchmark tools", "sampling": "pass@1", "harness": "hle" }, "notes": "Anthropic Claude Sonnet 5 System Card Figure 8.9.1.A, page 122: Claude Sonnet 5; Humanity’s Last Exam [with tools]=57.4%. Source setting: effort=max; tools=python / benchmark tools; sampling=pass@1; harness=hle." }, { "score": 36.5, "reference_url": "https://www.anthropic.com/claude-sonnet-5-system-card", "source_type": "model_card", "reported_setting": { "effort": "low", "tools": "web search, fetch, and code execution", "sampling": "pass@1", "harness": "hle" }, "notes": "Anthropic Claude Sonnet 5 System Card Figure 8.9.1.B, page 123: Claude Sonnet 5; Humanity’s Last Exam [low]=36.5%. Source setting: effort=low; tools=web search, fetch, and code execution; sampling=pass@1; harness=hle." }, { "score": 47.2, "reference_url": "https://www.anthropic.com/claude-sonnet-5-system-card", "source_type": "model_card", "reported_setting": { "effort": "medium", "tools": "web search, fetch, and code execution", "sampling": "pass@1", "harness": "hle" }, "notes": "Anthropic Claude Sonnet 5 System Card Figure 8.9.1.B, page 123: Claude Sonnet 5; Humanity’s Last Exam [med]=47.2%. Source setting: effort=medium; tools=web search, fetch, and code execution; sampling=pass@1; harness=hle." }, { "score": 54.6, "reference_url": "https://www.anthropic.com/claude-sonnet-5-system-card", "source_type": "model_card", "reported_setting": { "effort": "xhigh", "tools": "web search, fetch, and code execution", "sampling": "pass@1", "harness": "hle" }, "notes": "Anthropic Claude Sonnet 5 System Card Figure 8.9.1.B, page 123: Claude Sonnet 5; Humanity’s Last Exam [xhigh]=54.6%. Source setting: effort=xhigh; tools=web search, fetch, and code execution; sampling=pass@1; harness=hle." }, { "score": 36.5, "reference_url": "https://www.anthropic.com/claude-opus-5-system-card", "source_type": "model_card", "reported_setting": { "effort": "low", "tools": "benchmark-specific" }, "notes": "Figure 8.10.1.A / Claude Sonnet 5 / low: 2,500 questions; 1M token cap; web/programmatic/code tools; no context compaction." }, { "score": 47.2, "reference_url": "https://www.anthropic.com/claude-opus-5-system-card", "source_type": "model_card", "reported_setting": { "effort": "medium", "tools": "benchmark-specific" }, "notes": "Figure 8.10.1.A / Claude Sonnet 5 / medium: 2,500 questions; 1M token cap; web/programmatic/code tools; no context compaction." }, { "score": 54.6, "reference_url": "https://www.anthropic.com/claude-opus-5-system-card", "source_type": "model_card", "reported_setting": { "effort": "xhigh", "tools": "benchmark-specific" }, "notes": "Figure 8.10.1.A / Claude Sonnet 5 / xhigh: 2,500 questions; 1M token cap; web/programmatic/code tools; no context compaction." }, { "score": 57.4, "reference_url": "https://www.anthropic.com/claude-opus-5-system-card", "source_type": "model_card", "reported_setting": { "effort": "max", "tools": "benchmark-specific" }, "notes": "Figure 8.10.1.A / Claude Sonnet 5 / max: 2,500 questions; 1M token cap; web/programmatic/code tools; no context compaction." } ] }, { "model_id": "claude-sonnet-4.6", "benchmark_id": "gdp_pdf_mean_criteria_pass_rate", "score": 66.9, "reference_url": "https://www.anthropic.com/claude-sonnet-5-system-card", "reported_setting": { "effort": "max", "tools": "none", "sampling": "avg 5 trials unless source states otherwise", "harness": "gdp" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Anthropic Claude Sonnet 5 System Card Figure 8.10.1.A, page 125: Claude Sonnet 4.6; GDP.pdf [no tools]=66.9. Source setting: effort=max; tools=none; sampling=avg 5 trials unless source states otherwise; harness=gdp.", "candidates": [ { "score": 78.6, "reference_url": "https://www.anthropic.com/claude-sonnet-5-system-card", "source_type": "model_card", "reported_setting": { "effort": "max", "tools": "python / benchmark tools", "sampling": "avg 5 trials unless source states otherwise", "harness": "gdp" }, "notes": "Anthropic Claude Sonnet 5 System Card Figure 8.10.1.A, page 125: Claude Sonnet 4.6; GDP.pdf [Python tools]=78.6. Source setting: effort=max; tools=python / benchmark tools; sampling=avg 5 trials unless source states otherwise; harness=gdp." } ] }, { "model_id": "claude-mythos", "benchmark_id": "gdp_pdf_mean_criteria_pass_rate", "score": 70.6, "reference_url": "https://www.anthropic.com/claude-sonnet-5-system-card", "reported_setting": { "effort": "max", "tools": "none", "sampling": "avg 5 trials unless source states otherwise", "harness": "gdp" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Anthropic Claude Sonnet 5 System Card Figure 8.10.1.A, page 125: Claude Mythos Preview; GDP.pdf [no tools]=70.6. Source setting: effort=max; tools=none; sampling=avg 5 trials unless source states otherwise; harness=gdp.", "candidates": [ { "score": 85.6, "reference_url": "https://www.anthropic.com/claude-sonnet-5-system-card", "source_type": "model_card", "reported_setting": { "effort": "max", "tools": "python / benchmark tools", "sampling": "avg 5 trials unless source states otherwise", "harness": "gdp" }, "notes": "Anthropic Claude Sonnet 5 System Card Figure 8.10.1.A, page 125: Claude Mythos Preview; GDP.pdf [Python tools]=85.6. Source setting: effort=max; tools=python / benchmark tools; sampling=avg 5 trials unless source states otherwise; harness=gdp." } ] }, { "model_id": "claude-opus-4.8", "benchmark_id": "gdp_pdf_mean_criteria_pass_rate", "score": 71.2, "reference_url": "https://www.anthropic.com/claude-sonnet-5-system-card", "reported_setting": { "effort": "max", "tools": "none", "sampling": "avg 5 trials unless source states otherwise", "harness": "gdp" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Anthropic Claude Sonnet 5 System Card Figure 8.10.1.A, page 125: Claude Opus 4.8; GDP.pdf [no tools]=71.2. Source setting: effort=max; tools=none; sampling=avg 5 trials unless source states otherwise; harness=gdp.", "candidates": [ { "score": 85.7, "reference_url": "https://www.anthropic.com/claude-sonnet-5-system-card", "source_type": "model_card", "reported_setting": { "effort": "max", "tools": "python / benchmark tools", "sampling": "avg 5 trials unless source states otherwise", "harness": "gdp" }, "notes": "Anthropic Claude Sonnet 5 System Card Figure 8.10.1.A, page 125: Claude Opus 4.8; GDP.pdf [Python tools]=85.7. Source setting: effort=max; tools=python / benchmark tools; sampling=avg 5 trials unless source states otherwise; harness=gdp." }, { "score": 77.5, "reference_url": "https://www.anthropic.com/claude-opus-5-system-card", "source_type": "model_card", "reported_setting": { "effort": "max", "tools": "none", "harness": "corrected full 100-prompt run" }, "notes": "Figure 8.12.4.A narrative / Claude Opus 4.8 / no tools: 100 prompts, 10 domains, corrected PDF truncation bug, Opus 4.7 judge." }, { "score": 84.8, "reference_url": "https://www.anthropic.com/claude-opus-5-system-card", "source_type": "model_card", "reported_setting": { "effort": "max", "tools": "python / benchmark tools", "harness": "corrected full 100-prompt run" }, "notes": "Figure 8.12.4.A narrative / Claude Opus 4.8 / tools: 100 prompts, 10 domains, corrected PDF truncation bug, Opus 4.7 judge." } ] }, { "model_id": "claude-sonnet-5", "benchmark_id": "gdp_pdf_mean_criteria_pass_rate", "score": 67.5, "reference_url": "https://www.anthropic.com/claude-sonnet-5-system-card", "reported_setting": { "effort": "max", "tools": "none", "sampling": "avg 5 trials unless source states otherwise", "harness": "gdp" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Anthropic Claude Sonnet 5 System Card Figure 8.10.1.A, page 125: Claude Sonnet 5; GDP.pdf [no tools]=67.5. Source setting: effort=max; tools=none; sampling=avg 5 trials unless source states otherwise; harness=gdp.", "candidates": [ { "score": 81.6, "reference_url": "https://www.anthropic.com/claude-sonnet-5-system-card", "source_type": "model_card", "reported_setting": { "effort": "max", "tools": "python / benchmark tools", "sampling": "avg 5 trials unless source states otherwise", "harness": "gdp" }, "notes": "Anthropic Claude Sonnet 5 System Card Figure 8.10.1.A, page 125: Claude Sonnet 5; GDP.pdf [Python tools]=81.6. Source setting: effort=max; tools=python / benchmark tools; sampling=avg 5 trials unless source states otherwise; harness=gdp." } ] }, { "model_id": "claude-sonnet-4.6", "benchmark_id": "chartmuseum", "score": 59.3, "reference_url": "https://www.anthropic.com/claude-sonnet-5-system-card", "reported_setting": { "effort": "max", "tools": "none", "sampling": "avg 5 trials unless source states otherwise", "harness": "vision" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Anthropic Claude Sonnet 5 System Card Figure 8.10.4.A, page 129: Claude Sonnet 4.6; ChartMuseum [no tools]=59.3. Source setting: effort=max; tools=none; sampling=avg 5 trials unless source states otherwise; harness=vision.", "candidates": [ { "score": 80.9, "reference_url": "https://www.anthropic.com/claude-sonnet-5-system-card", "source_type": "model_card", "reported_setting": { "effort": "max", "tools": "python / benchmark tools", "sampling": "avg 5 trials unless source states otherwise", "harness": "vision" }, "notes": "Anthropic Claude Sonnet 5 System Card Figure 8.10.4.A, page 129: Claude Sonnet 4.6; ChartMuseum [Python tools]=80.9. Source setting: effort=max; tools=python / benchmark tools; sampling=avg 5 trials unless source states otherwise; harness=vision." } ] }, { "model_id": "claude-mythos", "benchmark_id": "chartmuseum", "score": 80.7, "reference_url": "https://www.anthropic.com/claude-sonnet-5-system-card", "reported_setting": { "effort": "max", "tools": "none", "sampling": "avg 5 trials unless source states otherwise", "harness": "vision" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Anthropic Claude Sonnet 5 System Card Figure 8.10.4.A, page 129: Claude Mythos Preview; ChartMuseum [no tools]=80.7. Source setting: effort=max; tools=none; sampling=avg 5 trials unless source states otherwise; harness=vision.", "candidates": [ { "score": 92.2, "reference_url": "https://www.anthropic.com/claude-sonnet-5-system-card", "source_type": "model_card", "reported_setting": { "effort": "max", "tools": "python / benchmark tools", "sampling": "avg 5 trials unless source states otherwise", "harness": "vision" }, "notes": "Anthropic Claude Sonnet 5 System Card Figure 8.10.4.A, page 129: Claude Mythos Preview; ChartMuseum [Python tools]=92.2. Source setting: effort=max; tools=python / benchmark tools; sampling=avg 5 trials unless source states otherwise; harness=vision." } ] }, { "model_id": "claude-opus-4.8", "benchmark_id": "chartmuseum", "score": 75.8, "reference_url": "https://www.anthropic.com/claude-sonnet-5-system-card", "reported_setting": { "effort": "max", "tools": "none", "sampling": "avg 5 trials unless source states otherwise", "harness": "vision" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Anthropic Claude Sonnet 5 System Card Figure 8.10.4.A, page 129: Claude Opus 4.8; ChartMuseum [no tools]=75.8. Source setting: effort=max; tools=none; sampling=avg 5 trials unless source states otherwise; harness=vision.", "candidates": [ { "score": 89.7, "reference_url": "https://www.anthropic.com/claude-sonnet-5-system-card", "source_type": "model_card", "reported_setting": { "effort": "max", "tools": "python / benchmark tools", "sampling": "avg 5 trials unless source states otherwise", "harness": "vision" }, "notes": "Anthropic Claude Sonnet 5 System Card Figure 8.10.4.A, page 129: Claude Opus 4.8; ChartMuseum [Python tools]=89.7. Source setting: effort=max; tools=python / benchmark tools; sampling=avg 5 trials unless source states otherwise; harness=vision." } ] }, { "model_id": "claude-sonnet-5", "benchmark_id": "chartmuseum", "score": 70.1, "reference_url": "https://www.anthropic.com/claude-sonnet-5-system-card", "reported_setting": { "effort": "max", "tools": "none", "sampling": "avg 5 trials unless source states otherwise", "harness": "vision" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Anthropic Claude Sonnet 5 System Card Figure 8.10.4.A, page 129: Claude Sonnet 5; ChartMuseum [no tools]=70.1. Source setting: effort=max; tools=none; sampling=avg 5 trials unless source states otherwise; harness=vision.", "candidates": [ { "score": 86.7, "reference_url": "https://www.anthropic.com/claude-sonnet-5-system-card", "source_type": "model_card", "reported_setting": { "effort": "max", "tools": "python / benchmark tools", "sampling": "avg 5 trials unless source states otherwise", "harness": "vision" }, "notes": "Anthropic Claude Sonnet 5 System Card Figure 8.10.4.A, page 129: Claude Sonnet 5; ChartMuseum [Python tools]=86.7. Source setting: effort=max; tools=python / benchmark tools; sampling=avg 5 trials unless source states otherwise; harness=vision." } ] }, { "model_id": "claude-sonnet-4.6", "benchmark_id": "charxiv_reasoning", "score": 71.6, "reference_url": "https://www.anthropic.com/claude-sonnet-5-system-card", "reported_setting": { "effort": "max", "tools": "none", "sampling": "avg 5 trials unless source states otherwise", "harness": "vision" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Anthropic Claude Sonnet 5 System Card Figure 8.10.5.A, page 130: Claude Sonnet 4.6; CharXiv Reasoning [no tools]=71.6. Source setting: effort=max; tools=none; sampling=avg 5 trials unless source states otherwise; harness=vision.", "candidates": [ { "score": 85.3, "reference_url": "https://www.anthropic.com/claude-sonnet-5-system-card", "source_type": "model_card", "reported_setting": { "effort": "max", "tools": "python / benchmark tools", "sampling": "avg 5 trials unless source states otherwise", "harness": "vision" }, "notes": "Anthropic Claude Sonnet 5 System Card Figure 8.10.5.A, page 130: Claude Sonnet 4.6; CharXiv Reasoning [Python tools]=85.3. Source setting: effort=max; tools=python / benchmark tools; sampling=avg 5 trials unless source states otherwise; harness=vision." }, { "score": 72.4, "reference_url": "https://deepmind.google/models/model-cards/gemini-3-5-flash/", "source_type": "official_model_card", "reported_setting": { "mode": "thinking/reasoning", "effort": "max/best available", "tools": "none", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "Google self-computed for Gemini/GPT; provider for Claude", "temperature": "default", "snapshot": "May 2026" }, "notes": "Google Gemini 3.5 Flash May 2026 table: CharXiv Reasoning; no tools. Settings and provenance are preserved per cell." } ] }, { "model_id": "claude-sonnet-5", "benchmark_id": "charxiv_reasoning", "score": 77.0, "reference_url": "https://www.anthropic.com/claude-sonnet-5-system-card", "reported_setting": { "effort": "max", "tools": "none", "sampling": "avg 5 trials unless source states otherwise", "harness": "vision" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Anthropic Claude Sonnet 5 System Card Figure 8.10.5.A, page 130: Claude Sonnet 5; CharXiv Reasoning [no tools]=77.0. Source setting: effort=max; tools=none; sampling=avg 5 trials unless source states otherwise; harness=vision.", "candidates": [ { "score": 88.3, "reference_url": "https://www.anthropic.com/claude-sonnet-5-system-card", "source_type": "model_card", "reported_setting": { "effort": "max", "tools": "python / benchmark tools", "sampling": "avg 5 trials unless source states otherwise", "harness": "vision" }, "notes": "Anthropic Claude Sonnet 5 System Card Figure 8.10.5.A, page 130: Claude Sonnet 5; CharXiv Reasoning [Python tools]=88.3. Source setting: effort=max; tools=python / benchmark tools; sampling=avg 5 trials unless source states otherwise; harness=vision." } ] }, { "model_id": "claude-sonnet-4.6", "benchmark_id": "officeqa", "score": 68.7, "reference_url": "https://www.anthropic.com/claude-sonnet-5-system-card", "reported_setting": { "effort": "source does not state", "tools": "agentic benchmark harness", "sampling": "pass@1", "harness": "office" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Anthropic Claude Sonnet 5 System Card Figure 8.11.1.A, page 131: Claude Sonnet 4.6; OfficeQA / OfficeQA Pro [OfficeQA full]=68.7. Source setting: effort=source does not state; tools=agentic benchmark harness; sampling=pass@1; harness=office." }, { "model_id": "claude-opus-4.7", "benchmark_id": "officeqa", "score": 76.3, "reference_url": "https://www.anthropic.com/claude-sonnet-5-system-card", "reported_setting": { "effort": "source does not state", "tools": "agentic benchmark harness", "sampling": "pass@1", "harness": "office" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Anthropic Claude Sonnet 5 System Card Figure 8.11.1.A, page 131: Claude Opus 4.7; OfficeQA / OfficeQA Pro [OfficeQA full]=76.3. Source setting: effort=source does not state; tools=agentic benchmark harness; sampling=pass@1; harness=office." }, { "model_id": "claude-opus-4.8", "benchmark_id": "officeqa", "score": 77.6, "reference_url": "https://www.anthropic.com/claude-sonnet-5-system-card", "reported_setting": { "effort": "source does not state", "tools": "agentic benchmark harness", "sampling": "pass@1", "harness": "office" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Anthropic Claude Sonnet 5 System Card Figure 8.11.1.A, page 131: Claude Opus 4.8; OfficeQA / OfficeQA Pro [OfficeQA full]=77.6. Source setting: effort=source does not state; tools=agentic benchmark harness; sampling=pass@1; harness=office." }, { "model_id": "claude-sonnet-5", "benchmark_id": "officeqa", "score": 73.3, "reference_url": "https://www.anthropic.com/claude-sonnet-5-system-card", "reported_setting": { "effort": "source does not state", "tools": "agentic benchmark harness", "sampling": "pass@1", "harness": "office" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Anthropic Claude Sonnet 5 System Card Figure 8.11.1.A, page 131: Claude Sonnet 5; OfficeQA / OfficeQA Pro [OfficeQA full]=73.3. Source setting: effort=source does not state; tools=agentic benchmark harness; sampling=pass@1; harness=office." }, { "model_id": "claude-sonnet-4.6", "benchmark_id": "officeqa_pro", "score": 53.4, "reference_url": "https://www.anthropic.com/claude-sonnet-5-system-card", "reported_setting": { "effort": "source does not state", "tools": "agentic benchmark harness", "sampling": "pass@1", "harness": "office" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Anthropic Claude Sonnet 5 System Card Figure 8.11.1.A, page 131: Claude Sonnet 4.6; OfficeQA / OfficeQA Pro [OfficeQA Pro]=53.4. Source setting: effort=source does not state; tools=agentic benchmark harness; sampling=pass@1; harness=office." }, { "model_id": "claude-sonnet-5", "benchmark_id": "officeqa_pro", "score": 59.4, "reference_url": "https://www.anthropic.com/claude-sonnet-5-system-card", "reported_setting": { "effort": "source does not state", "tools": "agentic benchmark harness", "sampling": "pass@1", "harness": "office" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Anthropic Claude Sonnet 5 System Card Figure 8.11.1.A, page 131: Claude Sonnet 5; OfficeQA / OfficeQA Pro [OfficeQA Pro]=59.4. Source setting: effort=source does not state; tools=agentic benchmark harness; sampling=pass@1; harness=office." }, { "model_id": "claude-sonnet-4.6", "benchmark_id": "real_world_finance_v2_elo", "score": 1000.0, "reference_url": "https://www.anthropic.com/claude-sonnet-5-system-card", "reported_setting": { "effort": "max", "tools": "agentic benchmark harness", "sampling": "pass@1", "harness": "finance" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Anthropic Claude Sonnet 5 System Card Figure 8.11.2.A, page 133: Sonnet 4.6; Real-World Finance v2 [summary]=1000. Source setting: effort=max; tools=agentic benchmark harness; sampling=pass@1; harness=finance." }, { "model_id": "claude-opus-4.7", "benchmark_id": "real_world_finance_v2_elo", "score": 1207.0, "reference_url": "https://www.anthropic.com/claude-sonnet-5-system-card", "reported_setting": { "effort": "max", "tools": "agentic benchmark harness", "sampling": "pass@1", "harness": "finance" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Anthropic Claude Sonnet 5 System Card Figure 8.11.2.A, page 133: Opus 4.7; Real-World Finance v2 [summary]=1207. Source setting: effort=max; tools=agentic benchmark harness; sampling=pass@1; harness=finance." }, { "model_id": "claude-opus-4.8", "benchmark_id": "real_world_finance_v2_elo", "score": 1222.0, "reference_url": "https://www.anthropic.com/claude-sonnet-5-system-card", "reported_setting": { "effort": "max", "tools": "agentic benchmark harness", "sampling": "pass@1", "harness": "finance" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Anthropic Claude Sonnet 5 System Card Figure 8.11.2.A, page 133: Opus 4.8; Real-World Finance v2 [summary]=1222. Source setting: effort=max; tools=agentic benchmark harness; sampling=pass@1; harness=finance." }, { "model_id": "claude-fable-5", "benchmark_id": "real_world_finance_v2_elo", "score": 1374.0, "reference_url": "https://www.anthropic.com/claude-sonnet-5-system-card", "reported_setting": { "effort": "max", "tools": "agentic benchmark harness", "sampling": "pass@1", "harness": "finance" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Anthropic Claude Sonnet 5 System Card Figure 8.11.2.A, page 133: Fable 5; Real-World Finance v2 [summary]=1374. Source setting: effort=max; tools=agentic benchmark harness; sampling=pass@1; harness=finance." }, { "model_id": "claude-sonnet-5", "benchmark_id": "real_world_finance_v2_elo", "score": 1219.0, "reference_url": "https://www.anthropic.com/claude-sonnet-5-system-card", "reported_setting": { "effort": "max", "tools": "agentic benchmark harness", "sampling": "pass@1", "harness": "finance" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Anthropic Claude Sonnet 5 System Card Figure 8.11.2.A, page 133: Sonnet 5; Real-World Finance v2 [summary]=1219. Source setting: effort=max; tools=agentic benchmark harness; sampling=pass@1; harness=finance." }, { "model_id": "claude-sonnet-5", "benchmark_id": "gdpval_aa_elo", "score": 1618.0, "reference_url": "https://www.anthropic.com/claude-sonnet-5-system-card", "reported_setting": { "effort": "source does not state", "tools": "none", "sampling": "pass@1", "harness": "gdpval" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Anthropic Claude Sonnet 5 System Card Section 8.11.4, page 134: Claude Sonnet 5; GDPval-AA v2 [summary]=1618. Source setting: effort=source does not state; tools=none; sampling=pass@1; harness=gdpval.", "candidates": [ { "score": 1618.0, "reference_url": "https://www.anthropic.com/claude-sonnet-5-system-card", "source_type": "model_card", "reported_setting": { "effort": "max", "tools": "none", "sampling": "avg 5 trials unless source states otherwise", "harness": "table8" }, "notes": "Anthropic Claude Sonnet 5 System Card Table 8.1.A, page 115: Claude Sonnet 5; GDPval-AA v2 [summary]=1618. Source setting: effort=max; tools=none; sampling=avg 5 trials unless source states otherwise; harness=table8." } ] }, { "model_id": "claude-sonnet-5", "benchmark_id": "aa_briefcase_elo", "score": 1393.0, "reference_url": "https://www.anthropic.com/claude-sonnet-5-system-card", "reported_setting": { "effort": "source does not state", "tools": "none", "sampling": "pass@1", "harness": "briefcase" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Anthropic Claude Sonnet 5 System Card Section 8.11.7, page 137: Claude Sonnet 5; AA-Briefcase [summary]=1393. Source setting: effort=source does not state; tools=none; sampling=pass@1; harness=briefcase.", "candidates": [ { "score": 1383, "reference_url": "https://media.x.ai/v1/website/card-4p6-4cd2dc57.pdf", "source_type": "model_card", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "offline multi-file knowledge-work tools", "sampling": "unknown", "judge": "rubric plus analytical and presentation pairwise Elo", "harness": "Artificial Analysis AA-Briefcase harness", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown" }, "notes": "card46 AA Briefcase AA-Briefcase Elo; source effort=max; benchmark protocol matches canonical=True. Unknown source fields remain unknown." } ] }, { "model_id": "claude-sonnet-5", "benchmark_id": "toolathlon", "score": 54.3, "reference_url": "https://www.anthropic.com/claude-sonnet-5-system-card", "reported_setting": { "effort": "max", "tools": "agentic benchmark harness", "sampling": "3 trials/task", "harness": "toolathlon" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Anthropic Claude Sonnet 5 System Card Table 8.11.5.A, page 135: Claude Sonnet 5; Toolathlon [Pass@1]=54.3. Source setting: effort=max; tools=agentic benchmark harness; sampling=3 trials/task; harness=toolathlon." }, { "model_id": "claude-opus-4.7", "benchmark_id": "toolathlon", "score": 59.3, "reference_url": "https://www.anthropic.com/claude-sonnet-5-system-card", "reported_setting": { "effort": "max", "tools": "agentic benchmark harness", "sampling": "3 trials/task", "harness": "toolathlon" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Anthropic Claude Sonnet 5 System Card Table 8.11.5.A, page 135: Claude Opus 4.7; Toolathlon [Pass@1]=59.3. Source setting: effort=max; tools=agentic benchmark harness; sampling=3 trials/task; harness=toolathlon.", "candidates": [ { "score": 52.8, "reference_url": "https://seed.bytedance.com/en/blog/seed2-1-officially-released-advancing-ai-productivity", "source_type": "official_blog", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "agentic tool use", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official", "prompt_style": "official default", "temperature": "unknown" }, "notes": "Release blog EN: Daily-life/tool-use chart (https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/user-upload/4xfa4mqq4ujsp.png)." } ] }, { "model_id": "claude-sonnet-4.6", "benchmark_id": "toolathlon", "score": 49.4, "reference_url": "https://www.anthropic.com/claude-sonnet-5-system-card", "reported_setting": { "effort": "max", "tools": "agentic benchmark harness", "sampling": "3 trials/task", "harness": "toolathlon" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Anthropic Claude Sonnet 5 System Card Table 8.11.5.A, page 135: Claude Sonnet 4.6; Toolathlon [Pass@1]=49.4. Source setting: effort=max; tools=agentic benchmark harness; sampling=3 trials/task; harness=toolathlon." }, { "model_id": "gemini-3.5-flash", "benchmark_id": "automation_bench_private_heldout", "score": 12.6, "reference_url": "https://www.anthropic.com/claude-sonnet-5-system-card", "reported_setting": { "effort": "high", "tools": "agentic benchmark harness", "sampling": "pass@1", "harness": "automation" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Anthropic Claude Sonnet 5 System Card Figure 8.11.6.A, page 136: Gemini 3.5 Flash; AutomationBench [high]=12.6%. Source setting: effort=high; tools=agentic benchmark harness; sampling=pass@1; harness=automation.", "candidates": [ { "score": 14.5, "reference_url": "https://www.anthropic.com/claude-sonnet-5-system-card", "source_type": "model_card", "reported_setting": { "effort": "source does not state", "tools": "none", "sampling": "avg 5 trials unless source states otherwise", "harness": "table8" }, "notes": "Anthropic Claude Sonnet 5 System Card Table 8.1.A, page 115: Gemini 3.5 Flash; AutomationBench [summary]=14.5. Source setting: effort=source does not state; tools=none; sampling=avg 5 trials unless source states otherwise; harness=table8." }, { "score": 12.2, "reference_url": "https://www.anthropic.com/claude-sonnet-5-system-card", "source_type": "model_card", "reported_setting": { "effort": "low", "tools": "agentic benchmark harness", "sampling": "pass@1", "harness": "automation" }, "notes": "Anthropic Claude Sonnet 5 System Card Figure 8.11.6.A, page 136: Gemini 3.5 Flash; AutomationBench [low]=12.2%. Source setting: effort=low; tools=agentic benchmark harness; sampling=pass@1; harness=automation." }, { "score": 14.5, "reference_url": "https://www.anthropic.com/claude-sonnet-5-system-card", "source_type": "model_card", "reported_setting": { "effort": "medium", "tools": "agentic benchmark harness", "sampling": "pass@1", "harness": "automation" }, "notes": "Anthropic Claude Sonnet 5 System Card Figure 8.11.6.A, page 136: Gemini 3.5 Flash; AutomationBench [medium]=14.5%. Source setting: effort=medium; tools=agentic benchmark harness; sampling=pass@1; harness=automation." } ] }, { "model_id": "gpt-5.5", "benchmark_id": "automation_bench_private_heldout", "score": 12.9, "reference_url": "https://www.anthropic.com/claude-sonnet-5-system-card", "reported_setting": { "effort": "xhigh", "tools": "agentic benchmark harness", "sampling": "pass@1", "harness": "automation" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Anthropic Claude Sonnet 5 System Card Figure 8.11.6.A, page 136: GPT-5.5; AutomationBench [xhigh]=12.9%. Source setting: effort=xhigh; tools=agentic benchmark harness; sampling=pass@1; harness=automation.", "candidates": [ { "score": 12.9, "reference_url": "https://www.anthropic.com/claude-sonnet-5-system-card", "source_type": "model_card", "reported_setting": { "effort": "source does not state", "tools": "none", "sampling": "avg 5 trials unless source states otherwise", "harness": "table8" }, "notes": "Anthropic Claude Sonnet 5 System Card Table 8.1.A, page 115: GPT-5.5; AutomationBench [summary]=12.9. Source setting: effort=source does not state; tools=none; sampling=avg 5 trials unless source states otherwise; harness=table8." } ] }, { "model_id": "claude-opus-4.8", "benchmark_id": "automation_bench_private_heldout", "score": 15.5, "reference_url": "https://www.anthropic.com/claude-sonnet-5-system-card", "reported_setting": { "effort": "max", "tools": "agentic benchmark harness", "sampling": "pass@1", "harness": "automation" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Anthropic Claude Sonnet 5 System Card Figure 8.11.6.A, page 136: Claude Opus 4.8; AutomationBench [max]=15.5%. Source setting: effort=max; tools=agentic benchmark harness; sampling=pass@1; harness=automation.", "candidates": [ { "score": 17.0, "reference_url": "https://www.anthropic.com/claude-opus-5-system-card", "source_type": "model_card", "reported_setting": { "effort": "max" }, "notes": "Figure 8.13.7.A / Claude Opus 4.8 / max: Private held-out set across 47 applications; deterministic assertions." } ] }, { "model_id": "claude-fable-5", "benchmark_id": "automation_bench_private_heldout", "score": 17.4, "reference_url": "https://www.anthropic.com/claude-sonnet-5-system-card", "reported_setting": { "effort": "max", "tools": "agentic benchmark harness", "sampling": "pass@1", "harness": "automation" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Anthropic Claude Sonnet 5 System Card Figure 8.11.6.A, page 136: Claude Fable 5; AutomationBench [max]=17.4%. Source setting: effort=max; tools=agentic benchmark harness; sampling=pass@1; harness=automation." }, { "model_id": "claude-sonnet-4.6", "benchmark_id": "healthbench_length_adjusted", "score": 49.9, "reference_url": "https://www.anthropic.com/claude-sonnet-5-system-card", "reported_setting": { "effort": "max", "tools": "none", "sampling": "avg 5 trials unless source states otherwise", "harness": "health" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Anthropic Claude Sonnet 5 System Card Figure 8.12.1.A, page 138: Claude Sonnet 4.6; HealthBench [summary]=49.9%. Source setting: effort=max; tools=none; sampling=avg 5 trials unless source states otherwise; harness=health." }, { "model_id": "claude-mythos", "benchmark_id": "healthbench_length_adjusted", "score": 61.1, "reference_url": "https://www.anthropic.com/claude-sonnet-5-system-card", "reported_setting": { "effort": "max", "tools": "none", "sampling": "avg 5 trials unless source states otherwise", "harness": "health" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Anthropic Claude Sonnet 5 System Card Figure 8.12.1.A, page 138: Claude Mythos Preview; HealthBench [summary]=61.1%. Source setting: effort=max; tools=none; sampling=avg 5 trials unless source states otherwise; harness=health." }, { "model_id": "claude-opus-4.8", "benchmark_id": "healthbench_length_adjusted", "score": 59.3, "reference_url": "https://www.anthropic.com/claude-sonnet-5-system-card", "reported_setting": { "effort": "max", "tools": "none", "sampling": "avg 5 trials unless source states otherwise", "harness": "health" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Anthropic Claude Sonnet 5 System Card Figure 8.12.1.A, page 138: Claude Opus 4.8; HealthBench [summary]=59.3%. Source setting: effort=max; tools=none; sampling=avg 5 trials unless source states otherwise; harness=health." }, { "model_id": "claude-sonnet-5", "benchmark_id": "healthbench_length_adjusted", "score": 58.7, "reference_url": "https://www.anthropic.com/claude-sonnet-5-system-card", "reported_setting": { "effort": "max", "tools": "none", "sampling": "avg 5 trials unless source states otherwise", "harness": "health" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Anthropic Claude Sonnet 5 System Card Figure 8.12.1.A, page 138: Claude Sonnet 5; HealthBench [summary]=58.7%. Source setting: effort=max; tools=none; sampling=avg 5 trials unless source states otherwise; harness=health." }, { "model_id": "claude-mythos", "benchmark_id": "healthbench_professional_length_adjusted", "score": 64.0, "reference_url": "https://www.anthropic.com/claude-sonnet-5-system-card", "reported_setting": { "effort": "max", "tools": "none", "sampling": "avg 5 trials unless source states otherwise", "harness": "health" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Anthropic Claude Sonnet 5 System Card Figure 8.12.2.A, page 139: Claude Mythos Preview; HealthBench Professional [summary]=64.0%. Source setting: effort=max; tools=none; sampling=avg 5 trials unless source states otherwise; harness=health." }, { "model_id": "claude-sonnet-4.6", "benchmark_id": "global_mmlu", "score": 88.4, "reference_url": "https://www.anthropic.com/claude-sonnet-5-system-card", "reported_setting": { "effort": "max", "tools": "none", "sampling": "pass@1", "harness": "gmmlu" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Anthropic Claude Sonnet 5 System Card Figure 8.13.1.A, page 140: Claude Sonnet 4.6; GMMLU [summary]=88.4. Source setting: effort=max; tools=none; sampling=pass@1; harness=gmmlu." }, { "model_id": "claude-mythos", "benchmark_id": "global_mmlu", "score": 92.9, "reference_url": "https://www.anthropic.com/claude-sonnet-5-system-card", "reported_setting": { "effort": "max", "tools": "none", "sampling": "pass@1", "harness": "gmmlu" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Anthropic Claude Sonnet 5 System Card Figure 8.13.1.A, page 140: Claude Mythos Preview; GMMLU [summary]=92.9. Source setting: effort=max; tools=none; sampling=pass@1; harness=gmmlu." }, { "model_id": "claude-opus-4.8", "benchmark_id": "global_mmlu", "score": 90.9, "reference_url": "https://www.anthropic.com/claude-sonnet-5-system-card", "reported_setting": { "effort": "max", "tools": "none", "sampling": "pass@1", "harness": "gmmlu" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Anthropic Claude Sonnet 5 System Card Figure 8.13.1.A, page 140: Claude Opus 4.8; GMMLU [summary]=90.9. Source setting: effort=max; tools=none; sampling=pass@1; harness=gmmlu." }, { "model_id": "claude-sonnet-5", "benchmark_id": "global_mmlu", "score": 89.0, "reference_url": "https://www.anthropic.com/claude-sonnet-5-system-card", "reported_setting": { "effort": "max", "tools": "none", "sampling": "pass@1", "harness": "gmmlu" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Anthropic Claude Sonnet 5 System Card Figure 8.13.1.A, page 140: Claude Sonnet 5; GMMLU [summary]=89.0. Source setting: effort=max; tools=none; sampling=pass@1; harness=gmmlu." }, { "model_id": "claude-sonnet-4.6", "benchmark_id": "milu", "score": 89.4, "reference_url": "https://www.anthropic.com/claude-sonnet-5-system-card", "reported_setting": { "effort": "max", "tools": "none", "sampling": "avg 5 trials unless source states otherwise", "harness": "multilingual" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Anthropic Claude Sonnet 5 System Card Figure 8.13.2.A, page 141: Claude Sonnet 4.6; MILU [summary]=89.4. Source setting: effort=max; tools=none; sampling=avg 5 trials unless source states otherwise; harness=multilingual." }, { "model_id": "claude-mythos", "benchmark_id": "milu", "score": 92.7, "reference_url": "https://www.anthropic.com/claude-sonnet-5-system-card", "reported_setting": { "effort": "max", "tools": "none", "sampling": "avg 5 trials unless source states otherwise", "harness": "multilingual" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Anthropic Claude Sonnet 5 System Card Figure 8.13.2.A, page 141: Claude Mythos Preview; MILU [summary]=92.7. Source setting: effort=max; tools=none; sampling=avg 5 trials unless source states otherwise; harness=multilingual." }, { "model_id": "claude-opus-4.8", "benchmark_id": "milu", "score": 91.1, "reference_url": "https://www.anthropic.com/claude-sonnet-5-system-card", "reported_setting": { "effort": "max", "tools": "none", "sampling": "avg 5 trials unless source states otherwise", "harness": "multilingual" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Anthropic Claude Sonnet 5 System Card Figure 8.13.2.A, page 141: Claude Opus 4.8; MILU [summary]=91.1. Source setting: effort=max; tools=none; sampling=avg 5 trials unless source states otherwise; harness=multilingual." }, { "model_id": "claude-sonnet-5", "benchmark_id": "milu", "score": 89.3, "reference_url": "https://www.anthropic.com/claude-sonnet-5-system-card", "reported_setting": { "effort": "max", "tools": "none", "sampling": "avg 5 trials unless source states otherwise", "harness": "multilingual" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Anthropic Claude Sonnet 5 System Card Figure 8.13.2.A, page 141: Claude Sonnet 5; MILU [summary]=89.3. Source setting: effort=max; tools=none; sampling=avg 5 trials unless source states otherwise; harness=multilingual." }, { "model_id": "claude-sonnet-4.6", "benchmark_id": "include_base_44", "score": 85.7, "reference_url": "https://www.anthropic.com/claude-sonnet-5-system-card", "reported_setting": { "effort": "max", "tools": "none", "sampling": "avg 5 trials unless source states otherwise", "harness": "multilingual" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Anthropic Claude Sonnet 5 System Card Figure 8.13.3.A, page 142: Claude Sonnet 4.6; INCLUDE [summary]=85.7. Source setting: effort=max; tools=none; sampling=avg 5 trials unless source states otherwise; harness=multilingual." }, { "model_id": "claude-mythos", "benchmark_id": "include_base_44", "score": 90.1, "reference_url": "https://www.anthropic.com/claude-sonnet-5-system-card", "reported_setting": { "effort": "max", "tools": "none", "sampling": "avg 5 trials unless source states otherwise", "harness": "multilingual" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Anthropic Claude Sonnet 5 System Card Figure 8.13.3.A, page 142: Claude Mythos Preview; INCLUDE [summary]=90.1. Source setting: effort=max; tools=none; sampling=avg 5 trials unless source states otherwise; harness=multilingual." }, { "model_id": "claude-opus-4.8", "benchmark_id": "include_base_44", "score": 88.5, "reference_url": "https://www.anthropic.com/claude-sonnet-5-system-card", "reported_setting": { "effort": "max", "tools": "none", "sampling": "avg 5 trials unless source states otherwise", "harness": "multilingual" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Anthropic Claude Sonnet 5 System Card Figure 8.13.3.A, page 142: Claude Opus 4.8; INCLUDE [summary]=88.5. Source setting: effort=max; tools=none; sampling=avg 5 trials unless source states otherwise; harness=multilingual." }, { "model_id": "claude-sonnet-5", "benchmark_id": "include_base_44", "score": 86.5, "reference_url": "https://www.anthropic.com/claude-sonnet-5-system-card", "reported_setting": { "effort": "max", "tools": "none", "sampling": "avg 5 trials unless source states otherwise", "harness": "multilingual" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Anthropic Claude Sonnet 5 System Card Figure 8.13.3.A, page 142: Claude Sonnet 5; INCLUDE [summary]=86.5. Source setting: effort=max; tools=none; sampling=avg 5 trials unless source states otherwise; harness=multilingual." }, { "model_id": "claude-sonnet-4.6", "benchmark_id": "biomysterybench_human_solvable", "score": 0.78, "reference_url": "https://www.anthropic.com/claude-sonnet-5-system-card", "reported_setting": { "effort": "source does not state", "tools": "none", "sampling": "pass@1", "harness": "life" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Anthropic Claude Sonnet 5 System Card Figure 8.14.6.A, page 144: Claude Sonnet 4.6; BioMysteryBench / Human solvable [summary]=0.78. Source setting: effort=source does not state; tools=none; sampling=pass@1; harness=life." }, { "model_id": "claude-opus-4.8", "benchmark_id": "biomysterybench_human_solvable", "score": 0.8, "reference_url": "https://www.anthropic.com/claude-sonnet-5-system-card", "reported_setting": { "effort": "source does not state", "tools": "none", "sampling": "pass@1", "harness": "life" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Anthropic Claude Sonnet 5 System Card Figure 8.14.6.A, page 144: Claude Opus 4.8; BioMysteryBench / Human solvable [summary]=0.80. Source setting: effort=source does not state; tools=none; sampling=pass@1; harness=life." }, { "model_id": "claude-fable-5", "benchmark_id": "biomysterybench_human_solvable", "score": 0.84, "reference_url": "https://www.anthropic.com/claude-sonnet-5-system-card", "reported_setting": { "effort": "source does not state", "tools": "none", "sampling": "pass@1", "harness": "life", "deployment": "claude-mythos-5" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Anthropic Claude Sonnet 5 System Card Figure 8.14.6.A, page 144: Claude Mythos 5; BioMysteryBench / Human solvable [summary]=0.84. Source setting: effort=source does not state; tools=none; sampling=pass@1; harness=life; deployment=claude-mythos-5." }, { "model_id": "claude-sonnet-5", "benchmark_id": "biomysterybench_human_solvable", "score": 0.82, "reference_url": "https://www.anthropic.com/claude-sonnet-5-system-card", "reported_setting": { "effort": "source does not state", "tools": "none", "sampling": "pass@1", "harness": "life" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Anthropic Claude Sonnet 5 System Card Figure 8.14.6.A, page 144: Claude Sonnet 5; BioMysteryBench / Human solvable [summary]=0.82. Source setting: effort=source does not state; tools=none; sampling=pass@1; harness=life." }, { "model_id": "claude-sonnet-4.6", "benchmark_id": "biomysterybench_human_difficult", "score": 0.31, "reference_url": "https://www.anthropic.com/claude-sonnet-5-system-card", "reported_setting": { "effort": "source does not state", "tools": "none", "sampling": "pass@1", "harness": "life" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Anthropic Claude Sonnet 5 System Card Figure 8.14.6.A, page 144: Claude Sonnet 4.6; BioMysteryBench / Human difficult [summary]=0.31. Source setting: effort=source does not state; tools=none; sampling=pass@1; harness=life." }, { "model_id": "claude-opus-4.8", "benchmark_id": "biomysterybench_human_difficult", "score": 0.4, "reference_url": "https://www.anthropic.com/claude-sonnet-5-system-card", "reported_setting": { "effort": "source does not state", "tools": "none", "sampling": "pass@1", "harness": "life" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Anthropic Claude Sonnet 5 System Card Figure 8.14.6.A, page 144: Claude Opus 4.8; BioMysteryBench / Human difficult [summary]=0.40. Source setting: effort=source does not state; tools=none; sampling=pass@1; harness=life." }, { "model_id": "claude-fable-5", "benchmark_id": "biomysterybench_human_difficult", "score": 0.46, "reference_url": "https://www.anthropic.com/claude-sonnet-5-system-card", "reported_setting": { "effort": "source does not state", "tools": "none", "sampling": "pass@1", "harness": "life", "deployment": "claude-mythos-5" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Anthropic Claude Sonnet 5 System Card Figure 8.14.6.A, page 144: Claude Mythos 5; BioMysteryBench / Human difficult [summary]=0.46. Source setting: effort=source does not state; tools=none; sampling=pass@1; harness=life; deployment=claude-mythos-5." }, { "model_id": "claude-sonnet-5", "benchmark_id": "biomysterybench_human_difficult", "score": 0.35, "reference_url": "https://www.anthropic.com/claude-sonnet-5-system-card", "reported_setting": { "effort": "source does not state", "tools": "none", "sampling": "pass@1", "harness": "life" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Anthropic Claude Sonnet 5 System Card Figure 8.14.6.A, page 144: Claude Sonnet 5; BioMysteryBench / Human difficult [summary]=0.35. Source setting: effort=source does not state; tools=none; sampling=pass@1; harness=life." }, { "model_id": "claude-sonnet-4.6", "benchmark_id": "spatialbench_verified", "score": 0.6, "reference_url": "https://www.anthropic.com/claude-sonnet-5-system-card", "reported_setting": { "effort": "source does not state", "tools": "none", "sampling": "pass@1", "harness": "life" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Anthropic Claude Sonnet 5 System Card Figure 8.14.6.A, page 144: Claude Sonnet 4.6; LatchBio Bioinformatics / SpatialBench Verified [summary]=0.60. Source setting: effort=source does not state; tools=none; sampling=pass@1; harness=life." }, { "model_id": "claude-opus-4.8", "benchmark_id": "spatialbench_verified", "score": 0.67, "reference_url": "https://www.anthropic.com/claude-sonnet-5-system-card", "reported_setting": { "effort": "source does not state", "tools": "none", "sampling": "pass@1", "harness": "life" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Anthropic Claude Sonnet 5 System Card Figure 8.14.6.A, page 144: Claude Opus 4.8; LatchBio Bioinformatics / SpatialBench Verified [summary]=0.67. Source setting: effort=source does not state; tools=none; sampling=pass@1; harness=life.", "candidates": [ { "score": 0.666, "reference_url": "https://www.anthropic.com/claude-opus-5-system-card", "source_type": "model_card", "reported_setting": { "effort": "max", "tools": "benchmark-specific" }, "notes": "Figure 8.17.6.A / SpatialBench Verified / Claude Opus 4.8: 115 problems. Source label 66.6%; normalized to 0.666 for the 0-1 score metric." } ] }, { "model_id": "claude-fable-5", "benchmark_id": "spatialbench_verified", "score": 0.7, "reference_url": "https://www.anthropic.com/claude-sonnet-5-system-card", "reported_setting": { "effort": "source does not state", "tools": "none", "sampling": "pass@1", "harness": "life", "deployment": "claude-mythos-5" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Anthropic Claude Sonnet 5 System Card Figure 8.14.6.A, page 144: Claude Mythos 5; LatchBio Bioinformatics / SpatialBench Verified [summary]=0.70. Source setting: effort=source does not state; tools=none; sampling=pass@1; harness=life; deployment=claude-mythos-5.", "candidates": [ { "score": 0.692, "reference_url": "https://www.anthropic.com/claude-opus-5-system-card", "source_type": "model_card", "reported_setting": { "effort": "max", "tools": "benchmark-specific", "deployment": "claude-mythos-5" }, "notes": "Figure 8.17.6.A / SpatialBench Verified / Claude Mythos 5: 115 problems. Source label 69.2%; normalized to 0.692 for the 0-1 score metric." } ] }, { "model_id": "claude-sonnet-5", "benchmark_id": "spatialbench_verified", "score": 0.7, "reference_url": "https://www.anthropic.com/claude-sonnet-5-system-card", "reported_setting": { "effort": "source does not state", "tools": "none", "sampling": "pass@1", "harness": "life" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Anthropic Claude Sonnet 5 System Card Figure 8.14.6.A, page 144: Claude Sonnet 5; LatchBio Bioinformatics / SpatialBench Verified [summary]=0.70. Source setting: effort=source does not state; tools=none; sampling=pass@1; harness=life.", "candidates": [ { "score": 0.678, "reference_url": "https://www.anthropic.com/claude-opus-5-system-card", "source_type": "model_card", "reported_setting": { "effort": "max", "tools": "benchmark-specific" }, "notes": "Figure 8.17.6.A / SpatialBench Verified / Claude Sonnet 5: 115 problems. Source label 67.8%; normalized to 0.678 for the 0-1 score metric." } ] }, { "model_id": "claude-sonnet-4.6", "benchmark_id": "singlecellbench", "score": 0.5, "reference_url": "https://www.anthropic.com/claude-sonnet-5-system-card", "reported_setting": { "effort": "source does not state", "tools": "none", "sampling": "pass@1", "harness": "life" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Anthropic Claude Sonnet 5 System Card Figure 8.14.6.A, page 144: Claude Sonnet 4.6; LatchBio Bioinformatics / SingleCellBench [summary]=0.50. Source setting: effort=source does not state; tools=none; sampling=pass@1; harness=life." }, { "model_id": "claude-opus-4.8", "benchmark_id": "singlecellbench", "score": 0.58, "reference_url": "https://www.anthropic.com/claude-sonnet-5-system-card", "reported_setting": { "effort": "source does not state", "tools": "none", "sampling": "pass@1", "harness": "life" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Anthropic Claude Sonnet 5 System Card Figure 8.14.6.A, page 144: Claude Opus 4.8; LatchBio Bioinformatics / SingleCellBench [summary]=0.58. Source setting: effort=source does not state; tools=none; sampling=pass@1; harness=life.", "candidates": [ { "score": 0.582, "reference_url": "https://www.anthropic.com/claude-opus-5-system-card", "source_type": "model_card", "reported_setting": { "effort": "max", "tools": "benchmark-specific" }, "notes": "Figure 8.17.6.A / SingleCellBench / Claude Opus 4.8: 195 problems. Source label 58.2%; normalized to 0.582 for the 0-1 score metric." } ] }, { "model_id": "claude-fable-5", "benchmark_id": "singlecellbench", "score": 0.59, "reference_url": "https://www.anthropic.com/claude-sonnet-5-system-card", "reported_setting": { "effort": "source does not state", "tools": "none", "sampling": "pass@1", "harness": "life", "deployment": "claude-mythos-5" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Anthropic Claude Sonnet 5 System Card Figure 8.14.6.A, page 144: Claude Mythos 5; LatchBio Bioinformatics / SingleCellBench [summary]=0.59. Source setting: effort=source does not state; tools=none; sampling=pass@1; harness=life; deployment=claude-mythos-5.", "candidates": [ { "score": 0.593, "reference_url": "https://www.anthropic.com/claude-opus-5-system-card", "source_type": "model_card", "reported_setting": { "effort": "max", "tools": "benchmark-specific", "deployment": "claude-mythos-5" }, "notes": "Figure 8.17.6.A / SingleCellBench / Claude Mythos 5: 195 problems. Source label 59.3%; normalized to 0.593 for the 0-1 score metric." } ] }, { "model_id": "claude-sonnet-5", "benchmark_id": "singlecellbench", "score": 0.56, "reference_url": "https://www.anthropic.com/claude-sonnet-5-system-card", "reported_setting": { "effort": "source does not state", "tools": "none", "sampling": "pass@1", "harness": "life" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Anthropic Claude Sonnet 5 System Card Figure 8.14.6.A, page 144: Claude Sonnet 5; LatchBio Bioinformatics / SingleCellBench [summary]=0.56. Source setting: effort=source does not state; tools=none; sampling=pass@1; harness=life.", "candidates": [ { "score": 0.562, "reference_url": "https://www.anthropic.com/claude-opus-5-system-card", "source_type": "model_card", "reported_setting": { "effort": "max", "tools": "benchmark-specific" }, "notes": "Figure 8.17.6.A / SingleCellBench / Claude Sonnet 5: 195 problems. Source label 56.2%; normalized to 0.562 for the 0-1 score metric." } ] }, { "model_id": "claude-sonnet-4.6", "benchmark_id": "structural_biology_open_ended_internal", "score": 0.32, "reference_url": "https://www.anthropic.com/claude-sonnet-5-system-card", "reported_setting": { "effort": "source does not state", "tools": "none", "sampling": "pass@1", "harness": "life" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Anthropic Claude Sonnet 5 System Card Figure 8.14.6.A, page 144: Claude Sonnet 4.6; Structural biology open-ended [summary]=0.32. Source setting: effort=source does not state; tools=none; sampling=pass@1; harness=life." }, { "model_id": "claude-opus-4.8", "benchmark_id": "structural_biology_open_ended_internal", "score": 0.77, "reference_url": "https://www.anthropic.com/claude-sonnet-5-system-card", "reported_setting": { "effort": "source does not state", "tools": "none", "sampling": "pass@1", "harness": "life" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Anthropic Claude Sonnet 5 System Card Figure 8.14.6.A, page 144: Claude Opus 4.8; Structural biology open-ended [summary]=0.77. Source setting: effort=source does not state; tools=none; sampling=pass@1; harness=life." }, { "model_id": "claude-fable-5", "benchmark_id": "structural_biology_open_ended_internal", "score": 0.87, "reference_url": "https://www.anthropic.com/claude-sonnet-5-system-card", "reported_setting": { "effort": "source does not state", "tools": "none", "sampling": "pass@1", "harness": "life", "deployment": "claude-mythos-5" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Anthropic Claude Sonnet 5 System Card Figure 8.14.6.A, page 144: Claude Mythos 5; Structural biology open-ended [summary]=0.87. Source setting: effort=source does not state; tools=none; sampling=pass@1; harness=life; deployment=claude-mythos-5." }, { "model_id": "claude-sonnet-5", "benchmark_id": "structural_biology_open_ended_internal", "score": 0.69, "reference_url": "https://www.anthropic.com/claude-sonnet-5-system-card", "reported_setting": { "effort": "source does not state", "tools": "none", "sampling": "pass@1", "harness": "life" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Anthropic Claude Sonnet 5 System Card Figure 8.14.6.A, page 144: Claude Sonnet 5; Structural biology open-ended [summary]=0.69. Source setting: effort=source does not state; tools=none; sampling=pass@1; harness=life." }, { "model_id": "claude-sonnet-4.6", "benchmark_id": "proteingym_hard", "score": 0.35, "reference_url": "https://www.anthropic.com/claude-sonnet-5-system-card", "reported_setting": { "effort": "source does not state", "tools": "none", "sampling": "pass@1", "harness": "life" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Anthropic Claude Sonnet 5 System Card Figure 8.14.6.A, page 144: Claude Sonnet 4.6; ProteinGym Hard [summary]=0.35. Source setting: effort=source does not state; tools=none; sampling=pass@1; harness=life." }, { "model_id": "claude-opus-4.8", "benchmark_id": "proteingym_hard", "score": 0.4, "reference_url": "https://www.anthropic.com/claude-sonnet-5-system-card", "reported_setting": { "effort": "source does not state", "tools": "none", "sampling": "pass@1", "harness": "life" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Anthropic Claude Sonnet 5 System Card Figure 8.14.6.A, page 144: Claude Opus 4.8; ProteinGym Hard [summary]=0.40. Source setting: effort=source does not state; tools=none; sampling=pass@1; harness=life." }, { "model_id": "claude-fable-5", "benchmark_id": "proteingym_hard", "score": 0.45, "reference_url": "https://www.anthropic.com/claude-sonnet-5-system-card", "reported_setting": { "effort": "source does not state", "tools": "none", "sampling": "pass@1", "harness": "life", "deployment": "claude-mythos-5" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Anthropic Claude Sonnet 5 System Card Figure 8.14.6.A, page 144: Claude Mythos 5; ProteinGym Hard [summary]=0.45. Source setting: effort=source does not state; tools=none; sampling=pass@1; harness=life; deployment=claude-mythos-5.", "candidates": [ { "score": 0.458, "reference_url": "https://www.anthropic.com/claude-opus-5-system-card", "source_type": "model_card", "reported_setting": { "effort": "max", "tools": "benchmark-specific", "deployment": "claude-mythos-5" }, "notes": "Figure 8.17.6.A / ProteinGym Hard / Claude Mythos 5: Rank correlation. Source label 45.8%; normalized to 0.458 for the 0-1 score metric." } ] }, { "model_id": "claude-sonnet-5", "benchmark_id": "proteingym_hard", "score": 0.37, "reference_url": "https://www.anthropic.com/claude-sonnet-5-system-card", "reported_setting": { "effort": "source does not state", "tools": "none", "sampling": "pass@1", "harness": "life" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Anthropic Claude Sonnet 5 System Card Figure 8.14.6.A, page 144: Claude Sonnet 5; ProteinGym Hard [summary]=0.37. Source setting: effort=source does not state; tools=none; sampling=pass@1; harness=life.", "candidates": [ { "score": 0.366, "reference_url": "https://www.anthropic.com/claude-opus-5-system-card", "source_type": "model_card", "reported_setting": { "effort": "max", "tools": "benchmark-specific" }, "notes": "Figure 8.17.6.A / ProteinGym Hard / Claude Sonnet 5: Rank correlation. Source label 36.6%; normalized to 0.366 for the 0-1 score metric." } ] }, { "model_id": "claude-sonnet-4.6", "benchmark_id": "organic_chemistry_internal", "score": 0.56, "reference_url": "https://www.anthropic.com/claude-sonnet-5-system-card", "reported_setting": { "effort": "source does not state", "tools": "none", "sampling": "pass@1", "harness": "life" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Anthropic Claude Sonnet 5 System Card Figure 8.14.6.A, page 144: Claude Sonnet 4.6; Organic chemistry [summary]=0.56. Source setting: effort=source does not state; tools=none; sampling=pass@1; harness=life." }, { "model_id": "claude-opus-4.8", "benchmark_id": "organic_chemistry_internal", "score": 0.86, "reference_url": "https://www.anthropic.com/claude-sonnet-5-system-card", "reported_setting": { "effort": "source does not state", "tools": "none", "sampling": "pass@1", "harness": "life" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Anthropic Claude Sonnet 5 System Card Figure 8.14.6.A, page 144: Claude Opus 4.8; Organic chemistry [summary]=0.86. Source setting: effort=source does not state; tools=none; sampling=pass@1; harness=life." }, { "model_id": "claude-fable-5", "benchmark_id": "organic_chemistry_internal", "score": 0.9, "reference_url": "https://www.anthropic.com/claude-sonnet-5-system-card", "reported_setting": { "effort": "source does not state", "tools": "none", "sampling": "pass@1", "harness": "life", "deployment": "claude-mythos-5" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Anthropic Claude Sonnet 5 System Card Figure 8.14.6.A, page 144: Claude Mythos 5; Organic chemistry [summary]=0.90. Source setting: effort=source does not state; tools=none; sampling=pass@1; harness=life; deployment=claude-mythos-5." }, { "model_id": "claude-sonnet-5", "benchmark_id": "organic_chemistry_internal", "score": 0.81, "reference_url": "https://www.anthropic.com/claude-sonnet-5-system-card", "reported_setting": { "effort": "source does not state", "tools": "none", "sampling": "pass@1", "harness": "life" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Anthropic Claude Sonnet 5 System Card Figure 8.14.6.A, page 144: Claude Sonnet 5; Organic chemistry [summary]=0.81. Source setting: effort=source does not state; tools=none; sampling=pass@1; harness=life." }, { "model_id": "claude-sonnet-4.6", "benchmark_id": "protocol_troubleshooting_internal", "score": 0.42, "reference_url": "https://www.anthropic.com/claude-sonnet-5-system-card", "reported_setting": { "effort": "source does not state", "tools": "none", "sampling": "pass@1", "harness": "life" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Anthropic Claude Sonnet 5 System Card Figure 8.14.6.A, page 144: Claude Sonnet 4.6; Protocol troubleshooting [summary]=0.42. Source setting: effort=source does not state; tools=none; sampling=pass@1; harness=life." }, { "model_id": "claude-opus-4.8", "benchmark_id": "protocol_troubleshooting_internal", "score": 0.6, "reference_url": "https://www.anthropic.com/claude-sonnet-5-system-card", "reported_setting": { "effort": "source does not state", "tools": "none", "sampling": "pass@1", "harness": "life" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Anthropic Claude Sonnet 5 System Card Figure 8.14.6.A, page 144: Claude Opus 4.8; Protocol troubleshooting [summary]=0.60. Source setting: effort=source does not state; tools=none; sampling=pass@1; harness=life.", "candidates": [ { "score": 0.596, "reference_url": "https://www.anthropic.com/claude-opus-5-system-card", "source_type": "model_card", "reported_setting": { "effort": "max", "tools": "benchmark-specific" }, "notes": "Figure 8.17.6.A / Protocol Troubleshooting / Claude Opus 4.8: Bash, file editor, and web search. Source label 59.6%; normalized to 0.596 for the 0-1 score metric." } ] }, { "model_id": "claude-fable-5", "benchmark_id": "protocol_troubleshooting_internal", "score": 0.67, "reference_url": "https://www.anthropic.com/claude-sonnet-5-system-card", "reported_setting": { "effort": "source does not state", "tools": "none", "sampling": "pass@1", "harness": "life", "deployment": "claude-mythos-5" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Anthropic Claude Sonnet 5 System Card Figure 8.14.6.A, page 144: Claude Mythos 5; Protocol troubleshooting [summary]=0.67. Source setting: effort=source does not state; tools=none; sampling=pass@1; harness=life; deployment=claude-mythos-5.", "candidates": [ { "score": 0.667, "reference_url": "https://www.anthropic.com/claude-opus-5-system-card", "source_type": "model_card", "reported_setting": { "effort": "max", "tools": "benchmark-specific", "deployment": "claude-mythos-5" }, "notes": "Figure 8.17.6.A / Protocol Troubleshooting / Claude Mythos 5: Bash, file editor, and web search. Source label 66.7%; normalized to 0.667 for the 0-1 score metric." } ] }, { "model_id": "claude-sonnet-5", "benchmark_id": "protocol_troubleshooting_internal", "score": 0.62, "reference_url": "https://www.anthropic.com/claude-sonnet-5-system-card", "reported_setting": { "effort": "source does not state", "tools": "none", "sampling": "pass@1", "harness": "life" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Anthropic Claude Sonnet 5 System Card Figure 8.14.6.A, page 144: Claude Sonnet 5; Protocol troubleshooting [summary]=0.62. Source setting: effort=source does not state; tools=none; sampling=pass@1; harness=life.", "candidates": [ { "score": 0.623, "reference_url": "https://www.anthropic.com/claude-opus-5-system-card", "source_type": "model_card", "reported_setting": { "effort": "max", "tools": "benchmark-specific" }, "notes": "Figure 8.17.6.A / Protocol Troubleshooting / Claude Sonnet 5: Bash, file editor, and web search. Source label 62.3%; normalized to 0.623 for the 0-1 score metric." } ] }, { "model_id": "claude-sonnet-5", "benchmark_id": "swe_bench_verified", "score": 85.2, "reference_url": "https://www.anthropic.com/claude-sonnet-5-system-card", "reported_setting": { "effort": "max", "tools": "none", "sampling": "avg 5 trials unless source states otherwise", "harness": "standard" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Anthropic Claude Sonnet 5 System Card Section 8.2, page 116: Claude Sonnet 5; SWE-bench Verified / Multilingual / Multimodal [Verified]=85.2%. Source setting: effort=max; tools=none; sampling=avg 5 trials unless source states otherwise; harness=standard." }, { "model_id": "claude-sonnet-5", "benchmark_id": "swe_bench_multilingual", "score": 78.3, "reference_url": "https://www.anthropic.com/claude-sonnet-5-system-card", "reported_setting": { "effort": "max", "tools": "none", "sampling": "avg 5 trials unless source states otherwise", "harness": "standard" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Anthropic Claude Sonnet 5 System Card Section 8.2, page 116: Claude Sonnet 5; SWE-bench Verified / Multilingual / Multimodal [Multilingual]=78.3%. Source setting: effort=max; tools=none; sampling=avg 5 trials unless source states otherwise; harness=standard." }, { "model_id": "claude-sonnet-5", "benchmark_id": "swe_bench_multimodal", "score": 28.1, "reference_url": "https://www.anthropic.com/claude-sonnet-5-system-card", "reported_setting": { "effort": "max", "tools": "none", "sampling": "avg 5 trials unless source states otherwise", "harness": "standard" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Anthropic Claude Sonnet 5 System Card Section 8.2, page 116: Claude Sonnet 5; SWE-bench Verified / Multilingual / Multimodal [Multimodal]=28.1%. Source setting: effort=max; tools=none; sampling=avg 5 trials unless source states otherwise; harness=standard." }, { "model_id": "claude-sonnet-5", "benchmark_id": "usamo_2026", "score": 79.5, "reference_url": "https://www.anthropic.com/claude-sonnet-5-system-card", "reported_setting": { "effort": "high", "tools": "none", "sampling": "10 attempts/problem", "harness": "usamo" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Anthropic Claude Sonnet 5 System Card Section 8.6, page 120: Claude Sonnet 5; USAMO 2026 [summary]=79.5%. Source setting: effort=high; tools=none; sampling=10 attempts/problem; harness=usamo." }, { "model_id": "claude-sonnet-4.6", "benchmark_id": "usamo_2026", "score": 55.0, "reference_url": "https://www.anthropic.com/claude-sonnet-5-system-card", "reported_setting": { "effort": "high", "tools": "none", "sampling": "10 attempts/problem", "harness": "usamo" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Anthropic Claude Sonnet 5 System Card Section 8.6, page 120: Claude Sonnet 4.6; USAMO 2026 [summary]=55.0%. Source setting: effort=high; tools=none; sampling=10 attempts/problem; harness=usamo." }, { "model_id": "claude-fable-5", "benchmark_id": "usamo_2026", "score": 99.8, "reference_url": "https://www.anthropic.com/claude-sonnet-5-system-card", "reported_setting": { "effort": "high", "tools": "none", "sampling": "10 attempts/problem", "harness": "usamo", "deployment": "claude-mythos-5" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Anthropic Claude Sonnet 5 System Card Section 8.6, page 120: Claude Mythos 5; USAMO 2026 [summary]=99.8%. Source setting: effort=high; tools=none; sampling=10 attempts/problem; harness=usamo; deployment=claude-mythos-5." }, { "model_id": "claude-opus-5", "benchmark_id": "swe_bench_pro", "score": 79.2, "reference_url": "https://www.anthropic.com/claude-opus-5-system-card", "reported_setting": { "effort": "max", "harness": "official or cited leaderboard" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Table 8.1.A / SWE-bench Pro: Unless noted, Anthropic values use adaptive thinking at max effort, default sampling, mean over five trials. Table 8.1.A card summary." }, { "model_id": "claude-opus-5", "benchmark_id": "swe_bench_multilingual", "score": 89.5, "reference_url": "https://www.anthropic.com/claude-opus-5-system-card", "reported_setting": { "effort": "max", "harness": "official or cited leaderboard" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Table 8.1.A / SWE-bench Multilingual: Unless noted, Anthropic values use adaptive thinking at max effort, default sampling, mean over five trials. Table 8.1.A card summary." }, { "model_id": "claude-fable-5", "benchmark_id": "swe_bench_multilingual", "score": 86.6, "reference_url": "https://www.anthropic.com/claude-opus-5-system-card", "reported_setting": { "effort": "max", "harness": "official or cited leaderboard" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Table 8.1.A / SWE-bench Multilingual: Unless noted, Anthropic values use adaptive thinking at max effort, default sampling, mean over five trials. Table 8.1.A card summary." }, { "model_id": "claude-opus-5", "benchmark_id": "swe_bench_multimodal", "score": 59.4, "reference_url": "https://www.anthropic.com/claude-opus-5-system-card", "reported_setting": { "effort": "max", "harness": "official or cited leaderboard" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Table 8.1.A / SWE-bench Multimodal: Unless noted, Anthropic values use adaptive thinking at max effort, default sampling, mean over five trials. Table 8.1.A card summary." }, { "model_id": "claude-opus-4.8", "benchmark_id": "swe_bench_multimodal", "score": 38.4, "reference_url": "https://www.anthropic.com/claude-opus-5-system-card", "reported_setting": { "effort": "max", "harness": "official or cited leaderboard" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Table 8.1.A / SWE-bench Multimodal: Unless noted, Anthropic values use adaptive thinking at max effort, default sampling, mean over five trials. Table 8.1.A card summary." }, { "model_id": "claude-fable-5", "benchmark_id": "swe_bench_multimodal", "score": 54.1, "reference_url": "https://www.anthropic.com/claude-opus-5-system-card", "reported_setting": { "effort": "max", "harness": "official or cited leaderboard" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Table 8.1.A / SWE-bench Multimodal: Unless noted, Anthropic values use adaptive thinking at max effort, default sampling, mean over five trials. Table 8.1.A card summary." }, { "model_id": "claude-opus-5", "benchmark_id": "healthbench_professional_length_adjusted", "score": 59.8, "reference_url": "https://www.anthropic.com/claude-opus-5-system-card", "reported_setting": { "effort": "max", "harness": "official or cited leaderboard" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Table 8.1.A / HealthBench Professional Length-Adjusted: Unless noted, Anthropic values use adaptive thinking at max effort, default sampling, mean over five trials. Table 8.1.A card summary.", "candidates": [ { "score": 59.8, "reference_url": "https://www.anthropic.com/claude-opus-5-system-card", "source_type": "model_card", "reported_setting": { "effort": "max", "tools": "none" }, "notes": "Figure 8.15 / HealthBench Professional Length-Adjusted / Claude Opus 5: Five trials; Opus 4.8 grader; no customized system prompt." } ] }, { "model_id": "gpt-5.6-sol", "benchmark_id": "automation_bench_private_heldout", "score": 18.1, "reference_url": "https://www.anthropic.com/claude-opus-5-system-card", "reported_setting": { "effort": "max", "harness": "official or cited leaderboard" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Table 8.1.A / AutomationBench Private Held-Out: Unless noted, Anthropic values use adaptive thinking at max effort, default sampling, mean over five trials. Table 8.1.A card summary." }, { "model_id": "claude-opus-4.8", "benchmark_id": "arc_agi_1", "score": 92.5, "reference_url": "https://www.anthropic.com/claude-opus-5-system-card", "reported_setting": { "effort": "max", "harness": "official or cited leaderboard" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Table 8.1.A / ARC-AGI-1: Unless noted, Anthropic values use adaptive thinking at max effort, default sampling, mean over five trials. Table 8.1.A card summary." }, { "model_id": "gpt-5.6-sol", "benchmark_id": "arc_agi_1", "score": 97.5, "reference_url": "https://www.anthropic.com/claude-opus-5-system-card", "reported_setting": { "effort": "xhigh", "harness": "official or cited leaderboard" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Table 8.1.A / ARC-AGI-1: Unless noted, Anthropic values use adaptive thinking at max effort, default sampling, mean over five trials. Table 8.1.A card summary." }, { "model_id": "claude-opus-4.8", "benchmark_id": "arc_agi_2", "score": 72.1, "reference_url": "https://www.anthropic.com/claude-opus-5-system-card", "reported_setting": { "effort": "max", "harness": "official or cited leaderboard" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Table 8.1.A / ARC-AGI-2: Unless noted, Anthropic values use adaptive thinking at max effort, default sampling, mean over five trials. Table 8.1.A card summary." }, { "model_id": "gpt-5.6-sol", "benchmark_id": "arc_agi_2", "score": 92.5, "reference_url": "https://www.anthropic.com/claude-opus-5-system-card", "reported_setting": { "effort": "max", "harness": "official or cited leaderboard" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Table 8.1.A / ARC-AGI-2: Unless noted, Anthropic values use adaptive thinking at max effort, default sampling, mean over five trials. Table 8.1.A card summary." }, { "model_id": "claude-opus-5", "benchmark_id": "swe_bench_verified", "score": 96.0, "reference_url": "https://www.anthropic.com/claude-opus-5-system-card", "reported_setting": { "effort": "max", "harness": "official" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "SWE-bench Verified narrative: 500 problems; average over five trials." }, { "model_id": "claude-sonnet-5", "benchmark_id": "deep_swe_v1_1", "score": 48.2, "reference_url": "https://www.anthropic.com/claude-opus-5-system-card", "reported_setting": { "effort": "high", "tools": "benchmark-specific" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Figure 8.3.A / Claude Sonnet 5 / high: 113 tasks; five trials.", "candidates": [ { "score": 30.5, "reference_url": "https://www.anthropic.com/claude-opus-5-system-card", "source_type": "model_card", "reported_setting": { "effort": "low", "tools": "benchmark-specific" }, "notes": "Figure 8.3.A / Claude Sonnet 5 / low: 113 tasks; five trials." }, { "score": 39.8, "reference_url": "https://www.anthropic.com/claude-opus-5-system-card", "source_type": "model_card", "reported_setting": { "effort": "medium", "tools": "benchmark-specific" }, "notes": "Figure 8.3.A / Claude Sonnet 5 / medium: 113 tasks; five trials." }, { "score": 49.7, "reference_url": "https://www.anthropic.com/claude-opus-5-system-card", "source_type": "model_card", "reported_setting": { "effort": "xhigh", "tools": "benchmark-specific" }, "notes": "Figure 8.3.A / Claude Sonnet 5 / xhigh: 113 tasks; five trials." }, { "score": 53.8, "reference_url": "https://www.anthropic.com/claude-opus-5-system-card", "source_type": "model_card", "reported_setting": { "effort": "max", "tools": "benchmark-specific" }, "notes": "Figure 8.3.A / Claude Sonnet 5 / max: 113 tasks; five trials." }, { "score": 54.0, "reference_url": "https://deepmind.google/models/model-cards/gemini-3-6-flash/", "source_type": "official_model_card", "reported_setting": { "mode": "thinking/reasoning", "effort": "maximum/best available", "tools": "code execution", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "DataCurve DeepSWE v1.1 leaderboard", "temperature": "default", "snapshot": "July 2026" }, "notes": "Google Gemini 3.6 Flash July 2026 matrix: DeepSWE v1.1. Exact effort, tools, sampling and harness are preserved." }, { "score": 54.0, "reference_url": "https://media.x.ai/v1/website/card-4p6-4cd2dc57.pdf", "source_type": "model_card", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "agentic repository shell/editor", "sampling": "pass@1", "judge": "isolated repository verifier", "harness": "mini-swe-agent run by Datacurve", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown" }, "notes": "card46 DeepSWE v1.1 Pass@1 (%); source effort=max; benchmark protocol matches canonical=True. Unknown source fields remain unknown." } ] }, { "model_id": "claude-opus-5", "benchmark_id": "deep_swe_v1_1", "score": 68.0, "reference_url": "https://www.anthropic.com/claude-opus-5-system-card", "reported_setting": { "effort": "high", "tools": "benchmark-specific" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Figure 8.3.A / Claude Opus 5 / high: 113 tasks; five trials.", "candidates": [ { "score": 57.7, "reference_url": "https://www.anthropic.com/claude-opus-5-system-card", "source_type": "model_card", "reported_setting": { "effort": "low", "tools": "benchmark-specific" }, "notes": "Figure 8.3.A / Claude Opus 5 / low: 113 tasks; five trials." }, { "score": 66.9, "reference_url": "https://www.anthropic.com/claude-opus-5-system-card", "source_type": "model_card", "reported_setting": { "effort": "medium", "tools": "benchmark-specific" }, "notes": "Figure 8.3.A / Claude Opus 5 / medium: 113 tasks; five trials." }, { "score": 69.7, "reference_url": "https://www.anthropic.com/claude-opus-5-system-card", "source_type": "model_card", "reported_setting": { "effort": "xhigh", "tools": "benchmark-specific" }, "notes": "Figure 8.3.A / Claude Opus 5 / xhigh: 113 tasks; five trials." }, { "score": 68.8, "reference_url": "https://www.anthropic.com/claude-opus-5-system-card", "source_type": "model_card", "reported_setting": { "effort": "max", "tools": "benchmark-specific" }, "notes": "Figure 8.3.A / Claude Opus 5 / max: 113 tasks; five trials." }, { "score": 74.0, "reference_url": "https://media.x.ai/v1/website/card-4p6-4cd2dc57.pdf", "source_type": "model_card", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "agentic repository shell/editor", "sampling": "pass@1", "judge": "isolated repository verifier", "harness": "mini-swe-agent run by Datacurve", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown" }, "notes": "card46 DeepSWE v1.1 Pass@1 (%); source effort=max; benchmark protocol matches canonical=True. Unknown source fields remain unknown." }, { "score": 65.0, "reference_url": "https://research.meta.ai/articles/introducing-muse-code-and-muse-spark-1-2/evaluations/deepswe-1-1-v1.png", "source_type": "official_blog", "reported_setting": { "mode": "thinking", "effort": "max", "prompt_style": "not disclosed", "temperature": "not disclosed", "context": "not disclosed", "agent": "Claude Code", "attempts": 5, "tools": "Claude Code repository shell/editor toolset", "sampling": "pass@1 averaged across five attempts per task", "judge": "functional verifier plus regression checks in a pristine verifier container", "harness": "Meta internal agent evaluation framework; Claude Code; isolated Daytona sandbox; Harbor conversion", "internet": "disabled during rollout and grading; model endpoint only", "configuration": "five-language Harbor conversion with pristine verifier" }, "notes": "Exact printed value from the locked Meta release chart; evaluation settings come from the locked official methodology." } ] }, { "model_id": "claude-opus-5", "benchmark_id": "hle_tools", "score": 63.2, "reference_url": "https://www.anthropic.com/claude-opus-5-system-card", "reported_setting": { "effort": "high", "tools": "benchmark-specific" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Figure 8.10.1.A / Claude Opus 5 / high: 2,500 questions; 1M token cap; web/programmatic/code tools; no context compaction.", "candidates": [ { "score": 56.1, "reference_url": "https://www.anthropic.com/claude-opus-5-system-card", "source_type": "model_card", "reported_setting": { "effort": "low", "tools": "benchmark-specific" }, "notes": "Figure 8.10.1.A / Claude Opus 5 / low: 2,500 questions; 1M token cap; web/programmatic/code tools; no context compaction." }, { "score": 61.3, "reference_url": "https://www.anthropic.com/claude-opus-5-system-card", "source_type": "model_card", "reported_setting": { "effort": "medium", "tools": "benchmark-specific" }, "notes": "Figure 8.10.1.A / Claude Opus 5 / medium: 2,500 questions; 1M token cap; web/programmatic/code tools; no context compaction." }, { "score": 64.8, "reference_url": "https://www.anthropic.com/claude-opus-5-system-card", "source_type": "model_card", "reported_setting": { "effort": "xhigh", "tools": "benchmark-specific" }, "notes": "Figure 8.10.1.A / Claude Opus 5 / xhigh: 2,500 questions; 1M token cap; web/programmatic/code tools; no context compaction." }, { "score": 64.7, "reference_url": "https://www.anthropic.com/claude-opus-5-system-card", "source_type": "model_card", "reported_setting": { "effort": "max", "tools": "benchmark-specific" }, "notes": "Figure 8.10.1.A / Claude Opus 5 / max: 2,500 questions; 1M token cap; web/programmatic/code tools; no context compaction." } ] }, { "model_id": "claude-opus-5", "benchmark_id": "hle", "score": 56.0, "reference_url": "https://www.anthropic.com/claude-opus-5-system-card", "reported_setting": { "effort": "high", "tools": "benchmark-specific" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Figure 8.10.1.B / Claude Opus 5 / high: 2,500 questions; 1M token cap; no tools and no context compaction.", "candidates": [ { "score": 47.8, "reference_url": "https://www.anthropic.com/claude-opus-5-system-card", "source_type": "model_card", "reported_setting": { "effort": "low", "tools": "benchmark-specific" }, "notes": "Figure 8.10.1.B / Claude Opus 5 / low: 2,500 questions; 1M token cap; no tools and no context compaction." }, { "score": 54.2, "reference_url": "https://www.anthropic.com/claude-opus-5-system-card", "source_type": "model_card", "reported_setting": { "effort": "medium", "tools": "benchmark-specific" }, "notes": "Figure 8.10.1.B / Claude Opus 5 / medium: 2,500 questions; 1M token cap; no tools and no context compaction." }, { "score": 56.4, "reference_url": "https://www.anthropic.com/claude-opus-5-system-card", "source_type": "model_card", "reported_setting": { "effort": "xhigh", "tools": "benchmark-specific" }, "notes": "Figure 8.10.1.B / Claude Opus 5 / xhigh: 2,500 questions; 1M token cap; no tools and no context compaction." }, { "score": 56.3, "reference_url": "https://www.anthropic.com/claude-opus-5-system-card", "source_type": "model_card", "reported_setting": { "effort": "max", "tools": "benchmark-specific" }, "notes": "Figure 8.10.1.B / Claude Opus 5 / max: 2,500 questions; 1M token cap; no tools and no context compaction." } ] }, { "model_id": "claude-opus-5", "benchmark_id": "browsecomp", "score": 90.2, "reference_url": "https://www.anthropic.com/claude-opus-5-system-card", "reported_setting": { "effort": "high", "tools": "benchmark-specific" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Figure 8.10.2.B / Claude Opus 5 / high: Fixed 10M token budget with web/programmatic/code tools and context compaction.", "candidates": [ { "score": 84.4, "reference_url": "https://www.anthropic.com/claude-opus-5-system-card", "source_type": "model_card", "reported_setting": { "effort": "low", "tools": "benchmark-specific" }, "notes": "Figure 8.10.2.B / Claude Opus 5 / low: Fixed 10M token budget with web/programmatic/code tools and context compaction." }, { "score": 88.9, "reference_url": "https://www.anthropic.com/claude-opus-5-system-card", "source_type": "model_card", "reported_setting": { "effort": "medium", "tools": "benchmark-specific" }, "notes": "Figure 8.10.2.B / Claude Opus 5 / medium: Fixed 10M token budget with web/programmatic/code tools and context compaction." }, { "score": 90.7, "reference_url": "https://www.anthropic.com/claude-opus-5-system-card", "source_type": "model_card", "reported_setting": { "effort": "xhigh", "tools": "benchmark-specific" }, "notes": "Figure 8.10.2.B / Claude Opus 5 / xhigh: Fixed 10M token budget with web/programmatic/code tools and context compaction." }, { "score": 90.8, "reference_url": "https://www.anthropic.com/claude-opus-5-system-card", "source_type": "model_card", "reported_setting": { "effort": "max", "tools": "benchmark-specific" }, "notes": "Figure 8.10.2.B / Claude Opus 5 / max: Fixed 10M token budget with web/programmatic/code tools and context compaction." }, { "score": 86.8, "reference_url": "https://www.anthropic.com/claude-opus-5-system-card", "source_type": "model_card", "reported_setting": { "effort": "max", "tools": "web/programmatic/code", "harness": "token-budget scaling" }, "notes": "Figure 8.10.2.A / Claude Opus 5 / 1M: Token-budget scaling; source reports exact labels at 1M, 3M, and 10M. Budget variants are candidates; fixed 10M effort ladder supplies the preferred canonical-setting value." }, { "score": 89.8, "reference_url": "https://www.anthropic.com/claude-opus-5-system-card", "source_type": "model_card", "reported_setting": { "effort": "max", "tools": "web/programmatic/code", "harness": "token-budget scaling" }, "notes": "Figure 8.10.2.A / Claude Opus 5 / 3M: Token-budget scaling; source reports exact labels at 1M, 3M, and 10M. Budget variants are candidates; fixed 10M effort ladder supplies the preferred canonical-setting value." }, { "score": 90.2, "reference_url": "https://www.anthropic.com/claude-opus-5-system-card", "source_type": "model_card", "reported_setting": { "effort": "max", "tools": "web/programmatic/code", "harness": "token-budget scaling" }, "notes": "Figure 8.10.2.A / Claude Opus 5 / 10M: Token-budget scaling; source reports exact labels at 1M, 3M, and 10M. Budget variants are candidates; fixed 10M effort ladder supplies the preferred canonical-setting value." } ] }, { "model_id": "claude-sonnet-5", "benchmark_id": "deepsearchqa_f1", "score": 89.4, "reference_url": "https://www.anthropic.com/claude-opus-5-system-card", "reported_setting": { "effort": "high", "tools": "benchmark-specific" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Figure 8.10.3.A / Claude Sonnet 5 / high: Mean F1 over 900 prompts in 17 fields; web tools; nominal 1M token budget.", "candidates": [ { "score": 83.9, "reference_url": "https://www.anthropic.com/claude-opus-5-system-card", "source_type": "model_card", "reported_setting": { "effort": "low", "tools": "benchmark-specific" }, "notes": "Figure 8.10.3.A / Claude Sonnet 5 / low: Mean F1 over 900 prompts in 17 fields; web tools; nominal 1M token budget." }, { "score": 87.8, "reference_url": "https://www.anthropic.com/claude-opus-5-system-card", "source_type": "model_card", "reported_setting": { "effort": "medium", "tools": "benchmark-specific" }, "notes": "Figure 8.10.3.A / Claude Sonnet 5 / medium: Mean F1 over 900 prompts in 17 fields; web tools; nominal 1M token budget." }, { "score": 90.4, "reference_url": "https://www.anthropic.com/claude-opus-5-system-card", "source_type": "model_card", "reported_setting": { "effort": "xhigh", "tools": "benchmark-specific" }, "notes": "Figure 8.10.3.A / Claude Sonnet 5 / xhigh: Mean F1 over 900 prompts in 17 fields; web tools; nominal 1M token budget." }, { "score": 92.4, "reference_url": "https://www.anthropic.com/claude-opus-5-system-card", "source_type": "model_card", "reported_setting": { "effort": "max", "tools": "benchmark-specific" }, "notes": "Figure 8.10.3.A / Claude Sonnet 5 / max: Mean F1 over 900 prompts in 17 fields; web tools; nominal 1M token budget." } ] }, { "model_id": "claude-opus-5", "benchmark_id": "deepsearchqa_f1", "score": 94.1, "reference_url": "https://www.anthropic.com/claude-opus-5-system-card", "reported_setting": { "effort": "high", "tools": "benchmark-specific" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Figure 8.10.3.A / Claude Opus 5 / high: Mean F1 over 900 prompts in 17 fields; web tools; nominal 1M token budget.", "candidates": [ { "score": 91.0, "reference_url": "https://www.anthropic.com/claude-opus-5-system-card", "source_type": "model_card", "reported_setting": { "effort": "low", "tools": "benchmark-specific" }, "notes": "Figure 8.10.3.A / Claude Opus 5 / low: Mean F1 over 900 prompts in 17 fields; web tools; nominal 1M token budget." }, { "score": 92.8, "reference_url": "https://www.anthropic.com/claude-opus-5-system-card", "source_type": "model_card", "reported_setting": { "effort": "medium", "tools": "benchmark-specific" }, "notes": "Figure 8.10.3.A / Claude Opus 5 / medium: Mean F1 over 900 prompts in 17 fields; web tools; nominal 1M token budget." }, { "score": 94.6, "reference_url": "https://www.anthropic.com/claude-opus-5-system-card", "source_type": "model_card", "reported_setting": { "effort": "xhigh", "tools": "benchmark-specific" }, "notes": "Figure 8.10.3.A / Claude Opus 5 / xhigh: Mean F1 over 900 prompts in 17 fields; web tools; nominal 1M token budget." }, { "score": 95.0, "reference_url": "https://www.anthropic.com/claude-opus-5-system-card", "source_type": "model_card", "reported_setting": { "effort": "max", "tools": "benchmark-specific" }, "notes": "Figure 8.10.3.A / Claude Opus 5 / max: Mean F1 over 900 prompts in 17 fields; web tools; nominal 1M token budget." } ] }, { "model_id": "claude-opus-4.8", "benchmark_id": "draco", "score": 76.3, "reference_url": "https://www.anthropic.com/claude-opus-5-system-card", "reported_setting": { "effort": "high", "tools": "benchmark-specific" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Figure 8.10.4.A / Claude Opus 4.8 / high: 100 curated agentic data-analysis tasks; Opus 4.6 judge; five grading runs.", "candidates": [ { "score": 70.8, "reference_url": "https://www.anthropic.com/claude-opus-5-system-card", "source_type": "model_card", "reported_setting": { "effort": "low", "tools": "benchmark-specific" }, "notes": "Figure 8.10.4.A / Claude Opus 4.8 / low: 100 curated agentic data-analysis tasks; Opus 4.6 judge; five grading runs." }, { "score": 74.6, "reference_url": "https://www.anthropic.com/claude-opus-5-system-card", "source_type": "model_card", "reported_setting": { "effort": "medium", "tools": "benchmark-specific" }, "notes": "Figure 8.10.4.A / Claude Opus 4.8 / medium: 100 curated agentic data-analysis tasks; Opus 4.6 judge; five grading runs." }, { "score": 79.4, "reference_url": "https://www.anthropic.com/claude-opus-5-system-card", "source_type": "model_card", "reported_setting": { "effort": "xhigh", "tools": "benchmark-specific" }, "notes": "Figure 8.10.4.A / Claude Opus 4.8 / xhigh: 100 curated agentic data-analysis tasks; Opus 4.6 judge; five grading runs." }, { "score": 80.6, "reference_url": "https://www.anthropic.com/claude-opus-5-system-card", "source_type": "model_card", "reported_setting": { "effort": "max", "tools": "benchmark-specific" }, "notes": "Figure 8.10.4.A / Claude Opus 4.8 / max: 100 curated agentic data-analysis tasks; Opus 4.6 judge; five grading runs." } ] }, { "model_id": "claude-fable-5", "benchmark_id": "draco", "score": 87.7, "reference_url": "https://www.anthropic.com/claude-opus-5-system-card", "reported_setting": { "effort": "max", "tools": "benchmark-specific" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Figure 8.10.4.A / Claude Fable 5 / max: 100 curated agentic data-analysis tasks; Opus 4.6 judge; five grading runs.", "candidates": [ { "score": 78.3, "reference_url": "https://www.anthropic.com/claude-opus-5-system-card", "source_type": "model_card", "reported_setting": { "effort": "low", "tools": "benchmark-specific" }, "notes": "Figure 8.10.4.A / Claude Fable 5 / low: 100 curated agentic data-analysis tasks; Opus 4.6 judge; five grading runs." }, { "score": 81.0, "reference_url": "https://www.anthropic.com/claude-opus-5-system-card", "source_type": "model_card", "reported_setting": { "effort": "medium", "tools": "benchmark-specific" }, "notes": "Figure 8.10.4.A / Claude Fable 5 / medium: 100 curated agentic data-analysis tasks; Opus 4.6 judge; five grading runs." }, { "score": 83.5, "reference_url": "https://www.anthropic.com/claude-opus-5-system-card", "source_type": "model_card", "reported_setting": { "effort": "high", "tools": "benchmark-specific" }, "notes": "Figure 8.10.4.A / Claude Fable 5 / high: 100 curated agentic data-analysis tasks; Opus 4.6 judge; five grading runs." }, { "score": 86.3, "reference_url": "https://www.anthropic.com/claude-opus-5-system-card", "source_type": "model_card", "reported_setting": { "effort": "xhigh", "tools": "benchmark-specific" }, "notes": "Figure 8.10.4.A / Claude Fable 5 / xhigh: 100 curated agentic data-analysis tasks; Opus 4.6 judge; five grading runs." } ] }, { "model_id": "claude-sonnet-5", "benchmark_id": "draco", "score": 76.4, "reference_url": "https://www.anthropic.com/claude-opus-5-system-card", "reported_setting": { "effort": "high", "tools": "benchmark-specific" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Figure 8.10.4.A / Claude Sonnet 5 / high: 100 curated agentic data-analysis tasks; Opus 4.6 judge; five grading runs.", "candidates": [ { "score": 66.8, "reference_url": "https://www.anthropic.com/claude-opus-5-system-card", "source_type": "model_card", "reported_setting": { "effort": "low", "tools": "benchmark-specific" }, "notes": "Figure 8.10.4.A / Claude Sonnet 5 / low: 100 curated agentic data-analysis tasks; Opus 4.6 judge; five grading runs." }, { "score": 72.7, "reference_url": "https://www.anthropic.com/claude-opus-5-system-card", "source_type": "model_card", "reported_setting": { "effort": "medium", "tools": "benchmark-specific" }, "notes": "Figure 8.10.4.A / Claude Sonnet 5 / medium: 100 curated agentic data-analysis tasks; Opus 4.6 judge; five grading runs." }, { "score": 81.6, "reference_url": "https://www.anthropic.com/claude-opus-5-system-card", "source_type": "model_card", "reported_setting": { "effort": "xhigh", "tools": "benchmark-specific" }, "notes": "Figure 8.10.4.A / Claude Sonnet 5 / xhigh: 100 curated agentic data-analysis tasks; Opus 4.6 judge; five grading runs." }, { "score": 84.3, "reference_url": "https://www.anthropic.com/claude-opus-5-system-card", "source_type": "model_card", "reported_setting": { "effort": "max", "tools": "benchmark-specific" }, "notes": "Figure 8.10.4.A / Claude Sonnet 5 / max: 100 curated agentic data-analysis tasks; Opus 4.6 judge; five grading runs." } ] }, { "model_id": "claude-opus-5", "benchmark_id": "draco", "score": 87.0, "reference_url": "https://www.anthropic.com/claude-opus-5-system-card", "reported_setting": { "effort": "high", "tools": "benchmark-specific" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Figure 8.10.4.A / Claude Opus 5 / high: 100 curated agentic data-analysis tasks; Opus 4.6 judge; five grading runs.", "candidates": [ { "score": 83.2, "reference_url": "https://www.anthropic.com/claude-opus-5-system-card", "source_type": "model_card", "reported_setting": { "effort": "low", "tools": "benchmark-specific" }, "notes": "Figure 8.10.4.A / Claude Opus 5 / low: 100 curated agentic data-analysis tasks; Opus 4.6 judge; five grading runs." }, { "score": 86.8, "reference_url": "https://www.anthropic.com/claude-opus-5-system-card", "source_type": "model_card", "reported_setting": { "effort": "medium", "tools": "benchmark-specific" }, "notes": "Figure 8.10.4.A / Claude Opus 5 / medium: 100 curated agentic data-analysis tasks; Opus 4.6 judge; five grading runs." }, { "score": 88.1, "reference_url": "https://www.anthropic.com/claude-opus-5-system-card", "source_type": "model_card", "reported_setting": { "effort": "xhigh", "tools": "benchmark-specific" }, "notes": "Figure 8.10.4.A / Claude Opus 5 / xhigh: 100 curated agentic data-analysis tasks; Opus 4.6 judge; five grading runs." }, { "score": 88.6, "reference_url": "https://www.anthropic.com/claude-opus-5-system-card", "source_type": "model_card", "reported_setting": { "effort": "max", "tools": "benchmark-specific" }, "notes": "Figure 8.10.4.A / Claude Opus 5 / max: 100 curated agentic data-analysis tasks; Opus 4.6 judge; five grading runs." } ] }, { "model_id": "gpt-5.6-sol", "benchmark_id": "frontiercode_main_v1_1", "score": 47.5, "reference_url": "https://www.anthropic.com/claude-opus-5-system-card", "reported_setting": { "effort": "max", "harness": "Cognition" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Figure 8.4.A / GPT 5.6 Sol / max: Five effort settings; score is averaged over five trials.", "candidates": [ { "score": 35.4, "reference_url": "https://www.anthropic.com/claude-opus-5-system-card", "source_type": "model_card", "reported_setting": { "effort": "low", "harness": "Cognition" }, "notes": "Figure 8.4.A / GPT 5.6 Sol / low: Five effort settings; score is averaged over five trials." }, { "score": 39.9, "reference_url": "https://www.anthropic.com/claude-opus-5-system-card", "source_type": "model_card", "reported_setting": { "effort": "medium", "harness": "Cognition" }, "notes": "Figure 8.4.A / GPT 5.6 Sol / medium: Five effort settings; score is averaged over five trials." }, { "score": 45.1, "reference_url": "https://www.anthropic.com/claude-opus-5-system-card", "source_type": "model_card", "reported_setting": { "effort": "high", "harness": "Cognition" }, "notes": "Figure 8.4.A / GPT 5.6 Sol / high: Five effort settings; score is averaged over five trials." }, { "score": 46.8, "reference_url": "https://www.anthropic.com/claude-opus-5-system-card", "source_type": "model_card", "reported_setting": { "effort": "xhigh", "harness": "Cognition" }, "notes": "Figure 8.4.A / GPT 5.6 Sol / xhigh: Five effort settings; score is averaged over five trials." } ] }, { "model_id": "claude-opus-4.8", "benchmark_id": "frontiercode_main_v1_1", "score": 41.0, "reference_url": "https://www.anthropic.com/claude-opus-5-system-card", "reported_setting": { "effort": "high", "harness": "Cognition" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Figure 8.4.A / Claude Opus 4.8 / high: Five effort settings; score is averaged over five trials.", "candidates": [ { "score": 35.1, "reference_url": "https://www.anthropic.com/claude-opus-5-system-card", "source_type": "model_card", "reported_setting": { "effort": "low", "harness": "Cognition" }, "notes": "Figure 8.4.A / Claude Opus 4.8 / low: Five effort settings; score is averaged over five trials." }, { "score": 40.5, "reference_url": "https://www.anthropic.com/claude-opus-5-system-card", "source_type": "model_card", "reported_setting": { "effort": "medium", "harness": "Cognition" }, "notes": "Figure 8.4.A / Claude Opus 4.8 / medium: Five effort settings; score is averaged over five trials." }, { "score": 45.5, "reference_url": "https://www.anthropic.com/claude-opus-5-system-card", "source_type": "model_card", "reported_setting": { "effort": "xhigh", "harness": "Cognition" }, "notes": "Figure 8.4.A / Claude Opus 4.8 / xhigh: Five effort settings; score is averaged over five trials." }, { "score": 46.5, "reference_url": "https://www.anthropic.com/claude-opus-5-system-card", "source_type": "model_card", "reported_setting": { "effort": "max", "harness": "Cognition" }, "notes": "Figure 8.4.A / Claude Opus 4.8 / max: Five effort settings; score is averaged over five trials." } ] }, { "model_id": "claude-fable-5", "benchmark_id": "frontiercode_main_v1_1", "score": 51.6, "reference_url": "https://www.anthropic.com/claude-opus-5-system-card", "reported_setting": { "effort": "max", "harness": "Cognition" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Figure 8.4.A / Claude Fable 5 / max: Five effort settings; score is averaged over five trials.", "candidates": [ { "score": 48.0, "reference_url": "https://www.anthropic.com/claude-opus-5-system-card", "source_type": "model_card", "reported_setting": { "effort": "low", "harness": "Cognition" }, "notes": "Figure 8.4.A / Claude Fable 5 / low: Five effort settings; score is averaged over five trials." }, { "score": 49.8, "reference_url": "https://www.anthropic.com/claude-opus-5-system-card", "source_type": "model_card", "reported_setting": { "effort": "medium", "harness": "Cognition" }, "notes": "Figure 8.4.A / Claude Fable 5 / medium: Five effort settings; score is averaged over five trials." }, { "score": 52.7, "reference_url": "https://www.anthropic.com/claude-opus-5-system-card", "source_type": "model_card", "reported_setting": { "effort": "high", "harness": "Cognition" }, "notes": "Figure 8.4.A / Claude Fable 5 / high: Five effort settings; score is averaged over five trials." }, { "score": 53.5, "reference_url": "https://www.anthropic.com/claude-opus-5-system-card", "source_type": "model_card", "reported_setting": { "effort": "xhigh", "harness": "Cognition" }, "notes": "Figure 8.4.A / Claude Fable 5 / xhigh: Five effort settings; score is averaged over five trials." } ] }, { "model_id": "claude-opus-5", "benchmark_id": "frontiercode_main_v1_1", "score": 48.0, "reference_url": "https://www.anthropic.com/claude-opus-5-system-card", "reported_setting": { "effort": "high", "harness": "Cognition" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Figure 8.4.A / Claude Opus 5 / high: Five effort settings; score is averaged over five trials.", "candidates": [ { "score": 41.9, "reference_url": "https://www.anthropic.com/claude-opus-5-system-card", "source_type": "model_card", "reported_setting": { "effort": "low", "harness": "Cognition" }, "notes": "Figure 8.4.A / Claude Opus 5 / low: Five effort settings; score is averaged over five trials." }, { "score": 53.4, "reference_url": "https://www.anthropic.com/claude-opus-5-system-card", "source_type": "model_card", "reported_setting": { "effort": "medium", "harness": "Cognition" }, "notes": "Figure 8.4.A / Claude Opus 5 / medium: Five effort settings; score is averaged over five trials." }, { "score": 43.6, "reference_url": "https://www.anthropic.com/claude-opus-5-system-card", "source_type": "model_card", "reported_setting": { "effort": "xhigh", "harness": "Cognition" }, "notes": "Figure 8.4.A / Claude Opus 5 / xhigh: Five effort settings; score is averaged over five trials." }, { "score": 48.0, "reference_url": "https://www.anthropic.com/claude-opus-5-system-card", "source_type": "model_card", "reported_setting": { "effort": "max", "harness": "Cognition" }, "notes": "Figure 8.4.A / Claude Opus 5 / max: Five effort settings; score is averaged over five trials." } ] }, { "model_id": "gpt-5.6-sol", "benchmark_id": "frontiercode_extended_v1_1", "score": 60.6, "reference_url": "https://www.anthropic.com/claude-opus-5-system-card", "reported_setting": { "effort": "max", "harness": "Cognition" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Figure 8.4.B / GPT 5.6 Sol / max: Five effort settings; score is averaged over five trials.", "candidates": [ { "score": 50.0, "reference_url": "https://www.anthropic.com/claude-opus-5-system-card", "source_type": "model_card", "reported_setting": { "effort": "low", "harness": "Cognition" }, "notes": "Figure 8.4.B / GPT 5.6 Sol / low: Five effort settings; score is averaged over five trials." }, { "score": 54.7, "reference_url": "https://www.anthropic.com/claude-opus-5-system-card", "source_type": "model_card", "reported_setting": { "effort": "medium", "harness": "Cognition" }, "notes": "Figure 8.4.B / GPT 5.6 Sol / medium: Five effort settings; score is averaged over five trials." }, { "score": 58.7, "reference_url": "https://www.anthropic.com/claude-opus-5-system-card", "source_type": "model_card", "reported_setting": { "effort": "high", "harness": "Cognition" }, "notes": "Figure 8.4.B / GPT 5.6 Sol / high: Five effort settings; score is averaged over five trials." }, { "score": 60.0, "reference_url": "https://www.anthropic.com/claude-opus-5-system-card", "source_type": "model_card", "reported_setting": { "effort": "xhigh", "harness": "Cognition" }, "notes": "Figure 8.4.B / GPT 5.6 Sol / xhigh: Five effort settings; score is averaged over five trials." }, { "score": 60.6, "reference_url": "https://x.ai/news/grok-4-6", "source_type": "official_blog", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "fixed coding-agent scaffold", "sampling": "unknown", "judge": "maintainer-authored rubrics, verifiers, and hard blockers", "harness": "FrontierCode v1.1 Extended harness", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown" }, "notes": "release46 FrontierCode v1.1 Extended Extended score (%); source effort=max; benchmark protocol matches canonical=True. Unknown source fields remain unknown." } ] }, { "model_id": "claude-opus-4.8", "benchmark_id": "frontiercode_extended_v1_1", "score": 55.6, "reference_url": "https://www.anthropic.com/claude-opus-5-system-card", "reported_setting": { "effort": "high", "harness": "Cognition" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Figure 8.4.B / Claude Opus 4.8 / high: Five effort settings; score is averaged over five trials.", "candidates": [ { "score": 50.8, "reference_url": "https://www.anthropic.com/claude-opus-5-system-card", "source_type": "model_card", "reported_setting": { "effort": "low", "harness": "Cognition" }, "notes": "Figure 8.4.B / Claude Opus 4.8 / low: Five effort settings; score is averaged over five trials." }, { "score": 55.4, "reference_url": "https://www.anthropic.com/claude-opus-5-system-card", "source_type": "model_card", "reported_setting": { "effort": "medium", "harness": "Cognition" }, "notes": "Figure 8.4.B / Claude Opus 4.8 / medium: Five effort settings; score is averaged over five trials." }, { "score": 59.5, "reference_url": "https://www.anthropic.com/claude-opus-5-system-card", "source_type": "model_card", "reported_setting": { "effort": "xhigh", "harness": "Cognition" }, "notes": "Figure 8.4.B / Claude Opus 4.8 / xhigh: Five effort settings; score is averaged over five trials." }, { "score": 59.6, "reference_url": "https://www.anthropic.com/claude-opus-5-system-card", "source_type": "model_card", "reported_setting": { "effort": "max", "harness": "Cognition" }, "notes": "Figure 8.4.B / Claude Opus 4.8 / max: Five effort settings; score is averaged over five trials." }, { "score": 59.6, "reference_url": "https://media.x.ai/v1/website/card-4p6-4cd2dc57.pdf", "source_type": "model_card", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "fixed coding-agent scaffold", "sampling": "unknown", "judge": "maintainer-authored rubrics, verifiers, and hard blockers", "harness": "FrontierCode v1.1 Extended harness", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown" }, "notes": "card46 FrontierCode v1.1 Extended Extended score (%); source effort=max; benchmark protocol matches canonical=True. Unknown source fields remain unknown." } ] }, { "model_id": "claude-fable-5", "benchmark_id": "frontiercode_extended_v1_1", "score": 64.9, "reference_url": "https://media.x.ai/v1/website/card-4p6-4cd2dc57.pdf", "reported_setting": { "mode": "thinking", "effort": "max, with fallback", "tools": "fixed coding-agent scaffold", "sampling": "unknown", "judge": "maintainer-authored rubrics, verifiers, and hard blockers", "harness": "FrontierCode v1.1 Extended harness", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "card46 FrontierCode v1.1 Extended Extended score (%); source effort=max, with fallback; benchmark protocol matches canonical=True. Unknown source fields remain unknown.", "candidates": [ { "score": 60.8, "reference_url": "https://www.anthropic.com/claude-opus-5-system-card", "source_type": "model_card", "reported_setting": { "effort": "low", "harness": "Cognition" }, "notes": "Figure 8.4.B / Claude Fable 5 / low: Five effort settings; score is averaged over five trials." }, { "score": 62.8, "reference_url": "https://www.anthropic.com/claude-opus-5-system-card", "source_type": "model_card", "reported_setting": { "effort": "medium", "harness": "Cognition" }, "notes": "Figure 8.4.B / Claude Fable 5 / medium: Five effort settings; score is averaged over five trials." }, { "score": 64.2, "reference_url": "https://www.anthropic.com/claude-opus-5-system-card", "source_type": "model_card", "reported_setting": { "effort": "high", "harness": "Cognition" }, "notes": "Figure 8.4.B / Claude Fable 5 / high: Five effort settings; score is averaged over five trials." }, { "score": 64.9, "reference_url": "https://www.anthropic.com/claude-opus-5-system-card", "source_type": "model_card", "reported_setting": { "effort": "xhigh", "harness": "Cognition" }, "notes": "Figure 8.4.B / Claude Fable 5 / xhigh: Five effort settings; score is averaged over five trials." }, { "score": 63.6, "reference_url": "https://www.anthropic.com/claude-opus-5-system-card", "source_type": "model_card", "reported_setting": { "effort": "max", "harness": "Cognition" }, "notes": "Figure 8.4.B / Claude Fable 5 / max: Five effort settings; score is averaged over five trials." }, { "score": 63.6, "reference_url": "https://x.ai/news/grok-4-6", "source_type": "official_blog", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "fixed coding-agent scaffold", "sampling": "unknown", "judge": "maintainer-authored rubrics, verifiers, and hard blockers", "harness": "FrontierCode v1.1 Extended harness", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown" }, "notes": "release46 FrontierCode v1.1 Extended Extended score (%); source effort=max; benchmark protocol matches canonical=True. Unknown source fields remain unknown." } ] }, { "model_id": "claude-opus-5", "benchmark_id": "frontiercode_extended_v1_1", "score": 58.5, "reference_url": "https://www.anthropic.com/claude-opus-5-system-card", "reported_setting": { "effort": "high", "harness": "Cognition" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Figure 8.4.B / Claude Opus 5 / high: Five effort settings; score is averaged over five trials.", "candidates": [ { "score": 55.8, "reference_url": "https://www.anthropic.com/claude-opus-5-system-card", "source_type": "model_card", "reported_setting": { "effort": "low", "harness": "Cognition" }, "notes": "Figure 8.4.B / Claude Opus 5 / low: Five effort settings; score is averaged over five trials." }, { "score": 63.6, "reference_url": "https://www.anthropic.com/claude-opus-5-system-card", "source_type": "model_card", "reported_setting": { "effort": "medium", "harness": "Cognition" }, "notes": "Figure 8.4.B / Claude Opus 5 / medium: Five effort settings; score is averaged over five trials." }, { "score": 56.9, "reference_url": "https://www.anthropic.com/claude-opus-5-system-card", "source_type": "model_card", "reported_setting": { "effort": "xhigh", "harness": "Cognition" }, "notes": "Figure 8.4.B / Claude Opus 5 / xhigh: Five effort settings; score is averaged over five trials." }, { "score": 58.9, "reference_url": "https://www.anthropic.com/claude-opus-5-system-card", "source_type": "model_card", "reported_setting": { "effort": "max", "harness": "Cognition" }, "notes": "Figure 8.4.B / Claude Opus 5 / max: Five effort settings; score is averaged over five trials." }, { "score": 63.6, "reference_url": "https://media.x.ai/v1/website/card-4p6-4cd2dc57.pdf", "source_type": "model_card", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "fixed coding-agent scaffold", "sampling": "unknown", "judge": "maintainer-authored rubrics, verifiers, and hard blockers", "harness": "FrontierCode v1.1 Extended harness", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown" }, "notes": "card46 FrontierCode v1.1 Extended Extended score (%); source effort=max; benchmark protocol matches canonical=True. Unknown source fields remain unknown." } ] }, { "model_id": "claude-opus-5", "benchmark_id": "frontierbench_v0_1", "score": 39.0, "reference_url": "https://www.anthropic.com/claude-opus-5-system-card", "reported_setting": { "effort": "high", "harness": "mini-SWE-agent on GKE" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "FrontierBench narrative / Claude Opus 5 / high: 74 tasks; five attempts per task.", "candidates": [ { "score": 43.3, "reference_url": "https://www.anthropic.com/claude-opus-5-system-card", "source_type": "model_card", "reported_setting": { "effort": "source does not state", "harness": "Harbor leaderboard" }, "notes": "Table 8.1.A / FrontierBench v0.1: Unless noted, Anthropic values use adaptive thinking at max effort, default sampling, mean over five trials. Table 8.1.A card summary." }, { "score": 25.0, "reference_url": "https://www.anthropic.com/claude-opus-5-system-card", "source_type": "model_card", "reported_setting": { "effort": "low", "harness": "mini-SWE-agent on GKE" }, "notes": "FrontierBench narrative / Claude Opus 5 / low: 74 tasks; five attempts per task." }, { "score": 44.4, "reference_url": "https://www.anthropic.com/claude-opus-5-system-card", "source_type": "model_card", "reported_setting": { "effort": "xhigh", "harness": "mini-SWE-agent on GKE" }, "notes": "FrontierBench narrative / Claude Opus 5 / xhigh: 74 tasks; five attempts per task." }, { "score": 43.0, "reference_url": "https://www.anthropic.com/claude-opus-5-system-card", "source_type": "model_card", "reported_setting": { "effort": "max", "harness": "mini-SWE-agent on GKE" }, "notes": "FrontierBench narrative / Claude Opus 5 / max: 74 tasks; five attempts per task." } ] }, { "model_id": "claude-fable-5", "benchmark_id": "frontierbench_v0_1", "score": 33.7, "reference_url": "https://www.anthropic.com/claude-opus-5-system-card", "reported_setting": { "effort": "max", "harness": "mini-SWE-agent on GKE" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "FrontierBench narrative / Claude Fable 5: 74 tasks; five attempts per task.", "candidates": [ { "score": 33.8, "reference_url": "https://www.anthropic.com/claude-opus-5-system-card", "source_type": "model_card", "reported_setting": { "effort": "source does not state", "harness": "Harbor leaderboard" }, "notes": "Table 8.1.A / FrontierBench v0.1: Unless noted, Anthropic values use adaptive thinking at max effort, default sampling, mean over five trials. Table 8.1.A card summary." } ] }, { "model_id": "claude-sonnet-5", "benchmark_id": "frontierbench_v0_1", "score": 17.0, "reference_url": "https://www.anthropic.com/claude-opus-5-system-card", "reported_setting": { "effort": "max", "harness": "mini-SWE-agent on GKE" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "FrontierBench narrative / Claude Sonnet 5: 74 tasks; five attempts per task." }, { "model_id": "claude-opus-4.8", "benchmark_id": "frontierbench_v0_1", "score": 18.7, "reference_url": "https://www.anthropic.com/claude-opus-5-system-card", "reported_setting": { "effort": "max", "harness": "mini-SWE-agent on GKE" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "FrontierBench narrative / Claude Opus 4.8: 74 tasks; five attempts per task.", "candidates": [ { "score": 21.1, "reference_url": "https://www.anthropic.com/claude-opus-5-system-card", "source_type": "model_card", "reported_setting": { "effort": "source does not state", "harness": "Harbor leaderboard" }, "notes": "Table 8.1.A / FrontierBench v0.1: Unless noted, Anthropic values use adaptive thinking at max effort, default sampling, mean over five trials. Table 8.1.A card summary." } ] }, { "model_id": "gpt-5.6-sol", "benchmark_id": "frontierbench_v0_1", "score": 37.5, "reference_url": "https://www.anthropic.com/claude-opus-5-system-card", "reported_setting": { "effort": "max", "harness": "mini-SWE-agent on GKE" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "FrontierBench narrative / GPT 5.6 Sol: 74 tasks; five attempts per task.", "candidates": [ { "score": 34.4, "reference_url": "https://www.anthropic.com/claude-opus-5-system-card", "source_type": "model_card", "reported_setting": { "effort": "source does not state", "harness": "Harbor leaderboard" }, "notes": "Table 8.1.A / FrontierBench v0.1: Unless noted, Anthropic values use adaptive thinking at max effort, default sampling, mean over five trials. Table 8.1.A card summary." } ] }, { "model_id": "claude-mythos", "benchmark_id": "riemannbench_no_tools", "score": 53, "reference_url": "https://www.anthropic.com/claude-opus-5-system-card", "reported_setting": { "effort": "max", "tools": "none" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Figure 8.7.A / Claude Mythos Preview / no tools: Private 25-problem benchmark; corrected references/grading; four attempts per problem." }, { "model_id": "claude-mythos", "benchmark_id": "riemannbench_with_tools", "score": 67, "reference_url": "https://www.anthropic.com/claude-opus-5-system-card", "reported_setting": { "effort": "max", "tools": "benchmark-specified" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Figure 8.7.A / Claude Mythos Preview / with tools: Private 25-problem benchmark; corrected references/grading; four attempts per problem." }, { "model_id": "claude-opus-4.8", "benchmark_id": "riemannbench_no_tools", "score": 48, "reference_url": "https://www.anthropic.com/claude-opus-5-system-card", "reported_setting": { "effort": "max", "tools": "none" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Figure 8.7.A / Claude Opus 4.8 / no tools: Private 25-problem benchmark; corrected references/grading; four attempts per problem." }, { "model_id": "claude-opus-4.8", "benchmark_id": "riemannbench_with_tools", "score": 66, "reference_url": "https://www.anthropic.com/claude-opus-5-system-card", "reported_setting": { "effort": "max", "tools": "benchmark-specified" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Figure 8.7.A / Claude Opus 4.8 / with tools: Private 25-problem benchmark; corrected references/grading; four attempts per problem." }, { "model_id": "claude-opus-5", "benchmark_id": "riemannbench_no_tools", "score": 60, "reference_url": "https://www.anthropic.com/claude-opus-5-system-card", "reported_setting": { "effort": "max", "tools": "none" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Figure 8.7.A / Claude Opus 5 / no tools: Private 25-problem benchmark; corrected references/grading; four attempts per problem." }, { "model_id": "claude-opus-5", "benchmark_id": "riemannbench_with_tools", "score": 79, "reference_url": "https://www.anthropic.com/claude-opus-5-system-card", "reported_setting": { "effort": "max", "tools": "benchmark-specified" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Figure 8.7.A / Claude Opus 5 / with tools: Private 25-problem benchmark; corrected references/grading; four attempts per problem." }, { "model_id": "grok-4.5", "benchmark_id": "arxivmath_2026_06_no_tools", "score": 65.99, "reference_url": "https://www.anthropic.com/claude-opus-5-system-card", "reported_setting": { "effort": "source does not state", "tools": "none" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Figure 8.8.A / Grok 4.5 / no tools: 49 June 2026 problems; Anthropic internal values average four runs." }, { "model_id": "gemini-3.1-pro", "benchmark_id": "arxivmath_2026_06_no_tools", "score": 65.99, "reference_url": "https://www.anthropic.com/claude-opus-5-system-card", "reported_setting": { "effort": "source does not state", "tools": "none" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Figure 8.8.A / Gemini 3.1 Pro Preview / no tools: 49 June 2026 problems; Anthropic internal values average four runs." }, { "model_id": "gpt-5.6-sol", "benchmark_id": "arxivmath_2026_06_no_tools", "score": 86.73, "reference_url": "https://www.anthropic.com/claude-opus-5-system-card", "reported_setting": { "effort": "source does not state", "tools": "none" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Figure 8.8.A / GPT-5.6-Sol / no tools: 49 June 2026 problems; Anthropic internal values average four runs." }, { "model_id": "claude-opus-4.8", "benchmark_id": "arxivmath_2026_06_no_tools", "score": 82.65, "reference_url": "https://www.anthropic.com/claude-opus-5-system-card", "reported_setting": { "effort": "max", "tools": "none" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Figure 8.8.A / Claude Opus 4.8 / no tools: 49 June 2026 problems; Anthropic internal values average four runs." }, { "model_id": "claude-opus-5", "benchmark_id": "arxivmath_2026_06_no_tools", "score": 90.82, "reference_url": "https://www.anthropic.com/claude-opus-5-system-card", "reported_setting": { "effort": "max", "tools": "none" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Figure 8.8.A / Claude Opus 5 / no tools: 49 June 2026 problems; Anthropic internal values average four runs." }, { "model_id": "claude-opus-5", "benchmark_id": "arxivmath_2026_06_with_tools", "score": 91.33, "reference_url": "https://www.anthropic.com/claude-opus-5-system-card", "reported_setting": { "effort": "max", "tools": "benchmark-specified" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Figure 8.8.A / Claude Opus 5 / with tools: 49 June 2026 problems; four runs per problem." }, { "model_id": "gemini-3.5-flash", "benchmark_id": "chartography_no_tools", "score": 35.9, "reference_url": "https://www.anthropic.com/claude-opus-5-system-card", "reported_setting": { "effort": "max", "tools": "none" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Figure 8.12.1.A / Gemini 3.5 Flash / no tools: 100 specialized chart types; five runs; 95% CI." }, { "model_id": "gpt-5.6-sol", "benchmark_id": "chartography_no_tools", "score": 45.0, "reference_url": "https://www.anthropic.com/claude-opus-5-system-card", "reported_setting": { "effort": "max", "tools": "none" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Figure 8.12.1.A / GPT 5.6 Sol / no tools: 100 specialized chart types; five runs; 95% CI." }, { "model_id": "claude-opus-4.8", "benchmark_id": "chartography_no_tools", "score": 17.0, "reference_url": "https://www.anthropic.com/claude-opus-5-system-card", "reported_setting": { "effort": "max", "tools": "none" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Figure 8.12.1.A / Claude Opus 4.8 / no tools: 100 specialized chart types; five runs; 95% CI." }, { "model_id": "claude-opus-4.8", "benchmark_id": "chartography_with_tools", "score": 75.0, "reference_url": "https://www.anthropic.com/claude-opus-5-system-card", "reported_setting": { "effort": "max", "tools": "benchmark-specified" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Figure 8.12.1.A / Claude Opus 4.8 / tools: 100 specialized chart types; five runs; 95% CI." }, { "model_id": "claude-sonnet-5", "benchmark_id": "chartography_no_tools", "score": 16.0, "reference_url": "https://www.anthropic.com/claude-opus-5-system-card", "reported_setting": { "effort": "max", "tools": "none" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Figure 8.12.1.A / Claude Sonnet 5 / no tools: 100 specialized chart types; five runs; 95% CI." }, { "model_id": "claude-sonnet-5", "benchmark_id": "chartography_with_tools", "score": 71.8, "reference_url": "https://www.anthropic.com/claude-opus-5-system-card", "reported_setting": { "effort": "max", "tools": "benchmark-specified" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Figure 8.12.1.A / Claude Sonnet 5 / tools: 100 specialized chart types; five runs; 95% CI." }, { "model_id": "claude-opus-5", "benchmark_id": "chartography_no_tools", "score": 29.6, "reference_url": "https://www.anthropic.com/claude-opus-5-system-card", "reported_setting": { "effort": "max", "tools": "none" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Figure 8.12.1.A / Claude Opus 5 / no tools: 100 specialized chart types; five runs; 95% CI." }, { "model_id": "claude-opus-5", "benchmark_id": "chartography_with_tools", "score": 83.0, "reference_url": "https://www.anthropic.com/claude-opus-5-system-card", "reported_setting": { "effort": "max", "tools": "benchmark-specified" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Figure 8.12.1.A / Claude Opus 5 / tools: 100 specialized chart types; five runs; 95% CI." }, { "model_id": "claude-opus-5", "benchmark_id": "gdp_pdf_mean_criteria_pass_rate", "score": 83.4, "reference_url": "https://www.anthropic.com/claude-opus-5-system-card", "reported_setting": { "effort": "max", "tools": "none", "harness": "corrected full 100-prompt run" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Figure 8.12.4.A narrative / Claude Opus 5 / no tools: 100 prompts, 10 domains, corrected PDF truncation bug, Opus 4.7 judge.", "candidates": [ { "score": 85.5, "reference_url": "https://www.anthropic.com/claude-opus-5-system-card", "source_type": "model_card", "reported_setting": { "effort": "max", "tools": "python / benchmark tools", "harness": "corrected full 100-prompt run" }, "notes": "Figure 8.12.4.A narrative / Claude Opus 5 / tools: 100 prompts, 10 domains, corrected PDF truncation bug, Opus 4.7 judge." } ] }, { "model_id": "claude-opus-5", "benchmark_id": "officeqa", "score": 78.1, "reference_url": "https://www.anthropic.com/claude-opus-5-system-card", "reported_setting": { "effort": "max" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "OfficeQA narrative / Claude Opus 5: Public Messages API for Opus 5; production safeguards and Opus 4.8 classifier fallback." }, { "model_id": "claude-opus-5", "benchmark_id": "officeqa_pro", "score": 66.9, "reference_url": "https://www.anthropic.com/claude-opus-5-system-card", "reported_setting": { "effort": "max" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "OfficeQA Pro narrative / Claude Opus 5: 133-question subset; public Messages API for Opus 5.", "candidates": [ { "score": 60.9, "reference_url": "https://media.x.ai/v1/website/card-4p6-4cd2dc57.pdf", "source_type": "model_card", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "document and rendered-image analysis", "sampling": "unknown", "judge": "OfficeQA Pro exact-match accuracy", "harness": "OfficeQA Pro official harness", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown" }, "notes": "card46 OfficeQA Pro Accuracy (%); source effort=max; benchmark protocol matches canonical=True. Unknown source fields remain unknown." } ] }, { "model_id": "claude-opus-5", "benchmark_id": "legal_agent_benchmark_public", "score": 23.58, "reference_url": "https://www.anthropic.com/claude-opus-5-system-card", "reported_setting": { "effort": "max", "tools": "bash and Python" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Legal Agent Benchmark / public all-pass: 1,235 tested tasks after 16 exclusions; n=5; source reports +/-0.48." }, { "model_id": "claude-opus-5", "benchmark_id": "legal_agent_benchmark_harvey_held_out", "score": 11.7, "reference_url": "https://www.anthropic.com/claude-opus-5-system-card", "reported_setting": { "effort": "max" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Legal Agent Benchmark / Harvey held-out all-pass: Proprietary Harvey held-out evaluation." }, { "model_id": "claude-opus-5", "benchmark_id": "gdpval_aa_elo", "score": 1861, "reference_url": "https://www.anthropic.com/claude-opus-5-system-card", "reported_setting": { "effort": "max", "tools": "shell and web browsing" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "GDPval-AA v2 narrative / Claude Opus 5 / max: 220 tasks, 44 occupations, 9 industries; Artificial Analysis.", "candidates": [ { "score": 1827, "reference_url": "https://www.anthropic.com/claude-opus-5-system-card", "source_type": "model_card", "reported_setting": { "effort": "xhigh", "tools": "shell and web browsing" }, "notes": "GDPval-AA v2 narrative / Claude Opus 5 / xhigh: 220 tasks, 44 occupations, 9 industries; Artificial Analysis." } ] }, { "model_id": "claude-opus-5", "benchmark_id": "aa_briefcase_elo", "score": 1606, "reference_url": "https://www.anthropic.com/claude-opus-5-system-card", "reported_setting": { "effort": "high" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "AA-Briefcase narrative / Claude Opus 5 / high: Multi-week professional projects; Artificial Analysis.", "candidates": [ { "score": 1720, "reference_url": "https://www.anthropic.com/claude-opus-5-system-card", "source_type": "model_card", "reported_setting": { "effort": "max" }, "notes": "AA-Briefcase narrative / Claude Opus 5 / max: Multi-week professional projects; Artificial Analysis." }, { "score": 1693, "reference_url": "https://www.anthropic.com/claude-opus-5-system-card", "source_type": "model_card", "reported_setting": { "effort": "xhigh" }, "notes": "AA-Briefcase narrative / Claude Opus 5 / xhigh: Multi-week professional projects; Artificial Analysis." }, { "score": 1715, "reference_url": "https://media.x.ai/v1/website/card-4p6-4cd2dc57.pdf", "source_type": "model_card", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "offline multi-file knowledge-work tools", "sampling": "unknown", "judge": "rubric plus analytical and presentation pairwise Elo", "harness": "Artificial Analysis AA-Briefcase harness", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown" }, "notes": "card46 AA Briefcase AA-Briefcase Elo; source effort=max; benchmark protocol matches canonical=True. Unknown source fields remain unknown." } ] }, { "model_id": "claude-opus-5", "benchmark_id": "toolathlon_verified", "score": 80.6, "reference_url": "https://www.anthropic.com/claude-opus-5-system-card", "reported_setting": { "effort": "max", "tools": "600+ tools across 32 applications" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Table 8.13.6.A / Claude Opus 5 / Pass@1: 108 tasks; three trials; internal table excludes null attempts. Opus 4.8 and Sonnet 5 internal values are candidates because official published leaderboard values include null attempts." }, { "model_id": "claude-sonnet-5", "benchmark_id": "toolathlon_verified", "score": 71.6, "reference_url": "https://www.anthropic.com/claude-opus-5-system-card", "reported_setting": { "effort": "max", "tools": "600+ tools across 32 applications" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Toolathlon published leaderboard narrative / Claude Sonnet 5: Official published leaderboard including null attempts.", "candidates": [ { "score": 74.7, "reference_url": "https://www.anthropic.com/claude-opus-5-system-card", "source_type": "model_card", "reported_setting": { "effort": "max", "tools": "600+ tools across 32 applications" }, "notes": "Table 8.13.6.A / Claude Sonnet 5 / Pass@1: 108 tasks; three trials; internal table excludes null attempts. Opus 4.8 and Sonnet 5 internal values are candidates because official published leaderboard values include null attempts." } ] }, { "model_id": "claude-opus-5", "benchmark_id": "automation_bench_private_heldout", "score": 26.0, "reference_url": "https://www.anthropic.com/claude-opus-5-system-card", "reported_setting": { "effort": "max" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Figure 8.13.7.A / Claude Opus 5 / max: Private held-out set across 47 applications; deterministic assertions.", "candidates": [ { "score": 24.0, "reference_url": "https://www.anthropic.com/claude-opus-5-system-card", "source_type": "model_card", "reported_setting": { "effort": "medium" }, "notes": "Figure 8.13.7.A / Claude Opus 5 / medium: Private held-out set across 47 applications; deterministic assertions. Exact cost/task: $0.89." } ] }, { "model_id": "claude-opus-5", "benchmark_id": "arc_agi_1", "score": 97.5, "reference_url": "https://arcprize.org/leaderboard", "reported_setting": { "effort": "max" }, "matches_canonical": false, "source_type": "leaderboard", "audit_status": "verified", "notes": "Figure 8.14 narrative / ARC-AGI-1 / Claude Opus 5: Verified semi-private ARC Prize score." }, { "model_id": "claude-opus-5", "benchmark_id": "arc_agi_2", "score": 90.42, "reference_url": "https://arcprize.org/leaderboard", "reported_setting": { "effort": "max" }, "matches_canonical": false, "source_type": "leaderboard", "audit_status": "verified", "notes": "Figure 8.14 narrative / ARC-AGI-2 / Claude Opus 5: Verified semi-private ARC Prize score." }, { "model_id": "claude-mythos", "benchmark_id": "healthbench", "score": 59.6, "reference_url": "https://www.anthropic.com/claude-opus-5-system-card", "reported_setting": { "effort": "max", "tools": "none" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Figure 8.15 / HealthBench / Claude Mythos Preview: Five trials; Opus 4.8 grader; no customized system prompt." }, { "model_id": "claude-opus-4.8", "benchmark_id": "healthbench", "score": 58.8, "reference_url": "https://www.anthropic.com/claude-opus-5-system-card", "reported_setting": { "effort": "max", "tools": "none" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Figure 8.15 / HealthBench / Claude Opus 4.8: Five trials; Opus 4.8 grader; no customized system prompt." }, { "model_id": "claude-sonnet-5", "benchmark_id": "healthbench", "score": 59.2, "reference_url": "https://www.anthropic.com/claude-opus-5-system-card", "reported_setting": { "effort": "max", "tools": "none" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Figure 8.15 / HealthBench / Claude Sonnet 5: Five trials; Opus 4.8 grader; no customized system prompt." }, { "model_id": "claude-opus-5", "benchmark_id": "healthbench", "score": 67.1, "reference_url": "https://www.anthropic.com/claude-opus-5-system-card", "reported_setting": { "effort": "max", "tools": "none" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Figure 8.15 / HealthBench / Claude Opus 5: Five trials; Opus 4.8 grader; no customized system prompt." }, { "model_id": "claude-opus-5", "benchmark_id": "healthbench_length_adjusted", "score": 57.8, "reference_url": "https://www.anthropic.com/claude-opus-5-system-card", "reported_setting": { "effort": "max", "tools": "none" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Figure 8.15 / HealthBench Length-Adjusted / Claude Opus 5: Five trials; Opus 4.8 grader; no customized system prompt." }, { "model_id": "claude-mythos", "benchmark_id": "healthbench_professional", "score": 67.2, "reference_url": "https://www.anthropic.com/claude-opus-5-system-card", "reported_setting": { "effort": "max", "tools": "none" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Figure 8.15 / HealthBench Professional / Claude Mythos Preview: Five trials; Opus 4.8 grader; no customized system prompt." }, { "model_id": "claude-opus-4.8", "benchmark_id": "healthbench_professional", "score": 60.3, "reference_url": "https://www.anthropic.com/claude-opus-5-system-card", "reported_setting": { "effort": "max", "tools": "none" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Figure 8.15 / HealthBench Professional / Claude Opus 4.8: Five trials; Opus 4.8 grader; no customized system prompt." }, { "model_id": "claude-sonnet-5", "benchmark_id": "healthbench_professional", "score": 62.4, "reference_url": "https://www.anthropic.com/claude-opus-5-system-card", "reported_setting": { "effort": "max", "tools": "none" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Figure 8.15 / HealthBench Professional / Claude Sonnet 5: Five trials; Opus 4.8 grader; no customized system prompt." }, { "model_id": "claude-opus-5", "benchmark_id": "healthbench_professional", "score": 73.4, "reference_url": "https://www.anthropic.com/claude-opus-5-system-card", "reported_setting": { "effort": "max", "tools": "none" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Figure 8.15 / HealthBench Professional / Claude Opus 5: Five trials; Opus 4.8 grader; no customized system prompt." }, { "model_id": "claude-mythos", "benchmark_id": "gmmlu", "score": 92.9, "reference_url": "https://www.anthropic.com/claude-opus-5-system-card", "reported_setting": { "effort": "max", "tools": "none" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Figure 8.16 / GMMLU / Claude Mythos Preview: 42 languages; one trial." }, { "model_id": "claude-opus-4.8", "benchmark_id": "gmmlu", "score": 90.9, "reference_url": "https://www.anthropic.com/claude-opus-5-system-card", "reported_setting": { "effort": "max", "tools": "none" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Figure 8.16 / GMMLU / Claude Opus 4.8: 42 languages; one trial." }, { "model_id": "claude-sonnet-5", "benchmark_id": "gmmlu", "score": 89.0, "reference_url": "https://www.anthropic.com/claude-opus-5-system-card", "reported_setting": { "effort": "max", "tools": "none" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Figure 8.16 / GMMLU / Claude Sonnet 5: 42 languages; one trial." }, { "model_id": "claude-opus-5", "benchmark_id": "gmmlu", "score": 92.5, "reference_url": "https://www.anthropic.com/claude-opus-5-system-card", "reported_setting": { "effort": "max", "tools": "none" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Figure 8.16 / GMMLU / Claude Opus 5: 42 languages; one trial." }, { "model_id": "claude-opus-5", "benchmark_id": "milu", "score": 92.1, "reference_url": "https://www.anthropic.com/claude-opus-5-system-card", "reported_setting": { "effort": "max", "tools": "none" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Figure 8.16 / MILU / Claude Opus 5: 11 languages; five trials." }, { "model_id": "claude-opus-5", "benchmark_id": "include_base_44", "score": 89.8, "reference_url": "https://www.anthropic.com/claude-opus-5-system-card", "reported_setting": { "effort": "max", "tools": "none" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Figure 8.16 / INCLUDE-base-44 / Claude Opus 5: 44 languages; five trials." }, { "model_id": "claude-opus-4.8", "benchmark_id": "biomysterybench_human_solvable_revised_2026_07", "score": 0.885, "reference_url": "https://www.anthropic.com/claude-opus-5-system-card", "reported_setting": { "effort": "max", "tools": "benchmark-specific" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Figure 8.17.6.A / BioMysteryBench Human Solvable Revised / Claude Opus 4.8: Revised subset after 3 removals. Source label 88.5%; normalized to 0.885 for the 0-1 score metric." }, { "model_id": "claude-sonnet-5", "benchmark_id": "biomysterybench_human_solvable_revised_2026_07", "score": 0.875, "reference_url": "https://www.anthropic.com/claude-opus-5-system-card", "reported_setting": { "effort": "max", "tools": "benchmark-specific" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Figure 8.17.6.A / BioMysteryBench Human Solvable Revised / Claude Sonnet 5: Revised subset after 3 removals. Source label 87.5%; normalized to 0.875 for the 0-1 score metric." }, { "model_id": "claude-opus-5", "benchmark_id": "biomysterybench_human_solvable_revised_2026_07", "score": 0.901, "reference_url": "https://www.anthropic.com/claude-opus-5-system-card", "reported_setting": { "effort": "max", "tools": "benchmark-specific" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Figure 8.17.6.A / BioMysteryBench Human Solvable Revised / Claude Opus 5: Revised subset after 3 removals. Source label 90.1%; normalized to 0.901 for the 0-1 score metric." }, { "model_id": "claude-opus-4.8", "benchmark_id": "biomysterybench_human_difficult_revised_2026_07", "score": 0.424, "reference_url": "https://www.anthropic.com/claude-opus-5-system-card", "reported_setting": { "effort": "max", "tools": "benchmark-specific" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Figure 8.17.6.A / BioMysteryBench Human Difficult Revised / Claude Opus 4.8: Revised subset after 6 removals. Source label 42.4%; normalized to 0.424 for the 0-1 score metric." }, { "model_id": "claude-sonnet-5", "benchmark_id": "biomysterybench_human_difficult_revised_2026_07", "score": 0.341, "reference_url": "https://www.anthropic.com/claude-opus-5-system-card", "reported_setting": { "effort": "max", "tools": "benchmark-specific" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Figure 8.17.6.A / BioMysteryBench Human Difficult Revised / Claude Sonnet 5: Revised subset after 6 removals. Source label 34.1%; normalized to 0.341 for the 0-1 score metric." }, { "model_id": "claude-opus-5", "benchmark_id": "biomysterybench_human_difficult_revised_2026_07", "score": 0.494, "reference_url": "https://www.anthropic.com/claude-opus-5-system-card", "reported_setting": { "effort": "max", "tools": "benchmark-specific" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Figure 8.17.6.A / BioMysteryBench Human Difficult Revised / Claude Opus 5: Revised subset after 6 removals. Source label 49.4%; normalized to 0.494 for the 0-1 score metric." }, { "model_id": "claude-opus-5", "benchmark_id": "spatialbench_verified", "score": 0.725, "reference_url": "https://www.anthropic.com/claude-opus-5-system-card", "reported_setting": { "effort": "max", "tools": "benchmark-specific" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Figure 8.17.6.A / SpatialBench Verified / Claude Opus 5: 115 problems. Source label 72.5%; normalized to 0.725 for the 0-1 score metric." }, { "model_id": "claude-opus-5", "benchmark_id": "singlecellbench", "score": 0.606, "reference_url": "https://www.anthropic.com/claude-opus-5-system-card", "reported_setting": { "effort": "max", "tools": "benchmark-specific" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Figure 8.17.6.A / SingleCellBench / Claude Opus 5: 195 problems. Source label 60.6%; normalized to 0.606 for the 0-1 score metric." }, { "model_id": "claude-opus-5", "benchmark_id": "proteingym_hard", "score": 0.477, "reference_url": "https://www.anthropic.com/claude-opus-5-system-card", "reported_setting": { "effort": "max", "tools": "benchmark-specific" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Figure 8.17.6.A / ProteinGym Hard / Claude Opus 5: Rank correlation. Source label 47.7%; normalized to 0.477 for the 0-1 score metric." }, { "model_id": "claude-opus-4.8", "benchmark_id": "protein_design_internal", "score": 0.32, "reference_url": "https://www.anthropic.com/claude-opus-5-system-card", "reported_setting": { "effort": "max", "tools": "benchmark-specific" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Figure 8.17.6.A / Protein Design / Claude Opus 4.8: Combined constraint-satisfaction, folding-confidence, and novelty score. Source label 32.0%; normalized to 0.32 for the 0-1 score metric." }, { "model_id": "claude-sonnet-5", "benchmark_id": "protein_design_internal", "score": 0.212, "reference_url": "https://www.anthropic.com/claude-opus-5-system-card", "reported_setting": { "effort": "max", "tools": "benchmark-specific" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Figure 8.17.6.A / Protein Design / Claude Sonnet 5: Combined constraint-satisfaction, folding-confidence, and novelty score. Source label 21.2%; normalized to 0.212 for the 0-1 score metric." }, { "model_id": "claude-opus-5", "benchmark_id": "protein_design_internal", "score": 0.425, "reference_url": "https://www.anthropic.com/claude-opus-5-system-card", "reported_setting": { "effort": "max", "tools": "benchmark-specific" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Figure 8.17.6.A / Protein Design / Claude Opus 5: Combined constraint-satisfaction, folding-confidence, and novelty score. Source label 42.5%; normalized to 0.425 for the 0-1 score metric." }, { "model_id": "claude-opus-4.8", "benchmark_id": "organic_chemistry_v2_internal", "score": 0.504, "reference_url": "https://www.anthropic.com/claude-opus-5-system-card", "reported_setting": { "effort": "max", "tools": "benchmark-specific" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Figure 8.17.6.A / Organic Chemistry V2 / Claude Opus 4.8: Opus 5 timed out on 10/1,260 attempts; excluded; stated effect <= +/-0.006. Source label 50.4%; normalized to 0.504 for the 0-1 score metric." }, { "model_id": "claude-sonnet-5", "benchmark_id": "organic_chemistry_v2_internal", "score": 0.406, "reference_url": "https://www.anthropic.com/claude-opus-5-system-card", "reported_setting": { "effort": "max", "tools": "benchmark-specific" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Figure 8.17.6.A / Organic Chemistry V2 / Claude Sonnet 5: Opus 5 timed out on 10/1,260 attempts; excluded; stated effect <= +/-0.006. Source label 40.6%; normalized to 0.406 for the 0-1 score metric." }, { "model_id": "claude-opus-5", "benchmark_id": "organic_chemistry_v2_internal", "score": 0.616, "reference_url": "https://www.anthropic.com/claude-opus-5-system-card", "reported_setting": { "effort": "max", "tools": "benchmark-specific" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Figure 8.17.6.A / Organic Chemistry V2 / Claude Opus 5: Opus 5 timed out on 10/1,260 attempts; excluded; stated effect <= +/-0.006. Source label 61.6%; normalized to 0.616 for the 0-1 score metric." }, { "model_id": "claude-opus-5", "benchmark_id": "protocol_troubleshooting_internal", "score": 0.611, "reference_url": "https://www.anthropic.com/claude-opus-5-system-card", "reported_setting": { "effort": "max", "tools": "benchmark-specific" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Figure 8.17.6.A / Protocol Troubleshooting / Claude Opus 5: Bash, file editor, and web search. Source label 61.1%; normalized to 0.611 for the 0-1 score metric." }, { "model_id": "claude-opus-4.8", "benchmark_id": "protocol_understanding_internal", "score": 0.623, "reference_url": "https://www.anthropic.com/claude-opus-5-system-card", "reported_setting": { "effort": "max", "tools": "benchmark-specific" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Figure 8.17.6.A / Protocol Understanding / Claude Opus 4.8: Benchling protocol-understanding variant. Source label 62.3%; normalized to 0.623 for the 0-1 score metric." }, { "model_id": "claude-sonnet-5", "benchmark_id": "protocol_understanding_internal", "score": 0.605, "reference_url": "https://www.anthropic.com/claude-opus-5-system-card", "reported_setting": { "effort": "max", "tools": "benchmark-specific" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Figure 8.17.6.A / Protocol Understanding / Claude Sonnet 5: Benchling protocol-understanding variant. Source label 60.5%; normalized to 0.605 for the 0-1 score metric." }, { "model_id": "claude-opus-5", "benchmark_id": "protocol_understanding_internal", "score": 0.784, "reference_url": "https://www.anthropic.com/claude-opus-5-system-card", "reported_setting": { "effort": "max", "tools": "benchmark-specific" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Figure 8.17.6.A / Protocol Understanding / Claude Opus 5: Benchling protocol-understanding variant. Source label 78.4%; normalized to 0.784 for the 0-1 score metric." }, { "model_id": "claude-opus-5", "benchmark_id": "long_form_virology_task_1_sequence_design", "score": 0.89, "reference_url": "https://www.anthropic.com/claude-opus-5-system-card", "reported_setting": { "effort": "source does not state" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Figure 2.2.4.A / Long-form virology / Task 1 / Sequence design / Claude Opus 5: Two long-form virology tasks." }, { "model_id": "claude-opus-5", "benchmark_id": "long_form_virology_task_1_protocol_design", "score": 0.91, "reference_url": "https://www.anthropic.com/claude-opus-5-system-card", "reported_setting": { "effort": "source does not state" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Figure 2.2.4.A / Long-form virology / Task 1 / Protocol design / Claude Opus 5: Two long-form virology tasks." }, { "model_id": "claude-opus-5", "benchmark_id": "long_form_virology_task_1_end_to_end", "score": 0.802, "reference_url": "https://www.anthropic.com/claude-opus-5-system-card", "reported_setting": { "effort": "source does not state" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Figure 2.2.4.A / Long-form virology / Task 1 / End-to-end / Claude Opus 5: Two long-form virology tasks." }, { "model_id": "claude-opus-5", "benchmark_id": "long_form_virology_task_2_sequence_design", "score": 1.0, "reference_url": "https://www.anthropic.com/claude-opus-5-system-card", "reported_setting": { "effort": "source does not state" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Figure 2.2.4.A / Long-form virology / Task 2 / Sequence design / Claude Opus 5: Two long-form virology tasks." }, { "model_id": "claude-opus-5", "benchmark_id": "long_form_virology_task_2_protocol_design", "score": 0.87, "reference_url": "https://www.anthropic.com/claude-opus-5-system-card", "reported_setting": { "effort": "source does not state" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Figure 2.2.4.A / Long-form virology / Task 2 / Protocol design / Claude Opus 5: Two long-form virology tasks." }, { "model_id": "claude-opus-5", "benchmark_id": "long_form_virology_task_2_end_to_end", "score": 0.872, "reference_url": "https://www.anthropic.com/claude-opus-5-system-card", "reported_setting": { "effort": "source does not state" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Figure 2.2.4.A / Long-form virology / Task 2 / End-to-end / Claude Opus 5: Two long-form virology tasks. Task 2 Opus 5 narrative reports 0.872; the rendered figure rounds it to 0.87." }, { "model_id": "claude-opus-5", "benchmark_id": "ai_rd_kernel_best_speedup", "score": 449.46, "reference_url": "https://www.anthropic.com/claude-opus-5-system-card", "reported_setting": { "effort": "source does not state" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Table 2.3.5.A / Kernel best speedup / Claude Opus 5: Anthropic internal AI R&D task-based evaluation." }, { "model_id": "claude-opus-5", "benchmark_id": "ai_rd_time_series_forecasting_mse", "score": 5.68, "reference_url": "https://www.anthropic.com/claude-opus-5-system-card", "reported_setting": { "effort": "source does not state" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Table 2.3.5.A / Time-Series Forecasting MSE / Claude Opus 5: Anthropic internal AI R&D task-based evaluation." }, { "model_id": "claude-opus-5", "benchmark_id": "ai_rd_llm_training_speedup", "score": 68.54, "reference_url": "https://www.anthropic.com/claude-opus-5-system-card", "reported_setting": { "effort": "source does not state" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Table 2.3.5.A / LLM Training easy speedup / Claude Opus 5: Anthropic internal AI R&D task-based evaluation." }, { "model_id": "claude-opus-5", "benchmark_id": "ai_rd_llm_training_hard_speedup", "score": 14.19, "reference_url": "https://www.anthropic.com/claude-opus-5-system-card", "reported_setting": { "effort": "source does not state" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Table 2.3.5.A / LLM Training hard speedup / Claude Opus 5: Anthropic internal AI R&D task-based evaluation." }, { "model_id": "claude-opus-5", "benchmark_id": "ai_rd_quadruped_rl", "score": 31.3, "reference_url": "https://www.anthropic.com/claude-opus-5-system-card", "reported_setting": { "effort": "source does not state" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Table 2.3.5.A / Quadruped RL / Claude Opus 5: Anthropic internal AI R&D task-based evaluation." }, { "model_id": "claude-opus-5", "benchmark_id": "ai_rd_novel_compiler", "score": 80.91, "reference_url": "https://www.anthropic.com/claude-opus-5-system-card", "reported_setting": { "effort": "source does not state" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Table 2.3.5.A / Novel Compiler complex-test pass rate / Claude Opus 5: Anthropic internal AI R&D task-based evaluation." }, { "model_id": "claude-opus-5", "benchmark_id": "exploitbench", "score": 70, "reference_url": "https://www.anthropic.com/claude-opus-5-system-card", "reported_setting": { "effort": "source does not state" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Figure 3.3.1.A / ExploitBench Cap% / Claude Opus 5: 41 V8 vulnerabilities; five trials; Cap% uses a random three-trial subset; safeguards off." }, { "model_id": "claude-opus-5", "benchmark_id": "oss_fuzz_any_progress", "score": 79.4, "reference_url": "https://www.anthropic.com/claude-opus-5-system-card", "reported_setting": { "effort": "source does not state" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Figure 3.3.2.A / OSS-Fuzz any progress / Claude Opus 5: Approximately 830 entry points from 228 projects; safeguards off." }, { "model_id": "claude-opus-5", "benchmark_id": "oss_fuzz_control_flow_hijack_count", "score": 4, "reference_url": "https://www.anthropic.com/claude-opus-5-system-card", "reported_setting": { "effort": "source does not state" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Figure 3.3.2.A / OSS-Fuzz control-flow hijack / Claude Opus 5: Count reaching grade 1.0 across approximately 830 entry points." }, { "model_id": "claude-opus-5", "benchmark_id": "firefox_147_exploit_development_working_exploit", "score": 52.4, "reference_url": "https://www.anthropic.com/claude-opus-5-system-card", "reported_setting": { "effort": "source does not state" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Figure 3.3.3.A / Firefox 147 working exploit / Claude Opus 5: 50 crash categories, five trials each, 250 trials total; mitigations off." }, { "model_id": "claude-opus-5", "benchmark_id": "firefox_147_exploit_development_any_success", "score": 87.2, "reference_url": "https://www.anthropic.com/claude-opus-5-system-card", "reported_setting": { "effort": "source does not state" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Figure 3.3.3.A / Firefox 147 any success / Claude Opus 5: Grade >=0.5 across 250 trials; mitigations off." }, { "model_id": "claude-opus-4.8", "benchmark_id": "exploitgym", "score": 13.808976, "reference_url": "https://www.anthropic.com/claude-opus-5-system-card", "reported_setting": { "effort": "max", "harness": "6h wall-clock budget" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Figure 3.3.5.A / ExploitGym / Claude Opus 4.8 / 6h: 120/869 successful exploits; mitigations off; agent-judge verification. Exact percentage derived deterministically from printed count 120/869.", "candidates": [ { "score": 9.205984, "reference_url": "https://www.anthropic.com/claude-opus-5-system-card", "source_type": "model_card", "reported_setting": { "effort": "max", "harness": "2h wall-clock budget" }, "notes": "Figure 3.3.5.A / ExploitGym / Claude Opus 4.8 / 2h: 80/869 successful exploits; mitigations off; agent-judge verification. Exact percentage derived deterministically from printed count 80/869." } ] }, { "model_id": "claude-opus-5", "benchmark_id": "exploitgym", "score": 21.979287, "reference_url": "https://www.anthropic.com/claude-opus-5-system-card", "reported_setting": { "effort": "max", "harness": "6h wall-clock budget" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Figure 3.3.5.A / ExploitGym / Claude Opus 5 / 6h: 191/869 successful exploits; mitigations off; agent-judge verification. Exact percentage derived deterministically from printed count 191/869.", "candidates": [ { "score": 19.677791, "reference_url": "https://www.anthropic.com/claude-opus-5-system-card", "source_type": "model_card", "reported_setting": { "effort": "max", "harness": "2h wall-clock budget" }, "notes": "Figure 3.3.5.A / ExploitGym / Claude Opus 5 / 2h: 171/869 successful exploits; mitigations off; agent-judge verification. Exact percentage derived deterministically from printed count 171/869." } ] }, { "model_id": "claude-fable-5", "benchmark_id": "cursorbench_3_2", "score": 70.5, "reference_url": "https://cursor.com/cursorbench", "reported_setting": { "effort": "max", "harness": "Cursor agent" }, "matches_canonical": true, "source_type": "leaderboard", "audit_status": "verified", "notes": "Official CursorBench table / Fable 5 Max: Official score; average cost/task $17.32; 103,525 tokens/task; 72 steps/task.", "candidates": [ { "score": 68.4, "reference_url": "https://cursor.com/cursorbench", "source_type": "leaderboard", "reported_setting": { "effort": "xhigh", "harness": "Cursor agent" }, "notes": "Official CursorBench table / Fable 5 Extra High: Official score; average cost/task $11.73; 64,971 tokens/task; 56 steps/task." }, { "score": 66.5, "reference_url": "https://cursor.com/cursorbench", "source_type": "leaderboard", "reported_setting": { "effort": "high", "harness": "Cursor agent" }, "notes": "Official CursorBench table / Fable 5 High: Official score; average cost/task $8.77; 43,747 tokens/task; 48 steps/task." }, { "score": 65.2, "reference_url": "https://cursor.com/cursorbench", "source_type": "leaderboard", "reported_setting": { "effort": "medium", "harness": "Cursor agent" }, "notes": "Official CursorBench table / Fable 5 Medium: Official score; average cost/task $6.8; 30,366 tokens/task; 41 steps/task." }, { "score": 62.1, "reference_url": "https://cursor.com/cursorbench", "source_type": "leaderboard", "reported_setting": { "effort": "low", "harness": "Cursor agent" }, "notes": "Official CursorBench table / Fable 5 Low: Official score; average cost/task $4.46; 18,182 tokens/task; 31 steps/task." }, { "score": 70.5, "reference_url": "https://x.ai/news/grok-4-6", "source_type": "official_blog", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "Cursor coding-agent tools", "sampling": "unknown", "judge": "CursorBench task score", "harness": "Cursor first-party agent harness", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown" }, "notes": "release46 CursorBench v3.2 Score (%); source effort=max; benchmark protocol matches canonical=True. Unknown source fields remain unknown." } ] }, { "model_id": "gpt-5.6-sol", "benchmark_id": "cursorbench_3_2", "score": 67.2, "reference_url": "https://cursor.com/cursorbench", "reported_setting": { "effort": "max", "harness": "Cursor agent" }, "matches_canonical": true, "source_type": "leaderboard", "audit_status": "verified", "notes": "Official CursorBench table / GPT-5.6 Sol Max: Official score; average cost/task $5.69; 28,320 tokens/task; 48 steps/task.", "candidates": [ { "score": 64.5, "reference_url": "https://cursor.com/cursorbench", "source_type": "leaderboard", "reported_setting": { "effort": "xhigh", "harness": "Cursor agent" }, "notes": "Official CursorBench table / GPT-5.6 Sol Extra High: Official score; average cost/task $3.88; 19,699 tokens/task; 38 steps/task." }, { "score": 63.5, "reference_url": "https://cursor.com/cursorbench", "source_type": "leaderboard", "reported_setting": { "effort": "high", "harness": "Cursor agent" }, "notes": "Official CursorBench table / GPT-5.6 Sol High: Official score; average cost/task $2.79; 13,867 tokens/task; 32 steps/task." }, { "score": 60.0, "reference_url": "https://cursor.com/cursorbench", "source_type": "leaderboard", "reported_setting": { "effort": "medium", "harness": "Cursor agent" }, "notes": "Official CursorBench table / GPT-5.6 Sol Medium: Official score; average cost/task $1.95; 9,747 tokens/task; 27 steps/task." }, { "score": 52.6, "reference_url": "https://cursor.com/cursorbench", "source_type": "leaderboard", "reported_setting": { "effort": "low", "harness": "Cursor agent" }, "notes": "Official CursorBench table / GPT-5.6 Sol Low: Official score; average cost/task $1.01; 5,104 tokens/task; 19 steps/task." }, { "score": 67.2, "reference_url": "https://x.ai/news/grok-4-6", "source_type": "official_blog", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "Cursor coding-agent tools", "sampling": "unknown", "judge": "CursorBench task score", "harness": "Cursor first-party agent harness", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown" }, "notes": "release46 CursorBench v3.2 Score (%); source effort=max; benchmark protocol matches canonical=True. Unknown source fields remain unknown." } ] }, { "model_id": "claude-opus-5", "benchmark_id": "cursorbench_3_2", "score": 66.7, "reference_url": "https://cursor.com/cursorbench", "reported_setting": { "effort": "high", "harness": "Cursor agent" }, "matches_canonical": true, "source_type": "leaderboard", "audit_status": "verified", "notes": "Official CursorBench table / Opus 5 High: Official score; average cost/task $3.91; 27,932 tokens/task; 48 steps/task.", "candidates": [ { "score": 70.0, "reference_url": "https://cursor.com/cursorbench", "source_type": "leaderboard", "reported_setting": { "effort": "max", "harness": "Cursor agent" }, "notes": "Official CursorBench table / Opus 5 Max: Official score; average cost/task $8.23; 61,838 tokens/task; 78 steps/task." }, { "score": 69.3, "reference_url": "https://cursor.com/cursorbench", "source_type": "leaderboard", "reported_setting": { "effort": "xhigh", "harness": "Cursor agent" }, "notes": "Official CursorBench table / Opus 5 Extra High: Official score; average cost/task $7.35; 54,239 tokens/task; 72 steps/task." }, { "score": 64.3, "reference_url": "https://cursor.com/cursorbench", "source_type": "leaderboard", "reported_setting": { "effort": "medium", "harness": "Cursor agent" }, "notes": "Official CursorBench table / Opus 5 Medium: Official score; average cost/task $3.29; 23,612 tokens/task; 44 steps/task." }, { "score": 62.8, "reference_url": "https://cursor.com/cursorbench", "source_type": "leaderboard", "reported_setting": { "effort": "low", "harness": "Cursor agent" }, "notes": "Official CursorBench table / Opus 5 Low: Official score; average cost/task $2.55; 18,529 tokens/task; 37 steps/task." } ] }, { "model_id": "gpt-5.6-terra", "benchmark_id": "cursorbench_3_2", "score": 64.9, "reference_url": "https://cursor.com/cursorbench", "reported_setting": { "effort": "max", "harness": "Cursor agent" }, "matches_canonical": true, "source_type": "leaderboard", "audit_status": "verified", "notes": "Official CursorBench table / GPT-5.6 Terra Max: Official score; average cost/task $2.31; 32,969 tokens/task; 47 steps/task.", "candidates": [ { "score": 59.2, "reference_url": "https://cursor.com/cursorbench", "source_type": "leaderboard", "reported_setting": { "effort": "xhigh", "harness": "Cursor agent" }, "notes": "Official CursorBench table / GPT-5.6 Terra Extra High: Official score; average cost/task $1.15; 16,089 tokens/task; 29 steps/task." }, { "score": 54.2, "reference_url": "https://cursor.com/cursorbench", "source_type": "leaderboard", "reported_setting": { "effort": "high", "harness": "Cursor agent" }, "notes": "Official CursorBench table / GPT-5.6 Terra High: Official score; average cost/task $0.71; 9,468 tokens/task; 23 steps/task." }, { "score": 50.3, "reference_url": "https://cursor.com/cursorbench", "source_type": "leaderboard", "reported_setting": { "effort": "medium", "harness": "Cursor agent" }, "notes": "Official CursorBench table / GPT-5.6 Terra Medium: Official score; average cost/task $0.49; 6,222 tokens/task; 20 steps/task." }, { "score": 46.9, "reference_url": "https://cursor.com/cursorbench", "source_type": "leaderboard", "reported_setting": { "effort": "low", "harness": "Cursor agent" }, "notes": "Official CursorBench table / GPT-5.6 Terra Low: Official score; average cost/task $0.42; 5,312 tokens/task; 19 steps/task." } ] }, { "model_id": "gpt-5.6-luna", "benchmark_id": "cursorbench_3_2", "score": 61.1, "reference_url": "https://cursor.com/cursorbench", "reported_setting": { "effort": "max", "harness": "Cursor agent" }, "matches_canonical": true, "source_type": "leaderboard", "audit_status": "verified", "notes": "Official CursorBench table / GPT-5.6 Luna Max: Official score; average cost/task $0.39; 87,973 tokens/task; 61 steps/task.", "candidates": [ { "score": 57.7, "reference_url": "https://cursor.com/cursorbench", "source_type": "leaderboard", "reported_setting": { "effort": "xhigh", "harness": "Cursor agent" }, "notes": "Official CursorBench table / GPT-5.6 Luna Extra High: Official score; average cost/task $0.23; 22,480 tokens/task; 48 steps/task." }, { "score": 56.8, "reference_url": "https://cursor.com/cursorbench", "source_type": "leaderboard", "reported_setting": { "effort": "high", "harness": "Cursor agent" }, "notes": "Official CursorBench table / GPT-5.6 Luna High: Official score; average cost/task $0.16; 15,141 tokens/task; 40 steps/task." }, { "score": 47.7, "reference_url": "https://cursor.com/cursorbench", "source_type": "leaderboard", "reported_setting": { "effort": "medium", "harness": "Cursor agent" }, "notes": "Official CursorBench table / GPT-5.6 Luna Medium: Official score; average cost/task $0.08; 7,095 tokens/task; 28 steps/task." }, { "score": 37.6, "reference_url": "https://cursor.com/cursorbench", "source_type": "leaderboard", "reported_setting": { "effort": "low", "harness": "Cursor agent" }, "notes": "Official CursorBench table / GPT-5.6 Luna Low: Official score; average cost/task $0.03; 3,209 tokens/task; 17 steps/task." } ] }, { "model_id": "kimi-k3", "benchmark_id": "cursorbench_3_2", "score": 60.8, "reference_url": "https://cursor.com/cursorbench", "reported_setting": { "effort": "max", "harness": "Cursor agent" }, "matches_canonical": true, "source_type": "leaderboard", "audit_status": "verified", "notes": "Official CursorBench table / Kimi K3 Max: Official score; average cost/task $2.7; 38,428 tokens/task; 57 steps/task.", "candidates": [ { "score": 59.7, "reference_url": "https://cursor.com/cursorbench", "source_type": "leaderboard", "reported_setting": { "effort": "high", "harness": "Cursor agent" }, "notes": "Official CursorBench table / Kimi K3 High: Official score; average cost/task $1.89; 26,846 tokens/task; 47 steps/task." }, { "score": 50.5, "reference_url": "https://cursor.com/cursorbench", "source_type": "leaderboard", "reported_setting": { "effort": "low", "harness": "Cursor agent" }, "notes": "Official CursorBench table / Kimi K3 Low: Official score; average cost/task $0.99; 13,007 tokens/task; 33 steps/task." } ] }, { "model_id": "gpt-5.5", "benchmark_id": "cursorbench_3_2", "score": 58.4, "reference_url": "https://cursor.com/cursorbench", "reported_setting": { "effort": "xhigh", "harness": "Cursor agent" }, "matches_canonical": true, "source_type": "leaderboard", "audit_status": "verified", "notes": "Official CursorBench table / GPT-5.5 Extra High: Official score; average cost/task $2.85; 17,534 tokens/task; 32 steps/task.", "candidates": [ { "score": 58.4, "reference_url": "https://cursor.com/cursorbench", "source_type": "leaderboard", "reported_setting": { "effort": "high", "harness": "Cursor agent" }, "notes": "Official CursorBench table / GPT-5.5 High: Official score; average cost/task $2.05; 12,183 tokens/task; 28 steps/task." }, { "score": 53.8, "reference_url": "https://cursor.com/cursorbench", "source_type": "leaderboard", "reported_setting": { "effort": "medium", "harness": "Cursor agent" }, "notes": "Official CursorBench table / GPT-5.5 Medium: Official score; average cost/task $1.51; 8,522 tokens/task; 25 steps/task." }, { "score": 46.6, "reference_url": "https://cursor.com/cursorbench", "source_type": "leaderboard", "reported_setting": { "effort": "low", "harness": "Cursor agent" }, "notes": "Official CursorBench table / GPT-5.5 Low: Official score; average cost/task $0.98; 5,168 tokens/task; 20 steps/task." } ] }, { "model_id": "claude-opus-4.8", "benchmark_id": "cursorbench_3_2", "score": 58.0, "reference_url": "https://cursor.com/cursorbench", "reported_setting": { "effort": "high", "harness": "Cursor agent" }, "matches_canonical": true, "source_type": "leaderboard", "audit_status": "verified", "notes": "Official CursorBench table / Opus 4.8 High: Official score; average cost/task $3.15; 33,548 tokens/task; 33 steps/task.", "candidates": [ { "score": 62.3, "reference_url": "https://cursor.com/cursorbench", "source_type": "leaderboard", "reported_setting": { "effort": "max", "harness": "Cursor agent" }, "notes": "Official CursorBench table / Opus 4.8 Max: Official score; average cost/task $5.77; 71,411 tokens/task; 44 steps/task." }, { "score": 59.4, "reference_url": "https://cursor.com/cursorbench", "source_type": "leaderboard", "reported_setting": { "effort": "xhigh", "harness": "Cursor agent" }, "notes": "Official CursorBench table / Opus 4.8 Extra High: Official score; average cost/task $4.5; 51,121 tokens/task; 40 steps/task." }, { "score": 56.1, "reference_url": "https://cursor.com/cursorbench", "source_type": "leaderboard", "reported_setting": { "effort": "medium", "harness": "Cursor agent" }, "notes": "Official CursorBench table / Opus 4.8 Medium: Official score; average cost/task $2.81; 28,384 tokens/task; 32 steps/task." }, { "score": 53.1, "reference_url": "https://cursor.com/cursorbench", "source_type": "leaderboard", "reported_setting": { "effort": "low", "harness": "Cursor agent" }, "notes": "Official CursorBench table / Opus 4.8 Low: Official score; average cost/task $2.02; 19,624 tokens/task; 27 steps/task." } ] }, { "model_id": "claude-sonnet-5", "benchmark_id": "cursorbench_3_2", "score": 56.9, "reference_url": "https://cursor.com/cursorbench", "reported_setting": { "effort": "high", "harness": "Cursor agent" }, "matches_canonical": true, "source_type": "leaderboard", "audit_status": "verified", "notes": "Official CursorBench table / Sonnet 5 High: Official score; average cost/task $2.13; 39,483 tokens/task; 57 steps/task.", "candidates": [ { "score": 61.5, "reference_url": "https://cursor.com/cursorbench", "source_type": "leaderboard", "reported_setting": { "effort": "max", "harness": "Cursor agent" }, "notes": "Official CursorBench table / Sonnet 5 Max: Official score; average cost/task $4.3; 92,882 tokens/task; 86 steps/task." }, { "score": 58.7, "reference_url": "https://cursor.com/cursorbench", "source_type": "leaderboard", "reported_setting": { "effort": "xhigh", "harness": "Cursor agent" }, "notes": "Official CursorBench table / Sonnet 5 Extra High: Official score; average cost/task $2.77; 52,871 tokens/task; 67 steps/task." }, { "score": 52.4, "reference_url": "https://cursor.com/cursorbench", "source_type": "leaderboard", "reported_setting": { "effort": "medium", "harness": "Cursor agent" }, "notes": "Official CursorBench table / Sonnet 5 Medium: Official score; average cost/task $1.44; 26,200 tokens/task; 46 steps/task." }, { "score": 47.7, "reference_url": "https://cursor.com/cursorbench", "source_type": "leaderboard", "reported_setting": { "effort": "low", "harness": "Cursor agent" }, "notes": "Official CursorBench table / Sonnet 5 Low: Official score; average cost/task $0.87; 16,269 tokens/task; 33 steps/task." } ] }, { "model_id": "composer-2.5", "benchmark_id": "cursorbench_3_2", "score": 56.1, "reference_url": "https://cursor.com/cursorbench", "reported_setting": { "harness": "Cursor agent" }, "matches_canonical": false, "source_type": "leaderboard", "audit_status": "verified", "notes": "Official CursorBench table / Composer 2.5: Official score; average cost/task $0.44; 14,286 tokens/task; 33 steps/task." }, { "model_id": "glm-5.2", "benchmark_id": "cursorbench_3_2", "score": 55.0, "reference_url": "https://cursor.com/cursorbench", "reported_setting": { "effort": "max", "harness": "Cursor agent" }, "matches_canonical": true, "source_type": "leaderboard", "audit_status": "verified", "notes": "Official CursorBench table / GLM 5.2 Max: Official score; average cost/task $1.76; 35,946 tokens/task; 58 steps/task.", "candidates": [ { "score": 51.5, "reference_url": "https://cursor.com/cursorbench", "source_type": "leaderboard", "reported_setting": { "effort": "high", "harness": "Cursor agent" }, "notes": "Official CursorBench table / GLM 5.2 High: Official score; average cost/task $1.19; 21,829 tokens/task; 49 steps/task." } ] }, { "model_id": "gemini-3.6-flash", "benchmark_id": "cursorbench_3_2", "score": 51.2, "reference_url": "https://cursor.com/cursorbench", "reported_setting": { "effort": "medium", "harness": "Cursor agent" }, "matches_canonical": true, "source_type": "leaderboard", "audit_status": "verified", "notes": "Official CursorBench table / Gemini 3.6 Flash Medium: Official score; average cost/task $1.48; 28,511 tokens/task; 62 steps/task.", "candidates": [ { "score": 53.5, "reference_url": "https://cursor.com/cursorbench", "source_type": "leaderboard", "reported_setting": { "effort": "high", "harness": "Cursor agent" }, "notes": "Official CursorBench table / Gemini 3.6 Flash High: Official score; average cost/task $1.56; 30,436 tokens/task; 64 steps/task." }, { "score": 47.4, "reference_url": "https://cursor.com/cursorbench", "source_type": "leaderboard", "reported_setting": { "effort": "low", "harness": "Cursor agent" }, "notes": "Official CursorBench table / Gemini 3.6 Flash Low: Official score; average cost/task $1.13; 20,529 tokens/task; 50 steps/task." } ] }, { "model_id": "kimi-k2.7-code", "benchmark_id": "cursorbench_3_2", "score": 49.7, "reference_url": "https://cursor.com/cursorbench", "reported_setting": { "harness": "Cursor agent" }, "matches_canonical": false, "source_type": "leaderboard", "audit_status": "verified", "notes": "Official CursorBench table / Kimi K2.7 Code: Official score; average cost/task $1.43; 31,247 tokens/task; 58 steps/task." }, { "model_id": "claude-opus-4.6", "benchmark_id": "arc_agi_3", "score": 0.51, "reference_url": "https://arcprize.org/leaderboard", "reported_setting": { "effort": "max" }, "matches_canonical": true, "source_type": "leaderboard", "audit_status": "verified", "notes": "ARC Prize leaderboard / Anthropic Opus 4.6 (Max): Relative Human Action Efficiency; total cost $8866.2." }, { "model_id": "grok-4.20", "benchmark_id": "arc_agi_3", "score": 0.09, "reference_url": "https://arcprize.org/leaderboard", "reported_setting": {}, "matches_canonical": false, "source_type": "leaderboard", "audit_status": "verified", "notes": "ARC Prize leaderboard / Grok 4.20 (Beta Reasoning): Relative Human Action Efficiency; total cost $3775." }, { "model_id": "gpt-5.4", "benchmark_id": "arc_agi_3", "score": 0.21, "reference_url": "https://arcprize.org/leaderboard", "reported_setting": { "effort": "high" }, "matches_canonical": false, "source_type": "leaderboard", "audit_status": "verified", "notes": "ARC Prize leaderboard / GPT-5.4 (High): Relative Human Action Efficiency; total cost $5187.41." }, { "model_id": "claude-opus-4.7", "benchmark_id": "arc_agi_3", "score": 0.18, "reference_url": "https://arcprize.org/leaderboard", "reported_setting": { "effort": "high" }, "matches_canonical": false, "source_type": "leaderboard", "audit_status": "verified", "notes": "ARC Prize leaderboard / Opus 4.7 (High): Relative Human Action Efficiency; total cost $10000." }, { "model_id": "grok-4.5", "benchmark_id": "arc_agi_3", "score": 0.3, "reference_url": "https://arcprize.org/leaderboard", "reported_setting": { "effort": "high" }, "matches_canonical": true, "source_type": "leaderboard", "audit_status": "verified", "notes": "ARC Prize leaderboard / Grok 4.5 (High): Relative Human Action Efficiency; total cost $6892.87.", "candidates": [ { "score": 0.32, "reference_url": "https://arcprize.org/leaderboard", "source_type": "leaderboard", "reported_setting": { "effort": "medium" }, "notes": "ARC Prize leaderboard / Grok 4.5 (Medium): Relative Human Action Efficiency; total cost $8458.3." }, { "score": 0.26, "reference_url": "https://arcprize.org/leaderboard", "source_type": "leaderboard", "reported_setting": { "effort": "low" }, "notes": "ARC Prize leaderboard / Grok 4.5 (Low): Relative Human Action Efficiency; total cost $8718.48." } ] }, { "model_id": "claude-opus-5", "benchmark_id": "arc_agi_3", "score": 30.16, "reference_url": "https://arcprize.org/leaderboard", "reported_setting": { "effort": "high" }, "matches_canonical": true, "source_type": "leaderboard", "audit_status": "verified", "notes": "ARC Prize leaderboard / Claude Opus 5 (High): Relative Human Action Efficiency; total cost $20657.4." }, { "model_id": "composer-2.5", "benchmark_id": "terminal_bench", "score": 69.3, "reference_url": "https://cursor.com/blog/composer-2-5", "reported_setting": { "effort": "source does not state", "tools": "agentic", "sampling": "pass@1", "harness": "Cursor" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Composer 2.5 release benchmark image / Terminal-Bench 2.0 / Composer 2.5: Cursor release benchmark image. Opus 4.7 and GPT-5.5 public evaluation values are self-reported." }, { "model_id": "composer-2", "benchmark_id": "terminal_bench", "score": 61.7, "reference_url": "https://cursor.com/blog/composer-2-5", "reported_setting": { "effort": "source does not state", "tools": "agentic", "sampling": "pass@1", "harness": "Cursor" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Composer 2.5 release benchmark image / Terminal-Bench 2.0 / Composer 2: Cursor release benchmark image. Opus 4.7 and GPT-5.5 public evaluation values are self-reported." }, { "model_id": "composer-2.5", "benchmark_id": "swe_bench_multilingual", "score": 79.8, "reference_url": "https://cursor.com/blog/composer-2-5", "reported_setting": { "effort": "source does not state", "tools": "agentic", "sampling": "pass@1", "harness": "Cursor" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Composer 2.5 release benchmark image / SWE-Bench Multilingual / Composer 2.5: Cursor release benchmark image. Opus 4.7 and GPT-5.5 public evaluation values are self-reported." }, { "model_id": "composer-2", "benchmark_id": "swe_bench_multilingual", "score": 73.7, "reference_url": "https://cursor.com/blog/composer-2-5", "reported_setting": { "effort": "source does not state", "tools": "agentic", "sampling": "pass@1", "harness": "Cursor" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Composer 2.5 release benchmark image / SWE-Bench Multilingual / Composer 2: Cursor release benchmark image. Opus 4.7 and GPT-5.5 public evaluation values are self-reported." }, { "model_id": "composer-2.5", "benchmark_id": "cursorbench_3_1", "score": 63.2, "reference_url": "https://cursor.com/blog/composer-2-5", "reported_setting": { "effort": "source does not state", "tools": "agentic", "sampling": "pass@1", "harness": "Cursor" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Composer 2.5 release benchmark image / CursorBench 3.1 / Composer 2.5: Cursor release benchmark image. Opus 4.7 and GPT-5.5 public evaluation values are self-reported." }, { "model_id": "composer-2", "benchmark_id": "cursorbench_3_1", "score": 52.2, "reference_url": "https://cursor.com/blog/composer-2-5", "reported_setting": { "effort": "source does not state", "tools": "agentic", "sampling": "pass@1", "harness": "Cursor" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Composer 2.5 release benchmark image / CursorBench 3.1 / Composer 2: Cursor release benchmark image. Opus 4.7 and GPT-5.5 public evaluation values are self-reported." }, { "model_id": "grok-4.6", "benchmark_id": "cursorbench_3_2", "score": 69.9, "reference_url": "https://x.ai/news/grok-4-6", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "Cursor coding-agent tools", "sampling": "unknown", "judge": "CursorBench task score", "harness": "Cursor first-party agent harness", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "release46 CursorBench v3.2 Score (%); source effort=high; benchmark protocol matches canonical=True. Unknown source fields remain unknown.", "candidates": [ { "score": 70.8, "reference_url": "https://cursor.com/cursorbench", "source_type": "leaderboard", "reported_setting": { "harness": "Cursor agent", "effort": "xhigh" }, "notes": "CursorBench 3.2 current leaderboard / rank 1 / Grok 4.6 Extra High: Average cost/task $ 2.81; tokens/task 41,136; steps/task 46." }, { "score": 67.1, "reference_url": "https://cursor.com/cursorbench", "source_type": "leaderboard", "reported_setting": { "harness": "Cursor agent", "effort": "medium" }, "notes": "CursorBench 3.2 current leaderboard / rank 8 / Grok 4.6 Medium: Average cost/task $ 1.28; tokens/task 17,942; steps/task 29." }, { "score": 61.0, "reference_url": "https://cursor.com/cursorbench", "source_type": "leaderboard", "reported_setting": { "harness": "Cursor agent", "effort": "low" }, "notes": "CursorBench 3.2 current leaderboard / rank 22 / Grok 4.6 Low: Average cost/task $ 0.70; tokens/task 10,658; steps/task 23." }, { "score": 69.9, "reference_url": "https://cursor.com/cursorbench", "source_type": "leaderboard", "reported_setting": { "harness": "Cursor agent", "effort": "high" }, "notes": "CursorBench 3.2 current leaderboard / rank 4 / Grok 4.6 High: Average cost/task $ 2.34; tokens/task 32,449; steps/task 39." } ] }, { "model_id": "gemini-3.7-flash", "benchmark_id": "cursorbench_3_2", "score": 59.0, "reference_url": "https://cursor.com/cursorbench", "reported_setting": { "harness": "Cursor agent", "effort": "medium" }, "matches_canonical": true, "source_type": "leaderboard", "audit_status": "verified", "notes": "CursorBench 3.2 current leaderboard / rank 28 / Gemini 3.7 Flash Medium: Average cost/task $ 0.95; tokens/task 30,953; steps/task 82.", "candidates": [ { "score": 61.6, "reference_url": "https://cursor.com/cursorbench", "source_type": "leaderboard", "reported_setting": { "harness": "Cursor agent", "effort": "high" }, "notes": "CursorBench 3.2 current leaderboard / rank 19 / Gemini 3.7 Flash High: Average cost/task $ 1.20; tokens/task 38,448; steps/task 99." }, { "score": 53.8, "reference_url": "https://cursor.com/cursorbench", "source_type": "leaderboard", "reported_setting": { "harness": "Cursor agent", "effort": "low" }, "notes": "CursorBench 3.2 current leaderboard / rank 41 / Gemini 3.7 Flash Low: Average cost/task $ 0.74; tokens/task 20,594; steps/task 68." } ] }, { "model_id": "composer-2", "benchmark_id": "cursorbench_3_0", "score": 61.3, "reference_url": "https://cursor.com/resources/Composer2.pdf", "reported_setting": { "effort": "source does not state", "tools": "Cursor coding agent", "sampling": "pass@1", "harness": "Cursor" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Composer 2 technical report Table 1 / CursorBench 3.0 / Composer 2 / value 1: For third-party public benchmarks, slash-separated values are Cursor/official-harness then self-reported. SWE prompts prepend 'please solve this github issue'; Terminal prompts add solution-format instructions." }, { "model_id": "composer-1.5", "benchmark_id": "cursorbench_3_0", "score": 44.2, "reference_url": "https://cursor.com/resources/Composer2.pdf", "reported_setting": { "effort": "source does not state", "tools": "Cursor coding agent", "sampling": "pass@1", "harness": "Cursor" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Composer 2 technical report Table 1 / CursorBench 3.0 / Composer 1.5 / value 1: For third-party public benchmarks, slash-separated values are Cursor/official-harness then self-reported. SWE prompts prepend 'please solve this github issue'; Terminal prompts add solution-format instructions.", "candidates": [ { "score": 44.8, "reference_url": "https://cursor.com/resources/Composer2.pdf", "source_type": "tech_report", "reported_setting": { "effort": "source does not state", "tools": "Cursor coding agent", "sampling": "pass@1", "harness": "Cursor" }, "notes": "Composer 2 technical report Figure 1 / CursorBench 3.0 / Composer 1.5: Figure 1 exact printed bar value." } ] }, { "model_id": "composer-1.5", "benchmark_id": "swe_bench_multilingual", "score": 65.9, "reference_url": "https://cursor.com/resources/Composer2.pdf", "reported_setting": { "effort": "source does not state", "tools": "Cursor coding agent", "sampling": "pass@1", "harness": "Cursor harness" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Composer 2 technical report Table 1 / SWE-Bench Multilingual / Composer 1.5 / value 1: For third-party public benchmarks, slash-separated values are Cursor/official-harness then self-reported. SWE prompts prepend 'please solve this github issue'; Terminal prompts add solution-format instructions." }, { "model_id": "composer-1.5", "benchmark_id": "terminal_bench", "score": 47.9, "reference_url": "https://cursor.com/resources/Composer2.pdf", "reported_setting": { "effort": "source does not state", "tools": "Cursor coding agent", "sampling": "pass@1", "harness": "Harbor / Cursor-selected official value" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Composer 2 technical report Table 1 / Terminal-Bench 2.0 / Composer 1.5 / value 1: For third-party public benchmarks, slash-separated values are Cursor/official-harness then self-reported. SWE prompts prepend 'please solve this github issue'; Terminal prompts add solution-format instructions." }, { "model_id": "composer-1", "benchmark_id": "cursorbench_3_0", "score": 38.0, "reference_url": "https://cursor.com/resources/Composer2.pdf", "reported_setting": { "effort": "source does not state", "tools": "Cursor coding agent", "sampling": "pass@1", "harness": "Cursor" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Composer 2 technical report Table 1 / CursorBench 3.0 / Composer 1 / value 1: For third-party public benchmarks, slash-separated values are Cursor/official-harness then self-reported. SWE prompts prepend 'please solve this github issue'; Terminal prompts add solution-format instructions." }, { "model_id": "composer-1", "benchmark_id": "swe_bench_multilingual", "score": 56.9, "reference_url": "https://cursor.com/resources/Composer2.pdf", "reported_setting": { "effort": "source does not state", "tools": "Cursor coding agent", "sampling": "pass@1", "harness": "Cursor harness" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Composer 2 technical report Table 1 / SWE-Bench Multilingual / Composer 1 / value 1: For third-party public benchmarks, slash-separated values are Cursor/official-harness then self-reported. SWE prompts prepend 'please solve this github issue'; Terminal prompts add solution-format instructions." }, { "model_id": "composer-1", "benchmark_id": "terminal_bench", "score": 40.0, "reference_url": "https://cursor.com/resources/Composer2.pdf", "reported_setting": { "effort": "source does not state", "tools": "Cursor coding agent", "sampling": "pass@1", "harness": "Harbor / Cursor-selected official value" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Composer 2 technical report Table 1 / Terminal-Bench 2.0 / Composer 1 / value 1: For third-party public benchmarks, slash-separated values are Cursor/official-harness then self-reported. SWE prompts prepend 'please solve this github issue'; Terminal prompts add solution-format instructions." }, { "model_id": "claude-opus-4.6", "benchmark_id": "cursorbench_3_0", "score": 58.2, "reference_url": "https://cursor.com/resources/Composer2.pdf", "reported_setting": { "effort": "high", "tools": "Cursor coding agent", "sampling": "pass@1", "harness": "Cursor" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Composer 2 technical report Table 1 / CursorBench 3.0 / Opus 4.6 High / value 1: For third-party public benchmarks, slash-separated values are Cursor/official-harness then self-reported. SWE prompts prepend 'please solve this github issue'; Terminal prompts add solution-format instructions." }, { "model_id": "claude-opus-4.5", "benchmark_id": "cursorbench_3_0", "score": 48.4, "reference_url": "https://cursor.com/resources/Composer2.pdf", "reported_setting": { "effort": "high", "tools": "Cursor coding agent", "sampling": "pass@1", "harness": "Cursor" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Composer 2 technical report Table 1 / CursorBench 3.0 / Opus 4.5 High / value 1: For third-party public benchmarks, slash-separated values are Cursor/official-harness then self-reported. SWE prompts prepend 'please solve this github issue'; Terminal prompts add solution-format instructions." }, { "model_id": "gpt-5.4", "benchmark_id": "cursorbench_3_0", "score": 63.9, "reference_url": "https://cursor.com/resources/Composer2.pdf", "reported_setting": { "effort": "high", "tools": "Cursor coding agent", "sampling": "pass@1", "harness": "Cursor" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Composer 2 technical report Table 1 / CursorBench 3.0 / GPT-5.4 / value 1: For third-party public benchmarks, slash-separated values are Cursor/official-harness then self-reported. SWE prompts prepend 'please solve this github issue'; Terminal prompts add solution-format instructions." }, { "model_id": "gpt-5.4", "benchmark_id": "swe_bench_multilingual", "score": 76.8, "reference_url": "https://cursor.com/resources/Composer2.pdf", "reported_setting": { "effort": "high", "tools": "Cursor coding agent", "sampling": "pass@1", "harness": "Cursor harness" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Composer 2 technical report Table 1 / SWE-Bench Multilingual / GPT-5.4 / value 1: For third-party public benchmarks, slash-separated values are Cursor/official-harness then self-reported. SWE prompts prepend 'please solve this github issue'; Terminal prompts add solution-format instructions." }, { "model_id": "gpt-5.3-codex", "benchmark_id": "cursorbench_3_0", "score": 59.1, "reference_url": "https://cursor.com/resources/Composer2.pdf", "reported_setting": { "effort": "high", "tools": "Cursor coding agent", "sampling": "pass@1", "harness": "Cursor" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Composer 2 technical report Table 1 / CursorBench 3.0 / GPT-5.3 Codex / value 1: For third-party public benchmarks, slash-separated values are Cursor/official-harness then self-reported. SWE prompts prepend 'please solve this github issue'; Terminal prompts add solution-format instructions." }, { "model_id": "gpt-5.3-codex", "benchmark_id": "swe_bench_multilingual", "score": 74.8, "reference_url": "https://cursor.com/resources/Composer2.pdf", "reported_setting": { "effort": "high", "tools": "Cursor coding agent", "sampling": "pass@1", "harness": "Cursor harness" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Composer 2 technical report Table 1 / SWE-Bench Multilingual / GPT-5.3 Codex / value 1: For third-party public benchmarks, slash-separated values are Cursor/official-harness then self-reported. SWE prompts prepend 'please solve this github issue'; Terminal prompts add solution-format instructions." }, { "model_id": "gpt-5.2", "benchmark_id": "cursorbench_3_0", "score": 56.5, "reference_url": "https://cursor.com/resources/Composer2.pdf", "reported_setting": { "effort": "high", "tools": "Cursor coding agent", "sampling": "pass@1", "harness": "Cursor" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Composer 2 technical report Table 1 / CursorBench 3.0 / GPT-5.2 / value 1: For third-party public benchmarks, slash-separated values are Cursor/official-harness then self-reported. SWE prompts prepend 'please solve this github issue'; Terminal prompts add solution-format instructions." }, { "model_id": "glm-5", "benchmark_id": "cursorbench_3_0", "score": 42.7, "reference_url": "https://cursor.com/resources/Composer2.pdf", "reported_setting": { "effort": "source does not state", "tools": "Cursor coding agent", "sampling": "pass@1", "harness": "Cursor" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Composer 2 technical report Table 1 / CursorBench 3.0 / GLM-5 / value 1: For third-party public benchmarks, slash-separated values are Cursor/official-harness then self-reported. SWE prompts prepend 'please solve this github issue'; Terminal prompts add solution-format instructions." }, { "model_id": "kimi-k2.5", "benchmark_id": "cursorbench_3_0", "score": 36.0, "reference_url": "https://cursor.com/resources/Composer2.pdf", "reported_setting": { "effort": "source does not state", "tools": "Cursor coding agent", "sampling": "pass@1", "harness": "Cursor" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Composer 2 technical report Table 1 / CursorBench 3.0 / Kimi K2.5 / value 1: For third-party public benchmarks, slash-separated values are Cursor/official-harness then self-reported. SWE prompts prepend 'please solve this github issue'; Terminal prompts add solution-format instructions." }, { "model_id": "muse-spark-1.1", "benchmark_id": "wmdp_bio_accuracy", "score": 89.0, "reference_url": "https://ai.meta.com/static-resource/muse-spark-1.1-evaluation-report", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "benchmark-specified", "sampling": "source-specific", "harness": "Meta safety/preparedness evaluation", "temperature": "1.0 for Muse; provider/source setting for comparisons", "context": "source-specific", "configuration": "default", "notes": "Capability scorecard; Muse evaluated without system prompt; high reasoning." }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Table 1 · p6; metric=accuracy_percent. Claude cyber-benchmark cells omitted where refusal compromised capability measurement." }, { "model_id": "muse-spark", "benchmark_id": "wmdp_bio_accuracy", "score": 88.4, "reference_url": "https://ai.meta.com/static-resource/muse-spark-1.1-evaluation-report", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "benchmark-specified", "sampling": "source-specific", "harness": "Meta safety/preparedness evaluation", "temperature": "1.0 for Muse; provider/source setting for comparisons", "context": "source-specific", "configuration": "default", "notes": "Capability scorecard; Muse evaluated without system prompt; high reasoning." }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Table 1 · p6; metric=accuracy_percent. Claude cyber-benchmark cells omitted where refusal compromised capability measurement." }, { "model_id": "gpt-5.5", "benchmark_id": "wmdp_bio_accuracy", "score": 90.4, "reference_url": "https://ai.meta.com/static-resource/muse-spark-1.1-evaluation-report", "reported_setting": { "mode": "thinking", "effort": "xhigh", "tools": "benchmark-specified", "sampling": "source-specific", "harness": "Meta safety/preparedness evaluation", "temperature": "1.0 for Muse; provider/source setting for comparisons", "context": "source-specific", "configuration": "default", "notes": "Capability scorecard; Muse evaluated without system prompt; high reasoning." }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Table 1 · p6; metric=accuracy_percent. Claude cyber-benchmark cells omitted where refusal compromised capability measurement." }, { "model_id": "gemini-3.1-pro", "benchmark_id": "wmdp_bio_accuracy", "score": 89.5, "reference_url": "https://ai.meta.com/static-resource/muse-spark-1.1-evaluation-report", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "benchmark-specified", "sampling": "source-specific", "harness": "Meta safety/preparedness evaluation", "temperature": "1.0 for Muse; provider/source setting for comparisons", "context": "source-specific", "configuration": "default", "notes": "Capability scorecard; Muse evaluated without system prompt; high reasoning." }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Table 1 · p6; metric=accuracy_percent. Claude cyber-benchmark cells omitted where refusal compromised capability measurement." }, { "model_id": "muse-spark-1.1", "benchmark_id": "wmdp_chem_accuracy", "score": 87.0, "reference_url": "https://ai.meta.com/static-resource/muse-spark-1.1-evaluation-report", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "benchmark-specified", "sampling": "source-specific", "harness": "Meta safety/preparedness evaluation", "temperature": "1.0 for Muse; provider/source setting for comparisons", "context": "source-specific", "configuration": "default", "notes": "Capability scorecard; Muse evaluated without system prompt; high reasoning." }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Table 1 · p6; metric=accuracy_percent. Claude cyber-benchmark cells omitted where refusal compromised capability measurement." }, { "model_id": "muse-spark", "benchmark_id": "wmdp_chem_accuracy", "score": 85.6, "reference_url": "https://ai.meta.com/static-resource/muse-spark-1.1-evaluation-report", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "benchmark-specified", "sampling": "source-specific", "harness": "Meta safety/preparedness evaluation", "temperature": "1.0 for Muse; provider/source setting for comparisons", "context": "source-specific", "configuration": "default", "notes": "Capability scorecard; Muse evaluated without system prompt; high reasoning." }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Table 1 · p6; metric=accuracy_percent. Claude cyber-benchmark cells omitted where refusal compromised capability measurement." }, { "model_id": "gpt-5.5", "benchmark_id": "wmdp_chem_accuracy", "score": 85.5, "reference_url": "https://ai.meta.com/static-resource/muse-spark-1.1-evaluation-report", "reported_setting": { "mode": "thinking", "effort": "xhigh", "tools": "benchmark-specified", "sampling": "source-specific", "harness": "Meta safety/preparedness evaluation", "temperature": "1.0 for Muse; provider/source setting for comparisons", "context": "source-specific", "configuration": "default", "notes": "Capability scorecard; Muse evaluated without system prompt; high reasoning." }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Table 1 · p6; metric=accuracy_percent. Claude cyber-benchmark cells omitted where refusal compromised capability measurement." }, { "model_id": "gemini-3.1-pro", "benchmark_id": "wmdp_chem_accuracy", "score": 86.3, "reference_url": "https://ai.meta.com/static-resource/muse-spark-1.1-evaluation-report", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "benchmark-specified", "sampling": "source-specific", "harness": "Meta safety/preparedness evaluation", "temperature": "1.0 for Muse; provider/source setting for comparisons", "context": "source-specific", "configuration": "default", "notes": "Capability scorecard; Muse evaluated without system prompt; high reasoning." }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Table 1 · p6; metric=accuracy_percent. Claude cyber-benchmark cells omitted where refusal compromised capability measurement." }, { "model_id": "muse-spark-1.1", "benchmark_id": "lab_bench_protocolqa_accuracy", "score": 88.0, "reference_url": "https://ai.meta.com/static-resource/muse-spark-1.1-evaluation-report", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "benchmark-specified", "sampling": "source-specific", "harness": "Meta safety/preparedness evaluation", "temperature": "1.0 for Muse; provider/source setting for comparisons", "context": "source-specific", "configuration": "default", "notes": "Capability scorecard; Muse evaluated without system prompt; high reasoning." }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Table 1 · p6; metric=accuracy_percent. Claude cyber-benchmark cells omitted where refusal compromised capability measurement." }, { "model_id": "muse-spark", "benchmark_id": "lab_bench_protocolqa_accuracy", "score": 87.3, "reference_url": "https://ai.meta.com/static-resource/muse-spark-1.1-evaluation-report", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "benchmark-specified", "sampling": "source-specific", "harness": "Meta safety/preparedness evaluation", "temperature": "1.0 for Muse; provider/source setting for comparisons", "context": "source-specific", "configuration": "default", "notes": "Capability scorecard; Muse evaluated without system prompt; high reasoning." }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Table 1 · p6; metric=accuracy_percent. Claude cyber-benchmark cells omitted where refusal compromised capability measurement." }, { "model_id": "gpt-5.5", "benchmark_id": "lab_bench_protocolqa_accuracy", "score": 78.2, "reference_url": "https://ai.meta.com/static-resource/muse-spark-1.1-evaluation-report", "reported_setting": { "mode": "thinking", "effort": "xhigh", "tools": "benchmark-specified", "sampling": "source-specific", "harness": "Meta safety/preparedness evaluation", "temperature": "1.0 for Muse; provider/source setting for comparisons", "context": "source-specific", "configuration": "default", "notes": "Capability scorecard; Muse evaluated without system prompt; high reasoning." }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Table 1 · p6; metric=accuracy_percent. Claude cyber-benchmark cells omitted where refusal compromised capability measurement." }, { "model_id": "gemini-3.1-pro", "benchmark_id": "lab_bench_protocolqa_accuracy", "score": 88.9, "reference_url": "https://ai.meta.com/static-resource/muse-spark-1.1-evaluation-report", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "benchmark-specified", "sampling": "source-specific", "harness": "Meta safety/preparedness evaluation", "temperature": "1.0 for Muse; provider/source setting for comparisons", "context": "source-specific", "configuration": "default", "notes": "Capability scorecard; Muse evaluated without system prompt; high reasoning." }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Table 1 · p6; metric=accuracy_percent. Claude cyber-benchmark cells omitted where refusal compromised capability measurement." }, { "model_id": "muse-spark-1.1", "benchmark_id": "cybench", "score": 92.9, "reference_url": "https://ai.meta.com/static-resource/muse-spark-1.1-evaluation-report", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "benchmark-specified", "sampling": "pass@1", "harness": "Meta safety/preparedness evaluation", "temperature": "1.0 for Muse; provider/source setting for comparisons", "context": "source-specific", "configuration": "default", "notes": "Capability scorecard; Muse evaluated without system prompt; high reasoning." }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Table 1 · p6; metric=pass@1_percent. Claude cyber-benchmark cells omitted where refusal compromised capability measurement.", "candidates": [ { "score": 97.0, "reference_url": "https://ai.meta.com/static-resource/muse-spark-1.1-evaluation-report", "source_type": "tech_report", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "benchmark-specified", "sampling": "pass@10", "harness": "Meta safety/preparedness evaluation", "temperature": "1.0 for Muse; provider/source setting for comparisons", "context": "source-specific", "configuration": "default", "notes": "Only printed endpoints; intermediate curve points unlabeled." }, "notes": "Figure 8 · p21; metric=pass@10_percent." }, { "score": 42.5, "reference_url": "https://ai.meta.com/static-resource/muse-spark-1.1-evaluation-report", "source_type": "tech_report", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "benchmark-specified", "sampling": "success by 10,000 output tokens", "harness": "Meta safety/preparedness evaluation", "temperature": "1.0 for Muse; provider/source setting for comparisons", "context": "source-specific", "configuration": "default", "notes": "Only printed endpoints; intermediate curve points unlabeled." }, "notes": "Figure 8 · p21; metric=10k_token_success_percent." } ] }, { "model_id": "muse-spark", "benchmark_id": "cybench", "score": 65.4, "reference_url": "https://ai.meta.com/static-resource/muse-spark-1.1-evaluation-report", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "benchmark-specified", "sampling": "pass@1", "harness": "Meta safety/preparedness evaluation", "temperature": "1.0 for Muse; provider/source setting for comparisons", "context": "source-specific", "configuration": "default", "notes": "Capability scorecard; Muse evaluated without system prompt; high reasoning." }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Table 1 · p6; metric=pass@1_percent. Claude cyber-benchmark cells omitted where refusal compromised capability measurement.", "candidates": [ { "score": 79.0, "reference_url": "https://ai.meta.com/static-resource/muse-spark-1.1-evaluation-report", "source_type": "tech_report", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "benchmark-specified", "sampling": "pass@10", "harness": "Meta safety/preparedness evaluation", "temperature": "1.0 for Muse; provider/source setting for comparisons", "context": "source-specific", "configuration": "default", "notes": "Only printed endpoints; intermediate curve points unlabeled." }, "notes": "Figure 8 · p21; metric=pass@10_percent." }, { "score": 35.4, "reference_url": "https://ai.meta.com/static-resource/muse-spark-1.1-evaluation-report", "source_type": "tech_report", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "benchmark-specified", "sampling": "success by 10,000 output tokens", "harness": "Meta safety/preparedness evaluation", "temperature": "1.0 for Muse; provider/source setting for comparisons", "context": "source-specific", "configuration": "default", "notes": "Only printed endpoints; intermediate curve points unlabeled." }, "notes": "Figure 8 · p21; metric=10k_token_success_percent." } ] }, { "model_id": "gpt-5.5", "benchmark_id": "cybench", "score": 100.0, "reference_url": "https://ai.meta.com/static-resource/muse-spark-1.1-evaluation-report", "reported_setting": { "mode": "thinking", "effort": "xhigh", "tools": "benchmark-specified", "sampling": "pass@1", "harness": "Meta safety/preparedness evaluation", "temperature": "1.0 for Muse; provider/source setting for comparisons", "context": "source-specific", "configuration": "default", "notes": "Capability scorecard; Muse evaluated without system prompt; high reasoning." }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Table 1 · p6; metric=pass@1_percent. Claude cyber-benchmark cells omitted where refusal compromised capability measurement." }, { "model_id": "claude-opus-4.8", "benchmark_id": "cybench", "score": 95.0, "reference_url": "https://ai.meta.com/static-resource/muse-spark-1.1-evaluation-report", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "benchmark-specified", "sampling": "pass@1", "harness": "Meta safety/preparedness evaluation", "temperature": "1.0 for Muse; provider/source setting for comparisons", "context": "source-specific", "configuration": "default", "notes": "Capability scorecard; Muse evaluated without system prompt; high reasoning." }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Table 1 · p6; metric=pass@1_percent. Claude cyber-benchmark cells omitted where refusal compromised capability measurement." }, { "model_id": "muse-spark-1.1", "benchmark_id": "cybergym", "score": 59.0, "reference_url": "https://ai.meta.com/static-resource/muse-spark-1.1-evaluation-report", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "benchmark-specified", "sampling": "pass@1", "harness": "Meta safety/preparedness evaluation", "temperature": "1.0 for Muse; provider/source setting for comparisons", "context": "source-specific", "configuration": "default", "notes": "Capability scorecard; Muse evaluated without system prompt; high reasoning." }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Table 1 · p6; metric=pass@1_percent. Claude cyber-benchmark cells omitted where refusal compromised capability measurement." }, { "model_id": "muse-spark", "benchmark_id": "cybergym", "score": 43.5, "reference_url": "https://ai.meta.com/static-resource/muse-spark-1.1-evaluation-report", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "benchmark-specified", "sampling": "pass@1", "harness": "Meta safety/preparedness evaluation", "temperature": "1.0 for Muse; provider/source setting for comparisons", "context": "source-specific", "configuration": "default", "notes": "Capability scorecard; Muse evaluated without system prompt; high reasoning." }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Table 1 · p6; metric=pass@1_percent. Claude cyber-benchmark cells omitted where refusal compromised capability measurement." }, { "model_id": "muse-spark-1.1", "benchmark_id": "exploitgym", "score": 0.8, "reference_url": "https://ai.meta.com/static-resource/muse-spark-1.1-evaluation-report", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "benchmark-specified", "sampling": "pass@1", "harness": "Meta safety/preparedness evaluation", "temperature": "1.0 for Muse; provider/source setting for comparisons", "context": "source-specific", "configuration": "default", "notes": "Capability scorecard; Muse evaluated without system prompt; high reasoning." }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Table 1 · p6; metric=pass@1_percent. Claude cyber-benchmark cells omitted where refusal compromised capability measurement.", "candidates": [ { "score": 0.6, "reference_url": "https://ai.meta.com/static-resource/muse-spark-1.1-evaluation-report", "source_type": "tech_report", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "benchmark-specified", "sampling": "source-specific", "harness": "Meta safety/preparedness evaluation", "temperature": "1.0 for Muse; provider/source setting for comparisons", "context": "source-specific", "configuration": "default", "notes": "Prose rounded rate." }, "notes": "Section 2 · p24; metric=2h_pass_rate_percent. Two-hour result differs from Table 1; four-hour result aligns with Table 1 after rounding." } ] }, { "model_id": "gemini-3.1-pro", "benchmark_id": "exploitgym", "score": 1.4, "reference_url": "https://ai.meta.com/static-resource/muse-spark-1.1-evaluation-report", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "benchmark-specified", "sampling": "pass@1", "harness": "Meta safety/preparedness evaluation", "temperature": "1.0 for Muse; provider/source setting for comparisons", "context": "source-specific", "configuration": "default", "notes": "Capability scorecard; Muse evaluated without system prompt; high reasoning." }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Table 1 · p6; metric=pass@1_percent. Claude cyber-benchmark cells omitted where refusal compromised capability measurement." }, { "model_id": "kimi-k2.5", "benchmark_id": "wmdp_bio_accuracy", "score": 86.8, "reference_url": "https://ai.meta.com/static-resource/muse-spark-1.1-evaluation-report", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "benchmark-specified", "sampling": "source-specific", "harness": "Meta safety/preparedness evaluation", "temperature": "1.0 for Muse; provider/source setting for comparisons", "context": "source-specific", "configuration": "default", "notes": "Exact bar labels." }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Figure 2 · p13; metric=accuracy_percent." }, { "model_id": "kimi-k2.5", "benchmark_id": "wmdp_chem_accuracy", "score": 82.0, "reference_url": "https://ai.meta.com/static-resource/muse-spark-1.1-evaluation-report", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "benchmark-specified", "sampling": "source-specific", "harness": "Meta safety/preparedness evaluation", "temperature": "1.0 for Muse; provider/source setting for comparisons", "context": "source-specific", "configuration": "default", "notes": "Exact bar labels." }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Figure 2 · p13; metric=accuracy_percent." }, { "model_id": "kimi-k2.5", "benchmark_id": "lab_bench_protocolqa_accuracy", "score": 79.5, "reference_url": "https://ai.meta.com/static-resource/muse-spark-1.1-evaluation-report", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "benchmark-specified", "sampling": "source-specific", "harness": "Meta safety/preparedness evaluation", "temperature": "1.0 for Muse; provider/source setting for comparisons", "context": "source-specific", "configuration": "default", "notes": "Muse Spark and Kimi include an abstention option; accuracy excludes abstentions." }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Figure 3 · p14; metric=accuracy_percent. Asterisk marks abstention-option variants." }, { "model_id": "muse-spark-1.1", "benchmark_id": "hle_tools", "score": 62.1, "reference_url": "https://ai.meta.com/static-resource/muse-spark-1.1-evaluation-report", "reported_setting": { "mode": "thinking", "effort": "xhigh", "tools": "general tools", "sampling": "pass@1 unless source row states otherwise", "harness": "Meta MSL general-capability evaluation", "temperature": "1.0 for Muse; provider/source setting for comparisons", "context": "benchmark-specific", "configuration": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Figure 44 · HLE w/tools; metric=headline_metric. Exact printed value in the general-capability summary." }, { "model_id": "muse-spark-1.1", "benchmark_id": "hle", "score": 52.2, "reference_url": "https://ai.meta.com/static-resource/muse-spark-1.1-evaluation-report", "reported_setting": { "mode": "thinking", "effort": "xhigh", "tools": "none", "sampling": "pass@1 unless source row states otherwise", "harness": "Meta MSL general-capability evaluation", "temperature": "1.0 for Muse; provider/source setting for comparisons", "context": "benchmark-specific", "configuration": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Figure 44 · HLE no tools; metric=headline_metric. Exact printed value in the general-capability summary." }, { "model_id": "muse-spark-1.1", "benchmark_id": "mrcr_v2_8needle_512k_1m", "score": 54.1, "reference_url": "https://ai.meta.com/static-resource/muse-spark-1.1-evaluation-report", "reported_setting": { "mode": "thinking", "effort": "xhigh", "tools": "none", "sampling": "pass@1 unless source row states otherwise", "harness": "Meta MSL general-capability evaluation", "temperature": "1.0 for Muse; provider/source setting for comparisons", "context": "benchmark-specific", "configuration": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Figure 44 · MRCR v2 8-needle, 512K-1M; metric=headline_metric. Exact printed value in the general-capability summary." }, { "model_id": "gemini-3.1-pro", "benchmark_id": "mrcr_v2_8needle_512k_1m", "score": 26.3, "reference_url": "https://ai.meta.com/static-resource/muse-spark-1.1-evaluation-report", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1 unless source row states otherwise", "harness": "Meta MSL general-capability evaluation", "temperature": "1.0 for Muse; provider/source setting for comparisons", "context": "benchmark-specific", "configuration": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Figure 44 · MRCR v2 8-needle, 512K-1M; metric=headline_metric. Exact printed value in the general-capability summary." }, { "model_id": "gpt-5.5", "benchmark_id": "mrcr_v2_8needle_512k_1m", "score": 74.0, "reference_url": "https://ai.meta.com/static-resource/muse-spark-1.1-evaluation-report", "reported_setting": { "mode": "thinking", "effort": "xhigh", "tools": "none", "sampling": "pass@1 unless source row states otherwise", "harness": "Meta MSL general-capability evaluation", "temperature": "1.0 for Muse; provider/source setting for comparisons", "context": "benchmark-specific", "configuration": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Figure 44 · MRCR v2 8-needle, 512K-1M; metric=headline_metric. Exact printed value in the general-capability summary." }, { "model_id": "muse-spark-1.1", "benchmark_id": "mcpatlas_full_1000", "score": 88.1, "reference_url": "https://ai.meta.com/static-resource/muse-spark-1.1-evaluation-report", "reported_setting": { "mode": "thinking", "effort": "xhigh", "tools": "MCP tool servers", "sampling": "pass@1 unless source row states otherwise", "harness": "Meta MSL general-capability evaluation", "temperature": "1.0 for Muse; provider/source setting for comparisons", "context": "benchmark-specific", "configuration": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Figure 44 · MCP-Atlas full 1,000 tasks; metric=headline_metric. Exact printed value in the general-capability summary.", "candidates": [ { "score": 88.1, "reference_url": "https://research.meta.ai/articles/introducing-muse-code-and-muse-spark-1-2/evaluations/mcp-atlas-v1.png", "source_type": "official_blog", "reported_setting": { "mode": "thinking", "effort": "xhigh", "prompt_style": "not disclosed", "temperature": "not disclosed", "context": "not disclosed", "agent": "Scale AI MCP-Atlas agent harness", "attempts": 1, "tools": "controlled target and distractor tools across 36 MCP servers and 220 tools", "sampling": "pass@1 over all 1,000 tasks", "judge": "claim-level 1/0.5/0; pass at coverage >=0.75; Gemini 3.1 Pro Preview primary", "harness": "Scale AI MCP-Atlas agent harness and scoring pipeline", "internet": "containerized real MCP servers under benchmark controls", "configuration": "500 public plus 500 held-out private tasks" }, "notes": "Exact printed value from the locked Meta release chart; evaluation settings come from the locked official methodology." } ] }, { "model_id": "muse-spark", "benchmark_id": "mcpatlas_full_1000", "score": 82.2, "reference_url": "https://ai.meta.com/static-resource/muse-spark-1.1-evaluation-report", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "MCP tool servers", "sampling": "pass@1 unless source row states otherwise", "harness": "Meta MSL general-capability evaluation", "temperature": "1.0 for Muse; provider/source setting for comparisons", "context": "benchmark-specific", "configuration": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Figure 44 · MCP-Atlas full 1,000 tasks; metric=headline_metric. Exact printed value in the general-capability summary." }, { "model_id": "gemini-3.1-pro", "benchmark_id": "mcpatlas_full_1000", "score": 78.2, "reference_url": "https://ai.meta.com/static-resource/muse-spark-1.1-evaluation-report", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "MCP tool servers", "sampling": "pass@1 unless source row states otherwise", "harness": "Meta MSL general-capability evaluation", "temperature": "1.0 for Muse; provider/source setting for comparisons", "context": "benchmark-specific", "configuration": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Figure 44 · MCP-Atlas full 1,000 tasks; metric=headline_metric. Exact printed value in the general-capability summary." }, { "model_id": "claude-opus-4.8", "benchmark_id": "mcpatlas_full_1000", "score": 82.2, "reference_url": "https://ai.meta.com/static-resource/muse-spark-1.1-evaluation-report", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "MCP tool servers", "sampling": "pass@1 unless source row states otherwise", "harness": "Meta MSL general-capability evaluation", "temperature": "1.0 for Muse; provider/source setting for comparisons", "context": "benchmark-specific", "configuration": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Figure 44 · MCP-Atlas full 1,000 tasks; metric=headline_metric. Exact printed value in the general-capability summary." }, { "model_id": "gpt-5.5", "benchmark_id": "mcpatlas_full_1000", "score": 75.3, "reference_url": "https://ai.meta.com/static-resource/muse-spark-1.1-evaluation-report", "reported_setting": { "mode": "thinking", "effort": "xhigh", "tools": "MCP tool servers", "sampling": "pass@1 unless source row states otherwise", "harness": "Meta MSL general-capability evaluation", "temperature": "1.0 for Muse; provider/source setting for comparisons", "context": "benchmark-specific", "configuration": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Figure 44 · MCP-Atlas full 1,000 tasks; metric=headline_metric. Exact printed value in the general-capability summary." }, { "model_id": "muse-spark-1.1", "benchmark_id": "toolathlon_verified", "score": 75.6, "reference_url": "https://ai.meta.com/static-resource/muse-spark-1.1-evaluation-report", "reported_setting": { "mode": "thinking", "effort": "xhigh", "tools": "benchmark tools", "sampling": "pass@1 unless source row states otherwise", "harness": "Meta MSL general-capability evaluation", "temperature": "1.0 for Muse; provider/source setting for comparisons", "context": "benchmark-specific", "configuration": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Figure 44 · Toolathlon-Verified; metric=headline_metric. Exact printed value in the general-capability summary." }, { "model_id": "muse-spark", "benchmark_id": "toolathlon_verified", "score": 49.4, "reference_url": "https://ai.meta.com/static-resource/muse-spark-1.1-evaluation-report", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "benchmark tools", "sampling": "pass@1 unless source row states otherwise", "harness": "Meta MSL general-capability evaluation", "temperature": "1.0 for Muse; provider/source setting for comparisons", "context": "benchmark-specific", "configuration": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Figure 44 · Toolathlon-Verified; metric=headline_metric. Exact printed value in the general-capability summary." }, { "model_id": "gemini-3.1-pro", "benchmark_id": "toolathlon_verified", "score": 61.1, "reference_url": "https://ai.meta.com/static-resource/muse-spark-1.1-evaluation-report", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "benchmark tools", "sampling": "pass@1 unless source row states otherwise", "harness": "Meta MSL general-capability evaluation", "temperature": "1.0 for Muse; provider/source setting for comparisons", "context": "benchmark-specific", "configuration": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Figure 44 · Toolathlon-Verified; metric=headline_metric. Exact printed value in the general-capability summary." }, { "model_id": "muse-spark-1.1", "benchmark_id": "osworld_verified", "score": 80.8, "reference_url": "https://ai.meta.com/static-resource/muse-spark-1.1-evaluation-report", "reported_setting": { "mode": "thinking", "effort": "xhigh", "tools": "computer use", "sampling": "pass@1 unless source row states otherwise", "harness": "Meta MSL general-capability evaluation", "temperature": "1.0 for Muse; provider/source setting for comparisons", "context": "benchmark-specific", "configuration": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Figure 44 · OSWorld-Verified; metric=headline_metric. Exact printed value in the general-capability summary." }, { "model_id": "muse-spark", "benchmark_id": "osworld_verified", "score": 53.3, "reference_url": "https://ai.meta.com/static-resource/muse-spark-1.1-evaluation-report", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "computer use", "sampling": "pass@1 unless source row states otherwise", "harness": "Meta MSL general-capability evaluation", "temperature": "1.0 for Muse; provider/source setting for comparisons", "context": "benchmark-specific", "configuration": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Figure 44 · OSWorld-Verified; metric=headline_metric. Exact printed value in the general-capability summary." }, { "model_id": "gemini-3.1-pro", "benchmark_id": "osworld_verified", "score": 76.2, "reference_url": "https://ai.meta.com/static-resource/muse-spark-1.1-evaluation-report", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "computer use", "sampling": "pass@1 unless source row states otherwise", "harness": "Meta MSL general-capability evaluation", "temperature": "1.0 for Muse; provider/source setting for comparisons", "context": "benchmark-specific", "configuration": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Figure 44 · OSWorld-Verified; metric=headline_metric. Exact printed value in the general-capability summary." }, { "model_id": "muse-spark-1.1", "benchmark_id": "osworld_2_0_binary", "score": 14.2, "reference_url": "https://ai.meta.com/static-resource/muse-spark-1.1-evaluation-report", "reported_setting": { "mode": "thinking", "effort": "xhigh", "tools": "computer use", "sampling": "pass@1 unless source row states otherwise", "harness": "Meta MSL general-capability evaluation", "temperature": "1.0 for Muse; provider/source setting for comparisons", "context": "benchmark-specific", "configuration": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Figure 44 · OSWorld 2.0 binary; metric=headline_metric. Exact printed value in the general-capability summary." }, { "model_id": "gemini-3.1-pro", "benchmark_id": "osworld_2_0_binary", "score": 7.8, "reference_url": "https://ai.meta.com/static-resource/muse-spark-1.1-evaluation-report", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "computer use", "sampling": "pass@1 unless source row states otherwise", "harness": "Meta MSL general-capability evaluation", "temperature": "1.0 for Muse; provider/source setting for comparisons", "context": "benchmark-specific", "configuration": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Figure 44 · OSWorld 2.0 binary; metric=headline_metric. Exact printed value in the general-capability summary." }, { "model_id": "claude-opus-4.8", "benchmark_id": "osworld_2_0_binary", "score": 20.6, "reference_url": "https://ai.meta.com/static-resource/muse-spark-1.1-evaluation-report", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "computer use", "sampling": "pass@1 unless source row states otherwise", "harness": "Meta MSL general-capability evaluation", "temperature": "1.0 for Muse; provider/source setting for comparisons", "context": "benchmark-specific", "configuration": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Figure 44 · OSWorld 2.0 binary; metric=headline_metric. Exact printed value in the general-capability summary." }, { "model_id": "gpt-5.5", "benchmark_id": "osworld_2_0_binary", "score": 13.9, "reference_url": "https://ai.meta.com/static-resource/muse-spark-1.1-evaluation-report", "reported_setting": { "mode": "thinking", "effort": "xhigh", "tools": "computer use", "sampling": "pass@1 unless source row states otherwise", "harness": "Meta MSL general-capability evaluation", "temperature": "1.0 for Muse; provider/source setting for comparisons", "context": "benchmark-specific", "configuration": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Figure 44 · OSWorld 2.0 binary; metric=headline_metric. Exact printed value in the general-capability summary." }, { "model_id": "muse-spark-1.1", "benchmark_id": "osworld_2_0", "score": 47.3, "reference_url": "https://ai.meta.com/static-resource/muse-spark-1.1-evaluation-report", "reported_setting": { "mode": "thinking", "effort": "xhigh", "tools": "computer use", "sampling": "pass@1 unless source row states otherwise", "harness": "Meta MSL general-capability evaluation", "temperature": "1.0 for Muse; provider/source setting for comparisons", "context": "benchmark-specific", "configuration": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Figure 44 · OSWorld 2.0 partial; metric=headline_metric. Exact printed value in the general-capability summary." }, { "model_id": "gemini-3.1-pro", "benchmark_id": "osworld_2_0", "score": 30.6, "reference_url": "https://ai.meta.com/static-resource/muse-spark-1.1-evaluation-report", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "computer use", "sampling": "pass@1 unless source row states otherwise", "harness": "Meta MSL general-capability evaluation", "temperature": "1.0 for Muse; provider/source setting for comparisons", "context": "benchmark-specific", "configuration": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Figure 44 · OSWorld 2.0 partial; metric=headline_metric. Exact printed value in the general-capability summary." }, { "model_id": "muse-spark-1.1", "benchmark_id": "webarena_verified_full", "score": 69.0, "reference_url": "https://ai.meta.com/static-resource/muse-spark-1.1-evaluation-report", "reported_setting": { "mode": "thinking", "effort": "xhigh", "tools": "browser", "sampling": "pass@1 unless source row states otherwise", "harness": "Meta MSL general-capability evaluation", "temperature": "1.0 for Muse; provider/source setting for comparisons", "context": "benchmark-specific", "configuration": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Figure 44 · WebArena-Verified; metric=headline_metric. Exact printed value in the general-capability summary." }, { "model_id": "muse-spark", "benchmark_id": "webarena_verified_full", "score": 59.0, "reference_url": "https://ai.meta.com/static-resource/muse-spark-1.1-evaluation-report", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "browser", "sampling": "pass@1 unless source row states otherwise", "harness": "Meta MSL general-capability evaluation", "temperature": "1.0 for Muse; provider/source setting for comparisons", "context": "benchmark-specific", "configuration": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Figure 44 · WebArena-Verified; metric=headline_metric. Exact printed value in the general-capability summary." }, { "model_id": "gemini-3.1-pro", "benchmark_id": "webarena_verified_full", "score": 69.0, "reference_url": "https://ai.meta.com/static-resource/muse-spark-1.1-evaluation-report", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "browser", "sampling": "pass@1 unless source row states otherwise", "harness": "Meta MSL general-capability evaluation", "temperature": "1.0 for Muse; provider/source setting for comparisons", "context": "benchmark-specific", "configuration": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Figure 44 · WebArena-Verified; metric=headline_metric. Exact printed value in the general-capability summary." }, { "model_id": "claude-opus-4.8", "benchmark_id": "webarena_verified_full", "score": 71.2, "reference_url": "https://ai.meta.com/static-resource/muse-spark-1.1-evaluation-report", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "browser", "sampling": "pass@1 unless source row states otherwise", "harness": "Meta MSL general-capability evaluation", "temperature": "1.0 for Muse; provider/source setting for comparisons", "context": "benchmark-specific", "configuration": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Figure 44 · WebArena-Verified; metric=headline_metric. Exact printed value in the general-capability summary." }, { "model_id": "gpt-5.5", "benchmark_id": "webarena_verified_full", "score": 67.0, "reference_url": "https://ai.meta.com/static-resource/muse-spark-1.1-evaluation-report", "reported_setting": { "mode": "thinking", "effort": "xhigh", "tools": "browser", "sampling": "pass@1 unless source row states otherwise", "harness": "Meta MSL general-capability evaluation", "temperature": "1.0 for Muse; provider/source setting for comparisons", "context": "benchmark-specific", "configuration": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Figure 44 · WebArena-Verified; metric=headline_metric. Exact printed value in the general-capability summary." }, { "model_id": "muse-spark-1.1", "benchmark_id": "deepsearchqa_f1", "score": 84.9, "reference_url": "https://ai.meta.com/static-resource/muse-spark-1.1-evaluation-report", "reported_setting": { "mode": "thinking", "effort": "xhigh", "tools": "search/browser", "sampling": "pass@1 unless source row states otherwise", "harness": "Meta MSL general-capability evaluation", "temperature": "1.0 for Muse; provider/source setting for comparisons", "context": "benchmark-specific", "configuration": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Figure 44 · DeepSearchQA F1; metric=headline_metric. Exact printed value in the general-capability summary." }, { "model_id": "muse-spark-1.1", "benchmark_id": "gdpval_aa_v2_elo", "score": 1381, "reference_url": "https://ai.meta.com/static-resource/muse-spark-1.1-evaluation-report", "reported_setting": { "mode": "thinking", "effort": "xhigh", "tools": "benchmark-specified", "sampling": "pass@1 unless source row states otherwise", "harness": "Meta MSL general-capability evaluation", "temperature": "1.0 for Muse; provider/source setting for comparisons", "context": "benchmark-specific", "configuration": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Figure 44 · GDPval-AA v2 Elo; metric=headline_metric. Exact printed value in the general-capability summary.", "candidates": [ { "score": 1371, "reference_url": "https://research.meta.ai/articles/introducing-muse-code-and-muse-spark-1-2/evaluations/gdpval-aa-v2-v1.png", "source_type": "official_blog", "reported_setting": { "mode": "thinking", "effort": "xhigh", "prompt_style": "not disclosed", "temperature": "not disclosed", "context": "not disclosed", "agent": "Artificial Analysis Stirrup", "attempts": 1, "tools": "Web Fetch, Web Search, View Image, Code Exec, Finish, Abandon Task", "sampling": "one agentic submission per each of 220 tasks", "judge": "blind pairwise panel of three frontier LLM judges; Bradley-Terry Elo", "harness": "Artificial Analysis Stirrup; fresh E2B sandbox; 250-turn limit", "internet": "web fetch and Brave-backed web search available", "configuration": "fresh E2B sandbox per task; 250-turn limit" }, "notes": "Exact printed value from the locked Meta release chart; evaluation settings come from the locked official methodology." } ] }, { "model_id": "muse-spark", "benchmark_id": "gdpval_aa_v2_elo", "score": 1145, "reference_url": "https://ai.meta.com/static-resource/muse-spark-1.1-evaluation-report", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "benchmark-specified", "sampling": "pass@1 unless source row states otherwise", "harness": "Meta MSL general-capability evaluation", "temperature": "1.0 for Muse; provider/source setting for comparisons", "context": "benchmark-specific", "configuration": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Figure 44 · GDPval-AA v2 Elo; metric=headline_metric. Exact printed value in the general-capability summary." }, { "model_id": "gemini-3.1-pro", "benchmark_id": "gdpval_aa_v2_elo", "score": 963, "reference_url": "https://ai.meta.com/static-resource/muse-spark-1.1-evaluation-report", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "benchmark-specified", "sampling": "pass@1 unless source row states otherwise", "harness": "Meta MSL general-capability evaluation", "temperature": "1.0 for Muse; provider/source setting for comparisons", "context": "benchmark-specific", "configuration": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Figure 44 · GDPval-AA v2 Elo; metric=headline_metric. Exact printed value in the general-capability summary.", "candidates": [ { "score": 1314, "reference_url": "https://deepmind.google/models/model-cards/gemini-3-5-flash/", "source_type": "official_model_card", "reported_setting": { "mode": "thinking/reasoning", "effort": "source does not state", "tools": "Stirrup/E2B agent sandbox", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "Artificial Analysis GDPval-AA v2 leaderboard", "temperature": "default", "snapshot": "May 2026" }, "notes": "Google Gemini 3.5 Flash May 2026 table: GDPval-AA v2; Artificial Analysis Elo. Settings and provenance are preserved per cell." }, { "score": 965, "reference_url": "https://deepmind.google/models/model-cards/gemini-3-6-flash/", "source_type": "official_model_card", "reported_setting": { "mode": "thinking/reasoning", "effort": "high/default", "tools": "Stirrup/E2B agent sandbox", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "Artificial Analysis GDPval-AA v2 leaderboard", "temperature": "default", "snapshot": "July 2026" }, "notes": "Google Gemini 3.6 Flash July 2026 matrix: GDPval-AA v2 Elo. Exact effort, tools, sampling and harness are preserved." } ] }, { "model_id": "claude-opus-4.8", "benchmark_id": "gdpval_aa_v2_elo", "score": 1600, "reference_url": "https://ai.meta.com/static-resource/muse-spark-1.1-evaluation-report", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "benchmark-specified", "sampling": "pass@1 unless source row states otherwise", "harness": "Meta MSL general-capability evaluation", "temperature": "1.0 for Muse; provider/source setting for comparisons", "context": "benchmark-specific", "configuration": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Figure 44 · GDPval-AA v2 Elo; metric=headline_metric. Exact printed value in the general-capability summary.", "candidates": [ { "score": 1588, "reference_url": "https://media.x.ai/v1/website/card-4p6-4cd2dc57.pdf", "source_type": "model_card", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "benchmark-specified knowledge-work tools", "sampling": "unknown", "judge": "pairwise Bradley-Terry Elo", "harness": "Artificial Analysis GDPval-AA v2 harness", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown" }, "notes": "card46 AA GDPVal v2 AA GDPVal Elo; source effort=max; benchmark protocol matches canonical=True. Unknown source fields remain unknown." }, { "score": 1600, "reference_url": "https://media.x.ai/v1/website/4p5-5184fdf9.pdf", "source_type": "model_card", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "benchmark-specified knowledge-work tools", "sampling": "unknown", "judge": "pairwise Bradley-Terry Elo", "harness": "Artificial Analysis GDPval-AA v2 harness", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown" }, "notes": "card45 GDPval-AA v2 AA score; source effort=max; benchmark protocol matches canonical=True. Unknown source fields remain unknown." } ] }, { "model_id": "gpt-5.5", "benchmark_id": "gdpval_aa_v2_elo", "score": 1494, "reference_url": "https://ai.meta.com/static-resource/muse-spark-1.1-evaluation-report", "reported_setting": { "mode": "thinking", "effort": "xhigh", "tools": "benchmark-specified", "sampling": "pass@1 unless source row states otherwise", "harness": "Meta MSL general-capability evaluation", "temperature": "1.0 for Muse; provider/source setting for comparisons", "context": "benchmark-specific", "configuration": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Figure 44 · GDPval-AA v2 Elo; metric=headline_metric. Exact printed value in the general-capability summary.", "candidates": [ { "score": 1769, "reference_url": "https://deepmind.google/models/model-cards/gemini-3-5-flash/", "source_type": "official_model_card", "reported_setting": { "mode": "thinking/reasoning", "effort": "xhigh/best available", "tools": "Stirrup/E2B agent sandbox", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "Artificial Analysis GDPval-AA v2 leaderboard", "temperature": "default", "snapshot": "May 2026" }, "notes": "Google Gemini 3.5 Flash May 2026 table: GDPval-AA v2; Artificial Analysis Elo. Settings and provenance are preserved per cell." }, { "score": 1490, "reference_url": "https://media.x.ai/v1/website/card-4p6-4cd2dc57.pdf", "source_type": "model_card", "reported_setting": { "mode": "thinking", "effort": "xhigh", "tools": "benchmark-specified knowledge-work tools", "sampling": "unknown", "judge": "pairwise Bradley-Terry Elo", "harness": "Artificial Analysis GDPval-AA v2 harness", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown" }, "notes": "card46 AA GDPVal v2 AA GDPVal Elo; source effort=xhigh; benchmark protocol matches canonical=True. Unknown source fields remain unknown." }, { "score": 1493, "reference_url": "https://media.x.ai/v1/website/4p5-5184fdf9.pdf", "source_type": "model_card", "reported_setting": { "mode": "thinking", "effort": "xhigh", "tools": "benchmark-specified knowledge-work tools", "sampling": "unknown", "judge": "pairwise Bradley-Terry Elo", "harness": "Artificial Analysis GDPval-AA v2 harness", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown" }, "notes": "card45 GDPval-AA v2 AA score; source effort=xhigh; benchmark protocol matches canonical=True. Unknown source fields remain unknown." } ] }, { "model_id": "muse-spark-1.1", "benchmark_id": "job_bench", "score": 54.7, "reference_url": "https://ai.meta.com/static-resource/muse-spark-1.1-evaluation-report", "reported_setting": { "mode": "thinking", "effort": "xhigh", "tools": "benchmark-specified", "sampling": "pass@1 unless source row states otherwise", "harness": "Meta MSL general-capability evaluation", "temperature": "1.0 for Muse; provider/source setting for comparisons", "context": "benchmark-specific", "configuration": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Figure 44 · JobBench; metric=headline_metric. Exact printed value in the general-capability summary." }, { "model_id": "muse-spark", "benchmark_id": "job_bench", "score": 17.0, "reference_url": "https://ai.meta.com/static-resource/muse-spark-1.1-evaluation-report", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "benchmark-specified", "sampling": "pass@1 unless source row states otherwise", "harness": "Meta MSL general-capability evaluation", "temperature": "1.0 for Muse; provider/source setting for comparisons", "context": "benchmark-specific", "configuration": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Figure 44 · JobBench; metric=headline_metric. Exact printed value in the general-capability summary." }, { "model_id": "gemini-3.1-pro", "benchmark_id": "job_bench", "score": 15.9, "reference_url": "https://ai.meta.com/static-resource/muse-spark-1.1-evaluation-report", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "benchmark-specified", "sampling": "pass@1 unless source row states otherwise", "harness": "Meta MSL general-capability evaluation", "temperature": "1.0 for Muse; provider/source setting for comparisons", "context": "benchmark-specific", "configuration": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Figure 44 · JobBench; metric=headline_metric. Exact printed value in the general-capability summary." }, { "model_id": "muse-spark-1.1", "benchmark_id": "finance_agent_v2", "score": 57.2, "reference_url": "https://ai.meta.com/static-resource/muse-spark-1.1-evaluation-report", "reported_setting": { "mode": "thinking", "effort": "xhigh", "tools": "finance-agent tools", "sampling": "pass@1 unless source row states otherwise", "harness": "Meta MSL general-capability evaluation", "temperature": "1.0 for Muse; provider/source setting for comparisons", "context": "benchmark-specific", "configuration": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Figure 44 · Finance Agent v2; metric=headline_metric. Exact printed value in the general-capability summary." }, { "model_id": "gemini-3.1-pro", "benchmark_id": "finance_agent_v2", "score": 43.0, "reference_url": "https://ai.meta.com/static-resource/muse-spark-1.1-evaluation-report", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "finance-agent tools", "sampling": "pass@1 unless source row states otherwise", "harness": "Meta MSL general-capability evaluation", "temperature": "1.0 for Muse; provider/source setting for comparisons", "context": "benchmark-specific", "configuration": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Figure 44 · Finance Agent v2; metric=headline_metric. Exact printed value in the general-capability summary." }, { "model_id": "muse-spark-1.1", "benchmark_id": "terminal_bench_2_1", "score": 80.0, "reference_url": "https://ai.meta.com/static-resource/muse-spark-1.1-evaluation-report", "reported_setting": { "mode": "thinking", "effort": "xhigh", "tools": "terminal", "sampling": "pass@1 unless source row states otherwise", "harness": "Meta MSL general-capability evaluation", "temperature": "1.0 for Muse; provider/source setting for comparisons", "context": "benchmark-specific", "configuration": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Figure 44 · Terminal-Bench 2.1; metric=headline_metric. Exact printed value in the general-capability summary.", "candidates": [ { "score": 76.2, "reference_url": "https://research.meta.ai/articles/introducing-muse-code-and-muse-spark-1-2/evaluations/terminal-bench-2-1-v1.png", "source_type": "official_blog", "reported_setting": { "mode": "thinking", "effort": "xhigh", "prompt_style": "not disclosed", "temperature": "not disclosed", "context": "not disclosed", "agent": "mini-swe-agent", "attempts": 5, "tools": "mini-swe-agent terminal coding toolset", "sampling": "pass@1 averaged across five attempts per task", "judge": "Terminal-Bench 2.1 executable task verifier", "harness": "Meta internal agent evaluation framework; mini-swe-agent; isolated Daytona sandbox", "internet": "not disclosed", "configuration": "selected official agent product in isolated Daytona" }, "notes": "Exact printed value from the locked Meta release chart; evaluation settings come from the locked official methodology." } ] }, { "model_id": "muse-spark", "benchmark_id": "terminal_bench_2_1", "score": 67.3, "reference_url": "https://ai.meta.com/static-resource/muse-spark-1.1-evaluation-report", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "terminal", "sampling": "pass@1 unless source row states otherwise", "harness": "Meta MSL general-capability evaluation", "temperature": "1.0 for Muse; provider/source setting for comparisons", "context": "benchmark-specific", "configuration": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Figure 44 · Terminal-Bench 2.1; metric=headline_metric. Exact printed value in the general-capability summary." }, { "model_id": "muse-spark-1.1", "benchmark_id": "swe_bench_pro", "score": 61.5, "reference_url": "https://ai.meta.com/static-resource/muse-spark-1.1-evaluation-report", "reported_setting": { "mode": "thinking", "effort": "xhigh", "tools": "agentic coding", "sampling": "pass@1 unless source row states otherwise", "harness": "Meta MSL general-capability evaluation", "temperature": "1.0 for Muse; provider/source setting for comparisons", "context": "benchmark-specific", "configuration": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Figure 44 · SWE-Bench Pro; metric=headline_metric. Exact printed value in the general-capability summary." }, { "model_id": "muse-spark-1.1", "benchmark_id": "deep_swe_v1_1", "score": 53.3, "reference_url": "https://ai.meta.com/static-resource/muse-spark-1.1-evaluation-report", "reported_setting": { "mode": "thinking", "effort": "xhigh", "tools": "agentic coding", "sampling": "pass@1 unless source row states otherwise", "harness": "Meta MSL general-capability evaluation", "temperature": "1.0 for Muse; provider/source setting for comparisons", "context": "benchmark-specific", "configuration": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Figure 44 · DeepSWE 1.1; metric=headline_metric. Exact printed value in the general-capability summary.", "candidates": [ { "score": 53.0, "reference_url": "https://research.meta.ai/articles/introducing-muse-code-and-muse-spark-1-2/evaluations/deepswe-1-1-v1.png", "source_type": "official_blog", "reported_setting": { "mode": "thinking", "effort": "xhigh", "prompt_style": "not disclosed", "temperature": "not disclosed", "context": "not disclosed", "agent": "mini-swe-agent", "attempts": 5, "tools": "mini-swe-agent repository shell/editor toolset", "sampling": "pass@1 averaged across five attempts per task", "judge": "functional verifier plus regression checks in a pristine verifier container", "harness": "Meta internal agent evaluation framework; mini-swe-agent; isolated Daytona sandbox; Harbor conversion", "internet": "disabled during rollout and grading; model endpoint only", "configuration": "five-language Harbor conversion with pristine verifier" }, "notes": "Exact printed value from the locked Meta release chart; evaluation settings come from the locked official methodology." } ] }, { "model_id": "muse-spark", "benchmark_id": "deep_swe_v1_1", "score": 10.0, "reference_url": "https://ai.meta.com/static-resource/muse-spark-1.1-evaluation-report", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "agentic coding", "sampling": "pass@1 unless source row states otherwise", "harness": "Meta MSL general-capability evaluation", "temperature": "1.0 for Muse; provider/source setting for comparisons", "context": "benchmark-specific", "configuration": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Figure 44 · DeepSWE 1.1; metric=headline_metric. Exact printed value in the general-capability summary." }, { "model_id": "muse-spark-1.1", "benchmark_id": "healthbench_professional_length_adjusted", "score": 59.3, "reference_url": "https://ai.meta.com/static-resource/muse-spark-1.1-evaluation-report", "reported_setting": { "mode": "thinking", "effort": "xhigh", "tools": "none", "sampling": "pass@1 unless source row states otherwise", "harness": "Meta MSL general-capability evaluation", "temperature": "1.0 for Muse; provider/source setting for comparisons", "context": "benchmark-specific", "configuration": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Figure 44 · HealthBench Professional length-adjusted; metric=headline_metric. Exact printed value in the general-capability summary." }, { "model_id": "muse-spark", "benchmark_id": "healthbench_professional_length_adjusted", "score": 54.1, "reference_url": "https://ai.meta.com/static-resource/muse-spark-1.1-evaluation-report", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1 unless source row states otherwise", "harness": "Meta MSL general-capability evaluation", "temperature": "1.0 for Muse; provider/source setting for comparisons", "context": "benchmark-specific", "configuration": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Figure 44 · HealthBench Professional length-adjusted; metric=headline_metric. Exact printed value in the general-capability summary." }, { "model_id": "gemini-3.1-pro", "benchmark_id": "healthbench_professional_length_adjusted", "score": 41.6, "reference_url": "https://ai.meta.com/static-resource/muse-spark-1.1-evaluation-report", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1 unless source row states otherwise", "harness": "Meta MSL general-capability evaluation", "temperature": "1.0 for Muse; provider/source setting for comparisons", "context": "benchmark-specific", "configuration": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Figure 44 · HealthBench Professional length-adjusted; metric=headline_metric. Exact printed value in the general-capability summary." }, { "model_id": "muse-spark-1.1", "benchmark_id": "charxiv_reasoning", "score": 88.4, "reference_url": "https://ai.meta.com/static-resource/muse-spark-1.1-evaluation-report", "reported_setting": { "mode": "thinking", "effort": "xhigh", "tools": "Python", "sampling": "pass@1 unless source row states otherwise", "harness": "Meta MSL general-capability evaluation", "temperature": "1.0 for Muse; provider/source setting for comparisons", "context": "benchmark-specific", "configuration": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Figure 44 · CharXiv Reasoning w/code execution; metric=headline_metric. Exact printed value in the general-capability summary." }, { "model_id": "muse-spark-1.1", "benchmark_id": "babyvision", "score": 76.3, "reference_url": "https://ai.meta.com/static-resource/muse-spark-1.1-evaluation-report", "reported_setting": { "mode": "thinking", "effort": "xhigh", "tools": "Python", "sampling": "pass@1 unless source row states otherwise", "harness": "Meta MSL general-capability evaluation", "temperature": "1.0 for Muse; provider/source setting for comparisons", "context": "benchmark-specific", "configuration": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Figure 44 · BabyVision w/code execution; metric=headline_metric. Exact printed value in the general-capability summary." }, { "model_id": "muse-spark", "benchmark_id": "babyvision", "score": 39.9, "reference_url": "https://ai.meta.com/static-resource/muse-spark-1.1-evaluation-report", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "Python", "sampling": "pass@1 unless source row states otherwise", "harness": "Meta MSL general-capability evaluation", "temperature": "1.0 for Muse; provider/source setting for comparisons", "context": "benchmark-specific", "configuration": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Figure 44 · BabyVision w/code execution; metric=headline_metric. Exact printed value in the general-capability summary." }, { "model_id": "claude-opus-4.8", "benchmark_id": "meta_internal_coding_bench", "score": 69.0, "reference_url": "https://ai.meta.com/blog/introducing-muse-spark-meta-model-api/", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "agentic coding", "sampling": "pass@1 unless source row states otherwise", "harness": "Meta MSL general-capability evaluation", "temperature": "1.0 for Muse; provider/source setting for comparisons", "context": "benchmark-specific", "configuration": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "launch asset 04 · Meta Internal Coding Bench; metric=headline_metric. Exact numeric label visually read from the official launch raster." }, { "model_id": "muse-spark-1.1", "benchmark_id": "meta_internal_coding_bench", "score": 68.3, "reference_url": "https://ai.meta.com/blog/introducing-muse-spark-meta-model-api/", "reported_setting": { "mode": "thinking", "effort": "xhigh", "tools": "agentic coding", "sampling": "pass@1 unless source row states otherwise", "harness": "Meta MSL general-capability evaluation", "temperature": "1.0 for Muse; provider/source setting for comparisons", "context": "benchmark-specific", "configuration": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "launch asset 04 · Meta Internal Coding Bench; metric=headline_metric. Exact numeric label visually read from the official launch raster.", "candidates": [ { "score": 68.3, "reference_url": "https://research.meta.ai/articles/introducing-muse-code-and-muse-spark-1-2/evaluations/meta-internal-coding-bench-v1.png", "source_type": "official_blog", "reported_setting": { "mode": "thinking", "effort": "xhigh", "prompt_style": "not disclosed", "temperature": "not disclosed", "context": "not disclosed", "agent": "not disclosed", "attempts": 2, "tools": "Meta internal agentic coding environment", "sampling": "two attempts per task; average task-level success rate", "judge": "compile and unit tests in dedicated grading containers", "harness": "Meta internal agentic coding harness and isolated task sandbox", "internet": "disabled", "configuration": "440 internal pull-request tasks" }, "notes": "Exact printed value from the locked Meta release chart; evaluation settings come from the locked official methodology." } ] }, { "model_id": "gpt-5.5", "benchmark_id": "meta_internal_coding_bench", "score": 67.1, "reference_url": "https://ai.meta.com/blog/introducing-muse-spark-meta-model-api/", "reported_setting": { "mode": "thinking", "effort": "xhigh", "tools": "agentic coding", "sampling": "pass@1 unless source row states otherwise", "harness": "Meta MSL general-capability evaluation", "temperature": "1.0 for Muse; provider/source setting for comparisons", "context": "benchmark-specific", "configuration": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "launch asset 04 · Meta Internal Coding Bench; metric=headline_metric. Exact numeric label visually read from the official launch raster." }, { "model_id": "gemini-3.1-pro", "benchmark_id": "meta_internal_coding_bench", "score": 59.2, "reference_url": "https://ai.meta.com/blog/introducing-muse-spark-meta-model-api/", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "agentic coding", "sampling": "pass@1 unless source row states otherwise", "harness": "Meta MSL general-capability evaluation", "temperature": "1.0 for Muse; provider/source setting for comparisons", "context": "benchmark-specific", "configuration": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "launch asset 04 · Meta Internal Coding Bench; metric=headline_metric. Exact numeric label visually read from the official launch raster." }, { "model_id": "muse-spark", "benchmark_id": "meta_internal_coding_bench", "score": 58.8, "reference_url": "https://ai.meta.com/blog/introducing-muse-spark-meta-model-api/", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "agentic coding", "sampling": "pass@1 unless source row states otherwise", "harness": "Meta MSL general-capability evaluation", "temperature": "1.0 for Muse; provider/source setting for comparisons", "context": "benchmark-specific", "configuration": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "launch asset 04 · Meta Internal Coding Bench; metric=headline_metric. Exact numeric label visually read from the official launch raster." }, { "model_id": "muse-spark-1.1", "benchmark_id": "vibecodebench_v1_1_test", "score": 72.2, "reference_url": "https://ai.meta.com/blog/introducing-muse-spark-meta-model-api/", "reported_setting": { "mode": "thinking", "effort": "xhigh", "tools": "agentic coding", "sampling": "pass@1 unless source row states otherwise", "harness": "Meta MSL general-capability evaluation", "temperature": "1.0 for Muse; provider/source setting for comparisons", "context": "benchmark-specific", "configuration": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "launch asset 09 · Vibe Code Bench v1.1 test; metric=headline_metric. Exact numeric label visually read from the official launch raster." }, { "model_id": "muse-spark", "benchmark_id": "vibecodebench_v1_1_test", "score": 19.7, "reference_url": "https://ai.meta.com/blog/introducing-muse-spark-meta-model-api/", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "agentic coding", "sampling": "pass@1 unless source row states otherwise", "harness": "Meta MSL general-capability evaluation", "temperature": "1.0 for Muse; provider/source setting for comparisons", "context": "benchmark-specific", "configuration": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "launch asset 09 · Vibe Code Bench v1.1 test; metric=headline_metric. Exact numeric label visually read from the official launch raster." }, { "model_id": "muse-spark-1.1", "benchmark_id": "swe_atlas_codebase_qna", "score": 42.0, "reference_url": "https://ai.meta.com/blog/introducing-muse-spark-meta-model-api/", "reported_setting": { "mode": "thinking", "effort": "xhigh", "tools": "codebase tools", "sampling": "pass@1 unless source row states otherwise", "harness": "Meta MSL general-capability evaluation", "temperature": "1.0 for Muse; provider/source setting for comparisons", "context": "benchmark-specific", "configuration": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "launch asset 09 · SWE Atlas Codebase QnA; metric=headline_metric. Exact numeric label visually read from the official launch raster." }, { "model_id": "muse-spark", "benchmark_id": "swe_atlas_codebase_qna", "score": 24.2, "reference_url": "https://ai.meta.com/blog/introducing-muse-spark-meta-model-api/", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "codebase tools", "sampling": "pass@1 unless source row states otherwise", "harness": "Meta MSL general-capability evaluation", "temperature": "1.0 for Muse; provider/source setting for comparisons", "context": "benchmark-specific", "configuration": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "launch asset 09 · SWE Atlas Codebase QnA; metric=headline_metric. Exact numeric label visually read from the official launch raster." }, { "model_id": "gemini-3-pro-deep-think-2025-12", "benchmark_id": "hle", "score": 41.0, "reference_url": "https://blog.google/products-and-platforms/products/gemini/gemini-3/#gemini-3-deep-think", "reported_setting": { "mode": "deep-think", "effort": "vendor-specialized", "tools": "none", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official Google / ARC Prize", "temperature": "default", "snapshot": "Nov-Dec 2025" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "Google Gemini 3 launch: Humanity's Last Exam: no tools. Distinct Dec 2025 Deep Think evaluated configuration." }, { "model_id": "gemini-3-pro-deep-think-2025-12", "benchmark_id": "gpqa_diamond", "score": 93.8, "reference_url": "https://blog.google/products-and-platforms/products/gemini/gemini-3/#gemini-3-deep-think", "reported_setting": { "mode": "deep-think", "effort": "vendor-specialized", "tools": "none", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official Google / ARC Prize", "temperature": "default", "snapshot": "Nov-Dec 2025" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "Google Gemini 3 launch: GPQA Diamond. Distinct Dec 2025 Deep Think evaluated configuration." }, { "model_id": "gemini-3-pro-deep-think-2025-12", "benchmark_id": "arc_agi_2", "score": 45.1, "reference_url": "https://blog.google/products-and-platforms/products/gemini/gemini-3/#gemini-3-deep-think", "reported_setting": { "mode": "deep-think", "effort": "vendor-specialized", "tools": "code execution", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official Google / ARC Prize", "temperature": "default", "snapshot": "Nov-Dec 2025" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Google Gemini 3 launch: ARC-AGI-2: code execution. Distinct Dec 2025 Deep Think evaluated configuration." }, { "model_id": "gemini-3.1-pro-deep-think-2026-02", "benchmark_id": "arc_agi_2", "score": 84.6, "reference_url": "https://deepmind.google/models/gemini/deep-think/", "reported_setting": { "mode": "deep-think", "effort": "vendor-specialized", "tools": "none", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "Google Deep Think evaluation", "temperature": "default", "snapshot": "Feb 2026" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "Google February Deep Think table: ARC-AGI-2. ARC Prize Verified; v2 semi-private. Non-default sampling or Google cross-model harness is retained in reported_setting rather than changing benchmark identity." }, { "model_id": "gemini-3.1-pro-deep-think-2026-02", "benchmark_id": "hle", "score": 48.4, "reference_url": "https://deepmind.google/models/gemini/deep-think/", "reported_setting": { "mode": "deep-think", "effort": "vendor-specialized", "tools": "none", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "Google Deep Think evaluation", "temperature": "default", "snapshot": "Feb 2026" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "Google February Deep Think table: Humanity's Last Exam: full text+MM, no tools. Full 2,500-question text+multimodal set; no tools. Non-default sampling or Google cross-model harness is retained in reported_setting rather than changing benchmark identity." }, { "model_id": "gemini-3.1-pro-deep-think-2026-02", "benchmark_id": "hle_tools", "score": 53.4, "reference_url": "https://deepmind.google/models/gemini/deep-think/", "reported_setting": { "mode": "deep-think", "effort": "vendor-specialized", "tools": "search + code execution", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "Google Deep Think evaluation", "temperature": "default", "snapshot": "Feb 2026" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Google February Deep Think table: Humanity's Last Exam: search + code execution. Full text+multimodal set; search and code execution. Non-default sampling or Google cross-model harness is retained in reported_setting rather than changing benchmark identity." }, { "model_id": "gemini-3.1-pro-deep-think-2026-02", "benchmark_id": "mmmu_pro", "score": 81.5, "reference_url": "https://deepmind.google/models/gemini/deep-think/", "reported_setting": { "mode": "deep-think", "effort": "vendor-specialized", "tools": "none", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "Google Deep Think evaluation", "temperature": "default", "snapshot": "Feb 2026" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "Google February Deep Think table: MMMU-Pro: Standard 10-option + Vision average. No tools. Non-default sampling or Google cross-model harness is retained in reported_setting rather than changing benchmark identity." }, { "model_id": "gemini-3.1-pro-deep-think-2026-02", "benchmark_id": "imo_2025", "score": 81.5, "reference_url": "https://deepmind.google/models/gemini/deep-think/", "reported_setting": { "mode": "deep-think", "effort": "vendor-specialized", "tools": "none", "sampling": "trials=4 (average)", "judge": "benchmark-specified", "harness": "Google Deep Think evaluation", "temperature": "default", "snapshot": "Feb 2026" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Google February Deep Think table: International Math Olympiad 2025. All 6 problems; average across 4 runs; expert grading. Non-default sampling or Google cross-model harness is retained in reported_setting rather than changing benchmark identity.", "candidates": [ { "score": 81.5, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2.1/Seed2_1_Model_Card.pdf", "source_type": "tech_report", "reported_setting": { "mode": "deep-think", "effort": "unknown", "tools": "benchmark-specified", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official", "prompt_style": "official default", "temperature": "unknown" }, "notes": "Model card Table 7 (p24)." } ] }, { "model_id": "gemini-3-pro", "benchmark_id": "imo_2025", "score": 14.3, "reference_url": "https://deepmind.google/models/gemini/deep-think/", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "trials=4 (average)", "judge": "benchmark-specified", "harness": "Google self-computed cross-model evaluation", "temperature": "default", "snapshot": "Feb 2026" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Google February Deep Think table: International Math Olympiad 2025. All 6 problems; average across 4 runs; expert grading. Non-default sampling or Google cross-model harness is retained in reported_setting rather than changing benchmark identity." }, { "model_id": "gpt-5.2", "benchmark_id": "imo_2025", "score": 71.4, "reference_url": "https://deepmind.google/models/gemini/deep-think/", "reported_setting": { "mode": "thinking", "effort": "xhigh", "tools": "none", "sampling": "trials=4 (average)", "judge": "benchmark-specified", "harness": "Google self-computed cross-model evaluation", "temperature": "default", "snapshot": "Feb 2026" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Google February Deep Think table: International Math Olympiad 2025. All 6 problems; average across 4 runs; expert grading. Non-default sampling or Google cross-model harness is retained in reported_setting rather than changing benchmark identity." }, { "model_id": "gemini-3.1-pro-deep-think-2026-02", "benchmark_id": "ipho_2025_theory", "score": 87.7, "reference_url": "https://deepmind.google/models/gemini/deep-think/", "reported_setting": { "mode": "deep-think", "effort": "vendor-specialized", "tools": "none", "sampling": "trials=8 (average)", "judge": "benchmark-specified", "harness": "Google Deep Think evaluation", "temperature": "default", "snapshot": "Feb 2026" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Google February Deep Think table: International Physics Olympiad 2025 Theory. All 3 used problems; average across 8 runs; Gemini judge. Non-default sampling or Google cross-model harness is retained in reported_setting rather than changing benchmark identity.", "candidates": [ { "score": 87.7, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2.1/Seed2_1_Model_Card.pdf", "source_type": "tech_report", "reported_setting": { "mode": "deep-think", "effort": "unknown", "tools": "benchmark-specified", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official", "prompt_style": "official default", "temperature": "unknown" }, "notes": "Model card Table 7 (p24)." } ] }, { "model_id": "gemini-3-pro", "benchmark_id": "ipho_2025_theory", "score": 76.3, "reference_url": "https://deepmind.google/models/gemini/deep-think/", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "trials=8 (average)", "judge": "benchmark-specified", "harness": "Google self-computed cross-model evaluation", "temperature": "default", "snapshot": "Feb 2026" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Google February Deep Think table: International Physics Olympiad 2025 Theory. All 3 used problems; average across 8 runs; Gemini judge. Non-default sampling or Google cross-model harness is retained in reported_setting rather than changing benchmark identity." }, { "model_id": "claude-opus-4.6", "benchmark_id": "ipho_2025_theory", "score": 71.6, "reference_url": "https://deepmind.google/models/gemini/deep-think/", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "none", "sampling": "trials=8 (average)", "judge": "benchmark-specified", "harness": "Google self-computed cross-model evaluation", "temperature": "default", "snapshot": "Feb 2026" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Google February Deep Think table: International Physics Olympiad 2025 Theory. All 3 used problems; average across 8 runs; Gemini judge. Non-default sampling or Google cross-model harness is retained in reported_setting rather than changing benchmark identity." }, { "model_id": "gpt-5.2", "benchmark_id": "ipho_2025_theory", "score": 70.5, "reference_url": "https://deepmind.google/models/gemini/deep-think/", "reported_setting": { "mode": "thinking", "effort": "xhigh", "tools": "none", "sampling": "trials=8 (average)", "judge": "benchmark-specified", "harness": "Google self-computed cross-model evaluation", "temperature": "default", "snapshot": "Feb 2026" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Google February Deep Think table: International Physics Olympiad 2025 Theory. All 3 used problems; average across 8 runs; Gemini judge. Non-default sampling or Google cross-model harness is retained in reported_setting rather than changing benchmark identity." }, { "model_id": "gemini-3.1-pro-deep-think-2026-02", "benchmark_id": "cmt_benchmark", "score": 50.5, "reference_url": "https://deepmind.google/models/gemini/deep-think/", "reported_setting": { "mode": "deep-think", "effort": "vendor-specialized", "tools": "none", "sampling": "pass@8", "judge": "benchmark-specified", "harness": "Google Deep Think evaluation", "temperature": "default", "snapshot": "Feb 2026" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Google February Deep Think table: CMT-Benchmark. 50 problems; pass@8; Gemini judge. Non-default sampling or Google cross-model harness is retained in reported_setting rather than changing benchmark identity." }, { "model_id": "gemini-3-pro", "benchmark_id": "cmt_benchmark", "score": 39.5, "reference_url": "https://deepmind.google/models/gemini/deep-think/", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@8", "judge": "benchmark-specified", "harness": "Google self-computed cross-model evaluation", "temperature": "default", "snapshot": "Feb 2026" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Google February Deep Think table: CMT-Benchmark. 50 problems; pass@8; Gemini judge. Non-default sampling or Google cross-model harness is retained in reported_setting rather than changing benchmark identity." }, { "model_id": "claude-opus-4.6", "benchmark_id": "cmt_benchmark", "score": 17.1, "reference_url": "https://deepmind.google/models/gemini/deep-think/", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "none", "sampling": "pass@8", "judge": "benchmark-specified", "harness": "Google self-computed cross-model evaluation", "temperature": "default", "snapshot": "Feb 2026" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Google February Deep Think table: CMT-Benchmark. 50 problems; pass@8; Gemini judge. Non-default sampling or Google cross-model harness is retained in reported_setting rather than changing benchmark identity." }, { "model_id": "gpt-5.2", "benchmark_id": "cmt_benchmark", "score": 41.0, "reference_url": "https://deepmind.google/models/gemini/deep-think/", "reported_setting": { "mode": "thinking", "effort": "xhigh", "tools": "none", "sampling": "pass@8", "judge": "benchmark-specified", "harness": "Google self-computed cross-model evaluation", "temperature": "default", "snapshot": "Feb 2026" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Google February Deep Think table: CMT-Benchmark. 50 problems; pass@8; Gemini judge. Non-default sampling or Google cross-model harness is retained in reported_setting rather than changing benchmark identity." }, { "model_id": "gemini-3.1-pro-deep-think-2026-02", "benchmark_id": "icho_2025_theory", "score": 82.8, "reference_url": "https://deepmind.google/models/gemini/deep-think/", "reported_setting": { "mode": "deep-think", "effort": "vendor-specialized", "tools": "none", "sampling": "trials=8 (average)", "judge": "benchmark-specified", "harness": "Google Deep Think evaluation", "temperature": "default", "snapshot": "Feb 2026" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Google February Deep Think table: International Chemistry Olympiad 2025 Theory. All 9 problems; average across 8 runs; Gemini judge. Non-default sampling or Google cross-model harness is retained in reported_setting rather than changing benchmark identity." }, { "model_id": "gemini-3-pro", "benchmark_id": "icho_2025_theory", "score": 69.6, "reference_url": "https://deepmind.google/models/gemini/deep-think/", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "trials=8 (average)", "judge": "benchmark-specified", "harness": "Google self-computed cross-model evaluation", "temperature": "default", "snapshot": "Feb 2026" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Google February Deep Think table: International Chemistry Olympiad 2025 Theory. All 9 problems; average across 8 runs; Gemini judge. Non-default sampling or Google cross-model harness is retained in reported_setting rather than changing benchmark identity." }, { "model_id": "gpt-5.2", "benchmark_id": "icho_2025_theory", "score": 72.0, "reference_url": "https://deepmind.google/models/gemini/deep-think/", "reported_setting": { "mode": "thinking", "effort": "xhigh", "tools": "none", "sampling": "trials=8 (average)", "judge": "benchmark-specified", "harness": "Google self-computed cross-model evaluation", "temperature": "default", "snapshot": "Feb 2026" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Google February Deep Think table: International Chemistry Olympiad 2025 Theory. All 9 problems; average across 8 runs; Gemini judge. Non-default sampling or Google cross-model harness is retained in reported_setting rather than changing benchmark identity." }, { "model_id": "gemini-3-flash", "benchmark_id": "terminal_bench_2_1", "score": 58.0, "reference_url": "https://deepmind.google/models/model-cards/gemini-3-5-flash/", "reported_setting": { "mode": "thinking/reasoning", "effort": "source does not state", "tools": "terminal agent", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "Terminus-2; Gemini self-computed, others public leaderboard", "temperature": "default", "snapshot": "May 2026" }, "matches_canonical": false, "source_type": "official_model_card", "audit_status": "verified", "notes": "Google Gemini 3.5 Flash May 2026 table: Terminal-Bench 2.1; Terminus-2 harness. Settings and provenance are preserved per cell." }, { "model_id": "gemini-3.5-flash", "benchmark_id": "mcpatlas", "score": 83.6, "reference_url": "https://deepmind.google/models/model-cards/gemini-3-5-flash/", "reported_setting": { "mode": "thinking/reasoning", "effort": "medium (runtime default)", "tools": "MCP server tool use", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "ScaleAI official May 2026 leaderboard", "temperature": "default", "snapshot": "May 2026" }, "matches_canonical": true, "source_type": "official_model_card", "audit_status": "verified", "notes": "Google Gemini 3.5 Flash May 2026 table: MCP Atlas; ScaleAI official leaderboard. Settings and provenance are preserved per cell." }, { "model_id": "gemini-3.5-flash", "benchmark_id": "toolathlon", "score": 56.5, "reference_url": "https://deepmind.google/models/model-cards/gemini-3-5-flash/", "reported_setting": { "mode": "thinking/reasoning", "effort": "medium (runtime default)", "tools": "Toolathlon agent harness", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "benchmark-author supplied for Gemini", "temperature": "default", "snapshot": "May 2026" }, "matches_canonical": true, "source_type": "official_model_card", "audit_status": "verified", "notes": "Google Gemini 3.5 Flash May 2026 table: Toolathlon; original 108-task benchmark. Settings and provenance are preserved per cell." }, { "model_id": "gemini-3-flash", "benchmark_id": "osworld_verified", "score": 65.1, "reference_url": "https://deepmind.google/models/model-cards/gemini-3-5-flash/", "reported_setting": { "mode": "thinking/reasoning", "effort": "source does not state", "tools": "pyautogui + UI-specific function declarations", "sampling": "Gemini avg 5 runs, single attempt per run", "judge": "benchmark-specified", "harness": "OSWorld Docker, 1080p, max 100 steps", "temperature": "default", "snapshot": "May 2026" }, "matches_canonical": false, "source_type": "official_model_card", "audit_status": "verified", "notes": "Google Gemini 3.5 Flash May 2026 table: OSWorld-Verified; Docker, 1080p, max 100 steps. Settings and provenance are preserved per cell." }, { "model_id": "claude-opus-4.7", "benchmark_id": "osworld_verified", "score": 78.0, "reference_url": "https://deepmind.google/models/model-cards/gemini-3-5-flash/", "reported_setting": { "mode": "thinking/reasoning", "effort": "max/best available", "tools": "pyautogui + UI-specific function declarations", "sampling": "Gemini avg 5 runs, single attempt per run", "judge": "benchmark-specified", "harness": "OSWorld Docker, 1080p, max 100 steps", "temperature": "default", "snapshot": "May 2026" }, "matches_canonical": false, "source_type": "official_model_card", "audit_status": "verified", "notes": "Google Gemini 3.5 Flash May 2026 table: OSWorld-Verified; Docker, 1080p, max 100 steps. Settings and provenance are preserved per cell." }, { "model_id": "gemini-3.5-flash", "benchmark_id": "finance_agent_v2", "score": 57.9, "reference_url": "https://deepmind.google/models/model-cards/gemini-3-5-flash/", "reported_setting": { "mode": "thinking/reasoning", "effort": "medium (runtime default)", "tools": "Vals six-tool harness", "sampling": "mean of 3 runs", "judge": "benchmark-specified", "harness": "Vals Finance Agent v2 leaderboard", "temperature": "default", "snapshot": "May 2026" }, "matches_canonical": true, "source_type": "official_model_card", "audit_status": "verified", "notes": "Google Gemini 3.5 Flash May 2026 table: Finance Agent v2; Vals held-out leaderboard, weighted partial credit. Settings and provenance are preserved per cell." }, { "model_id": "gemini-3-flash", "benchmark_id": "finance_agent_v2", "score": 42.6, "reference_url": "https://deepmind.google/models/model-cards/gemini-3-5-flash/", "reported_setting": { "mode": "thinking/reasoning", "effort": "source does not state", "tools": "Vals six-tool harness", "sampling": "mean of 3 runs", "judge": "benchmark-specified", "harness": "Vals Finance Agent v2 leaderboard", "temperature": "default", "snapshot": "May 2026" }, "matches_canonical": true, "source_type": "official_model_card", "audit_status": "verified", "notes": "Google Gemini 3.5 Flash May 2026 table: Finance Agent v2; Vals held-out leaderboard, weighted partial credit. Settings and provenance are preserved per cell." }, { "model_id": "claude-sonnet-4.6", "benchmark_id": "finance_agent_v2", "score": 51.0, "reference_url": "https://deepmind.google/models/model-cards/gemini-3-5-flash/", "reported_setting": { "mode": "thinking/reasoning", "effort": "max/best available", "tools": "Vals six-tool harness", "sampling": "mean of 3 runs", "judge": "benchmark-specified", "harness": "Vals Finance Agent v2 leaderboard", "temperature": "default", "snapshot": "May 2026" }, "matches_canonical": true, "source_type": "official_model_card", "audit_status": "verified", "notes": "Google Gemini 3.5 Flash May 2026 table: Finance Agent v2; Vals held-out leaderboard, weighted partial credit. Settings and provenance are preserved per cell." }, { "model_id": "claude-opus-4.7", "benchmark_id": "finance_agent_v2", "score": 51.5, "reference_url": "https://deepmind.google/models/model-cards/gemini-3-5-flash/", "reported_setting": { "mode": "thinking/reasoning", "effort": "max/best available", "tools": "Vals six-tool harness", "sampling": "mean of 3 runs", "judge": "benchmark-specified", "harness": "Vals Finance Agent v2 leaderboard", "temperature": "default", "snapshot": "May 2026" }, "matches_canonical": true, "source_type": "official_model_card", "audit_status": "verified", "notes": "Google Gemini 3.5 Flash May 2026 table: Finance Agent v2; Vals held-out leaderboard, weighted partial credit. Settings and provenance are preserved per cell." }, { "model_id": "gemini-3.5-flash", "benchmark_id": "gdpval_aa_v2_elo", "score": 1656, "reference_url": "https://deepmind.google/models/model-cards/gemini-3-5-flash/", "reported_setting": { "mode": "thinking/reasoning", "effort": "medium (runtime default)", "tools": "Stirrup/E2B agent sandbox", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "Artificial Analysis GDPval-AA v2 leaderboard", "temperature": "default", "snapshot": "May 2026" }, "matches_canonical": true, "source_type": "official_model_card", "audit_status": "verified", "notes": "Google Gemini 3.5 Flash May 2026 table: GDPval-AA v2; Artificial Analysis Elo. Settings and provenance are preserved per cell.", "candidates": [ { "score": 1349, "reference_url": "https://deepmind.google/models/model-cards/gemini-3-6-flash/", "source_type": "official_model_card", "reported_setting": { "mode": "thinking/reasoning", "effort": "medium (default)", "tools": "Stirrup/E2B agent sandbox", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "Artificial Analysis GDPval-AA v2 leaderboard", "temperature": "default", "snapshot": "July 2026" }, "notes": "Google Gemini 3.6 Flash July 2026 matrix: GDPval-AA v2 Elo. Exact effort, tools, sampling and harness are preserved." } ] }, { "model_id": "gemini-3-flash", "benchmark_id": "gdpval_aa_v2_elo", "score": 1204, "reference_url": "https://deepmind.google/models/model-cards/gemini-3-5-flash/", "reported_setting": { "mode": "thinking/reasoning", "effort": "source does not state", "tools": "Stirrup/E2B agent sandbox", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "Artificial Analysis GDPval-AA v2 leaderboard", "temperature": "default", "snapshot": "May 2026" }, "matches_canonical": true, "source_type": "official_model_card", "audit_status": "verified", "notes": "Google Gemini 3.5 Flash May 2026 table: GDPval-AA v2; Artificial Analysis Elo. Settings and provenance are preserved per cell." }, { "model_id": "claude-sonnet-4.6", "benchmark_id": "gdpval_aa_v2_elo", "score": 1676, "reference_url": "https://deepmind.google/models/model-cards/gemini-3-5-flash/", "reported_setting": { "mode": "thinking/reasoning", "effort": "max/best available", "tools": "Stirrup/E2B agent sandbox", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "Artificial Analysis GDPval-AA v2 leaderboard", "temperature": "default", "snapshot": "May 2026" }, "matches_canonical": true, "source_type": "official_model_card", "audit_status": "verified", "notes": "Google Gemini 3.5 Flash May 2026 table: GDPval-AA v2; Artificial Analysis Elo. Settings and provenance are preserved per cell." }, { "model_id": "claude-opus-4.7", "benchmark_id": "gdpval_aa_v2_elo", "score": 1753, "reference_url": "https://deepmind.google/models/model-cards/gemini-3-5-flash/", "reported_setting": { "mode": "thinking/reasoning", "effort": "max/best available", "tools": "Stirrup/E2B agent sandbox", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "Artificial Analysis GDPval-AA v2 leaderboard", "temperature": "default", "snapshot": "May 2026" }, "matches_canonical": true, "source_type": "official_model_card", "audit_status": "verified", "notes": "Google Gemini 3.5 Flash May 2026 table: GDPval-AA v2; Artificial Analysis Elo. Settings and provenance are preserved per cell." }, { "model_id": "gemini-3.5-flash", "benchmark_id": "charxiv_reasoning", "score": 84.2, "reference_url": "https://deepmind.google/models/model-cards/gemini-3-5-flash/", "reported_setting": { "mode": "thinking/reasoning", "effort": "medium (runtime default)", "tools": "none", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "Google self-computed for Gemini/GPT; provider for Claude", "temperature": "default", "snapshot": "May 2026" }, "matches_canonical": false, "source_type": "official_model_card", "audit_status": "verified", "notes": "Google Gemini 3.5 Flash May 2026 table: CharXiv Reasoning; no tools. Settings and provenance are preserved per cell.", "candidates": [ { "score": 84.9, "reference_url": "https://deepmind.google/models/model-cards/gemini-3-6-flash/", "source_type": "official_model_card", "reported_setting": { "mode": "thinking/reasoning", "effort": "medium (default)", "tools": "search + code execution", "sampling": "pass@1", "judge": "official CharXiv binary answer judge", "harness": "Google self-computed; search/code details unpublished", "temperature": "default", "snapshot": "July 2026" }, "notes": "Same CharXiv reasoning set with search and code execution." } ] }, { "model_id": "gemini-3.5-flash", "benchmark_id": "mmmu_pro", "score": 83.6, "reference_url": "https://deepmind.google/models/model-cards/gemini-3-5-flash/", "reported_setting": { "mode": "thinking/reasoning", "effort": "medium (runtime default)", "tools": "none", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "Standard 10-option + Vision arithmetic mean", "temperature": "default", "snapshot": "May 2026" }, "matches_canonical": true, "source_type": "official_model_card", "audit_status": "verified", "notes": "Google Gemini 3.5 Flash May 2026 table: MMMU-Pro; Standard 10-option + Vision average; no tools. Settings and provenance are preserved per cell." }, { "model_id": "claude-opus-4.7", "benchmark_id": "mmmu_pro", "score": 75.2, "reference_url": "https://deepmind.google/models/model-cards/gemini-3-5-flash/", "reported_setting": { "mode": "thinking/reasoning", "effort": "max/best available", "tools": "none", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "Standard 10-option + Vision arithmetic mean", "temperature": "default", "snapshot": "May 2026" }, "matches_canonical": true, "source_type": "official_model_card", "audit_status": "verified", "notes": "Google Gemini 3.5 Flash May 2026 table: MMMU-Pro; Standard 10-option + Vision average; no tools. Settings and provenance are preserved per cell.", "candidates": [ { "score": 74.0, "reference_url": "https://seed.bytedance.com/en/blog/seed2-1-officially-released-advancing-ai-productivity", "source_type": "official_blog", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "none", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official", "prompt_style": "official default", "temperature": "unknown" }, "notes": "Release blog EN: Vision capability chart (https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/user-upload/4xfa4mqq4yer3.png)." } ] }, { "model_id": "gemini-3.5-flash", "benchmark_id": "blueprint_bench_2", "score": 33.6, "reference_url": "https://deepmind.google/models/model-cards/gemini-3-5-flash/", "reported_setting": { "mode": "thinking/reasoning", "effort": "medium (runtime default)", "tools": "persistent cross-apartment notepad", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "Andon Labs public leaderboard", "temperature": "default", "snapshot": "May 2026" }, "matches_canonical": false, "source_type": "official_model_card", "audit_status": "verified", "notes": "Google Gemini 3.5 Flash May 2026 table: Blueprint-Bench 2; normalized 0-100 leaderboard score. Settings and provenance are preserved per cell." }, { "model_id": "gemini-3-flash", "benchmark_id": "blueprint_bench_2", "score": 0.0, "reference_url": "https://deepmind.google/models/model-cards/gemini-3-5-flash/", "reported_setting": { "mode": "thinking/reasoning", "effort": "source does not state", "tools": "persistent cross-apartment notepad", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "Andon Labs public leaderboard", "temperature": "default", "snapshot": "May 2026" }, "matches_canonical": false, "source_type": "official_model_card", "audit_status": "verified", "notes": "Google Gemini 3.5 Flash May 2026 table: Blueprint-Bench 2; normalized 0-100 leaderboard score. Settings and provenance are preserved per cell." }, { "model_id": "gemini-3.1-pro", "benchmark_id": "blueprint_bench_2", "score": 26.5, "reference_url": "https://deepmind.google/models/model-cards/gemini-3-5-flash/", "reported_setting": { "mode": "thinking/reasoning", "effort": "source does not state", "tools": "persistent cross-apartment notepad", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "Andon Labs public leaderboard", "temperature": "default", "snapshot": "May 2026" }, "matches_canonical": false, "source_type": "official_model_card", "audit_status": "verified", "notes": "Google Gemini 3.5 Flash May 2026 table: Blueprint-Bench 2; normalized 0-100 leaderboard score. Settings and provenance are preserved per cell." }, { "model_id": "claude-sonnet-4.6", "benchmark_id": "blueprint_bench_2", "score": 6.7, "reference_url": "https://deepmind.google/models/model-cards/gemini-3-5-flash/", "reported_setting": { "mode": "thinking/reasoning", "effort": "max/best available", "tools": "persistent cross-apartment notepad", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "Andon Labs public leaderboard", "temperature": "default", "snapshot": "May 2026" }, "matches_canonical": false, "source_type": "official_model_card", "audit_status": "verified", "notes": "Google Gemini 3.5 Flash May 2026 table: Blueprint-Bench 2; normalized 0-100 leaderboard score. Settings and provenance are preserved per cell." }, { "model_id": "claude-opus-4.7", "benchmark_id": "blueprint_bench_2", "score": 24.5, "reference_url": "https://deepmind.google/models/model-cards/gemini-3-5-flash/", "reported_setting": { "mode": "thinking/reasoning", "effort": "max/best available", "tools": "persistent cross-apartment notepad", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "Andon Labs public leaderboard", "temperature": "default", "snapshot": "May 2026" }, "matches_canonical": false, "source_type": "official_model_card", "audit_status": "verified", "notes": "Google Gemini 3.5 Flash May 2026 table: Blueprint-Bench 2; normalized 0-100 leaderboard score. Settings and provenance are preserved per cell." }, { "model_id": "gpt-5.5", "benchmark_id": "blueprint_bench_2", "score": 36.2, "reference_url": "https://deepmind.google/models/model-cards/gemini-3-5-flash/", "reported_setting": { "mode": "thinking/reasoning", "effort": "xhigh/best available", "tools": "persistent cross-apartment notepad", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "Andon Labs public leaderboard", "temperature": "default", "snapshot": "May 2026" }, "matches_canonical": false, "source_type": "official_model_card", "audit_status": "verified", "notes": "Google Gemini 3.5 Flash May 2026 table: Blueprint-Bench 2; normalized 0-100 leaderboard score. Settings and provenance are preserved per cell." }, { "model_id": "gemini-3.5-flash", "benchmark_id": "gdm_mrcr_v2_8needle_upto_128k", "score": 77.3, "reference_url": "https://deepmind.google/models/model-cards/gemini-3-5-flash/", "reported_setting": { "mode": "thinking/reasoning", "effort": "medium (runtime default)", "tools": "none", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "Google DeepMind eval_hub MRCR v2 scorer", "temperature": "default", "snapshot": "May 2026" }, "matches_canonical": true, "source_type": "official_model_card", "audit_status": "verified", "notes": "Google Gemini 3.5 Flash May 2026 table: GDM MRCR v2 8-needle: 128k average; cumulative upto_128K. Settings and provenance are preserved per cell." }, { "model_id": "gemini-3-flash", "benchmark_id": "gdm_mrcr_v2_8needle_upto_128k", "score": 67.2, "reference_url": "https://deepmind.google/models/model-cards/gemini-3-5-flash/", "reported_setting": { "mode": "thinking/reasoning", "effort": "source does not state", "tools": "none", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "Google DeepMind eval_hub MRCR v2 scorer", "temperature": "default", "snapshot": "May 2026" }, "matches_canonical": true, "source_type": "official_model_card", "audit_status": "verified", "notes": "Google Gemini 3.5 Flash May 2026 table: GDM MRCR v2 8-needle: 128k average; cumulative upto_128K. Settings and provenance are preserved per cell." }, { "model_id": "gemini-3.1-pro", "benchmark_id": "gdm_mrcr_v2_8needle_upto_128k", "score": 84.9, "reference_url": "https://deepmind.google/models/model-cards/gemini-3-5-flash/", "reported_setting": { "mode": "thinking/reasoning", "effort": "source does not state", "tools": "none", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "Google DeepMind eval_hub MRCR v2 scorer", "temperature": "default", "snapshot": "May 2026" }, "matches_canonical": true, "source_type": "official_model_card", "audit_status": "verified", "notes": "Google Gemini 3.5 Flash May 2026 table: GDM MRCR v2 8-needle: 128k average; cumulative upto_128K. Settings and provenance are preserved per cell." }, { "model_id": "claude-sonnet-4.6", "benchmark_id": "gdm_mrcr_v2_8needle_upto_128k", "score": 84.9, "reference_url": "https://deepmind.google/models/model-cards/gemini-3-5-flash/", "reported_setting": { "mode": "thinking/reasoning", "effort": "max/best available", "tools": "none", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "Google DeepMind eval_hub MRCR v2 scorer", "temperature": "default", "snapshot": "May 2026" }, "matches_canonical": true, "source_type": "official_model_card", "audit_status": "verified", "notes": "Google Gemini 3.5 Flash May 2026 table: GDM MRCR v2 8-needle: 128k average; cumulative upto_128K. Settings and provenance are preserved per cell." }, { "model_id": "claude-opus-4.7", "benchmark_id": "gdm_mrcr_v2_8needle_upto_128k", "score": 59.3, "reference_url": "https://deepmind.google/models/model-cards/gemini-3-5-flash/", "reported_setting": { "mode": "thinking/reasoning", "effort": "max/best available", "tools": "none", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "Google DeepMind eval_hub MRCR v2 scorer", "temperature": "default", "snapshot": "May 2026" }, "matches_canonical": true, "source_type": "official_model_card", "audit_status": "verified", "notes": "Google Gemini 3.5 Flash May 2026 table: GDM MRCR v2 8-needle: 128k average; cumulative upto_128K. Settings and provenance are preserved per cell." }, { "model_id": "gpt-5.5", "benchmark_id": "gdm_mrcr_v2_8needle_upto_128k", "score": 94.8, "reference_url": "https://deepmind.google/models/model-cards/gemini-3-5-flash/", "reported_setting": { "mode": "thinking/reasoning", "effort": "xhigh/best available", "tools": "none", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "Google DeepMind eval_hub MRCR v2 scorer", "temperature": "default", "snapshot": "May 2026" }, "matches_canonical": true, "source_type": "official_model_card", "audit_status": "verified", "notes": "Google Gemini 3.5 Flash May 2026 table: GDM MRCR v2 8-needle: 128k average; cumulative upto_128K. Settings and provenance are preserved per cell." }, { "model_id": "gemini-3.5-flash", "benchmark_id": "gdm_mrcr_v2_8needle_1m", "score": 26.6, "reference_url": "https://deepmind.google/models/model-cards/gemini-3-5-flash/", "reported_setting": { "mode": "thinking/reasoning", "effort": "medium (runtime default)", "tools": "none", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "Google DeepMind eval_hub MRCR v2 scorer", "temperature": "default", "snapshot": "May 2026" }, "matches_canonical": true, "source_type": "official_model_card", "audit_status": "verified", "notes": "Google Gemini 3.5 Flash May 2026 table: GDM MRCR v2 8-needle: 1M pointwise; single (524288,1048576] context bin. Settings and provenance are preserved per cell." }, { "model_id": "gemini-3-flash", "benchmark_id": "gdm_mrcr_v2_8needle_1m", "score": 22.1, "reference_url": "https://deepmind.google/models/model-cards/gemini-3-5-flash/", "reported_setting": { "mode": "thinking/reasoning", "effort": "source does not state", "tools": "none", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "Google DeepMind eval_hub MRCR v2 scorer", "temperature": "default", "snapshot": "May 2026" }, "matches_canonical": true, "source_type": "official_model_card", "audit_status": "verified", "notes": "Google Gemini 3.5 Flash May 2026 table: GDM MRCR v2 8-needle: 1M pointwise; single (524288,1048576] context bin. Settings and provenance are preserved per cell." }, { "model_id": "gemini-3.1-pro", "benchmark_id": "gdm_mrcr_v2_8needle_1m", "score": 26.3, "reference_url": "https://deepmind.google/models/model-cards/gemini-3-5-flash/", "reported_setting": { "mode": "thinking/reasoning", "effort": "source does not state", "tools": "none", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "Google DeepMind eval_hub MRCR v2 scorer", "temperature": "default", "snapshot": "May 2026" }, "matches_canonical": true, "source_type": "official_model_card", "audit_status": "verified", "notes": "Google Gemini 3.5 Flash May 2026 table: GDM MRCR v2 8-needle: 1M pointwise; single (524288,1048576] context bin. Settings and provenance are preserved per cell." }, { "model_id": "gemini-3.5-flash", "benchmark_id": "arc_agi_2", "score": 72.1, "reference_url": "https://deepmind.google/models/model-cards/gemini-3-5-flash/", "reported_setting": { "mode": "thinking/reasoning", "effort": "medium (runtime default)", "tools": "none", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "ARC Prize verified leaderboard", "temperature": "default", "snapshot": "May 2026" }, "matches_canonical": true, "source_type": "official_model_card", "audit_status": "verified", "notes": "Google Gemini 3.5 Flash May 2026 table: ARC-AGI-2; ARC Prize Verified; semi-private. Settings and provenance are preserved per cell." }, { "model_id": "gemini-3.5-flash-lite", "benchmark_id": "swe_bench_pro", "score": 54.2, "reference_url": "https://deepmind.google/models/model-cards/gemini-3-5-flash-lite/", "reported_setting": { "mode": "thinking/reasoning", "effort": "high", "tools": "code execution", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "Google internal Antigravity for Gemini; provider for others", "temperature": "default", "snapshot": "July 2026" }, "matches_canonical": false, "source_type": "official_model_card", "audit_status": "verified", "notes": "Google Gemini 3.5 Flash-Lite July 2026 matrix: SWE-Bench Pro (Public). Exact source effort, tools, sampling and harness are preserved." }, { "model_id": "gemini-3.1-flash-lite", "benchmark_id": "swe_bench_pro", "score": 38.3, "reference_url": "https://deepmind.google/models/model-cards/gemini-3-5-flash-lite/", "reported_setting": { "mode": "thinking/reasoning", "effort": "high", "tools": "code execution", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "Google internal Antigravity for Gemini; provider for others", "temperature": "default", "snapshot": "July 2026" }, "matches_canonical": false, "source_type": "official_model_card", "audit_status": "verified", "notes": "Google Gemini 3.5 Flash-Lite July 2026 matrix: SWE-Bench Pro (Public). Exact source effort, tools, sampling and harness are preserved." }, { "model_id": "gpt-5.4-mini", "benchmark_id": "swe_bench_pro", "score": 54.4, "reference_url": "https://deepmind.google/models/model-cards/gemini-3-5-flash-lite/", "reported_setting": { "mode": "thinking/reasoning", "effort": "xhigh/best available", "tools": "code execution", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "Google internal Antigravity for Gemini; provider for others", "temperature": "default", "snapshot": "July 2026" }, "matches_canonical": false, "source_type": "official_model_card", "audit_status": "verified", "notes": "Google Gemini 3.5 Flash-Lite July 2026 matrix: SWE-Bench Pro (Public). Exact source effort, tools, sampling and harness are preserved." }, { "model_id": "gemini-3.5-flash-lite", "benchmark_id": "terminal_bench_2_1", "score": 54.0, "reference_url": "https://deepmind.google/models/model-cards/gemini-3-5-flash-lite/", "reported_setting": { "mode": "thinking/reasoning", "effort": "high", "tools": "terminal agent", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "Terminus-2", "temperature": "default", "snapshot": "July 2026" }, "matches_canonical": false, "source_type": "official_model_card", "audit_status": "verified", "notes": "Google Gemini 3.5 Flash-Lite July 2026 matrix: Terminal-Bench 2.1. Exact source effort, tools, sampling and harness are preserved." }, { "model_id": "gemini-3.1-flash-lite", "benchmark_id": "terminal_bench_2_1", "score": 31.0, "reference_url": "https://deepmind.google/models/model-cards/gemini-3-5-flash-lite/", "reported_setting": { "mode": "thinking/reasoning", "effort": "high", "tools": "terminal agent", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "Terminus-2", "temperature": "default", "snapshot": "July 2026" }, "matches_canonical": false, "source_type": "official_model_card", "audit_status": "verified", "notes": "Google Gemini 3.5 Flash-Lite July 2026 matrix: Terminal-Bench 2.1. Exact source effort, tools, sampling and harness are preserved." }, { "model_id": "gpt-5.4-mini", "benchmark_id": "terminal_bench_2_1", "score": 59.2, "reference_url": "https://deepmind.google/models/model-cards/gemini-3-5-flash-lite/", "reported_setting": { "mode": "thinking/reasoning", "effort": "xhigh/best available", "tools": "terminal agent", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "Terminus-2", "temperature": "default", "snapshot": "July 2026" }, "matches_canonical": false, "source_type": "official_model_card", "audit_status": "verified", "notes": "Google Gemini 3.5 Flash-Lite July 2026 matrix: Terminal-Bench 2.1. Exact source effort, tools, sampling and harness are preserved." }, { "model_id": "claude-haiku-4.5", "benchmark_id": "terminal_bench_2_1", "score": 44.2, "reference_url": "https://deepmind.google/models/model-cards/gemini-3-5-flash-lite/", "reported_setting": { "mode": "thinking/reasoning", "effort": "extended thinking/best available", "tools": "terminal agent", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "Terminus-2", "temperature": "default", "snapshot": "July 2026" }, "matches_canonical": false, "source_type": "official_model_card", "audit_status": "verified", "notes": "Google Gemini 3.5 Flash-Lite July 2026 matrix: Terminal-Bench 2.1. Exact source effort, tools, sampling and harness are preserved." }, { "model_id": "gemini-3.5-flash-lite", "benchmark_id": "mle_bench_partial30_avg_position_k2", "score": 39.2, "reference_url": "https://deepmind.google/models/model-cards/gemini-3-5-flash-lite/", "reported_setting": { "mode": "thinking/reasoning", "effort": "high", "tools": "interactive Bash + internet + H100", "sampling": "2 independent runs/problem", "judge": "Kaggle private-leaderboard rank", "harness": "Google MLE-Bench agent harness", "temperature": "default", "snapshot": "July 2026" }, "matches_canonical": false, "source_type": "official_model_card", "audit_status": "verified", "notes": "Google Gemini 3.5 Flash-Lite July 2026 matrix: MLE-Bench Partial 30: Average Position Score, k=2. Exact source effort, tools, sampling and harness are preserved." }, { "model_id": "gemini-3.1-flash-lite", "benchmark_id": "mle_bench_partial30_avg_position_k2", "score": 22.0, "reference_url": "https://deepmind.google/models/model-cards/gemini-3-5-flash-lite/", "reported_setting": { "mode": "thinking/reasoning", "effort": "high", "tools": "interactive Bash + internet + H100", "sampling": "2 independent runs/problem", "judge": "Kaggle private-leaderboard rank", "harness": "Google MLE-Bench agent harness", "temperature": "default", "snapshot": "July 2026" }, "matches_canonical": false, "source_type": "official_model_card", "audit_status": "verified", "notes": "Google Gemini 3.5 Flash-Lite July 2026 matrix: MLE-Bench Partial 30: Average Position Score, k=2. Exact source effort, tools, sampling and harness are preserved." }, { "model_id": "gemini-3.5-flash-lite", "benchmark_id": "gdpval_aa_v2_elo", "score": 1140, "reference_url": "https://deepmind.google/models/model-cards/gemini-3-5-flash-lite/", "reported_setting": { "mode": "thinking/reasoning", "effort": "high", "tools": "Stirrup/E2B agent sandbox", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "Artificial Analysis GDPval-AA v2 leaderboard", "temperature": "default", "snapshot": "July 2026" }, "matches_canonical": false, "source_type": "official_model_card", "audit_status": "verified", "notes": "Google Gemini 3.5 Flash-Lite July 2026 matrix: GDPval-AA v2 Elo. Exact source effort, tools, sampling and harness are preserved." }, { "model_id": "gemini-3.1-flash-lite", "benchmark_id": "gdpval_aa_v2_elo", "score": 642, "reference_url": "https://deepmind.google/models/model-cards/gemini-3-5-flash-lite/", "reported_setting": { "mode": "thinking/reasoning", "effort": "high", "tools": "Stirrup/E2B agent sandbox", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "Artificial Analysis GDPval-AA v2 leaderboard", "temperature": "default", "snapshot": "July 2026" }, "matches_canonical": false, "source_type": "official_model_card", "audit_status": "verified", "notes": "Google Gemini 3.5 Flash-Lite July 2026 matrix: GDPval-AA v2 Elo. Exact source effort, tools, sampling and harness are preserved." }, { "model_id": "gpt-5.4-mini", "benchmark_id": "gdpval_aa_v2_elo", "score": 1171, "reference_url": "https://deepmind.google/models/model-cards/gemini-3-5-flash-lite/", "reported_setting": { "mode": "thinking/reasoning", "effort": "xhigh/best available", "tools": "Stirrup/E2B agent sandbox", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "Artificial Analysis GDPval-AA v2 leaderboard", "temperature": "default", "snapshot": "July 2026" }, "matches_canonical": false, "source_type": "official_model_card", "audit_status": "verified", "notes": "Google Gemini 3.5 Flash-Lite July 2026 matrix: GDPval-AA v2 Elo. Exact source effort, tools, sampling and harness are preserved." }, { "model_id": "claude-haiku-4.5", "benchmark_id": "gdpval_aa_v2_elo", "score": 907, "reference_url": "https://deepmind.google/models/model-cards/gemini-3-5-flash-lite/", "reported_setting": { "mode": "thinking/reasoning", "effort": "extended thinking/best available", "tools": "Stirrup/E2B agent sandbox", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "Artificial Analysis GDPval-AA v2 leaderboard", "temperature": "default", "snapshot": "July 2026" }, "matches_canonical": true, "source_type": "official_model_card", "audit_status": "verified", "notes": "Google Gemini 3.5 Flash-Lite July 2026 matrix: GDPval-AA v2 Elo. Exact source effort, tools, sampling and harness are preserved." }, { "model_id": "gemini-3.5-flash-lite", "benchmark_id": "osworld_verified", "score": 74.0, "reference_url": "https://deepmind.google/models/model-cards/gemini-3-5-flash-lite/", "reported_setting": { "mode": "thinking/reasoning", "effort": "high", "tools": "pyautogui + UI-specific functions", "sampling": "Gemini avg 5 runs; single attempt/run", "judge": "benchmark-specified", "harness": "official OSWorld Docker, 1080p, max 100 steps", "temperature": "default", "snapshot": "July 2026" }, "matches_canonical": false, "source_type": "official_model_card", "audit_status": "verified", "notes": "Google Gemini 3.5 Flash-Lite July 2026 matrix: OSWorld-Verified. Exact source effort, tools, sampling and harness are preserved." }, { "model_id": "gemini-3.1-flash-lite", "benchmark_id": "osworld_verified", "score": 54.3, "reference_url": "https://deepmind.google/models/model-cards/gemini-3-5-flash-lite/", "reported_setting": { "mode": "thinking/reasoning", "effort": "high", "tools": "pyautogui + UI-specific functions", "sampling": "Gemini avg 5 runs; single attempt/run", "judge": "benchmark-specified", "harness": "official OSWorld Docker, 1080p, max 100 steps", "temperature": "default", "snapshot": "July 2026" }, "matches_canonical": false, "source_type": "official_model_card", "audit_status": "verified", "notes": "Google Gemini 3.5 Flash-Lite July 2026 matrix: OSWorld-Verified. Exact source effort, tools, sampling and harness are preserved." }, { "model_id": "gpt-5.4-mini", "benchmark_id": "osworld_verified", "score": 72.1, "reference_url": "https://deepmind.google/models/model-cards/gemini-3-5-flash-lite/", "reported_setting": { "mode": "thinking/reasoning", "effort": "xhigh/best available", "tools": "pyautogui + UI-specific functions", "sampling": "Gemini avg 5 runs; single attempt/run", "judge": "benchmark-specified", "harness": "official OSWorld Docker, 1080p, max 100 steps", "temperature": "default", "snapshot": "July 2026" }, "matches_canonical": false, "source_type": "official_model_card", "audit_status": "verified", "notes": "Google Gemini 3.5 Flash-Lite July 2026 matrix: OSWorld-Verified. Exact source effort, tools, sampling and harness are preserved." }, { "model_id": "claude-haiku-4.5", "benchmark_id": "osworld_verified", "score": 50.7, "reference_url": "https://deepmind.google/models/model-cards/gemini-3-5-flash-lite/", "reported_setting": { "mode": "thinking/reasoning", "effort": "extended thinking/best available", "tools": "pyautogui + UI-specific functions", "sampling": "Gemini avg 5 runs; single attempt/run", "judge": "benchmark-specified", "harness": "official OSWorld Docker, 1080p, max 100 steps", "temperature": "default", "snapshot": "July 2026" }, "matches_canonical": false, "source_type": "official_model_card", "audit_status": "verified", "notes": "Google Gemini 3.5 Flash-Lite July 2026 matrix: OSWorld-Verified. Exact source effort, tools, sampling and harness are preserved." }, { "model_id": "gemini-3.5-flash-lite", "benchmark_id": "charxiv_reasoning", "score": 74.5, "reference_url": "https://deepmind.google/models/model-cards/gemini-3-5-flash-lite/", "reported_setting": { "mode": "thinking/reasoning", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "official CharXiv binary answer judge", "harness": "Google self-computed", "temperature": "default", "snapshot": "July 2026" }, "matches_canonical": false, "source_type": "official_model_card", "audit_status": "verified", "notes": "Google Gemini 3.5 Flash-Lite July 2026 matrix: CharXiv Reasoning: no tools. Exact source effort, tools, sampling and harness are preserved.", "candidates": [ { "score": 76.5, "reference_url": "https://deepmind.google/models/model-cards/gemini-3-5-flash-lite/", "source_type": "official_model_card", "reported_setting": { "mode": "thinking/reasoning", "effort": "high", "tools": "search + code execution", "sampling": "pass@1", "judge": "official CharXiv binary answer judge", "harness": "Google self-computed; search/code details unpublished", "temperature": "default", "snapshot": "July 2026" }, "notes": "Same CharXiv validation reasoning set with search and code execution." } ] }, { "model_id": "gemini-3.1-flash-lite", "benchmark_id": "charxiv_reasoning", "score": 73.2, "reference_url": "https://deepmind.google/models/model-cards/gemini-3-5-flash-lite/", "reported_setting": { "mode": "thinking/reasoning", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "official CharXiv binary answer judge", "harness": "Google self-computed", "temperature": "default", "snapshot": "July 2026" }, "matches_canonical": false, "source_type": "official_model_card", "audit_status": "verified", "notes": "Google Gemini 3.5 Flash-Lite July 2026 matrix: CharXiv Reasoning: no tools. Exact source effort, tools, sampling and harness are preserved.", "candidates": [ { "score": 75.6, "reference_url": "https://deepmind.google/models/model-cards/gemini-3-5-flash-lite/", "source_type": "official_model_card", "reported_setting": { "mode": "thinking/reasoning", "effort": "high", "tools": "search + code execution", "sampling": "pass@1", "judge": "official CharXiv binary answer judge", "harness": "Google self-computed; search/code details unpublished", "temperature": "default", "snapshot": "July 2026" }, "notes": "Same CharXiv validation reasoning set with search and code execution." } ] }, { "model_id": "gpt-5.4-mini", "benchmark_id": "charxiv_reasoning", "score": 80.3, "reference_url": "https://deepmind.google/models/model-cards/gemini-3-5-flash-lite/", "reported_setting": { "mode": "thinking/reasoning", "effort": "xhigh/best available", "tools": "none", "sampling": "pass@1", "judge": "official CharXiv binary answer judge", "harness": "Google self-computed", "temperature": "default", "snapshot": "July 2026" }, "matches_canonical": false, "source_type": "official_model_card", "audit_status": "verified", "notes": "Google Gemini 3.5 Flash-Lite July 2026 matrix: CharXiv Reasoning: no tools. Exact source effort, tools, sampling and harness are preserved." }, { "model_id": "claude-haiku-4.5", "benchmark_id": "charxiv_reasoning", "score": 61.7, "reference_url": "https://deepmind.google/models/model-cards/gemini-3-5-flash-lite/", "reported_setting": { "mode": "thinking/reasoning", "effort": "extended thinking/best available", "tools": "none", "sampling": "pass@1", "judge": "official CharXiv binary answer judge", "harness": "Google self-computed", "temperature": "default", "snapshot": "July 2026" }, "matches_canonical": true, "source_type": "official_model_card", "audit_status": "verified", "notes": "Google Gemini 3.5 Flash-Lite July 2026 matrix: CharXiv Reasoning: no tools. Exact source effort, tools, sampling and harness are preserved." }, { "model_id": "gemini-3.5-flash-lite", "benchmark_id": "gdm_mrcr_v2_8needle_upto_128k", "score": 72.2, "reference_url": "https://deepmind.google/models/model-cards/gemini-3-5-flash-lite/", "reported_setting": { "mode": "thinking/reasoning", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "hash check + SequenceMatcher strict scorer", "harness": "Google DeepMind eval_hub MRCR v2", "temperature": "default", "snapshot": "July 2026" }, "matches_canonical": false, "source_type": "official_model_card", "audit_status": "verified", "notes": "Google Gemini 3.5 Flash-Lite July 2026 matrix: GDM MRCR v2 8-needle: 128k average. Exact source effort, tools, sampling and harness are preserved." }, { "model_id": "gemini-3.1-flash-lite", "benchmark_id": "gdm_mrcr_v2_8needle_upto_128k", "score": 60.1, "reference_url": "https://deepmind.google/models/model-cards/gemini-3-5-flash-lite/", "reported_setting": { "mode": "thinking/reasoning", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "hash check + SequenceMatcher strict scorer", "harness": "Google DeepMind eval_hub MRCR v2", "temperature": "default", "snapshot": "July 2026" }, "matches_canonical": false, "source_type": "official_model_card", "audit_status": "verified", "notes": "Google Gemini 3.5 Flash-Lite July 2026 matrix: GDM MRCR v2 8-needle: 128k average. Exact source effort, tools, sampling and harness are preserved." }, { "model_id": "gpt-5.4-mini", "benchmark_id": "gdm_mrcr_v2_8needle_upto_128k", "score": 42.7, "reference_url": "https://deepmind.google/models/model-cards/gemini-3-5-flash-lite/", "reported_setting": { "mode": "thinking/reasoning", "effort": "xhigh/best available", "tools": "none", "sampling": "pass@1", "judge": "hash check + SequenceMatcher strict scorer", "harness": "Google DeepMind eval_hub MRCR v2", "temperature": "default", "snapshot": "July 2026" }, "matches_canonical": false, "source_type": "official_model_card", "audit_status": "verified", "notes": "Google Gemini 3.5 Flash-Lite July 2026 matrix: GDM MRCR v2 8-needle: 128k average. Exact source effort, tools, sampling and harness are preserved." }, { "model_id": "claude-haiku-4.5", "benchmark_id": "gdm_mrcr_v2_8needle_upto_128k", "score": 35.3, "reference_url": "https://deepmind.google/models/model-cards/gemini-3-5-flash-lite/", "reported_setting": { "mode": "thinking/reasoning", "effort": "extended thinking/best available", "tools": "none", "sampling": "pass@1", "judge": "hash check + SequenceMatcher strict scorer", "harness": "Google DeepMind eval_hub MRCR v2", "temperature": "default", "snapshot": "July 2026" }, "matches_canonical": true, "source_type": "official_model_card", "audit_status": "verified", "notes": "Google Gemini 3.5 Flash-Lite July 2026 matrix: GDM MRCR v2 8-needle: 128k average. Exact source effort, tools, sampling and harness are preserved." }, { "model_id": "gemini-3.5-flash-lite", "benchmark_id": "gdm_mrcr_v2_8needle_1m", "score": 21.3, "reference_url": "https://deepmind.google/models/model-cards/gemini-3-5-flash-lite/", "reported_setting": { "mode": "thinking/reasoning", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "hash check + SequenceMatcher strict scorer", "harness": "Google DeepMind eval_hub MRCR v2", "temperature": "default", "snapshot": "July 2026" }, "matches_canonical": false, "source_type": "official_model_card", "audit_status": "verified", "notes": "Google Gemini 3.5 Flash-Lite July 2026 matrix: GDM MRCR v2 8-needle: 1M pointwise. Exact source effort, tools, sampling and harness are preserved." }, { "model_id": "gemini-3.1-flash-lite", "benchmark_id": "gdm_mrcr_v2_8needle_1m", "score": 12.3, "reference_url": "https://deepmind.google/models/model-cards/gemini-3-5-flash-lite/", "reported_setting": { "mode": "thinking/reasoning", "effort": "high", "tools": "none", "sampling": "pass@1", "judge": "hash check + SequenceMatcher strict scorer", "harness": "Google DeepMind eval_hub MRCR v2", "temperature": "default", "snapshot": "July 2026" }, "matches_canonical": false, "source_type": "official_model_card", "audit_status": "verified", "notes": "Google Gemini 3.5 Flash-Lite July 2026 matrix: GDM MRCR v2 8-needle: 1M pointwise. Exact source effort, tools, sampling and harness are preserved." }, { "model_id": "gemini-3.6-flash", "benchmark_id": "swe_bench_pro", "score": 58.7, "reference_url": "https://deepmind.google/models/model-cards/gemini-3-6-flash/", "reported_setting": { "mode": "thinking/reasoning", "effort": "medium (default)", "tools": "code execution", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "Google internal Antigravity for Gemini; provider for others", "temperature": "default", "snapshot": "July 2026" }, "matches_canonical": false, "source_type": "official_model_card", "audit_status": "verified", "notes": "Google Gemini 3.6 Flash July 2026 matrix: SWE-Bench Pro (Public). Exact effort, tools, sampling and harness are preserved." }, { "model_id": "grok-4.5", "benchmark_id": "swe_bench_pro", "score": 64.7, "reference_url": "https://media.x.ai/v1/website/4p5-5184fdf9.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "agentic repository shell/editor", "sampling": "pass@1", "judge": "verification reward and repository tests", "harness": "fixed SWE-bench Pro agent scaffold", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "card45 SWE-Bench Pro Resolve rate (%); source effort=high; benchmark protocol matches canonical=True. Unknown source fields remain unknown.", "candidates": [ { "score": 64.7, "reference_url": "https://deepmind.google/models/model-cards/gemini-3-6-flash/", "source_type": "official_model_card", "reported_setting": { "mode": "thinking/reasoning", "effort": "maximum/best available", "tools": "code execution", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "Google internal Antigravity for Gemini; provider for others", "temperature": "default", "snapshot": "July 2026" }, "notes": "Google Gemini 3.6 Flash July 2026 matrix: SWE-Bench Pro (Public). Exact effort, tools, sampling and harness are preserved." }, { "score": 64.7, "reference_url": "https://x.ai/news/grok-4-5", "source_type": "official_blog", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "agentic repository shell/editor", "sampling": "pass@1", "judge": "verification reward and repository tests", "harness": "fixed SWE-bench Pro agent scaffold", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown" }, "notes": "release45 SWE-Bench Pro Resolve rate (%); source effort=unknown; benchmark protocol matches canonical=True. Unknown source fields remain unknown." } ] }, { "model_id": "gemini-3.6-flash", "benchmark_id": "deep_swe_v1_1", "score": 49.0, "reference_url": "https://deepmind.google/models/model-cards/gemini-3-6-flash/", "reported_setting": { "mode": "thinking/reasoning", "effort": "high", "tools": "code execution", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "DataCurve DeepSWE v1.1 leaderboard", "temperature": "default", "snapshot": "July 2026" }, "matches_canonical": false, "source_type": "official_model_card", "audit_status": "verified", "notes": "Google Gemini 3.6 Flash July 2026 matrix: DeepSWE v1.1. Exact effort, tools, sampling and harness are preserved.", "candidates": [ { "score": 48.6, "reference_url": "https://storage.googleapis.com/deepmind-media/gemini/gemini_3-7_flash_model_evaluation.pdf", "source_type": "tech_report", "reported_setting": { "mode": "thinking", "effort": "high", "prompt_style": "not disclosed", "temperature": "not disclosed", "context": "not disclosed", "output_limit": "not disclosed", "trials": "not disclosed", "dataset_version": "DeepSWE v1.1", "dataset_split": "113 tasks across 91 repositories", "tools": "mini-swe-agent repository shell/editor toolset", "sampling": "pass@1", "judge": "DeepSWE programmatic verifier", "harness": "historic official DeepSWE v1.1 mini-swe-agent leaderboard observation" }, "notes": "Long-horizon software engineering. Physical rendering retained even when semantically identical to another official rendering." }, { "score": 47.0, "reference_url": "https://deepswe.datacurve.ai/", "source_type": "leaderboard", "reported_setting": { "mode": "thinking", "effort": "high", "prompt_style": "not disclosed", "temperature": "not disclosed", "context": "not disclosed", "output_limit": "not disclosed", "trials": "not disclosed", "dataset_version": "DeepSWE v1.1", "dataset_split": "113 tasks across 91 repositories", "tools": "mini-swe-agent repository shell/editor toolset", "sampling": "pass@1", "judge": "functional verifier plus regression checks", "harness": "official DeepSWE v1.1 mini-swe-agent leaderboard" }, "notes": "Current official leaderboard update; distinct harness or rounding/revision is preserved as an alternative." }, { "score": 40.0, "reference_url": "https://research.meta.ai/articles/introducing-muse-code-and-muse-spark-1-2/evaluations/deepswe-1-1-v1.png", "source_type": "official_blog", "reported_setting": { "mode": "thinking", "effort": "high", "prompt_style": "not disclosed", "temperature": "not disclosed", "context": "not disclosed", "agent": "Antigravity CLI", "attempts": 5, "tools": "Antigravity CLI repository shell/editor toolset", "sampling": "pass@1 averaged across five attempts per task", "judge": "functional verifier plus regression checks in a pristine verifier container", "harness": "Meta internal agent evaluation framework; Antigravity CLI; isolated Daytona sandbox; Harbor conversion", "internet": "disabled during rollout and grading; model endpoint only", "configuration": "five-language Harbor conversion with pristine verifier" }, "notes": "Exact printed value from the locked Meta release chart; evaluation settings come from the locked official methodology." } ] }, { "model_id": "gemini-3.5-flash", "benchmark_id": "deep_swe_v1_1", "score": 37.0, "reference_url": "https://deepmind.google/models/model-cards/gemini-3-6-flash/", "reported_setting": { "mode": "thinking/reasoning", "effort": "medium", "tools": "code execution", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "DataCurve DeepSWE v1.1 leaderboard", "temperature": "default", "snapshot": "July 2026" }, "matches_canonical": true, "source_type": "official_model_card", "audit_status": "verified", "notes": "Google Gemini 3.6 Flash July 2026 matrix: DeepSWE v1.1. Exact effort, tools, sampling and harness are preserved." }, { "model_id": "grok-4.5", "benchmark_id": "deep_swe_v1_1", "score": 54.0, "reference_url": "https://x.ai/news/grok-4-6", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "agentic repository shell/editor", "sampling": "pass@1", "judge": "isolated repository verifier", "harness": "mini-swe-agent run by Datacurve", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "release46 DeepSWE v1.1 Pass@1 (%); source effort=high; benchmark protocol matches canonical=True. Unknown source fields remain unknown.", "candidates": [ { "score": 54.0, "reference_url": "https://deepmind.google/models/model-cards/gemini-3-6-flash/", "source_type": "official_model_card", "reported_setting": { "mode": "thinking/reasoning", "effort": "maximum/best available", "tools": "code execution", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "DataCurve DeepSWE v1.1 leaderboard", "temperature": "default", "snapshot": "July 2026" }, "notes": "Google Gemini 3.6 Flash July 2026 matrix: DeepSWE v1.1. Exact effort, tools, sampling and harness are preserved." }, { "score": 53.0, "reference_url": "https://media.x.ai/v1/website/4p5-5184fdf9.pdf", "source_type": "model_card", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "agentic repository shell/editor", "sampling": "pass@1", "judge": "isolated repository verifier", "harness": "mini-swe-agent run by Datacurve", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown" }, "notes": "card45 DeepSWE v1.1 Pass@1 (%); source effort=high; benchmark protocol matches canonical=True. Unknown source fields remain unknown." }, { "score": 53.0, "reference_url": "https://x.ai/news/grok-4-5", "source_type": "official_blog", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "agentic repository shell/editor", "sampling": "pass@1", "judge": "isolated repository verifier", "harness": "mini-swe-agent run by Datacurve", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown" }, "notes": "release45 DeepSWE 1.1 Pass@1 (%); source effort=unknown; benchmark protocol matches canonical=True. Unknown source fields remain unknown." }, { "score": 56.6, "reference_url": "https://research.meta.ai/articles/introducing-muse-code-and-muse-spark-1-2/evaluations/deepswe-1-1-v1.png", "source_type": "official_blog", "reported_setting": { "mode": "thinking", "effort": "high", "prompt_style": "not disclosed", "temperature": "not disclosed", "context": "not disclosed", "agent": "Grok Build", "attempts": 5, "tools": "Grok Build repository shell/editor toolset", "sampling": "pass@1 averaged across five attempts per task", "judge": "functional verifier plus regression checks in a pristine verifier container", "harness": "Meta internal agent evaluation framework; Grok Build; isolated Daytona sandbox; Harbor conversion", "internet": "disabled during rollout and grading; model endpoint only", "configuration": "five-language Harbor conversion with pristine verifier" }, "notes": "Exact printed value from the locked Meta release chart; evaluation settings come from the locked official methodology." } ] }, { "model_id": "gemini-3.6-flash", "benchmark_id": "terminal_bench_2_1", "score": 78.0, "reference_url": "https://deepmind.google/models/model-cards/gemini-3-6-flash/", "reported_setting": { "mode": "thinking/reasoning", "effort": "medium (default)", "tools": "terminal agent", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "Terminus-2", "temperature": "default", "snapshot": "July 2026" }, "matches_canonical": false, "source_type": "official_model_card", "audit_status": "verified", "notes": "Google Gemini 3.6 Flash July 2026 matrix: Terminal-Bench 2.1. Exact effort, tools, sampling and harness are preserved.", "candidates": [ { "score": 78.9, "reference_url": "https://research.meta.ai/articles/introducing-muse-code-and-muse-spark-1-2/evaluations/terminal-bench-2-1-v1.png", "source_type": "official_blog", "reported_setting": { "mode": "thinking", "effort": "high", "prompt_style": "not disclosed", "temperature": "not disclosed", "context": "not disclosed", "agent": "Antigravity CLI", "attempts": 5, "tools": "Antigravity CLI terminal coding toolset", "sampling": "pass@1 averaged across five attempts per task", "judge": "Terminal-Bench 2.1 executable task verifier", "harness": "Meta internal agent evaluation framework; Antigravity CLI; isolated Daytona sandbox", "internet": "not disclosed", "configuration": "selected official agent product in isolated Daytona" }, "notes": "Exact printed value from the locked Meta release chart; evaluation settings come from the locked official methodology." } ] }, { "model_id": "grok-4.5", "benchmark_id": "terminal_bench_2_1", "score": 83.3, "reference_url": "https://media.x.ai/v1/website/4p5-5184fdf9.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "container terminal", "sampling": "pass@1", "judge": "verified task-success evaluator", "harness": "Grok Build for Grok; source-reported peer harnesses", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "card45 Terminal-Bench 2.1 Task success rate (%); source effort=high; benchmark protocol matches canonical=True. Unknown source fields remain unknown.", "candidates": [ { "score": 83.3, "reference_url": "https://deepmind.google/models/model-cards/gemini-3-6-flash/", "source_type": "official_model_card", "reported_setting": { "mode": "thinking/reasoning", "effort": "maximum/best available", "tools": "terminal agent", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "Terminus-2", "temperature": "default", "snapshot": "July 2026" }, "notes": "Google Gemini 3.6 Flash July 2026 matrix: Terminal-Bench 2.1. Exact effort, tools, sampling and harness are preserved." }, { "score": 83.3, "reference_url": "https://x.ai/news/grok-4-5", "source_type": "official_blog", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "container terminal", "sampling": "pass@1", "judge": "verified task-success evaluator", "harness": "Grok Build for Grok; source-reported peer harnesses", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown" }, "notes": "release45 Terminal-Bench 2.1 Task success rate (%); source effort=unknown; benchmark protocol matches canonical=True. Unknown source fields remain unknown." }, { "score": 81.6, "reference_url": "https://research.meta.ai/articles/introducing-muse-code-and-muse-spark-1-2/evaluations/terminal-bench-2-1-v1.png", "source_type": "official_blog", "reported_setting": { "mode": "thinking", "effort": "high", "prompt_style": "not disclosed", "temperature": "not disclosed", "context": "not disclosed", "agent": "Grok Build", "attempts": 5, "tools": "Grok Build terminal coding toolset", "sampling": "pass@1 averaged across five attempts per task", "judge": "Terminal-Bench 2.1 executable task verifier", "harness": "Meta internal agent evaluation framework; Grok Build; isolated Daytona sandbox", "internet": "not disclosed", "configuration": "selected official agent product in isolated Daytona" }, "notes": "Exact printed value from the locked Meta release chart; evaluation settings come from the locked official methodology." } ] }, { "model_id": "gemini-3.6-flash", "benchmark_id": "mle_bench_partial30_avg_position_k2", "score": 63.9, "reference_url": "https://deepmind.google/models/model-cards/gemini-3-6-flash/", "reported_setting": { "mode": "thinking/reasoning", "effort": "medium (default)", "tools": "interactive Bash + internet + H100", "sampling": "2 independent runs/problem", "judge": "Kaggle private-leaderboard rank", "harness": "Google MLE-Bench agent harness", "temperature": "default", "snapshot": "July 2026" }, "matches_canonical": true, "source_type": "official_model_card", "audit_status": "verified", "notes": "Google Gemini 3.6 Flash July 2026 matrix: MLE-Bench Partial 30: Average Position Score, k=2. Exact effort, tools, sampling and harness are preserved." }, { "model_id": "gemini-3.5-flash", "benchmark_id": "mle_bench_partial30_avg_position_k2", "score": 49.7, "reference_url": "https://deepmind.google/models/model-cards/gemini-3-6-flash/", "reported_setting": { "mode": "thinking/reasoning", "effort": "medium (default)", "tools": "interactive Bash + internet + H100", "sampling": "2 independent runs/problem", "judge": "Kaggle private-leaderboard rank", "harness": "Google MLE-Bench agent harness", "temperature": "default", "snapshot": "July 2026" }, "matches_canonical": true, "source_type": "official_model_card", "audit_status": "verified", "notes": "Google Gemini 3.6 Flash July 2026 matrix: MLE-Bench Partial 30: Average Position Score, k=2. Exact effort, tools, sampling and harness are preserved." }, { "model_id": "gemini-3.1-pro", "benchmark_id": "mle_bench_partial30_avg_position_k2", "score": 42.6, "reference_url": "https://deepmind.google/models/model-cards/gemini-3-6-flash/", "reported_setting": { "mode": "thinking/reasoning", "effort": "high/default", "tools": "interactive Bash + internet + H100", "sampling": "2 independent runs/problem", "judge": "Kaggle private-leaderboard rank", "harness": "Google MLE-Bench agent harness", "temperature": "default", "snapshot": "July 2026" }, "matches_canonical": true, "source_type": "official_model_card", "audit_status": "verified", "notes": "Google Gemini 3.6 Flash July 2026 matrix: MLE-Bench Partial 30: Average Position Score, k=2. Exact effort, tools, sampling and harness are preserved." }, { "model_id": "gpt-5.6-luna", "benchmark_id": "mle_bench_partial30_avg_position_k2", "score": 47.6, "reference_url": "https://deepmind.google/models/model-cards/gemini-3-6-flash/", "reported_setting": { "mode": "thinking/reasoning", "effort": "maximum/best available", "tools": "interactive Bash + internet + H100", "sampling": "2 independent runs/problem", "judge": "Kaggle private-leaderboard rank", "harness": "Google MLE-Bench agent harness", "temperature": "default", "snapshot": "July 2026" }, "matches_canonical": true, "source_type": "official_model_card", "audit_status": "verified", "notes": "Google Gemini 3.6 Flash July 2026 matrix: MLE-Bench Partial 30: Average Position Score, k=2. Exact effort, tools, sampling and harness are preserved." }, { "model_id": "grok-4.5", "benchmark_id": "mle_bench_partial30_avg_position_k2", "score": 43.2, "reference_url": "https://deepmind.google/models/model-cards/gemini-3-6-flash/", "reported_setting": { "mode": "thinking/reasoning", "effort": "maximum/best available", "tools": "interactive Bash + internet + H100", "sampling": "2 independent runs/problem", "judge": "Kaggle private-leaderboard rank", "harness": "Google MLE-Bench agent harness", "temperature": "default", "snapshot": "July 2026" }, "matches_canonical": true, "source_type": "official_model_card", "audit_status": "verified", "notes": "Google Gemini 3.6 Flash July 2026 matrix: MLE-Bench Partial 30: Average Position Score, k=2. Exact effort, tools, sampling and harness are preserved." }, { "model_id": "claude-sonnet-5", "benchmark_id": "mle_bench_partial30_avg_position_k2", "score": 66.9, "reference_url": "https://deepmind.google/models/model-cards/gemini-3-6-flash/", "reported_setting": { "mode": "thinking/reasoning", "effort": "maximum/best available", "tools": "interactive Bash + internet + H100", "sampling": "2 independent runs/problem", "judge": "Kaggle private-leaderboard rank", "harness": "Google MLE-Bench agent harness", "temperature": "default", "snapshot": "July 2026" }, "matches_canonical": true, "source_type": "official_model_card", "audit_status": "verified", "notes": "Google Gemini 3.6 Flash July 2026 matrix: MLE-Bench Partial 30: Average Position Score, k=2. Exact effort, tools, sampling and harness are preserved." }, { "model_id": "gemini-3.6-flash", "benchmark_id": "gdpval_aa_v2_elo", "score": 1421, "reference_url": "https://deepmind.google/models/model-cards/gemini-3-6-flash/", "reported_setting": { "mode": "thinking/reasoning", "effort": "medium (default)", "tools": "Stirrup/E2B agent sandbox", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "Artificial Analysis GDPval-AA v2 leaderboard", "temperature": "default", "snapshot": "July 2026" }, "matches_canonical": true, "source_type": "official_model_card", "audit_status": "verified", "notes": "Google Gemini 3.6 Flash July 2026 matrix: GDPval-AA v2 Elo. Exact effort, tools, sampling and harness are preserved.", "candidates": [ { "score": 1422, "reference_url": "https://storage.googleapis.com/deepmind-media/gemini/gemini_3-7_flash_model_evaluation.pdf", "source_type": "tech_report", "reported_setting": { "mode": "thinking", "effort": "not disclosed", "prompt_style": "not disclosed", "temperature": "not disclosed", "context": "not disclosed", "output_limit": "not disclosed", "trials": 1, "tools": "Web Fetch, Web Search, View Image, Code Exec, Finish, Abandon Task", "sampling": "pass@1", "aggregation": "blind pairwise Bradley-Terry Elo over one agentic submission per task", "judge": "panel of three frontier LLM judges sampled per comparison", "harness": "Artificial Analysis Stirrup; fresh E2B sandbox; 250-turn limit", "dataset_version": "GDPval-AA v2", "dataset_split": "220 tasks" }, "notes": "Elo; knowledge work. Physical rendering retained even when semantically identical to another official rendering." }, { "score": 1423, "reference_url": "https://research.meta.ai/articles/introducing-muse-code-and-muse-spark-1-2/evaluations/gdpval-aa-v2-v1.png", "source_type": "official_blog", "reported_setting": { "mode": "thinking", "effort": "high", "prompt_style": "not disclosed", "temperature": "not disclosed", "context": "not disclosed", "agent": "Artificial Analysis Stirrup", "attempts": 1, "tools": "Web Fetch, Web Search, View Image, Code Exec, Finish, Abandon Task", "sampling": "one agentic submission per each of 220 tasks", "judge": "blind pairwise panel of three frontier LLM judges; Bradley-Terry Elo", "harness": "Artificial Analysis Stirrup; fresh E2B sandbox; 250-turn limit", "internet": "web fetch and Brave-backed web search available", "configuration": "fresh E2B sandbox per task; 250-turn limit" }, "notes": "Exact printed value from the locked Meta release chart; evaluation settings come from the locked official methodology." } ] }, { "model_id": "gpt-5.6-luna", "benchmark_id": "gdpval_aa_v2_elo", "score": 1584, "reference_url": "https://deepmind.google/models/model-cards/gemini-3-6-flash/", "reported_setting": { "mode": "thinking/reasoning", "effort": "maximum/best available", "tools": "Stirrup/E2B agent sandbox", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "Artificial Analysis GDPval-AA v2 leaderboard", "temperature": "default", "snapshot": "July 2026" }, "matches_canonical": true, "source_type": "official_model_card", "audit_status": "verified", "notes": "Google Gemini 3.6 Flash July 2026 matrix: GDPval-AA v2 Elo. Exact effort, tools, sampling and harness are preserved." }, { "model_id": "grok-4.5", "benchmark_id": "gdpval_aa_v2_elo", "score": 1526, "reference_url": "https://x.ai/news/grok-4-6", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "benchmark-specified knowledge-work tools", "sampling": "unknown", "judge": "pairwise Bradley-Terry Elo", "harness": "Artificial Analysis GDPval-AA v2 harness", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "release46 GDPVal-AA v2 Elo; source effort=high; benchmark protocol matches canonical=True. Unknown source fields remain unknown.", "candidates": [ { "score": 1535, "reference_url": "https://deepmind.google/models/model-cards/gemini-3-6-flash/", "source_type": "official_model_card", "reported_setting": { "mode": "thinking/reasoning", "effort": "maximum/best available", "tools": "Stirrup/E2B agent sandbox", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "Artificial Analysis GDPval-AA v2 leaderboard", "temperature": "default", "snapshot": "July 2026" }, "notes": "Google Gemini 3.6 Flash July 2026 matrix: GDPval-AA v2 Elo. Exact effort, tools, sampling and harness are preserved." }, { "score": 1535, "reference_url": "https://media.x.ai/v1/website/4p5-5184fdf9.pdf", "source_type": "model_card", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "benchmark-specified knowledge-work tools", "sampling": "unknown", "judge": "pairwise Bradley-Terry Elo", "harness": "Artificial Analysis GDPval-AA v2 harness", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown" }, "notes": "card45 GDPval-AA v2 AA score; source effort=high; benchmark protocol matches canonical=True. Unknown source fields remain unknown." }, { "score": 1526, "reference_url": "https://research.meta.ai/articles/introducing-muse-code-and-muse-spark-1-2/evaluations/gdpval-aa-v2-v1.png", "source_type": "official_blog", "reported_setting": { "mode": "thinking", "effort": "high", "prompt_style": "not disclosed", "temperature": "not disclosed", "context": "not disclosed", "agent": "Artificial Analysis Stirrup", "attempts": 1, "tools": "Web Fetch, Web Search, View Image, Code Exec, Finish, Abandon Task", "sampling": "one agentic submission per each of 220 tasks", "judge": "blind pairwise panel of three frontier LLM judges; Bradley-Terry Elo", "harness": "Artificial Analysis Stirrup; fresh E2B sandbox; 250-turn limit", "internet": "web fetch and Brave-backed web search available", "configuration": "fresh E2B sandbox per task; 250-turn limit" }, "notes": "Exact printed value from the locked Meta release chart; evaluation settings come from the locked official methodology." } ] }, { "model_id": "claude-sonnet-5", "benchmark_id": "gdpval_aa_v2_elo", "score": 1607, "reference_url": "https://deepmind.google/models/model-cards/gemini-3-6-flash/", "reported_setting": { "mode": "thinking/reasoning", "effort": "maximum/best available", "tools": "Stirrup/E2B agent sandbox", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "Artificial Analysis GDPval-AA v2 leaderboard", "temperature": "default", "snapshot": "July 2026" }, "matches_canonical": true, "source_type": "official_model_card", "audit_status": "verified", "notes": "Google Gemini 3.6 Flash July 2026 matrix: GDPval-AA v2 Elo. Exact effort, tools, sampling and harness are preserved.", "candidates": [ { "score": 1601, "reference_url": "https://media.x.ai/v1/website/card-4p6-4cd2dc57.pdf", "source_type": "model_card", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "benchmark-specified knowledge-work tools", "sampling": "unknown", "judge": "pairwise Bradley-Terry Elo", "harness": "Artificial Analysis GDPval-AA v2 harness", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown" }, "notes": "card46 AA GDPVal v2 AA GDPVal Elo; source effort=max; benchmark protocol matches canonical=True. Unknown source fields remain unknown." }, { "score": 1607, "reference_url": "https://media.x.ai/v1/website/4p5-5184fdf9.pdf", "source_type": "model_card", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "benchmark-specified knowledge-work tools", "sampling": "unknown", "judge": "pairwise Bradley-Terry Elo", "harness": "Artificial Analysis GDPval-AA v2 harness", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown" }, "notes": "card45 GDPval-AA v2 AA score; source effort=max; benchmark protocol matches canonical=True. Unknown source fields remain unknown." }, { "score": 1598, "reference_url": "https://storage.googleapis.com/deepmind-media/gemini/gemini_3-7_flash_model_evaluation.pdf", "source_type": "tech_report", "reported_setting": { "mode": "thinking", "effort": "max", "prompt_style": "not disclosed", "temperature": "not disclosed", "context": "not disclosed", "output_limit": "not disclosed", "trials": 1, "tools": "Web Fetch, Web Search, View Image, Code Exec, Finish, Abandon Task", "sampling": "pass@1", "aggregation": "blind pairwise Bradley-Terry Elo over one agentic submission per task", "judge": "panel of three frontier LLM judges sampled per comparison", "harness": "Artificial Analysis Stirrup; fresh E2B sandbox; 250-turn limit", "dataset_version": "GDPval-AA v2", "dataset_split": "220 tasks" }, "notes": "Elo; knowledge work. Physical rendering retained even when semantically identical to another official rendering." } ] }, { "model_id": "gemini-3.6-flash", "benchmark_id": "osworld_verified", "score": 83.0, "reference_url": "https://deepmind.google/models/model-cards/gemini-3-6-flash/", "reported_setting": { "mode": "thinking/reasoning", "effort": "medium (default)", "tools": "pyautogui + UI-specific functions", "sampling": "Google self-computed avg 5; single attempt/run", "judge": "benchmark-specified", "harness": "official OSWorld Docker, 1080p, max 100 steps", "temperature": "default", "snapshot": "July 2026" }, "matches_canonical": true, "source_type": "official_model_card", "audit_status": "verified", "notes": "Google Gemini 3.6 Flash July 2026 matrix: OSWorld-Verified. Exact effort, tools, sampling and harness are preserved." }, { "model_id": "gpt-5.6-luna", "benchmark_id": "osworld_verified", "score": 72.6, "reference_url": "https://deepmind.google/models/model-cards/gemini-3-6-flash/", "reported_setting": { "mode": "thinking/reasoning", "effort": "maximum/best available", "tools": "pyautogui + UI-specific functions", "sampling": "Google self-computed avg 5; single attempt/run", "judge": "benchmark-specified", "harness": "official OSWorld Docker, 1080p, max 100 steps", "temperature": "default", "snapshot": "July 2026" }, "matches_canonical": true, "source_type": "official_model_card", "audit_status": "verified", "notes": "Google Gemini 3.6 Flash July 2026 matrix: OSWorld-Verified. Exact effort, tools, sampling and harness are preserved." }, { "model_id": "gemini-3.6-flash", "benchmark_id": "charxiv_reasoning", "score": 85.2, "reference_url": "https://deepmind.google/models/model-cards/gemini-3-6-flash/", "reported_setting": { "mode": "thinking/reasoning", "effort": "medium (default)", "tools": "none", "sampling": "pass@1", "judge": "official CharXiv binary answer judge", "harness": "Google self-computed for Gemini/Luna/Grok", "temperature": "default", "snapshot": "July 2026" }, "matches_canonical": true, "source_type": "official_model_card", "audit_status": "verified", "notes": "Google Gemini 3.6 Flash July 2026 matrix: CharXiv Reasoning: no tools. Exact effort, tools, sampling and harness are preserved.", "candidates": [ { "score": 89.4, "reference_url": "https://deepmind.google/models/model-cards/gemini-3-6-flash/", "source_type": "official_model_card", "reported_setting": { "mode": "thinking/reasoning", "effort": "medium (default)", "tools": "search + code execution", "sampling": "pass@1", "judge": "official CharXiv binary answer judge", "harness": "Google self-computed; search/code details unpublished", "temperature": "default", "snapshot": "July 2026" }, "notes": "Same CharXiv reasoning set with search and code execution." } ] }, { "model_id": "gpt-5.6-luna", "benchmark_id": "charxiv_reasoning", "score": 82.7, "reference_url": "https://deepmind.google/models/model-cards/gemini-3-6-flash/", "reported_setting": { "mode": "thinking/reasoning", "effort": "maximum/best available", "tools": "none", "sampling": "pass@1", "judge": "official CharXiv binary answer judge", "harness": "Google self-computed for Gemini/Luna/Grok", "temperature": "default", "snapshot": "July 2026" }, "matches_canonical": true, "source_type": "official_model_card", "audit_status": "verified", "notes": "Google Gemini 3.6 Flash July 2026 matrix: CharXiv Reasoning: no tools. Exact effort, tools, sampling and harness are preserved." }, { "model_id": "grok-4.5", "benchmark_id": "charxiv_reasoning", "score": 81.6, "reference_url": "https://deepmind.google/models/model-cards/gemini-3-6-flash/", "reported_setting": { "mode": "thinking/reasoning", "effort": "maximum/best available", "tools": "none", "sampling": "pass@1", "judge": "official CharXiv binary answer judge", "harness": "Google self-computed for Gemini/Luna/Grok", "temperature": "default", "snapshot": "July 2026" }, "matches_canonical": true, "source_type": "official_model_card", "audit_status": "verified", "notes": "Google Gemini 3.6 Flash July 2026 matrix: CharXiv Reasoning: no tools. Exact effort, tools, sampling and harness are preserved." }, { "model_id": "gemini-3.6-flash", "benchmark_id": "gdm_mrcr_v2_8needle_upto_128k", "score": 91.8, "reference_url": "https://deepmind.google/models/model-cards/gemini-3-6-flash/", "reported_setting": { "mode": "thinking/reasoning", "effort": "medium (default)", "tools": "none", "sampling": "pass@1", "judge": "hash check + SequenceMatcher strict scorer", "harness": "Google DeepMind eval_hub / Context Arena", "temperature": "default", "snapshot": "July 2026" }, "matches_canonical": true, "source_type": "official_model_card", "audit_status": "verified", "notes": "Google Gemini 3.6 Flash July 2026 matrix: GDM MRCR v2 8-needle: 128k average. Exact effort, tools, sampling and harness are preserved." }, { "model_id": "gpt-5.6-luna", "benchmark_id": "gdm_mrcr_v2_8needle_upto_128k", "score": 74.8, "reference_url": "https://deepmind.google/models/model-cards/gemini-3-6-flash/", "reported_setting": { "mode": "thinking/reasoning", "effort": "maximum/best available", "tools": "none", "sampling": "pass@1", "judge": "hash check + SequenceMatcher strict scorer", "harness": "Google DeepMind eval_hub / Context Arena", "temperature": "default", "snapshot": "July 2026" }, "matches_canonical": true, "source_type": "official_model_card", "audit_status": "verified", "notes": "Google Gemini 3.6 Flash July 2026 matrix: GDM MRCR v2 8-needle: 128k average. Exact effort, tools, sampling and harness are preserved." }, { "model_id": "grok-4.5", "benchmark_id": "gdm_mrcr_v2_8needle_upto_128k", "score": 81.4, "reference_url": "https://deepmind.google/models/model-cards/gemini-3-6-flash/", "reported_setting": { "mode": "thinking/reasoning", "effort": "maximum/best available", "tools": "none", "sampling": "pass@1", "judge": "hash check + SequenceMatcher strict scorer", "harness": "Google DeepMind eval_hub / Context Arena", "temperature": "default", "snapshot": "July 2026" }, "matches_canonical": true, "source_type": "official_model_card", "audit_status": "verified", "notes": "Google Gemini 3.6 Flash July 2026 matrix: GDM MRCR v2 8-needle: 128k average. Exact effort, tools, sampling and harness are preserved." }, { "model_id": "claude-sonnet-5", "benchmark_id": "gdm_mrcr_v2_8needle_upto_128k", "score": 71.6, "reference_url": "https://deepmind.google/models/model-cards/gemini-3-6-flash/", "reported_setting": { "mode": "thinking/reasoning", "effort": "maximum/best available", "tools": "none", "sampling": "pass@1", "judge": "hash check + SequenceMatcher strict scorer", "harness": "Google DeepMind eval_hub / Context Arena", "temperature": "default", "snapshot": "July 2026" }, "matches_canonical": true, "source_type": "official_model_card", "audit_status": "verified", "notes": "Google Gemini 3.6 Flash July 2026 matrix: GDM MRCR v2 8-needle: 128k average. Exact effort, tools, sampling and harness are preserved.", "candidates": [ { "score": 81.5, "reference_url": "https://storage.googleapis.com/deepmind-media/gemini/gemini_3-7_flash_model_evaluation.pdf", "source_type": "tech_report", "reported_setting": { "mode": "thinking", "effort": "max", "prompt_style": "not disclosed", "temperature": "not disclosed", "context": "not disclosed", "output_limit": "not disclosed", "trials": "not disclosed", "tools": "none", "sampling": "not disclosed", "judge": "strict MRCR scorer", "harness": "Google self-computed GDM eval_hub", "dataset_version": "GDM MRCR v2.1, 8 needles", "dataset_split": "cumulative through 128K; 484 examples", "sequence_length": "cumulative through 128K" }, "notes": "Cumulative through 128K. Physical rendering retained even when semantically identical to another official rendering." } ] }, { "model_id": "gemini-3.6-flash", "benchmark_id": "gdm_mrcr_v2_8needle_1m", "score": 54.0, "reference_url": "https://deepmind.google/models/model-cards/gemini-3-6-flash/", "reported_setting": { "mode": "thinking/reasoning", "effort": "medium (default)", "tools": "none", "sampling": "pass@1", "judge": "hash check + SequenceMatcher strict scorer", "harness": "Google DeepMind eval_hub / Context Arena", "temperature": "default", "snapshot": "July 2026" }, "matches_canonical": true, "source_type": "official_model_card", "audit_status": "verified", "notes": "Google Gemini 3.6 Flash July 2026 matrix: GDM MRCR v2 8-needle: 1M pointwise. Exact effort, tools, sampling and harness are preserved." }, { "model_id": "deepseek-v3.2-exp-thinking", "benchmark_id": "text_arena_elo", "score": 1425, "reference_url": "https://deepmind.google/models/gemma/gemma-4/", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "none", "sampling": "live human pairwise votes", "judge": "blind human raters", "harness": "Arena Text", "prompt_style": "live user prompts", "temperature": "unknown", "context": "default", "snapshot_date": "2026-04-02" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "DeepMind page accessibility text provides the exact plotted Elo value." }, { "model_id": "deepseek-v4-flash", "benchmark_id": "text_arena_elo", "score": 1436, "reference_url": "https://arxiv.org/abs/2607.02770", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "none", "sampling": "live human pairwise votes", "judge": "blind human raters", "harness": "Arena Text", "prompt_style": "live user prompts", "temperature": "unknown", "context": "default", "snapshot_date": "2026-06-19" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Gemma 4 Technical Report Table 4 dated Arena Text Elo snapshot." }, { "model_id": "deepseek-v4-pro", "benchmark_id": "text_arena_elo", "score": 1458, "reference_url": "https://arxiv.org/abs/2607.02770", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "none", "sampling": "live human pairwise votes", "judge": "blind human raters", "harness": "Arena Text", "prompt_style": "live user prompts", "temperature": "unknown", "context": "default", "snapshot_date": "2026-06-19" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Gemma 4 Technical Report Table 4 dated Arena Text Elo snapshot." }, { "model_id": "deepseek-v4-pro-non-thinking", "benchmark_id": "text_arena_elo", "score": 1456, "reference_url": "https://arxiv.org/abs/2607.02770", "reported_setting": { "mode": "non-thinking", "effort": "unknown", "tools": "none", "sampling": "live human pairwise votes", "judge": "blind human raters", "harness": "Arena Text", "prompt_style": "live user prompts", "temperature": "unknown", "context": "default", "snapshot_date": "2026-06-19" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Gemma 4 Technical Report Table 4 dated Arena Text Elo snapshot." }, { "model_id": "gemma-3-27b", "benchmark_id": "aime_2026", "score": 20.8, "reference_url": "https://blog.google/innovation-and-ai/technology/developers-tools/gemma-4/", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "provider-reported", "judge": "benchmark-specified", "harness": "official Google evaluation", "prompt_style": "official", "temperature": "not disclosed", "context": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "Google Gemma 4 family launch table, duplicated by the DeepMind semantic table." }, { "model_id": "gemma-3-27b", "benchmark_id": "codeforces_rating", "score": 110, "reference_url": "https://arxiv.org/abs/2607.02770", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "code execution in undisclosed harness", "sampling": "provider-reported", "judge": "benchmark-specified", "harness": "official Google evaluation", "prompt_style": "official", "temperature": "not disclosed", "context": "default" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Gemma 4 Technical Report Table 5; Gemma 4 thinking unless explicitly stated, Gemma 3 27B non-thinking." }, { "model_id": "gemma-3-27b", "benchmark_id": "gdm_mrcr_v2_8needle_upto_128k", "score": 13.5, "reference_url": "https://arxiv.org/abs/2607.02770", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "provider-reported", "judge": "benchmark-specified", "harness": "official Google evaluation", "prompt_style": "official", "temperature": "not disclosed", "context": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Gemma 4 Technical Report Table 5; Gemma 4 thinking unless explicitly stated, Gemma 3 27B non-thinking." }, { "model_id": "gemma-3-27b", "benchmark_id": "graphwalks_lt128k_combined", "score": 32.8, "reference_url": "https://arxiv.org/abs/2607.02770", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "provider-reported", "judge": "benchmark-specified", "harness": "official Google evaluation", "prompt_style": "official", "temperature": "not disclosed", "context": "below 128K" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Gemma 4 Technical Report Table 9; final IT models evaluated without thinking." }, { "model_id": "gemma-3-27b", "benchmark_id": "ifbench", "score": 32.0, "reference_url": "https://arxiv.org/abs/2607.02770", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "provider-reported", "judge": "benchmark-specified", "harness": "official Google evaluation", "prompt_style": "official", "temperature": "not disclosed", "context": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Gemma 4 Technical Report Table 5; Gemma 4 thinking unless explicitly stated, Gemma 3 27B non-thinking." }, { "model_id": "gemma-3-27b", "benchmark_id": "infographicvqa", "score": 70.6, "reference_url": "https://arxiv.org/abs/2607.02770", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "provider-reported", "judge": "benchmark-specified", "harness": "official Google evaluation", "prompt_style": "official", "temperature": "not disclosed", "context": "default", "visual_tokens": "Pan & Scan" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Gemma 4 Technical Report Table 6; maximum supported resolution, 1,120 visual tokens." }, { "model_id": "gemma-3-27b", "benchmark_id": "livecodebench_v6", "score": 29.1, "reference_url": "https://blog.google/innovation-and-ai/technology/developers-tools/gemma-4/", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "provider-reported", "judge": "benchmark-specified", "harness": "official Google evaluation", "prompt_style": "official", "temperature": "not disclosed", "context": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "Google Gemma 4 family launch table, duplicated by the DeepMind semantic table." }, { "model_id": "gemma-3-27b", "benchmark_id": "loft_text_retrieval_128k", "score": 8.6, "reference_url": "https://arxiv.org/abs/2607.02770", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "provider-reported", "judge": "benchmark-specified", "harness": "official Google evaluation", "prompt_style": "official", "temperature": "not disclosed", "context": "128K" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Gemma 4 Technical Report Table 9; final IT models evaluated without thinking." }, { "model_id": "gemma-3-27b", "benchmark_id": "mathvision", "score": 46.0, "reference_url": "https://arxiv.org/abs/2607.02770", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "provider-reported", "judge": "benchmark-specified", "harness": "official Google evaluation", "prompt_style": "official", "temperature": "not disclosed", "context": "default", "visual_tokens": "Pan & Scan" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Gemma 4 Technical Report Table 6; maximum supported resolution, 1,120 visual tokens." }, { "model_id": "gemma-3-27b", "benchmark_id": "mmmlu", "score": 70.7, "reference_url": "https://arxiv.org/abs/2607.02770", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "provider-reported", "judge": "benchmark-specified", "harness": "official Google evaluation", "prompt_style": "official", "temperature": "not disclosed", "context": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Gemma 4 Technical Report Table 5; Gemma 4 thinking unless explicitly stated, Gemma 3 27B non-thinking." }, { "model_id": "gemma-3-27b", "benchmark_id": "mmmu_pro", "score": 49.7, "reference_url": "https://blog.google/innovation-and-ai/technology/developers-tools/gemma-4/", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "provider-reported", "judge": "benchmark-specified", "harness": "official Google evaluation", "prompt_style": "official", "temperature": "not disclosed", "context": "default", "visual_tokens": "Pan & Scan" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "Google Gemma 4 family launch table, duplicated by the DeepMind semantic table." }, { "model_id": "gemma-3-27b", "benchmark_id": "mtob_eng_kgv_half_book", "score": 41.0, "reference_url": "https://arxiv.org/abs/2607.02770", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "provider-reported", "judge": "benchmark-specified", "harness": "official Google evaluation", "prompt_style": "official", "temperature": "not disclosed", "context": "about 128K" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Gemma 4 Technical Report Table 9; final IT models evaluated without thinking." }, { "model_id": "gemma-3-27b", "benchmark_id": "mtob_kgv_eng_half_book", "score": 31.2, "reference_url": "https://arxiv.org/abs/2607.02770", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "provider-reported", "judge": "benchmark-specified", "harness": "official Google evaluation", "prompt_style": "official", "temperature": "not disclosed", "context": "about 128K" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Gemma 4 Technical Report Table 9; final IT models evaluated without thinking." }, { "model_id": "gemma-3-27b", "benchmark_id": "omnidocbench_1.5", "score": 0.365, "reference_url": "https://arxiv.org/abs/2607.02770", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "provider-reported", "judge": "benchmark-specified", "harness": "official Google evaluation", "prompt_style": "official", "temperature": "not disclosed", "context": "default", "visual_tokens": "Pan & Scan" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Gemma 4 Technical Report Table 6; maximum supported resolution, 1,120 visual tokens." }, { "model_id": "gemma-3-27b", "benchmark_id": "ruler_128k", "score": 66.0, "reference_url": "https://arxiv.org/abs/2607.02770", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "provider-reported", "judge": "benchmark-specified", "harness": "official Google evaluation", "prompt_style": "official", "temperature": "not disclosed", "context": "128K" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Gemma 4 Technical Report Table 9; final IT models evaluated without thinking." }, { "model_id": "gemma-3-27b", "benchmark_id": "ruler_32k", "score": 91.1, "reference_url": "https://arxiv.org/abs/2607.02770", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "provider-reported", "judge": "benchmark-specified", "harness": "official Google evaluation", "prompt_style": "official", "temperature": "not disclosed", "context": "32K" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Gemma 4 Technical Report Table 9; final IT models evaluated without thinking." }, { "model_id": "gemma-3-27b", "benchmark_id": "scicode", "score": 21.0, "reference_url": "https://arxiv.org/abs/2607.02770", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "provider-reported", "judge": "unit tests", "harness": "official Google evaluation", "prompt_style": "official", "temperature": "not disclosed", "context": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Gemma 4 Technical Report Table 5; Gemma 4 thinking unless explicitly stated, Gemma 3 27B non-thinking." }, { "model_id": "gemma-3-27b", "benchmark_id": "tau2_bench_airline", "score": 39.0, "reference_url": "https://arxiv.org/abs/2607.02770", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "benchmark APIs + simulated user", "sampling": "provider-reported", "judge": "state-based task success", "harness": "Tau2-bench", "prompt_style": "official", "temperature": "not disclosed", "context": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Gemma 4 Technical Report Table 5; Gemma 4 thinking unless explicitly stated, Gemma 3 27B non-thinking." }, { "model_id": "gemma-3-27b", "benchmark_id": "tau2_bench_avg", "score": 16.2, "reference_url": "https://ai.google.dev/gemma/docs/core/model_card_4", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "benchmark APIs + simulated user", "sampling": "provider-reported", "judge": "state-based task success", "harness": "Tau2-bench", "prompt_style": "official", "temperature": "not disclosed", "context": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Google model card matrix; all values are instruction-tuned despite the duplicated base-card surface." }, { "model_id": "gemma-3-27b", "benchmark_id": "tau2_bench_retail", "score": 6.6, "reference_url": "https://blog.google/innovation-and-ai/technology/developers-tools/gemma-4/", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "benchmark APIs + simulated user", "sampling": "provider-reported", "judge": "state-based task success", "harness": "Tau2-bench", "prompt_style": "official", "temperature": "not disclosed", "context": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "Google Gemma 4 family launch table, duplicated by the DeepMind semantic table." }, { "model_id": "gemma-3-27b", "benchmark_id": "tau2_bench_telecom", "score": 3.1, "reference_url": "https://arxiv.org/abs/2607.02770", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "benchmark APIs + simulated user", "sampling": "provider-reported", "judge": "state-based task success", "harness": "Tau2-bench", "prompt_style": "official", "temperature": "not disclosed", "context": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Gemma 4 Technical Report Table 5; Gemma 4 thinking unless explicitly stated, Gemma 3 27B non-thinking." }, { "model_id": "gemma-3-27b", "benchmark_id": "terminal_bench_hard", "score": 4.0, "reference_url": "https://arxiv.org/abs/2607.02770", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "agentic terminal", "sampling": "provider-reported", "judge": "benchmark-specified", "harness": "Terminal-Bench Hard provider harness", "prompt_style": "official", "temperature": "not disclosed", "context": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Gemma 4 Technical Report Table 5; Gemma 4 thinking unless explicitly stated, Gemma 3 27B non-thinking." }, { "model_id": "gemma-3-27b", "benchmark_id": "text_arena_elo", "score": 1365, "reference_url": "https://blog.google/innovation-and-ai/technology/developers-tools/gemma-4/", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "live human pairwise votes", "judge": "blind human raters", "harness": "Arena AI text", "prompt_style": "live user prompts", "temperature": "unknown", "context": "default", "snapshot_date": "2026-04-02" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Google Gemma 4 family launch table, duplicated by the DeepMind semantic table.", "candidates": [ { "score": 1366, "reference_url": "https://arxiv.org/abs/2607.02770", "source_type": "tech_report", "reported_setting": { "mode": "unknown", "effort": "unknown", "tools": "none", "sampling": "live human pairwise votes", "judge": "blind human raters", "harness": "Arena Text", "prompt_style": "live user prompts", "temperature": "unknown", "context": "default", "snapshot_date": "2026-06-19" }, "notes": "Gemma 4 Technical Report Table 4 dated Arena Text Elo snapshot." } ] }, { "model_id": "gemma-3n-e2b-instructed", "benchmark_id": "covost2_de_en", "score": 36.5, "reference_url": "https://arxiv.org/abs/2607.02770", "reported_setting": { "mode": "unknown", "effort": "n/a", "tools": "audio input; no external tools", "sampling": "provider-reported", "judge": "CorpusBLEU", "harness": "official Google audio evaluation", "prompt_style": "official", "temperature": "not disclosed", "context": "default" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Gemma 4 Technical Report Table 7, transcribe-then-translate CorpusBLEU." }, { "model_id": "gemma-3n-e2b-instructed", "benchmark_id": "covost2_es_en", "score": 39.9, "reference_url": "https://arxiv.org/abs/2607.02770", "reported_setting": { "mode": "unknown", "effort": "n/a", "tools": "audio input; no external tools", "sampling": "provider-reported", "judge": "CorpusBLEU", "harness": "official Google audio evaluation", "prompt_style": "official", "temperature": "not disclosed", "context": "default" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Gemma 4 Technical Report Table 7, transcribe-then-translate CorpusBLEU." }, { "model_id": "gemma-3n-e2b-instructed", "benchmark_id": "covost2_fr_en", "score": 35.7, "reference_url": "https://arxiv.org/abs/2607.02770", "reported_setting": { "mode": "unknown", "effort": "n/a", "tools": "audio input; no external tools", "sampling": "provider-reported", "judge": "CorpusBLEU", "harness": "official Google audio evaluation", "prompt_style": "official", "temperature": "not disclosed", "context": "default" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Gemma 4 Technical Report Table 7, transcribe-then-translate CorpusBLEU." }, { "model_id": "gemma-3n-e2b-instructed", "benchmark_id": "covost2_it_en", "score": 38.5, "reference_url": "https://arxiv.org/abs/2607.02770", "reported_setting": { "mode": "unknown", "effort": "n/a", "tools": "audio input; no external tools", "sampling": "provider-reported", "judge": "CorpusBLEU", "harness": "official Google audio evaluation", "prompt_style": "official", "temperature": "not disclosed", "context": "default" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Gemma 4 Technical Report Table 7, transcribe-then-translate CorpusBLEU." }, { "model_id": "gemma-3n-e2b-instructed", "benchmark_id": "covost2_ja_en", "score": 17.7, "reference_url": "https://arxiv.org/abs/2607.02770", "reported_setting": { "mode": "unknown", "effort": "n/a", "tools": "audio input; no external tools", "sampling": "provider-reported", "judge": "CorpusBLEU", "harness": "official Google audio evaluation", "prompt_style": "official", "temperature": "not disclosed", "context": "default" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Gemma 4 Technical Report Table 7, transcribe-then-translate CorpusBLEU." }, { "model_id": "gemma-3n-e2b-instructed", "benchmark_id": "covost2_ru_en", "score": 39.2, "reference_url": "https://arxiv.org/abs/2607.02770", "reported_setting": { "mode": "unknown", "effort": "n/a", "tools": "audio input; no external tools", "sampling": "provider-reported", "judge": "CorpusBLEU", "harness": "official Google audio evaluation", "prompt_style": "official", "temperature": "not disclosed", "context": "default" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Gemma 4 Technical Report Table 7, transcribe-then-translate CorpusBLEU." }, { "model_id": "gemma-3n-e2b-instructed", "benchmark_id": "covost2_xx_en_7lang_macro", "score": 31.6, "reference_url": "https://arxiv.org/abs/2607.02770", "reported_setting": { "mode": "unknown", "effort": "n/a", "tools": "audio input; no external tools", "sampling": "provider-reported", "judge": "CorpusBLEU", "harness": "official Google audio evaluation", "prompt_style": "official", "temperature": "not disclosed", "context": "default" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Gemma 4 Technical Report Table 7, transcribe-then-translate CorpusBLEU." }, { "model_id": "gemma-3n-e2b-instructed", "benchmark_id": "covost2_zh_cn_en", "score": 13.9, "reference_url": "https://arxiv.org/abs/2607.02770", "reported_setting": { "mode": "unknown", "effort": "n/a", "tools": "audio input; no external tools", "sampling": "provider-reported", "judge": "CorpusBLEU", "harness": "official Google audio evaluation", "prompt_style": "official", "temperature": "not disclosed", "context": "default" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Gemma 4 Technical Report Table 7, transcribe-then-translate CorpusBLEU." }, { "model_id": "gemma-3n-e2b-instructed", "benchmark_id": "fleurs", "score": 0.108, "reference_url": "https://arxiv.org/abs/2607.02770", "reported_setting": { "mode": "unknown", "effort": "n/a", "tools": "audio input; no external tools", "sampling": "provider-reported", "judge": "WER/CER", "harness": "official Google audio evaluation", "prompt_style": "official", "temperature": "not disclosed", "context": "default" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Gemma 4 Technical Report Table 7; WER except CER for Korean, Japanese and Chinese." }, { "model_id": "gemma-3n-e2b-instructed", "benchmark_id": "fleurs_asr_ar_eg", "score": 0.131, "reference_url": "https://arxiv.org/abs/2607.02770", "reported_setting": { "mode": "unknown", "effort": "n/a", "tools": "audio input; no external tools", "sampling": "provider-reported", "judge": "WER/CER", "harness": "official Google audio evaluation", "prompt_style": "official", "temperature": "not disclosed", "context": "default" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Gemma 4 Technical Report Table 7; WER except CER for Korean, Japanese and Chinese." }, { "model_id": "gemma-3n-e2b-instructed", "benchmark_id": "fleurs_asr_de_de", "score": 0.079, "reference_url": "https://arxiv.org/abs/2607.02770", "reported_setting": { "mode": "unknown", "effort": "n/a", "tools": "audio input; no external tools", "sampling": "provider-reported", "judge": "WER/CER", "harness": "official Google audio evaluation", "prompt_style": "official", "temperature": "not disclosed", "context": "default" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Gemma 4 Technical Report Table 7; WER except CER for Korean, Japanese and Chinese." }, { "model_id": "gemma-3n-e2b-instructed", "benchmark_id": "fleurs_asr_en_us", "score": 0.076, "reference_url": "https://arxiv.org/abs/2607.02770", "reported_setting": { "mode": "unknown", "effort": "n/a", "tools": "audio input; no external tools", "sampling": "provider-reported", "judge": "WER/CER", "harness": "official Google audio evaluation", "prompt_style": "official", "temperature": "not disclosed", "context": "default" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Gemma 4 Technical Report Table 7; WER except CER for Korean, Japanese and Chinese." }, { "model_id": "gemma-3n-e2b-instructed", "benchmark_id": "fleurs_asr_es_419", "score": 0.051, "reference_url": "https://arxiv.org/abs/2607.02770", "reported_setting": { "mode": "unknown", "effort": "n/a", "tools": "audio input; no external tools", "sampling": "provider-reported", "judge": "WER/CER", "harness": "official Google audio evaluation", "prompt_style": "official", "temperature": "not disclosed", "context": "default" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Gemma 4 Technical Report Table 7; WER except CER for Korean, Japanese and Chinese." }, { "model_id": "gemma-3n-e2b-instructed", "benchmark_id": "fleurs_asr_fr_fr", "score": 0.13, "reference_url": "https://arxiv.org/abs/2607.02770", "reported_setting": { "mode": "unknown", "effort": "n/a", "tools": "audio input; no external tools", "sampling": "provider-reported", "judge": "WER/CER", "harness": "official Google audio evaluation", "prompt_style": "official", "temperature": "not disclosed", "context": "default" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Gemma 4 Technical Report Table 7; WER except CER for Korean, Japanese and Chinese." }, { "model_id": "gemma-3n-e2b-instructed", "benchmark_id": "fleurs_asr_hi_in", "score": 0.106, "reference_url": "https://arxiv.org/abs/2607.02770", "reported_setting": { "mode": "unknown", "effort": "n/a", "tools": "audio input; no external tools", "sampling": "provider-reported", "judge": "WER/CER", "harness": "official Google audio evaluation", "prompt_style": "official", "temperature": "not disclosed", "context": "default" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Gemma 4 Technical Report Table 7; WER except CER for Korean, Japanese and Chinese." }, { "model_id": "gemma-3n-e2b-instructed", "benchmark_id": "fleurs_asr_it_it", "score": 0.044, "reference_url": "https://arxiv.org/abs/2607.02770", "reported_setting": { "mode": "unknown", "effort": "n/a", "tools": "audio input; no external tools", "sampling": "provider-reported", "judge": "WER/CER", "harness": "official Google audio evaluation", "prompt_style": "official", "temperature": "not disclosed", "context": "default" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Gemma 4 Technical Report Table 7; WER except CER for Korean, Japanese and Chinese." }, { "model_id": "gemma-3n-e2b-instructed", "benchmark_id": "fleurs_asr_ja_jp_cer", "score": 0.163, "reference_url": "https://arxiv.org/abs/2607.02770", "reported_setting": { "mode": "unknown", "effort": "n/a", "tools": "audio input; no external tools", "sampling": "provider-reported", "judge": "WER/CER", "harness": "official Google audio evaluation", "prompt_style": "official", "temperature": "not disclosed", "context": "default" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Gemma 4 Technical Report Table 7; WER except CER for Korean, Japanese and Chinese." }, { "model_id": "gemma-3n-e2b-instructed", "benchmark_id": "fleurs_asr_ko_kr_cer", "score": 0.101, "reference_url": "https://arxiv.org/abs/2607.02770", "reported_setting": { "mode": "unknown", "effort": "n/a", "tools": "audio input; no external tools", "sampling": "provider-reported", "judge": "WER/CER", "harness": "official Google audio evaluation", "prompt_style": "official", "temperature": "not disclosed", "context": "default" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Gemma 4 Technical Report Table 7; WER except CER for Korean, Japanese and Chinese." }, { "model_id": "gemma-3n-e2b-instructed", "benchmark_id": "fleurs_asr_pt_br", "score": 0.067, "reference_url": "https://arxiv.org/abs/2607.02770", "reported_setting": { "mode": "unknown", "effort": "n/a", "tools": "audio input; no external tools", "sampling": "provider-reported", "judge": "WER/CER", "harness": "official Google audio evaluation", "prompt_style": "official", "temperature": "not disclosed", "context": "default" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Gemma 4 Technical Report Table 7; WER except CER for Korean, Japanese and Chinese." }, { "model_id": "gemma-3n-e2b-instructed", "benchmark_id": "fleurs_asr_ru_ru", "score": 0.112, "reference_url": "https://arxiv.org/abs/2607.02770", "reported_setting": { "mode": "unknown", "effort": "n/a", "tools": "audio input; no external tools", "sampling": "provider-reported", "judge": "WER/CER", "harness": "official Google audio evaluation", "prompt_style": "official", "temperature": "not disclosed", "context": "default" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Gemma 4 Technical Report Table 7; WER except CER for Korean, Japanese and Chinese." }, { "model_id": "gemma-3n-e2b-instructed", "benchmark_id": "fleurs_asr_zh_cn_cer", "score": 0.235, "reference_url": "https://arxiv.org/abs/2607.02770", "reported_setting": { "mode": "unknown", "effort": "n/a", "tools": "audio input; no external tools", "sampling": "provider-reported", "judge": "WER/CER", "harness": "official Google audio evaluation", "prompt_style": "official", "temperature": "not disclosed", "context": "default" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Gemma 4 Technical Report Table 7; WER except CER for Korean, Japanese and Chinese." }, { "model_id": "gemma-3n-e4b-instructed", "benchmark_id": "covost2_de_en", "score": 39.1, "reference_url": "https://arxiv.org/abs/2607.02770", "reported_setting": { "mode": "unknown", "effort": "n/a", "tools": "audio input; no external tools", "sampling": "provider-reported", "judge": "CorpusBLEU", "harness": "official Google audio evaluation", "prompt_style": "official", "temperature": "not disclosed", "context": "default" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Gemma 4 Technical Report Table 7, transcribe-then-translate CorpusBLEU." }, { "model_id": "gemma-3n-e4b-instructed", "benchmark_id": "covost2_es_en", "score": 41.8, "reference_url": "https://arxiv.org/abs/2607.02770", "reported_setting": { "mode": "unknown", "effort": "n/a", "tools": "audio input; no external tools", "sampling": "provider-reported", "judge": "CorpusBLEU", "harness": "official Google audio evaluation", "prompt_style": "official", "temperature": "not disclosed", "context": "default" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Gemma 4 Technical Report Table 7, transcribe-then-translate CorpusBLEU." }, { "model_id": "gemma-3n-e4b-instructed", "benchmark_id": "covost2_fr_en", "score": 38.4, "reference_url": "https://arxiv.org/abs/2607.02770", "reported_setting": { "mode": "unknown", "effort": "n/a", "tools": "audio input; no external tools", "sampling": "provider-reported", "judge": "CorpusBLEU", "harness": "official Google audio evaluation", "prompt_style": "official", "temperature": "not disclosed", "context": "default" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Gemma 4 Technical Report Table 7, transcribe-then-translate CorpusBLEU." }, { "model_id": "gemma-3n-e4b-instructed", "benchmark_id": "covost2_it_en", "score": 40.4, "reference_url": "https://arxiv.org/abs/2607.02770", "reported_setting": { "mode": "unknown", "effort": "n/a", "tools": "audio input; no external tools", "sampling": "provider-reported", "judge": "CorpusBLEU", "harness": "official Google audio evaluation", "prompt_style": "official", "temperature": "not disclosed", "context": "default" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Gemma 4 Technical Report Table 7, transcribe-then-translate CorpusBLEU." }, { "model_id": "gemma-3n-e4b-instructed", "benchmark_id": "covost2_ja_en", "score": 22.3, "reference_url": "https://arxiv.org/abs/2607.02770", "reported_setting": { "mode": "unknown", "effort": "n/a", "tools": "audio input; no external tools", "sampling": "provider-reported", "judge": "CorpusBLEU", "harness": "official Google audio evaluation", "prompt_style": "official", "temperature": "not disclosed", "context": "default" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Gemma 4 Technical Report Table 7, transcribe-then-translate CorpusBLEU." }, { "model_id": "gemma-3n-e4b-instructed", "benchmark_id": "covost2_ru_en", "score": 43.7, "reference_url": "https://arxiv.org/abs/2607.02770", "reported_setting": { "mode": "unknown", "effort": "n/a", "tools": "audio input; no external tools", "sampling": "provider-reported", "judge": "CorpusBLEU", "harness": "official Google audio evaluation", "prompt_style": "official", "temperature": "not disclosed", "context": "default" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Gemma 4 Technical Report Table 7, transcribe-then-translate CorpusBLEU." }, { "model_id": "gemma-3n-e4b-instructed", "benchmark_id": "covost2_xx_en_7lang_macro", "score": 34.7, "reference_url": "https://arxiv.org/abs/2607.02770", "reported_setting": { "mode": "unknown", "effort": "n/a", "tools": "audio input; no external tools", "sampling": "provider-reported", "judge": "CorpusBLEU", "harness": "official Google audio evaluation", "prompt_style": "official", "temperature": "not disclosed", "context": "default" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Gemma 4 Technical Report Table 7, transcribe-then-translate CorpusBLEU." }, { "model_id": "gemma-3n-e4b-instructed", "benchmark_id": "covost2_zh_cn_en", "score": 17.4, "reference_url": "https://arxiv.org/abs/2607.02770", "reported_setting": { "mode": "unknown", "effort": "n/a", "tools": "audio input; no external tools", "sampling": "provider-reported", "judge": "CorpusBLEU", "harness": "official Google audio evaluation", "prompt_style": "official", "temperature": "not disclosed", "context": "default" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Gemma 4 Technical Report Table 7, transcribe-then-translate CorpusBLEU." }, { "model_id": "gemma-3n-e4b-instructed", "benchmark_id": "fleurs", "score": 0.085, "reference_url": "https://arxiv.org/abs/2607.02770", "reported_setting": { "mode": "unknown", "effort": "n/a", "tools": "audio input; no external tools", "sampling": "provider-reported", "judge": "WER/CER", "harness": "official Google audio evaluation", "prompt_style": "official", "temperature": "not disclosed", "context": "default" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Gemma 4 Technical Report Table 7; WER except CER for Korean, Japanese and Chinese." }, { "model_id": "gemma-3n-e4b-instructed", "benchmark_id": "fleurs_asr_ar_eg", "score": 0.101, "reference_url": "https://arxiv.org/abs/2607.02770", "reported_setting": { "mode": "unknown", "effort": "n/a", "tools": "audio input; no external tools", "sampling": "provider-reported", "judge": "WER/CER", "harness": "official Google audio evaluation", "prompt_style": "official", "temperature": "not disclosed", "context": "default" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Gemma 4 Technical Report Table 7; WER except CER for Korean, Japanese and Chinese." }, { "model_id": "gemma-3n-e4b-instructed", "benchmark_id": "fleurs_asr_de_de", "score": 0.065, "reference_url": "https://arxiv.org/abs/2607.02770", "reported_setting": { "mode": "unknown", "effort": "n/a", "tools": "audio input; no external tools", "sampling": "provider-reported", "judge": "WER/CER", "harness": "official Google audio evaluation", "prompt_style": "official", "temperature": "not disclosed", "context": "default" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Gemma 4 Technical Report Table 7; WER except CER for Korean, Japanese and Chinese." }, { "model_id": "gemma-3n-e4b-instructed", "benchmark_id": "fleurs_asr_en_us", "score": 0.066, "reference_url": "https://arxiv.org/abs/2607.02770", "reported_setting": { "mode": "unknown", "effort": "n/a", "tools": "audio input; no external tools", "sampling": "provider-reported", "judge": "WER/CER", "harness": "official Google audio evaluation", "prompt_style": "official", "temperature": "not disclosed", "context": "default" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Gemma 4 Technical Report Table 7; WER except CER for Korean, Japanese and Chinese." }, { "model_id": "gemma-3n-e4b-instructed", "benchmark_id": "fleurs_asr_es_419", "score": 0.041, "reference_url": "https://arxiv.org/abs/2607.02770", "reported_setting": { "mode": "unknown", "effort": "n/a", "tools": "audio input; no external tools", "sampling": "provider-reported", "judge": "WER/CER", "harness": "official Google audio evaluation", "prompt_style": "official", "temperature": "not disclosed", "context": "default" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Gemma 4 Technical Report Table 7; WER except CER for Korean, Japanese and Chinese." }, { "model_id": "gemma-3n-e4b-instructed", "benchmark_id": "fleurs_asr_fr_fr", "score": 0.098, "reference_url": "https://arxiv.org/abs/2607.02770", "reported_setting": { "mode": "unknown", "effort": "n/a", "tools": "audio input; no external tools", "sampling": "provider-reported", "judge": "WER/CER", "harness": "official Google audio evaluation", "prompt_style": "official", "temperature": "not disclosed", "context": "default" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Gemma 4 Technical Report Table 7; WER except CER for Korean, Japanese and Chinese." }, { "model_id": "gemma-3n-e4b-instructed", "benchmark_id": "fleurs_asr_hi_in", "score": 0.089, "reference_url": "https://arxiv.org/abs/2607.02770", "reported_setting": { "mode": "unknown", "effort": "n/a", "tools": "audio input; no external tools", "sampling": "provider-reported", "judge": "WER/CER", "harness": "official Google audio evaluation", "prompt_style": "official", "temperature": "not disclosed", "context": "default" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Gemma 4 Technical Report Table 7; WER except CER for Korean, Japanese and Chinese." }, { "model_id": "gemma-3n-e4b-instructed", "benchmark_id": "fleurs_asr_it_it", "score": 0.034, "reference_url": "https://arxiv.org/abs/2607.02770", "reported_setting": { "mode": "unknown", "effort": "n/a", "tools": "audio input; no external tools", "sampling": "provider-reported", "judge": "WER/CER", "harness": "official Google audio evaluation", "prompt_style": "official", "temperature": "not disclosed", "context": "default" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Gemma 4 Technical Report Table 7; WER except CER for Korean, Japanese and Chinese." }, { "model_id": "gemma-3n-e4b-instructed", "benchmark_id": "fleurs_asr_ja_jp_cer", "score": 0.111, "reference_url": "https://arxiv.org/abs/2607.02770", "reported_setting": { "mode": "unknown", "effort": "n/a", "tools": "audio input; no external tools", "sampling": "provider-reported", "judge": "WER/CER", "harness": "official Google audio evaluation", "prompt_style": "official", "temperature": "not disclosed", "context": "default" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Gemma 4 Technical Report Table 7; WER except CER for Korean, Japanese and Chinese." }, { "model_id": "gemma-3n-e4b-instructed", "benchmark_id": "fleurs_asr_ko_kr_cer", "score": 0.073, "reference_url": "https://arxiv.org/abs/2607.02770", "reported_setting": { "mode": "unknown", "effort": "n/a", "tools": "audio input; no external tools", "sampling": "provider-reported", "judge": "WER/CER", "harness": "official Google audio evaluation", "prompt_style": "official", "temperature": "not disclosed", "context": "default" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Gemma 4 Technical Report Table 7; WER except CER for Korean, Japanese and Chinese." }, { "model_id": "gemma-3n-e4b-instructed", "benchmark_id": "fleurs_asr_pt_br", "score": 0.053, "reference_url": "https://arxiv.org/abs/2607.02770", "reported_setting": { "mode": "unknown", "effort": "n/a", "tools": "audio input; no external tools", "sampling": "provider-reported", "judge": "WER/CER", "harness": "official Google audio evaluation", "prompt_style": "official", "temperature": "not disclosed", "context": "default" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Gemma 4 Technical Report Table 7; WER except CER for Korean, Japanese and Chinese." }, { "model_id": "gemma-3n-e4b-instructed", "benchmark_id": "fleurs_asr_ru_ru", "score": 0.087, "reference_url": "https://arxiv.org/abs/2607.02770", "reported_setting": { "mode": "unknown", "effort": "n/a", "tools": "audio input; no external tools", "sampling": "provider-reported", "judge": "WER/CER", "harness": "official Google audio evaluation", "prompt_style": "official", "temperature": "not disclosed", "context": "default" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Gemma 4 Technical Report Table 7; WER except CER for Korean, Japanese and Chinese." }, { "model_id": "gemma-3n-e4b-instructed", "benchmark_id": "fleurs_asr_zh_cn_cer", "score": 0.203, "reference_url": "https://arxiv.org/abs/2607.02770", "reported_setting": { "mode": "unknown", "effort": "n/a", "tools": "audio input; no external tools", "sampling": "provider-reported", "judge": "WER/CER", "harness": "official Google audio evaluation", "prompt_style": "official", "temperature": "not disclosed", "context": "default" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Gemma 4 Technical Report Table 7; WER except CER for Korean, Japanese and Chinese." }, { "model_id": "gemma-4-12b", "benchmark_id": "aime_2026", "score": 77.5, "reference_url": "https://arxiv.org/abs/2607.02770", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "provider-reported", "judge": "benchmark-specified", "harness": "official Google evaluation", "prompt_style": "official", "temperature": "not disclosed", "context": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Gemma 4 Technical Report Table 5; Gemma 4 thinking unless explicitly stated, Gemma 3 27B non-thinking." }, { "model_id": "gemma-4-12b", "benchmark_id": "bigbench_extra_hard", "score": 53.0, "reference_url": "https://blog.google/innovation-and-ai/technology/developers-tools/introducing-gemma-4-12B/", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "provider-reported", "judge": "benchmark-specified", "harness": "official Google evaluation", "prompt_style": "official", "temperature": "not disclosed", "context": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "Google June 3 Gemma 4 12B launch raster; value is printed above the 12B bar in the original 1920x1080 asset." }, { "model_id": "gemma-4-12b", "benchmark_id": "codeforces_rating", "score": 1659, "reference_url": "https://arxiv.org/abs/2607.02770", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "code execution in undisclosed harness", "sampling": "provider-reported", "judge": "benchmark-specified", "harness": "official Google evaluation", "prompt_style": "official", "temperature": "not disclosed", "context": "default" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Gemma 4 Technical Report Table 5; Gemma 4 thinking unless explicitly stated, Gemma 3 27B non-thinking." }, { "model_id": "gemma-4-12b", "benchmark_id": "covost2", "score": 38.5, "reference_url": "https://ai.google.dev/gemma/docs/core/model_card_4", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "provider-reported", "judge": "benchmark-specified", "harness": "official Google evaluation", "prompt_style": "official", "temperature": "not disclosed", "context": "default", "language_scope": "excluding Chinese" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Google model card matrix; all values are instruction-tuned despite the duplicated base-card surface." }, { "model_id": "gemma-4-12b", "benchmark_id": "covost2_de_en", "score": 41.9, "reference_url": "https://arxiv.org/abs/2607.02770", "reported_setting": { "mode": "unknown", "effort": "n/a", "tools": "audio input; no external tools", "sampling": "provider-reported", "judge": "CorpusBLEU", "harness": "official Google audio evaluation", "prompt_style": "official", "temperature": "not disclosed", "context": "default" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Gemma 4 Technical Report Table 8, encoder-free 12B audio evaluation." }, { "model_id": "gemma-4-12b", "benchmark_id": "covost2_es_en", "score": 44.6, "reference_url": "https://arxiv.org/abs/2607.02770", "reported_setting": { "mode": "unknown", "effort": "n/a", "tools": "audio input; no external tools", "sampling": "provider-reported", "judge": "CorpusBLEU", "harness": "official Google audio evaluation", "prompt_style": "official", "temperature": "not disclosed", "context": "default" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Gemma 4 Technical Report Table 8, encoder-free 12B audio evaluation." }, { "model_id": "gemma-4-12b", "benchmark_id": "covost2_fr_en", "score": 42.5, "reference_url": "https://arxiv.org/abs/2607.02770", "reported_setting": { "mode": "unknown", "effort": "n/a", "tools": "audio input; no external tools", "sampling": "provider-reported", "judge": "CorpusBLEU", "harness": "official Google audio evaluation", "prompt_style": "official", "temperature": "not disclosed", "context": "default" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Gemma 4 Technical Report Table 8, encoder-free 12B audio evaluation." }, { "model_id": "gemma-4-12b", "benchmark_id": "covost2_it_en", "score": 43.3, "reference_url": "https://arxiv.org/abs/2607.02770", "reported_setting": { "mode": "unknown", "effort": "n/a", "tools": "audio input; no external tools", "sampling": "provider-reported", "judge": "CorpusBLEU", "harness": "official Google audio evaluation", "prompt_style": "official", "temperature": "not disclosed", "context": "default" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Gemma 4 Technical Report Table 8, encoder-free 12B audio evaluation." }, { "model_id": "gemma-4-12b", "benchmark_id": "covost2_ja_en", "score": 26.4, "reference_url": "https://arxiv.org/abs/2607.02770", "reported_setting": { "mode": "unknown", "effort": "n/a", "tools": "audio input; no external tools", "sampling": "provider-reported", "judge": "CorpusBLEU", "harness": "official Google audio evaluation", "prompt_style": "official", "temperature": "not disclosed", "context": "default" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Gemma 4 Technical Report Table 8, encoder-free 12B audio evaluation." }, { "model_id": "gemma-4-12b", "benchmark_id": "covost2_ru_en", "score": 50.5, "reference_url": "https://arxiv.org/abs/2607.02770", "reported_setting": { "mode": "unknown", "effort": "n/a", "tools": "audio input; no external tools", "sampling": "provider-reported", "judge": "CorpusBLEU", "harness": "official Google audio evaluation", "prompt_style": "official", "temperature": "not disclosed", "context": "default" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Gemma 4 Technical Report Table 8, encoder-free 12B audio evaluation." }, { "model_id": "gemma-4-12b", "benchmark_id": "docvqa", "score": 94.9, "reference_url": "https://blog.google/innovation-and-ai/technology/developers-tools/introducing-gemma-4-12B/", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "provider-reported", "judge": "benchmark-specified", "harness": "official Google evaluation", "prompt_style": "official", "temperature": "not disclosed", "context": "default", "visual_tokens": 1120 }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "Google June 3 Gemma 4 12B launch raster; value is printed above the 12B bar in the original 1920x1080 asset." }, { "model_id": "gemma-4-12b", "benchmark_id": "fleurs", "score": 0.069, "reference_url": "https://ai.google.dev/gemma/docs/core/model_card_4", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "provider-reported", "judge": "benchmark-specified", "harness": "official Google evaluation", "prompt_style": "official", "temperature": "not disclosed", "context": "default", "language_scope": "excluding Chinese" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Google model card matrix; all values are instruction-tuned despite the duplicated base-card surface." }, { "model_id": "gemma-4-12b", "benchmark_id": "fleurs_asr_ar_eg", "score": 0.07, "reference_url": "https://arxiv.org/abs/2607.02770", "reported_setting": { "mode": "unknown", "effort": "n/a", "tools": "audio input; no external tools", "sampling": "provider-reported", "judge": "WER/CER", "harness": "official Google audio evaluation", "prompt_style": "official", "temperature": "not disclosed", "context": "default" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Gemma 4 Technical Report Table 8, encoder-free 12B audio evaluation." }, { "model_id": "gemma-4-12b", "benchmark_id": "fleurs_asr_de_de", "score": 0.053, "reference_url": "https://arxiv.org/abs/2607.02770", "reported_setting": { "mode": "unknown", "effort": "n/a", "tools": "audio input; no external tools", "sampling": "provider-reported", "judge": "WER/CER", "harness": "official Google audio evaluation", "prompt_style": "official", "temperature": "not disclosed", "context": "default" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Gemma 4 Technical Report Table 8, encoder-free 12B audio evaluation." }, { "model_id": "gemma-4-12b", "benchmark_id": "fleurs_asr_en_us", "score": 0.063, "reference_url": "https://arxiv.org/abs/2607.02770", "reported_setting": { "mode": "unknown", "effort": "n/a", "tools": "audio input; no external tools", "sampling": "provider-reported", "judge": "WER/CER", "harness": "official Google audio evaluation", "prompt_style": "official", "temperature": "not disclosed", "context": "default" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Gemma 4 Technical Report Table 8, encoder-free 12B audio evaluation." }, { "model_id": "gemma-4-12b", "benchmark_id": "fleurs_asr_es_419", "score": 0.038, "reference_url": "https://arxiv.org/abs/2607.02770", "reported_setting": { "mode": "unknown", "effort": "n/a", "tools": "audio input; no external tools", "sampling": "provider-reported", "judge": "WER/CER", "harness": "official Google audio evaluation", "prompt_style": "official", "temperature": "not disclosed", "context": "default" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Gemma 4 Technical Report Table 8, encoder-free 12B audio evaluation." }, { "model_id": "gemma-4-12b", "benchmark_id": "fleurs_asr_fr_fr", "score": 0.081, "reference_url": "https://arxiv.org/abs/2607.02770", "reported_setting": { "mode": "unknown", "effort": "n/a", "tools": "audio input; no external tools", "sampling": "provider-reported", "judge": "WER/CER", "harness": "official Google audio evaluation", "prompt_style": "official", "temperature": "not disclosed", "context": "default" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Gemma 4 Technical Report Table 8, encoder-free 12B audio evaluation." }, { "model_id": "gemma-4-12b", "benchmark_id": "fleurs_asr_it_it", "score": 0.03, "reference_url": "https://arxiv.org/abs/2607.02770", "reported_setting": { "mode": "unknown", "effort": "n/a", "tools": "audio input; no external tools", "sampling": "provider-reported", "judge": "WER/CER", "harness": "official Google audio evaluation", "prompt_style": "official", "temperature": "not disclosed", "context": "default" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Gemma 4 Technical Report Table 8, encoder-free 12B audio evaluation." }, { "model_id": "gemma-4-12b", "benchmark_id": "fleurs_asr_ja_jp_cer", "score": 0.08, "reference_url": "https://arxiv.org/abs/2607.02770", "reported_setting": { "mode": "unknown", "effort": "n/a", "tools": "audio input; no external tools", "sampling": "provider-reported", "judge": "WER/CER", "harness": "official Google audio evaluation", "prompt_style": "official", "temperature": "not disclosed", "context": "default" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Gemma 4 Technical Report Table 8, encoder-free 12B audio evaluation." }, { "model_id": "gemma-4-12b", "benchmark_id": "fleurs_asr_ko_kr_cer", "score": 0.057, "reference_url": "https://arxiv.org/abs/2607.02770", "reported_setting": { "mode": "unknown", "effort": "n/a", "tools": "audio input; no external tools", "sampling": "provider-reported", "judge": "WER/CER", "harness": "official Google audio evaluation", "prompt_style": "official", "temperature": "not disclosed", "context": "default" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Gemma 4 Technical Report Table 8, encoder-free 12B audio evaluation." }, { "model_id": "gemma-4-12b", "benchmark_id": "fleurs_asr_pt_br", "score": 0.047, "reference_url": "https://arxiv.org/abs/2607.02770", "reported_setting": { "mode": "unknown", "effort": "n/a", "tools": "audio input; no external tools", "sampling": "provider-reported", "judge": "WER/CER", "harness": "official Google audio evaluation", "prompt_style": "official", "temperature": "not disclosed", "context": "default" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Gemma 4 Technical Report Table 8, encoder-free 12B audio evaluation." }, { "model_id": "gemma-4-12b", "benchmark_id": "fleurs_asr_ru_ru", "score": 0.068, "reference_url": "https://arxiv.org/abs/2607.02770", "reported_setting": { "mode": "unknown", "effort": "n/a", "tools": "audio input; no external tools", "sampling": "provider-reported", "judge": "WER/CER", "harness": "official Google audio evaluation", "prompt_style": "official", "temperature": "not disclosed", "context": "default" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Gemma 4 Technical Report Table 8, encoder-free 12B audio evaluation." }, { "model_id": "gemma-4-12b", "benchmark_id": "gdm_mrcr_v2_8needle_upto_128k", "score": 43.4, "reference_url": "https://blog.google/innovation-and-ai/technology/developers-tools/introducing-gemma-4-12B/", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "provider-reported", "judge": "benchmark-specified", "harness": "official Google evaluation", "prompt_style": "official", "temperature": "not disclosed", "context": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "Google June 3 Gemma 4 12B launch raster; value is printed above the 12B bar in the original 1920x1080 asset." }, { "model_id": "gemma-4-12b", "benchmark_id": "gpqa_diamond", "score": 78.8, "reference_url": "https://blog.google/innovation-and-ai/technology/developers-tools/introducing-gemma-4-12B/", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "provider-reported", "judge": "benchmark-specified", "harness": "official Google evaluation", "prompt_style": "official", "temperature": "not disclosed", "context": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "Google June 3 Gemma 4 12B launch raster; value is printed above the 12B bar in the original 1920x1080 asset." }, { "model_id": "gemma-4-12b", "benchmark_id": "graphwalks_lt128k_combined", "score": 71.0, "reference_url": "https://arxiv.org/abs/2607.02770", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "provider-reported", "judge": "benchmark-specified", "harness": "official Google evaluation", "prompt_style": "official", "temperature": "not disclosed", "context": "below 128K" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Gemma 4 Technical Report Table 9; final IT models evaluated without thinking." }, { "model_id": "gemma-4-12b", "benchmark_id": "hle", "score": 5.2, "reference_url": "https://arxiv.org/abs/2607.02770", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "provider-reported", "judge": "benchmark-specified", "harness": "official Google evaluation", "prompt_style": "official", "temperature": "not disclosed", "context": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Gemma 4 Technical Report Table 5; Gemma 4 thinking unless explicitly stated, Gemma 3 27B non-thinking." }, { "model_id": "gemma-4-12b", "benchmark_id": "ifbench", "score": 74.0, "reference_url": "https://arxiv.org/abs/2607.02770", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "provider-reported", "judge": "benchmark-specified", "harness": "official Google evaluation", "prompt_style": "official", "temperature": "not disclosed", "context": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Gemma 4 Technical Report Table 5; Gemma 4 thinking unless explicitly stated, Gemma 3 27B non-thinking." }, { "model_id": "gemma-4-12b", "benchmark_id": "ifeval", "score": 97.2, "reference_url": "https://arxiv.org/abs/2607.02770", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "provider-reported", "judge": "benchmark-specified", "harness": "official Google evaluation", "prompt_style": "official", "temperature": "not disclosed", "context": "default" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Gemma 4 Technical Report Table 5; Gemma 4 thinking unless explicitly stated, Gemma 3 27B non-thinking." }, { "model_id": "gemma-4-12b", "benchmark_id": "infographicvqa", "score": 88.4, "reference_url": "https://blog.google/innovation-and-ai/technology/developers-tools/introducing-gemma-4-12B/", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "provider-reported", "judge": "benchmark-specified", "harness": "official Google evaluation", "prompt_style": "official", "temperature": "not disclosed", "context": "default", "visual_tokens": 1120 }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "Google June 3 Gemma 4 12B launch raster; value is printed above the 12B bar in the original 1920x1080 asset.", "candidates": [ { "score": 58.7, "reference_url": "https://arxiv.org/abs/2607.02770", "source_type": "tech_report", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "provider-reported", "judge": "benchmark-specified", "harness": "official Google evaluation", "prompt_style": "official", "temperature": "not disclosed", "context": "default", "visual_tokens": 280 }, "notes": "Gemma 4 Technical Report Appendix Table 12; thinking with 280 visual tokens." } ] }, { "model_id": "gemma-4-12b", "benchmark_id": "livecodebench_v6", "score": 72.0, "reference_url": "https://blog.google/innovation-and-ai/technology/developers-tools/introducing-gemma-4-12B/", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "provider-reported", "judge": "benchmark-specified", "harness": "official Google evaluation", "prompt_style": "official", "temperature": "not disclosed", "context": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "Google June 3 Gemma 4 12B launch raster; value is printed above the 12B bar in the original 1920x1080 asset." }, { "model_id": "gemma-4-12b", "benchmark_id": "loft_text_retrieval_128k", "score": 66.4, "reference_url": "https://arxiv.org/abs/2607.02770", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "provider-reported", "judge": "benchmark-specified", "harness": "official Google evaluation", "prompt_style": "official", "temperature": "not disclosed", "context": "128K" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Gemma 4 Technical Report Table 9; final IT models evaluated without thinking." }, { "model_id": "gemma-4-12b", "benchmark_id": "mathvision", "score": 79.7, "reference_url": "https://arxiv.org/abs/2607.02770", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "provider-reported", "judge": "benchmark-specified", "harness": "official Google evaluation", "prompt_style": "official", "temperature": "not disclosed", "context": "default", "visual_tokens": 1120 }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Gemma 4 Technical Report Table 6; maximum supported resolution, 1,120 visual tokens.", "candidates": [ { "score": 76.7, "reference_url": "https://arxiv.org/abs/2607.02770", "source_type": "tech_report", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "provider-reported", "judge": "benchmark-specified", "harness": "official Google evaluation", "prompt_style": "official", "temperature": "not disclosed", "context": "default", "visual_tokens": 280 }, "notes": "Gemma 4 Technical Report Appendix Table 12; thinking with 280 visual tokens." } ] }, { "model_id": "gemma-4-12b", "benchmark_id": "medxpertqa_mm", "score": 48.7, "reference_url": "https://arxiv.org/abs/2607.02770", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "provider-reported", "judge": "benchmark-specified", "harness": "official Google evaluation", "prompt_style": "official", "temperature": "not disclosed", "context": "default", "visual_tokens": 1120 }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Gemma 4 Technical Report Table 6; maximum supported resolution, 1,120 visual tokens.", "candidates": [ { "score": 47.4, "reference_url": "https://arxiv.org/abs/2607.02770", "source_type": "tech_report", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "provider-reported", "judge": "benchmark-specified", "harness": "official Google evaluation", "prompt_style": "official", "temperature": "not disclosed", "context": "default", "visual_tokens": 280 }, "notes": "Gemma 4 Technical Report Appendix Table 12; thinking with 280 visual tokens." } ] }, { "model_id": "gemma-4-12b", "benchmark_id": "mmlu_pro", "score": 77.2, "reference_url": "https://blog.google/innovation-and-ai/technology/developers-tools/introducing-gemma-4-12B/", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "provider-reported", "judge": "benchmark-specified", "harness": "official Google evaluation", "prompt_style": "official", "temperature": "not disclosed", "context": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "Google June 3 Gemma 4 12B launch raster; value is printed above the 12B bar in the original 1920x1080 asset." }, { "model_id": "gemma-4-12b", "benchmark_id": "mmmlu", "score": 83.4, "reference_url": "https://arxiv.org/abs/2607.02770", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "provider-reported", "judge": "benchmark-specified", "harness": "official Google evaluation", "prompt_style": "official", "temperature": "not disclosed", "context": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Gemma 4 Technical Report Table 5; Gemma 4 thinking unless explicitly stated, Gemma 3 27B non-thinking." }, { "model_id": "gemma-4-12b", "benchmark_id": "mmmu_pro", "score": 69.1, "reference_url": "https://blog.google/innovation-and-ai/technology/developers-tools/introducing-gemma-4-12B/", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "provider-reported", "judge": "benchmark-specified", "harness": "official Google evaluation", "prompt_style": "official", "temperature": "not disclosed", "context": "default", "visual_tokens": 1120 }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "Google June 3 Gemma 4 12B launch raster; value is printed above the 12B bar in the original 1920x1080 asset.", "candidates": [ { "score": 67.7, "reference_url": "https://arxiv.org/abs/2607.02770", "source_type": "tech_report", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "provider-reported", "judge": "benchmark-specified", "harness": "official Google evaluation", "prompt_style": "official", "temperature": "not disclosed", "context": "default", "visual_tokens": 280 }, "notes": "Gemma 4 Technical Report Appendix Table 12; thinking with 280 visual tokens." } ] }, { "model_id": "gemma-4-12b", "benchmark_id": "mtob_eng_kgv_full_book", "score": 41.9, "reference_url": "https://arxiv.org/abs/2607.02770", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "provider-reported", "judge": "benchmark-specified", "harness": "official Google evaluation", "prompt_style": "official", "temperature": "not disclosed", "context": "about 256K" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Gemma 4 Technical Report Table 9; final IT models evaluated without thinking." }, { "model_id": "gemma-4-12b", "benchmark_id": "mtob_eng_kgv_half_book", "score": 45.1, "reference_url": "https://arxiv.org/abs/2607.02770", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "provider-reported", "judge": "benchmark-specified", "harness": "official Google evaluation", "prompt_style": "official", "temperature": "not disclosed", "context": "about 128K" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Gemma 4 Technical Report Table 9; final IT models evaluated without thinking." }, { "model_id": "gemma-4-12b", "benchmark_id": "mtob_kgv_eng_full_book", "score": 32.9, "reference_url": "https://arxiv.org/abs/2607.02770", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "provider-reported", "judge": "benchmark-specified", "harness": "official Google evaluation", "prompt_style": "official", "temperature": "not disclosed", "context": "about 256K" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Gemma 4 Technical Report Table 9; final IT models evaluated without thinking." }, { "model_id": "gemma-4-12b", "benchmark_id": "mtob_kgv_eng_half_book", "score": 37.3, "reference_url": "https://arxiv.org/abs/2607.02770", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "provider-reported", "judge": "benchmark-specified", "harness": "official Google evaluation", "prompt_style": "official", "temperature": "not disclosed", "context": "about 128K" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Gemma 4 Technical Report Table 9; final IT models evaluated without thinking." }, { "model_id": "gemma-4-12b", "benchmark_id": "omnidocbench_1.5", "score": 0.164, "reference_url": "https://arxiv.org/abs/2607.02770", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "provider-reported", "judge": "benchmark-specified", "harness": "official Google evaluation", "prompt_style": "official", "temperature": "not disclosed", "context": "default", "visual_tokens": 1120 }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Gemma 4 Technical Report Table 6; maximum supported resolution, 1,120 visual tokens.", "candidates": [ { "score": 0.408, "reference_url": "https://arxiv.org/abs/2607.02770", "source_type": "tech_report", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "provider-reported", "judge": "benchmark-specified", "harness": "official Google evaluation", "prompt_style": "official", "temperature": "not disclosed", "context": "default", "visual_tokens": 280 }, "notes": "Gemma 4 Technical Report Appendix Table 12; thinking with 280 visual tokens." } ] }, { "model_id": "gemma-4-12b", "benchmark_id": "ruler_128k", "score": 91.2, "reference_url": "https://arxiv.org/abs/2607.02770", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "provider-reported", "judge": "benchmark-specified", "harness": "official Google evaluation", "prompt_style": "official", "temperature": "not disclosed", "context": "128K" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Gemma 4 Technical Report Table 9; final IT models evaluated without thinking." }, { "model_id": "gemma-4-12b", "benchmark_id": "ruler_32k", "score": 96.4, "reference_url": "https://arxiv.org/abs/2607.02770", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "provider-reported", "judge": "benchmark-specified", "harness": "official Google evaluation", "prompt_style": "official", "temperature": "not disclosed", "context": "32K" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Gemma 4 Technical Report Table 9; final IT models evaluated without thinking." }, { "model_id": "gemma-4-12b", "benchmark_id": "scicode", "score": 38.0, "reference_url": "https://arxiv.org/abs/2607.02770", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "provider-reported", "judge": "unit tests", "harness": "official Google evaluation", "prompt_style": "official", "temperature": "not disclosed", "context": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Gemma 4 Technical Report Table 5; Gemma 4 thinking unless explicitly stated, Gemma 3 27B non-thinking." }, { "model_id": "gemma-4-12b", "benchmark_id": "tau2_bench_airline", "score": 75.0, "reference_url": "https://arxiv.org/abs/2607.02770", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "benchmark APIs + simulated user", "sampling": "provider-reported", "judge": "state-based task success", "harness": "Tau2-bench", "prompt_style": "official", "temperature": "not disclosed", "context": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Gemma 4 Technical Report Table 5; Gemma 4 thinking unless explicitly stated, Gemma 3 27B non-thinking." }, { "model_id": "gemma-4-12b", "benchmark_id": "tau2_bench_avg", "score": 69.0, "reference_url": "https://ai.google.dev/gemma/docs/core/model_card_4", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "benchmark APIs + simulated user", "sampling": "provider-reported", "judge": "state-based task success", "harness": "Tau2-bench", "prompt_style": "official", "temperature": "not disclosed", "context": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Google model card matrix; all values are instruction-tuned despite the duplicated base-card surface." }, { "model_id": "gemma-4-12b", "benchmark_id": "tau2_bench_retail", "score": 77.6, "reference_url": "https://arxiv.org/abs/2607.02770", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "benchmark APIs + simulated user", "sampling": "provider-reported", "judge": "state-based task success", "harness": "Tau2-bench", "prompt_style": "official", "temperature": "not disclosed", "context": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Gemma 4 Technical Report Table 5; Gemma 4 thinking unless explicitly stated, Gemma 3 27B non-thinking." }, { "model_id": "gemma-4-12b", "benchmark_id": "tau2_bench_telecom", "score": 54.4, "reference_url": "https://arxiv.org/abs/2607.02770", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "benchmark APIs + simulated user", "sampling": "provider-reported", "judge": "state-based task success", "harness": "Tau2-bench", "prompt_style": "official", "temperature": "not disclosed", "context": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Gemma 4 Technical Report Table 5; Gemma 4 thinking unless explicitly stated, Gemma 3 27B non-thinking." }, { "model_id": "gemma-4-12b", "benchmark_id": "terminal_bench_hard", "score": 18.0, "reference_url": "https://arxiv.org/abs/2607.02770", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "agentic terminal", "sampling": "provider-reported", "judge": "benchmark-specified", "harness": "Terminal-Bench Hard provider harness", "prompt_style": "official", "temperature": "not disclosed", "context": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Gemma 4 Technical Report Table 5; Gemma 4 thinking unless explicitly stated, Gemma 3 27B non-thinking." }, { "model_id": "gemma-4-26b-a4b", "benchmark_id": "gdm_mrcr_v2_8needle_upto_128k", "score": 44.1, "reference_url": "https://arxiv.org/abs/2607.02770", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "provider-reported", "judge": "benchmark-specified", "harness": "official Google evaluation", "prompt_style": "official", "temperature": "not disclosed", "context": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Gemma 4 Technical Report Table 5; Gemma 4 thinking unless explicitly stated, Gemma 3 27B non-thinking.", "candidates": [ { "score": 44.1, "reference_url": "https://huggingface.co/google/diffusiongemma-26B-A4B-it", "source_type": "model_card", "reported_setting": { "mode": "thinking", "decoding": "AR MTP", "tools": "none", "sampling": "replicate count N undisclosed", "judge": "official MRCR strict scorer", "harness": "official Google model-card evaluation", "prompt_style": "official provider chat template", "context": "up to 128K" }, "notes": "Official HF model-card matrix; Kaggle is a duplicate physical mirror." } ] }, { "model_id": "gemma-4-26b-a4b", "benchmark_id": "graphwalks_lt128k_combined", "score": 72.6, "reference_url": "https://arxiv.org/abs/2607.02770", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "provider-reported", "judge": "benchmark-specified", "harness": "official Google evaluation", "prompt_style": "official", "temperature": "not disclosed", "context": "below 128K" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Gemma 4 Technical Report Table 9; final IT models evaluated without thinking." }, { "model_id": "gemma-4-26b-a4b", "benchmark_id": "ifbench", "score": 72.0, "reference_url": "https://arxiv.org/abs/2607.02770", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "provider-reported", "judge": "benchmark-specified", "harness": "official Google evaluation", "prompt_style": "official", "temperature": "not disclosed", "context": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Gemma 4 Technical Report Table 5; Gemma 4 thinking unless explicitly stated, Gemma 3 27B non-thinking.", "candidates": [ { "score": 77.74, "reference_url": "https://huggingface.co/internlm/Intern-S2-Preview/resolve/3b53ea8d2f8ed46abefa507e3f6f56ee0c662c4a/figs/performance.png", "source_type": "official_model_card", "reported_setting": { "table_framework_claim": "OpenCompass and VLMEvalKit used to evaluate all models", "exact_per_cell_settings": "not stated" }, "notes": "Displayed exactly: 77.74" }, { "score": 47.25, "reference_url": "https://www.liquid.ai/blog/lfm2-5-8b-a1b", "source_type": "official_blog", "reported_setting": { "mode": "source/model default", "effort": "source does not state", "tools": "none", "sampling": "source does not state trial count", "judge": "benchmark-specified", "harness": "Liquid AI official comparative evaluation", "prompt_style": "source does not state", "temperature": "source does not state evaluation parameters", "context": "source does not state evaluation context" }, "notes": "Displayed exactly as 47.25. Official Liquid AI reported result." } ] }, { "model_id": "gemma-4-26b-a4b", "benchmark_id": "ifeval", "score": 98.7, "reference_url": "https://arxiv.org/abs/2608.00146", "reported_setting": { "mode": "thinking", "decoding": "AR MTP", "tools": "none", "sampling": "replicate count N undisclosed", "judge": "IFEval strict/loose subtype undisclosed", "harness": "official Google cross-model evaluation", "prompt_style": "official provider chat template", "context": "default" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "DiffusionGemma Technical Report Table 3; replicate count N is undisclosed.", "candidates": [ { "score": 98.5, "reference_url": "https://arxiv.org/abs/2607.02770", "source_type": "tech_report", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "provider-reported", "judge": "benchmark-specified", "harness": "official Google evaluation", "prompt_style": "official", "temperature": "not disclosed", "context": "default" }, "notes": "Gemma 4 Technical Report Table 5; Gemma 4 thinking unless explicitly stated, Gemma 3 27B non-thinking." }, { "score": 97.8, "reference_url": "https://arxiv.org/abs/2608.00146", "source_type": "tech_report", "reported_setting": { "mode": "non-thinking", "decoding": "AR MTP", "tools": "none", "sampling": "replicate count N undisclosed", "judge": "IFEval strict/loose subtype undisclosed", "harness": "official Google cross-model evaluation", "prompt_style": "official provider chat template", "context": "default" }, "notes": "DiffusionGemma Technical Report Table 3; replicate count N is undisclosed." }, { "score": 91.4, "reference_url": "https://www.liquid.ai/blog/lfm2-5-8b-a1b", "source_type": "official_blog", "reported_setting": { "mode": "source/model default", "effort": "source does not state", "tools": "none", "sampling": "source does not state trial count", "judge": "benchmark-specified", "harness": "Liquid AI official comparative evaluation", "prompt_style": "source does not state", "temperature": "source does not state evaluation parameters", "context": "source does not state evaluation context" }, "notes": "Displayed exactly as 91.40. Official Liquid AI reported result." } ] }, { "model_id": "gemma-4-26b-a4b", "benchmark_id": "infographicvqa", "score": 89.3, "reference_url": "https://arxiv.org/abs/2607.02770", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "provider-reported", "judge": "benchmark-specified", "harness": "official Google evaluation", "prompt_style": "official", "temperature": "not disclosed", "context": "default", "visual_tokens": 1120 }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Gemma 4 Technical Report Table 6; maximum supported resolution, 1,120 visual tokens.", "candidates": [ { "score": 77.8, "reference_url": "https://arxiv.org/abs/2607.02770", "source_type": "tech_report", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "provider-reported", "judge": "benchmark-specified", "harness": "official Google evaluation", "prompt_style": "official", "temperature": "not disclosed", "context": "default", "visual_tokens": 280 }, "notes": "Gemma 4 Technical Report Appendix Table 12; thinking with 280 visual tokens." } ] }, { "model_id": "gemma-4-26b-a4b", "benchmark_id": "livecodebench_v6", "score": 77.1, "reference_url": "https://blog.google/innovation-and-ai/technology/developers-tools/gemma-4/", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "provider-reported", "judge": "benchmark-specified", "harness": "official Google evaluation", "prompt_style": "official", "temperature": "not disclosed", "context": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "Google Gemma 4 family launch table, duplicated by the DeepMind semantic table.", "candidates": [ { "score": 72.6, "reference_url": "https://arxiv.org/abs/2608.00146", "source_type": "tech_report", "reported_setting": { "mode": "non-thinking", "decoding": "AR MTP", "tools": "none", "sampling": "replicate count N undisclosed", "judge": "benchmark-specified", "harness": "official Google cross-model evaluation", "prompt_style": "official provider chat template", "context": "default" }, "notes": "DiffusionGemma Technical Report Table 3; replicate count N is undisclosed." } ] }, { "model_id": "gemma-4-26b-a4b", "benchmark_id": "loft_text_retrieval_128k", "score": 66.3, "reference_url": "https://arxiv.org/abs/2607.02770", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "provider-reported", "judge": "benchmark-specified", "harness": "official Google evaluation", "prompt_style": "official", "temperature": "not disclosed", "context": "128K" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Gemma 4 Technical Report Table 9; final IT models evaluated without thinking." }, { "model_id": "gemma-4-26b-a4b", "benchmark_id": "mtob_eng_kgv_full_book", "score": 48.9, "reference_url": "https://arxiv.org/abs/2607.02770", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "provider-reported", "judge": "benchmark-specified", "harness": "official Google evaluation", "prompt_style": "official", "temperature": "not disclosed", "context": "about 256K" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Gemma 4 Technical Report Table 9; final IT models evaluated without thinking." }, { "model_id": "gemma-4-26b-a4b", "benchmark_id": "mtob_eng_kgv_half_book", "score": 50.0, "reference_url": "https://arxiv.org/abs/2607.02770", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "provider-reported", "judge": "benchmark-specified", "harness": "official Google evaluation", "prompt_style": "official", "temperature": "not disclosed", "context": "about 128K" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Gemma 4 Technical Report Table 9; final IT models evaluated without thinking." }, { "model_id": "gemma-4-26b-a4b", "benchmark_id": "mtob_kgv_eng_full_book", "score": 42.7, "reference_url": "https://arxiv.org/abs/2607.02770", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "provider-reported", "judge": "benchmark-specified", "harness": "official Google evaluation", "prompt_style": "official", "temperature": "not disclosed", "context": "about 256K" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Gemma 4 Technical Report Table 9; final IT models evaluated without thinking." }, { "model_id": "gemma-4-26b-a4b", "benchmark_id": "mtob_kgv_eng_half_book", "score": 45.0, "reference_url": "https://arxiv.org/abs/2607.02770", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "provider-reported", "judge": "benchmark-specified", "harness": "official Google evaluation", "prompt_style": "official", "temperature": "not disclosed", "context": "about 128K" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Gemma 4 Technical Report Table 9; final IT models evaluated without thinking." }, { "model_id": "gemma-4-26b-a4b", "benchmark_id": "omnidocbench_1.5", "score": 0.149, "reference_url": "https://arxiv.org/abs/2607.02770", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "provider-reported", "judge": "benchmark-specified", "harness": "official Google evaluation", "prompt_style": "official", "temperature": "not disclosed", "context": "default", "visual_tokens": 1120 }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Gemma 4 Technical Report Table 6; maximum supported resolution, 1,120 visual tokens.", "candidates": [ { "score": 0.269, "reference_url": "https://arxiv.org/abs/2607.02770", "source_type": "tech_report", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "provider-reported", "judge": "benchmark-specified", "harness": "official Google evaluation", "prompt_style": "official", "temperature": "not disclosed", "context": "default", "visual_tokens": 280 }, "notes": "Gemma 4 Technical Report Appendix Table 12; thinking with 280 visual tokens." } ] }, { "model_id": "gemma-4-26b-a4b", "benchmark_id": "ruler_128k", "score": 89.8, "reference_url": "https://arxiv.org/abs/2607.02770", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "provider-reported", "judge": "benchmark-specified", "harness": "official Google evaluation", "prompt_style": "official", "temperature": "not disclosed", "context": "128K" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Gemma 4 Technical Report Table 9; final IT models evaluated without thinking." }, { "model_id": "gemma-4-26b-a4b", "benchmark_id": "ruler_32k", "score": 97.3, "reference_url": "https://arxiv.org/abs/2607.02770", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "provider-reported", "judge": "benchmark-specified", "harness": "official Google evaluation", "prompt_style": "official", "temperature": "not disclosed", "context": "32K" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Gemma 4 Technical Report Table 9; final IT models evaluated without thinking." }, { "model_id": "gemma-4-26b-a4b", "benchmark_id": "scicode", "score": 40.0, "reference_url": "https://arxiv.org/abs/2607.02770", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "provider-reported", "judge": "unit tests", "harness": "official Google evaluation", "prompt_style": "official", "temperature": "not disclosed", "context": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Gemma 4 Technical Report Table 5; Gemma 4 thinking unless explicitly stated, Gemma 3 27B non-thinking.", "candidates": [ { "score": 42.6, "reference_url": "https://huggingface.co/internlm/Intern-S2-Preview/resolve/3b53ea8d2f8ed46abefa507e3f6f56ee0c662c4a/figs/performance.png", "source_type": "official_model_card", "reported_setting": { "table_framework_claim": "OpenCompass and VLMEvalKit used to evaluate all models", "exact_per_cell_settings": "not stated" }, "notes": "Displayed exactly: 42.60" } ] }, { "model_id": "gemma-4-26b-a4b", "benchmark_id": "tau2_bench_airline", "score": 76.0, "reference_url": "https://arxiv.org/abs/2607.02770", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "benchmark APIs + simulated user", "sampling": "provider-reported", "judge": "state-based task success", "harness": "Tau2-bench", "prompt_style": "official", "temperature": "not disclosed", "context": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Gemma 4 Technical Report Table 5; Gemma 4 thinking unless explicitly stated, Gemma 3 27B non-thinking.", "candidates": [ { "score": 51.0, "reference_url": "https://arxiv.org/abs/2608.00146", "source_type": "tech_report", "reported_setting": { "mode": "non-thinking", "decoding": "AR MTP", "tools": "benchmark APIs + simulated user", "sampling": "trials=4 per task inferred from score totals; N otherwise undisclosed", "judge": "state-based task success", "harness": "official Google cross-model evaluation", "prompt_style": "official provider chat template", "context": "default" }, "notes": "DiffusionGemma Technical Report Table 3; replicate count N is undisclosed." } ] }, { "model_id": "gemma-4-26b-a4b", "benchmark_id": "tau2_bench_avg", "score": 68.2, "reference_url": "https://ai.google.dev/gemma/docs/core/model_card_4", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "benchmark APIs + simulated user", "sampling": "provider-reported", "judge": "state-based task success", "harness": "Tau2-bench", "prompt_style": "official", "temperature": "not disclosed", "context": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Google model card matrix; all values are instruction-tuned despite the duplicated base-card surface." }, { "model_id": "gemma-4-26b-a4b", "benchmark_id": "tau2_bench_retail", "score": 85.5, "reference_url": "https://blog.google/innovation-and-ai/technology/developers-tools/gemma-4/", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "benchmark APIs + simulated user", "sampling": "provider-reported", "judge": "state-based task success", "harness": "Tau2-bench", "prompt_style": "official", "temperature": "not disclosed", "context": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "Google Gemma 4 family launch table, duplicated by the DeepMind semantic table.", "candidates": [ { "score": 79.0, "reference_url": "https://arxiv.org/abs/2608.00146", "source_type": "tech_report", "reported_setting": { "mode": "non-thinking", "decoding": "AR MTP", "tools": "benchmark APIs + simulated user", "sampling": "trials=4 per task inferred from score totals; N otherwise undisclosed", "judge": "state-based task success", "harness": "official Google cross-model evaluation", "prompt_style": "official provider chat template", "context": "default" }, "notes": "DiffusionGemma Technical Report Table 3; replicate count N is undisclosed." }, { "score": 55.26, "reference_url": "https://www.liquid.ai/blog/lfm2-5-8b-a1b", "source_type": "official_blog", "reported_setting": { "mode": "source/model default", "effort": "source does not state", "tools": "benchmark-provided tools", "sampling": "source does not state trial count", "judge": "benchmark-specified", "harness": "Liquid AI official comparative evaluation", "prompt_style": "source does not state", "temperature": "source does not state evaluation parameters", "context": "source does not state evaluation context" }, "notes": "Displayed exactly as 55.26. Official Liquid AI reported result." } ] }, { "model_id": "gemma-4-26b-a4b", "benchmark_id": "tau2_bench_telecom", "score": 43.0, "reference_url": "https://arxiv.org/abs/2607.02770", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "benchmark APIs + simulated user", "sampling": "provider-reported", "judge": "state-based task success", "harness": "Tau2-bench", "prompt_style": "official", "temperature": "not disclosed", "context": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Gemma 4 Technical Report Table 5; Gemma 4 thinking unless explicitly stated, Gemma 3 27B non-thinking.", "candidates": [ { "score": 34.2, "reference_url": "https://arxiv.org/abs/2608.00146", "source_type": "tech_report", "reported_setting": { "mode": "non-thinking", "decoding": "AR MTP", "tools": "benchmark APIs + simulated user", "sampling": "trials=4 per task inferred from score totals; N otherwise undisclosed", "judge": "state-based task success", "harness": "official Google cross-model evaluation", "prompt_style": "official provider chat template", "context": "default" }, "notes": "DiffusionGemma Technical Report Table 3; replicate count N is undisclosed." }, { "score": 42.11, "reference_url": "https://www.liquid.ai/blog/lfm2-5-8b-a1b", "source_type": "official_blog", "reported_setting": { "mode": "source/model default", "effort": "source does not state", "tools": "benchmark-provided tools", "sampling": "source does not state trial count", "judge": "benchmark-specified", "harness": "Liquid AI official comparative evaluation", "prompt_style": "source does not state", "temperature": "source does not state evaluation parameters", "context": "source does not state evaluation context" }, "notes": "Displayed exactly as 42.11. Official Liquid AI reported result." } ] }, { "model_id": "gemma-4-26b-a4b", "benchmark_id": "terminal_bench_hard", "score": 14.0, "reference_url": "https://arxiv.org/abs/2607.02770", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "agentic terminal", "sampling": "provider-reported", "judge": "benchmark-specified", "harness": "Terminal-Bench Hard provider harness", "prompt_style": "official", "temperature": "not disclosed", "context": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Gemma 4 Technical Report Table 5; Gemma 4 thinking unless explicitly stated, Gemma 3 27B non-thinking." }, { "model_id": "gemma-4-26b-a4b", "benchmark_id": "text_arena_elo", "score": 1441, "reference_url": "https://blog.google/innovation-and-ai/technology/developers-tools/gemma-4/", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "live human pairwise votes", "judge": "blind human raters", "harness": "Arena AI text", "prompt_style": "live user prompts", "temperature": "unknown", "context": "default", "snapshot_date": "2026-04-02" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Google Gemma 4 family launch table, duplicated by the DeepMind semantic table.", "candidates": [ { "score": 1442, "reference_url": "https://deepmind.google/models/gemma/gemma-4/", "source_type": "official_blog", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "none", "sampling": "live human pairwise votes", "judge": "blind human raters", "harness": "Arena Text", "prompt_style": "live user prompts", "temperature": "unknown", "context": "default", "snapshot_date": "2026-04-02" }, "notes": "DeepMind page accessibility text provides the exact plotted Elo value." }, { "score": 1438, "reference_url": "https://arxiv.org/abs/2607.02770", "source_type": "tech_report", "reported_setting": { "mode": "unknown", "effort": "unknown", "tools": "none", "sampling": "live human pairwise votes", "judge": "blind human raters", "harness": "Arena Text", "prompt_style": "live user prompts", "temperature": "unknown", "context": "default", "snapshot_date": "2026-06-19" }, "notes": "Gemma 4 Technical Report Table 4 dated Arena Text Elo snapshot." } ] }, { "model_id": "gemma-4-31b", "benchmark_id": "gdm_mrcr_v2_8needle_upto_128k", "score": 66.4, "reference_url": "https://arxiv.org/abs/2607.02770", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "provider-reported", "judge": "benchmark-specified", "harness": "official Google evaluation", "prompt_style": "official", "temperature": "not disclosed", "context": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Gemma 4 Technical Report Table 5; Gemma 4 thinking unless explicitly stated, Gemma 3 27B non-thinking." }, { "model_id": "gemma-4-31b", "benchmark_id": "graphwalks_lt128k_combined", "score": 82.3, "reference_url": "https://arxiv.org/abs/2607.02770", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "provider-reported", "judge": "benchmark-specified", "harness": "official Google evaluation", "prompt_style": "official", "temperature": "not disclosed", "context": "below 128K" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Gemma 4 Technical Report Table 9; final IT models evaluated without thinking." }, { "model_id": "gemma-4-31b", "benchmark_id": "ifbench", "score": 76.0, "reference_url": "https://arxiv.org/abs/2607.02770", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "provider-reported", "judge": "benchmark-specified", "harness": "official Google evaluation", "prompt_style": "official", "temperature": "not disclosed", "context": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Gemma 4 Technical Report Table 5; Gemma 4 thinking unless explicitly stated, Gemma 3 27B non-thinking." }, { "model_id": "gemma-4-31b", "benchmark_id": "ifeval", "score": 98.9, "reference_url": "https://arxiv.org/abs/2607.02770", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "provider-reported", "judge": "benchmark-specified", "harness": "official Google evaluation", "prompt_style": "official", "temperature": "not disclosed", "context": "default" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Gemma 4 Technical Report Table 5; Gemma 4 thinking unless explicitly stated, Gemma 3 27B non-thinking." }, { "model_id": "gemma-4-31b", "benchmark_id": "infographicvqa", "score": 92.0, "reference_url": "https://arxiv.org/abs/2607.02770", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "provider-reported", "judge": "benchmark-specified", "harness": "official Google evaluation", "prompt_style": "official", "temperature": "not disclosed", "context": "default", "visual_tokens": 1120 }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Gemma 4 Technical Report Table 6; maximum supported resolution, 1,120 visual tokens.", "candidates": [ { "score": 82.8, "reference_url": "https://arxiv.org/abs/2607.02770", "source_type": "tech_report", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "provider-reported", "judge": "benchmark-specified", "harness": "official Google evaluation", "prompt_style": "official", "temperature": "not disclosed", "context": "default", "visual_tokens": 280 }, "notes": "Gemma 4 Technical Report Appendix Table 12; thinking with 280 visual tokens." } ] }, { "model_id": "gemma-4-31b", "benchmark_id": "livecodebench_v6", "score": 80.0, "reference_url": "https://blog.google/innovation-and-ai/technology/developers-tools/gemma-4/", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "provider-reported", "judge": "benchmark-specified", "harness": "official Google evaluation", "prompt_style": "official", "temperature": "not disclosed", "context": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "Google Gemma 4 family launch table, duplicated by the DeepMind semantic table." }, { "model_id": "gemma-4-31b", "benchmark_id": "loft_text_retrieval_128k", "score": 79.5, "reference_url": "https://arxiv.org/abs/2607.02770", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "provider-reported", "judge": "benchmark-specified", "harness": "official Google evaluation", "prompt_style": "official", "temperature": "not disclosed", "context": "128K" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Gemma 4 Technical Report Table 9; final IT models evaluated without thinking." }, { "model_id": "gemma-4-31b", "benchmark_id": "mtob_eng_kgv_full_book", "score": 54.3, "reference_url": "https://arxiv.org/abs/2607.02770", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "provider-reported", "judge": "benchmark-specified", "harness": "official Google evaluation", "prompt_style": "official", "temperature": "not disclosed", "context": "about 256K" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Gemma 4 Technical Report Table 9; final IT models evaluated without thinking." }, { "model_id": "gemma-4-31b", "benchmark_id": "mtob_eng_kgv_half_book", "score": 52.9, "reference_url": "https://arxiv.org/abs/2607.02770", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "provider-reported", "judge": "benchmark-specified", "harness": "official Google evaluation", "prompt_style": "official", "temperature": "not disclosed", "context": "about 128K" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Gemma 4 Technical Report Table 9; final IT models evaluated without thinking." }, { "model_id": "gemma-4-31b", "benchmark_id": "mtob_kgv_eng_full_book", "score": 46.2, "reference_url": "https://arxiv.org/abs/2607.02770", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "provider-reported", "judge": "benchmark-specified", "harness": "official Google evaluation", "prompt_style": "official", "temperature": "not disclosed", "context": "about 256K" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Gemma 4 Technical Report Table 9; final IT models evaluated without thinking." }, { "model_id": "gemma-4-31b", "benchmark_id": "mtob_kgv_eng_half_book", "score": 48.6, "reference_url": "https://arxiv.org/abs/2607.02770", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "provider-reported", "judge": "benchmark-specified", "harness": "official Google evaluation", "prompt_style": "official", "temperature": "not disclosed", "context": "about 128K" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Gemma 4 Technical Report Table 9; final IT models evaluated without thinking." }, { "model_id": "gemma-4-31b", "benchmark_id": "omnidocbench_1.5", "score": 0.131, "reference_url": "https://arxiv.org/abs/2607.02770", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "provider-reported", "judge": "benchmark-specified", "harness": "official Google evaluation", "prompt_style": "official", "temperature": "not disclosed", "context": "default", "visual_tokens": 1120 }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Gemma 4 Technical Report Table 6; maximum supported resolution, 1,120 visual tokens.", "candidates": [ { "score": 0.201, "reference_url": "https://arxiv.org/abs/2607.02770", "source_type": "tech_report", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "provider-reported", "judge": "benchmark-specified", "harness": "official Google evaluation", "prompt_style": "official", "temperature": "not disclosed", "context": "default", "visual_tokens": 280 }, "notes": "Gemma 4 Technical Report Appendix Table 12; thinking with 280 visual tokens." } ] }, { "model_id": "gemma-4-31b", "benchmark_id": "ruler_128k", "score": 96.4, "reference_url": "https://arxiv.org/abs/2607.02770", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "provider-reported", "judge": "benchmark-specified", "harness": "official Google evaluation", "prompt_style": "official", "temperature": "not disclosed", "context": "128K" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Gemma 4 Technical Report Table 9; final IT models evaluated without thinking." }, { "model_id": "gemma-4-31b", "benchmark_id": "ruler_32k", "score": 96.8, "reference_url": "https://arxiv.org/abs/2607.02770", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "provider-reported", "judge": "benchmark-specified", "harness": "official Google evaluation", "prompt_style": "official", "temperature": "not disclosed", "context": "32K" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Gemma 4 Technical Report Table 9; final IT models evaluated without thinking." }, { "model_id": "gemma-4-31b", "benchmark_id": "scicode", "score": 43.0, "reference_url": "https://arxiv.org/abs/2607.02770", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "provider-reported", "judge": "unit tests", "harness": "official Google evaluation", "prompt_style": "official", "temperature": "not disclosed", "context": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Gemma 4 Technical Report Table 5; Gemma 4 thinking unless explicitly stated, Gemma 3 27B non-thinking." }, { "model_id": "gemma-4-31b", "benchmark_id": "tau2_bench_airline", "score": 75.0, "reference_url": "https://arxiv.org/abs/2607.02770", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "benchmark APIs + simulated user", "sampling": "provider-reported", "judge": "state-based task success", "harness": "Tau2-bench", "prompt_style": "official", "temperature": "not disclosed", "context": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Gemma 4 Technical Report Table 5; Gemma 4 thinking unless explicitly stated, Gemma 3 27B non-thinking." }, { "model_id": "gemma-4-31b", "benchmark_id": "tau2_bench_avg", "score": 76.9, "reference_url": "https://ai.google.dev/gemma/docs/core/model_card_4", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "benchmark APIs + simulated user", "sampling": "provider-reported", "judge": "state-based task success", "harness": "Tau2-bench", "prompt_style": "official", "temperature": "not disclosed", "context": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Google model card matrix; all values are instruction-tuned despite the duplicated base-card surface." }, { "model_id": "gemma-4-31b", "benchmark_id": "tau2_bench_retail", "score": 86.4, "reference_url": "https://blog.google/innovation-and-ai/technology/developers-tools/gemma-4/", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "benchmark APIs + simulated user", "sampling": "provider-reported", "judge": "state-based task success", "harness": "Tau2-bench", "prompt_style": "official", "temperature": "not disclosed", "context": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "Google Gemma 4 family launch table, duplicated by the DeepMind semantic table." }, { "model_id": "gemma-4-31b", "benchmark_id": "tau2_bench_telecom", "score": 69.3, "reference_url": "https://arxiv.org/abs/2607.02770", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "benchmark APIs + simulated user", "sampling": "provider-reported", "judge": "state-based task success", "harness": "Tau2-bench", "prompt_style": "official", "temperature": "not disclosed", "context": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Gemma 4 Technical Report Table 5; Gemma 4 thinking unless explicitly stated, Gemma 3 27B non-thinking." }, { "model_id": "gemma-4-31b", "benchmark_id": "terminal_bench_hard", "score": 36.0, "reference_url": "https://arxiv.org/abs/2607.02770", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "agentic terminal", "sampling": "provider-reported", "judge": "benchmark-specified", "harness": "Terminal-Bench Hard provider harness", "prompt_style": "official", "temperature": "not disclosed", "context": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Gemma 4 Technical Report Table 5; Gemma 4 thinking unless explicitly stated, Gemma 3 27B non-thinking." }, { "model_id": "gemma-4-31b", "benchmark_id": "text_arena_elo", "score": 1452, "reference_url": "https://blog.google/innovation-and-ai/technology/developers-tools/gemma-4/", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "live human pairwise votes", "judge": "blind human raters", "harness": "Arena AI text", "prompt_style": "live user prompts", "temperature": "unknown", "context": "default", "snapshot_date": "2026-04-02" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Google Gemma 4 family launch table, duplicated by the DeepMind semantic table.", "candidates": [ { "score": 1451, "reference_url": "https://deepmind.google/models/gemma/gemma-4/", "source_type": "official_blog", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "none", "sampling": "live human pairwise votes", "judge": "blind human raters", "harness": "Arena Text", "prompt_style": "live user prompts", "temperature": "unknown", "context": "default", "snapshot_date": "2026-04-02" }, "notes": "DeepMind page accessibility text provides the exact plotted Elo value." }, { "score": 1451, "reference_url": "https://arxiv.org/abs/2607.02770", "source_type": "tech_report", "reported_setting": { "mode": "unknown", "effort": "unknown", "tools": "none", "sampling": "live human pairwise votes", "judge": "blind human raters", "harness": "Arena Text", "prompt_style": "live user prompts", "temperature": "unknown", "context": "default", "snapshot_date": "2026-06-19" }, "notes": "Gemma 4 Technical Report Table 4 dated Arena Text Elo snapshot." } ] }, { "model_id": "gemma-4-e2b", "benchmark_id": "covost2", "score": 33.47, "reference_url": "https://ai.google.dev/gemma/docs/core/model_card_4", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "provider-reported", "judge": "benchmark-specified", "harness": "official Google evaluation", "prompt_style": "official", "temperature": "not disclosed", "context": "default", "language_scope": "provider-reported aggregate" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Google model card matrix; all values are instruction-tuned despite the duplicated base-card surface." }, { "model_id": "gemma-4-e2b", "benchmark_id": "covost2_de_en", "score": 39.2, "reference_url": "https://arxiv.org/abs/2607.02770", "reported_setting": { "mode": "unknown", "effort": "n/a", "tools": "audio input; no external tools", "sampling": "provider-reported", "judge": "CorpusBLEU", "harness": "official Google audio evaluation", "prompt_style": "official", "temperature": "not disclosed", "context": "default" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Gemma 4 Technical Report Table 7, transcribe-then-translate CorpusBLEU." }, { "model_id": "gemma-4-e2b", "benchmark_id": "covost2_es_en", "score": 43.2, "reference_url": "https://arxiv.org/abs/2607.02770", "reported_setting": { "mode": "unknown", "effort": "n/a", "tools": "audio input; no external tools", "sampling": "provider-reported", "judge": "CorpusBLEU", "harness": "official Google audio evaluation", "prompt_style": "official", "temperature": "not disclosed", "context": "default" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Gemma 4 Technical Report Table 7, transcribe-then-translate CorpusBLEU." }, { "model_id": "gemma-4-e2b", "benchmark_id": "covost2_fr_en", "score": 39.2, "reference_url": "https://arxiv.org/abs/2607.02770", "reported_setting": { "mode": "unknown", "effort": "n/a", "tools": "audio input; no external tools", "sampling": "provider-reported", "judge": "CorpusBLEU", "harness": "official Google audio evaluation", "prompt_style": "official", "temperature": "not disclosed", "context": "default" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Gemma 4 Technical Report Table 7, transcribe-then-translate CorpusBLEU." }, { "model_id": "gemma-4-e2b", "benchmark_id": "covost2_it_en", "score": 40.8, "reference_url": "https://arxiv.org/abs/2607.02770", "reported_setting": { "mode": "unknown", "effort": "n/a", "tools": "audio input; no external tools", "sampling": "provider-reported", "judge": "CorpusBLEU", "harness": "official Google audio evaluation", "prompt_style": "official", "temperature": "not disclosed", "context": "default" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Gemma 4 Technical Report Table 7, transcribe-then-translate CorpusBLEU." }, { "model_id": "gemma-4-e2b", "benchmark_id": "covost2_ja_en", "score": 21.4, "reference_url": "https://arxiv.org/abs/2607.02770", "reported_setting": { "mode": "unknown", "effort": "n/a", "tools": "audio input; no external tools", "sampling": "provider-reported", "judge": "CorpusBLEU", "harness": "official Google audio evaluation", "prompt_style": "official", "temperature": "not disclosed", "context": "default" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Gemma 4 Technical Report Table 7, transcribe-then-translate CorpusBLEU." }, { "model_id": "gemma-4-e2b", "benchmark_id": "covost2_ru_en", "score": 46.4, "reference_url": "https://arxiv.org/abs/2607.02770", "reported_setting": { "mode": "unknown", "effort": "n/a", "tools": "audio input; no external tools", "sampling": "provider-reported", "judge": "CorpusBLEU", "harness": "official Google audio evaluation", "prompt_style": "official", "temperature": "not disclosed", "context": "default" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Gemma 4 Technical Report Table 7, transcribe-then-translate CorpusBLEU." }, { "model_id": "gemma-4-e2b", "benchmark_id": "covost2_xx_en_7lang_macro", "score": 35.4, "reference_url": "https://arxiv.org/abs/2607.02770", "reported_setting": { "mode": "unknown", "effort": "n/a", "tools": "audio input; no external tools", "sampling": "provider-reported", "judge": "CorpusBLEU", "harness": "official Google audio evaluation", "prompt_style": "official", "temperature": "not disclosed", "context": "default" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Gemma 4 Technical Report Table 7, transcribe-then-translate CorpusBLEU." }, { "model_id": "gemma-4-e2b", "benchmark_id": "covost2_zh_cn_en", "score": 17.9, "reference_url": "https://arxiv.org/abs/2607.02770", "reported_setting": { "mode": "unknown", "effort": "n/a", "tools": "audio input; no external tools", "sampling": "provider-reported", "judge": "CorpusBLEU", "harness": "official Google audio evaluation", "prompt_style": "official", "temperature": "not disclosed", "context": "default" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Gemma 4 Technical Report Table 7, transcribe-then-translate CorpusBLEU." }, { "model_id": "gemma-4-e2b", "benchmark_id": "fleurs_asr_ar_eg", "score": 0.143, "reference_url": "https://arxiv.org/abs/2607.02770", "reported_setting": { "mode": "unknown", "effort": "n/a", "tools": "audio input; no external tools", "sampling": "provider-reported", "judge": "WER/CER", "harness": "official Google audio evaluation", "prompt_style": "official", "temperature": "not disclosed", "context": "default" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Gemma 4 Technical Report Table 7; WER except CER for Korean, Japanese and Chinese." }, { "model_id": "gemma-4-e2b", "benchmark_id": "fleurs_asr_de_de", "score": 0.076, "reference_url": "https://arxiv.org/abs/2607.02770", "reported_setting": { "mode": "unknown", "effort": "n/a", "tools": "audio input; no external tools", "sampling": "provider-reported", "judge": "WER/CER", "harness": "official Google audio evaluation", "prompt_style": "official", "temperature": "not disclosed", "context": "default" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Gemma 4 Technical Report Table 7; WER except CER for Korean, Japanese and Chinese." }, { "model_id": "gemma-4-e2b", "benchmark_id": "fleurs_asr_en_us", "score": 0.08, "reference_url": "https://arxiv.org/abs/2607.02770", "reported_setting": { "mode": "unknown", "effort": "n/a", "tools": "audio input; no external tools", "sampling": "provider-reported", "judge": "WER/CER", "harness": "official Google audio evaluation", "prompt_style": "official", "temperature": "not disclosed", "context": "default" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Gemma 4 Technical Report Table 7; WER except CER for Korean, Japanese and Chinese." }, { "model_id": "gemma-4-e2b", "benchmark_id": "fleurs_asr_es_419", "score": 0.042, "reference_url": "https://arxiv.org/abs/2607.02770", "reported_setting": { "mode": "unknown", "effort": "n/a", "tools": "audio input; no external tools", "sampling": "provider-reported", "judge": "WER/CER", "harness": "official Google audio evaluation", "prompt_style": "official", "temperature": "not disclosed", "context": "default" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Gemma 4 Technical Report Table 7; WER except CER for Korean, Japanese and Chinese." }, { "model_id": "gemma-4-e2b", "benchmark_id": "fleurs_asr_fr_fr", "score": 0.101, "reference_url": "https://arxiv.org/abs/2607.02770", "reported_setting": { "mode": "unknown", "effort": "n/a", "tools": "audio input; no external tools", "sampling": "provider-reported", "judge": "WER/CER", "harness": "official Google audio evaluation", "prompt_style": "official", "temperature": "not disclosed", "context": "default" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Gemma 4 Technical Report Table 7; WER except CER for Korean, Japanese and Chinese." }, { "model_id": "gemma-4-e2b", "benchmark_id": "fleurs_asr_hi_in", "score": 0.101, "reference_url": "https://arxiv.org/abs/2607.02770", "reported_setting": { "mode": "unknown", "effort": "n/a", "tools": "audio input; no external tools", "sampling": "provider-reported", "judge": "WER/CER", "harness": "official Google audio evaluation", "prompt_style": "official", "temperature": "not disclosed", "context": "default" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Gemma 4 Technical Report Table 7; WER except CER for Korean, Japanese and Chinese." }, { "model_id": "gemma-4-e2b", "benchmark_id": "fleurs_asr_it_it", "score": 0.041, "reference_url": "https://arxiv.org/abs/2607.02770", "reported_setting": { "mode": "unknown", "effort": "n/a", "tools": "audio input; no external tools", "sampling": "provider-reported", "judge": "WER/CER", "harness": "official Google audio evaluation", "prompt_style": "official", "temperature": "not disclosed", "context": "default" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Gemma 4 Technical Report Table 7; WER except CER for Korean, Japanese and Chinese." }, { "model_id": "gemma-4-e2b", "benchmark_id": "fleurs_asr_ja_jp_cer", "score": 0.107, "reference_url": "https://arxiv.org/abs/2607.02770", "reported_setting": { "mode": "unknown", "effort": "n/a", "tools": "audio input; no external tools", "sampling": "provider-reported", "judge": "WER/CER", "harness": "official Google audio evaluation", "prompt_style": "official", "temperature": "not disclosed", "context": "default" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Gemma 4 Technical Report Table 7; WER except CER for Korean, Japanese and Chinese." }, { "model_id": "gemma-4-e2b", "benchmark_id": "fleurs_asr_ko_kr_cer", "score": 0.066, "reference_url": "https://arxiv.org/abs/2607.02770", "reported_setting": { "mode": "unknown", "effort": "n/a", "tools": "audio input; no external tools", "sampling": "provider-reported", "judge": "WER/CER", "harness": "official Google audio evaluation", "prompt_style": "official", "temperature": "not disclosed", "context": "default" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Gemma 4 Technical Report Table 7; WER except CER for Korean, Japanese and Chinese." }, { "model_id": "gemma-4-e2b", "benchmark_id": "fleurs_asr_pt_br", "score": 0.056, "reference_url": "https://arxiv.org/abs/2607.02770", "reported_setting": { "mode": "unknown", "effort": "n/a", "tools": "audio input; no external tools", "sampling": "provider-reported", "judge": "WER/CER", "harness": "official Google audio evaluation", "prompt_style": "official", "temperature": "not disclosed", "context": "default" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Gemma 4 Technical Report Table 7; WER except CER for Korean, Japanese and Chinese." }, { "model_id": "gemma-4-e2b", "benchmark_id": "fleurs_asr_ru_ru", "score": 0.084, "reference_url": "https://arxiv.org/abs/2607.02770", "reported_setting": { "mode": "unknown", "effort": "n/a", "tools": "audio input; no external tools", "sampling": "provider-reported", "judge": "WER/CER", "harness": "official Google audio evaluation", "prompt_style": "official", "temperature": "not disclosed", "context": "default" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Gemma 4 Technical Report Table 7; WER except CER for Korean, Japanese and Chinese." }, { "model_id": "gemma-4-e2b", "benchmark_id": "fleurs_asr_zh_cn_cer", "score": 0.187, "reference_url": "https://arxiv.org/abs/2607.02770", "reported_setting": { "mode": "unknown", "effort": "n/a", "tools": "audio input; no external tools", "sampling": "provider-reported", "judge": "WER/CER", "harness": "official Google audio evaluation", "prompt_style": "official", "temperature": "not disclosed", "context": "default" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Gemma 4 Technical Report Table 7; WER except CER for Korean, Japanese and Chinese." }, { "model_id": "gemma-4-e2b", "benchmark_id": "gdm_mrcr_v2_8needle_upto_128k", "score": 19.1, "reference_url": "https://arxiv.org/abs/2607.02770", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "provider-reported", "judge": "benchmark-specified", "harness": "official Google evaluation", "prompt_style": "official", "temperature": "not disclosed", "context": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Gemma 4 Technical Report Table 5; Gemma 4 thinking unless explicitly stated, Gemma 3 27B non-thinking." }, { "model_id": "gemma-4-e2b", "benchmark_id": "graphwalks_lt128k_combined", "score": 4.1, "reference_url": "https://arxiv.org/abs/2607.02770", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "provider-reported", "judge": "benchmark-specified", "harness": "official Google evaluation", "prompt_style": "official", "temperature": "not disclosed", "context": "below 128K" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Gemma 4 Technical Report Table 9; final IT models evaluated without thinking." }, { "model_id": "gemma-4-e2b", "benchmark_id": "ifbench", "score": 38.0, "reference_url": "https://arxiv.org/abs/2607.02770", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "provider-reported", "judge": "benchmark-specified", "harness": "official Google evaluation", "prompt_style": "official", "temperature": "not disclosed", "context": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Gemma 4 Technical Report Table 5; Gemma 4 thinking unless explicitly stated, Gemma 3 27B non-thinking.", "candidates": [ { "score": 33.53, "reference_url": "https://www.liquid.ai/blog/lfm2-5-8b-a1b", "source_type": "official_blog", "reported_setting": { "mode": "source/model default", "effort": "source does not state", "tools": "none", "sampling": "source does not state trial count", "judge": "benchmark-specified", "harness": "Liquid AI official comparative evaluation", "prompt_style": "source does not state", "temperature": "source does not state evaluation parameters", "context": "source does not state evaluation context" }, "notes": "Displayed exactly as 33.53. Official Liquid AI reported result." }, { "score": 34.08, "reference_url": "https://www.liquid.ai/blog/lfm2-5-2-6b", "source_type": "official_blog", "reported_setting": { "mode": "source/model default", "effort": "source does not state", "tools": "none", "sampling": "source does not state trial count", "judge": "benchmark-specified", "harness": "Liquid AI official comparative evaluation", "prompt_style": "source does not state", "temperature": "source does not state evaluation parameters", "context": "source does not state evaluation context" }, "notes": "Displayed exactly as 34.08. Official Liquid AI reported result." } ] }, { "model_id": "gemma-4-e2b", "benchmark_id": "ifeval", "score": 94.6, "reference_url": "https://arxiv.org/abs/2607.02770", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "provider-reported", "judge": "benchmark-specified", "harness": "official Google evaluation", "prompt_style": "official", "temperature": "not disclosed", "context": "default" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Gemma 4 Technical Report Table 5; Gemma 4 thinking unless explicitly stated, Gemma 3 27B non-thinking.", "candidates": [ { "score": 82.93, "reference_url": "https://www.liquid.ai/blog/lfm2-5-8b-a1b", "source_type": "official_blog", "reported_setting": { "mode": "source/model default", "effort": "source does not state", "tools": "none", "sampling": "source does not state trial count", "judge": "benchmark-specified", "harness": "Liquid AI official comparative evaluation", "prompt_style": "source does not state", "temperature": "source does not state evaluation parameters", "context": "source does not state evaluation context" }, "notes": "Displayed exactly as 82.93. Official Liquid AI reported result." } ] }, { "model_id": "gemma-4-e2b", "benchmark_id": "infographicvqa", "score": 63.9, "reference_url": "https://arxiv.org/abs/2607.02770", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "provider-reported", "judge": "benchmark-specified", "harness": "official Google evaluation", "prompt_style": "official", "temperature": "not disclosed", "context": "default", "visual_tokens": 1120 }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Gemma 4 Technical Report Table 6; maximum supported resolution, 1,120 visual tokens.", "candidates": [ { "score": 44.6, "reference_url": "https://arxiv.org/abs/2607.02770", "source_type": "tech_report", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "provider-reported", "judge": "benchmark-specified", "harness": "official Google evaluation", "prompt_style": "official", "temperature": "not disclosed", "context": "default", "visual_tokens": 280 }, "notes": "Gemma 4 Technical Report Appendix Table 12; thinking with 280 visual tokens." } ] }, { "model_id": "gemma-4-e2b", "benchmark_id": "livecodebench_v6", "score": 44.0, "reference_url": "https://blog.google/innovation-and-ai/technology/developers-tools/gemma-4/", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "provider-reported", "judge": "benchmark-specified", "harness": "official Google evaluation", "prompt_style": "official", "temperature": "not disclosed", "context": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "Google Gemma 4 family launch table, duplicated by the DeepMind semantic table.", "candidates": [ { "score": 54.92, "reference_url": "https://www.liquid.ai/blog/lfm2-5-2-6b", "source_type": "official_blog", "reported_setting": { "mode": "source/model default", "effort": "source does not state", "tools": "none", "sampling": "source does not state trial count", "judge": "benchmark-specified", "harness": "Liquid AI official comparative evaluation", "prompt_style": "source does not state", "temperature": "source does not state evaluation parameters", "context": "source does not state evaluation context" }, "notes": "Displayed exactly as 54.92. Official Liquid AI reported result." } ] }, { "model_id": "gemma-4-e2b", "benchmark_id": "loft_text_retrieval_128k", "score": 50.5, "reference_url": "https://arxiv.org/abs/2607.02770", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "provider-reported", "judge": "benchmark-specified", "harness": "official Google evaluation", "prompt_style": "official", "temperature": "not disclosed", "context": "128K" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Gemma 4 Technical Report Table 9; final IT models evaluated without thinking." }, { "model_id": "gemma-4-e2b", "benchmark_id": "mtob_eng_kgv_half_book", "score": 15.4, "reference_url": "https://arxiv.org/abs/2607.02770", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "provider-reported", "judge": "benchmark-specified", "harness": "official Google evaluation", "prompt_style": "official", "temperature": "not disclosed", "context": "about 128K" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Gemma 4 Technical Report Table 9; final IT models evaluated without thinking." }, { "model_id": "gemma-4-e2b", "benchmark_id": "mtob_kgv_eng_half_book", "score": 28.2, "reference_url": "https://arxiv.org/abs/2607.02770", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "provider-reported", "judge": "benchmark-specified", "harness": "official Google evaluation", "prompt_style": "official", "temperature": "not disclosed", "context": "about 128K" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Gemma 4 Technical Report Table 9; final IT models evaluated without thinking." }, { "model_id": "gemma-4-e2b", "benchmark_id": "omnidocbench_1.5", "score": 0.29, "reference_url": "https://arxiv.org/abs/2607.02770", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "provider-reported", "judge": "benchmark-specified", "harness": "official Google evaluation", "prompt_style": "official", "temperature": "not disclosed", "context": "default", "visual_tokens": 1120 }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Gemma 4 Technical Report Table 6; maximum supported resolution, 1,120 visual tokens.", "candidates": [ { "score": 0.496, "reference_url": "https://arxiv.org/abs/2607.02770", "source_type": "tech_report", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "provider-reported", "judge": "benchmark-specified", "harness": "official Google evaluation", "prompt_style": "official", "temperature": "not disclosed", "context": "default", "visual_tokens": 280 }, "notes": "Gemma 4 Technical Report Appendix Table 12; thinking with 280 visual tokens." } ] }, { "model_id": "gemma-4-e2b", "benchmark_id": "ruler_128k", "score": 70.4, "reference_url": "https://arxiv.org/abs/2607.02770", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "provider-reported", "judge": "benchmark-specified", "harness": "official Google evaluation", "prompt_style": "official", "temperature": "not disclosed", "context": "128K" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Gemma 4 Technical Report Table 9; final IT models evaluated without thinking." }, { "model_id": "gemma-4-e2b", "benchmark_id": "ruler_32k", "score": 83.0, "reference_url": "https://arxiv.org/abs/2607.02770", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "provider-reported", "judge": "benchmark-specified", "harness": "official Google evaluation", "prompt_style": "official", "temperature": "not disclosed", "context": "32K" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Gemma 4 Technical Report Table 9; final IT models evaluated without thinking." }, { "model_id": "gemma-4-e2b", "benchmark_id": "scicode", "score": 21.0, "reference_url": "https://arxiv.org/abs/2607.02770", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "provider-reported", "judge": "unit tests", "harness": "official Google evaluation", "prompt_style": "official", "temperature": "not disclosed", "context": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Gemma 4 Technical Report Table 5; Gemma 4 thinking unless explicitly stated, Gemma 3 27B non-thinking." }, { "model_id": "gemma-4-e2b", "benchmark_id": "tau2_bench_airline", "score": 31.0, "reference_url": "https://arxiv.org/abs/2607.02770", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "benchmark APIs + simulated user", "sampling": "provider-reported", "judge": "state-based task success", "harness": "Tau2-bench", "prompt_style": "official", "temperature": "not disclosed", "context": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Gemma 4 Technical Report Table 5; Gemma 4 thinking unless explicitly stated, Gemma 3 27B non-thinking." }, { "model_id": "gemma-4-e2b", "benchmark_id": "tau2_bench_avg", "score": 24.5, "reference_url": "https://ai.google.dev/gemma/docs/core/model_card_4", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "benchmark APIs + simulated user", "sampling": "provider-reported", "judge": "state-based task success", "harness": "Tau2-bench", "prompt_style": "official", "temperature": "not disclosed", "context": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Google model card matrix; all values are instruction-tuned despite the duplicated base-card surface." }, { "model_id": "gemma-4-e2b", "benchmark_id": "tau2_bench_retail", "score": 29.4, "reference_url": "https://blog.google/innovation-and-ai/technology/developers-tools/gemma-4/", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "benchmark APIs + simulated user", "sampling": "provider-reported", "judge": "state-based task success", "harness": "Tau2-bench", "prompt_style": "official", "temperature": "not disclosed", "context": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "Google Gemma 4 family launch table, duplicated by the DeepMind semantic table.", "candidates": [ { "score": 34.6, "reference_url": "https://arxiv.org/abs/2607.02770", "source_type": "tech_report", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "benchmark APIs + simulated user", "sampling": "provider-reported", "judge": "state-based task success", "harness": "Tau2-bench", "prompt_style": "official", "temperature": "not disclosed", "context": "default" }, "notes": "Gemma 4 Technical Report Table 5; Gemma 4 thinking unless explicitly stated, Gemma 3 27B non-thinking." }, { "score": 18.95, "reference_url": "https://www.liquid.ai/blog/lfm2-5-8b-a1b", "source_type": "official_blog", "reported_setting": { "mode": "source/model default", "effort": "source does not state", "tools": "benchmark-provided tools", "sampling": "source does not state trial count", "judge": "benchmark-specified", "harness": "Liquid AI official comparative evaluation", "prompt_style": "source does not state", "temperature": "source does not state evaluation parameters", "context": "source does not state evaluation context" }, "notes": "Displayed exactly as 18.95. Official Liquid AI reported result." } ] }, { "model_id": "gemma-4-e2b", "benchmark_id": "tau2_bench_telecom", "score": 19.7, "reference_url": "https://arxiv.org/abs/2607.02770", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "benchmark APIs + simulated user", "sampling": "provider-reported", "judge": "state-based task success", "harness": "Tau2-bench", "prompt_style": "official", "temperature": "not disclosed", "context": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Gemma 4 Technical Report Table 5; Gemma 4 thinking unless explicitly stated, Gemma 3 27B non-thinking.", "candidates": [ { "score": 22.37, "reference_url": "https://www.liquid.ai/blog/lfm2-5-8b-a1b", "source_type": "official_blog", "reported_setting": { "mode": "source/model default", "effort": "source does not state", "tools": "benchmark-provided tools", "sampling": "source does not state trial count", "judge": "benchmark-specified", "harness": "Liquid AI official comparative evaluation", "prompt_style": "source does not state", "temperature": "source does not state evaluation parameters", "context": "source does not state evaluation context" }, "notes": "Displayed exactly as 22.37. Official Liquid AI reported result." } ] }, { "model_id": "gemma-4-e2b", "benchmark_id": "terminal_bench_hard", "score": 3.0, "reference_url": "https://arxiv.org/abs/2607.02770", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "agentic terminal", "sampling": "provider-reported", "judge": "benchmark-specified", "harness": "Terminal-Bench Hard provider harness", "prompt_style": "official", "temperature": "not disclosed", "context": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Gemma 4 Technical Report Table 5; Gemma 4 thinking unless explicitly stated, Gemma 3 27B non-thinking." }, { "model_id": "gemma-4-e4b", "benchmark_id": "covost2", "score": 35.54, "reference_url": "https://ai.google.dev/gemma/docs/core/model_card_4", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "provider-reported", "judge": "benchmark-specified", "harness": "official Google evaluation", "prompt_style": "official", "temperature": "not disclosed", "context": "default", "language_scope": "provider-reported aggregate" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Google model card matrix; all values are instruction-tuned despite the duplicated base-card surface." }, { "model_id": "gemma-4-e4b", "benchmark_id": "covost2_de_en", "score": 42.0, "reference_url": "https://arxiv.org/abs/2607.02770", "reported_setting": { "mode": "unknown", "effort": "n/a", "tools": "audio input; no external tools", "sampling": "provider-reported", "judge": "CorpusBLEU", "harness": "official Google audio evaluation", "prompt_style": "official", "temperature": "not disclosed", "context": "default" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Gemma 4 Technical Report Table 7, transcribe-then-translate CorpusBLEU." }, { "model_id": "gemma-4-e4b", "benchmark_id": "covost2_es_en", "score": 44.8, "reference_url": "https://arxiv.org/abs/2607.02770", "reported_setting": { "mode": "unknown", "effort": "n/a", "tools": "audio input; no external tools", "sampling": "provider-reported", "judge": "CorpusBLEU", "harness": "official Google audio evaluation", "prompt_style": "official", "temperature": "not disclosed", "context": "default" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Gemma 4 Technical Report Table 7, transcribe-then-translate CorpusBLEU." }, { "model_id": "gemma-4-e4b", "benchmark_id": "covost2_fr_en", "score": 41.0, "reference_url": "https://arxiv.org/abs/2607.02770", "reported_setting": { "mode": "unknown", "effort": "n/a", "tools": "audio input; no external tools", "sampling": "provider-reported", "judge": "CorpusBLEU", "harness": "official Google audio evaluation", "prompt_style": "official", "temperature": "not disclosed", "context": "default" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Gemma 4 Technical Report Table 7, transcribe-then-translate CorpusBLEU." }, { "model_id": "gemma-4-e4b", "benchmark_id": "covost2_it_en", "score": 43.0, "reference_url": "https://arxiv.org/abs/2607.02770", "reported_setting": { "mode": "unknown", "effort": "n/a", "tools": "audio input; no external tools", "sampling": "provider-reported", "judge": "CorpusBLEU", "harness": "official Google audio evaluation", "prompt_style": "official", "temperature": "not disclosed", "context": "default" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Gemma 4 Technical Report Table 7, transcribe-then-translate CorpusBLEU." }, { "model_id": "gemma-4-e4b", "benchmark_id": "covost2_ja_en", "score": 25.5, "reference_url": "https://arxiv.org/abs/2607.02770", "reported_setting": { "mode": "unknown", "effort": "n/a", "tools": "audio input; no external tools", "sampling": "provider-reported", "judge": "CorpusBLEU", "harness": "official Google audio evaluation", "prompt_style": "official", "temperature": "not disclosed", "context": "default" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Gemma 4 Technical Report Table 7, transcribe-then-translate CorpusBLEU." }, { "model_id": "gemma-4-e4b", "benchmark_id": "covost2_ru_en", "score": 49.4, "reference_url": "https://arxiv.org/abs/2607.02770", "reported_setting": { "mode": "unknown", "effort": "n/a", "tools": "audio input; no external tools", "sampling": "provider-reported", "judge": "CorpusBLEU", "harness": "official Google audio evaluation", "prompt_style": "official", "temperature": "not disclosed", "context": "default" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Gemma 4 Technical Report Table 7, transcribe-then-translate CorpusBLEU." }, { "model_id": "gemma-4-e4b", "benchmark_id": "covost2_xx_en_7lang_macro", "score": 38.2, "reference_url": "https://arxiv.org/abs/2607.02770", "reported_setting": { "mode": "unknown", "effort": "n/a", "tools": "audio input; no external tools", "sampling": "provider-reported", "judge": "CorpusBLEU", "harness": "official Google audio evaluation", "prompt_style": "official", "temperature": "not disclosed", "context": "default" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Gemma 4 Technical Report Table 7, transcribe-then-translate CorpusBLEU." }, { "model_id": "gemma-4-e4b", "benchmark_id": "covost2_zh_cn_en", "score": 21.9, "reference_url": "https://arxiv.org/abs/2607.02770", "reported_setting": { "mode": "unknown", "effort": "n/a", "tools": "audio input; no external tools", "sampling": "provider-reported", "judge": "CorpusBLEU", "harness": "official Google audio evaluation", "prompt_style": "official", "temperature": "not disclosed", "context": "default" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Gemma 4 Technical Report Table 7, transcribe-then-translate CorpusBLEU." }, { "model_id": "gemma-4-e4b", "benchmark_id": "fleurs_asr_ar_eg", "score": 0.162, "reference_url": "https://arxiv.org/abs/2607.02770", "reported_setting": { "mode": "unknown", "effort": "n/a", "tools": "audio input; no external tools", "sampling": "provider-reported", "judge": "WER/CER", "harness": "official Google audio evaluation", "prompt_style": "official", "temperature": "not disclosed", "context": "default" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Gemma 4 Technical Report Table 7; WER except CER for Korean, Japanese and Chinese." }, { "model_id": "gemma-4-e4b", "benchmark_id": "fleurs_asr_de_de", "score": 0.061, "reference_url": "https://arxiv.org/abs/2607.02770", "reported_setting": { "mode": "unknown", "effort": "n/a", "tools": "audio input; no external tools", "sampling": "provider-reported", "judge": "WER/CER", "harness": "official Google audio evaluation", "prompt_style": "official", "temperature": "not disclosed", "context": "default" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Gemma 4 Technical Report Table 7; WER except CER for Korean, Japanese and Chinese." }, { "model_id": "gemma-4-e4b", "benchmark_id": "fleurs_asr_en_us", "score": 0.065, "reference_url": "https://arxiv.org/abs/2607.02770", "reported_setting": { "mode": "unknown", "effort": "n/a", "tools": "audio input; no external tools", "sampling": "provider-reported", "judge": "WER/CER", "harness": "official Google audio evaluation", "prompt_style": "official", "temperature": "not disclosed", "context": "default" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Gemma 4 Technical Report Table 7; WER except CER for Korean, Japanese and Chinese." }, { "model_id": "gemma-4-e4b", "benchmark_id": "fleurs_asr_es_419", "score": 0.035, "reference_url": "https://arxiv.org/abs/2607.02770", "reported_setting": { "mode": "unknown", "effort": "n/a", "tools": "audio input; no external tools", "sampling": "provider-reported", "judge": "WER/CER", "harness": "official Google audio evaluation", "prompt_style": "official", "temperature": "not disclosed", "context": "default" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Gemma 4 Technical Report Table 7; WER except CER for Korean, Japanese and Chinese." }, { "model_id": "gemma-4-e4b", "benchmark_id": "fleurs_asr_fr_fr", "score": 0.08, "reference_url": "https://arxiv.org/abs/2607.02770", "reported_setting": { "mode": "unknown", "effort": "n/a", "tools": "audio input; no external tools", "sampling": "provider-reported", "judge": "WER/CER", "harness": "official Google audio evaluation", "prompt_style": "official", "temperature": "not disclosed", "context": "default" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Gemma 4 Technical Report Table 7; WER except CER for Korean, Japanese and Chinese." }, { "model_id": "gemma-4-e4b", "benchmark_id": "fleurs_asr_hi_in", "score": 0.086, "reference_url": "https://arxiv.org/abs/2607.02770", "reported_setting": { "mode": "unknown", "effort": "n/a", "tools": "audio input; no external tools", "sampling": "provider-reported", "judge": "WER/CER", "harness": "official Google audio evaluation", "prompt_style": "official", "temperature": "not disclosed", "context": "default" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Gemma 4 Technical Report Table 7; WER except CER for Korean, Japanese and Chinese." }, { "model_id": "gemma-4-e4b", "benchmark_id": "fleurs_asr_it_it", "score": 0.032, "reference_url": "https://arxiv.org/abs/2607.02770", "reported_setting": { "mode": "unknown", "effort": "n/a", "tools": "audio input; no external tools", "sampling": "provider-reported", "judge": "WER/CER", "harness": "official Google audio evaluation", "prompt_style": "official", "temperature": "not disclosed", "context": "default" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Gemma 4 Technical Report Table 7; WER except CER for Korean, Japanese and Chinese." }, { "model_id": "gemma-4-e4b", "benchmark_id": "fleurs_asr_ja_jp_cer", "score": 0.078, "reference_url": "https://arxiv.org/abs/2607.02770", "reported_setting": { "mode": "unknown", "effort": "n/a", "tools": "audio input; no external tools", "sampling": "provider-reported", "judge": "WER/CER", "harness": "official Google audio evaluation", "prompt_style": "official", "temperature": "not disclosed", "context": "default" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Gemma 4 Technical Report Table 7; WER except CER for Korean, Japanese and Chinese." }, { "model_id": "gemma-4-e4b", "benchmark_id": "fleurs_asr_ko_kr_cer", "score": 0.053, "reference_url": "https://arxiv.org/abs/2607.02770", "reported_setting": { "mode": "unknown", "effort": "n/a", "tools": "audio input; no external tools", "sampling": "provider-reported", "judge": "WER/CER", "harness": "official Google audio evaluation", "prompt_style": "official", "temperature": "not disclosed", "context": "default" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Gemma 4 Technical Report Table 7; WER except CER for Korean, Japanese and Chinese." }, { "model_id": "gemma-4-e4b", "benchmark_id": "fleurs_asr_pt_br", "score": 0.046, "reference_url": "https://arxiv.org/abs/2607.02770", "reported_setting": { "mode": "unknown", "effort": "n/a", "tools": "audio input; no external tools", "sampling": "provider-reported", "judge": "WER/CER", "harness": "official Google audio evaluation", "prompt_style": "official", "temperature": "not disclosed", "context": "default" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Gemma 4 Technical Report Table 7; WER except CER for Korean, Japanese and Chinese." }, { "model_id": "gemma-4-e4b", "benchmark_id": "fleurs_asr_ru_ru", "score": 0.068, "reference_url": "https://arxiv.org/abs/2607.02770", "reported_setting": { "mode": "unknown", "effort": "n/a", "tools": "audio input; no external tools", "sampling": "provider-reported", "judge": "WER/CER", "harness": "official Google audio evaluation", "prompt_style": "official", "temperature": "not disclosed", "context": "default" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Gemma 4 Technical Report Table 7; WER except CER for Korean, Japanese and Chinese." }, { "model_id": "gemma-4-e4b", "benchmark_id": "fleurs_asr_zh_cn_cer", "score": 0.136, "reference_url": "https://arxiv.org/abs/2607.02770", "reported_setting": { "mode": "unknown", "effort": "n/a", "tools": "audio input; no external tools", "sampling": "provider-reported", "judge": "WER/CER", "harness": "official Google audio evaluation", "prompt_style": "official", "temperature": "not disclosed", "context": "default" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Gemma 4 Technical Report Table 7; WER except CER for Korean, Japanese and Chinese." }, { "model_id": "gemma-4-e4b", "benchmark_id": "gdm_mrcr_v2_8needle_upto_128k", "score": 25.4, "reference_url": "https://arxiv.org/abs/2607.02770", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "provider-reported", "judge": "benchmark-specified", "harness": "official Google evaluation", "prompt_style": "official", "temperature": "not disclosed", "context": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Gemma 4 Technical Report Table 5; Gemma 4 thinking unless explicitly stated, Gemma 3 27B non-thinking." }, { "model_id": "gemma-4-e4b", "benchmark_id": "graphwalks_lt128k_combined", "score": 50.9, "reference_url": "https://arxiv.org/abs/2607.02770", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "provider-reported", "judge": "benchmark-specified", "harness": "official Google evaluation", "prompt_style": "official", "temperature": "not disclosed", "context": "below 128K" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Gemma 4 Technical Report Table 9; final IT models evaluated without thinking." }, { "model_id": "gemma-4-e4b", "benchmark_id": "ifbench", "score": 44.0, "reference_url": "https://arxiv.org/abs/2607.02770", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "provider-reported", "judge": "benchmark-specified", "harness": "official Google evaluation", "prompt_style": "official", "temperature": "not disclosed", "context": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Gemma 4 Technical Report Table 5; Gemma 4 thinking unless explicitly stated, Gemma 3 27B non-thinking.", "candidates": [ { "score": 39.48, "reference_url": "https://www.liquid.ai/blog/lfm2-5-8b-a1b", "source_type": "official_blog", "reported_setting": { "mode": "source/model default", "effort": "source does not state", "tools": "none", "sampling": "source does not state trial count", "judge": "benchmark-specified", "harness": "Liquid AI official comparative evaluation", "prompt_style": "source does not state", "temperature": "source does not state evaluation parameters", "context": "source does not state evaluation context" }, "notes": "Displayed exactly as 39.48. Official Liquid AI reported result." }, { "score": 39.24, "reference_url": "https://www.liquid.ai/blog/lfm2-5-2-6b", "source_type": "official_blog", "reported_setting": { "mode": "source/model default", "effort": "source does not state", "tools": "none", "sampling": "source does not state trial count", "judge": "benchmark-specified", "harness": "Liquid AI official comparative evaluation", "prompt_style": "source does not state", "temperature": "source does not state evaluation parameters", "context": "source does not state evaluation context" }, "notes": "Displayed exactly as 39.24. Official Liquid AI reported result." } ] }, { "model_id": "gemma-4-e4b", "benchmark_id": "ifeval", "score": 96.7, "reference_url": "https://arxiv.org/abs/2607.02770", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "provider-reported", "judge": "benchmark-specified", "harness": "official Google evaluation", "prompt_style": "official", "temperature": "not disclosed", "context": "default" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Gemma 4 Technical Report Table 5; Gemma 4 thinking unless explicitly stated, Gemma 3 27B non-thinking.", "candidates": [ { "score": 87.74, "reference_url": "https://www.liquid.ai/blog/lfm2-5-8b-a1b", "source_type": "official_blog", "reported_setting": { "mode": "source/model default", "effort": "source does not state", "tools": "none", "sampling": "source does not state trial count", "judge": "benchmark-specified", "harness": "Liquid AI official comparative evaluation", "prompt_style": "source does not state", "temperature": "source does not state evaluation parameters", "context": "source does not state evaluation context" }, "notes": "Displayed exactly as 87.74. Official Liquid AI reported result." } ] }, { "model_id": "gemma-4-e4b", "benchmark_id": "infographicvqa", "score": 70.0, "reference_url": "https://arxiv.org/abs/2607.02770", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "provider-reported", "judge": "benchmark-specified", "harness": "official Google evaluation", "prompt_style": "official", "temperature": "not disclosed", "context": "default", "visual_tokens": 1120 }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Gemma 4 Technical Report Table 6; maximum supported resolution, 1,120 visual tokens.", "candidates": [ { "score": 54.8, "reference_url": "https://arxiv.org/abs/2607.02770", "source_type": "tech_report", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "provider-reported", "judge": "benchmark-specified", "harness": "official Google evaluation", "prompt_style": "official", "temperature": "not disclosed", "context": "default", "visual_tokens": 280 }, "notes": "Gemma 4 Technical Report Appendix Table 12; thinking with 280 visual tokens." } ] }, { "model_id": "gemma-4-e4b", "benchmark_id": "livecodebench_v6", "score": 52.0, "reference_url": "https://blog.google/innovation-and-ai/technology/developers-tools/gemma-4/", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "provider-reported", "judge": "benchmark-specified", "harness": "official Google evaluation", "prompt_style": "official", "temperature": "not disclosed", "context": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "Google Gemma 4 family launch table, duplicated by the DeepMind semantic table.", "candidates": [ { "score": 63.77, "reference_url": "https://www.liquid.ai/blog/lfm2-5-2-6b", "source_type": "official_blog", "reported_setting": { "mode": "source/model default", "effort": "source does not state", "tools": "none", "sampling": "source does not state trial count", "judge": "benchmark-specified", "harness": "Liquid AI official comparative evaluation", "prompt_style": "source does not state", "temperature": "source does not state evaluation parameters", "context": "source does not state evaluation context" }, "notes": "Displayed exactly as 63.77. Official Liquid AI reported result." } ] }, { "model_id": "gemma-4-e4b", "benchmark_id": "loft_text_retrieval_128k", "score": 58.5, "reference_url": "https://arxiv.org/abs/2607.02770", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "provider-reported", "judge": "benchmark-specified", "harness": "official Google evaluation", "prompt_style": "official", "temperature": "not disclosed", "context": "128K" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Gemma 4 Technical Report Table 9; final IT models evaluated without thinking." }, { "model_id": "gemma-4-e4b", "benchmark_id": "mtob_eng_kgv_half_book", "score": 37.8, "reference_url": "https://arxiv.org/abs/2607.02770", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "provider-reported", "judge": "benchmark-specified", "harness": "official Google evaluation", "prompt_style": "official", "temperature": "not disclosed", "context": "about 128K" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Gemma 4 Technical Report Table 9; final IT models evaluated without thinking." }, { "model_id": "gemma-4-e4b", "benchmark_id": "mtob_kgv_eng_half_book", "score": 34.6, "reference_url": "https://arxiv.org/abs/2607.02770", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "provider-reported", "judge": "benchmark-specified", "harness": "official Google evaluation", "prompt_style": "official", "temperature": "not disclosed", "context": "about 128K" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Gemma 4 Technical Report Table 9; final IT models evaluated without thinking." }, { "model_id": "gemma-4-e4b", "benchmark_id": "omnidocbench_1.5", "score": 0.181, "reference_url": "https://arxiv.org/abs/2607.02770", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "provider-reported", "judge": "benchmark-specified", "harness": "official Google evaluation", "prompt_style": "official", "temperature": "not disclosed", "context": "default", "visual_tokens": 1120 }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Gemma 4 Technical Report Table 6; maximum supported resolution, 1,120 visual tokens.", "candidates": [ { "score": 0.307, "reference_url": "https://arxiv.org/abs/2607.02770", "source_type": "tech_report", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "provider-reported", "judge": "benchmark-specified", "harness": "official Google evaluation", "prompt_style": "official", "temperature": "not disclosed", "context": "default", "visual_tokens": 280 }, "notes": "Gemma 4 Technical Report Appendix Table 12; thinking with 280 visual tokens." } ] }, { "model_id": "gemma-4-e4b", "benchmark_id": "ruler_128k", "score": 86.6, "reference_url": "https://arxiv.org/abs/2607.02770", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "provider-reported", "judge": "benchmark-specified", "harness": "official Google evaluation", "prompt_style": "official", "temperature": "not disclosed", "context": "128K" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Gemma 4 Technical Report Table 9; final IT models evaluated without thinking." }, { "model_id": "gemma-4-e4b", "benchmark_id": "ruler_32k", "score": 95.2, "reference_url": "https://arxiv.org/abs/2607.02770", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "provider-reported", "judge": "benchmark-specified", "harness": "official Google evaluation", "prompt_style": "official", "temperature": "not disclosed", "context": "32K" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Gemma 4 Technical Report Table 9; final IT models evaluated without thinking." }, { "model_id": "gemma-4-e4b", "benchmark_id": "scicode", "score": 24.0, "reference_url": "https://arxiv.org/abs/2607.02770", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "provider-reported", "judge": "unit tests", "harness": "official Google evaluation", "prompt_style": "official", "temperature": "not disclosed", "context": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Gemma 4 Technical Report Table 5; Gemma 4 thinking unless explicitly stated, Gemma 3 27B non-thinking." }, { "model_id": "gemma-4-e4b", "benchmark_id": "tau2_bench_airline", "score": 52.0, "reference_url": "https://arxiv.org/abs/2607.02770", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "benchmark APIs + simulated user", "sampling": "provider-reported", "judge": "state-based task success", "harness": "Tau2-bench", "prompt_style": "official", "temperature": "not disclosed", "context": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Gemma 4 Technical Report Table 5; Gemma 4 thinking unless explicitly stated, Gemma 3 27B non-thinking." }, { "model_id": "gemma-4-e4b", "benchmark_id": "tau2_bench_avg", "score": 42.2, "reference_url": "https://ai.google.dev/gemma/docs/core/model_card_4", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "benchmark APIs + simulated user", "sampling": "provider-reported", "judge": "state-based task success", "harness": "Tau2-bench", "prompt_style": "official", "temperature": "not disclosed", "context": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Google model card matrix; all values are instruction-tuned despite the duplicated base-card surface." }, { "model_id": "gemma-4-e4b", "benchmark_id": "tau2_bench_retail", "score": 57.5, "reference_url": "https://blog.google/innovation-and-ai/technology/developers-tools/gemma-4/", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "benchmark APIs + simulated user", "sampling": "provider-reported", "judge": "state-based task success", "harness": "Tau2-bench", "prompt_style": "official", "temperature": "not disclosed", "context": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "Google Gemma 4 family launch table, duplicated by the DeepMind semantic table.", "candidates": [ { "score": 67.1, "reference_url": "https://arxiv.org/abs/2607.02770", "source_type": "tech_report", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "benchmark APIs + simulated user", "sampling": "provider-reported", "judge": "state-based task success", "harness": "Tau2-bench", "prompt_style": "official", "temperature": "not disclosed", "context": "default" }, "notes": "Gemma 4 Technical Report Table 5; Gemma 4 thinking unless explicitly stated, Gemma 3 27B non-thinking." }, { "score": 42.11, "reference_url": "https://www.liquid.ai/blog/lfm2-5-8b-a1b", "source_type": "official_blog", "reported_setting": { "mode": "source/model default", "effort": "source does not state", "tools": "benchmark-provided tools", "sampling": "source does not state trial count", "judge": "benchmark-specified", "harness": "Liquid AI official comparative evaluation", "prompt_style": "source does not state", "temperature": "source does not state evaluation parameters", "context": "source does not state evaluation context" }, "notes": "Displayed exactly as 42.11. Official Liquid AI reported result." } ] }, { "model_id": "gemma-4-e4b", "benchmark_id": "tau2_bench_telecom", "score": 18.4, "reference_url": "https://arxiv.org/abs/2607.02770", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "benchmark APIs + simulated user", "sampling": "provider-reported", "judge": "state-based task success", "harness": "Tau2-bench", "prompt_style": "official", "temperature": "not disclosed", "context": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Gemma 4 Technical Report Table 5; Gemma 4 thinking unless explicitly stated, Gemma 3 27B non-thinking.", "candidates": [ { "score": 26.75, "reference_url": "https://www.liquid.ai/blog/lfm2-5-8b-a1b", "source_type": "official_blog", "reported_setting": { "mode": "source/model default", "effort": "source does not state", "tools": "benchmark-provided tools", "sampling": "source does not state trial count", "judge": "benchmark-specified", "harness": "Liquid AI official comparative evaluation", "prompt_style": "source does not state", "temperature": "source does not state evaluation parameters", "context": "source does not state evaluation context" }, "notes": "Displayed exactly as 26.75. Official Liquid AI reported result." } ] }, { "model_id": "gemma-4-e4b", "benchmark_id": "terminal_bench_hard", "score": 8.0, "reference_url": "https://arxiv.org/abs/2607.02770", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "agentic terminal", "sampling": "provider-reported", "judge": "benchmark-specified", "harness": "Terminal-Bench Hard provider harness", "prompt_style": "official", "temperature": "not disclosed", "context": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Gemma 4 Technical Report Table 5; Gemma 4 thinking unless explicitly stated, Gemma 3 27B non-thinking." }, { "model_id": "glm-5", "benchmark_id": "text_arena_elo", "score": 1456, "reference_url": "https://deepmind.google/models/gemma/gemma-4/", "reported_setting": { "mode": "unknown", "effort": "unknown", "tools": "none", "sampling": "live human pairwise votes", "judge": "blind human raters", "harness": "Arena Text", "prompt_style": "live user prompts", "temperature": "unknown", "context": "default", "snapshot_date": "2026-04-02" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "DeepMind page accessibility text provides the exact plotted Elo value.", "candidates": [ { "score": 1457, "reference_url": "https://arxiv.org/abs/2607.02770", "source_type": "tech_report", "reported_setting": { "mode": "unknown", "effort": "unknown", "tools": "none", "sampling": "live human pairwise votes", "judge": "blind human raters", "harness": "Arena Text", "prompt_style": "live user prompts", "temperature": "unknown", "context": "default", "snapshot_date": "2026-06-19" }, "notes": "Gemma 4 Technical Report Table 4 dated Arena Text Elo snapshot." } ] }, { "model_id": "glm-5.1", "benchmark_id": "text_arena_elo", "score": 1475, "reference_url": "https://arxiv.org/abs/2607.02770", "reported_setting": { "mode": "unknown", "effort": "unknown", "tools": "none", "sampling": "live human pairwise votes", "judge": "blind human raters", "harness": "Arena Text", "prompt_style": "live user prompts", "temperature": "unknown", "context": "default", "snapshot_date": "2026-06-19" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Gemma 4 Technical Report Table 4 dated Arena Text Elo snapshot." }, { "model_id": "gpt-oss-120b", "benchmark_id": "text_arena_elo", "score": 1354, "reference_url": "https://deepmind.google/models/gemma/gemma-4/", "reported_setting": { "mode": "unknown", "effort": "unknown", "tools": "none", "sampling": "live human pairwise votes", "judge": "blind human raters", "harness": "Arena Text", "prompt_style": "live user prompts", "temperature": "unknown", "context": "default", "snapshot_date": "2026-04-02" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "DeepMind page accessibility text provides the exact plotted Elo value." }, { "model_id": "kimi-k2.5", "benchmark_id": "text_arena_elo", "score": 1455, "reference_url": "https://deepmind.google/models/gemma/gemma-4/", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "none", "sampling": "live human pairwise votes", "judge": "blind human raters", "harness": "Arena Text", "prompt_style": "live user prompts", "temperature": "unknown", "context": "default", "snapshot_date": "2026-04-02" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "DeepMind page accessibility text provides the exact plotted Elo value.", "candidates": [ { "score": 1450, "reference_url": "https://arxiv.org/abs/2607.02770", "source_type": "tech_report", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "none", "sampling": "live human pairwise votes", "judge": "blind human raters", "harness": "Arena Text", "prompt_style": "live user prompts", "temperature": "unknown", "context": "default", "snapshot_date": "2026-06-19" }, "notes": "Gemma 4 Technical Report Table 4 dated Arena Text Elo snapshot." }, { "score": 1454, "reference_url": "https://huggingface.co/blog/gemma4", "source_type": "official_blog", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "none", "sampling": "live human pairwise votes", "judge": "blind human raters", "harness": "Arena Text", "prompt_style": "live user prompts", "temperature": "unknown", "context": "default", "snapshot_date": "2026-04-02" }, "notes": "Hugging Face collaboration blog chart with numeric labels." } ] }, { "model_id": "kimi-k2.6", "benchmark_id": "text_arena_elo", "score": 1460, "reference_url": "https://arxiv.org/abs/2607.02770", "reported_setting": { "mode": "unknown", "effort": "unknown", "tools": "none", "sampling": "live human pairwise votes", "judge": "blind human raters", "harness": "Arena Text", "prompt_style": "live user prompts", "temperature": "unknown", "context": "default", "snapshot_date": "2026-06-19" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Gemma 4 Technical Report Table 4 dated Arena Text Elo snapshot." }, { "model_id": "mimo-v2.5-pro", "benchmark_id": "text_arena_elo", "score": 1466, "reference_url": "https://arxiv.org/abs/2607.02770", "reported_setting": { "mode": "unknown", "effort": "unknown", "tools": "none", "sampling": "live human pairwise votes", "judge": "blind human raters", "harness": "Arena Text", "prompt_style": "live user prompts", "temperature": "unknown", "context": "default", "snapshot_date": "2026-06-19" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Gemma 4 Technical Report Table 4 dated Arena Text Elo snapshot." }, { "model_id": "mistral-large-3", "benchmark_id": "text_arena_elo", "score": 1416, "reference_url": "https://deepmind.google/models/gemma/gemma-4/", "reported_setting": { "mode": "unknown", "effort": "unknown", "tools": "none", "sampling": "live human pairwise votes", "judge": "blind human raters", "harness": "Arena Text", "prompt_style": "live user prompts", "temperature": "unknown", "context": "default", "snapshot_date": "2026-04-02" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "DeepMind page accessibility text provides the exact plotted Elo value." }, { "model_id": "qwen3.5-122b-a10b", "benchmark_id": "text_arena_elo", "score": 1416, "reference_url": "https://deepmind.google/models/gemma/gemma-4/", "reported_setting": { "mode": "unknown", "effort": "unknown", "tools": "none", "sampling": "live human pairwise votes", "judge": "blind human raters", "harness": "Arena Text", "prompt_style": "live user prompts", "temperature": "unknown", "context": "default", "snapshot_date": "2026-04-02" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "DeepMind page accessibility text provides the exact plotted Elo value." }, { "model_id": "qwen3.5-27b", "benchmark_id": "text_arena_elo", "score": 1404, "reference_url": "https://deepmind.google/models/gemma/gemma-4/", "reported_setting": { "mode": "unknown", "effort": "unknown", "tools": "none", "sampling": "live human pairwise votes", "judge": "blind human raters", "harness": "Arena Text", "prompt_style": "live user prompts", "temperature": "unknown", "context": "default", "snapshot_date": "2026-04-02" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "DeepMind page accessibility text provides the exact plotted Elo value." }, { "model_id": "qwen3.5-397b", "benchmark_id": "text_arena_elo", "score": 1450, "reference_url": "https://deepmind.google/models/gemma/gemma-4/", "reported_setting": { "mode": "unknown", "effort": "unknown", "tools": "none", "sampling": "live human pairwise votes", "judge": "blind human raters", "harness": "Arena Text", "prompt_style": "live user prompts", "temperature": "unknown", "context": "default", "snapshot_date": "2026-04-02" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "DeepMind page accessibility text provides the exact plotted Elo value.", "candidates": [ { "score": 1444, "reference_url": "https://arxiv.org/abs/2607.02770", "source_type": "tech_report", "reported_setting": { "mode": "unknown", "effort": "unknown", "tools": "none", "sampling": "live human pairwise votes", "judge": "blind human raters", "harness": "Arena Text", "prompt_style": "live user prompts", "temperature": "unknown", "context": "default", "snapshot_date": "2026-06-19" }, "notes": "Gemma 4 Technical Report Table 4 dated Arena Text Elo snapshot." } ] }, { "model_id": "diffusiongemma-26b-a4b", "benchmark_id": "aime_2026", "score": 69.1, "reference_url": "https://blog.google/innovation-and-ai/technology/developers-tools/diffusion-gemma-faster-text-generation/", "reported_setting": { "mode": "thinking", "decoding": "TD", "sampler": "entropy-bounded adaptive", "max_denoising_steps": 48, "canvas": 256, "tools": "none", "sampling": "replicate count N undisclosed", "judge": "benchmark-specified", "harness": "official Google launch evaluation", "prompt_style": "official provider chat template", "context": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "Google release-blog benchmark chart; highest-priority duplicate for target quality values.", "candidates": [ { "score": 50.8, "reference_url": "https://arxiv.org/abs/2608.00146", "source_type": "tech_report", "reported_setting": { "mode": "non-thinking", "decoding": "TD", "sampler": "entropy-bounded adaptive", "max_denoising_steps": 48, "canvas": 256, "tools": "none", "sampling": "replicate count N undisclosed", "judge": "benchmark-specified", "harness": "official Google cross-model evaluation", "prompt_style": "official provider chat template", "context": "default" }, "notes": "DiffusionGemma Technical Report Table 3; replicate count N is undisclosed." }, { "score": 84.2, "reference_url": "https://arxiv.org/abs/2608.00146", "source_type": "tech_report", "reported_setting": { "mode": "thinking", "decoding": "AR", "canvas": 256, "tools": "none", "sampling": "replicate count N undisclosed", "judge": "benchmark-specified", "harness": "official Google cross-model evaluation", "prompt_style": "official provider chat template", "context": "default" }, "notes": "DiffusionGemma Technical Report Table 3; replicate count N is undisclosed." }, { "score": 57.5, "reference_url": "https://arxiv.org/abs/2608.00146", "source_type": "tech_report", "reported_setting": { "mode": "non-thinking", "decoding": "AR", "canvas": 256, "tools": "none", "sampling": "replicate count N undisclosed", "judge": "benchmark-specified", "harness": "official Google cross-model evaluation", "prompt_style": "official provider chat template", "context": "default" }, "notes": "DiffusionGemma Technical Report Table 3; replicate count N is undisclosed." } ] }, { "model_id": "diffusiongemma-26b-a4b", "benchmark_id": "bigbench_extra_hard", "score": 47.6, "reference_url": "https://arxiv.org/abs/2608.00146", "reported_setting": { "mode": "thinking", "decoding": "TD", "sampler": "entropy-bounded adaptive", "max_denoising_steps": 48, "canvas": 256, "tools": "none", "sampling": "replicate count N undisclosed", "judge": "benchmark-specified", "harness": "official Google cross-model evaluation", "prompt_style": "official provider chat template", "context": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "DiffusionGemma Technical Report Table 3; replicate count N is undisclosed.", "candidates": [ { "score": 40.0, "reference_url": "https://arxiv.org/abs/2608.00146", "source_type": "tech_report", "reported_setting": { "mode": "non-thinking", "decoding": "TD", "sampler": "entropy-bounded adaptive", "max_denoising_steps": 48, "canvas": 256, "tools": "none", "sampling": "replicate count N undisclosed", "judge": "benchmark-specified", "harness": "official Google cross-model evaluation", "prompt_style": "official provider chat template", "context": "default" }, "notes": "DiffusionGemma Technical Report Table 3; replicate count N is undisclosed." }, { "score": 59.1, "reference_url": "https://arxiv.org/abs/2608.00146", "source_type": "tech_report", "reported_setting": { "mode": "thinking", "decoding": "AR", "canvas": 256, "tools": "none", "sampling": "replicate count N undisclosed", "judge": "benchmark-specified", "harness": "official Google cross-model evaluation", "prompt_style": "official provider chat template", "context": "default" }, "notes": "DiffusionGemma Technical Report Table 3; replicate count N is undisclosed." }, { "score": 42.2, "reference_url": "https://arxiv.org/abs/2608.00146", "source_type": "tech_report", "reported_setting": { "mode": "non-thinking", "decoding": "AR", "canvas": 256, "tools": "none", "sampling": "replicate count N undisclosed", "judge": "benchmark-specified", "harness": "official Google cross-model evaluation", "prompt_style": "official provider chat template", "context": "default" }, "notes": "DiffusionGemma Technical Report Table 3; replicate count N is undisclosed." } ] }, { "model_id": "diffusiongemma-26b-a4b", "benchmark_id": "codeelo_rating", "score": 1429, "reference_url": "https://arxiv.org/abs/2608.00146", "reported_setting": { "mode": "thinking", "decoding": "TD", "sampler": "entropy-bounded adaptive", "max_denoising_steps": 48, "canvas": 256, "tools": "none", "sampling": "Google run count undisclosed; canonical samples=8", "judge": "Codeforces online judge", "harness": "official Google cross-model evaluation", "prompt_style": "official provider chat template", "context": "default" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "DiffusionGemma Technical Report Table 3; replicate count N is undisclosed.", "candidates": [ { "score": 959, "reference_url": "https://arxiv.org/abs/2608.00146", "source_type": "tech_report", "reported_setting": { "mode": "non-thinking", "decoding": "TD", "sampler": "entropy-bounded adaptive", "max_denoising_steps": 48, "canvas": 256, "tools": "none", "sampling": "Google run count undisclosed; canonical samples=8", "judge": "Codeforces online judge", "harness": "official Google cross-model evaluation", "prompt_style": "official provider chat template", "context": "default" }, "notes": "DiffusionGemma Technical Report Table 3; replicate count N is undisclosed." }, { "score": 1569, "reference_url": "https://arxiv.org/abs/2608.00146", "source_type": "tech_report", "reported_setting": { "mode": "thinking", "decoding": "AR", "canvas": 256, "tools": "none", "sampling": "Google run count undisclosed; canonical samples=8", "judge": "Codeforces online judge", "harness": "official Google cross-model evaluation", "prompt_style": "official provider chat template", "context": "default" }, "notes": "DiffusionGemma Technical Report Table 3; replicate count N is undisclosed." }, { "score": 1059, "reference_url": "https://arxiv.org/abs/2608.00146", "source_type": "tech_report", "reported_setting": { "mode": "non-thinking", "decoding": "AR", "canvas": 256, "tools": "none", "sampling": "Google run count undisclosed; canonical samples=8", "judge": "Codeforces online judge", "harness": "official Google cross-model evaluation", "prompt_style": "official provider chat template", "context": "default" }, "notes": "DiffusionGemma Technical Report Table 3; replicate count N is undisclosed." } ] }, { "model_id": "diffusiongemma-26b-a4b", "benchmark_id": "gdm_mrcr_v2_8needle_upto_128k", "score": 32.0, "reference_url": "https://huggingface.co/google/diffusiongemma-26B-A4B-it", "reported_setting": { "mode": "thinking", "decoding": "TD", "sampler": "entropy-bounded adaptive", "max_denoising_steps": 48, "canvas": 256, "tools": "none", "sampling": "replicate count N undisclosed", "judge": "official MRCR strict scorer", "harness": "official Google model-card evaluation", "prompt_style": "official provider chat template", "context": "up to 128K" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Official HF model-card matrix; Kaggle is a duplicate physical mirror." }, { "model_id": "diffusiongemma-26b-a4b", "benchmark_id": "gpqa_diamond", "score": 73.2, "reference_url": "https://blog.google/innovation-and-ai/technology/developers-tools/diffusion-gemma-faster-text-generation/", "reported_setting": { "mode": "thinking", "decoding": "TD", "sampler": "entropy-bounded adaptive", "max_denoising_steps": 48, "canvas": 256, "tools": "none", "sampling": "replicate count N undisclosed", "judge": "benchmark-specified", "harness": "official Google launch evaluation", "prompt_style": "official provider chat template", "context": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "Google release-blog benchmark chart; highest-priority duplicate for target quality values.", "candidates": [ { "score": 64.6, "reference_url": "https://arxiv.org/abs/2608.00146", "source_type": "tech_report", "reported_setting": { "mode": "non-thinking", "decoding": "TD", "sampler": "entropy-bounded adaptive", "max_denoising_steps": 48, "canvas": 256, "tools": "none", "sampling": "replicate count N undisclosed", "judge": "benchmark-specified", "harness": "official Google cross-model evaluation", "prompt_style": "official provider chat template", "context": "default" }, "notes": "DiffusionGemma Technical Report Table 3; replicate count N is undisclosed." }, { "score": 79.8, "reference_url": "https://arxiv.org/abs/2608.00146", "source_type": "tech_report", "reported_setting": { "mode": "thinking", "decoding": "AR", "canvas": 256, "tools": "none", "sampling": "replicate count N undisclosed", "judge": "benchmark-specified", "harness": "official Google cross-model evaluation", "prompt_style": "official provider chat template", "context": "default" }, "notes": "DiffusionGemma Technical Report Table 3; replicate count N is undisclosed." }, { "score": 67.2, "reference_url": "https://arxiv.org/abs/2608.00146", "source_type": "tech_report", "reported_setting": { "mode": "non-thinking", "decoding": "AR", "canvas": 256, "tools": "none", "sampling": "replicate count N undisclosed", "judge": "benchmark-specified", "harness": "official Google cross-model evaluation", "prompt_style": "official provider chat template", "context": "default" }, "notes": "DiffusionGemma Technical Report Table 3; replicate count N is undisclosed." } ] }, { "model_id": "diffusiongemma-26b-a4b", "benchmark_id": "gsm8k", "score": 96.3, "reference_url": "https://arxiv.org/abs/2608.00146", "reported_setting": { "mode": "thinking", "decoding": "TD", "sampler": "entropy-bounded adaptive", "max_denoising_steps": 48, "canvas": 256, "tools": "none", "sampling": "replicate count N undisclosed", "judge": "benchmark-specified", "harness": "official Google cross-model evaluation", "prompt_style": "official provider chat template", "context": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "DiffusionGemma Technical Report Table 3; replicate count N is undisclosed.", "candidates": [ { "score": 95.8, "reference_url": "https://arxiv.org/abs/2608.00146", "source_type": "tech_report", "reported_setting": { "mode": "non-thinking", "decoding": "TD", "sampler": "entropy-bounded adaptive", "max_denoising_steps": 48, "canvas": 256, "tools": "none", "sampling": "replicate count N undisclosed", "judge": "benchmark-specified", "harness": "official Google cross-model evaluation", "prompt_style": "official provider chat template", "context": "default" }, "notes": "DiffusionGemma Technical Report Table 3; replicate count N is undisclosed." }, { "score": 96.6, "reference_url": "https://arxiv.org/abs/2608.00146", "source_type": "tech_report", "reported_setting": { "mode": "thinking", "decoding": "AR", "canvas": 256, "tools": "none", "sampling": "replicate count N undisclosed", "judge": "benchmark-specified", "harness": "official Google cross-model evaluation", "prompt_style": "official provider chat template", "context": "default" }, "notes": "DiffusionGemma Technical Report Table 3; replicate count N is undisclosed." }, { "score": 96.1, "reference_url": "https://arxiv.org/abs/2608.00146", "source_type": "tech_report", "reported_setting": { "mode": "non-thinking", "decoding": "AR", "canvas": 256, "tools": "none", "sampling": "replicate count N undisclosed", "judge": "benchmark-specified", "harness": "official Google cross-model evaluation", "prompt_style": "official provider chat template", "context": "default" }, "notes": "DiffusionGemma Technical Report Table 3; replicate count N is undisclosed." } ] }, { "model_id": "diffusiongemma-26b-a4b", "benchmark_id": "hiddenmath", "score": 80.6, "reference_url": "https://arxiv.org/abs/2608.00146", "reported_setting": { "mode": "thinking", "decoding": "TD", "sampler": "entropy-bounded adaptive", "max_denoising_steps": 48, "canvas": 256, "tools": "none", "sampling": "replicate count N undisclosed", "judge": "benchmark-specified", "harness": "official Google cross-model evaluation", "prompt_style": "official provider chat template", "context": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "DiffusionGemma Technical Report Table 3; replicate count N is undisclosed.", "candidates": [ { "score": 74.3, "reference_url": "https://arxiv.org/abs/2608.00146", "source_type": "tech_report", "reported_setting": { "mode": "non-thinking", "decoding": "TD", "sampler": "entropy-bounded adaptive", "max_denoising_steps": 48, "canvas": 256, "tools": "none", "sampling": "replicate count N undisclosed", "judge": "benchmark-specified", "harness": "official Google cross-model evaluation", "prompt_style": "official provider chat template", "context": "default" }, "notes": "DiffusionGemma Technical Report Table 3; replicate count N is undisclosed." }, { "score": 85.4, "reference_url": "https://arxiv.org/abs/2608.00146", "source_type": "tech_report", "reported_setting": { "mode": "thinking", "decoding": "AR", "canvas": 256, "tools": "none", "sampling": "replicate count N undisclosed", "judge": "benchmark-specified", "harness": "official Google cross-model evaluation", "prompt_style": "official provider chat template", "context": "default" }, "notes": "DiffusionGemma Technical Report Table 3; replicate count N is undisclosed." }, { "score": 77.5, "reference_url": "https://arxiv.org/abs/2608.00146", "source_type": "tech_report", "reported_setting": { "mode": "non-thinking", "decoding": "AR", "canvas": 256, "tools": "none", "sampling": "replicate count N undisclosed", "judge": "benchmark-specified", "harness": "official Google cross-model evaluation", "prompt_style": "official provider chat template", "context": "default" }, "notes": "DiffusionGemma Technical Report Table 3; replicate count N is undisclosed." } ] }, { "model_id": "diffusiongemma-26b-a4b", "benchmark_id": "hle", "score": 11.0, "reference_url": "https://huggingface.co/google/diffusiongemma-26B-A4B-it", "reported_setting": { "mode": "thinking", "decoding": "TD", "sampler": "entropy-bounded adaptive", "max_denoising_steps": 48, "canvas": 256, "tools": "none", "sampling": "replicate count N undisclosed", "judge": "HLE official rubric", "harness": "official Google model-card evaluation", "prompt_style": "official provider chat template", "context": "default" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "Official HF model-card matrix; Kaggle is a duplicate physical mirror.", "candidates": [ { "score": 11.9, "reference_url": "https://huggingface.co/google/diffusiongemma-26B-A4B-it", "source_type": "model_card", "reported_setting": { "mode": "thinking", "decoding": "TD", "sampler": "entropy-bounded adaptive", "max_denoising_steps": 48, "canvas": 256, "tools": "web search", "sampling": "replicate count N undisclosed", "judge": "HLE official rubric", "harness": "official Google model-card evaluation", "prompt_style": "official provider chat template", "context": "default" }, "notes": "Official HF model-card matrix; Kaggle is a duplicate physical mirror." } ] }, { "model_id": "diffusiongemma-26b-a4b", "benchmark_id": "humaneval", "score": 94.5, "reference_url": "https://arxiv.org/abs/2608.00146", "reported_setting": { "mode": "thinking", "decoding": "TD", "sampler": "entropy-bounded adaptive", "max_denoising_steps": 48, "canvas": 256, "tools": "none", "sampling": "replicate count N undisclosed", "judge": "benchmark-specified", "harness": "official Google cross-model evaluation", "prompt_style": "official provider chat template", "context": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "DiffusionGemma Technical Report Table 3; replicate count N is undisclosed.", "candidates": [ { "score": 92.7, "reference_url": "https://arxiv.org/abs/2608.00146", "source_type": "tech_report", "reported_setting": { "mode": "non-thinking", "decoding": "TD", "sampler": "entropy-bounded adaptive", "max_denoising_steps": 48, "canvas": 256, "tools": "none", "sampling": "replicate count N undisclosed", "judge": "benchmark-specified", "harness": "official Google cross-model evaluation", "prompt_style": "official provider chat template", "context": "default" }, "notes": "DiffusionGemma Technical Report Table 3; replicate count N is undisclosed." }, { "score": 98.2, "reference_url": "https://arxiv.org/abs/2608.00146", "source_type": "tech_report", "reported_setting": { "mode": "thinking", "decoding": "AR", "canvas": 256, "tools": "none", "sampling": "replicate count N undisclosed", "judge": "benchmark-specified", "harness": "official Google cross-model evaluation", "prompt_style": "official provider chat template", "context": "default" }, "notes": "DiffusionGemma Technical Report Table 3; replicate count N is undisclosed." }, { "score": 97.6, "reference_url": "https://arxiv.org/abs/2608.00146", "source_type": "tech_report", "reported_setting": { "mode": "non-thinking", "decoding": "AR", "canvas": 256, "tools": "none", "sampling": "replicate count N undisclosed", "judge": "benchmark-specified", "harness": "official Google cross-model evaluation", "prompt_style": "official provider chat template", "context": "default" }, "notes": "DiffusionGemma Technical Report Table 3; replicate count N is undisclosed." } ] }, { "model_id": "diffusiongemma-26b-a4b", "benchmark_id": "ifeval", "score": 97.4, "reference_url": "https://arxiv.org/abs/2608.00146", "reported_setting": { "mode": "thinking", "decoding": "TD", "sampler": "entropy-bounded adaptive", "max_denoising_steps": 48, "canvas": 256, "tools": "none", "sampling": "replicate count N undisclosed", "judge": "IFEval strict/loose subtype undisclosed", "harness": "official Google cross-model evaluation", "prompt_style": "official provider chat template", "context": "default" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "DiffusionGemma Technical Report Table 3; replicate count N is undisclosed.", "candidates": [ { "score": 94.5, "reference_url": "https://arxiv.org/abs/2608.00146", "source_type": "tech_report", "reported_setting": { "mode": "non-thinking", "decoding": "TD", "sampler": "entropy-bounded adaptive", "max_denoising_steps": 48, "canvas": 256, "tools": "none", "sampling": "replicate count N undisclosed", "judge": "IFEval strict/loose subtype undisclosed", "harness": "official Google cross-model evaluation", "prompt_style": "official provider chat template", "context": "default" }, "notes": "DiffusionGemma Technical Report Table 3; replicate count N is undisclosed." }, { "score": 97.2, "reference_url": "https://arxiv.org/abs/2608.00146", "source_type": "tech_report", "reported_setting": { "mode": "thinking", "decoding": "AR", "canvas": 256, "tools": "none", "sampling": "replicate count N undisclosed", "judge": "IFEval strict/loose subtype undisclosed", "harness": "official Google cross-model evaluation", "prompt_style": "official provider chat template", "context": "default" }, "notes": "DiffusionGemma Technical Report Table 3; replicate count N is undisclosed." }, { "score": 95.7, "reference_url": "https://arxiv.org/abs/2608.00146", "source_type": "tech_report", "reported_setting": { "mode": "non-thinking", "decoding": "AR", "canvas": 256, "tools": "none", "sampling": "replicate count N undisclosed", "judge": "IFEval strict/loose subtype undisclosed", "harness": "official Google cross-model evaluation", "prompt_style": "official provider chat template", "context": "default" }, "notes": "DiffusionGemma Technical Report Table 3; replicate count N is undisclosed." } ] }, { "model_id": "diffusiongemma-26b-a4b", "benchmark_id": "lbpp_v2_multilingual", "score": 81.0, "reference_url": "https://arxiv.org/abs/2608.00146", "reported_setting": { "mode": "thinking", "decoding": "TD", "sampler": "entropy-bounded adaptive", "max_denoising_steps": 48, "canvas": 256, "tools": "none", "sampling": "replicate count N undisclosed", "judge": "benchmark-specified", "harness": "official Google cross-model evaluation", "prompt_style": "official provider chat template", "context": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "DiffusionGemma Technical Report Table 3; replicate count N is undisclosed.", "candidates": [ { "score": 68.9, "reference_url": "https://arxiv.org/abs/2608.00146", "source_type": "tech_report", "reported_setting": { "mode": "non-thinking", "decoding": "TD", "sampler": "entropy-bounded adaptive", "max_denoising_steps": 48, "canvas": 256, "tools": "none", "sampling": "replicate count N undisclosed", "judge": "benchmark-specified", "harness": "official Google cross-model evaluation", "prompt_style": "official provider chat template", "context": "default" }, "notes": "DiffusionGemma Technical Report Table 3; replicate count N is undisclosed." }, { "score": 86.3, "reference_url": "https://arxiv.org/abs/2608.00146", "source_type": "tech_report", "reported_setting": { "mode": "thinking", "decoding": "AR", "canvas": 256, "tools": "none", "sampling": "replicate count N undisclosed", "judge": "benchmark-specified", "harness": "official Google cross-model evaluation", "prompt_style": "official provider chat template", "context": "default" }, "notes": "DiffusionGemma Technical Report Table 3; replicate count N is undisclosed." }, { "score": 74.1, "reference_url": "https://arxiv.org/abs/2608.00146", "source_type": "tech_report", "reported_setting": { "mode": "non-thinking", "decoding": "AR", "canvas": 256, "tools": "none", "sampling": "replicate count N undisclosed", "judge": "benchmark-specified", "harness": "official Google cross-model evaluation", "prompt_style": "official provider chat template", "context": "default" }, "notes": "DiffusionGemma Technical Report Table 3; replicate count N is undisclosed." } ] }, { "model_id": "diffusiongemma-26b-a4b", "benchmark_id": "livecodebench_v6", "score": 69.1, "reference_url": "https://blog.google/innovation-and-ai/technology/developers-tools/diffusion-gemma-faster-text-generation/", "reported_setting": { "mode": "thinking", "decoding": "TD", "sampler": "entropy-bounded adaptive", "max_denoising_steps": 48, "canvas": 256, "tools": "none", "sampling": "replicate count N undisclosed", "judge": "benchmark-specified", "harness": "official Google launch evaluation", "prompt_style": "official provider chat template", "context": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "Google release-blog benchmark chart; highest-priority duplicate for target quality values.", "candidates": [ { "score": 60.6, "reference_url": "https://arxiv.org/abs/2608.00146", "source_type": "tech_report", "reported_setting": { "mode": "non-thinking", "decoding": "TD", "sampler": "entropy-bounded adaptive", "max_denoising_steps": 48, "canvas": 256, "tools": "none", "sampling": "replicate count N undisclosed", "judge": "benchmark-specified", "harness": "official Google cross-model evaluation", "prompt_style": "official provider chat template", "context": "default" }, "notes": "DiffusionGemma Technical Report Table 3; replicate count N is undisclosed." }, { "score": 71.4, "reference_url": "https://arxiv.org/abs/2608.00146", "source_type": "tech_report", "reported_setting": { "mode": "thinking", "decoding": "AR", "canvas": 256, "tools": "none", "sampling": "replicate count N undisclosed", "judge": "benchmark-specified", "harness": "official Google cross-model evaluation", "prompt_style": "official provider chat template", "context": "default" }, "notes": "DiffusionGemma Technical Report Table 3; replicate count N is undisclosed." }, { "score": 58.3, "reference_url": "https://arxiv.org/abs/2608.00146", "source_type": "tech_report", "reported_setting": { "mode": "non-thinking", "decoding": "AR", "canvas": 256, "tools": "none", "sampling": "replicate count N undisclosed", "judge": "benchmark-specified", "harness": "official Google cross-model evaluation", "prompt_style": "official provider chat template", "context": "default" }, "notes": "DiffusionGemma Technical Report Table 3; replicate count N is undisclosed." } ] }, { "model_id": "diffusiongemma-26b-a4b", "benchmark_id": "mathvision", "score": 70.5, "reference_url": "https://huggingface.co/google/diffusiongemma-26B-A4B-it", "reported_setting": { "mode": "thinking", "decoding": "TD", "sampler": "entropy-bounded adaptive", "max_denoising_steps": 48, "canvas": 256, "tools": "none", "sampling": "replicate count N undisclosed", "judge": "benchmark-specified", "harness": "official Google model-card evaluation", "prompt_style": "official provider chat template", "context": "default", "visual_tokens": "undisclosed" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Official HF model-card matrix; Kaggle is a duplicate physical mirror." }, { "model_id": "diffusiongemma-26b-a4b", "benchmark_id": "medxpertqa_mm", "score": 49.0, "reference_url": "https://huggingface.co/google/diffusiongemma-26B-A4B-it", "reported_setting": { "mode": "thinking", "decoding": "TD", "sampler": "entropy-bounded adaptive", "max_denoising_steps": 48, "canvas": 256, "tools": "none", "sampling": "replicate count N undisclosed", "judge": "benchmark-specified", "harness": "official Google model-card evaluation", "prompt_style": "official provider chat template", "context": "default", "visual_tokens": "undisclosed" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Official HF model-card matrix; Kaggle is a duplicate physical mirror." }, { "model_id": "diffusiongemma-26b-a4b", "benchmark_id": "mgsm", "score": 84.8, "reference_url": "https://arxiv.org/abs/2608.00146", "reported_setting": { "mode": "thinking", "decoding": "TD", "sampler": "entropy-bounded adaptive", "max_denoising_steps": 48, "canvas": 256, "tools": "none", "sampling": "replicate count N undisclosed", "judge": "benchmark-specified", "harness": "official Google cross-model evaluation", "prompt_style": "official provider chat template", "context": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "DiffusionGemma Technical Report Table 3; replicate count N is undisclosed.", "candidates": [ { "score": 80.7, "reference_url": "https://arxiv.org/abs/2608.00146", "source_type": "tech_report", "reported_setting": { "mode": "non-thinking", "decoding": "TD", "sampler": "entropy-bounded adaptive", "max_denoising_steps": 48, "canvas": 256, "tools": "none", "sampling": "replicate count N undisclosed", "judge": "benchmark-specified", "harness": "official Google cross-model evaluation", "prompt_style": "official provider chat template", "context": "default" }, "notes": "DiffusionGemma Technical Report Table 3; replicate count N is undisclosed." }, { "score": 87.9, "reference_url": "https://arxiv.org/abs/2608.00146", "source_type": "tech_report", "reported_setting": { "mode": "thinking", "decoding": "AR", "canvas": 256, "tools": "none", "sampling": "replicate count N undisclosed", "judge": "benchmark-specified", "harness": "official Google cross-model evaluation", "prompt_style": "official provider chat template", "context": "default" }, "notes": "DiffusionGemma Technical Report Table 3; replicate count N is undisclosed." }, { "score": 84.3, "reference_url": "https://arxiv.org/abs/2608.00146", "source_type": "tech_report", "reported_setting": { "mode": "non-thinking", "decoding": "AR", "canvas": 256, "tools": "none", "sampling": "replicate count N undisclosed", "judge": "benchmark-specified", "harness": "official Google cross-model evaluation", "prompt_style": "official provider chat template", "context": "default" }, "notes": "DiffusionGemma Technical Report Table 3; replicate count N is undisclosed." } ] }, { "model_id": "diffusiongemma-26b-a4b", "benchmark_id": "mmlu_pro", "score": 77.6, "reference_url": "https://blog.google/innovation-and-ai/technology/developers-tools/diffusion-gemma-faster-text-generation/", "reported_setting": { "mode": "thinking", "decoding": "TD", "sampler": "entropy-bounded adaptive", "max_denoising_steps": 48, "canvas": 256, "tools": "none", "sampling": "replicate count N undisclosed", "judge": "benchmark-specified", "harness": "official Google launch evaluation", "prompt_style": "official provider chat template", "context": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "Google release-blog benchmark chart; highest-priority duplicate for target quality values.", "candidates": [ { "score": 77.9, "reference_url": "https://arxiv.org/abs/2608.00146", "source_type": "tech_report", "reported_setting": { "mode": "non-thinking", "decoding": "TD", "sampler": "entropy-bounded adaptive", "max_denoising_steps": 48, "canvas": 256, "tools": "none", "sampling": "replicate count N undisclosed", "judge": "benchmark-specified", "harness": "official Google cross-model evaluation", "prompt_style": "official provider chat template", "context": "default" }, "notes": "DiffusionGemma Technical Report Table 3; replicate count N is undisclosed." }, { "score": 78.8, "reference_url": "https://arxiv.org/abs/2608.00146", "source_type": "tech_report", "reported_setting": { "mode": "thinking", "decoding": "AR", "canvas": 256, "tools": "none", "sampling": "replicate count N undisclosed", "judge": "benchmark-specified", "harness": "official Google cross-model evaluation", "prompt_style": "official provider chat template", "context": "default" }, "notes": "DiffusionGemma Technical Report Table 3; replicate count N is undisclosed." }, { "score": 79.1, "reference_url": "https://arxiv.org/abs/2608.00146", "source_type": "tech_report", "reported_setting": { "mode": "non-thinking", "decoding": "AR", "canvas": 256, "tools": "none", "sampling": "replicate count N undisclosed", "judge": "benchmark-specified", "harness": "official Google cross-model evaluation", "prompt_style": "official provider chat template", "context": "default" }, "notes": "DiffusionGemma Technical Report Table 3; replicate count N is undisclosed." } ] }, { "model_id": "diffusiongemma-26b-a4b", "benchmark_id": "mmmlu", "score": 81.5, "reference_url": "https://blog.google/innovation-and-ai/technology/developers-tools/diffusion-gemma-faster-text-generation/", "reported_setting": { "mode": "thinking", "decoding": "TD", "sampler": "entropy-bounded adaptive", "max_denoising_steps": 48, "canvas": 256, "tools": "none", "sampling": "replicate count N undisclosed", "judge": "benchmark-specified", "harness": "official Google launch evaluation", "prompt_style": "official provider chat template", "context": "default" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "Google release-blog benchmark chart; highest-priority duplicate for target quality values.", "candidates": [ { "score": 76.3, "reference_url": "https://arxiv.org/abs/2608.00146", "source_type": "tech_report", "reported_setting": { "mode": "non-thinking", "decoding": "TD", "sampler": "entropy-bounded adaptive", "max_denoising_steps": 48, "canvas": 256, "tools": "none", "sampling": "replicate count N undisclosed", "judge": "benchmark-specified", "harness": "official Google cross-model evaluation", "prompt_style": "official provider chat template", "context": "default" }, "notes": "DiffusionGemma Technical Report Table 3; replicate count N is undisclosed." }, { "score": 82.2, "reference_url": "https://arxiv.org/abs/2608.00146", "source_type": "tech_report", "reported_setting": { "mode": "thinking", "decoding": "AR", "canvas": 256, "tools": "none", "sampling": "replicate count N undisclosed", "judge": "benchmark-specified", "harness": "official Google cross-model evaluation", "prompt_style": "official provider chat template", "context": "default" }, "notes": "DiffusionGemma Technical Report Table 3; replicate count N is undisclosed." }, { "score": 78.0, "reference_url": "https://arxiv.org/abs/2608.00146", "source_type": "tech_report", "reported_setting": { "mode": "non-thinking", "decoding": "AR", "canvas": 256, "tools": "none", "sampling": "replicate count N undisclosed", "judge": "benchmark-specified", "harness": "official Google cross-model evaluation", "prompt_style": "official provider chat template", "context": "default" }, "notes": "DiffusionGemma Technical Report Table 3; replicate count N is undisclosed." } ] }, { "model_id": "diffusiongemma-26b-a4b", "benchmark_id": "mmmu_pro", "score": 54.3, "reference_url": "https://arxiv.org/abs/2608.00146", "reported_setting": { "mode": "thinking", "decoding": "TD", "sampler": "entropy-bounded adaptive", "max_denoising_steps": 48, "canvas": 256, "tools": "none", "sampling": "replicate count N undisclosed", "judge": "benchmark-specified", "harness": "official Google cross-model evaluation", "prompt_style": "official provider chat template", "context": "default", "visual_tokens": "undisclosed" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "DiffusionGemma Technical Report Table 3; replicate count N is undisclosed.", "candidates": [ { "score": 66.0, "reference_url": "https://arxiv.org/abs/2608.00146", "source_type": "tech_report", "reported_setting": { "mode": "non-thinking", "decoding": "TD", "sampler": "entropy-bounded adaptive", "max_denoising_steps": 48, "canvas": 256, "tools": "none", "sampling": "replicate count N undisclosed", "judge": "benchmark-specified", "harness": "official Google cross-model evaluation", "prompt_style": "official provider chat template", "context": "default", "visual_tokens": "undisclosed" }, "notes": "DiffusionGemma Technical Report Table 3; replicate count N is undisclosed." }, { "score": 63.3, "reference_url": "https://arxiv.org/abs/2608.00146", "source_type": "tech_report", "reported_setting": { "mode": "thinking", "decoding": "AR", "canvas": 256, "tools": "none", "sampling": "replicate count N undisclosed", "judge": "benchmark-specified", "harness": "official Google cross-model evaluation", "prompt_style": "official provider chat template", "context": "default", "visual_tokens": "undisclosed" }, "notes": "DiffusionGemma Technical Report Table 3; replicate count N is undisclosed." }, { "score": 66.7, "reference_url": "https://arxiv.org/abs/2608.00146", "source_type": "tech_report", "reported_setting": { "mode": "non-thinking", "decoding": "AR", "canvas": 256, "tools": "none", "sampling": "replicate count N undisclosed", "judge": "benchmark-specified", "harness": "official Google cross-model evaluation", "prompt_style": "official provider chat template", "context": "default", "visual_tokens": "undisclosed" }, "notes": "DiffusionGemma Technical Report Table 3; replicate count N is undisclosed." } ] }, { "model_id": "diffusiongemma-26b-a4b", "benchmark_id": "natural2code", "score": 94.0, "reference_url": "https://arxiv.org/abs/2608.00146", "reported_setting": { "mode": "thinking", "decoding": "TD", "sampler": "entropy-bounded adaptive", "max_denoising_steps": 48, "canvas": 256, "tools": "none", "sampling": "replicate count N undisclosed", "judge": "benchmark-specified", "harness": "official Google cross-model evaluation", "prompt_style": "official provider chat template", "context": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "DiffusionGemma Technical Report Table 3; replicate count N is undisclosed.", "candidates": [ { "score": 90.1, "reference_url": "https://arxiv.org/abs/2608.00146", "source_type": "tech_report", "reported_setting": { "mode": "non-thinking", "decoding": "TD", "sampler": "entropy-bounded adaptive", "max_denoising_steps": 48, "canvas": 256, "tools": "none", "sampling": "replicate count N undisclosed", "judge": "benchmark-specified", "harness": "official Google cross-model evaluation", "prompt_style": "official provider chat template", "context": "default" }, "notes": "DiffusionGemma Technical Report Table 3; replicate count N is undisclosed." }, { "score": 96.2, "reference_url": "https://arxiv.org/abs/2608.00146", "source_type": "tech_report", "reported_setting": { "mode": "thinking", "decoding": "AR", "canvas": 256, "tools": "none", "sampling": "replicate count N undisclosed", "judge": "benchmark-specified", "harness": "official Google cross-model evaluation", "prompt_style": "official provider chat template", "context": "default" }, "notes": "DiffusionGemma Technical Report Table 3; replicate count N is undisclosed." }, { "score": 92.3, "reference_url": "https://arxiv.org/abs/2608.00146", "source_type": "tech_report", "reported_setting": { "mode": "non-thinking", "decoding": "AR", "canvas": 256, "tools": "none", "sampling": "replicate count N undisclosed", "judge": "benchmark-specified", "harness": "official Google cross-model evaluation", "prompt_style": "official provider chat template", "context": "default" }, "notes": "DiffusionGemma Technical Report Table 3; replicate count N is undisclosed." } ] }, { "model_id": "diffusiongemma-26b-a4b", "benchmark_id": "omnidocbench_1.5", "score": 0.319, "reference_url": "https://huggingface.co/google/diffusiongemma-26B-A4B-it", "reported_setting": { "mode": "thinking", "decoding": "TD", "sampler": "entropy-bounded adaptive", "max_denoising_steps": 48, "canvas": 256, "tools": "none", "sampling": "replicate count N undisclosed", "judge": "benchmark-specified", "harness": "official Google model-card evaluation", "prompt_style": "official provider chat template", "context": "default", "visual_tokens": "undisclosed" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "Official HF model-card matrix; Kaggle is a duplicate physical mirror." }, { "model_id": "diffusiongemma-26b-a4b", "benchmark_id": "pubmedqa_pqal_decision_accuracy", "score": 75.6, "reference_url": "https://arxiv.org/abs/2608.00146", "reported_setting": { "mode": "thinking", "decoding": "TD", "sampler": "entropy-bounded adaptive", "max_denoising_steps": 48, "canvas": 128, "tools": "none", "sampling": "replicate count N undisclosed", "judge": "benchmark-specified", "harness": "official Google cross-model evaluation", "prompt_style": "official provider chat template", "context": "1024-token prompt", "number_of_canvases": 2 }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Released DiffusionGemma before downstream LoRA. TD-thinking is inherited globally, but the PubMedQA recipe uses a noncanonical 128-token canvas with two canvases." }, { "model_id": "diffusiongemma-26b-a4b", "benchmark_id": "pubmedqa_pqal_long_answer_bleu", "score": 10.76, "reference_url": "https://arxiv.org/abs/2608.00146", "reported_setting": { "mode": "thinking", "decoding": "TD", "sampler": "entropy-bounded adaptive", "max_denoising_steps": 48, "canvas": 128, "tools": "none", "sampling": "replicate count N undisclosed", "judge": "benchmark-specified", "harness": "official Google cross-model evaluation", "prompt_style": "official provider chat template", "context": "1024-token prompt", "number_of_canvases": 2 }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Released DiffusionGemma before downstream LoRA. TD-thinking is inherited globally, but the PubMedQA recipe uses a noncanonical 128-token canvas with two canvases." }, { "model_id": "diffusiongemma-26b-a4b", "benchmark_id": "tau2_bench_airline", "score": 69.0, "reference_url": "https://arxiv.org/abs/2608.00146", "reported_setting": { "mode": "thinking", "decoding": "TD", "sampler": "entropy-bounded adaptive", "max_denoising_steps": 48, "canvas": 256, "tools": "benchmark APIs + simulated user", "sampling": "trials=4 per task inferred from score totals; N otherwise undisclosed", "judge": "state-based task success", "harness": "official Google cross-model evaluation", "prompt_style": "official provider chat template", "context": "default" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "DiffusionGemma Technical Report Table 3; replicate count N is undisclosed.", "candidates": [ { "score": 49.0, "reference_url": "https://arxiv.org/abs/2608.00146", "source_type": "tech_report", "reported_setting": { "mode": "non-thinking", "decoding": "TD", "sampler": "entropy-bounded adaptive", "max_denoising_steps": 48, "canvas": 256, "tools": "benchmark APIs + simulated user", "sampling": "trials=4 per task inferred from score totals; N otherwise undisclosed", "judge": "state-based task success", "harness": "official Google cross-model evaluation", "prompt_style": "official provider chat template", "context": "default" }, "notes": "DiffusionGemma Technical Report Table 3; replicate count N is undisclosed." }, { "score": 72.0, "reference_url": "https://arxiv.org/abs/2608.00146", "source_type": "tech_report", "reported_setting": { "mode": "thinking", "decoding": "AR", "canvas": 256, "tools": "benchmark APIs + simulated user", "sampling": "trials=4 per task inferred from score totals; N otherwise undisclosed", "judge": "state-based task success", "harness": "official Google cross-model evaluation", "prompt_style": "official provider chat template", "context": "default" }, "notes": "DiffusionGemma Technical Report Table 3; replicate count N is undisclosed." }, { "score": 50.0, "reference_url": "https://arxiv.org/abs/2608.00146", "source_type": "tech_report", "reported_setting": { "mode": "non-thinking", "decoding": "AR", "canvas": 256, "tools": "benchmark APIs + simulated user", "sampling": "trials=4 per task inferred from score totals; N otherwise undisclosed", "judge": "state-based task success", "harness": "official Google cross-model evaluation", "prompt_style": "official provider chat template", "context": "default" }, "notes": "DiffusionGemma Technical Report Table 3; replicate count N is undisclosed." } ] }, { "model_id": "diffusiongemma-26b-a4b", "benchmark_id": "tau2_bench_avg", "score": 56.2, "reference_url": "https://blog.google/innovation-and-ai/technology/developers-tools/diffusion-gemma-faster-text-generation/", "reported_setting": { "mode": "thinking", "decoding": "TD", "sampler": "entropy-bounded adaptive", "max_denoising_steps": 48, "canvas": 256, "tools": "benchmark APIs + simulated user", "sampling": "trials=4 per task inferred from score totals; N otherwise undisclosed", "judge": "state-based task success", "harness": "official Google launch evaluation", "prompt_style": "official provider chat template", "context": "default" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Google release-blog benchmark chart; highest-priority duplicate for target quality values." }, { "model_id": "diffusiongemma-26b-a4b", "benchmark_id": "tau2_bench_retail", "score": 71.5, "reference_url": "https://arxiv.org/abs/2608.00146", "reported_setting": { "mode": "thinking", "decoding": "TD", "sampler": "entropy-bounded adaptive", "max_denoising_steps": 48, "canvas": 256, "tools": "benchmark APIs + simulated user", "sampling": "trials=4 per task inferred from score totals; N otherwise undisclosed", "judge": "state-based task success", "harness": "official Google cross-model evaluation", "prompt_style": "official provider chat template", "context": "default" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "DiffusionGemma Technical Report Table 3; replicate count N is undisclosed.", "candidates": [ { "score": 57.5, "reference_url": "https://arxiv.org/abs/2608.00146", "source_type": "tech_report", "reported_setting": { "mode": "non-thinking", "decoding": "TD", "sampler": "entropy-bounded adaptive", "max_denoising_steps": 48, "canvas": 256, "tools": "benchmark APIs + simulated user", "sampling": "trials=4 per task inferred from score totals; N otherwise undisclosed", "judge": "state-based task success", "harness": "official Google cross-model evaluation", "prompt_style": "official provider chat template", "context": "default" }, "notes": "DiffusionGemma Technical Report Table 3; replicate count N is undisclosed." }, { "score": 75.4, "reference_url": "https://arxiv.org/abs/2608.00146", "source_type": "tech_report", "reported_setting": { "mode": "thinking", "decoding": "AR", "canvas": 256, "tools": "benchmark APIs + simulated user", "sampling": "trials=4 per task inferred from score totals; N otherwise undisclosed", "judge": "state-based task success", "harness": "official Google cross-model evaluation", "prompt_style": "official provider chat template", "context": "default" }, "notes": "DiffusionGemma Technical Report Table 3; replicate count N is undisclosed." }, { "score": 61.0, "reference_url": "https://arxiv.org/abs/2608.00146", "source_type": "tech_report", "reported_setting": { "mode": "non-thinking", "decoding": "AR", "canvas": 256, "tools": "benchmark APIs + simulated user", "sampling": "trials=4 per task inferred from score totals; N otherwise undisclosed", "judge": "state-based task success", "harness": "official Google cross-model evaluation", "prompt_style": "official provider chat template", "context": "default" }, "notes": "DiffusionGemma Technical Report Table 3; replicate count N is undisclosed." } ] }, { "model_id": "diffusiongemma-26b-a4b", "benchmark_id": "tau2_bench_telecom", "score": 28.1, "reference_url": "https://arxiv.org/abs/2608.00146", "reported_setting": { "mode": "thinking", "decoding": "TD", "sampler": "entropy-bounded adaptive", "max_denoising_steps": 48, "canvas": 256, "tools": "benchmark APIs + simulated user", "sampling": "trials=4 per task inferred from score totals; N otherwise undisclosed", "judge": "state-based task success", "harness": "official Google cross-model evaluation", "prompt_style": "official provider chat template", "context": "default" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "DiffusionGemma Technical Report Table 3; replicate count N is undisclosed.", "candidates": [ { "score": 32.0, "reference_url": "https://arxiv.org/abs/2608.00146", "source_type": "tech_report", "reported_setting": { "mode": "non-thinking", "decoding": "TD", "sampler": "entropy-bounded adaptive", "max_denoising_steps": 48, "canvas": 256, "tools": "benchmark APIs + simulated user", "sampling": "trials=4 per task inferred from score totals; N otherwise undisclosed", "judge": "state-based task success", "harness": "official Google cross-model evaluation", "prompt_style": "official provider chat template", "context": "default" }, "notes": "DiffusionGemma Technical Report Table 3; replicate count N is undisclosed." }, { "score": 33.8, "reference_url": "https://arxiv.org/abs/2608.00146", "source_type": "tech_report", "reported_setting": { "mode": "thinking", "decoding": "AR", "canvas": 256, "tools": "benchmark APIs + simulated user", "sampling": "trials=4 per task inferred from score totals; N otherwise undisclosed", "judge": "state-based task success", "harness": "official Google cross-model evaluation", "prompt_style": "official provider chat template", "context": "default" }, "notes": "DiffusionGemma Technical Report Table 3; replicate count N is undisclosed." }, { "score": 32.0, "reference_url": "https://arxiv.org/abs/2608.00146", "source_type": "tech_report", "reported_setting": { "mode": "non-thinking", "decoding": "AR", "canvas": 256, "tools": "benchmark APIs + simulated user", "sampling": "trials=4 per task inferred from score totals; N otherwise undisclosed", "judge": "state-based task success", "harness": "official Google cross-model evaluation", "prompt_style": "official provider chat template", "context": "default" }, "notes": "DiffusionGemma Technical Report Table 3; replicate count N is undisclosed." } ] }, { "model_id": "gemma-4-26b-a4b", "benchmark_id": "codeelo_rating", "score": 1718, "reference_url": "https://arxiv.org/abs/2608.00146", "reported_setting": { "mode": "thinking", "decoding": "AR MTP", "tools": "none", "sampling": "Google run count undisclosed; canonical samples=8", "judge": "Codeforces online judge", "harness": "official Google cross-model evaluation", "prompt_style": "official provider chat template", "context": "default" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "DiffusionGemma Technical Report Table 3; replicate count N is undisclosed.", "candidates": [ { "score": 1529, "reference_url": "https://arxiv.org/abs/2608.00146", "source_type": "tech_report", "reported_setting": { "mode": "non-thinking", "decoding": "AR MTP", "tools": "none", "sampling": "Google run count undisclosed; canonical samples=8", "judge": "Codeforces online judge", "harness": "official Google cross-model evaluation", "prompt_style": "official provider chat template", "context": "default" }, "notes": "DiffusionGemma Technical Report Table 3; replicate count N is undisclosed." } ] }, { "model_id": "gemma-4-26b-a4b", "benchmark_id": "gsm8k", "score": 96.7, "reference_url": "https://arxiv.org/abs/2608.00146", "reported_setting": { "mode": "thinking", "decoding": "AR MTP", "tools": "none", "sampling": "replicate count N undisclosed", "judge": "benchmark-specified", "harness": "official Google cross-model evaluation", "prompt_style": "official provider chat template", "context": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "DiffusionGemma Technical Report Table 3; replicate count N is undisclosed.", "candidates": [ { "score": 96.4, "reference_url": "https://arxiv.org/abs/2608.00146", "source_type": "tech_report", "reported_setting": { "mode": "non-thinking", "decoding": "AR MTP", "tools": "none", "sampling": "replicate count N undisclosed", "judge": "benchmark-specified", "harness": "official Google cross-model evaluation", "prompt_style": "official provider chat template", "context": "default" }, "notes": "DiffusionGemma Technical Report Table 3; replicate count N is undisclosed." } ] }, { "model_id": "gemma-4-26b-a4b", "benchmark_id": "hiddenmath", "score": 87.2, "reference_url": "https://arxiv.org/abs/2608.00146", "reported_setting": { "mode": "thinking", "decoding": "AR MTP", "tools": "none", "sampling": "replicate count N undisclosed", "judge": "benchmark-specified", "harness": "official Google cross-model evaluation", "prompt_style": "official provider chat template", "context": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "DiffusionGemma Technical Report Table 3; replicate count N is undisclosed.", "candidates": [ { "score": 81.6, "reference_url": "https://arxiv.org/abs/2608.00146", "source_type": "tech_report", "reported_setting": { "mode": "non-thinking", "decoding": "AR MTP", "tools": "none", "sampling": "replicate count N undisclosed", "judge": "benchmark-specified", "harness": "official Google cross-model evaluation", "prompt_style": "official provider chat template", "context": "default" }, "notes": "DiffusionGemma Technical Report Table 3; replicate count N is undisclosed." } ] }, { "model_id": "gemma-4-26b-a4b", "benchmark_id": "humaneval", "score": 98.8, "reference_url": "https://arxiv.org/abs/2608.00146", "reported_setting": { "mode": "thinking", "decoding": "AR MTP", "tools": "none", "sampling": "replicate count N undisclosed", "judge": "benchmark-specified", "harness": "official Google cross-model evaluation", "prompt_style": "official provider chat template", "context": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "DiffusionGemma Technical Report Table 3; replicate count N is undisclosed.", "candidates": [ { "score": 97.6, "reference_url": "https://arxiv.org/abs/2608.00146", "source_type": "tech_report", "reported_setting": { "mode": "non-thinking", "decoding": "AR MTP", "tools": "none", "sampling": "replicate count N undisclosed", "judge": "benchmark-specified", "harness": "official Google cross-model evaluation", "prompt_style": "official provider chat template", "context": "default" }, "notes": "DiffusionGemma Technical Report Table 3; replicate count N is undisclosed." } ] }, { "model_id": "gemma-4-26b-a4b", "benchmark_id": "lbpp_v2_multilingual", "score": 89.5, "reference_url": "https://arxiv.org/abs/2608.00146", "reported_setting": { "mode": "thinking", "decoding": "AR MTP", "tools": "none", "sampling": "replicate count N undisclosed", "judge": "benchmark-specified", "harness": "official Google cross-model evaluation", "prompt_style": "official provider chat template", "context": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "DiffusionGemma Technical Report Table 3; replicate count N is undisclosed.", "candidates": [ { "score": 77.3, "reference_url": "https://arxiv.org/abs/2608.00146", "source_type": "tech_report", "reported_setting": { "mode": "non-thinking", "decoding": "AR MTP", "tools": "none", "sampling": "replicate count N undisclosed", "judge": "benchmark-specified", "harness": "official Google cross-model evaluation", "prompt_style": "official provider chat template", "context": "default" }, "notes": "DiffusionGemma Technical Report Table 3; replicate count N is undisclosed." } ] }, { "model_id": "gemma-4-26b-a4b", "benchmark_id": "mgsm", "score": 92.9, "reference_url": "https://arxiv.org/abs/2608.00146", "reported_setting": { "mode": "thinking", "decoding": "AR MTP", "tools": "none", "sampling": "replicate count N undisclosed", "judge": "benchmark-specified", "harness": "official Google cross-model evaluation", "prompt_style": "official provider chat template", "context": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "DiffusionGemma Technical Report Table 3; replicate count N is undisclosed.", "candidates": [ { "score": 91.5, "reference_url": "https://arxiv.org/abs/2608.00146", "source_type": "tech_report", "reported_setting": { "mode": "non-thinking", "decoding": "AR MTP", "tools": "none", "sampling": "replicate count N undisclosed", "judge": "benchmark-specified", "harness": "official Google cross-model evaluation", "prompt_style": "official provider chat template", "context": "default" }, "notes": "DiffusionGemma Technical Report Table 3; replicate count N is undisclosed." } ] }, { "model_id": "gemma-4-26b-a4b", "benchmark_id": "natural2code", "score": 96.3, "reference_url": "https://arxiv.org/abs/2608.00146", "reported_setting": { "mode": "thinking", "decoding": "AR MTP", "tools": "none", "sampling": "replicate count N undisclosed", "judge": "benchmark-specified", "harness": "official Google cross-model evaluation", "prompt_style": "official provider chat template", "context": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "DiffusionGemma Technical Report Table 3; replicate count N is undisclosed.", "candidates": [ { "score": 94.7, "reference_url": "https://arxiv.org/abs/2608.00146", "source_type": "tech_report", "reported_setting": { "mode": "non-thinking", "decoding": "AR MTP", "tools": "none", "sampling": "replicate count N undisclosed", "judge": "benchmark-specified", "harness": "official Google cross-model evaluation", "prompt_style": "official provider chat template", "context": "default" }, "notes": "DiffusionGemma Technical Report Table 3; replicate count N is undisclosed." } ] }, { "model_id": "llada-2.1-flash-100b", "benchmark_id": "aime_2026", "score": 80.0, "reference_url": "https://arxiv.org/abs/2608.00146", "reported_setting": { "mode": "unknown", "decoding": "TD S Mode", "thinking_state": "undisclosed", "threshold": 0.5, "editing_threshold": 0.0, "block_length": 32, "temperature": 0.0, "max_post_steps": 16, "tools": "none", "sampling": "replicate count N undisclosed", "judge": "benchmark-specified", "harness": "official Google cross-model evaluation", "prompt_style": "official provider chat template", "context": "default" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "DiffusionGemma Technical Report Table 3; replicate count N is undisclosed." }, { "model_id": "llada-2.1-flash-100b", "benchmark_id": "codeelo_rating", "score": 718, "reference_url": "https://arxiv.org/abs/2608.00146", "reported_setting": { "mode": "unknown", "decoding": "TD S Mode", "thinking_state": "undisclosed", "threshold": 0.5, "editing_threshold": 0.0, "block_length": 32, "temperature": 0.0, "max_post_steps": 16, "tools": "none", "sampling": "Google run count undisclosed; canonical samples=8", "judge": "Codeforces online judge", "harness": "official Google cross-model evaluation", "prompt_style": "official provider chat template", "context": "default" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "DiffusionGemma Technical Report Table 3; replicate count N is undisclosed." }, { "model_id": "llada-2.1-flash-100b", "benchmark_id": "gpqa_diamond", "score": 68.7, "reference_url": "https://arxiv.org/abs/2608.00146", "reported_setting": { "mode": "unknown", "decoding": "TD S Mode", "thinking_state": "undisclosed", "threshold": 0.5, "editing_threshold": 0.0, "block_length": 32, "temperature": 0.0, "max_post_steps": 16, "tools": "none", "sampling": "replicate count N undisclosed", "judge": "benchmark-specified", "harness": "official Google cross-model evaluation", "prompt_style": "official provider chat template", "context": "default" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "DiffusionGemma Technical Report Table 3; replicate count N is undisclosed." }, { "model_id": "llada-2.1-flash-100b", "benchmark_id": "gsm8k", "score": 45.0, "reference_url": "https://arxiv.org/abs/2608.00146", "reported_setting": { "mode": "unknown", "decoding": "TD S Mode", "thinking_state": "undisclosed", "threshold": 0.5, "editing_threshold": 0.0, "block_length": 32, "temperature": 0.0, "max_post_steps": 16, "tools": "none", "sampling": "replicate count N undisclosed", "judge": "benchmark-specified", "harness": "official Google cross-model evaluation", "prompt_style": "official provider chat template", "context": "default" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "DiffusionGemma Technical Report Table 3; replicate count N is undisclosed." }, { "model_id": "llada-2.1-flash-100b", "benchmark_id": "humaneval", "score": 90.2, "reference_url": "https://arxiv.org/abs/2608.00146", "reported_setting": { "mode": "unknown", "decoding": "TD S Mode", "thinking_state": "undisclosed", "threshold": 0.5, "editing_threshold": 0.0, "block_length": 32, "temperature": 0.0, "max_post_steps": 16, "tools": "none", "sampling": "replicate count N undisclosed", "judge": "benchmark-specified", "harness": "official Google cross-model evaluation", "prompt_style": "official provider chat template", "context": "default" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "DiffusionGemma Technical Report Table 3; replicate count N is undisclosed." }, { "model_id": "llada-2.1-flash-100b", "benchmark_id": "lbpp_v2_multilingual", "score": 45.7, "reference_url": "https://arxiv.org/abs/2608.00146", "reported_setting": { "mode": "unknown", "decoding": "TD S Mode", "thinking_state": "undisclosed", "threshold": 0.5, "editing_threshold": 0.0, "block_length": 32, "temperature": 0.0, "max_post_steps": 16, "tools": "none", "sampling": "replicate count N undisclosed", "judge": "benchmark-specified", "harness": "official Google cross-model evaluation", "prompt_style": "official provider chat template", "context": "default" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "DiffusionGemma Technical Report Table 3; replicate count N is undisclosed." }, { "model_id": "llada-2.1-flash-100b", "benchmark_id": "livecodebench_v6", "score": 39.4, "reference_url": "https://arxiv.org/abs/2608.00146", "reported_setting": { "mode": "unknown", "decoding": "TD S Mode", "thinking_state": "undisclosed", "threshold": 0.5, "editing_threshold": 0.0, "block_length": 32, "temperature": 0.0, "max_post_steps": 16, "tools": "none", "sampling": "replicate count N undisclosed", "judge": "benchmark-specified", "harness": "official Google cross-model evaluation", "prompt_style": "official provider chat template", "context": "default" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "DiffusionGemma Technical Report Table 3; replicate count N is undisclosed." }, { "model_id": "llada-2.1-flash-100b", "benchmark_id": "mgsm", "score": 6.8, "reference_url": "https://arxiv.org/abs/2608.00146", "reported_setting": { "mode": "unknown", "decoding": "TD S Mode", "thinking_state": "undisclosed", "threshold": 0.5, "editing_threshold": 0.0, "block_length": 32, "temperature": 0.0, "max_post_steps": 16, "tools": "none", "sampling": "replicate count N undisclosed", "judge": "benchmark-specified", "harness": "official Google cross-model evaluation", "prompt_style": "official provider chat template", "context": "default" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "DiffusionGemma Technical Report Table 3; replicate count N is undisclosed." }, { "model_id": "llada-2.1-flash-100b", "benchmark_id": "natural2code", "score": 86.9, "reference_url": "https://arxiv.org/abs/2608.00146", "reported_setting": { "mode": "unknown", "decoding": "TD S Mode", "thinking_state": "undisclosed", "threshold": 0.5, "editing_threshold": 0.0, "block_length": 32, "temperature": 0.0, "max_post_steps": 16, "tools": "none", "sampling": "replicate count N undisclosed", "judge": "benchmark-specified", "harness": "official Google cross-model evaluation", "prompt_style": "official provider chat template", "context": "default" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "DiffusionGemma Technical Report Table 3; replicate count N is undisclosed." }, { "model_id": "mercury-2", "benchmark_id": "aime_2026", "score": 82.5, "reference_url": "https://arxiv.org/abs/2608.00146", "reported_setting": { "mode": "thinking", "decoding": "diffusion", "effort": "medium", "temperature": 0.75, "tools": "none", "sampling": "replicate count N undisclosed", "judge": "benchmark-specified", "harness": "official Google cross-model evaluation", "prompt_style": "official provider chat template", "context": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "DiffusionGemma Technical Report Table 3; replicate count N is undisclosed.", "candidates": [ { "score": 91.7, "reference_url": "https://arxiv.org/abs/2608.00146", "source_type": "tech_report", "reported_setting": { "mode": "thinking", "decoding": "diffusion", "effort": "high", "temperature": 0.75, "tools": "none", "sampling": "replicate count N undisclosed", "judge": "benchmark-specified", "harness": "official Google cross-model evaluation", "prompt_style": "official provider chat template", "context": "default" }, "notes": "DiffusionGemma Technical Report Table 3; replicate count N is undisclosed." } ] }, { "model_id": "mercury-2", "benchmark_id": "bigbench_extra_hard", "score": 43.8, "reference_url": "https://arxiv.org/abs/2608.00146", "reported_setting": { "mode": "thinking", "decoding": "diffusion", "effort": "medium", "temperature": 0.75, "tools": "none", "sampling": "replicate count N undisclosed", "judge": "benchmark-specified", "harness": "official Google cross-model evaluation", "prompt_style": "official provider chat template", "context": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "DiffusionGemma Technical Report Table 3; replicate count N is undisclosed.", "candidates": [ { "score": 48.9, "reference_url": "https://arxiv.org/abs/2608.00146", "source_type": "tech_report", "reported_setting": { "mode": "thinking", "decoding": "diffusion", "effort": "high", "temperature": 0.75, "tools": "none", "sampling": "replicate count N undisclosed", "judge": "benchmark-specified", "harness": "official Google cross-model evaluation", "prompt_style": "official provider chat template", "context": "default" }, "notes": "DiffusionGemma Technical Report Table 3; replicate count N is undisclosed." } ] }, { "model_id": "mercury-2", "benchmark_id": "codeelo_rating", "score": 1629, "reference_url": "https://arxiv.org/abs/2608.00146", "reported_setting": { "mode": "thinking", "decoding": "diffusion", "effort": "medium", "temperature": 0.75, "tools": "none", "sampling": "Google run count undisclosed; canonical samples=8", "judge": "Codeforces online judge", "harness": "official Google cross-model evaluation", "prompt_style": "official provider chat template", "context": "default" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "DiffusionGemma Technical Report Table 3; replicate count N is undisclosed.", "candidates": [ { "score": 1986, "reference_url": "https://arxiv.org/abs/2608.00146", "source_type": "tech_report", "reported_setting": { "mode": "thinking", "decoding": "diffusion", "effort": "high", "temperature": 0.75, "tools": "none", "sampling": "Google run count undisclosed; canonical samples=8", "judge": "Codeforces online judge", "harness": "official Google cross-model evaluation", "prompt_style": "official provider chat template", "context": "default" }, "notes": "DiffusionGemma Technical Report Table 3; replicate count N is undisclosed." } ] }, { "model_id": "mercury-2", "benchmark_id": "gpqa_diamond", "score": 66.7, "reference_url": "https://arxiv.org/abs/2608.00146", "reported_setting": { "mode": "thinking", "decoding": "diffusion", "effort": "medium", "temperature": 0.75, "tools": "none", "sampling": "replicate count N undisclosed", "judge": "benchmark-specified", "harness": "official Google cross-model evaluation", "prompt_style": "official provider chat template", "context": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "DiffusionGemma Technical Report Table 3; replicate count N is undisclosed.", "candidates": [ { "score": 75.2, "reference_url": "https://arxiv.org/abs/2608.00146", "source_type": "tech_report", "reported_setting": { "mode": "thinking", "decoding": "diffusion", "effort": "high", "temperature": 0.75, "tools": "none", "sampling": "replicate count N undisclosed", "judge": "benchmark-specified", "harness": "official Google cross-model evaluation", "prompt_style": "official provider chat template", "context": "default" }, "notes": "DiffusionGemma Technical Report Table 3; replicate count N is undisclosed." } ] }, { "model_id": "mercury-2", "benchmark_id": "gsm8k", "score": 95.8, "reference_url": "https://arxiv.org/abs/2608.00146", "reported_setting": { "mode": "thinking", "decoding": "diffusion", "effort": "medium", "temperature": 0.75, "tools": "none", "sampling": "replicate count N undisclosed", "judge": "benchmark-specified", "harness": "official Google cross-model evaluation", "prompt_style": "official provider chat template", "context": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "DiffusionGemma Technical Report Table 3; replicate count N is undisclosed.", "candidates": [ { "score": 96.5, "reference_url": "https://arxiv.org/abs/2608.00146", "source_type": "tech_report", "reported_setting": { "mode": "thinking", "decoding": "diffusion", "effort": "high", "temperature": 0.75, "tools": "none", "sampling": "replicate count N undisclosed", "judge": "benchmark-specified", "harness": "official Google cross-model evaluation", "prompt_style": "official provider chat template", "context": "default" }, "notes": "DiffusionGemma Technical Report Table 3; replicate count N is undisclosed." } ] }, { "model_id": "mercury-2", "benchmark_id": "hiddenmath", "score": 82.3, "reference_url": "https://arxiv.org/abs/2608.00146", "reported_setting": { "mode": "thinking", "decoding": "diffusion", "effort": "medium", "temperature": 0.75, "tools": "none", "sampling": "replicate count N undisclosed", "judge": "benchmark-specified", "harness": "official Google cross-model evaluation", "prompt_style": "official provider chat template", "context": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "DiffusionGemma Technical Report Table 3; replicate count N is undisclosed.", "candidates": [ { "score": 82.7, "reference_url": "https://arxiv.org/abs/2608.00146", "source_type": "tech_report", "reported_setting": { "mode": "thinking", "decoding": "diffusion", "effort": "high", "temperature": 0.75, "tools": "none", "sampling": "replicate count N undisclosed", "judge": "benchmark-specified", "harness": "official Google cross-model evaluation", "prompt_style": "official provider chat template", "context": "default" }, "notes": "DiffusionGemma Technical Report Table 3; replicate count N is undisclosed." } ] }, { "model_id": "mercury-2", "benchmark_id": "humaneval", "score": 98.2, "reference_url": "https://arxiv.org/abs/2608.00146", "reported_setting": { "mode": "thinking", "decoding": "diffusion", "effort": "medium", "temperature": 0.75, "tools": "none", "sampling": "replicate count N undisclosed", "judge": "benchmark-specified", "harness": "official Google cross-model evaluation", "prompt_style": "official provider chat template", "context": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "DiffusionGemma Technical Report Table 3; replicate count N is undisclosed.", "candidates": [ { "score": 98.2, "reference_url": "https://arxiv.org/abs/2608.00146", "source_type": "tech_report", "reported_setting": { "mode": "thinking", "decoding": "diffusion", "effort": "high", "temperature": 0.75, "tools": "none", "sampling": "replicate count N undisclosed", "judge": "benchmark-specified", "harness": "official Google cross-model evaluation", "prompt_style": "official provider chat template", "context": "default" }, "notes": "DiffusionGemma Technical Report Table 3; replicate count N is undisclosed." } ] }, { "model_id": "mercury-2", "benchmark_id": "ifeval", "score": 94.5, "reference_url": "https://arxiv.org/abs/2608.00146", "reported_setting": { "mode": "thinking", "decoding": "diffusion", "effort": "medium", "temperature": 0.75, "tools": "none", "sampling": "replicate count N undisclosed", "judge": "IFEval strict/loose subtype undisclosed", "harness": "official Google cross-model evaluation", "prompt_style": "official provider chat template", "context": "default" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "DiffusionGemma Technical Report Table 3; replicate count N is undisclosed.", "candidates": [ { "score": 97.0, "reference_url": "https://arxiv.org/abs/2608.00146", "source_type": "tech_report", "reported_setting": { "mode": "thinking", "decoding": "diffusion", "effort": "high", "temperature": 0.75, "tools": "none", "sampling": "replicate count N undisclosed", "judge": "IFEval strict/loose subtype undisclosed", "harness": "official Google cross-model evaluation", "prompt_style": "official provider chat template", "context": "default" }, "notes": "DiffusionGemma Technical Report Table 3; replicate count N is undisclosed." } ] }, { "model_id": "mercury-2", "benchmark_id": "lbpp_v2_multilingual", "score": 85.0, "reference_url": "https://arxiv.org/abs/2608.00146", "reported_setting": { "mode": "thinking", "decoding": "diffusion", "effort": "medium", "temperature": 0.75, "tools": "none", "sampling": "replicate count N undisclosed", "judge": "benchmark-specified", "harness": "official Google cross-model evaluation", "prompt_style": "official provider chat template", "context": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "DiffusionGemma Technical Report Table 3; replicate count N is undisclosed.", "candidates": [ { "score": 89.2, "reference_url": "https://arxiv.org/abs/2608.00146", "source_type": "tech_report", "reported_setting": { "mode": "thinking", "decoding": "diffusion", "effort": "high", "temperature": 0.75, "tools": "none", "sampling": "replicate count N undisclosed", "judge": "benchmark-specified", "harness": "official Google cross-model evaluation", "prompt_style": "official provider chat template", "context": "default" }, "notes": "DiffusionGemma Technical Report Table 3; replicate count N is undisclosed." } ] }, { "model_id": "mercury-2", "benchmark_id": "livecodebench_v6", "score": 74.9, "reference_url": "https://arxiv.org/abs/2608.00146", "reported_setting": { "mode": "thinking", "decoding": "diffusion", "effort": "medium", "temperature": 0.75, "tools": "none", "sampling": "replicate count N undisclosed", "judge": "benchmark-specified", "harness": "official Google cross-model evaluation", "prompt_style": "official provider chat template", "context": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "DiffusionGemma Technical Report Table 3; replicate count N is undisclosed.", "candidates": [ { "score": 79.4, "reference_url": "https://arxiv.org/abs/2608.00146", "source_type": "tech_report", "reported_setting": { "mode": "thinking", "decoding": "diffusion", "effort": "high", "temperature": 0.75, "tools": "none", "sampling": "replicate count N undisclosed", "judge": "benchmark-specified", "harness": "official Google cross-model evaluation", "prompt_style": "official provider chat template", "context": "default" }, "notes": "DiffusionGemma Technical Report Table 3; replicate count N is undisclosed." } ] }, { "model_id": "mercury-2", "benchmark_id": "mgsm", "score": 91.2, "reference_url": "https://arxiv.org/abs/2608.00146", "reported_setting": { "mode": "thinking", "decoding": "diffusion", "effort": "medium", "temperature": 0.75, "tools": "none", "sampling": "replicate count N undisclosed", "judge": "benchmark-specified", "harness": "official Google cross-model evaluation", "prompt_style": "official provider chat template", "context": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "DiffusionGemma Technical Report Table 3; replicate count N is undisclosed.", "candidates": [ { "score": 91.9, "reference_url": "https://arxiv.org/abs/2608.00146", "source_type": "tech_report", "reported_setting": { "mode": "thinking", "decoding": "diffusion", "effort": "high", "temperature": 0.75, "tools": "none", "sampling": "replicate count N undisclosed", "judge": "benchmark-specified", "harness": "official Google cross-model evaluation", "prompt_style": "official provider chat template", "context": "default" }, "notes": "DiffusionGemma Technical Report Table 3; replicate count N is undisclosed." } ] }, { "model_id": "mercury-2", "benchmark_id": "mmlu_pro", "score": 75.5, "reference_url": "https://arxiv.org/abs/2608.00146", "reported_setting": { "mode": "thinking", "decoding": "diffusion", "effort": "medium", "temperature": 0.75, "tools": "none", "sampling": "replicate count N undisclosed", "judge": "benchmark-specified", "harness": "official Google cross-model evaluation", "prompt_style": "official provider chat template", "context": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "DiffusionGemma Technical Report Table 3; replicate count N is undisclosed.", "candidates": [ { "score": 77.6, "reference_url": "https://arxiv.org/abs/2608.00146", "source_type": "tech_report", "reported_setting": { "mode": "thinking", "decoding": "diffusion", "effort": "high", "temperature": 0.75, "tools": "none", "sampling": "replicate count N undisclosed", "judge": "benchmark-specified", "harness": "official Google cross-model evaluation", "prompt_style": "official provider chat template", "context": "default" }, "notes": "DiffusionGemma Technical Report Table 3; replicate count N is undisclosed." } ] }, { "model_id": "mercury-2", "benchmark_id": "mmmlu", "score": 80.6, "reference_url": "https://arxiv.org/abs/2608.00146", "reported_setting": { "mode": "thinking", "decoding": "diffusion", "effort": "medium", "temperature": 0.75, "tools": "none", "sampling": "replicate count N undisclosed", "judge": "benchmark-specified", "harness": "official Google cross-model evaluation", "prompt_style": "official provider chat template", "context": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "DiffusionGemma Technical Report Table 3; replicate count N is undisclosed.", "candidates": [ { "score": 81.9, "reference_url": "https://arxiv.org/abs/2608.00146", "source_type": "tech_report", "reported_setting": { "mode": "thinking", "decoding": "diffusion", "effort": "high", "temperature": 0.75, "tools": "none", "sampling": "replicate count N undisclosed", "judge": "benchmark-specified", "harness": "official Google cross-model evaluation", "prompt_style": "official provider chat template", "context": "default" }, "notes": "DiffusionGemma Technical Report Table 3; replicate count N is undisclosed." } ] }, { "model_id": "mercury-2", "benchmark_id": "natural2code", "score": 71.3, "reference_url": "https://arxiv.org/abs/2608.00146", "reported_setting": { "mode": "thinking", "decoding": "diffusion", "effort": "medium", "temperature": 0.75, "tools": "none", "sampling": "replicate count N undisclosed", "judge": "benchmark-specified", "harness": "official Google cross-model evaluation", "prompt_style": "official provider chat template", "context": "default" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "DiffusionGemma Technical Report Table 3; replicate count N is undisclosed.", "candidates": [ { "score": 79.1, "reference_url": "https://arxiv.org/abs/2608.00146", "source_type": "tech_report", "reported_setting": { "mode": "thinking", "decoding": "diffusion", "effort": "high", "temperature": 0.75, "tools": "none", "sampling": "replicate count N undisclosed", "judge": "benchmark-specified", "harness": "official Google cross-model evaluation", "prompt_style": "official provider chat template", "context": "default" }, "notes": "DiffusionGemma Technical Report Table 3; replicate count N is undisclosed." } ] }, { "model_id": "nemotron-diffusion-14b", "benchmark_id": "aime_2026", "score": 40.0, "reference_url": "https://arxiv.org/abs/2608.00146", "reported_setting": { "mode": "unknown", "decoding": "TD Diffusion Mode", "thinking_state": "undisclosed", "block_length": 32, "threshold": 0.9, "tools": "none", "sampling": "replicate count N undisclosed", "judge": "benchmark-specified", "harness": "official Google cross-model evaluation", "prompt_style": "official provider chat template", "context": "default" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "DiffusionGemma Technical Report Table 3; replicate count N is undisclosed." }, { "model_id": "nemotron-diffusion-14b", "benchmark_id": "gpqa_diamond", "score": 47.0, "reference_url": "https://arxiv.org/abs/2608.00146", "reported_setting": { "mode": "unknown", "decoding": "TD Diffusion Mode", "thinking_state": "undisclosed", "block_length": 32, "threshold": 0.9, "tools": "none", "sampling": "replicate count N undisclosed", "judge": "benchmark-specified", "harness": "official Google cross-model evaluation", "prompt_style": "official provider chat template", "context": "default" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "DiffusionGemma Technical Report Table 3; replicate count N is undisclosed." }, { "model_id": "nemotron-diffusion-14b", "benchmark_id": "hiddenmath", "score": 44.3, "reference_url": "https://arxiv.org/abs/2608.00146", "reported_setting": { "mode": "unknown", "decoding": "TD Diffusion Mode", "thinking_state": "undisclosed", "block_length": 32, "threshold": 0.9, "tools": "none", "sampling": "replicate count N undisclosed", "judge": "benchmark-specified", "harness": "official Google cross-model evaluation", "prompt_style": "official provider chat template", "context": "default" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "DiffusionGemma Technical Report Table 3; replicate count N is undisclosed." }, { "model_id": "nemotron-diffusion-14b", "benchmark_id": "humaneval", "score": 86.0, "reference_url": "https://arxiv.org/abs/2608.00146", "reported_setting": { "mode": "unknown", "decoding": "TD Diffusion Mode", "thinking_state": "undisclosed", "block_length": 32, "threshold": 0.9, "tools": "none", "sampling": "replicate count N undisclosed", "judge": "benchmark-specified", "harness": "official Google cross-model evaluation", "prompt_style": "official provider chat template", "context": "default" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "DiffusionGemma Technical Report Table 3; replicate count N is undisclosed." }, { "model_id": "nemotron-diffusion-14b", "benchmark_id": "ifeval", "score": 72.1, "reference_url": "https://arxiv.org/abs/2608.00146", "reported_setting": { "mode": "unknown", "decoding": "TD Diffusion Mode", "thinking_state": "undisclosed", "block_length": 32, "threshold": 0.9, "tools": "none", "sampling": "replicate count N undisclosed", "judge": "IFEval strict/loose subtype undisclosed", "harness": "official Google cross-model evaluation", "prompt_style": "official provider chat template", "context": "default" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "DiffusionGemma Technical Report Table 3; replicate count N is undisclosed." }, { "model_id": "nemotron-diffusion-14b", "benchmark_id": "lbpp_v2_multilingual", "score": 40.7, "reference_url": "https://arxiv.org/abs/2608.00146", "reported_setting": { "mode": "unknown", "decoding": "TD Diffusion Mode", "thinking_state": "undisclosed", "block_length": 32, "threshold": 0.9, "tools": "none", "sampling": "replicate count N undisclosed", "judge": "benchmark-specified", "harness": "official Google cross-model evaluation", "prompt_style": "official provider chat template", "context": "default" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "DiffusionGemma Technical Report Table 3; replicate count N is undisclosed." }, { "model_id": "nemotron-diffusion-14b", "benchmark_id": "livecodebench_v6", "score": 28.6, "reference_url": "https://arxiv.org/abs/2608.00146", "reported_setting": { "mode": "unknown", "decoding": "TD Diffusion Mode", "thinking_state": "undisclosed", "block_length": 32, "threshold": 0.9, "tools": "none", "sampling": "replicate count N undisclosed", "judge": "benchmark-specified", "harness": "official Google cross-model evaluation", "prompt_style": "official provider chat template", "context": "default" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "DiffusionGemma Technical Report Table 3; replicate count N is undisclosed." }, { "model_id": "nemotron-diffusion-14b", "benchmark_id": "mgsm", "score": 69.3, "reference_url": "https://arxiv.org/abs/2608.00146", "reported_setting": { "mode": "unknown", "decoding": "TD Diffusion Mode", "thinking_state": "undisclosed", "block_length": 32, "threshold": 0.9, "tools": "none", "sampling": "replicate count N undisclosed", "judge": "benchmark-specified", "harness": "official Google cross-model evaluation", "prompt_style": "official provider chat template", "context": "default" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "DiffusionGemma Technical Report Table 3; replicate count N is undisclosed." }, { "model_id": "nemotron-diffusion-14b", "benchmark_id": "natural2code", "score": 73.3, "reference_url": "https://arxiv.org/abs/2608.00146", "reported_setting": { "mode": "unknown", "decoding": "TD Diffusion Mode", "thinking_state": "undisclosed", "block_length": 32, "threshold": 0.9, "tools": "none", "sampling": "replicate count N undisclosed", "judge": "benchmark-specified", "harness": "official Google cross-model evaluation", "prompt_style": "official provider chat template", "context": "default" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "DiffusionGemma Technical Report Table 3; replicate count N is undisclosed." }, { "model_id": "claude-fable-5", "benchmark_id": "apex_swe", "score": 58.8, "reference_url": "https://media.x.ai/v1/website/card-4p6-4cd2dc57.pdf", "reported_setting": { "mode": "thinking", "effort": "max, with fallback", "tools": "agentic software-engineering environment", "sampling": "pass@1", "judge": "Mercor integration and observability task verifier", "harness": "Mercor APEX-SWE harness", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "card46 APEX-SWE Pass@1 (%); source effort=max, with fallback; benchmark protocol matches canonical=True. Unknown source fields remain unknown.", "candidates": [ { "score": 54.8, "reference_url": "https://media.x.ai/v1/website/4p5-5184fdf9.pdf", "source_type": "model_card", "reported_setting": { "mode": "thinking", "effort": "max, with fallback", "tools": "agentic software-engineering environment", "sampling": "pass@1", "judge": "Mercor integration and observability task verifier", "harness": "Mercor APEX-SWE harness", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown" }, "notes": "card45 APEX-SWE Pass@1 (%); source effort=max, with fallback; benchmark protocol matches canonical=True. Unknown source fields remain unknown." }, { "score": 58.8, "reference_url": "https://x.ai/news/grok-4-6", "source_type": "official_blog", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "agentic software-engineering environment", "sampling": "pass@1", "judge": "Mercor integration and observability task verifier", "harness": "Mercor APEX-SWE harness", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown" }, "notes": "release46 APEX-SWE Pass@1 (%); source effort=max; benchmark protocol matches canonical=True. Unknown source fields remain unknown." } ] }, { "model_id": "claude-fable-5", "benchmark_id": "deep_swe_v1_0", "score": 66.1, "reference_url": "https://x.ai/news/grok-4-5", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "agentic repository shell/editor", "sampling": "pass@1", "judge": "behavioral and correctness verifiers", "harness": "provider harnesses run by Artificial Analysis", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "release45 DeepSWE 1.0 Pass@1 (%); source effort=max; benchmark protocol matches canonical=False. Unknown source fields remain unknown.", "candidates": [ { "score": 66.1, "reference_url": "https://media.x.ai/v1/website/4p5-5184fdf9.pdf", "source_type": "model_card", "reported_setting": { "mode": "thinking", "effort": "max, with fallback", "tools": "agentic repository shell/editor", "sampling": "pass@1", "judge": "behavioral and correctness verifiers", "harness": "provider harnesses run by Artificial Analysis", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown" }, "notes": "card45 DeepSWE v1.0 Pass@1 (%); source effort=max, with fallback; benchmark protocol matches canonical=False. Unknown source fields remain unknown." } ] }, { "model_id": "claude-fable-5", "benchmark_id": "eebench_v1_core_reward", "score": 54.2, "reference_url": "https://media.x.ai/v1/website/card-4p6-4cd2dc57.pdf", "reported_setting": { "mode": "thinking", "effort": "max, with fallback", "tools": "electrical-engineering design tools", "sampling": "unknown", "judge": "physical correctness and functionality reward", "harness": "EEBench v1 core public leaderboard harness", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "card46 EEBench v1 core Reward (%); source effort=max, with fallback; benchmark protocol matches canonical=True. Unknown source fields remain unknown." }, { "model_id": "claude-fable-5", "benchmark_id": "gdpval_aa_v2_elo", "score": 1741, "reference_url": "https://media.x.ai/v1/website/card-4p6-4cd2dc57.pdf", "reported_setting": { "mode": "thinking", "effort": "max, with fallback", "tools": "benchmark-specified knowledge-work tools", "sampling": "unknown", "judge": "pairwise Bradley-Terry Elo", "harness": "Artificial Analysis GDPval-AA v2 harness", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "card46 AA GDPVal v2 AA GDPVal Elo; source effort=max, with fallback; benchmark protocol matches canonical=True. Unknown source fields remain unknown.", "candidates": [ { "score": 1760, "reference_url": "https://media.x.ai/v1/website/4p5-5184fdf9.pdf", "source_type": "model_card", "reported_setting": { "mode": "thinking", "effort": "max, with fallback", "tools": "benchmark-specified knowledge-work tools", "sampling": "unknown", "judge": "pairwise Bradley-Terry Elo", "harness": "Artificial Analysis GDPval-AA v2 harness", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown" }, "notes": "card45 GDPval-AA v2 AA score; source effort=max, with fallback; benchmark protocol matches canonical=True. Unknown source fields remain unknown." }, { "score": 1741, "reference_url": "https://x.ai/news/grok-4-6", "source_type": "official_blog", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "benchmark-specified knowledge-work tools", "sampling": "unknown", "judge": "pairwise Bradley-Terry Elo", "harness": "Artificial Analysis GDPval-AA v2 harness", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown" }, "notes": "release46 GDPVal-AA v2 Elo; source effort=max; benchmark protocol matches canonical=True. Unknown source fields remain unknown." } ] }, { "model_id": "claude-fable-5", "benchmark_id": "harvey_lab_vals_final_score", "score": 14.2, "reference_url": "https://media.x.ai/v1/website/card-4p6-4cd2dc57.pdf", "reported_setting": { "mode": "thinking", "effort": "max, with fallback", "tools": "Vals legal-agent file tools", "sampling": "unknown", "judge": "expert all-pass final score", "harness": "Vals implementation of Harvey LAB", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "card46 Harvey Legal Agent Benchmark Final score (%); source effort=max, with fallback; benchmark protocol matches canonical=True. Unknown source fields remain unknown.", "candidates": [ { "score": 11.3, "reference_url": "https://x.ai/news/grok-4-6", "source_type": "official_blog", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "Vals legal-agent file tools", "sampling": "unknown", "judge": "expert all-pass final score", "harness": "Vals implementation of Harvey LAB", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown" }, "notes": "release46 Harvey LAB (Vals) Final score (%); source effort=max; benchmark protocol matches canonical=True. Unknown source fields remain unknown." } ] }, { "model_id": "claude-fable-5", "benchmark_id": "parametric_cad_bench_combined", "score": 88.5, "reference_url": "https://media.x.ai/v1/website/4p5-5184fdf9.pdf", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "parametric CAD agent tools", "sampling": "unknown", "judge": "combined executability and geometric reward", "harness": "gNucleus Parametric CAD Bench harness", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "card45 Parametric CAD Bench Reward (%); source effort=max; benchmark protocol matches canonical=True. Unknown source fields remain unknown." }, { "model_id": "claude-fable-5", "benchmark_id": "swe_atlas_codebase_qna", "score": 83.0, "reference_url": "https://media.x.ai/v1/website/4p5-5184fdf9.pdf", "reported_setting": { "mode": "thinking", "effort": "max, with fallback", "tools": "codebase reading and exploration", "sampling": "unknown", "judge": "repository-ground-truth rubric", "harness": "SWE-Atlas-QnA tracked split", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "card45 SWE-Atlas-QnA Accuracy (%); source effort=max, with fallback; benchmark protocol matches canonical=True. Unknown source fields remain unknown." }, { "model_id": "claude-fable-5", "benchmark_id": "swe_marathon_v1_0", "score": 24.0, "reference_url": "https://media.x.ai/v1/website/4p5-5184fdf9.pdf", "reported_setting": { "mode": "thinking", "effort": "max, with fallback", "tools": "agentic code execution", "sampling": "pass@1", "judge": "multi-layer resolution verification", "harness": "SWE-Marathon v1.0 harness", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "card45 SWE-Marathon v1.0 Resolution rate (%); source effort=max, with fallback; benchmark protocol matches canonical=True. Unknown source fields remain unknown.", "candidates": [ { "score": 24.0, "reference_url": "https://x.ai/news/grok-4-5", "source_type": "official_blog", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "agentic code execution", "sampling": "pass@1", "judge": "multi-layer resolution verification", "harness": "SWE-Marathon v1.0 harness", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown" }, "notes": "release45 SWE-Marathon v1.0 Resolution rate (%); source effort=max; benchmark protocol matches canonical=True. Unknown source fields remain unknown." } ] }, { "model_id": "claude-fable-5", "benchmark_id": "swe_marathon_v1_1", "score": 45.0, "reference_url": "https://media.x.ai/v1/website/card-4p6-4cd2dc57.pdf", "reported_setting": { "mode": "thinking", "effort": "max, with fallback", "tools": "agentic code execution", "sampling": "pass@1", "judge": "multi-layer resolution verification", "harness": "SWE-Marathon v1.1 harness", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "card46 SWE-Marathon v1.1 Resolution rate (%); source effort=max, with fallback; benchmark protocol matches canonical=True. Unknown source fields remain unknown." }, { "model_id": "claude-fable-5", "benchmark_id": "terminal_bench_3_0", "score": 34.1, "reference_url": "https://x.ai/news/grok-4-6", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "container terminal", "sampling": "pass@1", "judge": "verified task-success evaluator", "harness": "Terminal-Bench 3.0 / FrontierBench fixed harness", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "release46 Terminal-Bench v3.0 Task success rate (%); source effort=max; benchmark protocol matches canonical=False. Unknown source fields remain unknown.", "candidates": [ { "score": 34.1, "reference_url": "https://media.x.ai/v1/website/card-4p6-4cd2dc57.pdf", "source_type": "model_card", "reported_setting": { "mode": "thinking", "effort": "max, with fallback", "tools": "container terminal", "sampling": "pass@1", "judge": "verified task-success evaluator", "harness": "Terminal-Bench 3.0 / FrontierBench fixed harness", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown" }, "notes": "card46 Terminal-Bench 3.0 Task success rate (%); source effort=max, with fallback; benchmark protocol matches canonical=False. Unknown source fields remain unknown." } ] }, { "model_id": "claude-opus-4.7", "benchmark_id": "deep_swe_v1_0", "score": 40.12, "reference_url": "https://x.ai/news/grok-4-5", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "agentic repository shell/editor", "sampling": "pass@1", "judge": "behavioral and correctness verifiers", "harness": "provider harnesses run by Artificial Analysis", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "release45 DeepSWE 1.0 Pass@1 (%); source effort=max; benchmark protocol matches canonical=False. Unknown source fields remain unknown.", "candidates": [ { "score": 40.1, "reference_url": "https://media.x.ai/v1/website/4p5-5184fdf9.pdf", "source_type": "model_card", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "agentic repository shell/editor", "sampling": "pass@1", "judge": "behavioral and correctness verifiers", "harness": "provider harnesses run by Artificial Analysis", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown" }, "notes": "card45 DeepSWE v1.0 Pass@1 (%); source effort=max; benchmark protocol matches canonical=False. Unknown source fields remain unknown." } ] }, { "model_id": "claude-opus-4.8", "benchmark_id": "apex_swe", "score": 47.3, "reference_url": "https://media.x.ai/v1/website/4p5-5184fdf9.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "agentic software-engineering environment", "sampling": "pass@1", "judge": "Mercor integration and observability task verifier", "harness": "Mercor APEX-SWE harness", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "card45 APEX-SWE Pass@1 (%); source effort=high; benchmark protocol matches canonical=True. Unknown source fields remain unknown.", "candidates": [ { "score": 43.9, "reference_url": "https://media.x.ai/v1/website/card-4p6-4cd2dc57.pdf", "source_type": "model_card", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "agentic software-engineering environment", "sampling": "pass@1", "judge": "Mercor integration and observability task verifier", "harness": "Mercor APEX-SWE harness", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown" }, "notes": "card46 APEX-SWE Pass@1 (%); source effort=max; benchmark protocol matches canonical=True. Unknown source fields remain unknown." } ] }, { "model_id": "claude-opus-4.8", "benchmark_id": "deep_swe_v1_0", "score": 55.75, "reference_url": "https://x.ai/news/grok-4-5", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "agentic repository shell/editor", "sampling": "pass@1", "judge": "behavioral and correctness verifiers", "harness": "provider harnesses run by Artificial Analysis", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "release45 DeepSWE 1.0 Pass@1 (%); source effort=max; benchmark protocol matches canonical=False. Unknown source fields remain unknown.", "candidates": [ { "score": 55.8, "reference_url": "https://media.x.ai/v1/website/4p5-5184fdf9.pdf", "source_type": "model_card", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "agentic repository shell/editor", "sampling": "pass@1", "judge": "behavioral and correctness verifiers", "harness": "provider harnesses run by Artificial Analysis", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown" }, "notes": "card45 DeepSWE v1.0 Pass@1 (%); source effort=max; benchmark protocol matches canonical=False. Unknown source fields remain unknown." } ] }, { "model_id": "claude-opus-4.8", "benchmark_id": "deepsearchqa_acc", "score": 84.8, "reference_url": "https://media.x.ai/v1/website/card-4p6-4cd2dc57.pdf", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "web search agent", "sampling": "unknown", "judge": "DeepSearchQA answer-accuracy grader", "harness": "SpaceXAI internal DeepSearchQA implementation", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "card46 DeepSearchQA Accuracy (%); source effort=max; benchmark protocol matches canonical=False. Unknown source fields remain unknown.", "candidates": [ { "score": 40.7, "reference_url": "https://media.x.ai/v1/website/4p5-5184fdf9.pdf", "source_type": "model_card", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "web search agent", "sampling": "unknown", "judge": "DeepSearchQA answer-accuracy grader", "harness": "SpaceXAI internal DeepSearchQA implementation", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown" }, "notes": "card45 DeepSearchQA Accuracy (%); source effort=max; benchmark protocol matches canonical=False. Unknown source fields remain unknown." } ] }, { "model_id": "claude-opus-4.8", "benchmark_id": "eebench_v1_core_reward", "score": 51.4, "reference_url": "https://media.x.ai/v1/website/card-4p6-4cd2dc57.pdf", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "electrical-engineering design tools", "sampling": "unknown", "judge": "physical correctness and functionality reward", "harness": "EEBench v1 core public leaderboard harness", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "card46 EEBench v1 core Reward (%); source effort=max; benchmark protocol matches canonical=True. Unknown source fields remain unknown." }, { "model_id": "claude-opus-4.8", "benchmark_id": "falseclaimbench_accuracy", "score": 59.0, "reference_url": "https://media.x.ai/v1/website/4p5-5184fdf9.pdf", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "workspace editing and command tools", "sampling": "unknown", "judge": "claim-to-final-workspace-state verifier", "harness": "SpaceXAI internal FalseClaimBench harness", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "card45 FalseClaimBench Accuracy (%); source effort=max; benchmark protocol matches canonical=True. Unknown source fields remain unknown." }, { "model_id": "claude-opus-4.8", "benchmark_id": "hackerbench_v0_2_benign_refusal", "score": 0.0, "reference_url": "https://media.x.ai/v1/website/card-4p6-4cd2dc57.pdf", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "constrained cyber task tools", "sampling": "unknown", "judge": "benign refusal grader", "harness": "HackerBench v0.2 with standard safeguards", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "card46 HackerBench v0.2 Benign Refusal (%); source effort=max; benchmark protocol matches canonical=True. Unknown source fields remain unknown." }, { "model_id": "claude-opus-4.8", "benchmark_id": "hackerbench_v0_2_harmful_compliance", "score": 22.8, "reference_url": "https://media.x.ai/v1/website/card-4p6-4cd2dc57.pdf", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "constrained cyber task tools", "sampling": "unknown", "judge": "harmful or dual-use compliance grader", "harness": "HackerBench v0.2 with standard safeguards", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "card46 HackerBench v0.2 Harmful/Dual-Use Compliance (%); source effort=max; benchmark protocol matches canonical=True. Unknown source fields remain unknown." }, { "model_id": "claude-opus-4.8", "benchmark_id": "parametric_cad_bench_combined", "score": 88.2, "reference_url": "https://media.x.ai/v1/website/4p5-5184fdf9.pdf", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "parametric CAD agent tools", "sampling": "unknown", "judge": "combined executability and geometric reward", "harness": "gNucleus Parametric CAD Bench harness", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "card45 Parametric CAD Bench Reward (%); source effort=max; benchmark protocol matches canonical=True. Unknown source fields remain unknown." }, { "model_id": "claude-opus-4.8", "benchmark_id": "single_turn_hallucination_rate", "score": 3.4, "reference_url": "https://media.x.ai/v1/website/card-4p6-4cd2dc57.pdf", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "none", "sampling": "unknown", "judge": "separate factual-support grader", "harness": "SpaceXAI internal single-turn factuality harness", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "card46 Single-turn hallucination Hallucination rate (%); source effort=max; benchmark protocol matches canonical=True. Unknown source fields remain unknown.", "candidates": [ { "score": 3.35, "reference_url": "https://media.x.ai/v1/website/4p5-5184fdf9.pdf", "source_type": "model_card", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "none", "sampling": "unknown", "judge": "separate factual-support grader", "harness": "SpaceXAI internal single-turn factuality harness", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown" }, "notes": "card45 Single-turn hallucination Hallucination rate (%); source effort=max; benchmark protocol matches canonical=True. Unknown source fields remain unknown." } ] }, { "model_id": "claude-opus-4.8", "benchmark_id": "spacexai_mts_accuracy", "score": 55.8, "reference_url": "https://media.x.ai/v1/website/card-4p6-4cd2dc57.pdf", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "Grok Build coding and research tools", "sampling": "unknown", "judge": "task correctness under fixed rollout budget", "harness": "Grok Build", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "card46 SpaceXAI MTS Eval Correct (%); source effort=max; benchmark protocol matches canonical=True. Unknown source fields remain unknown." }, { "model_id": "claude-opus-4.8", "benchmark_id": "swe_atlas_codebase_qna", "score": 82.0, "reference_url": "https://media.x.ai/v1/website/4p5-5184fdf9.pdf", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "codebase reading and exploration", "sampling": "unknown", "judge": "repository-ground-truth rubric", "harness": "SWE-Atlas-QnA tracked split", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "card45 SWE-Atlas-QnA Accuracy (%); source effort=max; benchmark protocol matches canonical=True. Unknown source fields remain unknown." }, { "model_id": "claude-opus-4.8", "benchmark_id": "swe_marathon_v1_1", "score": 48.8, "reference_url": "https://media.x.ai/v1/website/card-4p6-4cd2dc57.pdf", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "agentic code execution", "sampling": "pass@1", "judge": "multi-layer resolution verification", "harness": "SWE-Marathon v1.1 harness", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "card46 SWE-Marathon v1.1 Resolution rate (%); source effort=max; benchmark protocol matches canonical=True. Unknown source fields remain unknown." }, { "model_id": "claude-opus-4.8", "benchmark_id": "terminal_bench_3_0", "score": 21.1, "reference_url": "https://media.x.ai/v1/website/card-4p6-4cd2dc57.pdf", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "container terminal", "sampling": "pass@1", "judge": "verified task-success evaluator", "harness": "Terminal-Bench 3.0 / FrontierBench fixed harness", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "card46 Terminal-Bench 3.0 Task success rate (%); source effort=max; benchmark protocol matches canonical=False. Unknown source fields remain unknown." }, { "model_id": "claude-opus-4.8", "benchmark_id": "xai_inferenceeval_accuracy", "score": 40.7, "reference_url": "https://media.x.ai/v1/website/card-4p6-4cd2dc57.pdf", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "Grok Build coding tools and GPU tests", "sampling": "unknown", "judge": "hidden GPU unit score gated by integration probe", "harness": "SpaceXAI InferenceEval harness", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "card46 InferenceEval Accuracy (%); source effort=max; benchmark protocol matches canonical=True. Unknown source fields remain unknown." }, { "model_id": "claude-opus-4.8", "benchmark_id": "xai_kernelbenchinternal_accuracy_efficiency", "score": 67.6, "reference_url": "https://media.x.ai/v1/website/card-4p6-4cd2dc57.pdf", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "sandboxed GPU coding tools", "sampling": "unknown", "judge": "correctness and measured speedup composite", "harness": "SpaceXAI KernelBenchInternal harness", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "card46 KernelBenchInternal Accuracy / efficiency (%); source effort=max; benchmark protocol matches canonical=True. Unknown source fields remain unknown." }, { "model_id": "claude-opus-5", "benchmark_id": "apex_agents", "score": 60.6, "reference_url": "https://media.x.ai/v1/website/card-4p6-4cd2dc57.pdf", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "multi-application professional tools", "sampling": "pass@1", "judge": "expert all-criteria-pass rubric", "harness": "Mercor APEX-Agents harness", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "card46 APEX-Agents Pass@1 (%); source effort=max; benchmark protocol matches canonical=True. Unknown source fields remain unknown." }, { "model_id": "claude-opus-5", "benchmark_id": "apex_swe", "score": 63.7, "reference_url": "https://media.x.ai/v1/website/card-4p6-4cd2dc57.pdf", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "agentic software-engineering environment", "sampling": "pass@1", "judge": "Mercor integration and observability task verifier", "harness": "Mercor APEX-SWE harness", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "card46 APEX-SWE Pass@1 (%); source effort=max; benchmark protocol matches canonical=True. Unknown source fields remain unknown." }, { "model_id": "claude-opus-5", "benchmark_id": "cadgenbench_generation_cad_score", "score": 36.6, "reference_url": "https://media.x.ai/v1/website/card-4p6-4cd2dc57.pdf", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "CAD generation agent tools", "sampling": "unknown", "judge": "generation CAD score", "harness": "CadGenBench generation harness", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "card46 CadGenBench Generation Reward (%); source effort=max; benchmark protocol matches canonical=True. Unknown source fields remain unknown." }, { "model_id": "claude-opus-5", "benchmark_id": "eebench_v1_core_reward", "score": 61.6, "reference_url": "https://media.x.ai/v1/website/card-4p6-4cd2dc57.pdf", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "electrical-engineering design tools", "sampling": "unknown", "judge": "physical correctness and functionality reward", "harness": "EEBench v1 core public leaderboard harness", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "card46 EEBench v1 core Reward (%); source effort=max; benchmark protocol matches canonical=True. Unknown source fields remain unknown." }, { "model_id": "claude-opus-5", "benchmark_id": "gdpval_aa_v2_elo", "score": 1849, "reference_url": "https://media.x.ai/v1/website/card-4p6-4cd2dc57.pdf", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "benchmark-specified knowledge-work tools", "sampling": "unknown", "judge": "pairwise Bradley-Terry Elo", "harness": "Artificial Analysis GDPval-AA v2 harness", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "card46 AA GDPVal v2 AA GDPVal Elo; source effort=max; benchmark protocol matches canonical=True. Unknown source fields remain unknown.", "candidates": [ { "score": 1852, "reference_url": "https://research.meta.ai/articles/introducing-muse-code-and-muse-spark-1-2/evaluations/gdpval-aa-v2-v1.png", "source_type": "official_blog", "reported_setting": { "mode": "thinking", "effort": "max", "prompt_style": "not disclosed", "temperature": "not disclosed", "context": "not disclosed", "agent": "Artificial Analysis Stirrup", "attempts": 1, "tools": "Web Fetch, Web Search, View Image, Code Exec, Finish, Abandon Task", "sampling": "one agentic submission per each of 220 tasks", "judge": "blind pairwise panel of three frontier LLM judges; Bradley-Terry Elo", "harness": "Artificial Analysis Stirrup; fresh E2B sandbox; 250-turn limit", "internet": "web fetch and Brave-backed web search available", "configuration": "fresh E2B sandbox per task; 250-turn limit" }, "notes": "Exact printed value from the locked Meta release chart; evaluation settings come from the locked official methodology." } ] }, { "model_id": "claude-opus-5", "benchmark_id": "harvey_lab_vals_final_score", "score": 11.7, "reference_url": "https://media.x.ai/v1/website/card-4p6-4cd2dc57.pdf", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "Vals legal-agent file tools", "sampling": "unknown", "judge": "expert all-pass final score", "harness": "Vals implementation of Harvey LAB", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "card46 Harvey Legal Agent Benchmark Final score (%); source effort=max; benchmark protocol matches canonical=True. Unknown source fields remain unknown." }, { "model_id": "claude-opus-5", "benchmark_id": "parametric_cad_bench_combined", "score": 90.6, "reference_url": "https://media.x.ai/v1/website/card-4p6-4cd2dc57.pdf", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "parametric CAD agent tools", "sampling": "unknown", "judge": "combined executability and geometric reward", "harness": "gNucleus Parametric CAD Bench harness", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "card46 Parametric CAD Bench Reward (%); source effort=max; benchmark protocol matches canonical=True. Unknown source fields remain unknown." }, { "model_id": "claude-opus-5", "benchmark_id": "spacexai_mts_accuracy", "score": 52.6, "reference_url": "https://media.x.ai/v1/website/card-4p6-4cd2dc57.pdf", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "Grok Build coding and research tools", "sampling": "unknown", "judge": "task correctness under fixed rollout budget", "harness": "Grok Build", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "card46 SpaceXAI MTS Eval Correct (%); source effort=max; benchmark protocol matches canonical=True. Unknown source fields remain unknown." }, { "model_id": "claude-opus-5", "benchmark_id": "swe_marathon_v1_1", "score": 50.0, "reference_url": "https://media.x.ai/v1/website/card-4p6-4cd2dc57.pdf", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "agentic code execution", "sampling": "pass@1", "judge": "multi-layer resolution verification", "harness": "SWE-Marathon v1.1 harness", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "card46 SWE-Marathon v1.1 Resolution rate (%); source effort=max; benchmark protocol matches canonical=True. Unknown source fields remain unknown." }, { "model_id": "claude-opus-5", "benchmark_id": "terminal_bench_3_0", "score": 43.5, "reference_url": "https://media.x.ai/v1/website/card-4p6-4cd2dc57.pdf", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "container terminal", "sampling": "pass@1", "judge": "verified task-success evaluator", "harness": "Terminal-Bench 3.0 / FrontierBench fixed harness", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "card46 Terminal-Bench 3.0 Task success rate (%); source effort=max; benchmark protocol matches canonical=False. Unknown source fields remain unknown." }, { "model_id": "claude-opus-5", "benchmark_id": "vals_index", "score": 74.8, "reference_url": "https://media.x.ai/v1/website/card-4p6-4cd2dc57.pdf", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "benchmark-specified professional tools", "sampling": "unknown", "judge": "Vals weighted suite score", "harness": "Vals Index live harness", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "card46 Vals Index Suite score (%); source effort=max; benchmark protocol matches canonical=True. Unknown source fields remain unknown." }, { "model_id": "claude-opus-5", "benchmark_id": "xai_inferenceeval_accuracy", "score": 43.0, "reference_url": "https://media.x.ai/v1/website/card-4p6-4cd2dc57.pdf", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "Grok Build coding tools and GPU tests", "sampling": "unknown", "judge": "hidden GPU unit score gated by integration probe", "harness": "SpaceXAI InferenceEval harness", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "card46 InferenceEval Accuracy (%); source effort=max; benchmark protocol matches canonical=True. Unknown source fields remain unknown." }, { "model_id": "claude-opus-5", "benchmark_id": "xai_kernelbenchinternal_accuracy_efficiency", "score": 73.0, "reference_url": "https://media.x.ai/v1/website/card-4p6-4cd2dc57.pdf", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "sandboxed GPU coding tools", "sampling": "unknown", "judge": "correctness and measured speedup composite", "harness": "SpaceXAI KernelBenchInternal harness", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "card46 KernelBenchInternal Accuracy / efficiency (%); source effort=max; benchmark protocol matches canonical=True. Unknown source fields remain unknown." }, { "model_id": "claude-sonnet-5", "benchmark_id": "apex_agents", "score": 32.5, "reference_url": "https://media.x.ai/v1/website/card-4p6-4cd2dc57.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "multi-application professional tools", "sampling": "pass@1", "judge": "expert all-criteria-pass rubric", "harness": "Mercor APEX-Agents harness", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "card46 APEX-Agents Pass@1 (%); source effort=high; benchmark protocol matches canonical=True. Unknown source fields remain unknown." }, { "model_id": "claude-sonnet-5", "benchmark_id": "apex_swe", "score": 43.6, "reference_url": "https://media.x.ai/v1/website/4p5-5184fdf9.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "agentic software-engineering environment", "sampling": "pass@1", "judge": "Mercor integration and observability task verifier", "harness": "Mercor APEX-SWE harness", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "card45 APEX-SWE Pass@1 (%); source effort=high; benchmark protocol matches canonical=True. Unknown source fields remain unknown.", "candidates": [ { "score": 46.4, "reference_url": "https://media.x.ai/v1/website/card-4p6-4cd2dc57.pdf", "source_type": "model_card", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "agentic software-engineering environment", "sampling": "pass@1", "judge": "Mercor integration and observability task verifier", "harness": "Mercor APEX-SWE harness", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown" }, "notes": "card46 APEX-SWE Pass@1 (%); source effort=max; benchmark protocol matches canonical=True. Unknown source fields remain unknown." } ] }, { "model_id": "claude-sonnet-5", "benchmark_id": "frontiercode_extended_v1_1", "score": 56.2, "reference_url": "https://media.x.ai/v1/website/card-4p6-4cd2dc57.pdf", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "fixed coding-agent scaffold", "sampling": "unknown", "judge": "maintainer-authored rubrics, verifiers, and hard blockers", "harness": "FrontierCode v1.1 Extended harness", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "card46 FrontierCode v1.1 Extended Extended score (%); source effort=max; benchmark protocol matches canonical=True. Unknown source fields remain unknown." }, { "model_id": "claude-sonnet-5", "benchmark_id": "parametric_cad_bench_combined", "score": 86.9, "reference_url": "https://media.x.ai/v1/website/4p5-5184fdf9.pdf", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "parametric CAD agent tools", "sampling": "unknown", "judge": "combined executability and geometric reward", "harness": "gNucleus Parametric CAD Bench harness", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "card45 Parametric CAD Bench Reward (%); source effort=max; benchmark protocol matches canonical=True. Unknown source fields remain unknown." }, { "model_id": "claude-sonnet-5", "benchmark_id": "swe_marathon_v1_1", "score": 30.0, "reference_url": "https://media.x.ai/v1/website/card-4p6-4cd2dc57.pdf", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "agentic code execution", "sampling": "pass@1", "judge": "multi-layer resolution verification", "harness": "SWE-Marathon v1.1 harness", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "card46 SWE-Marathon v1.1 Resolution rate (%); source effort=max; benchmark protocol matches canonical=True. Unknown source fields remain unknown." }, { "model_id": "claude-sonnet-5", "benchmark_id": "tau3_banking", "score": 28.2, "reference_url": "https://media.x.ai/v1/website/4p5-5184fdf9.pdf", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "banking APIs and simulated user", "sampling": "pass@1", "judge": "final-state task accuracy", "harness": "Artificial Analysis tau3-banking harness", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "card45 tau3-banking Accuracy (%); source effort=max; benchmark protocol matches canonical=True. Unknown source fields remain unknown." }, { "model_id": "claude-sonnet-5", "benchmark_id": "terminal_bench_3_0", "score": 14.6, "reference_url": "https://media.x.ai/v1/website/card-4p6-4cd2dc57.pdf", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "container terminal", "sampling": "pass@1", "judge": "verified task-success evaluator", "harness": "Terminal-Bench 3.0 / FrontierBench fixed harness", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "card46 Terminal-Bench 3.0 Task success rate (%); source effort=max; benchmark protocol matches canonical=False. Unknown source fields remain unknown." }, { "model_id": "claude-sonnet-5", "benchmark_id": "vals_index", "score": 68.6, "reference_url": "https://media.x.ai/v1/website/card-4p6-4cd2dc57.pdf", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "benchmark-specified professional tools", "sampling": "unknown", "judge": "Vals weighted suite score", "harness": "Vals Index live harness", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "card46 Vals Index Suite score (%); source effort=max; benchmark protocol matches canonical=True. Unknown source fields remain unknown." }, { "model_id": "glm-5.2", "benchmark_id": "apex_swe", "score": 37.3, "reference_url": "https://media.x.ai/v1/website/4p5-5184fdf9.pdf", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "agentic software-engineering environment", "sampling": "pass@1", "judge": "Mercor integration and observability task verifier", "harness": "Mercor APEX-SWE harness", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "card45 APEX-SWE Pass@1 (%); source effort=max; benchmark protocol matches canonical=True. Unknown source fields remain unknown." }, { "model_id": "glm-5.2", "benchmark_id": "falseclaimbench_accuracy", "score": 38.0, "reference_url": "https://media.x.ai/v1/website/4p5-5184fdf9.pdf", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "workspace editing and command tools", "sampling": "unknown", "judge": "claim-to-final-workspace-state verifier", "harness": "SpaceXAI internal FalseClaimBench harness", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "card45 FalseClaimBench Accuracy (%); source effort=max; benchmark protocol matches canonical=True. Unknown source fields remain unknown." }, { "model_id": "glm-5.2", "benchmark_id": "gdpval_aa_v2_elo", "score": 1514, "reference_url": "https://media.x.ai/v1/website/4p5-5184fdf9.pdf", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "benchmark-specified knowledge-work tools", "sampling": "unknown", "judge": "pairwise Bradley-Terry Elo", "harness": "Artificial Analysis GDPval-AA v2 harness", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "card45 GDPval-AA v2 AA score; source effort=max; benchmark protocol matches canonical=True. Unknown source fields remain unknown." }, { "model_id": "glm-5.2", "benchmark_id": "swe_atlas_codebase_qna", "score": 74.0, "reference_url": "https://media.x.ai/v1/website/4p5-5184fdf9.pdf", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "codebase reading and exploration", "sampling": "unknown", "judge": "repository-ground-truth rubric", "harness": "SWE-Atlas-QnA tracked split", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "card45 SWE-Atlas-QnA Accuracy (%); source effort=max; benchmark protocol matches canonical=True. Unknown source fields remain unknown." }, { "model_id": "gpt-5.5", "benchmark_id": "apex_swe", "score": 37.0, "reference_url": "https://media.x.ai/v1/website/card-4p6-4cd2dc57.pdf", "reported_setting": { "mode": "thinking", "effort": "xhigh", "tools": "agentic software-engineering environment", "sampling": "pass@1", "judge": "Mercor integration and observability task verifier", "harness": "Mercor APEX-SWE harness", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "card46 APEX-SWE Pass@1 (%); source effort=xhigh; benchmark protocol matches canonical=True. Unknown source fields remain unknown.", "candidates": [ { "score": 40.8, "reference_url": "https://media.x.ai/v1/website/4p5-5184fdf9.pdf", "source_type": "model_card", "reported_setting": { "mode": "thinking", "effort": "xhigh", "tools": "agentic software-engineering environment", "sampling": "pass@1", "judge": "Mercor integration and observability task verifier", "harness": "Mercor APEX-SWE harness", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown" }, "notes": "card45 APEX-SWE Pass@1 (%); source effort=xhigh; benchmark protocol matches canonical=True. Unknown source fields remain unknown." } ] }, { "model_id": "gpt-5.5", "benchmark_id": "deep_swe_v1_0", "score": 64.31, "reference_url": "https://x.ai/news/grok-4-5", "reported_setting": { "mode": "thinking", "effort": "xhigh", "tools": "agentic repository shell/editor", "sampling": "pass@1", "judge": "behavioral and correctness verifiers", "harness": "provider harnesses run by Artificial Analysis", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "release45 DeepSWE 1.0 Pass@1 (%); source effort=xhigh; benchmark protocol matches canonical=False. Unknown source fields remain unknown.", "candidates": [ { "score": 64.3, "reference_url": "https://media.x.ai/v1/website/4p5-5184fdf9.pdf", "source_type": "model_card", "reported_setting": { "mode": "thinking", "effort": "xhigh", "tools": "agentic repository shell/editor", "sampling": "pass@1", "judge": "behavioral and correctness verifiers", "harness": "provider harnesses run by Artificial Analysis", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown" }, "notes": "card45 DeepSWE v1.0 Pass@1 (%); source effort=xhigh; benchmark protocol matches canonical=False. Unknown source fields remain unknown." } ] }, { "model_id": "gpt-5.5", "benchmark_id": "deepsearchqa_acc", "score": 63.7, "reference_url": "https://media.x.ai/v1/website/card-4p6-4cd2dc57.pdf", "reported_setting": { "mode": "thinking", "effort": "xhigh", "tools": "web search agent", "sampling": "unknown", "judge": "DeepSearchQA answer-accuracy grader", "harness": "SpaceXAI internal DeepSearchQA implementation", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "card46 DeepSearchQA Accuracy (%); source effort=xhigh; benchmark protocol matches canonical=False. Unknown source fields remain unknown.", "candidates": [ { "score": 85.3, "reference_url": "https://static.stepfun.com/blog/step-3.7-flash/", "source_type": "official_blog", "reported_setting": { "quantization_config": "common release evaluation; artifact precision not established", "configuration": "base Step evaluation configuration", "mode": "source does not state", "effort": "source does not state", "tools_search": "web search/research agent", "shots": "source does not state", "trials_samples": "source does not state", "aggregation_pass_k": "source-reported score", "context": "source does not state", "output_cap": "source does not state", "temperature": "source does not state", "top_p": "source does not state", "judge": "Gemini 2.5 Flash with official Kaggle grading prompt", "harness_agent": "official StepFun evaluation; exact harness not stated", "dataset_version_split": "DeepSearchQA 900-prompt eval split", "multimodal_input": false }, "notes": "Official StepFun launch comparison table." } ] }, { "model_id": "gpt-5.5", "benchmark_id": "eebench_v1_core_reward", "score": 42.3, "reference_url": "https://media.x.ai/v1/website/card-4p6-4cd2dc57.pdf", "reported_setting": { "mode": "thinking", "effort": "xhigh", "tools": "electrical-engineering design tools", "sampling": "unknown", "judge": "physical correctness and functionality reward", "harness": "EEBench v1 core public leaderboard harness", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "card46 EEBench v1 core Reward (%); source effort=xhigh; benchmark protocol matches canonical=True. Unknown source fields remain unknown." }, { "model_id": "gpt-5.5", "benchmark_id": "frontiercode_extended_v1_1", "score": 56.7, "reference_url": "https://media.x.ai/v1/website/card-4p6-4cd2dc57.pdf", "reported_setting": { "mode": "thinking", "effort": "xhigh", "tools": "fixed coding-agent scaffold", "sampling": "unknown", "judge": "maintainer-authored rubrics, verifiers, and hard blockers", "harness": "FrontierCode v1.1 Extended harness", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "card46 FrontierCode v1.1 Extended Extended score (%); source effort=xhigh; benchmark protocol matches canonical=True. Unknown source fields remain unknown." }, { "model_id": "gpt-5.5", "benchmark_id": "hackerbench_v0_2_benign_refusal", "score": 1.1, "reference_url": "https://media.x.ai/v1/website/card-4p6-4cd2dc57.pdf", "reported_setting": { "mode": "thinking", "effort": "xhigh", "tools": "constrained cyber task tools", "sampling": "unknown", "judge": "benign refusal grader", "harness": "HackerBench v0.2 with standard safeguards", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "card46 HackerBench v0.2 Benign Refusal (%); source effort=xhigh; benchmark protocol matches canonical=True. Unknown source fields remain unknown." }, { "model_id": "gpt-5.5", "benchmark_id": "hackerbench_v0_2_harmful_compliance", "score": 25.0, "reference_url": "https://media.x.ai/v1/website/card-4p6-4cd2dc57.pdf", "reported_setting": { "mode": "thinking", "effort": "xhigh", "tools": "constrained cyber task tools", "sampling": "unknown", "judge": "harmful or dual-use compliance grader", "harness": "HackerBench v0.2 with standard safeguards", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "card46 HackerBench v0.2 Harmful/Dual-Use Compliance (%); source effort=xhigh; benchmark protocol matches canonical=True. Unknown source fields remain unknown." }, { "model_id": "gpt-5.5", "benchmark_id": "single_turn_hallucination_rate", "score": 1.1, "reference_url": "https://media.x.ai/v1/website/card-4p6-4cd2dc57.pdf", "reported_setting": { "mode": "thinking", "effort": "xhigh", "tools": "none", "sampling": "unknown", "judge": "separate factual-support grader", "harness": "SpaceXAI internal single-turn factuality harness", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "card46 Single-turn hallucination Hallucination rate (%); source effort=xhigh; benchmark protocol matches canonical=True. Unknown source fields remain unknown.", "candidates": [ { "score": 1.14, "reference_url": "https://media.x.ai/v1/website/4p5-5184fdf9.pdf", "source_type": "model_card", "reported_setting": { "mode": "thinking", "effort": "xhigh", "tools": "none", "sampling": "unknown", "judge": "separate factual-support grader", "harness": "SpaceXAI internal single-turn factuality harness", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown" }, "notes": "card45 Single-turn hallucination Hallucination rate (%); source effort=xhigh; benchmark protocol matches canonical=True. Unknown source fields remain unknown." } ] }, { "model_id": "gpt-5.5", "benchmark_id": "spacexai_mts_accuracy", "score": 46.4, "reference_url": "https://media.x.ai/v1/website/card-4p6-4cd2dc57.pdf", "reported_setting": { "mode": "thinking", "effort": "xhigh", "tools": "Grok Build coding and research tools", "sampling": "unknown", "judge": "task correctness under fixed rollout budget", "harness": "Grok Build", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "card46 SpaceXAI MTS Eval Correct (%); source effort=xhigh; benchmark protocol matches canonical=True. Unknown source fields remain unknown." }, { "model_id": "gpt-5.5", "benchmark_id": "swe_atlas_codebase_qna", "score": 81.0, "reference_url": "https://media.x.ai/v1/website/4p5-5184fdf9.pdf", "reported_setting": { "mode": "thinking", "effort": "xhigh", "tools": "codebase reading and exploration", "sampling": "unknown", "judge": "repository-ground-truth rubric", "harness": "SWE-Atlas-QnA tracked split", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "card45 SWE-Atlas-QnA Accuracy (%); source effort=xhigh; benchmark protocol matches canonical=True. Unknown source fields remain unknown." }, { "model_id": "gpt-5.5", "benchmark_id": "xai_inferenceeval_accuracy", "score": 38.9, "reference_url": "https://media.x.ai/v1/website/card-4p6-4cd2dc57.pdf", "reported_setting": { "mode": "thinking", "effort": "xhigh", "tools": "Grok Build coding tools and GPU tests", "sampling": "unknown", "judge": "hidden GPU unit score gated by integration probe", "harness": "SpaceXAI InferenceEval harness", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "card46 InferenceEval Accuracy (%); source effort=xhigh; benchmark protocol matches canonical=True. Unknown source fields remain unknown." }, { "model_id": "gpt-5.5", "benchmark_id": "xai_securecodereview_reward", "score": 64.1, "reference_url": "https://media.x.ai/v1/website/card-4p6-4cd2dc57.pdf", "reported_setting": { "mode": "thinking", "effort": "xhigh", "tools": "fixed secure-code-review tools", "sampling": "unknown", "judge": "security-fix reward with regression checks", "harness": "SpaceXAI SecureCodeReview harness", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "card46 SecureCodeReview Reward (%); source effort=xhigh; benchmark protocol matches canonical=True. Unknown source fields remain unknown." }, { "model_id": "gpt-5.6-sol", "benchmark_id": "apex_swe", "score": 45.8, "reference_url": "https://media.x.ai/v1/website/card-4p6-4cd2dc57.pdf", "reported_setting": { "mode": "thinking", "effort": "xhigh", "tools": "agentic software-engineering environment", "sampling": "pass@1", "judge": "Mercor integration and observability task verifier", "harness": "Mercor APEX-SWE harness", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "card46 APEX-SWE Pass@1 (%); source effort=xhigh; benchmark protocol matches canonical=True. Unknown source fields remain unknown.", "candidates": [ { "score": 39.7, "reference_url": "https://media.x.ai/v1/website/4p5-5184fdf9.pdf", "source_type": "model_card", "reported_setting": { "mode": "thinking", "effort": "xhigh", "tools": "agentic software-engineering environment", "sampling": "pass@1", "judge": "Mercor integration and observability task verifier", "harness": "Mercor APEX-SWE harness", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown" }, "notes": "card45 APEX-SWE Pass@1 (%); source effort=xhigh; benchmark protocol matches canonical=True. Unknown source fields remain unknown." } ] }, { "model_id": "gpt-5.6-sol", "benchmark_id": "cadgenbench_generation_cad_score", "score": 37.1, "reference_url": "https://media.x.ai/v1/website/card-4p6-4cd2dc57.pdf", "reported_setting": { "mode": "thinking", "effort": "xhigh", "tools": "CAD generation agent tools", "sampling": "unknown", "judge": "generation CAD score", "harness": "CadGenBench generation harness", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "card46 CadGenBench Generation Reward (%); source effort=xhigh; benchmark protocol matches canonical=True. Unknown source fields remain unknown." }, { "model_id": "gpt-5.6-sol", "benchmark_id": "cybergym", "score": 83.6, "reference_url": "https://media.x.ai/v1/website/card-4p6-4cd2dc57.pdf", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "unrestricted agentic code execution", "sampling": "unknown", "judge": "PoC reproduced on vulnerable but not fixed version", "harness": "CyberGym unrestricted capability harness", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "card46 CyberGym Mean Reproduced (%); source effort=max; benchmark protocol matches canonical=True. Unknown source fields remain unknown." }, { "model_id": "gpt-5.6-sol", "benchmark_id": "deepsearchqa_acc", "score": 75.0, "reference_url": "https://media.x.ai/v1/website/card-4p6-4cd2dc57.pdf", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "web search agent", "sampling": "unknown", "judge": "DeepSearchQA answer-accuracy grader", "harness": "SpaceXAI internal DeepSearchQA implementation", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "card46 DeepSearchQA Accuracy (%); source effort=max; benchmark protocol matches canonical=False. Unknown source fields remain unknown." }, { "model_id": "gpt-5.6-sol", "benchmark_id": "eebench_v1_core_reward", "score": 39.4, "reference_url": "https://media.x.ai/v1/website/card-4p6-4cd2dc57.pdf", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "electrical-engineering design tools", "sampling": "unknown", "judge": "physical correctness and functionality reward", "harness": "EEBench v1 core public leaderboard harness", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "card46 EEBench v1 core Reward (%); source effort=max; benchmark protocol matches canonical=True. Unknown source fields remain unknown." }, { "model_id": "gpt-5.6-sol", "benchmark_id": "gdpval_aa_v2_elo", "score": 1728, "reference_url": "https://x.ai/news/grok-4-6", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "benchmark-specified knowledge-work tools", "sampling": "unknown", "judge": "pairwise Bradley-Terry Elo", "harness": "Artificial Analysis GDPval-AA v2 harness", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "release46 GDPVal-AA v2 Elo; source effort=max; benchmark protocol matches canonical=True. Unknown source fields remain unknown.", "candidates": [ { "score": 1743, "reference_url": "https://media.x.ai/v1/website/4p5-5184fdf9.pdf", "source_type": "model_card", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "benchmark-specified knowledge-work tools", "sampling": "unknown", "judge": "pairwise Bradley-Terry Elo", "harness": "Artificial Analysis GDPval-AA v2 harness", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown" }, "notes": "card45 GDPval-AA v2 AA score; source effort=max; benchmark protocol matches canonical=True. Unknown source fields remain unknown." } ] }, { "model_id": "gpt-5.6-sol", "benchmark_id": "hackerbench_v0_2_benign_refusal", "score": 0.0, "reference_url": "https://media.x.ai/v1/website/card-4p6-4cd2dc57.pdf", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "constrained cyber task tools", "sampling": "unknown", "judge": "benign refusal grader", "harness": "HackerBench v0.2 with standard safeguards", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "card46 HackerBench v0.2 Benign Refusal (%); source effort=max; benchmark protocol matches canonical=True. Unknown source fields remain unknown." }, { "model_id": "gpt-5.6-sol", "benchmark_id": "hackerbench_v0_2_harmful_compliance", "score": 35.7, "reference_url": "https://media.x.ai/v1/website/card-4p6-4cd2dc57.pdf", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "constrained cyber task tools", "sampling": "unknown", "judge": "harmful or dual-use compliance grader", "harness": "HackerBench v0.2 with standard safeguards", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "card46 HackerBench v0.2 Harmful/Dual-Use Compliance (%); source effort=max; benchmark protocol matches canonical=True. Unknown source fields remain unknown." }, { "model_id": "gpt-5.6-sol", "benchmark_id": "harvey_lab_vals_final_score", "score": 2.5, "reference_url": "https://x.ai/news/grok-4-6", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "Vals legal-agent file tools", "sampling": "unknown", "judge": "expert all-pass final score", "harness": "Vals implementation of Harvey LAB", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "release46 Harvey LAB (Vals) Final score (%); source effort=max; benchmark protocol matches canonical=True. Unknown source fields remain unknown." }, { "model_id": "gpt-5.6-sol", "benchmark_id": "parametric_cad_bench_combined", "score": 86.5, "reference_url": "https://media.x.ai/v1/website/card-4p6-4cd2dc57.pdf", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "parametric CAD agent tools", "sampling": "unknown", "judge": "combined executability and geometric reward", "harness": "gNucleus Parametric CAD Bench harness", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "card46 Parametric CAD Bench Reward (%); source effort=max; benchmark protocol matches canonical=True. Unknown source fields remain unknown.", "candidates": [ { "score": 84.9, "reference_url": "https://media.x.ai/v1/website/4p5-5184fdf9.pdf", "source_type": "model_card", "reported_setting": { "mode": "thinking", "effort": "xhigh", "tools": "parametric CAD agent tools", "sampling": "unknown", "judge": "combined executability and geometric reward", "harness": "gNucleus Parametric CAD Bench harness", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown" }, "notes": "card45 Parametric CAD Bench Reward (%); source effort=xhigh; benchmark protocol matches canonical=True. Unknown source fields remain unknown." } ] }, { "model_id": "gpt-5.6-sol", "benchmark_id": "spacexai_mts_accuracy", "score": 52.1, "reference_url": "https://media.x.ai/v1/website/card-4p6-4cd2dc57.pdf", "reported_setting": { "mode": "thinking", "effort": "xhigh", "tools": "Grok Build coding and research tools", "sampling": "unknown", "judge": "task correctness under fixed rollout budget", "harness": "Grok Build", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "card46 SpaceXAI MTS Eval Correct (%); source effort=xhigh; benchmark protocol matches canonical=True. Unknown source fields remain unknown." }, { "model_id": "gpt-5.6-sol", "benchmark_id": "swe_atlas_codebase_qna", "score": 84.0, "reference_url": "https://media.x.ai/v1/website/4p5-5184fdf9.pdf", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "codebase reading and exploration", "sampling": "unknown", "judge": "repository-ground-truth rubric", "harness": "SWE-Atlas-QnA tracked split", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "card45 SWE-Atlas-QnA Accuracy (%); source effort=max; benchmark protocol matches canonical=True. Unknown source fields remain unknown." }, { "model_id": "gpt-5.6-sol", "benchmark_id": "swe_marathon_v1_1", "score": 42.5, "reference_url": "https://media.x.ai/v1/website/card-4p6-4cd2dc57.pdf", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "agentic code execution", "sampling": "pass@1", "judge": "multi-layer resolution verification", "harness": "SWE-Marathon v1.1 harness", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "card46 SWE-Marathon v1.1 Resolution rate (%); source effort=max; benchmark protocol matches canonical=True. Unknown source fields remain unknown." }, { "model_id": "gpt-5.6-sol", "benchmark_id": "terminal_bench_3_0", "score": 34.6, "reference_url": "https://x.ai/news/grok-4-6", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "container terminal", "sampling": "pass@1", "judge": "verified task-success evaluator", "harness": "Terminal-Bench 3.0 / FrontierBench fixed harness", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "release46 Terminal-Bench v3.0 Task success rate (%); source effort=max; benchmark protocol matches canonical=False. Unknown source fields remain unknown." }, { "model_id": "gpt-5.6-sol", "benchmark_id": "xai_inferenceeval_accuracy", "score": 44.1, "reference_url": "https://media.x.ai/v1/website/card-4p6-4cd2dc57.pdf", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "Grok Build coding tools and GPU tests", "sampling": "unknown", "judge": "hidden GPU unit score gated by integration probe", "harness": "SpaceXAI InferenceEval harness", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "card46 InferenceEval Accuracy (%); source effort=max; benchmark protocol matches canonical=True. Unknown source fields remain unknown." }, { "model_id": "gpt-5.6-sol", "benchmark_id": "xai_kernelbenchinternal_accuracy_efficiency", "score": 73.1, "reference_url": "https://media.x.ai/v1/website/card-4p6-4cd2dc57.pdf", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "sandboxed GPU coding tools", "sampling": "unknown", "judge": "correctness and measured speedup composite", "harness": "SpaceXAI KernelBenchInternal harness", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "card46 KernelBenchInternal Accuracy / efficiency (%); source effort=max; benchmark protocol matches canonical=True. Unknown source fields remain unknown." }, { "model_id": "gpt-5.6-sol", "benchmark_id": "xai_securecodereview_reward", "score": 57.5, "reference_url": "https://media.x.ai/v1/website/card-4p6-4cd2dc57.pdf", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "fixed secure-code-review tools", "sampling": "unknown", "judge": "security-fix reward with regression checks", "harness": "SpaceXAI SecureCodeReview harness", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "card46 SecureCodeReview Reward (%); source effort=max; benchmark protocol matches canonical=True. Unknown source fields remain unknown." }, { "model_id": "gpt-5.6-terra", "benchmark_id": "gdpval_aa_v2_elo", "score": 1578, "reference_url": "https://media.x.ai/v1/website/card-4p6-4cd2dc57.pdf", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "benchmark-specified knowledge-work tools", "sampling": "unknown", "judge": "pairwise Bradley-Terry Elo", "harness": "Artificial Analysis GDPval-AA v2 harness", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "card46 AA GDPVal v2 AA GDPVal Elo; source effort=max; benchmark protocol matches canonical=True. Unknown source fields remain unknown.", "candidates": [ { "score": 1577, "reference_url": "https://research.meta.ai/articles/introducing-muse-code-and-muse-spark-1-2/evaluations/gdpval-aa-v2-v1.png", "source_type": "official_blog", "reported_setting": { "mode": "thinking", "effort": "max", "prompt_style": "not disclosed", "temperature": "not disclosed", "context": "not disclosed", "agent": "Artificial Analysis Stirrup", "attempts": 1, "tools": "Web Fetch, Web Search, View Image, Code Exec, Finish, Abandon Task", "sampling": "one agentic submission per each of 220 tasks", "judge": "blind pairwise panel of three frontier LLM judges; Bradley-Terry Elo", "harness": "Artificial Analysis Stirrup; fresh E2B sandbox; 250-turn limit", "internet": "web fetch and Brave-backed web search available", "configuration": "fresh E2B sandbox per task; 250-turn limit" }, "notes": "Exact printed value from the locked Meta release chart; evaluation settings come from the locked official methodology." } ] }, { "model_id": "gpt-5.6-terra", "benchmark_id": "swe_atlas_codebase_qna", "score": 81.0, "reference_url": "https://media.x.ai/v1/website/4p5-5184fdf9.pdf", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "codebase reading and exploration", "sampling": "unknown", "judge": "repository-ground-truth rubric", "harness": "SWE-Atlas-QnA tracked split", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "card45 SWE-Atlas-QnA Accuracy (%); source effort=max; benchmark protocol matches canonical=True. Unknown source fields remain unknown." }, { "model_id": "gpt-5.6-terra", "benchmark_id": "tau3_banking", "score": 31.8, "reference_url": "https://media.x.ai/v1/website/4p5-5184fdf9.pdf", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "banking APIs and simulated user", "sampling": "pass@1", "judge": "final-state task accuracy", "harness": "Artificial Analysis tau3-banking harness", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "card45 tau3-banking Accuracy (%); source effort=max; benchmark protocol matches canonical=True. Unknown source fields remain unknown." }, { "model_id": "gpt-5.6-terra", "benchmark_id": "vals_index", "score": 65.1, "reference_url": "https://media.x.ai/v1/website/card-4p6-4cd2dc57.pdf", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "benchmark-specified professional tools", "sampling": "unknown", "judge": "Vals weighted suite score", "harness": "Vals Index live harness", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "card46 Vals Index Suite score (%); source effort=max; benchmark protocol matches canonical=True. Unknown source fields remain unknown." }, { "model_id": "grok-4.3", "benchmark_id": "gdpval_aa_v2_elo", "score": 1085, "reference_url": "https://media.x.ai/v1/website/card-4p6-4cd2dc57.pdf", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "benchmark-specified knowledge-work tools", "sampling": "unknown", "judge": "pairwise Bradley-Terry Elo", "harness": "Artificial Analysis GDPval-AA v2 harness", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "card46 AA GDPVal v2 AA GDPVal Elo; source effort=unknown; benchmark protocol matches canonical=True. Unknown source fields remain unknown.", "candidates": [ { "score": 1085, "reference_url": "https://media.x.ai/v1/website/4p5-5184fdf9.pdf", "source_type": "model_card", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "benchmark-specified knowledge-work tools", "sampling": "unknown", "judge": "pairwise Bradley-Terry Elo", "harness": "Artificial Analysis GDPval-AA v2 harness", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown" }, "notes": "card45 GDPval-AA v2 AA score; source effort=high; benchmark protocol matches canonical=True. Unknown source fields remain unknown." } ] }, { "model_id": "grok-4.3", "benchmark_id": "spacexai_mts_accuracy", "score": 40.6, "reference_url": "https://media.x.ai/v1/website/card-4p6-4cd2dc57.pdf", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "Grok Build coding and research tools", "sampling": "unknown", "judge": "task correctness under fixed rollout budget", "harness": "Grok Build", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "card46 SpaceXAI MTS Eval Correct (%); source effort=unknown; benchmark protocol matches canonical=True. Unknown source fields remain unknown." }, { "model_id": "grok-4.5", "benchmark_id": "aa_briefcase_elo", "score": 1313, "reference_url": "https://x.ai/news/grok-4-6", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "offline multi-file knowledge-work tools", "sampling": "unknown", "judge": "rubric plus analytical and presentation pairwise Elo", "harness": "Artificial Analysis AA-Briefcase harness", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "release46 AA-Briefcase Elo; source effort=high; benchmark protocol matches canonical=True. Unknown source fields remain unknown." }, { "model_id": "grok-4.5", "benchmark_id": "aa_intelligence_index_v4_1", "score": 56, "reference_url": "https://x.ai/news/grok-4-6", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "benchmark-specified", "sampling": "unknown", "judge": "Artificial Analysis weighted composite", "harness": "Artificial Analysis live v4.1 harness", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "release46 AA Intelligence Index Composite score; source effort=high; benchmark protocol matches canonical=True. Unknown source fields remain unknown." }, { "model_id": "grok-4.5", "benchmark_id": "apex_agents", "score": 47.1, "reference_url": "https://x.ai/news/grok-4-6", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "multi-application professional tools", "sampling": "pass@1", "judge": "expert all-criteria-pass rubric", "harness": "Mercor APEX-Agents harness", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "release46 APEX-Agents Pass@1 (%); source effort=high; benchmark protocol matches canonical=True. Unknown source fields remain unknown." }, { "model_id": "grok-4.5", "benchmark_id": "apex_swe", "score": 53.6, "reference_url": "https://x.ai/news/grok-4-6", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "agentic software-engineering environment", "sampling": "pass@1", "judge": "Mercor integration and observability task verifier", "harness": "Mercor APEX-SWE harness", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "release46 APEX-SWE Pass@1 (%); source effort=high; benchmark protocol matches canonical=True. Unknown source fields remain unknown.", "candidates": [ { "score": 51.2, "reference_url": "https://media.x.ai/v1/website/4p5-5184fdf9.pdf", "source_type": "model_card", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "agentic software-engineering environment", "sampling": "pass@1", "judge": "Mercor integration and observability task verifier", "harness": "Mercor APEX-SWE harness", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown" }, "notes": "card45 APEX-SWE Pass@1 (%); source effort=high; benchmark protocol matches canonical=True. Unknown source fields remain unknown." } ] }, { "model_id": "grok-4.5", "benchmark_id": "bixbench", "score": 93.8, "reference_url": "https://media.x.ai/v1/website/4p5-5184fdf9.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "analysis tools enabled by default", "sampling": "unknown", "judge": "zero-shot multiple-choice accuracy", "harness": "Grok Build default tool-on BixBench harness", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "card45 BixBench zero-shot MCQ Accuracy (%); source effort=high; benchmark protocol matches canonical=True. Unknown source fields remain unknown." }, { "model_id": "grok-4.5", "benchmark_id": "cadgenbench_generation_cad_score", "score": 33.2, "reference_url": "https://media.x.ai/v1/website/card-4p6-4cd2dc57.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "CAD generation agent tools", "sampling": "unknown", "judge": "generation CAD score", "harness": "CadGenBench generation harness", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "card46 CadGenBench Generation Reward (%); source effort=high; benchmark protocol matches canonical=True. Unknown source fields remain unknown." }, { "model_id": "grok-4.5", "benchmark_id": "cursorbench_3_2", "score": 66.7, "reference_url": "https://x.ai/news/grok-4-6", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "Cursor coding-agent tools", "sampling": "unknown", "judge": "CursorBench task score", "harness": "Cursor first-party agent harness", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "release46 CursorBench v3.2 Score (%); source effort=high; benchmark protocol matches canonical=True. Unknown source fields remain unknown." }, { "model_id": "grok-4.5", "benchmark_id": "cve_bench_reward", "score": 35.2, "reference_url": "https://media.x.ai/v1/website/card-4p6-4cd2dc57.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "unrestricted sandboxed web-application tools", "sampling": "unknown", "judge": "CVE-Bench exploit reward", "harness": "CVE-Bench sandbox", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "card46 CVE-Bench Reward (%); source effort=high; benchmark protocol matches canonical=True. Unknown source fields remain unknown." }, { "model_id": "grok-4.5", "benchmark_id": "cybergym", "score": 79.0, "reference_url": "https://media.x.ai/v1/website/card-4p6-4cd2dc57.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "unrestricted agentic code execution", "sampling": "unknown", "judge": "PoC reproduced on vulnerable but not fixed version", "harness": "CyberGym unrestricted capability harness", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "card46 CyberGym Mean Reproduced (%); source effort=high; benchmark protocol matches canonical=True. Unknown source fields remain unknown.", "candidates": [ { "score": 80.4, "reference_url": "https://media.x.ai/v1/website/4p5-5184fdf9.pdf", "source_type": "model_card", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "unrestricted agentic code execution", "sampling": "unknown", "judge": "PoC reproduced on vulnerable but not fixed version", "harness": "CyberGym unrestricted capability harness", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown" }, "notes": "card45 CyberGym Mean Reproduced (%); source effort=high; benchmark protocol matches canonical=True. Unknown source fields remain unknown." } ] }, { "model_id": "grok-4.5", "benchmark_id": "deep_swe_v1_0", "score": 62.0, "reference_url": "https://x.ai/news/grok-4-5", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "agentic repository shell/editor", "sampling": "pass@1", "judge": "behavioral and correctness verifiers", "harness": "provider harnesses run by Artificial Analysis", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "release45 DeepSWE 1.0 Pass@1 (%); source effort=unknown; benchmark protocol matches canonical=False. Unknown source fields remain unknown.", "candidates": [ { "score": 62.0, "reference_url": "https://media.x.ai/v1/website/4p5-5184fdf9.pdf", "source_type": "model_card", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "agentic repository shell/editor", "sampling": "pass@1", "judge": "behavioral and correctness verifiers", "harness": "provider harnesses run by Artificial Analysis", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown" }, "notes": "card45 DeepSWE v1.0 Pass@1 (%); source effort=high; benchmark protocol matches canonical=False. Unknown source fields remain unknown." } ] }, { "model_id": "grok-4.5", "benchmark_id": "deepsearchqa_acc", "score": 38.4, "reference_url": "https://media.x.ai/v1/website/card-4p6-4cd2dc57.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "web search agent", "sampling": "unknown", "judge": "DeepSearchQA answer-accuracy grader", "harness": "SpaceXAI internal DeepSearchQA implementation", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "card46 DeepSearchQA Accuracy (%); source effort=high; benchmark protocol matches canonical=False. Unknown source fields remain unknown." }, { "model_id": "grok-4.5", "benchmark_id": "eebench_v1_core_reward", "score": 50.9, "reference_url": "https://media.x.ai/v1/website/card-4p6-4cd2dc57.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "electrical-engineering design tools", "sampling": "unknown", "judge": "physical correctness and functionality reward", "harness": "EEBench v1 core public leaderboard harness", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "card46 EEBench v1 core Reward (%); source effort=high; benchmark protocol matches canonical=True. Unknown source fields remain unknown." }, { "model_id": "grok-4.5", "benchmark_id": "falseclaimbench_accuracy", "score": 84.0, "reference_url": "https://media.x.ai/v1/website/4p5-5184fdf9.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "workspace editing and command tools", "sampling": "unknown", "judge": "claim-to-final-workspace-state verifier", "harness": "SpaceXAI internal FalseClaimBench harness", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "card45 FalseClaimBench Accuracy (%); source effort=high; benchmark protocol matches canonical=True. Unknown source fields remain unknown." }, { "model_id": "grok-4.5", "benchmark_id": "frontier_swe", "score": 78.0, "reference_url": "https://media.x.ai/v1/website/4p5-5184fdf9.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "agentic code execution", "sampling": "mean@5", "judge": "continuous partial-credit task scoring and dominance", "harness": "public FrontierSWE harness; Grok CLI for Grok 4.5", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "card45 FrontierSWE Dominance (%); source effort=high; benchmark protocol matches canonical=True. Unknown source fields remain unknown." }, { "model_id": "grok-4.5", "benchmark_id": "frontiercode_extended_v1_1", "score": 56.6, "reference_url": "https://x.ai/news/grok-4-6", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "fixed coding-agent scaffold", "sampling": "unknown", "judge": "maintainer-authored rubrics, verifiers, and hard blockers", "harness": "FrontierCode v1.1 Extended harness", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "release46 FrontierCode v1.1 Extended Extended score (%); source effort=high; benchmark protocol matches canonical=True. Unknown source fields remain unknown." }, { "model_id": "grok-4.5", "benchmark_id": "hackerbench_v0_2_benign_refusal", "score": 0.0, "reference_url": "https://media.x.ai/v1/website/card-4p6-4cd2dc57.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "constrained cyber task tools", "sampling": "unknown", "judge": "benign refusal grader", "harness": "HackerBench v0.2 with standard safeguards", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "card46 HackerBench v0.2 Benign Refusal (%); source effort=high; benchmark protocol matches canonical=True. Unknown source fields remain unknown." }, { "model_id": "grok-4.5", "benchmark_id": "hackerbench_v0_2_harmful_compliance", "score": 7.8, "reference_url": "https://media.x.ai/v1/website/card-4p6-4cd2dc57.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "constrained cyber task tools", "sampling": "unknown", "judge": "harmful or dual-use compliance grader", "harness": "HackerBench v0.2 with standard safeguards", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "card46 HackerBench v0.2 Harmful/Dual-Use Compliance (%); source effort=high; benchmark protocol matches canonical=True. Unknown source fields remain unknown." }, { "model_id": "grok-4.5", "benchmark_id": "harvey_lab_vals_final_score", "score": 12.9, "reference_url": "https://x.ai/news/grok-4-6", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "Vals legal-agent file tools", "sampling": "unknown", "judge": "expert all-pass final score", "harness": "Vals implementation of Harvey LAB", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "release46 Harvey LAB (Vals) Final score (%); source effort=high; benchmark protocol matches canonical=True. Unknown source fields remain unknown." }, { "model_id": "grok-4.5", "benchmark_id": "lab_bench_protocolqa_accuracy", "score": 87.0, "reference_url": "https://media.x.ai/v1/website/card-4p6-4cd2dc57.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none; no safeguards", "sampling": "unknown", "judge": "open-ended single-error accuracy grader", "harness": "LAB-Bench ProtocolQA harness", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "card46 ProtocolQA Open-Ended Accuracy (%); source effort=high; benchmark protocol matches canonical=False. Unknown source fields remain unknown." }, { "model_id": "grok-4.5", "benchmark_id": "parametric_cad_bench_combined", "score": 83.7, "reference_url": "https://media.x.ai/v1/website/card-4p6-4cd2dc57.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "parametric CAD agent tools", "sampling": "unknown", "judge": "combined executability and geometric reward", "harness": "gNucleus Parametric CAD Bench harness", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "card46 Parametric CAD Bench Reward (%); source effort=high; benchmark protocol matches canonical=True. Unknown source fields remain unknown.", "candidates": [ { "score": 88.1, "reference_url": "https://media.x.ai/v1/website/4p5-5184fdf9.pdf", "source_type": "model_card", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "parametric CAD agent tools", "sampling": "unknown", "judge": "combined executability and geometric reward", "harness": "gNucleus Parametric CAD Bench harness", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown" }, "notes": "card45 Parametric CAD Bench Reward (%); source effort=high; benchmark protocol matches canonical=True. Unknown source fields remain unknown." } ] }, { "model_id": "grok-4.5", "benchmark_id": "program_bench", "score": 57.2, "reference_url": "https://media.x.ai/v1/website/4p5-5184fdf9.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "compiled executable and documentation; no internet or decompilation", "sampling": "unknown", "judge": "248,000+ fuzz-generated behavioral tests", "harness": "official ProgramBench sandbox", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "card45 ProgramBench Tests-passed rate (%); source effort=high; benchmark protocol matches canonical=True. Unknown source fields remain unknown." }, { "model_id": "grok-4.5", "benchmark_id": "single_turn_hallucination_rate", "score": 0.98, "reference_url": "https://media.x.ai/v1/website/card-4p6-4cd2dc57.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "unknown", "judge": "separate factual-support grader", "harness": "SpaceXAI internal single-turn factuality harness", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "card46 Single-turn hallucination Hallucination rate (%); source effort=high; benchmark protocol matches canonical=True. Unknown source fields remain unknown." }, { "model_id": "grok-4.5", "benchmark_id": "spacexai_mts_accuracy", "score": 57.1, "reference_url": "https://media.x.ai/v1/website/card-4p6-4cd2dc57.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "Grok Build coding and research tools", "sampling": "unknown", "judge": "task correctness under fixed rollout budget", "harness": "Grok Build", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "card46 SpaceXAI MTS Eval Correct (%); source effort=high; benchmark protocol matches canonical=True. Unknown source fields remain unknown." }, { "model_id": "grok-4.5", "benchmark_id": "swe_atlas_codebase_qna", "score": 84.0, "reference_url": "https://media.x.ai/v1/website/4p5-5184fdf9.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "codebase reading and exploration", "sampling": "unknown", "judge": "repository-ground-truth rubric", "harness": "SWE-Atlas-QnA tracked split", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "card45 SWE-Atlas-QnA Accuracy (%); source effort=high; benchmark protocol matches canonical=True. Unknown source fields remain unknown." }, { "model_id": "grok-4.5", "benchmark_id": "swe_bench_multilingual", "score": 78.0, "reference_url": "https://media.x.ai/v1/website/4p5-5184fdf9.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "agentic repository shell/editor", "sampling": "pass@1", "judge": "language-appropriate isolated tests", "harness": "fixed SWE-bench Multilingual agent scaffold", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "card45 SWE-Bench Multilingual Resolve rate (%); source effort=high; benchmark protocol matches canonical=True. Unknown source fields remain unknown." }, { "model_id": "grok-4.5", "benchmark_id": "swe_marathon_v1_0", "score": 29.0, "reference_url": "https://media.x.ai/v1/website/4p5-5184fdf9.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "agentic code execution", "sampling": "pass@1", "judge": "multi-layer resolution verification", "harness": "SWE-Marathon v1.0 harness", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "card45 SWE-Marathon v1.0 Resolution rate (%); source effort=high; benchmark protocol matches canonical=True. Unknown source fields remain unknown.", "candidates": [ { "score": 29.0, "reference_url": "https://x.ai/news/grok-4-5", "source_type": "official_blog", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "agentic code execution", "sampling": "pass@1", "judge": "multi-layer resolution verification", "harness": "SWE-Marathon v1.0 harness", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown" }, "notes": "release45 SWE-Marathon v1.0 Resolution rate (%); source effort=unknown; benchmark protocol matches canonical=True. Unknown source fields remain unknown." } ] }, { "model_id": "grok-4.5", "benchmark_id": "swe_marathon_v1_1", "score": 29.4, "reference_url": "https://media.x.ai/v1/website/card-4p6-4cd2dc57.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "agentic code execution", "sampling": "pass@1", "judge": "multi-layer resolution verification", "harness": "SWE-Marathon v1.1 harness", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "card46 SWE-Marathon v1.1 Resolution rate (%); source effort=high; benchmark protocol matches canonical=True. Unknown source fields remain unknown." }, { "model_id": "grok-4.5", "benchmark_id": "tau3_banking", "score": 32.6, "reference_url": "https://media.x.ai/v1/website/4p5-5184fdf9.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "banking APIs and simulated user", "sampling": "pass@1", "judge": "final-state task accuracy", "harness": "Artificial Analysis tau3-banking harness", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "card45 tau3-banking Accuracy (%); source effort=high; benchmark protocol matches canonical=True. Unknown source fields remain unknown." }, { "model_id": "grok-4.5", "benchmark_id": "terminal_bench_3_0", "score": 15.7, "reference_url": "https://x.ai/news/grok-4-6", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "container terminal", "sampling": "pass@1", "judge": "verified task-success evaluator", "harness": "Terminal-Bench 3.0 / FrontierBench fixed harness", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "release46 Terminal-Bench v3.0 Task success rate (%); source effort=high; benchmark protocol matches canonical=False. Unknown source fields remain unknown." }, { "model_id": "grok-4.5", "benchmark_id": "vals_index", "score": 65.3, "reference_url": "https://media.x.ai/v1/website/card-4p6-4cd2dc57.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "benchmark-specified professional tools", "sampling": "unknown", "judge": "Vals weighted suite score", "harness": "Vals Index live harness", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "card46 Vals Index Suite score (%); source effort=high; benchmark protocol matches canonical=True. Unknown source fields remain unknown." }, { "model_id": "grok-4.5", "benchmark_id": "vct_accuracy", "score": 65.5, "reference_url": "https://media.x.ai/v1/website/card-4p6-4cd2dc57.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "multimodal input; no safeguards", "sampling": "unknown", "judge": "exact VCT accuracy", "harness": "VCT capability harness", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "card46 Virology Capabilities Test Accuracy (%); source effort=high; benchmark protocol matches canonical=True. Unknown source fields remain unknown." }, { "model_id": "grok-4.5", "benchmark_id": "wmdp_bio_accuracy", "score": 90.9, "reference_url": "https://media.x.ai/v1/website/card-4p6-4cd2dc57.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none; no safeguards", "sampling": "unknown", "judge": "multiple-choice exact accuracy", "harness": "WMDP official evaluation", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "card46 WMDP WMDP-Bio accuracy (%); source effort=high; benchmark protocol matches canonical=True. Unknown source fields remain unknown." }, { "model_id": "grok-4.5", "benchmark_id": "wmdp_chem_accuracy", "score": 87.3, "reference_url": "https://media.x.ai/v1/website/card-4p6-4cd2dc57.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none; no safeguards", "sampling": "unknown", "judge": "multiple-choice exact accuracy", "harness": "WMDP official evaluation", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "card46 WMDP WMDP-Chem accuracy (%); source effort=high; benchmark protocol matches canonical=True. Unknown source fields remain unknown." }, { "model_id": "grok-4.5", "benchmark_id": "xai_bio_refusal_accuracy", "score": 97.9, "reference_url": "https://media.x.ai/v1/website/card-4p6-4cd2dc57.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "full safeguards", "sampling": "unknown", "judge": "dangerous-query refusal accuracy", "harness": "SpaceXAI Autointent-Bio suite", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "card46 CBRN refusals Bio refusal accuracy (%); source effort=high; benchmark protocol matches canonical=True. Unknown source fields remain unknown." }, { "model_id": "grok-4.5", "benchmark_id": "xai_chem_refusal_accuracy", "score": 96.7, "reference_url": "https://media.x.ai/v1/website/card-4p6-4cd2dc57.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "full safeguards", "sampling": "unknown", "judge": "dangerous-query refusal accuracy", "harness": "SpaceXAI Autointent-Chem suite", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "card46 CBRN refusals Chem refusal accuracy (%); source effort=high; benchmark protocol matches canonical=True. Unknown source fields remain unknown." }, { "model_id": "grok-4.5", "benchmark_id": "xai_child_safety_compliance", "score": 0.0, "reference_url": "https://media.x.ai/v1/website/card-4p6-4cd2dc57.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "standard safeguards", "sampling": "unknown", "judge": "CSAM and child-safety compliance grader", "harness": "SpaceXAI child-safety suite", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "card46 Child safety Compliance (%); source effort=high; benchmark protocol matches canonical=True. Unknown source fields remain unknown." }, { "model_id": "grok-4.5", "benchmark_id": "xai_epistemic_bias_rate", "score": 20.4, "reference_url": "https://media.x.ai/v1/website/4p5-5184fdf9.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "unknown", "judge": "opposing-framing bias grader", "harness": "SpaceXAI epistemic-bias suite", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "card45 Epistemic Bias Bias rate (%); source effort=high; benchmark protocol matches canonical=True. Unknown source fields remain unknown." }, { "model_id": "grok-4.5", "benchmark_id": "xai_general_refusal_compliance", "score": 1.1, "reference_url": "https://media.x.ai/v1/website/card-4p6-4cd2dc57.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "standard safeguards", "sampling": "unknown", "judge": "multilingual should-refuse compliance grader", "harness": "SpaceXAI general-refusal suite", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "card46 General refusals Compliance (%); source effort=high; benchmark protocol matches canonical=True. Unknown source fields remain unknown." }, { "model_id": "grok-4.5", "benchmark_id": "xai_inferenceeval_accuracy", "score": 41.3, "reference_url": "https://media.x.ai/v1/website/card-4p6-4cd2dc57.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "Grok Build coding tools and GPU tests", "sampling": "unknown", "judge": "hidden GPU unit score gated by integration probe", "harness": "SpaceXAI InferenceEval harness", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "card46 InferenceEval Accuracy (%); source effort=high; benchmark protocol matches canonical=True. Unknown source fields remain unknown." }, { "model_id": "grok-4.5", "benchmark_id": "xai_kernelbenchinternal_accuracy_efficiency", "score": 58.4, "reference_url": "https://media.x.ai/v1/website/card-4p6-4cd2dc57.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "sandboxed GPU coding tools", "sampling": "unknown", "judge": "correctness and measured speedup composite", "harness": "SpaceXAI KernelBenchInternal harness", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "card46 KernelBenchInternal Accuracy / efficiency (%); source effort=high; benchmark protocol matches canonical=True. Unknown source fields remain unknown." }, { "model_id": "grok-4.5", "benchmark_id": "xai_mask_rectified_dishonesty", "score": 0.67, "reference_url": "https://media.x.ai/v1/website/card-4p6-4cd2dc57.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "unknown", "judge": "MASK-Rectified dishonesty grader", "harness": "SpaceXAI MASK-Rectified harness", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "card46 MASK-Rectified Dishonesty (%); source effort=high; benchmark protocol matches canonical=True. Unknown source fields remain unknown." }, { "model_id": "grok-4.5", "benchmark_id": "xai_rn_refusal_accuracy", "score": 97.9, "reference_url": "https://media.x.ai/v1/website/4p5-5184fdf9.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "full safeguards", "sampling": "unknown", "judge": "dangerous-query refusal accuracy", "harness": "FORTRESS-RN evaluation", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "card45 CBRN refusals R/N refusal accuracy (%); source effort=high; benchmark protocol matches canonical=True. Unknown source fields remain unknown." }, { "model_id": "grok-4.5", "benchmark_id": "xai_securecodereview_reward", "score": 49.4, "reference_url": "https://media.x.ai/v1/website/card-4p6-4cd2dc57.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "fixed secure-code-review tools", "sampling": "unknown", "judge": "security-fix reward with regression checks", "harness": "SpaceXAI SecureCodeReview harness", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "card46 SecureCodeReview Reward (%); source effort=high; benchmark protocol matches canonical=True. Unknown source fields remain unknown." }, { "model_id": "grok-4.5", "benchmark_id": "xai_self_harm_compliance", "score": 0.5, "reference_url": "https://media.x.ai/v1/website/card-4p6-4cd2dc57.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "standard safeguards", "sampling": "unknown", "judge": "self-harm assistance or failed-redirection compliance grader", "harness": "SpaceXAI self-harm suite", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "card46 Self-harm refusals Compliance (%); source effort=high; benchmark protocol matches canonical=True. Unknown source fields remain unknown." }, { "model_id": "grok-4.5", "benchmark_id": "xai_standard_jailbreak_compliance", "score": 0.73, "reference_url": "https://media.x.ai/v1/website/card-4p6-4cd2dc57.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "standard safeguards", "sampling": "unknown", "judge": "should-refuse compliance grader", "harness": "SpaceXAI jailbreak suite", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "card46 Jailbreaks Standard jailbreak compliance (%); source effort=high; benchmark protocol matches canonical=True. Unknown source fields remain unknown." }, { "model_id": "grok-4.5", "benchmark_id": "xai_sycophancy_rate", "score": 0.01, "reference_url": "https://media.x.ai/v1/website/card-4p6-4cd2dc57.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "unknown", "judge": "accuracy-drop sycophancy grader", "harness": "SpaceXAI internal sycophancy suite", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "card46 Sycophancy Sycophancy (%); source effort=high; benchmark protocol matches canonical=True. Unknown source fields remain unknown." }, { "model_id": "grok-4.6", "benchmark_id": "aa_briefcase_elo", "score": 1577, "reference_url": "https://x.ai/news/grok-4-6", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "offline multi-file knowledge-work tools", "sampling": "unknown", "judge": "rubric plus analytical and presentation pairwise Elo", "harness": "Artificial Analysis AA-Briefcase harness", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "release46 AA-Briefcase Elo; source effort=high; benchmark protocol matches canonical=True. Unknown source fields remain unknown." }, { "model_id": "grok-4.6", "benchmark_id": "aa_intelligence_index_v4_1", "score": 61, "reference_url": "https://x.ai/news/grok-4-6", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "benchmark-specified", "sampling": "unknown", "judge": "Artificial Analysis weighted composite", "harness": "Artificial Analysis live v4.1 harness", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "release46 AA Intelligence Index Composite score; source effort=high; benchmark protocol matches canonical=True. Unknown source fields remain unknown." }, { "model_id": "grok-4.6", "benchmark_id": "apex_agents", "score": 57.5, "reference_url": "https://x.ai/news/grok-4-6", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "multi-application professional tools", "sampling": "pass@1", "judge": "expert all-criteria-pass rubric", "harness": "Mercor APEX-Agents harness", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "release46 APEX-Agents Pass@1 (%); source effort=high; benchmark protocol matches canonical=True. Unknown source fields remain unknown." }, { "model_id": "grok-4.6", "benchmark_id": "apex_swe", "score": 56.4, "reference_url": "https://x.ai/news/grok-4-6", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "agentic software-engineering environment", "sampling": "pass@1", "judge": "Mercor integration and observability task verifier", "harness": "Mercor APEX-SWE harness", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "release46 APEX-SWE Pass@1 (%); source effort=high; benchmark protocol matches canonical=True. Unknown source fields remain unknown." }, { "model_id": "grok-4.6", "benchmark_id": "biosecurity_vct_accuracy", "score": 47.8, "reference_url": "https://media.x.ai/v1/website/card-4p6-4cd2dc57.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "no safeguards", "sampling": "unknown", "judge": "overall should-refuse and should-comply accuracy", "harness": "SpaceXAI internal Biosecurity VCT harness", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "card46 Biosecurity VCT Accuracy (%); source effort=high; benchmark protocol matches canonical=True. Unknown source fields remain unknown." }, { "model_id": "grok-4.6", "benchmark_id": "biousebench_severity5_refusal_rate", "score": 90.7, "reference_url": "https://media.x.ai/v1/website/card-4p6-4cd2dc57.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "standard safeguards", "sampling": "unknown", "judge": "severity-5 refusal grader", "harness": "SpaceXAI internal BioUseBench harness", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "card46 BioUseBench Severity-5 refusal rate (%); source effort=high; benchmark protocol matches canonical=True. Unknown source fields remain unknown." }, { "model_id": "grok-4.6", "benchmark_id": "cadgenbench_generation_cad_score", "score": 40.9, "reference_url": "https://media.x.ai/v1/website/card-4p6-4cd2dc57.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "CAD generation agent tools", "sampling": "unknown", "judge": "generation CAD score", "harness": "CadGenBench generation harness", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "card46 CadGenBench Generation Reward (%); source effort=high; benchmark protocol matches canonical=True. Unknown source fields remain unknown." }, { "model_id": "grok-4.6", "benchmark_id": "cve_bench_reward", "score": 39.8, "reference_url": "https://media.x.ai/v1/website/card-4p6-4cd2dc57.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "unrestricted sandboxed web-application tools", "sampling": "unknown", "judge": "CVE-Bench exploit reward", "harness": "CVE-Bench sandbox", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "card46 CVE-Bench Reward (%); source effort=high; benchmark protocol matches canonical=True. Unknown source fields remain unknown." }, { "model_id": "grok-4.6", "benchmark_id": "cybergym", "score": 79.7, "reference_url": "https://media.x.ai/v1/website/card-4p6-4cd2dc57.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "unrestricted agentic code execution", "sampling": "unknown", "judge": "PoC reproduced on vulnerable but not fixed version", "harness": "CyberGym unrestricted capability harness", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "card46 CyberGym Mean Reproduced (%); source effort=high; benchmark protocol matches canonical=True. Unknown source fields remain unknown." }, { "model_id": "grok-4.6", "benchmark_id": "deep_swe_v1_1", "score": 65.9, "reference_url": "https://x.ai/news/grok-4-6", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "agentic repository shell/editor", "sampling": "pass@1", "judge": "isolated repository verifier", "harness": "mini-swe-agent run by Datacurve", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "release46 DeepSWE v1.1 Pass@1 (%); source effort=high; benchmark protocol matches canonical=True. Unknown source fields remain unknown." }, { "model_id": "grok-4.6", "benchmark_id": "deepsearchqa_acc", "score": 81.6, "reference_url": "https://media.x.ai/v1/website/card-4p6-4cd2dc57.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "web search agent", "sampling": "unknown", "judge": "DeepSearchQA answer-accuracy grader", "harness": "SpaceXAI internal DeepSearchQA implementation", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "card46 DeepSearchQA Accuracy (%); source effort=high; benchmark protocol matches canonical=False. Unknown source fields remain unknown." }, { "model_id": "grok-4.6", "benchmark_id": "eebench_v1_core_reward", "score": 53.0, "reference_url": "https://media.x.ai/v1/website/card-4p6-4cd2dc57.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "electrical-engineering design tools", "sampling": "unknown", "judge": "physical correctness and functionality reward", "harness": "EEBench v1 core public leaderboard harness", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "card46 EEBench v1 core Reward (%); source effort=high; benchmark protocol matches canonical=True. Unknown source fields remain unknown.", "candidates": [ { "score": 60.0, "reference_url": "https://media.x.ai/v1/website/card-4p6-4cd2dc57.pdf", "source_type": "model_card", "reported_setting": { "mode": "thinking", "effort": "xhigh", "tools": "electrical-engineering design tools", "sampling": "unknown", "judge": "physical correctness and functionality reward", "harness": "EEBench v1 core public leaderboard harness", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown" }, "notes": "card46 EEBench v1 core Reward (%); source effort=xhigh; benchmark protocol matches canonical=True. Unknown source fields remain unknown." } ] }, { "model_id": "grok-4.6", "benchmark_id": "frontiercode_extended_v1_1", "score": 61.3, "reference_url": "https://x.ai/news/grok-4-6", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "fixed coding-agent scaffold", "sampling": "unknown", "judge": "maintainer-authored rubrics, verifiers, and hard blockers", "harness": "FrontierCode v1.1 Extended harness", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "release46 FrontierCode v1.1 Extended Extended score (%); source effort=high; benchmark protocol matches canonical=True. Unknown source fields remain unknown." }, { "model_id": "grok-4.6", "benchmark_id": "gdpval_aa_v2_elo", "score": 1753, "reference_url": "https://x.ai/news/grok-4-6", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "benchmark-specified knowledge-work tools", "sampling": "unknown", "judge": "pairwise Bradley-Terry Elo", "harness": "Artificial Analysis GDPval-AA v2 harness", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "release46 GDPVal-AA v2 Elo; source effort=high; benchmark protocol matches canonical=True. Unknown source fields remain unknown." }, { "model_id": "grok-4.6", "benchmark_id": "hackerbench_v0_2_benign_refusal", "score": 0.8, "reference_url": "https://media.x.ai/v1/website/card-4p6-4cd2dc57.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "constrained cyber task tools", "sampling": "unknown", "judge": "benign refusal grader", "harness": "HackerBench v0.2 with standard safeguards", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "card46 HackerBench v0.2 Benign Refusal (%); source effort=high; benchmark protocol matches canonical=True. Unknown source fields remain unknown." }, { "model_id": "grok-4.6", "benchmark_id": "hackerbench_v0_2_harmful_compliance", "score": 16.7, "reference_url": "https://media.x.ai/v1/website/card-4p6-4cd2dc57.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "constrained cyber task tools", "sampling": "unknown", "judge": "harmful or dual-use compliance grader", "harness": "HackerBench v0.2 with standard safeguards", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "card46 HackerBench v0.2 Harmful/Dual-Use Compliance (%); source effort=high; benchmark protocol matches canonical=True. Unknown source fields remain unknown." }, { "model_id": "grok-4.6", "benchmark_id": "harvey_lab_vals_final_score", "score": 15.8, "reference_url": "https://x.ai/news/grok-4-6", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "Vals legal-agent file tools", "sampling": "unknown", "judge": "expert all-pass final score", "harness": "Vals implementation of Harvey LAB", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "release46 Harvey LAB (Vals) Final score (%); source effort=high; benchmark protocol matches canonical=True. Unknown source fields remain unknown.", "candidates": [ { "score": 22.0, "reference_url": "https://media.x.ai/v1/website/card-4p6-4cd2dc57.pdf", "source_type": "model_card", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "Vals legal-agent file tools", "sampling": "unknown", "judge": "expert all-pass final score", "harness": "Vals implementation of Harvey LAB", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown" }, "notes": "card46 Harvey Legal Agent Benchmark Final score (%); source effort=high; benchmark protocol matches canonical=True. Unknown source fields remain unknown." } ] }, { "model_id": "grok-4.6", "benchmark_id": "lab_bench_protocolqa_accuracy", "score": 79.6, "reference_url": "https://media.x.ai/v1/website/card-4p6-4cd2dc57.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none; no safeguards", "sampling": "unknown", "judge": "open-ended single-error accuracy grader", "harness": "LAB-Bench ProtocolQA harness", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "card46 ProtocolQA Open-Ended Accuracy (%); source effort=high; benchmark protocol matches canonical=False. Unknown source fields remain unknown." }, { "model_id": "grok-4.6", "benchmark_id": "officeqa_pro", "score": 63.2, "reference_url": "https://media.x.ai/v1/website/card-4p6-4cd2dc57.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "document and rendered-image analysis", "sampling": "unknown", "judge": "OfficeQA Pro exact-match accuracy", "harness": "OfficeQA Pro official harness", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "card46 OfficeQA Pro Accuracy (%); source effort=high; benchmark protocol matches canonical=True. Unknown source fields remain unknown." }, { "model_id": "grok-4.6", "benchmark_id": "parametric_cad_bench_combined", "score": 87.8, "reference_url": "https://media.x.ai/v1/website/card-4p6-4cd2dc57.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "parametric CAD agent tools", "sampling": "unknown", "judge": "combined executability and geometric reward", "harness": "gNucleus Parametric CAD Bench harness", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "card46 Parametric CAD Bench Reward (%); source effort=high; benchmark protocol matches canonical=True. Unknown source fields remain unknown." }, { "model_id": "grok-4.6", "benchmark_id": "single_turn_hallucination_rate", "score": 1.7, "reference_url": "https://media.x.ai/v1/website/card-4p6-4cd2dc57.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "unknown", "judge": "separate factual-support grader", "harness": "SpaceXAI internal single-turn factuality harness", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "card46 Single-turn hallucination Hallucination rate (%); source effort=high; benchmark protocol matches canonical=True. Unknown source fields remain unknown." }, { "model_id": "grok-4.6", "benchmark_id": "spacexai_mts_accuracy", "score": 61.1, "reference_url": "https://media.x.ai/v1/website/card-4p6-4cd2dc57.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "Grok Build coding and research tools", "sampling": "unknown", "judge": "task correctness under fixed rollout budget", "harness": "Grok Build", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "card46 SpaceXAI MTS Eval Correct (%); source effort=high; benchmark protocol matches canonical=True. Unknown source fields remain unknown." }, { "model_id": "grok-4.6", "benchmark_id": "strongreject_compliance", "score": 3.9, "reference_url": "https://media.x.ai/v1/website/card-4p6-4cd2dc57.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "standard safeguards", "sampling": "unknown", "judge": "StrongREJECT compliance grader", "harness": "StrongREJECT evaluation", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "card46 Jailbreaks StrongREJECT compliance (%); source effort=high; benchmark protocol matches canonical=True. Unknown source fields remain unknown." }, { "model_id": "grok-4.6", "benchmark_id": "swe_marathon_v1_1", "score": 31.9, "reference_url": "https://media.x.ai/v1/website/card-4p6-4cd2dc57.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "agentic code execution", "sampling": "pass@1", "judge": "multi-layer resolution verification", "harness": "SWE-Marathon v1.1 harness", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "card46 SWE-Marathon v1.1 Resolution rate (%); source effort=high; benchmark protocol matches canonical=True. Unknown source fields remain unknown." }, { "model_id": "grok-4.6", "benchmark_id": "terminal_bench_3_0", "score": 26.0, "reference_url": "https://x.ai/news/grok-4-6", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "container terminal", "sampling": "pass@1", "judge": "verified task-success evaluator", "harness": "Terminal-Bench 3.0 / FrontierBench fixed harness", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "release46 Terminal-Bench v3.0 Task success rate (%); source effort=high; benchmark protocol matches canonical=False. Unknown source fields remain unknown." }, { "model_id": "grok-4.6", "benchmark_id": "vals_index", "score": 71.1, "reference_url": "https://media.x.ai/v1/website/card-4p6-4cd2dc57.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "benchmark-specified professional tools", "sampling": "unknown", "judge": "Vals weighted suite score", "harness": "Vals Index live harness", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "card46 Vals Index Suite score (%); source effort=high; benchmark protocol matches canonical=True. Unknown source fields remain unknown." }, { "model_id": "grok-4.6", "benchmark_id": "vct_accuracy", "score": 67.4, "reference_url": "https://media.x.ai/v1/website/card-4p6-4cd2dc57.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "multimodal input; no safeguards", "sampling": "unknown", "judge": "exact VCT accuracy", "harness": "VCT capability harness", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "card46 Virology Capabilities Test Accuracy (%); source effort=high; benchmark protocol matches canonical=True. Unknown source fields remain unknown." }, { "model_id": "grok-4.6", "benchmark_id": "wmdp_bio_accuracy", "score": 90.0, "reference_url": "https://media.x.ai/v1/website/card-4p6-4cd2dc57.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none; no safeguards", "sampling": "unknown", "judge": "multiple-choice exact accuracy", "harness": "WMDP official evaluation", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "card46 WMDP WMDP-Bio accuracy (%); source effort=high; benchmark protocol matches canonical=True. Unknown source fields remain unknown." }, { "model_id": "grok-4.6", "benchmark_id": "wmdp_chem_accuracy", "score": 85.3, "reference_url": "https://media.x.ai/v1/website/card-4p6-4cd2dc57.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none; no safeguards", "sampling": "unknown", "judge": "multiple-choice exact accuracy", "harness": "WMDP official evaluation", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "card46 WMDP WMDP-Chem accuracy (%); source effort=high; benchmark protocol matches canonical=True. Unknown source fields remain unknown." }, { "model_id": "grok-4.6", "benchmark_id": "wmdp_cyber_accuracy", "score": 90.1, "reference_url": "https://media.x.ai/v1/website/card-4p6-4cd2dc57.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none; no safeguards", "sampling": "unknown", "judge": "multiple-choice exact accuracy", "harness": "WMDP official evaluation", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "card46 WMDP WMDP-Cyber accuracy (%); source effort=high; benchmark protocol matches canonical=True. Unknown source fields remain unknown." }, { "model_id": "grok-4.6", "benchmark_id": "xai_bio_refusal_accuracy", "score": 100.0, "reference_url": "https://media.x.ai/v1/website/card-4p6-4cd2dc57.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "full safeguards", "sampling": "unknown", "judge": "dangerous-query refusal accuracy", "harness": "SpaceXAI Autointent-Bio suite", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "card46 CBRN refusals Bio refusal accuracy (%); source effort=high; benchmark protocol matches canonical=True. Unknown source fields remain unknown." }, { "model_id": "grok-4.6", "benchmark_id": "xai_chem_refusal_accuracy", "score": 100.0, "reference_url": "https://media.x.ai/v1/website/card-4p6-4cd2dc57.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "full safeguards", "sampling": "unknown", "judge": "dangerous-query refusal accuracy", "harness": "SpaceXAI Autointent-Chem suite", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "card46 CBRN refusals Chem refusal accuracy (%); source effort=high; benchmark protocol matches canonical=True. Unknown source fields remain unknown." }, { "model_id": "grok-4.6", "benchmark_id": "xai_child_safety_compliance", "score": 0.0, "reference_url": "https://media.x.ai/v1/website/card-4p6-4cd2dc57.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "standard safeguards", "sampling": "unknown", "judge": "CSAM and child-safety compliance grader", "harness": "SpaceXAI child-safety suite", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "card46 Child safety Compliance (%); source effort=high; benchmark protocol matches canonical=True. Unknown source fields remain unknown." }, { "model_id": "grok-4.6", "benchmark_id": "xai_general_refusal_compliance", "score": 0.93, "reference_url": "https://media.x.ai/v1/website/card-4p6-4cd2dc57.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "standard safeguards", "sampling": "unknown", "judge": "multilingual should-refuse compliance grader", "harness": "SpaceXAI general-refusal suite", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "card46 General refusals Compliance (%); source effort=high; benchmark protocol matches canonical=True. Unknown source fields remain unknown." }, { "model_id": "grok-4.6", "benchmark_id": "xai_inferenceeval_accuracy", "score": 46.9, "reference_url": "https://media.x.ai/v1/website/card-4p6-4cd2dc57.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "Grok Build coding tools and GPU tests", "sampling": "unknown", "judge": "hidden GPU unit score gated by integration probe", "harness": "SpaceXAI InferenceEval harness", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "card46 InferenceEval Accuracy (%); source effort=high; benchmark protocol matches canonical=True. Unknown source fields remain unknown." }, { "model_id": "grok-4.6", "benchmark_id": "xai_kernelbenchinternal_accuracy_efficiency", "score": 70.6, "reference_url": "https://media.x.ai/v1/website/card-4p6-4cd2dc57.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "sandboxed GPU coding tools", "sampling": "unknown", "judge": "correctness and measured speedup composite", "harness": "SpaceXAI KernelBenchInternal harness", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "card46 KernelBenchInternal Accuracy / efficiency (%); source effort=high; benchmark protocol matches canonical=True. Unknown source fields remain unknown." }, { "model_id": "grok-4.6", "benchmark_id": "xai_long_horizon_jailbreak_compliance", "score": 1.0, "reference_url": "https://media.x.ai/v1/website/card-4p6-4cd2dc57.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "standard safeguards", "sampling": "unknown", "judge": "long-horizon should-refuse compliance grader", "harness": "SpaceXAI long-horizon jailbreak suite", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "card46 Jailbreaks Long-horizon compliance (%); source effort=high; benchmark protocol matches canonical=True. Unknown source fields remain unknown." }, { "model_id": "grok-4.6", "benchmark_id": "xai_mask_rectified_dishonesty", "score": 3.8, "reference_url": "https://media.x.ai/v1/website/card-4p6-4cd2dc57.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "unknown", "judge": "MASK-Rectified dishonesty grader", "harness": "SpaceXAI MASK-Rectified harness", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "card46 MASK-Rectified Dishonesty (%); source effort=high; benchmark protocol matches canonical=True. Unknown source fields remain unknown." }, { "model_id": "grok-4.6", "benchmark_id": "xai_securecodereview_reward", "score": 58.7, "reference_url": "https://media.x.ai/v1/website/card-4p6-4cd2dc57.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "fixed secure-code-review tools", "sampling": "unknown", "judge": "security-fix reward with regression checks", "harness": "SpaceXAI SecureCodeReview harness", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "card46 SecureCodeReview Reward (%); source effort=high; benchmark protocol matches canonical=True. Unknown source fields remain unknown." }, { "model_id": "grok-4.6", "benchmark_id": "xai_self_harm_compliance", "score": 3.7, "reference_url": "https://media.x.ai/v1/website/card-4p6-4cd2dc57.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "standard safeguards", "sampling": "unknown", "judge": "self-harm assistance or failed-redirection compliance grader", "harness": "SpaceXAI self-harm suite", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "card46 Self-harm refusals Compliance (%); source effort=high; benchmark protocol matches canonical=True. Unknown source fields remain unknown." }, { "model_id": "grok-4.6", "benchmark_id": "xai_standard_jailbreak_compliance", "score": 0.04, "reference_url": "https://media.x.ai/v1/website/card-4p6-4cd2dc57.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "standard safeguards", "sampling": "unknown", "judge": "should-refuse compliance grader", "harness": "SpaceXAI jailbreak suite", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "card46 Jailbreaks Standard jailbreak compliance (%); source effort=high; benchmark protocol matches canonical=True. Unknown source fields remain unknown." }, { "model_id": "grok-4.6", "benchmark_id": "xai_sycophancy_rate", "score": 0.04, "reference_url": "https://media.x.ai/v1/website/card-4p6-4cd2dc57.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "unknown", "judge": "accuracy-drop sycophancy grader", "harness": "SpaceXAI internal sycophancy suite", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "card46 Sycophancy Sycophancy (%); source effort=high; benchmark protocol matches canonical=True. Unknown source fields remain unknown." }, { "model_id": "kimi-k3", "benchmark_id": "apex_swe", "score": 48.0, "reference_url": "https://media.x.ai/v1/website/card-4p6-4cd2dc57.pdf", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "agentic software-engineering environment", "sampling": "pass@1", "judge": "Mercor integration and observability task verifier", "harness": "Mercor APEX-SWE harness", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "card46 APEX-SWE Pass@1 (%); source effort=max; benchmark protocol matches canonical=True. Unknown source fields remain unknown." }, { "model_id": "kimi-k3", "benchmark_id": "eebench_v1_core_reward", "score": 38.3, "reference_url": "https://media.x.ai/v1/website/card-4p6-4cd2dc57.pdf", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "electrical-engineering design tools", "sampling": "unknown", "judge": "physical correctness and functionality reward", "harness": "EEBench v1 core public leaderboard harness", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "card46 EEBench v1 core Reward (%); source effort=max; benchmark protocol matches canonical=True. Unknown source fields remain unknown." }, { "model_id": "kimi-k3", "benchmark_id": "gdpval_aa_v2_elo", "score": 1682, "reference_url": "https://media.x.ai/v1/website/card-4p6-4cd2dc57.pdf", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "benchmark-specified knowledge-work tools", "sampling": "unknown", "judge": "pairwise Bradley-Terry Elo", "harness": "Artificial Analysis GDPval-AA v2 harness", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "card46 AA GDPVal v2 AA GDPVal Elo; source effort=max; benchmark protocol matches canonical=True. Unknown source fields remain unknown." }, { "model_id": "kimi-k3", "benchmark_id": "swe_marathon_v1_1", "score": 48.1, "reference_url": "https://media.x.ai/v1/website/card-4p6-4cd2dc57.pdf", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "agentic code execution", "sampling": "pass@1", "judge": "multi-layer resolution verification", "harness": "SWE-Marathon v1.1 harness", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "card46 SWE-Marathon v1.1 Resolution rate (%); source effort=max; benchmark protocol matches canonical=True. Unknown source fields remain unknown." }, { "model_id": "kimi-k3", "benchmark_id": "xai_inferenceeval_accuracy", "score": 39.8, "reference_url": "https://media.x.ai/v1/website/card-4p6-4cd2dc57.pdf", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "Grok Build coding tools and GPU tests", "sampling": "unknown", "judge": "hidden GPU unit score gated by integration probe", "harness": "SpaceXAI InferenceEval harness", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "card46 InferenceEval Accuracy (%); source effort=max; benchmark protocol matches canonical=True. Unknown source fields remain unknown." }, { "model_id": "kimi-k3", "benchmark_id": "xai_kernelbenchinternal_accuracy_efficiency", "score": 68.5, "reference_url": "https://media.x.ai/v1/website/card-4p6-4cd2dc57.pdf", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "sandboxed GPU coding tools", "sampling": "unknown", "judge": "correctness and measured speedup composite", "harness": "SpaceXAI KernelBenchInternal harness", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown" }, "matches_canonical": true, "source_type": "model_card", "audit_status": "verified", "notes": "card46 KernelBenchInternal Accuracy / efficiency (%); source effort=max; benchmark protocol matches canonical=True. Unknown source fields remain unknown." }, { "model_id": "swe-1.7", "benchmark_id": "frontiercode_extended_v1_1", "score": 54.3, "reference_url": "https://media.x.ai/v1/website/card-4p6-4cd2dc57.pdf", "reported_setting": { "mode": "agentic/reasoning", "effort": "unknown", "tools": "fixed coding-agent scaffold", "sampling": "unknown", "judge": "maintainer-authored rubrics, verifiers, and hard blockers", "harness": "FrontierCode v1.1 Extended harness", "prompt_style": "unknown", "temperature": "unknown", "context": "unknown" }, "matches_canonical": false, "source_type": "model_card", "audit_status": "verified", "notes": "card46 FrontierCode v1.1 Extended Extended score (%); source effort=unknown; benchmark protocol matches canonical=True. Unknown source fields remain unknown." }, { "model_id": "claude-fable-5", "benchmark_id": "code_arena_frontend_elo", "score": 1654, "reference_url": "https://seed.bytedance.com/en/blog/seed2-1-officially-released-advancing-ai-productivity", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "107,962-vote Arena snapshot", "judge": "human preference", "harness": "Arena Code leaderboard", "prompt_style": "official default", "temperature": "unknown" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Release blog EN: Code Arena: Frontend snapshot (https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/user-upload/4xfa4mqq4xsvz.jpg). Historical leaderboard snapshot reproduced by ByteDance. Fable 5 was marked not currently sampled." }, { "model_id": "claude-opus-4.6", "benchmark_id": "code_arena_frontend_elo", "score": 1542, "reference_url": "https://seed.bytedance.com/en/blog/seed2-1-officially-released-advancing-ai-productivity", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "none", "sampling": "107,962-vote Arena snapshot", "judge": "human preference", "harness": "Arena Code leaderboard", "prompt_style": "official default", "temperature": "unknown" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Release blog EN: Code Arena: Frontend snapshot (https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/user-upload/4xfa4mqq4xsvz.jpg). Historical leaderboard snapshot reproduced by ByteDance.", "candidates": [ { "score": 1538, "reference_url": "https://seed.bytedance.com/en/blog/seed2-1-officially-released-advancing-ai-productivity", "source_type": "official_blog", "reported_setting": { "mode": "non-thinking", "effort": "default", "tools": "none", "sampling": "107,962-vote Arena snapshot", "judge": "human preference", "harness": "Arena Code leaderboard", "prompt_style": "official default", "temperature": "unknown" }, "notes": "Release blog EN: Code Arena: Frontend snapshot (https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/user-upload/4xfa4mqq4xsvz.jpg). Historical leaderboard snapshot reproduced by ByteDance." } ] }, { "model_id": "claude-opus-4.6", "benchmark_id": "officeqa_pro", "score": 48.12, "reference_url": "https://arxiv.org/abs/2603.08655", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "Claude Code CLI", "sampling": "pass@1", "judge": "0% tolerance exact match", "harness": "OfficeQA Pro; corpus=full; format=PDF", "prompt_style": "official default", "temperature": "unknown" }, "matches_canonical": false, "source_type": "leaderboard", "audit_status": "verified", "notes": "Model card Figure 17 (p30). Baseline reproduced from OfficeQA Pro paper section 3.2.", "candidates": [ { "score": 54.14, "reference_url": "https://arxiv.org/abs/2603.08655", "source_type": "leaderboard", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "Claude Code CLI", "sampling": "pass@1", "judge": "0% tolerance exact match", "harness": "OfficeQA Pro; corpus=full; format=DBX parsed", "prompt_style": "official default", "temperature": "unknown" }, "notes": "Model card Figure 17 (p30). Baseline reproduced from OfficeQA Pro paper section 3.2." }, { "score": 60.9, "reference_url": "https://arxiv.org/abs/2603.08655", "source_type": "leaderboard", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "Claude Code CLI", "sampling": "pass@1", "judge": "0% tolerance exact match", "harness": "OfficeQA Pro; corpus=oracle; format=PDF", "prompt_style": "official default", "temperature": "unknown" }, "notes": "Model card Figure 17 (p30). Baseline reproduced from OfficeQA Pro paper section 3.2." }, { "score": 66.92, "reference_url": "https://arxiv.org/abs/2603.08655", "source_type": "leaderboard", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "Claude Code CLI", "sampling": "pass@1", "judge": "0% tolerance exact match", "harness": "OfficeQA Pro; corpus=oracle; format=DBX parsed", "prompt_style": "official default", "temperature": "unknown" }, "notes": "Model card Figure 17 (p30). Baseline reproduced from OfficeQA Pro paper section 3.2." } ] }, { "model_id": "claude-opus-4.7", "benchmark_id": "aethercode", "score": 52.0, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2.1/Seed2_1_Model_Card.pdf", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "none", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official", "prompt_style": "official default", "temperature": "unknown" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Model card Table 11 (p53)." }, { "model_id": "claude-opus-4.7", "benchmark_id": "agent_startup_bench", "score": 62.3, "reference_url": "https://seed.bytedance.com/en/blog/seed2-1-officially-released-advancing-ai-productivity", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "research and professional-deliverable tools", "sampling": "pass@1", "judge": "expert review", "harness": "official", "prompt_style": "official default", "temperature": "unknown" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Release blog EN: General-agent chart (https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/user-upload/4xfa4mqq4u89c.png)." }, { "model_id": "claude-opus-4.7", "benchmark_id": "agents_last_exam", "score": 18.4, "reference_url": "https://seed.bytedance.com/en/blog/seed2-1-officially-released-advancing-ai-productivity", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "computer-use environment", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official", "prompt_style": "official default", "temperature": "unknown" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Release blog EN: General-agent chart (https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/user-upload/4xfa4mqq4u89c.png). Turbo is a source dash, not zero." }, { "model_id": "claude-opus-4.7", "benchmark_id": "agents_last_exam_average_score", "score": 40.5, "reference_url": "https://seed.bytedance.com/en/blog/seed2-1-officially-released-advancing-ai-productivity", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "computer-use environment", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official", "prompt_style": "official default", "temperature": "unknown" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Release blog EN: General-agent chart (https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/user-upload/4xfa4mqq4u89c.png). Turbo is a source dash, not zero." }, { "model_id": "claude-opus-4.7", "benchmark_id": "apex_agents", "score": 33.9, "reference_url": "https://seed.bytedance.com/en/blog/seed2-1-officially-released-advancing-ai-productivity", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "long-horizon professional-task environment", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official", "prompt_style": "official default", "temperature": "unknown" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Release blog EN: General-agent chart (https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/user-upload/4xfa4mqq4u89c.png)." }, { "model_id": "claude-opus-4.7", "benchmark_id": "babyvision", "score": 22.2, "reference_url": "https://seed.bytedance.com/en/blog/seed2-1-officially-released-advancing-ai-productivity", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "none", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official", "prompt_style": "official default", "temperature": "unknown" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Release blog EN: Vision capability chart (https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/user-upload/4xfa4mqq4yer3.png)." }, { "model_id": "claude-opus-4.7", "benchmark_id": "beyond_aime", "score": 79.0, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2.1/Seed2_1_Model_Card.pdf", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "none", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official", "prompt_style": "official default", "temperature": "unknown" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Model card Table 11 (p53)." }, { "model_id": "claude-opus-4.7", "benchmark_id": "blink", "score": 70.4, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2.1/Seed2_1_Model_Card.pdf", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "none", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official", "prompt_style": "official default", "temperature": "unknown" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Model card Table 9 (p51)." }, { "model_id": "claude-opus-4.7", "benchmark_id": "chartqapro", "score": 65.5, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2.1/Seed2_1_Model_Card.pdf", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "none", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official", "prompt_style": "official default", "temperature": "unknown" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Model card Table 9 (p51)." }, { "model_id": "claude-opus-4.7", "benchmark_id": "charxiv_descriptive", "score": 93.1, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2.1/Seed2_1_Model_Card.pdf", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "none", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official", "prompt_style": "official default", "temperature": "unknown" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Model card Table 9 (p51)." }, { "model_id": "claude-opus-4.7", "benchmark_id": "claw_eval_multimodal_pass3", "score": 44.0, "reference_url": "https://seed.bytedance.com/en/blog/seed2-1-officially-released-advancing-ai-productivity", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "high-resolution multimodal agent harness", "sampling": "Pass^3", "judge": "benchmark-specified", "harness": "official", "prompt_style": "official default", "temperature": "unknown" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Release blog EN: Multimodal-agent chart (https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/user-upload/4xfa4mqq4v1ol.png)." }, { "model_id": "claude-opus-4.7", "benchmark_id": "code_arena_frontend_elo", "score": 1563, "reference_url": "https://seed.bytedance.com/en/blog/seed2-1-officially-released-advancing-ai-productivity", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "none", "sampling": "107,962-vote Arena snapshot", "judge": "human preference", "harness": "Arena Code leaderboard", "prompt_style": "official default", "temperature": "unknown" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Release blog EN: Code Arena: Frontend snapshot (https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/user-upload/4xfa4mqq4xsvz.jpg). Historical leaderboard snapshot reproduced by ByteDance.", "candidates": [ { "score": 1557, "reference_url": "https://seed.bytedance.com/en/blog/seed2-1-officially-released-advancing-ai-productivity", "source_type": "official_blog", "reported_setting": { "mode": "non-thinking", "effort": "default", "tools": "none", "sampling": "107,962-vote Arena snapshot", "judge": "human preference", "harness": "Arena Code leaderboard", "prompt_style": "official default", "temperature": "unknown" }, "notes": "Release blog EN: Code Arena: Frontend snapshot (https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/user-upload/4xfa4mqq4xsvz.jpg). Historical leaderboard snapshot reproduced by ByteDance." } ] }, { "model_id": "claude-opus-4.7", "benchmark_id": "creativework", "score": 28.3, "reference_url": "https://seed.bytedance.com/en/blog/seed2-1-officially-released-advancing-ai-productivity", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "GUI + MCP in Notion, Canva, and Figma", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official", "prompt_style": "official default", "temperature": "unknown" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Release blog EN: Computer-use chart (https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/user-upload/4xfa4mqq4w8x3.png)." }, { "model_id": "claude-opus-4.7", "benchmark_id": "deep_swe_v1_1", "score": 54.0, "reference_url": "https://seed.bytedance.com/en/blog/seed2-1-officially-released-advancing-ai-productivity", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "Trae or benchmark-specific coding-agent environment", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official", "prompt_style": "official default", "temperature": "unknown" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Release blog EN: Coding chart (https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/user-upload/4xfa4mqq4wlbt.png)." }, { "model_id": "claude-opus-4.7", "benchmark_id": "doubao_multi_turn_bench", "score": 49.8, "reference_url": "https://seed.bytedance.com/en/blog/seed2-1-officially-released-advancing-ai-productivity", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "none", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official", "prompt_style": "official default", "temperature": "unknown" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Release blog EN: Daily-life/tool-use chart (https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/user-upload/4xfa4mqq4ujsp.png)." }, { "model_id": "claude-opus-4.7", "benchmark_id": "dude", "score": 84.5, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2.1/Seed2_1_Model_Card.pdf", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "none", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official", "prompt_style": "official default", "temperature": "unknown" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Model card Table 9 (p51)." }, { "model_id": "claude-opus-4.7", "benchmark_id": "dynamath", "score": 63.1, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2.1/Seed2_1_Model_Card.pdf", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "none", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official", "prompt_style": "official default", "temperature": "unknown" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Model card Table 9 (p51)." }, { "model_id": "claude-opus-4.7", "benchmark_id": "embspatial_bench", "score": 77.2, "reference_url": "https://seed.bytedance.com/en/blog/seed2-1-officially-released-advancing-ai-productivity", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "none", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official", "prompt_style": "official default", "temperature": "unknown" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Release blog EN: Spatial/long-context chart (https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/user-upload/4xfa4mqq4z2l5.png)." }, { "model_id": "claude-opus-4.7", "benchmark_id": "emma", "score": 68.7, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2.1/Seed2_1_Model_Card.pdf", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "none", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official", "prompt_style": "official default", "temperature": "unknown" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Model card Table 9 (p51)." }, { "model_id": "claude-opus-4.7", "benchmark_id": "erqa", "score": 52.5, "reference_url": "https://seed.bytedance.com/en/blog/seed2-1-officially-released-advancing-ai-productivity", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "none", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official", "prompt_style": "official default", "temperature": "unknown" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Release blog EN: Spatial/long-context chart (https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/user-upload/4xfa4mqq4z2l5.png)." }, { "model_id": "claude-opus-4.7", "benchmark_id": "frontier_science_research", "score": 20.0, "reference_url": "https://seed.bytedance.com/en/blog/seed2-1-officially-released-advancing-ai-productivity", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "research-oriented agent environment", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official", "prompt_style": "official default", "temperature": "unknown" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Release blog EN: Frontier-research chart (https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/user-upload/4xfa4mqq52cdr.png)." }, { "model_id": "claude-opus-4.7", "benchmark_id": "frontiersci_olympiad", "score": 69.0, "reference_url": "https://seed.bytedance.com/en/blog/seed2-1-officially-released-advancing-ai-productivity", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "none", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official", "prompt_style": "official default", "temperature": "unknown" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Release blog EN: Language/search chart (https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/user-upload/4xfa4mqq510uc.png)." }, { "model_id": "claude-opus-4.7", "benchmark_id": "fs_researcher", "score": 20.0, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2.1/Seed2_1_Model_Card.pdf", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "none", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official", "prompt_style": "official default", "temperature": "unknown" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Model card Table 11 (p53)." }, { "model_id": "claude-opus-4.7", "benchmark_id": "gameworld", "score": 26.5, "reference_url": "https://seed.bytedance.com/en/blog/seed2-1-officially-released-advancing-ai-productivity", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "browser-game GUI actions", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official", "prompt_style": "official default", "temperature": "unknown" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Release blog EN: Computer-use chart (https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/user-upload/4xfa4mqq4w8x3.png)." }, { "model_id": "claude-opus-4.7", "benchmark_id": "gdpval_seed_reported_score", "score": 82.7, "reference_url": "https://seed.bytedance.com/en/blog/seed2-1-officially-released-advancing-ai-productivity", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "document/spreadsheet deliverable workflow", "sampling": "pass@1", "judge": "source-specific evaluator", "harness": "official", "prompt_style": "official default", "temperature": "unknown" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Release blog EN: General-agent chart (https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/user-upload/4xfa4mqq4u89c.png)." }, { "model_id": "claude-opus-4.7", "benchmark_id": "hle_tools_text", "score": 54.7, "reference_url": "https://seed.bytedance.com/en/blog/seed2-1-officially-released-advancing-ai-productivity", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "search", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official", "prompt_style": "official default", "temperature": "unknown" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Release blog EN: Language/search chart (https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/user-upload/4xfa4mqq510uc.png)." }, { "model_id": "claude-opus-4.7", "benchmark_id": "hle_verified", "score": 46.9, "reference_url": "https://seed.bytedance.com/en/blog/seed2-1-officially-released-advancing-ai-productivity", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "none", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official", "prompt_style": "official default", "temperature": "unknown" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Release blog EN: Language/search chart (https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/user-upload/4xfa4mqq510uc.png)." }, { "model_id": "claude-opus-4.7", "benchmark_id": "horizon_math_pass12", "score": 4.0, "reference_url": "https://seed.bytedance.com/en/blog/seed2-1-officially-released-advancing-ai-productivity", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "research-oriented agent environment", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official", "prompt_style": "official default", "temperature": "unknown" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Release blog EN: Frontier-research chart (https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/user-upload/4xfa4mqq52cdr.png)." }, { "model_id": "claude-opus-4.7", "benchmark_id": "image2floorplan_avg_score", "score": 50.2, "reference_url": "https://seed.bytedance.com/en/blog/seed2-1-officially-released-advancing-ai-productivity", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "high-resolution multimodal agent harness", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official", "prompt_style": "official default", "temperature": "unknown" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Release blog EN: Multimodal-agent chart (https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/user-upload/4xfa4mqq4v1ol.png)." }, { "model_id": "claude-opus-4.7", "benchmark_id": "kina", "score": 46.7, "reference_url": "https://seed.bytedance.com/en/blog/seed2-1-officially-released-advancing-ai-productivity", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "none", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official", "prompt_style": "official default", "temperature": "unknown" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Release blog EN: Language/search chart (https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/user-upload/4xfa4mqq510uc.png)." }, { "model_id": "claude-opus-4.7", "benchmark_id": "live_mathematician_bench_2026_06", "score": 36.7, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2.1/Seed2_1_Model_Card.pdf", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "none", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official", "prompt_style": "official default", "temperature": "unknown" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Model card Table 11 (p53)." }, { "model_id": "claude-opus-4.7", "benchmark_id": "mathverse_vision_only", "score": 77.4, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2.1/Seed2_1_Model_Card.pdf", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "none", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official", "prompt_style": "official default", "temperature": "unknown" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Model card Table 9 (p51)." }, { "model_id": "claude-opus-4.7", "benchmark_id": "mathvision", "score": 83.1, "reference_url": "https://seed.bytedance.com/en/blog/seed2-1-officially-released-advancing-ai-productivity", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "none", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official", "prompt_style": "official default", "temperature": "unknown" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Release blog EN: Vision capability chart (https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/user-upload/4xfa4mqq4yer3.png)." }, { "model_id": "claude-opus-4.7", "benchmark_id": "mathvista", "score": 84.4, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2.1/Seed2_1_Model_Card.pdf", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "none", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official", "prompt_style": "official default", "temperature": "unknown" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Model card Table 9 (p51)." }, { "model_id": "claude-opus-4.7", "benchmark_id": "measurebench", "score": 29.7, "reference_url": "https://seed.bytedance.com/en/blog/seed2-1-officially-released-advancing-ai-productivity", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "none", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official", "prompt_style": "official default", "temperature": "unknown" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Release blog EN: Vision capability chart (https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/user-upload/4xfa4mqq4yer3.png)." }, { "model_id": "claude-opus-4.7", "benchmark_id": "mmsibench_circular", "score": 17.4, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2.1/Seed2_1_Model_Card.pdf", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "none", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official", "prompt_style": "official default", "temperature": "unknown" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Model card Table 9 (p51)." }, { "model_id": "claude-opus-4.7", "benchmark_id": "mobileworld", "score": 57.1, "reference_url": "https://seed.bytedance.com/en/blog/seed2-1-officially-released-advancing-ai-productivity", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "mobile GUI actions", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official", "prompt_style": "official default", "temperature": "unknown" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Release blog EN: Computer-use chart (https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/user-upload/4xfa4mqq4w8x3.png)." }, { "model_id": "claude-opus-4.7", "benchmark_id": "msqa", "score": 42.8, "reference_url": "https://seed.bytedance.com/en/blog/seed2-1-officially-released-advancing-ai-productivity", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "none", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official", "prompt_style": "official default", "temperature": "unknown" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Release blog EN: Language/search chart (https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/user-upload/4xfa4mqq510uc.png)." }, { "model_id": "claude-opus-4.7", "benchmark_id": "nl2repo_bench", "score": 58.2, "reference_url": "https://seed.bytedance.com/en/blog/seed2-1-officially-released-advancing-ai-productivity", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "Trae or benchmark-specific coding-agent environment", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official", "prompt_style": "official default", "temperature": "unknown" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Release blog EN: Coding chart (https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/user-upload/4xfa4mqq4wlbt.png)." }, { "model_id": "claude-opus-4.7", "benchmark_id": "ocrbench_v2", "score": 56.9, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2.1/Seed2_1_Model_Card.pdf", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "none", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official", "prompt_style": "official default", "temperature": "unknown" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Model card Table 9 (p51)." }, { "model_id": "claude-opus-4.7", "benchmark_id": "one_million_bench", "score": 73.0, "reference_url": "https://seed.bytedance.com/en/blog/seed2-1-officially-released-advancing-ai-productivity", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "agentic professional-task environment", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official", "prompt_style": "official default", "temperature": "unknown" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Release blog EN: General-agent chart (https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/user-upload/4xfa4mqq4u89c.png)." }, { "model_id": "claude-opus-4.7", "benchmark_id": "presentbench", "score": 61.8, "reference_url": "https://seed.bytedance.com/en/blog/seed2-1-officially-released-advancing-ai-productivity", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "slide-generation environment", "sampling": "pass@1", "judge": "fine-grained rubric evaluator", "harness": "official", "prompt_style": "official default", "temperature": "unknown" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Release blog EN: General-agent chart (https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/user-upload/4xfa4mqq4u89c.png)." }, { "model_id": "claude-opus-4.7", "benchmark_id": "program_bench", "score": 52.1, "reference_url": "https://seed.bytedance.com/en/blog/seed2-1-officially-released-advancing-ai-productivity", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "Trae or benchmark-specific coding-agent environment", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official", "prompt_style": "official default", "temperature": "unknown" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Release blog EN: Coding chart (https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/user-upload/4xfa4mqq4wlbt.png)." }, { "model_id": "claude-opus-4.7", "benchmark_id": "program_bench_almost_resolved_95", "score": 2.5, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2.1/Seed2_1_Model_Card.pdf", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "compiled executable and documentation", "sampling": "pass@1", "judge": "248,000+ behavioral tests", "harness": "official ProgramBench sandbox", "prompt_style": "official default", "temperature": "unknown" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Model card Table 5 (p19)." }, { "model_id": "claude-opus-4.7", "benchmark_id": "program_bench_fully_resolved", "score": 0.0, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2.1/Seed2_1_Model_Card.pdf", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "compiled executable and documentation", "sampling": "pass@1", "judge": "248,000+ behavioral tests", "harness": "official ProgramBench sandbox", "prompt_style": "official default", "temperature": "unknown" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Model card Table 5 (p19)." }, { "model_id": "claude-opus-4.7", "benchmark_id": "realworldqa", "score": 75.6, "reference_url": "https://seed.bytedance.com/en/blog/seed2-1-officially-released-advancing-ai-productivity", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "none", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official", "prompt_style": "official default", "temperature": "unknown" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Release blog EN: Vision capability chart (https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/user-upload/4xfa4mqq4yer3.png)." }, { "model_id": "claude-opus-4.7", "benchmark_id": "scicode", "score": 56.4, "reference_url": "https://seed.bytedance.com/en/blog/seed2-1-officially-released-advancing-ai-productivity", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "none", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official", "prompt_style": "official default", "temperature": "unknown" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Release blog EN: Language/search chart (https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/user-upload/4xfa4mqq510uc.png)." }, { "model_id": "claude-opus-4.7", "benchmark_id": "seed21_trae_acceptable_delivery_rate", "score": 92.2, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2.1/Seed2_1_Model_Card.pdf", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "real-repository Trae coding environment", "sampling": "pass@1", "judge": "multidimensional human rating", "harness": "official", "prompt_style": "official default", "temperature": "unknown" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Model card textual metrics (pp15-16, p32). Acceptable delivery rate." }, { "model_id": "claude-opus-4.7", "benchmark_id": "seed21_trae_boundary_adherence", "score": 4.38, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2.1/Seed2_1_Model_Card.pdf", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "real-repository Trae coding environment", "sampling": "pass@1", "judge": "multidimensional human rating", "harness": "official", "prompt_style": "official default", "temperature": "unknown" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Model card textual metrics (pp15-16, p32). Boundary-adherence rating." }, { "model_id": "claude-opus-4.7", "benchmark_id": "seed21_trae_delivery_completeness", "score": 3.83, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2.1/Seed2_1_Model_Card.pdf", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "real-repository Trae coding environment", "sampling": "pass@1", "judge": "multidimensional human rating", "harness": "official", "prompt_style": "official default", "temperature": "unknown" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Model card textual metrics (pp15-16, p32). Delivery-completeness score." }, { "model_id": "claude-opus-4.7", "benchmark_id": "seed21_trae_fully_correct_rate", "score": 20.4, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2.1/Seed2_1_Model_Card.pdf", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "real-repository Trae coding environment", "sampling": "pass@1", "judge": "multidimensional human rating", "harness": "official", "prompt_style": "official default", "temperature": "unknown" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Model card textual metrics (pp15-16, p32). Fully correct, ready-to-use artifact rate." }, { "model_id": "claude-opus-4.7", "benchmark_id": "seed21_trae_fully_usable_rate", "score": 54.5, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2.1/Seed2_1_Model_Card.pdf", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "real-repository Trae coding environment", "sampling": "pass@1", "judge": "multidimensional human rating", "harness": "official", "prompt_style": "official default", "temperature": "unknown" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Model card textual metrics (pp15-16, p32). Top fully usable tier." }, { "model_id": "claude-opus-4.7", "benchmark_id": "seed21_trae_instruction_following", "score": 4.11, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2.1/Seed2_1_Model_Card.pdf", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "real-repository Trae coding environment", "sampling": "pass@1", "judge": "multidimensional human rating", "harness": "official", "prompt_style": "official default", "temperature": "unknown" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Model card textual metrics (pp15-16, p32). Instruction-following rating." }, { "model_id": "claude-opus-4.7", "benchmark_id": "seed21_trae_mean_rating", "score": 3.967, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2.1/Seed2_1_Model_Card.pdf", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "real-repository Trae coding environment", "sampling": "pass@1", "judge": "multidimensional human rating", "harness": "official", "prompt_style": "official default", "temperature": "unknown" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Model card textual metrics (pp15-16, p32). Mean across six rating dimensions." }, { "model_id": "claude-opus-4.7", "benchmark_id": "seed21_trae_severely_broken_rate", "score": 2.4, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2.1/Seed2_1_Model_Card.pdf", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "real-repository Trae coding environment", "sampling": "pass@1", "judge": "multidimensional human rating", "harness": "official", "prompt_style": "official default", "temperature": "unknown" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Model card textual metrics (pp15-16, p32). Severely broken delivery rate; lower is better." }, { "model_id": "claude-opus-4.7", "benchmark_id": "seed21_trae_unusable_rate", "score": 7.8, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2.1/Seed2_1_Model_Card.pdf", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "real-repository Trae coding environment", "sampling": "pass@1", "judge": "multidimensional human rating", "harness": "official", "prompt_style": "official default", "temperature": "unknown" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Model card textual metrics (pp15-16, p32). Unusable tier; lower is better." }, { "model_id": "claude-opus-4.7", "benchmark_id": "seedclawbench", "score": 64.1, "reference_url": "https://seed.bytedance.com/en/blog/seed2-1-officially-released-advancing-ai-productivity", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "OpenClaw-style tools and skills", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official", "prompt_style": "official default", "temperature": "unknown" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Release blog EN: Daily-life/tool-use chart (https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/user-upload/4xfa4mqq4ujsp.png)." }, { "model_id": "claude-opus-4.7", "benchmark_id": "seedkernelbench_avg_speedup", "score": 10.7, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2.1/Seed2_1_Model_Card.pdf", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "Trae or benchmark-specific coding-agent environment", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official", "prompt_style": "official default", "temperature": "unknown" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Model card Table 5 (p19). Source dashes are not zero." }, { "model_id": "claude-opus-4.7", "benchmark_id": "simplevqa", "score": 56.5, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2.1/Seed2_1_Model_Card.pdf", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "none", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official", "prompt_style": "official default", "temperature": "unknown" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Model card Table 9 (p51)." }, { "model_id": "claude-opus-4.7", "benchmark_id": "superchem", "score": 55.0, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2.1/Seed2_1_Model_Card.pdf", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "none", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official", "prompt_style": "official default", "temperature": "unknown" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Model card Table 11 (p53)." }, { "model_id": "claude-opus-4.7", "benchmark_id": "supergpqa", "score": 68.5, "reference_url": "https://seed.bytedance.com/en/blog/seed2-1-officially-released-advancing-ai-productivity", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "none", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official", "prompt_style": "official default", "temperature": "unknown" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Release blog EN: Language/search chart (https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/user-upload/4xfa4mqq510uc.png)." }, { "model_id": "claude-opus-4.7", "benchmark_id": "swe_atlas", "score": 38.7, "reference_url": "https://seed.bytedance.com/en/blog/seed2-1-officially-released-advancing-ai-productivity", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "Trae or benchmark-specific coding-agent environment", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official", "prompt_style": "official default", "temperature": "unknown" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Release blog EN: Coding chart (https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/user-upload/4xfa4mqq4wlbt.png)." }, { "model_id": "claude-opus-4.7", "benchmark_id": "trae_artifacts", "score": 53.0, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2.1/Seed2_1_Model_Card.pdf", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "Trae or benchmark-specific coding-agent environment", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official", "prompt_style": "official default", "temperature": "unknown" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Model card Table 5 (p19)." }, { "model_id": "claude-opus-4.7", "benchmark_id": "trae_code_gen_js", "score": 66.4, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2.1/Seed2_1_Model_Card.pdf", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "Trae or benchmark-specific coding-agent environment", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official", "prompt_style": "official default", "temperature": "unknown" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Model card Table 5 (p19)." }, { "model_id": "claude-opus-4.7", "benchmark_id": "trae_code_gen_python", "score": 83.3, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2.1/Seed2_1_Model_Card.pdf", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "Trae or benchmark-specific coding-agent environment", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official", "prompt_style": "official default", "temperature": "unknown" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Model card Table 5 (p19)." }, { "model_id": "claude-opus-4.7", "benchmark_id": "trae_error_fix_go", "score": 66.7, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2.1/Seed2_1_Model_Card.pdf", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "Trae or benchmark-specific coding-agent environment", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official", "prompt_style": "official default", "temperature": "unknown" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Model card Table 5 (p19)." }, { "model_id": "claude-opus-4.7", "benchmark_id": "trae_error_fix_java", "score": 73.3, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2.1/Seed2_1_Model_Card.pdf", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "Trae or benchmark-specific coding-agent environment", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official", "prompt_style": "official default", "temperature": "unknown" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Model card Table 5 (p19)." }, { "model_id": "claude-opus-4.7", "benchmark_id": "trae_error_fix_js", "score": 78.2, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2.1/Seed2_1_Model_Card.pdf", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "Trae or benchmark-specific coding-agent environment", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official", "prompt_style": "official default", "temperature": "unknown" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Model card Table 5 (p19)." }, { "model_id": "claude-opus-4.7", "benchmark_id": "trae_error_fix_python", "score": 76.0, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2.1/Seed2_1_Model_Card.pdf", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "Trae or benchmark-specific coding-agent environment", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official", "prompt_style": "official default", "temperature": "unknown" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Model card Table 5 (p19)." }, { "model_id": "claude-opus-4.7", "benchmark_id": "trae_repo_env", "score": 63.3, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2.1/Seed2_1_Model_Card.pdf", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "Trae or benchmark-specific coding-agent environment", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official", "prompt_style": "official default", "temperature": "unknown" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Model card Table 5 (p19)." }, { "model_id": "claude-opus-4.7", "benchmark_id": "treebench", "score": 36.8, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2.1/Seed2_1_Model_Card.pdf", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "none", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official", "prompt_style": "official default", "temperature": "unknown" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Model card Table 9 (p51)." }, { "model_id": "claude-opus-4.7", "benchmark_id": "visfactor", "score": 30.5, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2.1/Seed2_1_Model_Card.pdf", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "none", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official", "prompt_style": "official default", "temperature": "unknown" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Model card Table 9 (p51)." }, { "model_id": "claude-opus-4.7", "benchmark_id": "visulogic", "score": 32.6, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2.1/Seed2_1_Model_Card.pdf", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "none", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official", "prompt_style": "official default", "temperature": "unknown" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Model card Table 9 (p51)." }, { "model_id": "claude-opus-4.7", "benchmark_id": "vlms_are_biased", "score": 30.2, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2.1/Seed2_1_Model_Card.pdf", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "none", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official", "prompt_style": "official default", "temperature": "unknown" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Model card Table 9 (p51)." }, { "model_id": "claude-opus-4.7", "benchmark_id": "webbench", "score": 75.0, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2.1/Seed2_1_Model_Card.pdf", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "Trae or benchmark-specific coding-agent environment", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official", "prompt_style": "official default", "temperature": "unknown" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Model card Table 5 (p19)." }, { "model_id": "claude-opus-4.7", "benchmark_id": "wildclaw_bench_60_openclaw", "score": 67.0, "reference_url": "https://seed.bytedance.com/en/blog/seed2-1-officially-released-advancing-ai-productivity", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "high-resolution multimodal agent harness", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official", "prompt_style": "official default", "temperature": "unknown" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Release blog EN: Multimodal-agent chart (https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/user-upload/4xfa4mqq4v1ol.png)." }, { "model_id": "claude-opus-4.7", "benchmark_id": "workspace_bench_openclaw_100", "score": 55.1, "reference_url": "https://seed.bytedance.com/en/blog/seed2-1-officially-released-advancing-ai-productivity", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "OpenClaw workspace tools", "sampling": "pass@1", "judge": "rubric-based evaluator", "harness": "100-task OpenClaw setting", "prompt_style": "official default", "temperature": "unknown" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Release blog EN: General-agent chart (https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/user-upload/4xfa4mqq4u89c.png)." }, { "model_id": "claude-opus-4.7", "benchmark_id": "workspace_bench_openclaw_100_pass100", "score": 18.0, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2.1/Seed2_1_Model_Card.pdf", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "OpenClaw workspace tools", "sampling": "pass@1", "judge": "rubric-based evaluator", "harness": "100-task OpenClaw setting", "prompt_style": "official default", "temperature": "unknown" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Model card Table 1 (p9)." }, { "model_id": "claude-opus-4.7", "benchmark_id": "workspace_bench_openclaw_100_pass30", "score": 76.0, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2.1/Seed2_1_Model_Card.pdf", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "OpenClaw workspace tools", "sampling": "pass@1", "judge": "rubric-based evaluator", "harness": "100-task OpenClaw setting", "prompt_style": "official default", "temperature": "unknown" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Model card Table 1 (p9)." }, { "model_id": "claude-opus-4.7", "benchmark_id": "workspace_bench_openclaw_100_pass50", "score": 56.0, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2.1/Seed2_1_Model_Card.pdf", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "OpenClaw workspace tools", "sampling": "pass@1", "judge": "rubric-based evaluator", "harness": "100-task OpenClaw setting", "prompt_style": "official default", "temperature": "unknown" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Model card Table 1 (p9)." }, { "model_id": "claude-opus-4.7", "benchmark_id": "workspace_bench_openclaw_100_pass70", "score": 40.0, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2.1/Seed2_1_Model_Card.pdf", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "OpenClaw workspace tools", "sampling": "pass@1", "judge": "rubric-based evaluator", "harness": "100-task OpenClaw setting", "prompt_style": "official default", "temperature": "unknown" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Model card Table 1 (p9)." }, { "model_id": "claude-opus-4.7", "benchmark_id": "workspace_bench_openclaw_100_pass90", "score": 20.0, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2.1/Seed2_1_Model_Card.pdf", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "OpenClaw workspace tools", "sampling": "pass@1", "judge": "rubric-based evaluator", "harness": "100-task OpenClaw setting", "prompt_style": "official default", "temperature": "unknown" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Model card Table 1 (p9)." }, { "model_id": "claude-opus-4.7", "benchmark_id": "worldbench", "score": 52.9, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2.1/Seed2_1_Model_Card.pdf", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "none", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official", "prompt_style": "official default", "temperature": "unknown" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Model card Table 9 (p51)." }, { "model_id": "claude-opus-4.7", "benchmark_id": "worldvqa", "score": 35.9, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2.1/Seed2_1_Model_Card.pdf", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "none", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official", "prompt_style": "official default", "temperature": "unknown" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Model card Table 9 (p51)." }, { "model_id": "claude-opus-4.7", "benchmark_id": "xdailybench", "score": 69.0, "reference_url": "https://seed.bytedance.com/en/blog/seed2-1-officially-released-advancing-ai-productivity", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "none", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official", "prompt_style": "official default", "temperature": "unknown" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Release blog EN: Daily-life/tool-use chart (https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/user-upload/4xfa4mqq4ujsp.png)." }, { "model_id": "claude-opus-4.7", "benchmark_id": "zerobench_main", "score": 8.0, "reference_url": "https://seed.bytedance.com/en/blog/seed2-1-officially-released-advancing-ai-productivity", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "none", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official", "prompt_style": "official default", "temperature": "unknown" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Release blog EN: Vision capability chart (https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/user-upload/4xfa4mqq4yer3.png)." }, { "model_id": "claude-opus-4.7", "benchmark_id": "zerobench_sub", "score": 37.1, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2.1/Seed2_1_Model_Card.pdf", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "none", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official", "prompt_style": "official default", "temperature": "unknown" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Model card Table 9 (p51)." }, { "model_id": "claude-opus-4.8", "benchmark_id": "code_arena_frontend_elo", "score": 1565, "reference_url": "https://seed.bytedance.com/en/blog/seed2-1-officially-released-advancing-ai-productivity", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "none", "sampling": "107,962-vote Arena snapshot", "judge": "human preference", "harness": "Arena Code leaderboard", "prompt_style": "official default", "temperature": "unknown" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Release blog EN: Code Arena: Frontend snapshot (https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/user-upload/4xfa4mqq4xsvz.jpg). Historical leaderboard snapshot reproduced by ByteDance.", "candidates": [ { "score": 1542, "reference_url": "https://seed.bytedance.com/en/blog/seed2-1-officially-released-advancing-ai-productivity", "source_type": "official_blog", "reported_setting": { "mode": "non-thinking", "effort": "default", "tools": "none", "sampling": "107,962-vote Arena snapshot", "judge": "human preference", "harness": "Arena Code leaderboard", "prompt_style": "official default", "temperature": "unknown" }, "notes": "Release blog EN: Code Arena: Frontend snapshot (https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/user-upload/4xfa4mqq4xsvz.jpg). Historical leaderboard snapshot reproduced by ByteDance." } ] }, { "model_id": "claude-opus-4.8", "benchmark_id": "seed21_dirtyfilter_precision", "score": 94.8, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2.1/Seed2_1_Model_Card.pdf", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "rule-only evaluator/optimizer loop", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "300M-token budget; identical validation set", "prompt_style": "official default", "temperature": "unknown" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Model card Figure 19 (p32)." }, { "model_id": "claude-opus-4.8", "benchmark_id": "seed21_dirtyfilter_validation_recall", "score": 51.6, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2.1/Seed2_1_Model_Card.pdf", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "rule-only evaluator/optimizer loop", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "300M-token budget; identical validation set", "prompt_style": "official default", "temperature": "unknown" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Model card Figure 19 (p32)." }, { "model_id": "claude-sonnet-4.6", "benchmark_id": "code_arena_frontend_elo", "score": 1521, "reference_url": "https://seed.bytedance.com/en/blog/seed2-1-officially-released-advancing-ai-productivity", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "none", "sampling": "107,962-vote Arena snapshot", "judge": "human preference", "harness": "Arena Code leaderboard", "prompt_style": "official default", "temperature": "unknown" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Release blog EN: Code Arena: Frontend snapshot (https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/user-upload/4xfa4mqq4xsvz.jpg). Historical leaderboard snapshot reproduced by ByteDance." }, { "model_id": "doubao-seed-2.1-deep-think", "benchmark_id": "frontier_science_research", "score": 40.7, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2.1/Seed2_1_Model_Card.pdf", "reported_setting": { "mode": "deep-think", "effort": "unknown", "tools": "web search + sandboxed code execution", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "reason -> verify -> revise -> select loop", "prompt_style": "official default", "temperature": "unknown" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Model card Table 7 (p24)." }, { "model_id": "doubao-seed-2.1-deep-think", "benchmark_id": "imo_2025", "score": 81.0, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2.1/Seed2_1_Model_Card.pdf", "reported_setting": { "mode": "deep-think", "effort": "unknown", "tools": "web search + sandboxed code execution", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "reason -> verify -> revise -> select loop", "prompt_style": "official default", "temperature": "unknown" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Model card Table 7 (p24)." }, { "model_id": "doubao-seed-2.1-deep-think", "benchmark_id": "imo_proofbench_advanced", "score": 83.8, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2.1/Seed2_1_Model_Card.pdf", "reported_setting": { "mode": "deep-think", "effort": "unknown", "tools": "web search + sandboxed code execution", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "reason -> verify -> revise -> select loop", "prompt_style": "official default", "temperature": "unknown" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Model card Table 7 (p24)." }, { "model_id": "doubao-seed-2.1-deep-think", "benchmark_id": "ipho_2025_theory", "score": 89.0, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2.1/Seed2_1_Model_Card.pdf", "reported_setting": { "mode": "deep-think", "effort": "unknown", "tools": "web search + sandboxed code execution", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "reason -> verify -> revise -> select loop", "prompt_style": "official default", "temperature": "unknown" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Model card Table 7 (p24)." }, { "model_id": "doubao-seed-2.1-pro", "benchmark_id": "aethercode", "score": 65.8, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2.1/Seed2_1_Model_Card.pdf", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "none", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official", "prompt_style": "official default", "temperature": "unknown" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Model card Table 11 (p53)." }, { "model_id": "doubao-seed-2.1-pro", "benchmark_id": "agent_startup_bench", "score": 68.8, "reference_url": "https://seed.bytedance.com/en/blog/seed2-1-officially-released-advancing-ai-productivity", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "research and professional-deliverable tools", "sampling": "pass@1", "judge": "expert review", "harness": "official", "prompt_style": "official default", "temperature": "unknown" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "Release blog EN: General-agent chart (https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/user-upload/4xfa4mqq4u89c.png)." }, { "model_id": "doubao-seed-2.1-pro", "benchmark_id": "agents_last_exam", "score": 19.5, "reference_url": "https://seed.bytedance.com/en/blog/seed2-1-officially-released-advancing-ai-productivity", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "computer-use environment", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official", "prompt_style": "official default", "temperature": "unknown" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "Release blog EN: General-agent chart (https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/user-upload/4xfa4mqq4u89c.png). Turbo is a source dash, not zero." }, { "model_id": "doubao-seed-2.1-pro", "benchmark_id": "agents_last_exam_average_score", "score": 41.4, "reference_url": "https://seed.bytedance.com/en/blog/seed2-1-officially-released-advancing-ai-productivity", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "computer-use environment", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official", "prompt_style": "official default", "temperature": "unknown" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "Release blog EN: General-agent chart (https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/user-upload/4xfa4mqq4u89c.png). Turbo is a source dash, not zero." }, { "model_id": "doubao-seed-2.1-pro", "benchmark_id": "arc_agi_2", "score": 62.5, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2.1/Seed2_1_Model_Card.pdf", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "none", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official", "prompt_style": "official default", "temperature": "unknown" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Model card Table 11 (p53)." }, { "model_id": "doubao-seed-2.1-pro", "benchmark_id": "babyvision", "score": 73.7, "reference_url": "https://seed.bytedance.com/en/blog/seed2-1-officially-released-advancing-ai-productivity", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "none", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official", "prompt_style": "official default", "temperature": "unknown" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "Release blog EN: Vision capability chart (https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/user-upload/4xfa4mqq4yer3.png)." }, { "model_id": "doubao-seed-2.1-pro", "benchmark_id": "beyond_aime", "score": 87.0, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2.1/Seed2_1_Model_Card.pdf", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "none", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official", "prompt_style": "official default", "temperature": "unknown" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Model card Table 11 (p53)." }, { "model_id": "doubao-seed-2.1-pro", "benchmark_id": "blink", "score": 81.4, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2.1/Seed2_1_Model_Card.pdf", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "none", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official", "prompt_style": "official default", "temperature": "unknown" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Model card Table 9 (p51)." }, { "model_id": "doubao-seed-2.1-pro", "benchmark_id": "chartqapro", "score": 70.9, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2.1/Seed2_1_Model_Card.pdf", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "none", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official", "prompt_style": "official default", "temperature": "unknown" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Model card Table 9 (p51)." }, { "model_id": "doubao-seed-2.1-pro", "benchmark_id": "charxiv_descriptive", "score": 95.5, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2.1/Seed2_1_Model_Card.pdf", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "none", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official", "prompt_style": "official default", "temperature": "unknown" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Model card Table 9 (p51)." }, { "model_id": "doubao-seed-2.1-pro", "benchmark_id": "charxiv_reasoning", "score": 85.4, "reference_url": "https://seed.bytedance.com/en/blog/seed2-1-officially-released-advancing-ai-productivity", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "none", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official", "prompt_style": "official default", "temperature": "unknown" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "Release blog EN: Vision capability chart (https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/user-upload/4xfa4mqq4yer3.png).", "candidates": [ { "score": 86.4, "reference_url": "https://seed.bytedance.com/en/blog/seed2-1-officially-released-advancing-ai-productivity", "source_type": "official_blog", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "source tool-augmented setting; exact tools undisclosed", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official", "prompt_style": "official default", "temperature": "unknown" }, "notes": "Release blog EN: Vision capability chart (https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/user-upload/4xfa4mqq4yer3.png). Parenthesized source value." } ] }, { "model_id": "doubao-seed-2.1-pro", "benchmark_id": "claw_eval_multimodal_pass3", "score": 51.0, "reference_url": "https://seed.bytedance.com/en/blog/seed2-1-officially-released-advancing-ai-productivity", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "high-resolution multimodal agent harness", "sampling": "Pass^3", "judge": "benchmark-specified", "harness": "official", "prompt_style": "official default", "temperature": "unknown" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "Release blog EN: Multimodal-agent chart (https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/user-upload/4xfa4mqq4v1ol.png)." }, { "model_id": "doubao-seed-2.1-pro", "benchmark_id": "contphy", "score": 63.6, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2.1/Seed2_1_Model_Card.pdf", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "none", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official", "prompt_style": "official default", "temperature": "unknown" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Model card Table 10 (p52)." }, { "model_id": "doubao-seed-2.1-pro", "benchmark_id": "creativework", "score": 42.5, "reference_url": "https://seed.bytedance.com/en/blog/seed2-1-officially-released-advancing-ai-productivity", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "GUI + MCP in Notion, Canva, and Figma", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official", "prompt_style": "official default", "temperature": "unknown" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "Release blog EN: Computer-use chart (https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/user-upload/4xfa4mqq4w8x3.png)." }, { "model_id": "doubao-seed-2.1-pro", "benchmark_id": "crossvid", "score": 65.0, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2.1/Seed2_1_Model_Card.pdf", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "none", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official", "prompt_style": "official default", "temperature": "unknown" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Model card Table 10 (p52)." }, { "model_id": "doubao-seed-2.1-pro", "benchmark_id": "cybergym", "score": 68.7, "reference_url": "https://seed.bytedance.com/en/blog/seed2-1-officially-released-advancing-ai-productivity", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "cybersecurity agent environment", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official", "prompt_style": "official default", "temperature": "unknown" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "Release blog EN: Coding chart (https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/user-upload/4xfa4mqq4wlbt.png). Release blog reports Pro=68.7; PDF Table 5 reports Pro=70.2.", "candidates": [ { "score": 70.2, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2.1/Seed2_1_Model_Card.pdf", "source_type": "tech_report", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "Trae or benchmark-specific coding-agent environment", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official", "prompt_style": "official default", "temperature": "unknown" }, "notes": "Model card Table 5 (p19). Source dashes are not zero." } ] }, { "model_id": "doubao-seed-2.1-pro", "benchmark_id": "doubao_multi_turn_bench", "score": 52.5, "reference_url": "https://seed.bytedance.com/en/blog/seed2-1-officially-released-advancing-ai-productivity", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "none", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official", "prompt_style": "official default", "temperature": "unknown" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "Release blog EN: Daily-life/tool-use chart (https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/user-upload/4xfa4mqq4ujsp.png)." }, { "model_id": "doubao-seed-2.1-pro", "benchmark_id": "dude", "score": 82.8, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2.1/Seed2_1_Model_Card.pdf", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "none", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official", "prompt_style": "official default", "temperature": "unknown" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Model card Table 9 (p51)." }, { "model_id": "doubao-seed-2.1-pro", "benchmark_id": "dynamath", "score": 73.1, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2.1/Seed2_1_Model_Card.pdf", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "none", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official", "prompt_style": "official default", "temperature": "unknown" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Model card Table 9 (p51)." }, { "model_id": "doubao-seed-2.1-pro", "benchmark_id": "embspatial_bench", "score": 83.4, "reference_url": "https://seed.bytedance.com/en/blog/seed2-1-officially-released-advancing-ai-productivity", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "none", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official", "prompt_style": "official default", "temperature": "unknown" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "Release blog EN: Spatial/long-context chart (https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/user-upload/4xfa4mqq4z2l5.png)." }, { "model_id": "doubao-seed-2.1-pro", "benchmark_id": "emma", "score": 79.3, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2.1/Seed2_1_Model_Card.pdf", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "none", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official", "prompt_style": "official default", "temperature": "unknown" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Model card Table 9 (p51)." }, { "model_id": "doubao-seed-2.1-pro", "benchmark_id": "erqa", "score": 72.0, "reference_url": "https://seed.bytedance.com/en/blog/seed2-1-officially-released-advancing-ai-productivity", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "none", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official", "prompt_style": "official default", "temperature": "unknown" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "Release blog EN: Spatial/long-context chart (https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/user-upload/4xfa4mqq4z2l5.png)." }, { "model_id": "doubao-seed-2.1-pro", "benchmark_id": "finance_agent", "score": 60.7, "reference_url": "https://seed.bytedance.com/en/blog/seed2-1-officially-released-advancing-ai-productivity", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "financial-analysis agent tools", "sampling": "pass@1", "judge": "LLM-as-judge rubric and contradiction grader", "harness": "official", "prompt_style": "official default", "temperature": "unknown" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "Release blog EN: General-agent chart (https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/user-upload/4xfa4mqq4u89c.png)." }, { "model_id": "doubao-seed-2.1-pro", "benchmark_id": "frontiercs_algorithmic_v1", "score": 28.2, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2.1/Seed2_1_Model_Card.pdf", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "research and code-execution environment", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official", "prompt_style": "official default", "temperature": "unknown" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Model card Table 6 (p24). Claude Opus 4.7 is a source dash; official sub-score name unresolved." }, { "model_id": "doubao-seed-2.1-pro", "benchmark_id": "frontiercs_overall_v1", "score": 29.1, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2.1/Seed2_1_Model_Card.pdf", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "research and code-execution environment", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official", "prompt_style": "official default", "temperature": "unknown" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Model card Table 6 (p24). Claude Opus 4.7 is a source dash; official sub-score name unresolved." }, { "model_id": "doubao-seed-2.1-pro", "benchmark_id": "frontiercs_research_v1", "score": 46.3, "reference_url": "https://seed.bytedance.com/en/blog/seed2-1-officially-released-advancing-ai-productivity", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "research and code-execution environment", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official", "prompt_style": "official default", "temperature": "unknown" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "Release blog EN: Frontier-research chart (https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/user-upload/4xfa4mqq52cdr.png). Claude Opus 4.7 is a source dash; official sub-score name unresolved." }, { "model_id": "doubao-seed-2.1-pro", "benchmark_id": "fs_researcher", "score": 28.3, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2.1/Seed2_1_Model_Card.pdf", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "none", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official", "prompt_style": "official default", "temperature": "unknown" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Model card Table 11 (p53)." }, { "model_id": "doubao-seed-2.1-pro", "benchmark_id": "gameworld", "score": 31.2, "reference_url": "https://seed.bytedance.com/en/blog/seed2-1-officially-released-advancing-ai-productivity", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "browser-game GUI actions", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official", "prompt_style": "official default", "temperature": "unknown" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "Release blog EN: Computer-use chart (https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/user-upload/4xfa4mqq4w8x3.png)." }, { "model_id": "doubao-seed-2.1-pro", "benchmark_id": "gdpval_seed_reported_score", "score": 87.9, "reference_url": "https://seed.bytedance.com/en/blog/seed2-1-officially-released-advancing-ai-productivity", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "document/spreadsheet deliverable workflow", "sampling": "pass@1", "judge": "source-specific evaluator", "harness": "official", "prompt_style": "official default", "temperature": "unknown" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "Release blog EN: General-agent chart (https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/user-upload/4xfa4mqq4u89c.png)." }, { "model_id": "doubao-seed-2.1-pro", "benchmark_id": "hle_verified", "score": 42.9, "reference_url": "https://seed.bytedance.com/en/blog/seed2-1-officially-released-advancing-ai-productivity", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "none", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official", "prompt_style": "official default", "temperature": "unknown" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "Release blog EN: Language/search chart (https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/user-upload/4xfa4mqq510uc.png)." }, { "model_id": "doubao-seed-2.1-pro", "benchmark_id": "image2floorplan_avg_score", "score": 48.0, "reference_url": "https://seed.bytedance.com/en/blog/seed2-1-officially-released-advancing-ai-productivity", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "high-resolution multimodal agent harness", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official", "prompt_style": "official default", "temperature": "unknown" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "Release blog EN: Multimodal-agent chart (https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/user-upload/4xfa4mqq4v1ol.png)." }, { "model_id": "doubao-seed-2.1-pro", "benchmark_id": "imo_2025", "score": 65.2, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2.1/Seed2_1_Model_Card.pdf", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "Deep Think section baseline; exact tools undisclosed", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "Deep Think comparison baseline", "prompt_style": "official default", "temperature": "unknown" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Model card Table 7 (p24)." }, { "model_id": "doubao-seed-2.1-pro", "benchmark_id": "imo_proofbench_advanced", "score": 54.3, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2.1/Seed2_1_Model_Card.pdf", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "Deep Think section baseline; exact tools undisclosed", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "Deep Think comparison baseline", "prompt_style": "official default", "temperature": "unknown" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Model card Table 7 (p24)." }, { "model_id": "doubao-seed-2.1-pro", "benchmark_id": "ipho_2025_theory", "score": 79.3, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2.1/Seed2_1_Model_Card.pdf", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "Deep Think section baseline; exact tools undisclosed", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "Deep Think comparison baseline", "prompt_style": "official default", "temperature": "unknown" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Model card Table 7 (p24)." }, { "model_id": "doubao-seed-2.1-pro", "benchmark_id": "kina", "score": 48.3, "reference_url": "https://seed.bytedance.com/en/blog/seed2-1-officially-released-advancing-ai-productivity", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "none", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official", "prompt_style": "official default", "temperature": "unknown" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "Release blog EN: Language/search chart (https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/user-upload/4xfa4mqq510uc.png)." }, { "model_id": "doubao-seed-2.1-pro", "benchmark_id": "live_mathematician_bench_2026_06", "score": 20.9, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2.1/Seed2_1_Model_Card.pdf", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "none", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official", "prompt_style": "official default", "temperature": "unknown" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Model card Table 11 (p53)." }, { "model_id": "doubao-seed-2.1-pro", "benchmark_id": "livesports_3k", "score": 76.8, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2.1/Seed2_1_Model_Card.pdf", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "none", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official", "prompt_style": "official default", "temperature": "unknown" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Model card Table 10 (p52)." }, { "model_id": "doubao-seed-2.1-pro", "benchmark_id": "longvideobench", "score": 80.6, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2.1/Seed2_1_Model_Card.pdf", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "none", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official", "prompt_style": "official default", "temperature": "unknown" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Model card Table 10 (p52)." }, { "model_id": "doubao-seed-2.1-pro", "benchmark_id": "lvbench", "score": 78.0, "reference_url": "https://seed.bytedance.com/en/blog/seed2-1-officially-released-advancing-ai-productivity", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "none", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official", "prompt_style": "official default", "temperature": "unknown" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "Release blog EN: Long/streaming-video chart (https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/user-upload/4xfa4mqq50ia3.png)." }, { "model_id": "doubao-seed-2.1-pro", "benchmark_id": "mathverse_vision_only", "score": 89.7, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2.1/Seed2_1_Model_Card.pdf", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "none", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official", "prompt_style": "official default", "temperature": "unknown" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Model card Table 9 (p51)." }, { "model_id": "doubao-seed-2.1-pro", "benchmark_id": "mathvision", "score": 92.6, "reference_url": "https://seed.bytedance.com/en/blog/seed2-1-officially-released-advancing-ai-productivity", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "none", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official", "prompt_style": "official default", "temperature": "unknown" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "Release blog EN: Vision capability chart (https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/user-upload/4xfa4mqq4yer3.png).", "candidates": [ { "score": 94.5, "reference_url": "https://seed.bytedance.com/en/blog/seed2-1-officially-released-advancing-ai-productivity", "source_type": "official_blog", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "source tool-augmented setting; exact tools undisclosed", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official", "prompt_style": "official default", "temperature": "unknown" }, "notes": "Release blog EN: Vision capability chart (https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/user-upload/4xfa4mqq4yer3.png). Parenthesized source value." } ] }, { "model_id": "doubao-seed-2.1-pro", "benchmark_id": "mathvista", "score": 90.7, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2.1/Seed2_1_Model_Card.pdf", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "none", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official", "prompt_style": "official default", "temperature": "unknown" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Model card Table 9 (p51)." }, { "model_id": "doubao-seed-2.1-pro", "benchmark_id": "measurebench", "score": 62.9, "reference_url": "https://seed.bytedance.com/en/blog/seed2-1-officially-released-advancing-ai-productivity", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "none", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official", "prompt_style": "official default", "temperature": "unknown" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "Release blog EN: Vision capability chart (https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/user-upload/4xfa4mqq4yer3.png)." }, { "model_id": "doubao-seed-2.1-pro", "benchmark_id": "minerva", "score": 70.7, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2.1/Seed2_1_Model_Card.pdf", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "none", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official", "prompt_style": "official default", "temperature": "unknown" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Model card Table 10 (p52). Source marks this row with an unexplained dagger. Source marks this row with an unexplained dagger." }, { "model_id": "doubao-seed-2.1-pro", "benchmark_id": "mmlongbench", "score": 78.3, "reference_url": "https://seed.bytedance.com/en/blog/seed2-1-officially-released-advancing-ai-productivity", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "none", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official", "prompt_style": "official default", "temperature": "unknown" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "Release blog EN: Spatial/long-context chart (https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/user-upload/4xfa4mqq4z2l5.png). Source dashes are not zero." }, { "model_id": "doubao-seed-2.1-pro", "benchmark_id": "mmmu_pro", "score": 81.6, "reference_url": "https://seed.bytedance.com/en/blog/seed2-1-officially-released-advancing-ai-productivity", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "none", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official", "prompt_style": "official default", "temperature": "unknown" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "Release blog EN: Vision capability chart (https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/user-upload/4xfa4mqq4yer3.png).", "candidates": [ { "score": 82.7, "reference_url": "https://seed.bytedance.com/en/blog/seed2-1-officially-released-advancing-ai-productivity", "source_type": "official_blog", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "source tool-augmented setting; exact tools undisclosed", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official", "prompt_style": "official default", "temperature": "unknown" }, "notes": "Release blog EN: Vision capability chart (https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/user-upload/4xfa4mqq4yer3.png). Parenthesized source value." } ] }, { "model_id": "doubao-seed-2.1-pro", "benchmark_id": "mmsibench_circular", "score": 35.9, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2.1/Seed2_1_Model_Card.pdf", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "none", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official", "prompt_style": "official default", "temperature": "unknown" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Model card Table 9 (p51)." }, { "model_id": "doubao-seed-2.1-pro", "benchmark_id": "mobileworld", "score": 73.1, "reference_url": "https://seed.bytedance.com/en/blog/seed2-1-officially-released-advancing-ai-productivity", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "mobile GUI actions", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official", "prompt_style": "official default", "temperature": "unknown" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "Release blog EN: Computer-use chart (https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/user-upload/4xfa4mqq4w8x3.png)." }, { "model_id": "doubao-seed-2.1-pro", "benchmark_id": "motionbench", "score": 74.9, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2.1/Seed2_1_Model_Card.pdf", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "none", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official", "prompt_style": "official default", "temperature": "unknown" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Model card Table 10 (p52)." }, { "model_id": "doubao-seed-2.1-pro", "benchmark_id": "msqa", "score": 50.2, "reference_url": "https://seed.bytedance.com/en/blog/seed2-1-officially-released-advancing-ai-productivity", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "none", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official", "prompt_style": "official default", "temperature": "unknown" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "Release blog EN: Language/search chart (https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/user-upload/4xfa4mqq510uc.png)." }, { "model_id": "doubao-seed-2.1-pro", "benchmark_id": "ocrbench_v2", "score": 63.2, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2.1/Seed2_1_Model_Card.pdf", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "none", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official", "prompt_style": "official default", "temperature": "unknown" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Model card Table 9 (p51)." }, { "model_id": "doubao-seed-2.1-pro", "benchmark_id": "officeqa_pro", "score": 72.2, "reference_url": "https://seed.bytedance.com/en/blog/seed2-1-officially-released-advancing-ai-productivity", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "high-resolution multimodal agent harness", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official", "prompt_style": "official default", "temperature": "unknown" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "Release blog EN: Multimodal-agent chart (https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/user-upload/4xfa4mqq4v1ol.png).", "candidates": [ { "score": 70.9, "reference_url": "https://seed.bytedance.com/en/blog/seed2-1-officially-released-advancing-ai-productivity", "source_type": "official_blog", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "document and rendered-image analysis", "sampling": "pass@1", "judge": "OfficeQA Pro evaluator", "harness": "official", "prompt_style": "official default", "temperature": "unknown" }, "notes": "Release blog EN: General-agent chart (https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/user-upload/4xfa4mqq4u89c.png). The source says its internal implementations and adaptations can fluctuate relative to the official leaderboard." } ] }, { "model_id": "doubao-seed-2.1-pro", "benchmark_id": "one_million_bench", "score": 68.8, "reference_url": "https://seed.bytedance.com/en/blog/seed2-1-officially-released-advancing-ai-productivity", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "agentic professional-task environment", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official", "prompt_style": "official default", "temperature": "unknown" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "Release blog EN: General-agent chart (https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/user-upload/4xfa4mqq4u89c.png)." }, { "model_id": "doubao-seed-2.1-pro", "benchmark_id": "osworld", "score": 78.8, "reference_url": "https://seed.bytedance.com/en/blog/seed2-1-officially-released-advancing-ai-productivity", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "GUI + shell/commands + filesystem tools", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official", "prompt_style": "official default", "temperature": "unknown" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "Release blog EN: Computer-use chart (https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/user-upload/4xfa4mqq4w8x3.png).", "candidates": [ { "score": 72.6, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2.1/Seed2_1_Model_Card.pdf", "source_type": "tech_report", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "GUI-only", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "OSWorld GUI-only", "prompt_style": "official default", "temperature": "unknown" }, "notes": "Model card Figure 5 (p13). Figure 5 GUI-only operating point." } ] }, { "model_id": "doubao-seed-2.1-pro", "benchmark_id": "ovbench", "score": 70.0, "reference_url": "https://seed.bytedance.com/en/blog/seed2-1-officially-released-advancing-ai-productivity", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "none", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official", "prompt_style": "official default", "temperature": "unknown" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "Release blog EN: Long/streaming-video chart (https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/user-upload/4xfa4mqq50ia3.png)." }, { "model_id": "doubao-seed-2.1-pro", "benchmark_id": "ovobench", "score": 80.7, "reference_url": "https://seed.bytedance.com/en/blog/seed2-1-officially-released-advancing-ai-productivity", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "none", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official", "prompt_style": "official default", "temperature": "unknown" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "Release blog EN: Long/streaming-video chart (https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/user-upload/4xfa4mqq50ia3.png)." }, { "model_id": "doubao-seed-2.1-pro", "benchmark_id": "posttrain_bench", "score": 16.5, "reference_url": "https://seed.bytedance.com/en/blog/seed2-1-officially-released-advancing-ai-productivity", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "research-oriented agent environment", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official", "prompt_style": "official default", "temperature": "unknown" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "Release blog EN: Frontier-research chart (https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/user-upload/4xfa4mqq52cdr.png). Source dashes are not zero." }, { "model_id": "doubao-seed-2.1-pro", "benchmark_id": "presentbench", "score": 54.6, "reference_url": "https://seed.bytedance.com/en/blog/seed2-1-officially-released-advancing-ai-productivity", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "slide-generation environment", "sampling": "pass@1", "judge": "fine-grained rubric evaluator", "harness": "official", "prompt_style": "official default", "temperature": "unknown" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "Release blog EN: General-agent chart (https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/user-upload/4xfa4mqq4u89c.png)." }, { "model_id": "doubao-seed-2.1-pro", "benchmark_id": "program_bench", "score": 50.3, "reference_url": "https://seed.bytedance.com/en/blog/seed2-1-officially-released-advancing-ai-productivity", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "Trae or benchmark-specific coding-agent environment", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official", "prompt_style": "official default", "temperature": "unknown" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "Release blog EN: Coding chart (https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/user-upload/4xfa4mqq4wlbt.png)." }, { "model_id": "doubao-seed-2.1-pro", "benchmark_id": "program_bench_almost_resolved_95", "score": 1.0, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2.1/Seed2_1_Model_Card.pdf", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "compiled executable and documentation", "sampling": "pass@1", "judge": "248,000+ behavioral tests", "harness": "official ProgramBench sandbox", "prompt_style": "official default", "temperature": "unknown" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Model card Table 5 (p19)." }, { "model_id": "doubao-seed-2.1-pro", "benchmark_id": "program_bench_fully_resolved", "score": 0.0, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2.1/Seed2_1_Model_Card.pdf", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "compiled executable and documentation", "sampling": "pass@1", "judge": "248,000+ behavioral tests", "harness": "official ProgramBench sandbox", "prompt_style": "official default", "temperature": "unknown" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Model card Table 5 (p19)." }, { "model_id": "doubao-seed-2.1-pro", "benchmark_id": "realworldqa", "score": 86.7, "reference_url": "https://seed.bytedance.com/en/blog/seed2-1-officially-released-advancing-ai-productivity", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "none", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official", "prompt_style": "official default", "temperature": "unknown" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "Release blog EN: Vision capability chart (https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/user-upload/4xfa4mqq4yer3.png)." }, { "model_id": "doubao-seed-2.1-pro", "benchmark_id": "scicode", "score": 59.8, "reference_url": "https://seed.bytedance.com/en/blog/seed2-1-officially-released-advancing-ai-productivity", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "none", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official", "prompt_style": "official default", "temperature": "unknown" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "Release blog EN: Language/search chart (https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/user-upload/4xfa4mqq510uc.png)." }, { "model_id": "doubao-seed-2.1-pro", "benchmark_id": "seed21_claudecode_opus46_loss_rate", "score": 29.6, "reference_url": "https://seed.bytedance.com/en/blog/seed2-1-officially-released-advancing-ai-productivity", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "real-repository coding environment", "sampling": "pass@1", "judge": "anonymous developer preference", "harness": "official", "prompt_style": "official default", "temperature": "unknown" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "Release blog EN: Crowdsourced coding chart (https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/user-upload/4xfa4mqq4wx8r.png)." }, { "model_id": "doubao-seed-2.1-pro", "benchmark_id": "seed21_claudecode_opus46_net_win_rate", "score": 29.6, "reference_url": "https://seed.bytedance.com/en/blog/seed2-1-officially-released-advancing-ai-productivity", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "real-repository coding environment", "sampling": "pass@1", "judge": "anonymous developer preference", "harness": "official", "prompt_style": "official default", "temperature": "unknown" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "Release blog EN: Crowdsourced coding chart (https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/user-upload/4xfa4mqq4wx8r.png). Source prints +29.6 percentage points." }, { "model_id": "doubao-seed-2.1-pro", "benchmark_id": "seed21_claudecode_opus46_tie_rate", "score": 11.3, "reference_url": "https://seed.bytedance.com/en/blog/seed2-1-officially-released-advancing-ai-productivity", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "real-repository coding environment", "sampling": "pass@1", "judge": "anonymous developer preference", "harness": "official", "prompt_style": "official default", "temperature": "unknown" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "Release blog EN: Crowdsourced coding chart (https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/user-upload/4xfa4mqq4wx8r.png)." }, { "model_id": "doubao-seed-2.1-pro", "benchmark_id": "seed21_claudecode_opus46_win_rate", "score": 59.1, "reference_url": "https://seed.bytedance.com/en/blog/seed2-1-officially-released-advancing-ai-productivity", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "real-repository coding environment", "sampling": "pass@1", "judge": "anonymous developer preference", "harness": "official", "prompt_style": "official default", "temperature": "unknown" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "Release blog EN: Crowdsourced coding chart (https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/user-upload/4xfa4mqq4wx8r.png). 136 wins / 26 ties / 68 losses over 230 comparisons." }, { "model_id": "doubao-seed-2.1-pro", "benchmark_id": "seed21_dirtyfilter_precision", "score": 94.5, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2.1/Seed2_1_Model_Card.pdf", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "rule-only evaluator/optimizer loop", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "300M-token budget; identical validation set", "prompt_style": "official default", "temperature": "unknown" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Model card Figure 19 (p32)." }, { "model_id": "doubao-seed-2.1-pro", "benchmark_id": "seed21_dirtyfilter_validation_recall", "score": 43.3, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2.1/Seed2_1_Model_Card.pdf", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "rule-only evaluator/optimizer loop", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "300M-token budget; identical validation set", "prompt_style": "official default", "temperature": "unknown" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Model card Figure 19 (p32).", "candidates": [ { "score": 43.4, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2.1/Seed2_1_Model_Card.pdf", "source_type": "tech_report", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "rule-only evaluator/optimizer loop", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "Figure 19 green chart endpoint", "prompt_style": "official default", "temperature": "unknown" }, "notes": "Model card Figure 19 (p32). The plotted green endpoint is labelled 43.4%, while the final summary table reports 43.3%; retain both source observations." } ] }, { "model_id": "doubao-seed-2.1-pro", "benchmark_id": "seed21_trae_acceptable_delivery_rate", "score": 94.0, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2.1/Seed2_1_Model_Card.pdf", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "real-repository Trae coding environment", "sampling": "pass@1", "judge": "multidimensional human rating", "harness": "official", "prompt_style": "official default", "temperature": "unknown" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Model card textual metrics (pp15-16, p32). Acceptable delivery rate." }, { "model_id": "doubao-seed-2.1-pro", "benchmark_id": "seed21_trae_boundary_adherence", "score": 4.16, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2.1/Seed2_1_Model_Card.pdf", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "real-repository Trae coding environment", "sampling": "pass@1", "judge": "multidimensional human rating", "harness": "official", "prompt_style": "official default", "temperature": "unknown" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Model card textual metrics (pp15-16, p32). Boundary-adherence rating." }, { "model_id": "doubao-seed-2.1-pro", "benchmark_id": "seed21_trae_delivery_completeness", "score": 3.97, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2.1/Seed2_1_Model_Card.pdf", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "real-repository Trae coding environment", "sampling": "pass@1", "judge": "multidimensional human rating", "harness": "official", "prompt_style": "official default", "temperature": "unknown" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Model card textual metrics (pp15-16, p32). Delivery-completeness score." }, { "model_id": "doubao-seed-2.1-pro", "benchmark_id": "seed21_trae_fully_correct_rate", "score": 29.3, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2.1/Seed2_1_Model_Card.pdf", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "real-repository Trae coding environment", "sampling": "pass@1", "judge": "multidimensional human rating", "harness": "official", "prompt_style": "official default", "temperature": "unknown" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Model card textual metrics (pp15-16, p32). Fully correct, ready-to-use artifact rate." }, { "model_id": "doubao-seed-2.1-pro", "benchmark_id": "seed21_trae_fully_usable_rate", "score": 58.7, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2.1/Seed2_1_Model_Card.pdf", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "real-repository Trae coding environment", "sampling": "pass@1", "judge": "multidimensional human rating", "harness": "official", "prompt_style": "official default", "temperature": "unknown" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Model card textual metrics (pp15-16, p32). Top fully usable tier." }, { "model_id": "doubao-seed-2.1-pro", "benchmark_id": "seed21_trae_instruction_following", "score": 3.96, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2.1/Seed2_1_Model_Card.pdf", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "real-repository Trae coding environment", "sampling": "pass@1", "judge": "multidimensional human rating", "harness": "official", "prompt_style": "official default", "temperature": "unknown" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Model card textual metrics (pp15-16, p32). Instruction-following rating." }, { "model_id": "doubao-seed-2.1-pro", "benchmark_id": "seed21_trae_mean_rating", "score": 3.967, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2.1/Seed2_1_Model_Card.pdf", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "real-repository Trae coding environment", "sampling": "pass@1", "judge": "multidimensional human rating", "harness": "official", "prompt_style": "official default", "temperature": "unknown" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Model card textual metrics (pp15-16, p32). Mean across six rating dimensions." }, { "model_id": "doubao-seed-2.1-pro", "benchmark_id": "seed21_trae_opus47_preference_win_rate", "score": 48.5, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2.1/Seed2_1_Model_Card.pdf", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "real-repository Trae coding environment", "sampling": "pass@1", "judge": "multidimensional human rating", "harness": "official", "prompt_style": "official default", "temperature": "unknown" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Model card textual metrics (pp15-16, p32). 81 Seed wins and 86 Claude wins over 167 valid head-to-head tasks." }, { "model_id": "doubao-seed-2.1-pro", "benchmark_id": "seed21_trae_severely_broken_rate", "score": 0.0, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2.1/Seed2_1_Model_Card.pdf", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "real-repository Trae coding environment", "sampling": "pass@1", "judge": "multidimensional human rating", "harness": "official", "prompt_style": "official default", "temperature": "unknown" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Model card textual metrics (pp15-16, p32). Severely broken delivery rate; lower is better." }, { "model_id": "doubao-seed-2.1-pro", "benchmark_id": "seed21_trae_unusable_rate", "score": 11.4, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2.1/Seed2_1_Model_Card.pdf", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "real-repository Trae coding environment", "sampling": "pass@1", "judge": "multidimensional human rating", "harness": "official", "prompt_style": "official default", "temperature": "unknown" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Model card textual metrics (pp15-16, p32). Unusable tier; lower is better." }, { "model_id": "doubao-seed-2.1-pro", "benchmark_id": "seedclawbench", "score": 66.6, "reference_url": "https://seed.bytedance.com/en/blog/seed2-1-officially-released-advancing-ai-productivity", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "OpenClaw-style tools and skills", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official", "prompt_style": "official default", "temperature": "unknown" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "Release blog EN: Daily-life/tool-use chart (https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/user-upload/4xfa4mqq4ujsp.png)." }, { "model_id": "doubao-seed-2.1-pro", "benchmark_id": "seedkernelbench_avg_speedup", "score": 9.21, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2.1/Seed2_1_Model_Card.pdf", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "Trae or benchmark-specific coding-agent environment", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official", "prompt_style": "official default", "temperature": "unknown" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Model card Table 5 (p19). Source dashes are not zero." }, { "model_id": "doubao-seed-2.1-pro", "benchmark_id": "simplevqa", "score": 74.5, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2.1/Seed2_1_Model_Card.pdf", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "none", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official", "prompt_style": "official default", "temperature": "unknown" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Model card Table 9 (p51)." }, { "model_id": "doubao-seed-2.1-pro", "benchmark_id": "supergpqa", "score": 70.8, "reference_url": "https://seed.bytedance.com/en/blog/seed2-1-officially-released-advancing-ai-productivity", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "none", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official", "prompt_style": "official default", "temperature": "unknown" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "Release blog EN: Language/search chart (https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/user-upload/4xfa4mqq510uc.png)." }, { "model_id": "doubao-seed-2.1-pro", "benchmark_id": "swe_atlas", "score": 35.2, "reference_url": "https://seed.bytedance.com/en/blog/seed2-1-officially-released-advancing-ai-productivity", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "Trae or benchmark-specific coding-agent environment", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official", "prompt_style": "official default", "temperature": "unknown" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "Release blog EN: Coding chart (https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/user-upload/4xfa4mqq4wlbt.png)." }, { "model_id": "doubao-seed-2.1-pro", "benchmark_id": "tomato", "score": 79.5, "reference_url": "https://seed.bytedance.com/en/blog/seed2-1-officially-released-advancing-ai-productivity", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "none", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official", "prompt_style": "official default", "temperature": "unknown" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "Release blog EN: Motion/perception chart (https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/user-upload/4xfa4mqq4zh7f.png)." }, { "model_id": "doubao-seed-2.1-pro", "benchmark_id": "trae_artifacts", "score": 51.0, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2.1/Seed2_1_Model_Card.pdf", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "Trae or benchmark-specific coding-agent environment", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official", "prompt_style": "official default", "temperature": "unknown" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Model card Table 5 (p19)." }, { "model_id": "doubao-seed-2.1-pro", "benchmark_id": "trae_code_gen_js", "score": 62.4, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2.1/Seed2_1_Model_Card.pdf", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "Trae or benchmark-specific coding-agent environment", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official", "prompt_style": "official default", "temperature": "unknown" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Model card Table 5 (p19)." }, { "model_id": "doubao-seed-2.1-pro", "benchmark_id": "trae_code_gen_python", "score": 75.6, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2.1/Seed2_1_Model_Card.pdf", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "Trae or benchmark-specific coding-agent environment", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official", "prompt_style": "official default", "temperature": "unknown" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Model card Table 5 (p19)." }, { "model_id": "doubao-seed-2.1-pro", "benchmark_id": "trae_error_fix_go", "score": 63.3, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2.1/Seed2_1_Model_Card.pdf", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "Trae or benchmark-specific coding-agent environment", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official", "prompt_style": "official default", "temperature": "unknown" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Model card Table 5 (p19)." }, { "model_id": "doubao-seed-2.1-pro", "benchmark_id": "trae_error_fix_java", "score": 66.7, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2.1/Seed2_1_Model_Card.pdf", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "Trae or benchmark-specific coding-agent environment", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official", "prompt_style": "official default", "temperature": "unknown" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Model card Table 5 (p19)." }, { "model_id": "doubao-seed-2.1-pro", "benchmark_id": "trae_error_fix_js", "score": 74.6, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2.1/Seed2_1_Model_Card.pdf", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "Trae or benchmark-specific coding-agent environment", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official", "prompt_style": "official default", "temperature": "unknown" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Model card Table 5 (p19)." }, { "model_id": "doubao-seed-2.1-pro", "benchmark_id": "trae_error_fix_python", "score": 70.7, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2.1/Seed2_1_Model_Card.pdf", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "Trae or benchmark-specific coding-agent environment", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official", "prompt_style": "official default", "temperature": "unknown" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Model card Table 5 (p19)." }, { "model_id": "doubao-seed-2.1-pro", "benchmark_id": "trae_repo_env", "score": 55.0, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2.1/Seed2_1_Model_Card.pdf", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "Trae or benchmark-specific coding-agent environment", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official", "prompt_style": "official default", "temperature": "unknown" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Model card Table 5 (p19)." }, { "model_id": "doubao-seed-2.1-pro", "benchmark_id": "treebench", "score": 71.1, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2.1/Seed2_1_Model_Card.pdf", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "none", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official", "prompt_style": "official default", "temperature": "unknown" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Model card Table 9 (p51)." }, { "model_id": "doubao-seed-2.1-pro", "benchmark_id": "tvbench", "score": 80.5, "reference_url": "https://seed.bytedance.com/en/blog/seed2-1-officially-released-advancing-ai-productivity", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "none", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official", "prompt_style": "official default", "temperature": "unknown" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "Release blog EN: Motion/perception chart (https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/user-upload/4xfa4mqq4zh7f.png)." }, { "model_id": "doubao-seed-2.1-pro", "benchmark_id": "video_mme", "score": 89.2, "reference_url": "https://seed.bytedance.com/en/blog/seed2-1-officially-released-advancing-ai-productivity", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "none", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official", "prompt_style": "official default", "temperature": "unknown" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "Release blog EN: Long/streaming-video chart (https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/user-upload/4xfa4mqq50ia3.png). Source marks this row with an unexplained dagger. Source marks this row with an unexplained dagger." }, { "model_id": "doubao-seed-2.1-pro", "benchmark_id": "videoholmes", "score": 68.2, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2.1/Seed2_1_Model_Card.pdf", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "none", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official", "prompt_style": "official default", "temperature": "unknown" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Model card Table 10 (p52). Source marks this row with an unexplained dagger. Source marks this row with an unexplained dagger." }, { "model_id": "doubao-seed-2.1-pro", "benchmark_id": "videosimpleqa", "score": 76.4, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2.1/Seed2_1_Model_Card.pdf", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "none", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official", "prompt_style": "official default", "temperature": "unknown" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Model card Table 10 (p52).", "candidates": [ { "score": 75.5, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2.1/Seed2_1_Model_Card.pdf", "source_type": "tech_report", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "none", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official", "prompt_style": "official default", "temperature": "unknown" }, "notes": "Model card textual metrics (pp15-16, p32). Conflicts with the Table 10 Seed2.1 Pro value of 76.4. Accompanying prose reports 75.5 while Table 10 reports 76.4; retain the prose value as a conflicting candidate." } ] }, { "model_id": "doubao-seed-2.1-pro", "benchmark_id": "visfactor", "score": 51.4, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2.1/Seed2_1_Model_Card.pdf", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "none", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official", "prompt_style": "official default", "temperature": "unknown" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Model card Table 9 (p51)." }, { "model_id": "doubao-seed-2.1-pro", "benchmark_id": "visulogic", "score": 54.3, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2.1/Seed2_1_Model_Card.pdf", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "none", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official", "prompt_style": "official default", "temperature": "unknown" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Model card Table 9 (p51)." }, { "model_id": "doubao-seed-2.1-pro", "benchmark_id": "vlms_are_biased", "score": 83.6, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2.1/Seed2_1_Model_Card.pdf", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "none", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official", "prompt_style": "official default", "temperature": "unknown" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Model card Table 9 (p51)." }, { "model_id": "doubao-seed-2.1-pro", "benchmark_id": "webbench", "score": 78.4, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2.1/Seed2_1_Model_Card.pdf", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "Trae or benchmark-specific coding-agent environment", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official", "prompt_style": "official default", "temperature": "unknown" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Model card Table 5 (p19)." }, { "model_id": "doubao-seed-2.1-pro", "benchmark_id": "wildclaw_bench_60_openclaw", "score": 61.7, "reference_url": "https://seed.bytedance.com/en/blog/seed2-1-officially-released-advancing-ai-productivity", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "high-resolution multimodal agent harness", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official", "prompt_style": "official default", "temperature": "unknown" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "Release blog EN: Multimodal-agent chart (https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/user-upload/4xfa4mqq4v1ol.png)." }, { "model_id": "doubao-seed-2.1-pro", "benchmark_id": "workspace_bench_openclaw_100", "score": 53.0, "reference_url": "https://seed.bytedance.com/en/blog/seed2-1-officially-released-advancing-ai-productivity", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "OpenClaw workspace tools", "sampling": "pass@1", "judge": "rubric-based evaluator", "harness": "100-task OpenClaw setting", "prompt_style": "official default", "temperature": "unknown" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "Release blog EN: General-agent chart (https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/user-upload/4xfa4mqq4u89c.png)." }, { "model_id": "doubao-seed-2.1-pro", "benchmark_id": "workspace_bench_openclaw_100_pass100", "score": 20.0, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2.1/Seed2_1_Model_Card.pdf", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "OpenClaw workspace tools", "sampling": "pass@1", "judge": "rubric-based evaluator", "harness": "100-task OpenClaw setting", "prompt_style": "official default", "temperature": "unknown" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Model card Table 1 (p9)." }, { "model_id": "doubao-seed-2.1-pro", "benchmark_id": "workspace_bench_openclaw_100_pass30", "score": 78.0, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2.1/Seed2_1_Model_Card.pdf", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "OpenClaw workspace tools", "sampling": "pass@1", "judge": "rubric-based evaluator", "harness": "100-task OpenClaw setting", "prompt_style": "official default", "temperature": "unknown" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Model card Table 1 (p9)." }, { "model_id": "doubao-seed-2.1-pro", "benchmark_id": "workspace_bench_openclaw_100_pass50", "score": 60.0, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2.1/Seed2_1_Model_Card.pdf", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "OpenClaw workspace tools", "sampling": "pass@1", "judge": "rubric-based evaluator", "harness": "100-task OpenClaw setting", "prompt_style": "official default", "temperature": "unknown" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Model card Table 1 (p9)." }, { "model_id": "doubao-seed-2.1-pro", "benchmark_id": "workspace_bench_openclaw_100_pass70", "score": 38.0, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2.1/Seed2_1_Model_Card.pdf", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "OpenClaw workspace tools", "sampling": "pass@1", "judge": "rubric-based evaluator", "harness": "100-task OpenClaw setting", "prompt_style": "official default", "temperature": "unknown" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Model card Table 1 (p9)." }, { "model_id": "doubao-seed-2.1-pro", "benchmark_id": "workspace_bench_openclaw_100_pass90", "score": 25.0, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2.1/Seed2_1_Model_Card.pdf", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "OpenClaw workspace tools", "sampling": "pass@1", "judge": "rubric-based evaluator", "harness": "100-task OpenClaw setting", "prompt_style": "official default", "temperature": "unknown" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Model card Table 1 (p9)." }, { "model_id": "doubao-seed-2.1-pro", "benchmark_id": "worldbench", "score": 67.6, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2.1/Seed2_1_Model_Card.pdf", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "none", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official", "prompt_style": "official default", "temperature": "unknown" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Model card Table 9 (p51)." }, { "model_id": "doubao-seed-2.1-pro", "benchmark_id": "worldvqa", "score": 53.0, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2.1/Seed2_1_Model_Card.pdf", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "none", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official", "prompt_style": "official default", "temperature": "unknown" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Model card Table 9 (p51)." }, { "model_id": "doubao-seed-2.1-pro", "benchmark_id": "xdailybench", "score": 61.0, "reference_url": "https://seed.bytedance.com/en/blog/seed2-1-officially-released-advancing-ai-productivity", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "none", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official", "prompt_style": "official default", "temperature": "unknown" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "Release blog EN: Daily-life/tool-use chart (https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/user-upload/4xfa4mqq4ujsp.png)." }, { "model_id": "doubao-seed-2.1-pro", "benchmark_id": "zerobench_main", "score": 18.0, "reference_url": "https://seed.bytedance.com/en/blog/seed2-1-officially-released-advancing-ai-productivity", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "none", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official", "prompt_style": "official default", "temperature": "unknown" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "Release blog EN: Vision capability chart (https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/user-upload/4xfa4mqq4yer3.png)." }, { "model_id": "doubao-seed-2.1-pro", "benchmark_id": "zerobench_sub", "score": 49.4, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2.1/Seed2_1_Model_Card.pdf", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "none", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official", "prompt_style": "official default", "temperature": "unknown" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Model card Table 9 (p51)." }, { "model_id": "doubao-seed-2.1-pro", "benchmark_id": "zerobench_sub_tools", "score": 56.3, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2.1/Seed2_1_Model_Card.pdf", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "source tool-augmented setting; exact tools undisclosed", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official", "prompt_style": "official default", "temperature": "unknown" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Model card Table 9 (p51). Parenthesized source value." }, { "model_id": "doubao-seed-2.1-pro", "benchmark_id": "zerobench_tools", "score": 22.0, "reference_url": "https://seed.bytedance.com/en/blog/seed2-1-officially-released-advancing-ai-productivity", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "source tool-augmented setting; exact tools undisclosed", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official", "prompt_style": "official default", "temperature": "unknown" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "Release blog EN: Vision capability chart (https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/user-upload/4xfa4mqq4yer3.png). Parenthesized source value." }, { "model_id": "doubao-seed-2.1-pro-preview", "benchmark_id": "code_arena_frontend_elo", "score": 1539, "reference_url": "https://seed.bytedance.com/en/blog/seed2-1-officially-released-advancing-ai-productivity", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "none", "sampling": "107,962-vote Arena snapshot", "judge": "human preference", "harness": "Arena Code leaderboard", "prompt_style": "official default", "temperature": "unknown" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "Release blog EN: Code Arena: Frontend snapshot (https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/user-upload/4xfa4mqq4xsvz.jpg). Historical leaderboard snapshot reproduced by ByteDance." }, { "model_id": "doubao-seed-2.1-turbo", "benchmark_id": "aethercode", "score": 67.9, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2.1/Seed2_1_Model_Card.pdf", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "none", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official", "prompt_style": "official default", "temperature": "unknown" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Model card Table 11 (p53)." }, { "model_id": "doubao-seed-2.1-turbo", "benchmark_id": "agent_startup_bench", "score": 54.0, "reference_url": "https://seed.bytedance.com/en/blog/seed2-1-officially-released-advancing-ai-productivity", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "research and professional-deliverable tools", "sampling": "pass@1", "judge": "expert review", "harness": "official", "prompt_style": "official default", "temperature": "unknown" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "Release blog EN: General-agent chart (https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/user-upload/4xfa4mqq4u89c.png)." }, { "model_id": "doubao-seed-2.1-turbo", "benchmark_id": "apex_agents", "score": 29.2, "reference_url": "https://seed.bytedance.com/en/blog/seed2-1-officially-released-advancing-ai-productivity", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "long-horizon professional-task environment", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official", "prompt_style": "official default", "temperature": "unknown" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "Release blog EN: General-agent chart (https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/user-upload/4xfa4mqq4u89c.png)." }, { "model_id": "doubao-seed-2.1-turbo", "benchmark_id": "arc_agi_2", "score": 61.3, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2.1/Seed2_1_Model_Card.pdf", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "none", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official", "prompt_style": "official default", "temperature": "unknown" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Model card Table 11 (p53)." }, { "model_id": "doubao-seed-2.1-turbo", "benchmark_id": "babyvision", "score": 62.9, "reference_url": "https://seed.bytedance.com/en/blog/seed2-1-officially-released-advancing-ai-productivity", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "none", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official", "prompt_style": "official default", "temperature": "unknown" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "Release blog EN: Vision capability chart (https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/user-upload/4xfa4mqq4yer3.png)." }, { "model_id": "doubao-seed-2.1-turbo", "benchmark_id": "beyond_aime", "score": 88.0, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2.1/Seed2_1_Model_Card.pdf", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "none", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official", "prompt_style": "official default", "temperature": "unknown" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Model card Table 11 (p53)." }, { "model_id": "doubao-seed-2.1-turbo", "benchmark_id": "blink", "score": 79.4, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2.1/Seed2_1_Model_Card.pdf", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "none", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official", "prompt_style": "official default", "temperature": "unknown" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Model card Table 9 (p51)." }, { "model_id": "doubao-seed-2.1-turbo", "benchmark_id": "browsecomp", "score": 84.9, "reference_url": "https://seed.bytedance.com/en/blog/seed2-1-officially-released-advancing-ai-productivity", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "search", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official", "prompt_style": "official default", "temperature": "unknown" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "Release blog EN: Language/search chart (https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/user-upload/4xfa4mqq510uc.png)." }, { "model_id": "doubao-seed-2.1-turbo", "benchmark_id": "chartqapro", "score": 70.9, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2.1/Seed2_1_Model_Card.pdf", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "none", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official", "prompt_style": "official default", "temperature": "unknown" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Model card Table 9 (p51)." }, { "model_id": "doubao-seed-2.1-turbo", "benchmark_id": "charxiv_descriptive", "score": 94.6, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2.1/Seed2_1_Model_Card.pdf", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "none", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official", "prompt_style": "official default", "temperature": "unknown" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Model card Table 9 (p51)." }, { "model_id": "doubao-seed-2.1-turbo", "benchmark_id": "charxiv_reasoning", "score": 82.5, "reference_url": "https://seed.bytedance.com/en/blog/seed2-1-officially-released-advancing-ai-productivity", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "none", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official", "prompt_style": "official default", "temperature": "unknown" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "Release blog EN: Vision capability chart (https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/user-upload/4xfa4mqq4yer3.png).", "candidates": [ { "score": 83.6, "reference_url": "https://seed.bytedance.com/en/blog/seed2-1-officially-released-advancing-ai-productivity", "source_type": "official_blog", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "source tool-augmented setting; exact tools undisclosed", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official", "prompt_style": "official default", "temperature": "unknown" }, "notes": "Release blog EN: Vision capability chart (https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/user-upload/4xfa4mqq4yer3.png). Parenthesized source value." } ] }, { "model_id": "doubao-seed-2.1-turbo", "benchmark_id": "claw_eval_multimodal_pass3", "score": 46.0, "reference_url": "https://seed.bytedance.com/en/blog/seed2-1-officially-released-advancing-ai-productivity", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "high-resolution multimodal agent harness", "sampling": "Pass^3", "judge": "benchmark-specified", "harness": "official", "prompt_style": "official default", "temperature": "unknown" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "Release blog EN: Multimodal-agent chart (https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/user-upload/4xfa4mqq4v1ol.png)." }, { "model_id": "doubao-seed-2.1-turbo", "benchmark_id": "contphy", "score": 61.1, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2.1/Seed2_1_Model_Card.pdf", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "none", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official", "prompt_style": "official default", "temperature": "unknown" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Model card Table 10 (p52)." }, { "model_id": "doubao-seed-2.1-turbo", "benchmark_id": "creativework", "score": 34.5, "reference_url": "https://seed.bytedance.com/en/blog/seed2-1-officially-released-advancing-ai-productivity", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "GUI + MCP in Notion, Canva, and Figma", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official", "prompt_style": "official default", "temperature": "unknown" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "Release blog EN: Computer-use chart (https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/user-upload/4xfa4mqq4w8x3.png)." }, { "model_id": "doubao-seed-2.1-turbo", "benchmark_id": "crossvid", "score": 63.2, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2.1/Seed2_1_Model_Card.pdf", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "none", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official", "prompt_style": "official default", "temperature": "unknown" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Model card Table 10 (p52)." }, { "model_id": "doubao-seed-2.1-turbo", "benchmark_id": "cybergym", "score": 67.0, "reference_url": "https://seed.bytedance.com/en/blog/seed2-1-officially-released-advancing-ai-productivity", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "cybersecurity agent environment", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official", "prompt_style": "official default", "temperature": "unknown" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "Release blog EN: Coding chart (https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/user-upload/4xfa4mqq4wlbt.png). Release blog reports Pro=68.7; PDF Table 5 reports Pro=70.2.", "candidates": [ { "score": 67.0, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2.1/Seed2_1_Model_Card.pdf", "source_type": "tech_report", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "Trae or benchmark-specific coding-agent environment", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official", "prompt_style": "official default", "temperature": "unknown" }, "notes": "Model card Table 5 (p19). Source dashes are not zero." } ] }, { "model_id": "doubao-seed-2.1-turbo", "benchmark_id": "deep_swe_v1_1", "score": 23.0, "reference_url": "https://seed.bytedance.com/en/blog/seed2-1-officially-released-advancing-ai-productivity", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "Trae or benchmark-specific coding-agent environment", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official", "prompt_style": "official default", "temperature": "unknown" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "Release blog EN: Coding chart (https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/user-upload/4xfa4mqq4wlbt.png)." }, { "model_id": "doubao-seed-2.1-turbo", "benchmark_id": "doubao_multi_turn_bench", "score": 49.0, "reference_url": "https://seed.bytedance.com/en/blog/seed2-1-officially-released-advancing-ai-productivity", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "none", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official", "prompt_style": "official default", "temperature": "unknown" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "Release blog EN: Daily-life/tool-use chart (https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/user-upload/4xfa4mqq4ujsp.png)." }, { "model_id": "doubao-seed-2.1-turbo", "benchmark_id": "dude", "score": 83.1, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2.1/Seed2_1_Model_Card.pdf", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "none", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official", "prompt_style": "official default", "temperature": "unknown" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Model card Table 9 (p51)." }, { "model_id": "doubao-seed-2.1-turbo", "benchmark_id": "dynamath", "score": 68.1, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2.1/Seed2_1_Model_Card.pdf", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "none", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official", "prompt_style": "official default", "temperature": "unknown" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Model card Table 9 (p51)." }, { "model_id": "doubao-seed-2.1-turbo", "benchmark_id": "embspatial_bench", "score": 82.5, "reference_url": "https://seed.bytedance.com/en/blog/seed2-1-officially-released-advancing-ai-productivity", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "none", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official", "prompt_style": "official default", "temperature": "unknown" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "Release blog EN: Spatial/long-context chart (https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/user-upload/4xfa4mqq4z2l5.png)." }, { "model_id": "doubao-seed-2.1-turbo", "benchmark_id": "emma", "score": 78.4, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2.1/Seed2_1_Model_Card.pdf", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "none", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official", "prompt_style": "official default", "temperature": "unknown" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Model card Table 9 (p51)." }, { "model_id": "doubao-seed-2.1-turbo", "benchmark_id": "erqa", "score": 71.3, "reference_url": "https://seed.bytedance.com/en/blog/seed2-1-officially-released-advancing-ai-productivity", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "none", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official", "prompt_style": "official default", "temperature": "unknown" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "Release blog EN: Spatial/long-context chart (https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/user-upload/4xfa4mqq4z2l5.png)." }, { "model_id": "doubao-seed-2.1-turbo", "benchmark_id": "finance_agent", "score": 56.0, "reference_url": "https://seed.bytedance.com/en/blog/seed2-1-officially-released-advancing-ai-productivity", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "financial-analysis agent tools", "sampling": "pass@1", "judge": "LLM-as-judge rubric and contradiction grader", "harness": "official", "prompt_style": "official default", "temperature": "unknown" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "Release blog EN: General-agent chart (https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/user-upload/4xfa4mqq4u89c.png)." }, { "model_id": "doubao-seed-2.1-turbo", "benchmark_id": "frontier_science_research", "score": 33.3, "reference_url": "https://seed.bytedance.com/en/blog/seed2-1-officially-released-advancing-ai-productivity", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "research-oriented agent environment", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official", "prompt_style": "official default", "temperature": "unknown" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "Release blog EN: Frontier-research chart (https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/user-upload/4xfa4mqq52cdr.png)." }, { "model_id": "doubao-seed-2.1-turbo", "benchmark_id": "frontiercs_algorithmic_v1", "score": 33.4, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2.1/Seed2_1_Model_Card.pdf", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "research and code-execution environment", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official", "prompt_style": "official default", "temperature": "unknown" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Model card Table 6 (p24). Claude Opus 4.7 is a source dash; official sub-score name unresolved." }, { "model_id": "doubao-seed-2.1-turbo", "benchmark_id": "frontiercs_overall_v1", "score": 33.5, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2.1/Seed2_1_Model_Card.pdf", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "research and code-execution environment", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official", "prompt_style": "official default", "temperature": "unknown" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Model card Table 6 (p24). Claude Opus 4.7 is a source dash; official sub-score name unresolved." }, { "model_id": "doubao-seed-2.1-turbo", "benchmark_id": "frontiercs_research_v1", "score": 50.8, "reference_url": "https://seed.bytedance.com/en/blog/seed2-1-officially-released-advancing-ai-productivity", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "research and code-execution environment", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official", "prompt_style": "official default", "temperature": "unknown" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "Release blog EN: Frontier-research chart (https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/user-upload/4xfa4mqq52cdr.png). Claude Opus 4.7 is a source dash; official sub-score name unresolved." }, { "model_id": "doubao-seed-2.1-turbo", "benchmark_id": "frontiersci_olympiad", "score": 76.0, "reference_url": "https://seed.bytedance.com/en/blog/seed2-1-officially-released-advancing-ai-productivity", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "none", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official", "prompt_style": "official default", "temperature": "unknown" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "Release blog EN: Language/search chart (https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/user-upload/4xfa4mqq510uc.png)." }, { "model_id": "doubao-seed-2.1-turbo", "benchmark_id": "fs_researcher", "score": 23.3, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2.1/Seed2_1_Model_Card.pdf", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "none", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official", "prompt_style": "official default", "temperature": "unknown" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Model card Table 11 (p53)." }, { "model_id": "doubao-seed-2.1-turbo", "benchmark_id": "gameworld", "score": 25.9, "reference_url": "https://seed.bytedance.com/en/blog/seed2-1-officially-released-advancing-ai-productivity", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "browser-game GUI actions", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official", "prompt_style": "official default", "temperature": "unknown" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "Release blog EN: Computer-use chart (https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/user-upload/4xfa4mqq4w8x3.png)." }, { "model_id": "doubao-seed-2.1-turbo", "benchmark_id": "gdpval_seed_reported_score", "score": 82.7, "reference_url": "https://seed.bytedance.com/en/blog/seed2-1-officially-released-advancing-ai-productivity", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "document/spreadsheet deliverable workflow", "sampling": "pass@1", "judge": "source-specific evaluator", "harness": "official", "prompt_style": "official default", "temperature": "unknown" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "Release blog EN: General-agent chart (https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/user-upload/4xfa4mqq4u89c.png)." }, { "model_id": "doubao-seed-2.1-turbo", "benchmark_id": "hle_tools_text", "score": 54.6, "reference_url": "https://seed.bytedance.com/en/blog/seed2-1-officially-released-advancing-ai-productivity", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "search", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official", "prompt_style": "official default", "temperature": "unknown" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "Release blog EN: Language/search chart (https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/user-upload/4xfa4mqq510uc.png)." }, { "model_id": "doubao-seed-2.1-turbo", "benchmark_id": "hle_verified", "score": 42.4, "reference_url": "https://seed.bytedance.com/en/blog/seed2-1-officially-released-advancing-ai-productivity", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "none", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official", "prompt_style": "official default", "temperature": "unknown" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "Release blog EN: Language/search chart (https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/user-upload/4xfa4mqq510uc.png)." }, { "model_id": "doubao-seed-2.1-turbo", "benchmark_id": "horizon_math_pass12", "score": 2.0, "reference_url": "https://seed.bytedance.com/en/blog/seed2-1-officially-released-advancing-ai-productivity", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "research-oriented agent environment", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official", "prompt_style": "official default", "temperature": "unknown" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "Release blog EN: Frontier-research chart (https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/user-upload/4xfa4mqq52cdr.png)." }, { "model_id": "doubao-seed-2.1-turbo", "benchmark_id": "image2floorplan_avg_score", "score": 35.9, "reference_url": "https://seed.bytedance.com/en/blog/seed2-1-officially-released-advancing-ai-productivity", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "high-resolution multimodal agent harness", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official", "prompt_style": "official default", "temperature": "unknown" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "Release blog EN: Multimodal-agent chart (https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/user-upload/4xfa4mqq4v1ol.png)." }, { "model_id": "doubao-seed-2.1-turbo", "benchmark_id": "kina", "score": 46.6, "reference_url": "https://seed.bytedance.com/en/blog/seed2-1-officially-released-advancing-ai-productivity", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "none", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official", "prompt_style": "official default", "temperature": "unknown" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "Release blog EN: Language/search chart (https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/user-upload/4xfa4mqq510uc.png)." }, { "model_id": "doubao-seed-2.1-turbo", "benchmark_id": "live_mathematician_bench_2026_06", "score": 27.7, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2.1/Seed2_1_Model_Card.pdf", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "none", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official", "prompt_style": "official default", "temperature": "unknown" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Model card Table 11 (p53)." }, { "model_id": "doubao-seed-2.1-turbo", "benchmark_id": "livesports_3k", "score": 77.1, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2.1/Seed2_1_Model_Card.pdf", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "none", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official", "prompt_style": "official default", "temperature": "unknown" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Model card Table 10 (p52)." }, { "model_id": "doubao-seed-2.1-turbo", "benchmark_id": "longvideobench", "score": 80.6, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2.1/Seed2_1_Model_Card.pdf", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "none", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official", "prompt_style": "official default", "temperature": "unknown" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Model card Table 10 (p52)." }, { "model_id": "doubao-seed-2.1-turbo", "benchmark_id": "lvbench", "score": 76.8, "reference_url": "https://seed.bytedance.com/en/blog/seed2-1-officially-released-advancing-ai-productivity", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "none", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official", "prompt_style": "official default", "temperature": "unknown" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "Release blog EN: Long/streaming-video chart (https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/user-upload/4xfa4mqq50ia3.png)." }, { "model_id": "doubao-seed-2.1-turbo", "benchmark_id": "matharena_apex_2025", "score": 35.4, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2.1/Seed2_1_Model_Card.pdf", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "none", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official", "prompt_style": "official default", "temperature": "unknown" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Model card Table 11 (p53). Source dash is not zero." }, { "model_id": "doubao-seed-2.1-turbo", "benchmark_id": "mathverse_vision_only", "score": 89.2, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2.1/Seed2_1_Model_Card.pdf", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "none", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official", "prompt_style": "official default", "temperature": "unknown" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Model card Table 9 (p51)." }, { "model_id": "doubao-seed-2.1-turbo", "benchmark_id": "mathvision", "score": 90.1, "reference_url": "https://seed.bytedance.com/en/blog/seed2-1-officially-released-advancing-ai-productivity", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "none", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official", "prompt_style": "official default", "temperature": "unknown" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "Release blog EN: Vision capability chart (https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/user-upload/4xfa4mqq4yer3.png).", "candidates": [ { "score": 92.7, "reference_url": "https://seed.bytedance.com/en/blog/seed2-1-officially-released-advancing-ai-productivity", "source_type": "official_blog", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "source tool-augmented setting; exact tools undisclosed", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official", "prompt_style": "official default", "temperature": "unknown" }, "notes": "Release blog EN: Vision capability chart (https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/user-upload/4xfa4mqq4yer3.png). Parenthesized source value." } ] }, { "model_id": "doubao-seed-2.1-turbo", "benchmark_id": "mathvista", "score": 90.5, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2.1/Seed2_1_Model_Card.pdf", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "none", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official", "prompt_style": "official default", "temperature": "unknown" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Model card Table 9 (p51)." }, { "model_id": "doubao-seed-2.1-turbo", "benchmark_id": "mcpatlas", "score": 80.3, "reference_url": "https://seed.bytedance.com/en/blog/seed2-1-officially-released-advancing-ai-productivity", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "MCP servers in an agent harness", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official", "prompt_style": "official default", "temperature": "unknown" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "Release blog EN: Daily-life/tool-use chart (https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/user-upload/4xfa4mqq4ujsp.png)." }, { "model_id": "doubao-seed-2.1-turbo", "benchmark_id": "measurebench", "score": 58.9, "reference_url": "https://seed.bytedance.com/en/blog/seed2-1-officially-released-advancing-ai-productivity", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "none", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official", "prompt_style": "official default", "temperature": "unknown" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "Release blog EN: Vision capability chart (https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/user-upload/4xfa4mqq4yer3.png)." }, { "model_id": "doubao-seed-2.1-turbo", "benchmark_id": "minerva", "score": 65.9, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2.1/Seed2_1_Model_Card.pdf", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "none", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official", "prompt_style": "official default", "temperature": "unknown" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Model card Table 10 (p52). Source marks this row with an unexplained dagger. Source marks this row with an unexplained dagger." }, { "model_id": "doubao-seed-2.1-turbo", "benchmark_id": "mmlongbench", "score": 76.9, "reference_url": "https://seed.bytedance.com/en/blog/seed2-1-officially-released-advancing-ai-productivity", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "none", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official", "prompt_style": "official default", "temperature": "unknown" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "Release blog EN: Spatial/long-context chart (https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/user-upload/4xfa4mqq4z2l5.png). Source dashes are not zero." }, { "model_id": "doubao-seed-2.1-turbo", "benchmark_id": "mmmu_pro", "score": 80.1, "reference_url": "https://seed.bytedance.com/en/blog/seed2-1-officially-released-advancing-ai-productivity", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "none", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official", "prompt_style": "official default", "temperature": "unknown" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "Release blog EN: Vision capability chart (https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/user-upload/4xfa4mqq4yer3.png).", "candidates": [ { "score": 82.2, "reference_url": "https://seed.bytedance.com/en/blog/seed2-1-officially-released-advancing-ai-productivity", "source_type": "official_blog", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "source tool-augmented setting; exact tools undisclosed", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official", "prompt_style": "official default", "temperature": "unknown" }, "notes": "Release blog EN: Vision capability chart (https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/user-upload/4xfa4mqq4yer3.png). Parenthesized source value." } ] }, { "model_id": "doubao-seed-2.1-turbo", "benchmark_id": "mmsibench_circular", "score": 31.4, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2.1/Seed2_1_Model_Card.pdf", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "none", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official", "prompt_style": "official default", "temperature": "unknown" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Model card Table 9 (p51)." }, { "model_id": "doubao-seed-2.1-turbo", "benchmark_id": "mobileworld", "score": 70.0, "reference_url": "https://seed.bytedance.com/en/blog/seed2-1-officially-released-advancing-ai-productivity", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "mobile GUI actions", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official", "prompt_style": "official default", "temperature": "unknown" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "Release blog EN: Computer-use chart (https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/user-upload/4xfa4mqq4w8x3.png)." }, { "model_id": "doubao-seed-2.1-turbo", "benchmark_id": "motionbench", "score": 74.8, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2.1/Seed2_1_Model_Card.pdf", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "none", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official", "prompt_style": "official default", "temperature": "unknown" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Model card Table 10 (p52)." }, { "model_id": "doubao-seed-2.1-turbo", "benchmark_id": "msqa", "score": 42.0, "reference_url": "https://seed.bytedance.com/en/blog/seed2-1-officially-released-advancing-ai-productivity", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "none", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official", "prompt_style": "official default", "temperature": "unknown" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "Release blog EN: Language/search chart (https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/user-upload/4xfa4mqq510uc.png)." }, { "model_id": "doubao-seed-2.1-turbo", "benchmark_id": "nl2repo_bench", "score": 43.7, "reference_url": "https://seed.bytedance.com/en/blog/seed2-1-officially-released-advancing-ai-productivity", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "Trae or benchmark-specific coding-agent environment", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official", "prompt_style": "official default", "temperature": "unknown" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "Release blog EN: Coding chart (https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/user-upload/4xfa4mqq4wlbt.png)." }, { "model_id": "doubao-seed-2.1-turbo", "benchmark_id": "ocrbench_v2", "score": 62.8, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2.1/Seed2_1_Model_Card.pdf", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "none", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official", "prompt_style": "official default", "temperature": "unknown" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Model card Table 9 (p51)." }, { "model_id": "doubao-seed-2.1-turbo", "benchmark_id": "officeqa_pro", "score": 71.1, "reference_url": "https://seed.bytedance.com/en/blog/seed2-1-officially-released-advancing-ai-productivity", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "high-resolution multimodal agent harness", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official", "prompt_style": "official default", "temperature": "unknown" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "Release blog EN: Multimodal-agent chart (https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/user-upload/4xfa4mqq4v1ol.png).", "candidates": [ { "score": 62.8, "reference_url": "https://seed.bytedance.com/en/blog/seed2-1-officially-released-advancing-ai-productivity", "source_type": "official_blog", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "document and rendered-image analysis", "sampling": "pass@1", "judge": "OfficeQA Pro evaluator", "harness": "official", "prompt_style": "official default", "temperature": "unknown" }, "notes": "Release blog EN: General-agent chart (https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/user-upload/4xfa4mqq4u89c.png). The source says its internal implementations and adaptations can fluctuate relative to the official leaderboard." } ] }, { "model_id": "doubao-seed-2.1-turbo", "benchmark_id": "one_million_bench", "score": 66.6, "reference_url": "https://seed.bytedance.com/en/blog/seed2-1-officially-released-advancing-ai-productivity", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "agentic professional-task environment", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official", "prompt_style": "official default", "temperature": "unknown" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "Release blog EN: General-agent chart (https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/user-upload/4xfa4mqq4u89c.png)." }, { "model_id": "doubao-seed-2.1-turbo", "benchmark_id": "osworld", "score": 76.4, "reference_url": "https://seed.bytedance.com/en/blog/seed2-1-officially-released-advancing-ai-productivity", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "GUI + shell/commands + filesystem tools", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official", "prompt_style": "official default", "temperature": "unknown" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "Release blog EN: Computer-use chart (https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/user-upload/4xfa4mqq4w8x3.png).", "candidates": [ { "score": 73.2, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2.1/Seed2_1_Model_Card.pdf", "source_type": "tech_report", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "GUI-only", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "OSWorld GUI-only", "prompt_style": "official default", "temperature": "unknown" }, "notes": "Model card Figure 5 (p13). Figure 5 GUI-only operating point." } ] }, { "model_id": "doubao-seed-2.1-turbo", "benchmark_id": "ovbench", "score": 69.7, "reference_url": "https://seed.bytedance.com/en/blog/seed2-1-officially-released-advancing-ai-productivity", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "none", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official", "prompt_style": "official default", "temperature": "unknown" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "Release blog EN: Long/streaming-video chart (https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/user-upload/4xfa4mqq50ia3.png)." }, { "model_id": "doubao-seed-2.1-turbo", "benchmark_id": "ovobench", "score": 79.2, "reference_url": "https://seed.bytedance.com/en/blog/seed2-1-officially-released-advancing-ai-productivity", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "none", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official", "prompt_style": "official default", "temperature": "unknown" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "Release blog EN: Long/streaming-video chart (https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/user-upload/4xfa4mqq50ia3.png)." }, { "model_id": "doubao-seed-2.1-turbo", "benchmark_id": "posttrain_bench", "score": 18.3, "reference_url": "https://seed.bytedance.com/en/blog/seed2-1-officially-released-advancing-ai-productivity", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "research-oriented agent environment", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official", "prompt_style": "official default", "temperature": "unknown" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "Release blog EN: Frontier-research chart (https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/user-upload/4xfa4mqq52cdr.png). Source dashes are not zero." }, { "model_id": "doubao-seed-2.1-turbo", "benchmark_id": "presentbench", "score": 48.3, "reference_url": "https://seed.bytedance.com/en/blog/seed2-1-officially-released-advancing-ai-productivity", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "slide-generation environment", "sampling": "pass@1", "judge": "fine-grained rubric evaluator", "harness": "official", "prompt_style": "official default", "temperature": "unknown" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "Release blog EN: General-agent chart (https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/user-upload/4xfa4mqq4u89c.png)." }, { "model_id": "doubao-seed-2.1-turbo", "benchmark_id": "program_bench", "score": 49.4, "reference_url": "https://seed.bytedance.com/en/blog/seed2-1-officially-released-advancing-ai-productivity", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "Trae or benchmark-specific coding-agent environment", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official", "prompt_style": "official default", "temperature": "unknown" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "Release blog EN: Coding chart (https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/user-upload/4xfa4mqq4wlbt.png)." }, { "model_id": "doubao-seed-2.1-turbo", "benchmark_id": "program_bench_almost_resolved_95", "score": 0.0, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2.1/Seed2_1_Model_Card.pdf", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "compiled executable and documentation", "sampling": "pass@1", "judge": "248,000+ behavioral tests", "harness": "official ProgramBench sandbox", "prompt_style": "official default", "temperature": "unknown" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Model card Table 5 (p19)." }, { "model_id": "doubao-seed-2.1-turbo", "benchmark_id": "program_bench_fully_resolved", "score": 0.0, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2.1/Seed2_1_Model_Card.pdf", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "compiled executable and documentation", "sampling": "pass@1", "judge": "248,000+ behavioral tests", "harness": "official ProgramBench sandbox", "prompt_style": "official default", "temperature": "unknown" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Model card Table 5 (p19)." }, { "model_id": "doubao-seed-2.1-turbo", "benchmark_id": "realworldqa", "score": 86.3, "reference_url": "https://seed.bytedance.com/en/blog/seed2-1-officially-released-advancing-ai-productivity", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "none", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official", "prompt_style": "official default", "temperature": "unknown" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "Release blog EN: Vision capability chart (https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/user-upload/4xfa4mqq4yer3.png)." }, { "model_id": "doubao-seed-2.1-turbo", "benchmark_id": "scicode", "score": 57.8, "reference_url": "https://seed.bytedance.com/en/blog/seed2-1-officially-released-advancing-ai-productivity", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "none", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official", "prompt_style": "official default", "temperature": "unknown" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "Release blog EN: Language/search chart (https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/user-upload/4xfa4mqq510uc.png)." }, { "model_id": "doubao-seed-2.1-turbo", "benchmark_id": "seed21_claudecode_glm51_net_win_rate", "score": 23.0, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2.1/Seed2_1_Model_Card.pdf", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "real-repository coding environment", "sampling": "pass@1", "judge": "anonymous developer preference", "harness": "official", "prompt_style": "official default", "temperature": "unknown" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Model card textual metrics (pp15-16, p32). 100 wins / 19 ties / 59 losses over 178 comparisons." }, { "model_id": "doubao-seed-2.1-turbo", "benchmark_id": "seed21_claudecode_glm51_win_rate", "score": 56.2, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2.1/Seed2_1_Model_Card.pdf", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "real-repository coding environment", "sampling": "pass@1", "judge": "anonymous developer preference", "harness": "official", "prompt_style": "official default", "temperature": "unknown" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Model card textual metrics (pp15-16, p32). 100 wins / 19 ties / 59 losses over 178 comparisons." }, { "model_id": "doubao-seed-2.1-turbo", "benchmark_id": "seedclawbench", "score": 63.8, "reference_url": "https://seed.bytedance.com/en/blog/seed2-1-officially-released-advancing-ai-productivity", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "OpenClaw-style tools and skills", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official", "prompt_style": "official default", "temperature": "unknown" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "Release blog EN: Daily-life/tool-use chart (https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/user-upload/4xfa4mqq4ujsp.png)." }, { "model_id": "doubao-seed-2.1-turbo", "benchmark_id": "seedkernelbench_avg_speedup", "score": 8.6, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2.1/Seed2_1_Model_Card.pdf", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "Trae or benchmark-specific coding-agent environment", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official", "prompt_style": "official default", "temperature": "unknown" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Model card Table 5 (p19). Source dashes are not zero." }, { "model_id": "doubao-seed-2.1-turbo", "benchmark_id": "simplevqa", "score": 71.1, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2.1/Seed2_1_Model_Card.pdf", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "none", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official", "prompt_style": "official default", "temperature": "unknown" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Model card Table 9 (p51)." }, { "model_id": "doubao-seed-2.1-turbo", "benchmark_id": "superchem", "score": 56.6, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2.1/Seed2_1_Model_Card.pdf", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "none", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official", "prompt_style": "official default", "temperature": "unknown" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Model card Table 11 (p53)." }, { "model_id": "doubao-seed-2.1-turbo", "benchmark_id": "supergpqa", "score": 67.4, "reference_url": "https://seed.bytedance.com/en/blog/seed2-1-officially-released-advancing-ai-productivity", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "none", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official", "prompt_style": "official default", "temperature": "unknown" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "Release blog EN: Language/search chart (https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/user-upload/4xfa4mqq510uc.png)." }, { "model_id": "doubao-seed-2.1-turbo", "benchmark_id": "swe_atlas", "score": 30.6, "reference_url": "https://seed.bytedance.com/en/blog/seed2-1-officially-released-advancing-ai-productivity", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "Trae or benchmark-specific coding-agent environment", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official", "prompt_style": "official default", "temperature": "unknown" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "Release blog EN: Coding chart (https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/user-upload/4xfa4mqq4wlbt.png)." }, { "model_id": "doubao-seed-2.1-turbo", "benchmark_id": "swe_bench_pro", "score": 57.0, "reference_url": "https://seed.bytedance.com/en/blog/seed2-1-officially-released-advancing-ai-productivity", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "Trae or benchmark-specific coding-agent environment", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official", "prompt_style": "official default", "temperature": "unknown" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "Release blog EN: Coding chart (https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/user-upload/4xfa4mqq4wlbt.png)." }, { "model_id": "doubao-seed-2.1-turbo", "benchmark_id": "terminal_bench_2_1", "score": 67.6, "reference_url": "https://seed.bytedance.com/en/blog/seed2-1-officially-released-advancing-ai-productivity", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "Trae or benchmark-specific coding-agent environment", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official", "prompt_style": "official default", "temperature": "unknown" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "Release blog EN: Coding chart (https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/user-upload/4xfa4mqq4wlbt.png)." }, { "model_id": "doubao-seed-2.1-turbo", "benchmark_id": "tomato", "score": 56.8, "reference_url": "https://seed.bytedance.com/en/blog/seed2-1-officially-released-advancing-ai-productivity", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "none", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official", "prompt_style": "official default", "temperature": "unknown" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "Release blog EN: Motion/perception chart (https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/user-upload/4xfa4mqq4zh7f.png)." }, { "model_id": "doubao-seed-2.1-turbo", "benchmark_id": "toolathlon", "score": 49.1, "reference_url": "https://seed.bytedance.com/en/blog/seed2-1-officially-released-advancing-ai-productivity", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "agentic tool use", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official", "prompt_style": "official default", "temperature": "unknown" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "Release blog EN: Daily-life/tool-use chart (https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/user-upload/4xfa4mqq4ujsp.png)." }, { "model_id": "doubao-seed-2.1-turbo", "benchmark_id": "trae_artifacts", "score": 47.0, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2.1/Seed2_1_Model_Card.pdf", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "Trae or benchmark-specific coding-agent environment", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official", "prompt_style": "official default", "temperature": "unknown" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Model card Table 5 (p19)." }, { "model_id": "doubao-seed-2.1-turbo", "benchmark_id": "trae_code_gen_js", "score": 59.7, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2.1/Seed2_1_Model_Card.pdf", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "Trae or benchmark-specific coding-agent environment", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official", "prompt_style": "official default", "temperature": "unknown" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Model card Table 5 (p19)." }, { "model_id": "doubao-seed-2.1-turbo", "benchmark_id": "trae_code_gen_python", "score": 73.3, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2.1/Seed2_1_Model_Card.pdf", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "Trae or benchmark-specific coding-agent environment", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official", "prompt_style": "official default", "temperature": "unknown" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Model card Table 5 (p19)." }, { "model_id": "doubao-seed-2.1-turbo", "benchmark_id": "trae_error_fix_go", "score": 56.7, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2.1/Seed2_1_Model_Card.pdf", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "Trae or benchmark-specific coding-agent environment", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official", "prompt_style": "official default", "temperature": "unknown" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Model card Table 5 (p19)." }, { "model_id": "doubao-seed-2.1-turbo", "benchmark_id": "trae_error_fix_java", "score": 66.7, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2.1/Seed2_1_Model_Card.pdf", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "Trae or benchmark-specific coding-agent environment", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official", "prompt_style": "official default", "temperature": "unknown" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Model card Table 5 (p19)." }, { "model_id": "doubao-seed-2.1-turbo", "benchmark_id": "trae_error_fix_js", "score": 69.4, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2.1/Seed2_1_Model_Card.pdf", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "Trae or benchmark-specific coding-agent environment", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official", "prompt_style": "official default", "temperature": "unknown" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Model card Table 5 (p19)." }, { "model_id": "doubao-seed-2.1-turbo", "benchmark_id": "trae_error_fix_python", "score": 74.0, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2.1/Seed2_1_Model_Card.pdf", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "Trae or benchmark-specific coding-agent environment", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official", "prompt_style": "official default", "temperature": "unknown" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Model card Table 5 (p19)." }, { "model_id": "doubao-seed-2.1-turbo", "benchmark_id": "trae_repo_env", "score": 46.7, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2.1/Seed2_1_Model_Card.pdf", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "Trae or benchmark-specific coding-agent environment", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official", "prompt_style": "official default", "temperature": "unknown" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Model card Table 5 (p19)." }, { "model_id": "doubao-seed-2.1-turbo", "benchmark_id": "treebench", "score": 71.1, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2.1/Seed2_1_Model_Card.pdf", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "none", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official", "prompt_style": "official default", "temperature": "unknown" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Model card Table 9 (p51)." }, { "model_id": "doubao-seed-2.1-turbo", "benchmark_id": "tvbench", "score": 77.2, "reference_url": "https://seed.bytedance.com/en/blog/seed2-1-officially-released-advancing-ai-productivity", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "none", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official", "prompt_style": "official default", "temperature": "unknown" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "Release blog EN: Motion/perception chart (https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/user-upload/4xfa4mqq4zh7f.png)." }, { "model_id": "doubao-seed-2.1-turbo", "benchmark_id": "video_mme", "score": 89.0, "reference_url": "https://seed.bytedance.com/en/blog/seed2-1-officially-released-advancing-ai-productivity", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "none", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official", "prompt_style": "official default", "temperature": "unknown" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "Release blog EN: Long/streaming-video chart (https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/user-upload/4xfa4mqq50ia3.png). Source marks this row with an unexplained dagger. Source marks this row with an unexplained dagger." }, { "model_id": "doubao-seed-2.1-turbo", "benchmark_id": "videoholmes", "score": 67.6, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2.1/Seed2_1_Model_Card.pdf", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "none", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official", "prompt_style": "official default", "temperature": "unknown" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Model card Table 10 (p52). Source marks this row with an unexplained dagger. Source marks this row with an unexplained dagger." }, { "model_id": "doubao-seed-2.1-turbo", "benchmark_id": "videosimpleqa", "score": 71.4, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2.1/Seed2_1_Model_Card.pdf", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "none", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official", "prompt_style": "official default", "temperature": "unknown" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Model card Table 10 (p52)." }, { "model_id": "doubao-seed-2.1-turbo", "benchmark_id": "visfactor", "score": 43.9, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2.1/Seed2_1_Model_Card.pdf", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "none", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official", "prompt_style": "official default", "temperature": "unknown" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Model card Table 9 (p51)." }, { "model_id": "doubao-seed-2.1-turbo", "benchmark_id": "visulogic", "score": 52.9, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2.1/Seed2_1_Model_Card.pdf", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "none", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official", "prompt_style": "official default", "temperature": "unknown" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Model card Table 9 (p51)." }, { "model_id": "doubao-seed-2.1-turbo", "benchmark_id": "vlms_are_biased", "score": 68.3, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2.1/Seed2_1_Model_Card.pdf", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "none", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official", "prompt_style": "official default", "temperature": "unknown" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Model card Table 9 (p51)." }, { "model_id": "doubao-seed-2.1-turbo", "benchmark_id": "webbench", "score": 73.6, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2.1/Seed2_1_Model_Card.pdf", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "Trae or benchmark-specific coding-agent environment", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official", "prompt_style": "official default", "temperature": "unknown" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Model card Table 5 (p19)." }, { "model_id": "doubao-seed-2.1-turbo", "benchmark_id": "wildclaw_bench_60_openclaw", "score": 62.8, "reference_url": "https://seed.bytedance.com/en/blog/seed2-1-officially-released-advancing-ai-productivity", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "high-resolution multimodal agent harness", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official", "prompt_style": "official default", "temperature": "unknown" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "Release blog EN: Multimodal-agent chart (https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/user-upload/4xfa4mqq4v1ol.png)." }, { "model_id": "doubao-seed-2.1-turbo", "benchmark_id": "workspace_bench_openclaw_100", "score": 54.7, "reference_url": "https://seed.bytedance.com/en/blog/seed2-1-officially-released-advancing-ai-productivity", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "OpenClaw workspace tools", "sampling": "pass@1", "judge": "rubric-based evaluator", "harness": "100-task OpenClaw setting", "prompt_style": "official default", "temperature": "unknown" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "Release blog EN: General-agent chart (https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/user-upload/4xfa4mqq4u89c.png)." }, { "model_id": "doubao-seed-2.1-turbo", "benchmark_id": "workspace_bench_openclaw_100_pass100", "score": 16.0, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2.1/Seed2_1_Model_Card.pdf", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "OpenClaw workspace tools", "sampling": "pass@1", "judge": "rubric-based evaluator", "harness": "100-task OpenClaw setting", "prompt_style": "official default", "temperature": "unknown" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Model card Table 1 (p9)." }, { "model_id": "doubao-seed-2.1-turbo", "benchmark_id": "workspace_bench_openclaw_100_pass30", "score": 74.0, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2.1/Seed2_1_Model_Card.pdf", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "OpenClaw workspace tools", "sampling": "pass@1", "judge": "rubric-based evaluator", "harness": "100-task OpenClaw setting", "prompt_style": "official default", "temperature": "unknown" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Model card Table 1 (p9)." }, { "model_id": "doubao-seed-2.1-turbo", "benchmark_id": "workspace_bench_openclaw_100_pass50", "score": 59.0, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2.1/Seed2_1_Model_Card.pdf", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "OpenClaw workspace tools", "sampling": "pass@1", "judge": "rubric-based evaluator", "harness": "100-task OpenClaw setting", "prompt_style": "official default", "temperature": "unknown" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Model card Table 1 (p9)." }, { "model_id": "doubao-seed-2.1-turbo", "benchmark_id": "workspace_bench_openclaw_100_pass70", "score": 39.0, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2.1/Seed2_1_Model_Card.pdf", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "OpenClaw workspace tools", "sampling": "pass@1", "judge": "rubric-based evaluator", "harness": "100-task OpenClaw setting", "prompt_style": "official default", "temperature": "unknown" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Model card Table 1 (p9)." }, { "model_id": "doubao-seed-2.1-turbo", "benchmark_id": "workspace_bench_openclaw_100_pass90", "score": 20.0, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2.1/Seed2_1_Model_Card.pdf", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "OpenClaw workspace tools", "sampling": "pass@1", "judge": "rubric-based evaluator", "harness": "100-task OpenClaw setting", "prompt_style": "official default", "temperature": "unknown" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Model card Table 1 (p9)." }, { "model_id": "doubao-seed-2.1-turbo", "benchmark_id": "worldbench", "score": 63.7, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2.1/Seed2_1_Model_Card.pdf", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "none", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official", "prompt_style": "official default", "temperature": "unknown" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Model card Table 9 (p51)." }, { "model_id": "doubao-seed-2.1-turbo", "benchmark_id": "worldvqa", "score": 48.6, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2.1/Seed2_1_Model_Card.pdf", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "none", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official", "prompt_style": "official default", "temperature": "unknown" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Model card Table 9 (p51)." }, { "model_id": "doubao-seed-2.1-turbo", "benchmark_id": "xdailybench", "score": 56.4, "reference_url": "https://seed.bytedance.com/en/blog/seed2-1-officially-released-advancing-ai-productivity", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "none", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official", "prompt_style": "official default", "temperature": "unknown" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "Release blog EN: Daily-life/tool-use chart (https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/user-upload/4xfa4mqq4ujsp.png)." }, { "model_id": "doubao-seed-2.1-turbo", "benchmark_id": "zerobench_main", "score": 11.0, "reference_url": "https://seed.bytedance.com/en/blog/seed2-1-officially-released-advancing-ai-productivity", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "none", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official", "prompt_style": "official default", "temperature": "unknown" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "Release blog EN: Vision capability chart (https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/user-upload/4xfa4mqq4yer3.png)." }, { "model_id": "doubao-seed-2.1-turbo", "benchmark_id": "zerobench_sub", "score": 49.1, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2.1/Seed2_1_Model_Card.pdf", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "none", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official", "prompt_style": "official default", "temperature": "unknown" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Model card Table 9 (p51)." }, { "model_id": "doubao-seed-2.1-turbo", "benchmark_id": "zerobench_sub_tools", "score": 57.2, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2.1/Seed2_1_Model_Card.pdf", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "source tool-augmented setting; exact tools undisclosed", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official", "prompt_style": "official default", "temperature": "unknown" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Model card Table 9 (p51). Parenthesized source value." }, { "model_id": "doubao-seed-2.1-turbo", "benchmark_id": "zerobench_tools", "score": 20.0, "reference_url": "https://seed.bytedance.com/en/blog/seed2-1-officially-released-advancing-ai-productivity", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "source tool-augmented setting; exact tools undisclosed", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official", "prompt_style": "official default", "temperature": "unknown" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "Release blog EN: Vision capability chart (https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/user-upload/4xfa4mqq4yer3.png). Parenthesized source value." }, { "model_id": "gemini-3.1-pro", "benchmark_id": "aethercode", "score": 74.7, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2.1/Seed2_1_Model_Card.pdf", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "none", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official", "prompt_style": "official default", "temperature": "unknown" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Model card Table 11 (p53)." }, { "model_id": "gemini-3.1-pro", "benchmark_id": "agent_startup_bench", "score": 45.7, "reference_url": "https://seed.bytedance.com/en/blog/seed2-1-officially-released-advancing-ai-productivity", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "research and professional-deliverable tools", "sampling": "pass@1", "judge": "expert review", "harness": "official", "prompt_style": "official default", "temperature": "unknown" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Release blog EN: General-agent chart (https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/user-upload/4xfa4mqq4u89c.png)." }, { "model_id": "gemini-3.1-pro", "benchmark_id": "agents_last_exam_average_score", "score": 32.0, "reference_url": "https://seed.bytedance.com/en/blog/seed2-1-officially-released-advancing-ai-productivity", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "computer-use environment", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official", "prompt_style": "official default", "temperature": "unknown" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Release blog EN: General-agent chart (https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/user-upload/4xfa4mqq4u89c.png). Turbo is a source dash, not zero." }, { "model_id": "gemini-3.1-pro", "benchmark_id": "beyond_aime", "score": 90.0, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2.1/Seed2_1_Model_Card.pdf", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "none", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official", "prompt_style": "official default", "temperature": "unknown" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Model card Table 11 (p53)." }, { "model_id": "gemini-3.1-pro", "benchmark_id": "blink", "score": 79.1, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2.1/Seed2_1_Model_Card.pdf", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "none", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official", "prompt_style": "official default", "temperature": "unknown" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Model card Table 9 (p51)." }, { "model_id": "gemini-3.1-pro", "benchmark_id": "charxiv_descriptive", "score": 94.9, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2.1/Seed2_1_Model_Card.pdf", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "none", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official", "prompt_style": "official default", "temperature": "unknown" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Model card Table 9 (p51)." }, { "model_id": "gemini-3.1-pro", "benchmark_id": "claw_eval_multimodal_pass3", "score": 27.0, "reference_url": "https://seed.bytedance.com/en/blog/seed2-1-officially-released-advancing-ai-productivity", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "high-resolution multimodal agent harness", "sampling": "Pass^3", "judge": "benchmark-specified", "harness": "official", "prompt_style": "official default", "temperature": "unknown" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Release blog EN: Multimodal-agent chart (https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/user-upload/4xfa4mqq4v1ol.png)." }, { "model_id": "gemini-3.1-pro", "benchmark_id": "contphy", "score": 64.5, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2.1/Seed2_1_Model_Card.pdf", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "none", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official", "prompt_style": "official default", "temperature": "unknown" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Model card Table 10 (p52)." }, { "model_id": "gemini-3.1-pro", "benchmark_id": "creativework", "score": 27.4, "reference_url": "https://seed.bytedance.com/en/blog/seed2-1-officially-released-advancing-ai-productivity", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "GUI + MCP in Notion, Canva, and Figma", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official", "prompt_style": "official default", "temperature": "unknown" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Release blog EN: Computer-use chart (https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/user-upload/4xfa4mqq4w8x3.png)." }, { "model_id": "gemini-3.1-pro", "benchmark_id": "crossvid", "score": 48.8, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2.1/Seed2_1_Model_Card.pdf", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "none", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official", "prompt_style": "official default", "temperature": "unknown" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Model card Table 10 (p52)." }, { "model_id": "gemini-3.1-pro", "benchmark_id": "doubao_multi_turn_bench", "score": 52.0, "reference_url": "https://seed.bytedance.com/en/blog/seed2-1-officially-released-advancing-ai-productivity", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "none", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official", "prompt_style": "official default", "temperature": "unknown" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Release blog EN: Daily-life/tool-use chart (https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/user-upload/4xfa4mqq4ujsp.png)." }, { "model_id": "gemini-3.1-pro", "benchmark_id": "dude", "score": 82.1, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2.1/Seed2_1_Model_Card.pdf", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "none", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official", "prompt_style": "official default", "temperature": "unknown" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Model card Table 9 (p51)." }, { "model_id": "gemini-3.1-pro", "benchmark_id": "dynamath", "score": 72.1, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2.1/Seed2_1_Model_Card.pdf", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "none", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official", "prompt_style": "official default", "temperature": "unknown" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Model card Table 9 (p51)." }, { "model_id": "gemini-3.1-pro", "benchmark_id": "embspatial_bench", "score": 84.2, "reference_url": "https://seed.bytedance.com/en/blog/seed2-1-officially-released-advancing-ai-productivity", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "none", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official", "prompt_style": "official default", "temperature": "unknown" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Release blog EN: Spatial/long-context chart (https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/user-upload/4xfa4mqq4z2l5.png)." }, { "model_id": "gemini-3.1-pro", "benchmark_id": "emma", "score": 72.1, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2.1/Seed2_1_Model_Card.pdf", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "none", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official", "prompt_style": "official default", "temperature": "unknown" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Model card Table 9 (p51)." }, { "model_id": "gemini-3.1-pro", "benchmark_id": "erqa", "score": 70.8, "reference_url": "https://seed.bytedance.com/en/blog/seed2-1-officially-released-advancing-ai-productivity", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "none", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official", "prompt_style": "official default", "temperature": "unknown" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Release blog EN: Spatial/long-context chart (https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/user-upload/4xfa4mqq4z2l5.png)." }, { "model_id": "gemini-3.1-pro", "benchmark_id": "frontiercs_algorithmic_v1", "score": 44.1, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2.1/Seed2_1_Model_Card.pdf", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "research and code-execution environment", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official", "prompt_style": "official default", "temperature": "unknown" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Model card Table 6 (p24). Claude Opus 4.7 is a source dash; official sub-score name unresolved." }, { "model_id": "gemini-3.1-pro", "benchmark_id": "frontiercs_overall_v1", "score": 43.8, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2.1/Seed2_1_Model_Card.pdf", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "research and code-execution environment", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official", "prompt_style": "official default", "temperature": "unknown" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Model card Table 6 (p24). Claude Opus 4.7 is a source dash; official sub-score name unresolved." }, { "model_id": "gemini-3.1-pro", "benchmark_id": "frontiercs_research_v1", "score": 64.4, "reference_url": "https://seed.bytedance.com/en/blog/seed2-1-officially-released-advancing-ai-productivity", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "research and code-execution environment", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official", "prompt_style": "official default", "temperature": "unknown" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Release blog EN: Frontier-research chart (https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/user-upload/4xfa4mqq52cdr.png). Claude Opus 4.7 is a source dash; official sub-score name unresolved." }, { "model_id": "gemini-3.1-pro", "benchmark_id": "fs_researcher", "score": 16.7, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2.1/Seed2_1_Model_Card.pdf", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "none", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official", "prompt_style": "official default", "temperature": "unknown" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Model card Table 11 (p53)." }, { "model_id": "gemini-3.1-pro", "benchmark_id": "gameworld", "score": 21.2, "reference_url": "https://seed.bytedance.com/en/blog/seed2-1-officially-released-advancing-ai-productivity", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "browser-game GUI actions", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official", "prompt_style": "official default", "temperature": "unknown" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Release blog EN: Computer-use chart (https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/user-upload/4xfa4mqq4w8x3.png)." }, { "model_id": "gemini-3.1-pro", "benchmark_id": "gdpval_seed_reported_score", "score": 67.3, "reference_url": "https://seed.bytedance.com/en/blog/seed2-1-officially-released-advancing-ai-productivity", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "document/spreadsheet deliverable workflow", "sampling": "pass@1", "judge": "source-specific evaluator", "harness": "official", "prompt_style": "official default", "temperature": "unknown" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Release blog EN: General-agent chart (https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/user-upload/4xfa4mqq4u89c.png)." }, { "model_id": "gemini-3.1-pro", "benchmark_id": "hle_verified", "score": 48.2, "reference_url": "https://seed.bytedance.com/en/blog/seed2-1-officially-released-advancing-ai-productivity", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "none", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official", "prompt_style": "official default", "temperature": "unknown" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Release blog EN: Language/search chart (https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/user-upload/4xfa4mqq510uc.png)." }, { "model_id": "gemini-3.1-pro", "benchmark_id": "image2floorplan_avg_score", "score": 55.1, "reference_url": "https://seed.bytedance.com/en/blog/seed2-1-officially-released-advancing-ai-productivity", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "high-resolution multimodal agent harness", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official", "prompt_style": "official default", "temperature": "unknown" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Release blog EN: Multimodal-agent chart (https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/user-upload/4xfa4mqq4v1ol.png)." }, { "model_id": "gemini-3.1-pro", "benchmark_id": "imo_proofbench_advanced", "score": 49.0, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2.1/Seed2_1_Model_Card.pdf", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "benchmark-specified", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official", "prompt_style": "official default", "temperature": "unknown" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Model card Table 7 (p24)." }, { "model_id": "gemini-3.1-pro", "benchmark_id": "ipho_2025_theory", "score": 76.3, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2.1/Seed2_1_Model_Card.pdf", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "benchmark-specified", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official", "prompt_style": "official default", "temperature": "unknown" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Model card Table 7 (p24)." }, { "model_id": "gemini-3.1-pro", "benchmark_id": "kina", "score": 53.2, "reference_url": "https://seed.bytedance.com/en/blog/seed2-1-officially-released-advancing-ai-productivity", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "none", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official", "prompt_style": "official default", "temperature": "unknown" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Release blog EN: Language/search chart (https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/user-upload/4xfa4mqq510uc.png)." }, { "model_id": "gemini-3.1-pro", "benchmark_id": "live_mathematician_bench_2026_06", "score": 43.5, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2.1/Seed2_1_Model_Card.pdf", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "none", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official", "prompt_style": "official default", "temperature": "unknown" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Model card Table 11 (p53)." }, { "model_id": "gemini-3.1-pro", "benchmark_id": "livesports_3k", "score": 89.0, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2.1/Seed2_1_Model_Card.pdf", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "none", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official", "prompt_style": "official default", "temperature": "unknown" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Model card Table 10 (p52)." }, { "model_id": "gemini-3.1-pro", "benchmark_id": "longvideobench", "score": 76.5, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2.1/Seed2_1_Model_Card.pdf", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "none", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official", "prompt_style": "official default", "temperature": "unknown" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Model card Table 10 (p52)." }, { "model_id": "gemini-3.1-pro", "benchmark_id": "lvbench", "score": 66.2, "reference_url": "https://seed.bytedance.com/en/blog/seed2-1-officially-released-advancing-ai-productivity", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "none", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official", "prompt_style": "official default", "temperature": "unknown" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Release blog EN: Long/streaming-video chart (https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/user-upload/4xfa4mqq50ia3.png)." }, { "model_id": "gemini-3.1-pro", "benchmark_id": "mathverse_vision_only", "score": 87.7, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2.1/Seed2_1_Model_Card.pdf", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "none", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official", "prompt_style": "official default", "temperature": "unknown" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Model card Table 9 (p51)." }, { "model_id": "gemini-3.1-pro", "benchmark_id": "mathvista", "score": 90.2, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2.1/Seed2_1_Model_Card.pdf", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "none", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official", "prompt_style": "official default", "temperature": "unknown" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Model card Table 9 (p51)." }, { "model_id": "gemini-3.1-pro", "benchmark_id": "measurebench", "score": 44.4, "reference_url": "https://seed.bytedance.com/en/blog/seed2-1-officially-released-advancing-ai-productivity", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "none", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official", "prompt_style": "official default", "temperature": "unknown" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Release blog EN: Vision capability chart (https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/user-upload/4xfa4mqq4yer3.png)." }, { "model_id": "gemini-3.1-pro", "benchmark_id": "minerva", "score": 63.5, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2.1/Seed2_1_Model_Card.pdf", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "none", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official", "prompt_style": "official default", "temperature": "unknown" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Model card Table 10 (p52). Source marks this row with an unexplained dagger. Source marks this row with an unexplained dagger." }, { "model_id": "gemini-3.1-pro", "benchmark_id": "mmlongbench", "score": 70.7, "reference_url": "https://seed.bytedance.com/en/blog/seed2-1-officially-released-advancing-ai-productivity", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "none", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official", "prompt_style": "official default", "temperature": "unknown" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Release blog EN: Spatial/long-context chart (https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/user-upload/4xfa4mqq4z2l5.png). Source dashes are not zero." }, { "model_id": "gemini-3.1-pro", "benchmark_id": "mmsibench_circular", "score": 27.4, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2.1/Seed2_1_Model_Card.pdf", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "none", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official", "prompt_style": "official default", "temperature": "unknown" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Model card Table 9 (p51)." }, { "model_id": "gemini-3.1-pro", "benchmark_id": "mobileworld", "score": 48.4, "reference_url": "https://seed.bytedance.com/en/blog/seed2-1-officially-released-advancing-ai-productivity", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "mobile GUI actions", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official", "prompt_style": "official default", "temperature": "unknown" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Release blog EN: Computer-use chart (https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/user-upload/4xfa4mqq4w8x3.png)." }, { "model_id": "gemini-3.1-pro", "benchmark_id": "motionbench", "score": 69.9, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2.1/Seed2_1_Model_Card.pdf", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "none", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official", "prompt_style": "official default", "temperature": "unknown" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Model card Table 10 (p52)." }, { "model_id": "gemini-3.1-pro", "benchmark_id": "msqa", "score": 69.7, "reference_url": "https://seed.bytedance.com/en/blog/seed2-1-officially-released-advancing-ai-productivity", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "none", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official", "prompt_style": "official default", "temperature": "unknown" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Release blog EN: Language/search chart (https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/user-upload/4xfa4mqq510uc.png)." }, { "model_id": "gemini-3.1-pro", "benchmark_id": "ocrbench_v2", "score": 62.8, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2.1/Seed2_1_Model_Card.pdf", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "none", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official", "prompt_style": "official default", "temperature": "unknown" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Model card Table 9 (p51)." }, { "model_id": "gemini-3.1-pro", "benchmark_id": "one_million_bench", "score": 60.2, "reference_url": "https://seed.bytedance.com/en/blog/seed2-1-officially-released-advancing-ai-productivity", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "agentic professional-task environment", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official", "prompt_style": "official default", "temperature": "unknown" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Release blog EN: General-agent chart (https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/user-upload/4xfa4mqq4u89c.png)." }, { "model_id": "gemini-3.1-pro", "benchmark_id": "ovbench", "score": 58.8, "reference_url": "https://seed.bytedance.com/en/blog/seed2-1-officially-released-advancing-ai-productivity", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "none", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official", "prompt_style": "official default", "temperature": "unknown" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Release blog EN: Long/streaming-video chart (https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/user-upload/4xfa4mqq50ia3.png)." }, { "model_id": "gemini-3.1-pro", "benchmark_id": "ovobench", "score": 64.1, "reference_url": "https://seed.bytedance.com/en/blog/seed2-1-officially-released-advancing-ai-productivity", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "none", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official", "prompt_style": "official default", "temperature": "unknown" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Release blog EN: Long/streaming-video chart (https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/user-upload/4xfa4mqq50ia3.png)." }, { "model_id": "gemini-3.1-pro", "benchmark_id": "presentbench", "score": 52.1, "reference_url": "https://seed.bytedance.com/en/blog/seed2-1-officially-released-advancing-ai-productivity", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "slide-generation environment", "sampling": "pass@1", "judge": "fine-grained rubric evaluator", "harness": "official", "prompt_style": "official default", "temperature": "unknown" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Release blog EN: General-agent chart (https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/user-upload/4xfa4mqq4u89c.png)." }, { "model_id": "gemini-3.1-pro", "benchmark_id": "program_bench_almost_resolved_95", "score": 1.0, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2.1/Seed2_1_Model_Card.pdf", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "compiled executable and documentation", "sampling": "pass@1", "judge": "248,000+ behavioral tests", "harness": "official ProgramBench sandbox", "prompt_style": "official default", "temperature": "unknown" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Model card Table 5 (p19)." }, { "model_id": "gemini-3.1-pro", "benchmark_id": "program_bench_fully_resolved", "score": 0.0, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2.1/Seed2_1_Model_Card.pdf", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "compiled executable and documentation", "sampling": "pass@1", "judge": "248,000+ behavioral tests", "harness": "official ProgramBench sandbox", "prompt_style": "official default", "temperature": "unknown" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Model card Table 5 (p19)." }, { "model_id": "gemini-3.1-pro", "benchmark_id": "realworldqa", "score": 85.4, "reference_url": "https://seed.bytedance.com/en/blog/seed2-1-officially-released-advancing-ai-productivity", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "none", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official", "prompt_style": "official default", "temperature": "unknown" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Release blog EN: Vision capability chart (https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/user-upload/4xfa4mqq4yer3.png)." }, { "model_id": "gemini-3.1-pro", "benchmark_id": "seedclawbench", "score": 57.1, "reference_url": "https://seed.bytedance.com/en/blog/seed2-1-officially-released-advancing-ai-productivity", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "OpenClaw-style tools and skills", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official", "prompt_style": "official default", "temperature": "unknown" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Release blog EN: Daily-life/tool-use chart (https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/user-upload/4xfa4mqq4ujsp.png)." }, { "model_id": "gemini-3.1-pro", "benchmark_id": "simplevqa", "score": 69.9, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2.1/Seed2_1_Model_Card.pdf", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "none", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official", "prompt_style": "official default", "temperature": "unknown" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Model card Table 9 (p51)." }, { "model_id": "gemini-3.1-pro", "benchmark_id": "supergpqa", "score": 76.6, "reference_url": "https://seed.bytedance.com/en/blog/seed2-1-officially-released-advancing-ai-productivity", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "none", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official", "prompt_style": "official default", "temperature": "unknown" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Release blog EN: Language/search chart (https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/user-upload/4xfa4mqq510uc.png)." }, { "model_id": "gemini-3.1-pro", "benchmark_id": "swe_atlas", "score": 23.6, "reference_url": "https://seed.bytedance.com/en/blog/seed2-1-officially-released-advancing-ai-productivity", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "Trae or benchmark-specific coding-agent environment", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official", "prompt_style": "official default", "temperature": "unknown" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Release blog EN: Coding chart (https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/user-upload/4xfa4mqq4wlbt.png)." }, { "model_id": "gemini-3.1-pro", "benchmark_id": "tomato", "score": 60.4, "reference_url": "https://seed.bytedance.com/en/blog/seed2-1-officially-released-advancing-ai-productivity", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "none", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official", "prompt_style": "official default", "temperature": "unknown" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Release blog EN: Motion/perception chart (https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/user-upload/4xfa4mqq4zh7f.png)." }, { "model_id": "gemini-3.1-pro", "benchmark_id": "trae_artifacts", "score": 48.0, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2.1/Seed2_1_Model_Card.pdf", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "Trae or benchmark-specific coding-agent environment", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official", "prompt_style": "official default", "temperature": "unknown" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Model card Table 5 (p19)." }, { "model_id": "gemini-3.1-pro", "benchmark_id": "trae_code_gen_js", "score": 65.3, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2.1/Seed2_1_Model_Card.pdf", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "Trae or benchmark-specific coding-agent environment", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official", "prompt_style": "official default", "temperature": "unknown" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Model card Table 5 (p19)." }, { "model_id": "gemini-3.1-pro", "benchmark_id": "trae_code_gen_python", "score": 71.7, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2.1/Seed2_1_Model_Card.pdf", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "Trae or benchmark-specific coding-agent environment", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official", "prompt_style": "official default", "temperature": "unknown" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Model card Table 5 (p19)." }, { "model_id": "gemini-3.1-pro", "benchmark_id": "trae_error_fix_go", "score": 60.0, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2.1/Seed2_1_Model_Card.pdf", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "Trae or benchmark-specific coding-agent environment", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official", "prompt_style": "official default", "temperature": "unknown" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Model card Table 5 (p19)." }, { "model_id": "gemini-3.1-pro", "benchmark_id": "trae_error_fix_java", "score": 62.5, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2.1/Seed2_1_Model_Card.pdf", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "Trae or benchmark-specific coding-agent environment", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official", "prompt_style": "official default", "temperature": "unknown" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Model card Table 5 (p19)." }, { "model_id": "gemini-3.1-pro", "benchmark_id": "trae_error_fix_js", "score": 77.1, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2.1/Seed2_1_Model_Card.pdf", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "Trae or benchmark-specific coding-agent environment", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official", "prompt_style": "official default", "temperature": "unknown" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Model card Table 5 (p19)." }, { "model_id": "gemini-3.1-pro", "benchmark_id": "trae_error_fix_python", "score": 75.5, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2.1/Seed2_1_Model_Card.pdf", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "Trae or benchmark-specific coding-agent environment", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official", "prompt_style": "official default", "temperature": "unknown" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Model card Table 5 (p19)." }, { "model_id": "gemini-3.1-pro", "benchmark_id": "trae_repo_env", "score": 57.1, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2.1/Seed2_1_Model_Card.pdf", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "Trae or benchmark-specific coding-agent environment", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official", "prompt_style": "official default", "temperature": "unknown" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Model card Table 5 (p19)." }, { "model_id": "gemini-3.1-pro", "benchmark_id": "treebench", "score": 60.5, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2.1/Seed2_1_Model_Card.pdf", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "none", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official", "prompt_style": "official default", "temperature": "unknown" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Model card Table 9 (p51)." }, { "model_id": "gemini-3.1-pro", "benchmark_id": "tvbench", "score": 71.0, "reference_url": "https://seed.bytedance.com/en/blog/seed2-1-officially-released-advancing-ai-productivity", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "none", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official", "prompt_style": "official default", "temperature": "unknown" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Release blog EN: Motion/perception chart (https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/user-upload/4xfa4mqq4zh7f.png)." }, { "model_id": "gemini-3.1-pro", "benchmark_id": "video_mme", "score": 86.7, "reference_url": "https://seed.bytedance.com/en/blog/seed2-1-officially-released-advancing-ai-productivity", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "none", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official", "prompt_style": "official default", "temperature": "unknown" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Release blog EN: Long/streaming-video chart (https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/user-upload/4xfa4mqq50ia3.png). Source marks this row with an unexplained dagger. Source marks this row with an unexplained dagger." }, { "model_id": "gemini-3.1-pro", "benchmark_id": "videoholmes", "score": 65.9, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2.1/Seed2_1_Model_Card.pdf", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "none", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official", "prompt_style": "official default", "temperature": "unknown" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Model card Table 10 (p52). Source marks this row with an unexplained dagger. Source marks this row with an unexplained dagger." }, { "model_id": "gemini-3.1-pro", "benchmark_id": "videosimpleqa", "score": 70.0, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2.1/Seed2_1_Model_Card.pdf", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "none", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official", "prompt_style": "official default", "temperature": "unknown" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Model card Table 10 (p52)." }, { "model_id": "gemini-3.1-pro", "benchmark_id": "visfactor", "score": 54.0, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2.1/Seed2_1_Model_Card.pdf", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "none", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official", "prompt_style": "official default", "temperature": "unknown" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Model card Table 9 (p51)." }, { "model_id": "gemini-3.1-pro", "benchmark_id": "visulogic", "score": 44.8, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2.1/Seed2_1_Model_Card.pdf", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "none", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official", "prompt_style": "official default", "temperature": "unknown" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Model card Table 9 (p51)." }, { "model_id": "gemini-3.1-pro", "benchmark_id": "vlms_are_biased", "score": 73.5, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2.1/Seed2_1_Model_Card.pdf", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "none", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official", "prompt_style": "official default", "temperature": "unknown" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Model card Table 9 (p51)." }, { "model_id": "gemini-3.1-pro", "benchmark_id": "webbench", "score": 71.9, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2.1/Seed2_1_Model_Card.pdf", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "Trae or benchmark-specific coding-agent environment", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official", "prompt_style": "official default", "temperature": "unknown" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Model card Table 5 (p19)." }, { "model_id": "gemini-3.1-pro", "benchmark_id": "wildclaw_bench_60_openclaw", "score": 61.1, "reference_url": "https://seed.bytedance.com/en/blog/seed2-1-officially-released-advancing-ai-productivity", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "high-resolution multimodal agent harness", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official", "prompt_style": "official default", "temperature": "unknown" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Release blog EN: Multimodal-agent chart (https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/user-upload/4xfa4mqq4v1ol.png)." }, { "model_id": "gemini-3.1-pro", "benchmark_id": "workspace_bench_openclaw_100", "score": 32.8, "reference_url": "https://seed.bytedance.com/en/blog/seed2-1-officially-released-advancing-ai-productivity", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "OpenClaw workspace tools", "sampling": "pass@1", "judge": "rubric-based evaluator", "harness": "100-task OpenClaw setting", "prompt_style": "official default", "temperature": "unknown" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Release blog EN: General-agent chart (https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/user-upload/4xfa4mqq4u89c.png)." }, { "model_id": "gemini-3.1-pro", "benchmark_id": "workspace_bench_openclaw_100_pass100", "score": 4.0, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2.1/Seed2_1_Model_Card.pdf", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "OpenClaw workspace tools", "sampling": "pass@1", "judge": "rubric-based evaluator", "harness": "100-task OpenClaw setting", "prompt_style": "official default", "temperature": "unknown" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Model card Table 1 (p9)." }, { "model_id": "gemini-3.1-pro", "benchmark_id": "workspace_bench_openclaw_100_pass30", "score": 49.0, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2.1/Seed2_1_Model_Card.pdf", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "OpenClaw workspace tools", "sampling": "pass@1", "judge": "rubric-based evaluator", "harness": "100-task OpenClaw setting", "prompt_style": "official default", "temperature": "unknown" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Model card Table 1 (p9)." }, { "model_id": "gemini-3.1-pro", "benchmark_id": "workspace_bench_openclaw_100_pass50", "score": 29.0, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2.1/Seed2_1_Model_Card.pdf", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "OpenClaw workspace tools", "sampling": "pass@1", "judge": "rubric-based evaluator", "harness": "100-task OpenClaw setting", "prompt_style": "official default", "temperature": "unknown" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Model card Table 1 (p9)." }, { "model_id": "gemini-3.1-pro", "benchmark_id": "workspace_bench_openclaw_100_pass70", "score": 17.0, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2.1/Seed2_1_Model_Card.pdf", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "OpenClaw workspace tools", "sampling": "pass@1", "judge": "rubric-based evaluator", "harness": "100-task OpenClaw setting", "prompt_style": "official default", "temperature": "unknown" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Model card Table 1 (p9)." }, { "model_id": "gemini-3.1-pro", "benchmark_id": "workspace_bench_openclaw_100_pass90", "score": 9.0, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2.1/Seed2_1_Model_Card.pdf", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "OpenClaw workspace tools", "sampling": "pass@1", "judge": "rubric-based evaluator", "harness": "100-task OpenClaw setting", "prompt_style": "official default", "temperature": "unknown" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Model card Table 1 (p9)." }, { "model_id": "gemini-3.1-pro", "benchmark_id": "worldbench", "score": 62.9, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2.1/Seed2_1_Model_Card.pdf", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "none", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official", "prompt_style": "official default", "temperature": "unknown" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Model card Table 9 (p51)." }, { "model_id": "gemini-3.1-pro", "benchmark_id": "worldvqa", "score": 44.3, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2.1/Seed2_1_Model_Card.pdf", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "none", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official", "prompt_style": "official default", "temperature": "unknown" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Model card Table 9 (p51)." }, { "model_id": "gemini-3.1-pro", "benchmark_id": "xdailybench", "score": 35.2, "reference_url": "https://seed.bytedance.com/en/blog/seed2-1-officially-released-advancing-ai-productivity", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "none", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official", "prompt_style": "official default", "temperature": "unknown" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Release blog EN: Daily-life/tool-use chart (https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/user-upload/4xfa4mqq4ujsp.png)." }, { "model_id": "gemini-3.1-pro", "benchmark_id": "zerobench_main", "score": 12.0, "reference_url": "https://seed.bytedance.com/en/blog/seed2-1-officially-released-advancing-ai-productivity", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "none", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official", "prompt_style": "official default", "temperature": "unknown" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Release blog EN: Vision capability chart (https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/user-upload/4xfa4mqq4yer3.png)." }, { "model_id": "gemini-3.1-pro", "benchmark_id": "zerobench_sub", "score": 41.9, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2.1/Seed2_1_Model_Card.pdf", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "none", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official", "prompt_style": "official default", "temperature": "unknown" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Model card Table 9 (p51)." }, { "model_id": "gemini-3.1-pro-deep-think-2026-02", "benchmark_id": "frontier_science_research", "score": 23.3, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2.1/Seed2_1_Model_Card.pdf", "reported_setting": { "mode": "deep-think", "effort": "unknown", "tools": "benchmark-specified", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official", "prompt_style": "official default", "temperature": "unknown" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Model card Table 7 (p24)." }, { "model_id": "gemini-3.1-pro-deep-think-2026-02", "benchmark_id": "imo_proofbench_advanced", "score": 76.7, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2.1/Seed2_1_Model_Card.pdf", "reported_setting": { "mode": "deep-think", "effort": "unknown", "tools": "benchmark-specified", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official", "prompt_style": "official default", "temperature": "unknown" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Model card Table 7 (p24)." }, { "model_id": "gemini-3.5-flash", "benchmark_id": "code_arena_frontend_elo", "score": 1506, "reference_url": "https://seed.bytedance.com/en/blog/seed2-1-officially-released-advancing-ai-productivity", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "none", "sampling": "107,962-vote Arena snapshot", "judge": "human preference", "harness": "Arena Code leaderboard", "prompt_style": "official default", "temperature": "unknown" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Release blog EN: Code Arena: Frontend snapshot (https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/user-upload/4xfa4mqq4xsvz.jpg). Historical leaderboard snapshot reproduced by ByteDance." }, { "model_id": "gemini-3.5-flash", "benchmark_id": "contphy", "score": 63.8, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2.1/Seed2_1_Model_Card.pdf", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "none", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official", "prompt_style": "official default", "temperature": "unknown" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Model card Table 10 (p52)." }, { "model_id": "gemini-3.5-flash", "benchmark_id": "crossvid", "score": 58.6, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2.1/Seed2_1_Model_Card.pdf", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "none", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official", "prompt_style": "official default", "temperature": "unknown" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Model card Table 10 (p52)." }, { "model_id": "gemini-3.5-flash", "benchmark_id": "livesports_3k", "score": 86.5, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2.1/Seed2_1_Model_Card.pdf", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "none", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official", "prompt_style": "official default", "temperature": "unknown" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Model card Table 10 (p52)." }, { "model_id": "gemini-3.5-flash", "benchmark_id": "longvideobench", "score": 77.6, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2.1/Seed2_1_Model_Card.pdf", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "none", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official", "prompt_style": "official default", "temperature": "unknown" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Model card Table 10 (p52)." }, { "model_id": "gemini-3.5-flash", "benchmark_id": "lvbench", "score": 76.3, "reference_url": "https://seed.bytedance.com/en/blog/seed2-1-officially-released-advancing-ai-productivity", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "none", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official", "prompt_style": "official default", "temperature": "unknown" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Release blog EN: Long/streaming-video chart (https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/user-upload/4xfa4mqq50ia3.png)." }, { "model_id": "gemini-3.5-flash", "benchmark_id": "minerva", "score": 68.6, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2.1/Seed2_1_Model_Card.pdf", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "none", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official", "prompt_style": "official default", "temperature": "unknown" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Model card Table 10 (p52). Source marks this row with an unexplained dagger. Source marks this row with an unexplained dagger." }, { "model_id": "gemini-3.5-flash", "benchmark_id": "motionbench", "score": 70.6, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2.1/Seed2_1_Model_Card.pdf", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "none", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official", "prompt_style": "official default", "temperature": "unknown" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Model card Table 10 (p52)." }, { "model_id": "gemini-3.5-flash", "benchmark_id": "ovbench", "score": 56.5, "reference_url": "https://seed.bytedance.com/en/blog/seed2-1-officially-released-advancing-ai-productivity", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "none", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official", "prompt_style": "official default", "temperature": "unknown" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Release blog EN: Long/streaming-video chart (https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/user-upload/4xfa4mqq50ia3.png)." }, { "model_id": "gemini-3.5-flash", "benchmark_id": "ovobench", "score": 64.5, "reference_url": "https://seed.bytedance.com/en/blog/seed2-1-officially-released-advancing-ai-productivity", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "none", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official", "prompt_style": "official default", "temperature": "unknown" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Release blog EN: Long/streaming-video chart (https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/user-upload/4xfa4mqq50ia3.png)." }, { "model_id": "gemini-3.5-flash", "benchmark_id": "tomato", "score": 71.9, "reference_url": "https://seed.bytedance.com/en/blog/seed2-1-officially-released-advancing-ai-productivity", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "none", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official", "prompt_style": "official default", "temperature": "unknown" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Release blog EN: Motion/perception chart (https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/user-upload/4xfa4mqq4zh7f.png)." }, { "model_id": "gemini-3.5-flash", "benchmark_id": "tvbench", "score": 76.4, "reference_url": "https://seed.bytedance.com/en/blog/seed2-1-officially-released-advancing-ai-productivity", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "none", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official", "prompt_style": "official default", "temperature": "unknown" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Release blog EN: Motion/perception chart (https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/user-upload/4xfa4mqq4zh7f.png)." }, { "model_id": "gemini-3.5-flash", "benchmark_id": "video_mme", "score": 87.2, "reference_url": "https://seed.bytedance.com/en/blog/seed2-1-officially-released-advancing-ai-productivity", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "none", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official", "prompt_style": "official default", "temperature": "unknown" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Release blog EN: Long/streaming-video chart (https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/user-upload/4xfa4mqq50ia3.png). Source marks this row with an unexplained dagger. Source marks this row with an unexplained dagger." }, { "model_id": "gemini-3.5-flash", "benchmark_id": "videoholmes", "score": 67.1, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2.1/Seed2_1_Model_Card.pdf", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "none", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official", "prompt_style": "official default", "temperature": "unknown" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Model card Table 10 (p52). Source marks this row with an unexplained dagger. Source marks this row with an unexplained dagger." }, { "model_id": "gemini-3.5-flash", "benchmark_id": "videosimpleqa", "score": 76.0, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2.1/Seed2_1_Model_Card.pdf", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "none", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official", "prompt_style": "official default", "temperature": "unknown" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Model card Table 10 (p52)." }, { "model_id": "glm-5.1", "benchmark_id": "code_arena_frontend_elo", "score": 1529, "reference_url": "https://seed.bytedance.com/en/blog/seed2-1-officially-released-advancing-ai-productivity", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "none", "sampling": "107,962-vote Arena snapshot", "judge": "human preference", "harness": "Arena Code leaderboard", "prompt_style": "official default", "temperature": "unknown" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Release blog EN: Code Arena: Frontend snapshot (https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/user-upload/4xfa4mqq4xsvz.jpg). Historical leaderboard snapshot reproduced by ByteDance." }, { "model_id": "glm-5.2", "benchmark_id": "code_arena_frontend_elo", "score": 1593, "reference_url": "https://seed.bytedance.com/en/blog/seed2-1-officially-released-advancing-ai-productivity", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "none", "sampling": "107,962-vote Arena snapshot", "judge": "human preference", "harness": "Arena Code leaderboard", "prompt_style": "official default", "temperature": "unknown" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Release blog EN: Code Arena: Frontend snapshot (https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/user-upload/4xfa4mqq4xsvz.jpg). Historical leaderboard snapshot reproduced by ByteDance." }, { "model_id": "gpt-5.5", "benchmark_id": "aethercode", "score": 81.6, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2.1/Seed2_1_Model_Card.pdf", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "none", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official", "prompt_style": "official default", "temperature": "unknown" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Model card Table 11 (p53)." }, { "model_id": "gpt-5.5", "benchmark_id": "agent_startup_bench", "score": 68.1, "reference_url": "https://seed.bytedance.com/en/blog/seed2-1-officially-released-advancing-ai-productivity", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "research and professional-deliverable tools", "sampling": "pass@1", "judge": "expert review", "harness": "official", "prompt_style": "official default", "temperature": "unknown" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Release blog EN: General-agent chart (https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/user-upload/4xfa4mqq4u89c.png)." }, { "model_id": "gpt-5.5", "benchmark_id": "agents_last_exam_average_score", "score": 42.8, "reference_url": "https://seed.bytedance.com/en/blog/seed2-1-officially-released-advancing-ai-productivity", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "computer-use environment", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official", "prompt_style": "official default", "temperature": "unknown" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Release blog EN: General-agent chart (https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/user-upload/4xfa4mqq4u89c.png). Turbo is a source dash, not zero." }, { "model_id": "gpt-5.5", "benchmark_id": "beyond_aime", "score": 91.0, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2.1/Seed2_1_Model_Card.pdf", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "none", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official", "prompt_style": "official default", "temperature": "unknown" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Model card Table 11 (p53)." }, { "model_id": "gpt-5.5", "benchmark_id": "blink", "score": 78.3, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2.1/Seed2_1_Model_Card.pdf", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "none", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official", "prompt_style": "official default", "temperature": "unknown" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Model card Table 9 (p51)." }, { "model_id": "gpt-5.5", "benchmark_id": "charxiv_descriptive", "score": 95.0, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2.1/Seed2_1_Model_Card.pdf", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "none", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official", "prompt_style": "official default", "temperature": "unknown" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Model card Table 9 (p51)." }, { "model_id": "gpt-5.5", "benchmark_id": "claw_eval_multimodal_pass3", "score": 43.0, "reference_url": "https://seed.bytedance.com/en/blog/seed2-1-officially-released-advancing-ai-productivity", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "high-resolution multimodal agent harness", "sampling": "Pass^3", "judge": "benchmark-specified", "harness": "official", "prompt_style": "official default", "temperature": "unknown" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Release blog EN: Multimodal-agent chart (https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/user-upload/4xfa4mqq4v1ol.png)." }, { "model_id": "gpt-5.5", "benchmark_id": "creativework", "score": 30.5, "reference_url": "https://seed.bytedance.com/en/blog/seed2-1-officially-released-advancing-ai-productivity", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "GUI + MCP in Notion, Canva, and Figma", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official", "prompt_style": "official default", "temperature": "unknown" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Release blog EN: Computer-use chart (https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/user-upload/4xfa4mqq4w8x3.png)." }, { "model_id": "gpt-5.5", "benchmark_id": "doubao_multi_turn_bench", "score": 62.5, "reference_url": "https://seed.bytedance.com/en/blog/seed2-1-officially-released-advancing-ai-productivity", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "none", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official", "prompt_style": "official default", "temperature": "unknown" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Release blog EN: Daily-life/tool-use chart (https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/user-upload/4xfa4mqq4ujsp.png)." }, { "model_id": "gpt-5.5", "benchmark_id": "dude", "score": 81.7, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2.1/Seed2_1_Model_Card.pdf", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "none", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official", "prompt_style": "official default", "temperature": "unknown" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Model card Table 9 (p51)." }, { "model_id": "gpt-5.5", "benchmark_id": "dynamath", "score": 75.9, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2.1/Seed2_1_Model_Card.pdf", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "none", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official", "prompt_style": "official default", "temperature": "unknown" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Model card Table 9 (p51)." }, { "model_id": "gpt-5.5", "benchmark_id": "embspatial_bench", "score": 81.9, "reference_url": "https://seed.bytedance.com/en/blog/seed2-1-officially-released-advancing-ai-productivity", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "none", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official", "prompt_style": "official default", "temperature": "unknown" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Release blog EN: Spatial/long-context chart (https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/user-upload/4xfa4mqq4z2l5.png)." }, { "model_id": "gpt-5.5", "benchmark_id": "emma", "score": 79.0, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2.1/Seed2_1_Model_Card.pdf", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "none", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official", "prompt_style": "official default", "temperature": "unknown" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Model card Table 9 (p51)." }, { "model_id": "gpt-5.5", "benchmark_id": "erqa", "score": 64.5, "reference_url": "https://seed.bytedance.com/en/blog/seed2-1-officially-released-advancing-ai-productivity", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "none", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official", "prompt_style": "official default", "temperature": "unknown" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Release blog EN: Spatial/long-context chart (https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/user-upload/4xfa4mqq4z2l5.png)." }, { "model_id": "gpt-5.5", "benchmark_id": "frontiercs_algorithmic_v1", "score": 40.0, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2.1/Seed2_1_Model_Card.pdf", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "research and code-execution environment", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official", "prompt_style": "official default", "temperature": "unknown" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Model card Table 6 (p24). Claude Opus 4.7 is a source dash; official sub-score name unresolved." }, { "model_id": "gpt-5.5", "benchmark_id": "frontiercs_overall_v1", "score": 41.0, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2.1/Seed2_1_Model_Card.pdf", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "research and code-execution environment", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official", "prompt_style": "official default", "temperature": "unknown" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Model card Table 6 (p24). Claude Opus 4.7 is a source dash; official sub-score name unresolved." }, { "model_id": "gpt-5.5", "benchmark_id": "frontiercs_research_v1", "score": 58.6, "reference_url": "https://seed.bytedance.com/en/blog/seed2-1-officially-released-advancing-ai-productivity", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "research and code-execution environment", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official", "prompt_style": "official default", "temperature": "unknown" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Release blog EN: Frontier-research chart (https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/user-upload/4xfa4mqq52cdr.png). Claude Opus 4.7 is a source dash; official sub-score name unresolved." }, { "model_id": "gpt-5.5", "benchmark_id": "fs_researcher", "score": 33.9, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2.1/Seed2_1_Model_Card.pdf", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "none", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official", "prompt_style": "official default", "temperature": "unknown" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Model card Table 11 (p53)." }, { "model_id": "gpt-5.5", "benchmark_id": "gameworld", "score": 34.1, "reference_url": "https://seed.bytedance.com/en/blog/seed2-1-officially-released-advancing-ai-productivity", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "browser-game GUI actions", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official", "prompt_style": "official default", "temperature": "unknown" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Release blog EN: Computer-use chart (https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/user-upload/4xfa4mqq4w8x3.png)." }, { "model_id": "gpt-5.5", "benchmark_id": "gdpval_seed_reported_score", "score": 84.9, "reference_url": "https://seed.bytedance.com/en/blog/seed2-1-officially-released-advancing-ai-productivity", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "document/spreadsheet deliverable workflow", "sampling": "pass@1", "judge": "source-specific evaluator", "harness": "official", "prompt_style": "official default", "temperature": "unknown" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Release blog EN: General-agent chart (https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/user-upload/4xfa4mqq4u89c.png)." }, { "model_id": "gpt-5.5", "benchmark_id": "hle_verified", "score": 50.4, "reference_url": "https://seed.bytedance.com/en/blog/seed2-1-officially-released-advancing-ai-productivity", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "none", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official", "prompt_style": "official default", "temperature": "unknown" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Release blog EN: Language/search chart (https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/user-upload/4xfa4mqq510uc.png)." }, { "model_id": "gpt-5.5", "benchmark_id": "image2floorplan_avg_score", "score": 50.7, "reference_url": "https://seed.bytedance.com/en/blog/seed2-1-officially-released-advancing-ai-productivity", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "high-resolution multimodal agent harness", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official", "prompt_style": "official default", "temperature": "unknown" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Release blog EN: Multimodal-agent chart (https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/user-upload/4xfa4mqq4v1ol.png)." }, { "model_id": "gpt-5.5", "benchmark_id": "kina", "score": 52.6, "reference_url": "https://seed.bytedance.com/en/blog/seed2-1-officially-released-advancing-ai-productivity", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "none", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official", "prompt_style": "official default", "temperature": "unknown" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Release blog EN: Language/search chart (https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/user-upload/4xfa4mqq510uc.png)." }, { "model_id": "gpt-5.5", "benchmark_id": "live_mathematician_bench_2026_06", "score": 50.8, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2.1/Seed2_1_Model_Card.pdf", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "none", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official", "prompt_style": "official default", "temperature": "unknown" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Model card Table 11 (p53)." }, { "model_id": "gpt-5.5", "benchmark_id": "mathverse_vision_only", "score": 84.6, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2.1/Seed2_1_Model_Card.pdf", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "none", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official", "prompt_style": "official default", "temperature": "unknown" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Model card Table 9 (p51)." }, { "model_id": "gpt-5.5", "benchmark_id": "mathvista", "score": 84.2, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2.1/Seed2_1_Model_Card.pdf", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "none", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official", "prompt_style": "official default", "temperature": "unknown" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Model card Table 9 (p51)." }, { "model_id": "gpt-5.5", "benchmark_id": "measurebench", "score": 49.9, "reference_url": "https://seed.bytedance.com/en/blog/seed2-1-officially-released-advancing-ai-productivity", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "none", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official", "prompt_style": "official default", "temperature": "unknown" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Release blog EN: Vision capability chart (https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/user-upload/4xfa4mqq4yer3.png)." }, { "model_id": "gpt-5.5", "benchmark_id": "mmsibench_circular", "score": 36.0, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2.1/Seed2_1_Model_Card.pdf", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "none", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official", "prompt_style": "official default", "temperature": "unknown" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Model card Table 9 (p51)." }, { "model_id": "gpt-5.5", "benchmark_id": "mobileworld", "score": 54.7, "reference_url": "https://seed.bytedance.com/en/blog/seed2-1-officially-released-advancing-ai-productivity", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "mobile GUI actions", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official", "prompt_style": "official default", "temperature": "unknown" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Release blog EN: Computer-use chart (https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/user-upload/4xfa4mqq4w8x3.png)." }, { "model_id": "gpt-5.5", "benchmark_id": "msqa", "score": 57.4, "reference_url": "https://seed.bytedance.com/en/blog/seed2-1-officially-released-advancing-ai-productivity", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "none", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official", "prompt_style": "official default", "temperature": "unknown" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Release blog EN: Language/search chart (https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/user-upload/4xfa4mqq510uc.png)." }, { "model_id": "gpt-5.5", "benchmark_id": "ocrbench_v2", "score": 61.1, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2.1/Seed2_1_Model_Card.pdf", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "none", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official", "prompt_style": "official default", "temperature": "unknown" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Model card Table 9 (p51)." }, { "model_id": "gpt-5.5", "benchmark_id": "one_million_bench", "score": 69.6, "reference_url": "https://seed.bytedance.com/en/blog/seed2-1-officially-released-advancing-ai-productivity", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "agentic professional-task environment", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official", "prompt_style": "official default", "temperature": "unknown" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Release blog EN: General-agent chart (https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/user-upload/4xfa4mqq4u89c.png)." }, { "model_id": "gpt-5.5", "benchmark_id": "presentbench", "score": 68.9, "reference_url": "https://seed.bytedance.com/en/blog/seed2-1-officially-released-advancing-ai-productivity", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "slide-generation environment", "sampling": "pass@1", "judge": "fine-grained rubric evaluator", "harness": "official", "prompt_style": "official default", "temperature": "unknown" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Release blog EN: General-agent chart (https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/user-upload/4xfa4mqq4u89c.png)." }, { "model_id": "gpt-5.5", "benchmark_id": "program_bench_almost_resolved_95", "score": 5.5, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2.1/Seed2_1_Model_Card.pdf", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "compiled executable and documentation", "sampling": "pass@1", "judge": "248,000+ behavioral tests", "harness": "official ProgramBench sandbox", "prompt_style": "official default", "temperature": "unknown" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Model card Table 5 (p19)." }, { "model_id": "gpt-5.5", "benchmark_id": "program_bench_fully_resolved", "score": 0.5, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2.1/Seed2_1_Model_Card.pdf", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "compiled executable and documentation", "sampling": "pass@1", "judge": "248,000+ behavioral tests", "harness": "official ProgramBench sandbox", "prompt_style": "official default", "temperature": "unknown" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Model card Table 5 (p19)." }, { "model_id": "gpt-5.5", "benchmark_id": "realworldqa", "score": 82.2, "reference_url": "https://seed.bytedance.com/en/blog/seed2-1-officially-released-advancing-ai-productivity", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "none", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official", "prompt_style": "official default", "temperature": "unknown" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Release blog EN: Vision capability chart (https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/user-upload/4xfa4mqq4yer3.png)." }, { "model_id": "gpt-5.5", "benchmark_id": "seed21_dirtyfilter_precision", "score": 94.1, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2.1/Seed2_1_Model_Card.pdf", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "rule-only evaluator/optimizer loop", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "300M-token budget; identical validation set", "prompt_style": "official default", "temperature": "unknown" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Model card Figure 19 (p32)." }, { "model_id": "gpt-5.5", "benchmark_id": "seed21_dirtyfilter_validation_recall", "score": 44.5, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2.1/Seed2_1_Model_Card.pdf", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "rule-only evaluator/optimizer loop", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "300M-token budget; identical validation set", "prompt_style": "official default", "temperature": "unknown" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Model card Figure 19 (p32)." }, { "model_id": "gpt-5.5", "benchmark_id": "seedclawbench", "score": 66.4, "reference_url": "https://seed.bytedance.com/en/blog/seed2-1-officially-released-advancing-ai-productivity", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "OpenClaw-style tools and skills", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official", "prompt_style": "official default", "temperature": "unknown" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Release blog EN: Daily-life/tool-use chart (https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/user-upload/4xfa4mqq4ujsp.png)." }, { "model_id": "gpt-5.5", "benchmark_id": "seedkernelbench_avg_speedup", "score": 9.14, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2.1/Seed2_1_Model_Card.pdf", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "Trae or benchmark-specific coding-agent environment", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official", "prompt_style": "official default", "temperature": "unknown" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Model card Table 5 (p19). Source dashes are not zero." }, { "model_id": "gpt-5.5", "benchmark_id": "simplevqa", "score": 58.6, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2.1/Seed2_1_Model_Card.pdf", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "none", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official", "prompt_style": "official default", "temperature": "unknown" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Model card Table 9 (p51).", "candidates": [ { "score": 79.11, "reference_url": "https://static.stepfun.com/blog/step-3.7-flash/", "source_type": "official_blog", "reported_setting": { "quantization_config": "common release evaluation; artifact precision not established", "configuration": "base Step evaluation configuration", "mode": "source does not state", "effort": "source does not state", "tools_search": "Visual Search", "shots": "source does not state", "trials_samples": "source does not state", "aggregation_pass_k": "source-reported score", "context": "source does not state", "output_cap": "source does not state", "temperature": "source does not state", "top_p": "source does not state", "judge": "LLM-as-judge; exact StepFun judge not stated", "harness_agent": "official StepFun evaluation; exact harness not stated", "dataset_version_split": "SimpleVQA official 2,025-item test release", "multimodal_input": true }, "notes": "Official StepFun benchmark-specific tool table." } ] }, { "model_id": "gpt-5.5", "benchmark_id": "supergpqa", "score": 72.7, "reference_url": "https://seed.bytedance.com/en/blog/seed2-1-officially-released-advancing-ai-productivity", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "none", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official", "prompt_style": "official default", "temperature": "unknown" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Release blog EN: Language/search chart (https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/user-upload/4xfa4mqq510uc.png)." }, { "model_id": "gpt-5.5", "benchmark_id": "swe_atlas", "score": 44.7, "reference_url": "https://seed.bytedance.com/en/blog/seed2-1-officially-released-advancing-ai-productivity", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "Trae or benchmark-specific coding-agent environment", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official", "prompt_style": "official default", "temperature": "unknown" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Release blog EN: Coding chart (https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/user-upload/4xfa4mqq4wlbt.png)." }, { "model_id": "gpt-5.5", "benchmark_id": "trae_artifacts", "score": 56.0, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2.1/Seed2_1_Model_Card.pdf", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "Trae or benchmark-specific coding-agent environment", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official", "prompt_style": "official default", "temperature": "unknown" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Model card Table 5 (p19)." }, { "model_id": "gpt-5.5", "benchmark_id": "trae_code_gen_js", "score": 76.0, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2.1/Seed2_1_Model_Card.pdf", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "Trae or benchmark-specific coding-agent environment", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official", "prompt_style": "official default", "temperature": "unknown" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Model card Table 5 (p19)." }, { "model_id": "gpt-5.5", "benchmark_id": "trae_code_gen_python", "score": 73.3, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2.1/Seed2_1_Model_Card.pdf", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "Trae or benchmark-specific coding-agent environment", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official", "prompt_style": "official default", "temperature": "unknown" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Model card Table 5 (p19)." }, { "model_id": "gpt-5.5", "benchmark_id": "trae_error_fix_go", "score": 70.0, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2.1/Seed2_1_Model_Card.pdf", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "Trae or benchmark-specific coding-agent environment", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official", "prompt_style": "official default", "temperature": "unknown" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Model card Table 5 (p19)." }, { "model_id": "gpt-5.5", "benchmark_id": "trae_error_fix_java", "score": 65.0, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2.1/Seed2_1_Model_Card.pdf", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "Trae or benchmark-specific coding-agent environment", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official", "prompt_style": "official default", "temperature": "unknown" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Model card Table 5 (p19)." }, { "model_id": "gpt-5.5", "benchmark_id": "trae_error_fix_js", "score": 79.4, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2.1/Seed2_1_Model_Card.pdf", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "Trae or benchmark-specific coding-agent environment", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official", "prompt_style": "official default", "temperature": "unknown" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Model card Table 5 (p19)." }, { "model_id": "gpt-5.5", "benchmark_id": "trae_error_fix_python", "score": 76.0, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2.1/Seed2_1_Model_Card.pdf", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "Trae or benchmark-specific coding-agent environment", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official", "prompt_style": "official default", "temperature": "unknown" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Model card Table 5 (p19)." }, { "model_id": "gpt-5.5", "benchmark_id": "trae_repo_env", "score": 90.0, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2.1/Seed2_1_Model_Card.pdf", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "Trae or benchmark-specific coding-agent environment", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official", "prompt_style": "official default", "temperature": "unknown" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Model card Table 5 (p19)." }, { "model_id": "gpt-5.5", "benchmark_id": "treebench", "score": 61.5, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2.1/Seed2_1_Model_Card.pdf", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "none", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official", "prompt_style": "official default", "temperature": "unknown" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Model card Table 9 (p51)." }, { "model_id": "gpt-5.5", "benchmark_id": "visfactor", "score": 56.2, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2.1/Seed2_1_Model_Card.pdf", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "none", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official", "prompt_style": "official default", "temperature": "unknown" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Model card Table 9 (p51)." }, { "model_id": "gpt-5.5", "benchmark_id": "visulogic", "score": 43.5, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2.1/Seed2_1_Model_Card.pdf", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "none", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official", "prompt_style": "official default", "temperature": "unknown" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Model card Table 9 (p51)." }, { "model_id": "gpt-5.5", "benchmark_id": "vlms_are_biased", "score": 49.8, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2.1/Seed2_1_Model_Card.pdf", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "none", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official", "prompt_style": "official default", "temperature": "unknown" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Model card Table 9 (p51)." }, { "model_id": "gpt-5.5", "benchmark_id": "webbench", "score": 81.3, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2.1/Seed2_1_Model_Card.pdf", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "Trae or benchmark-specific coding-agent environment", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official", "prompt_style": "official default", "temperature": "unknown" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Model card Table 5 (p19)." }, { "model_id": "gpt-5.5", "benchmark_id": "wildclaw_bench_60_openclaw", "score": 65.6, "reference_url": "https://seed.bytedance.com/en/blog/seed2-1-officially-released-advancing-ai-productivity", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "high-resolution multimodal agent harness", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official", "prompt_style": "official default", "temperature": "unknown" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Release blog EN: Multimodal-agent chart (https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/user-upload/4xfa4mqq4v1ol.png)." }, { "model_id": "gpt-5.5", "benchmark_id": "workspace_bench_openclaw_100", "score": 58.7, "reference_url": "https://seed.bytedance.com/en/blog/seed2-1-officially-released-advancing-ai-productivity", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "OpenClaw workspace tools", "sampling": "pass@1", "judge": "rubric-based evaluator", "harness": "100-task OpenClaw setting", "prompt_style": "official default", "temperature": "unknown" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Release blog EN: General-agent chart (https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/user-upload/4xfa4mqq4u89c.png)." }, { "model_id": "gpt-5.5", "benchmark_id": "workspace_bench_openclaw_100_pass100", "score": 20.0, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2.1/Seed2_1_Model_Card.pdf", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "OpenClaw workspace tools", "sampling": "pass@1", "judge": "rubric-based evaluator", "harness": "100-task OpenClaw setting", "prompt_style": "official default", "temperature": "unknown" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Model card Table 1 (p9)." }, { "model_id": "gpt-5.5", "benchmark_id": "workspace_bench_openclaw_100_pass30", "score": 81.0, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2.1/Seed2_1_Model_Card.pdf", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "OpenClaw workspace tools", "sampling": "pass@1", "judge": "rubric-based evaluator", "harness": "100-task OpenClaw setting", "prompt_style": "official default", "temperature": "unknown" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Model card Table 1 (p9)." }, { "model_id": "gpt-5.5", "benchmark_id": "workspace_bench_openclaw_100_pass50", "score": 67.0, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2.1/Seed2_1_Model_Card.pdf", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "OpenClaw workspace tools", "sampling": "pass@1", "judge": "rubric-based evaluator", "harness": "100-task OpenClaw setting", "prompt_style": "official default", "temperature": "unknown" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Model card Table 1 (p9)." }, { "model_id": "gpt-5.5", "benchmark_id": "workspace_bench_openclaw_100_pass70", "score": 44.0, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2.1/Seed2_1_Model_Card.pdf", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "OpenClaw workspace tools", "sampling": "pass@1", "judge": "rubric-based evaluator", "harness": "100-task OpenClaw setting", "prompt_style": "official default", "temperature": "unknown" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Model card Table 1 (p9)." }, { "model_id": "gpt-5.5", "benchmark_id": "workspace_bench_openclaw_100_pass90", "score": 25.0, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2.1/Seed2_1_Model_Card.pdf", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "OpenClaw workspace tools", "sampling": "pass@1", "judge": "rubric-based evaluator", "harness": "100-task OpenClaw setting", "prompt_style": "official default", "temperature": "unknown" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Model card Table 1 (p9)." }, { "model_id": "gpt-5.5", "benchmark_id": "worldbench", "score": 58.4, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2.1/Seed2_1_Model_Card.pdf", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "none", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official", "prompt_style": "official default", "temperature": "unknown" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Model card Table 9 (p51)." }, { "model_id": "gpt-5.5", "benchmark_id": "xdailybench", "score": 73.0, "reference_url": "https://seed.bytedance.com/en/blog/seed2-1-officially-released-advancing-ai-productivity", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "none", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official", "prompt_style": "official default", "temperature": "unknown" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Release blog EN: Daily-life/tool-use chart (https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/user-upload/4xfa4mqq4ujsp.png)." }, { "model_id": "gpt-5.5", "benchmark_id": "zerobench_sub", "score": 41.0, "reference_url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2.1/Seed2_1_Model_Card.pdf", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "none", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official", "prompt_style": "official default", "temperature": "unknown" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Model card Table 9 (p51)." }, { "model_id": "kimi-k2.6", "benchmark_id": "code_arena_frontend_elo", "score": 1513, "reference_url": "https://seed.bytedance.com/en/blog/seed2-1-officially-released-advancing-ai-productivity", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "none", "sampling": "107,962-vote Arena snapshot", "judge": "human preference", "harness": "Arena Code leaderboard", "prompt_style": "official default", "temperature": "unknown" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Release blog EN: Code Arena: Frontend snapshot (https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/user-upload/4xfa4mqq4xsvz.jpg). Historical leaderboard snapshot reproduced by ByteDance." }, { "model_id": "minimax-m3", "benchmark_id": "code_arena_frontend_elo", "score": 1505, "reference_url": "https://seed.bytedance.com/en/blog/seed2-1-officially-released-advancing-ai-productivity", "reported_setting": { "mode": "thinking", "effort": "unknown", "tools": "none", "sampling": "107,962-vote Arena snapshot", "judge": "human preference", "harness": "Arena Code leaderboard", "prompt_style": "official default", "temperature": "unknown" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Release blog EN: Code Arena: Frontend snapshot (https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/user-upload/4xfa4mqq4xsvz.jpg). Historical leaderboard snapshot reproduced by ByteDance." }, { "model_id": "qwen3.7-max", "benchmark_id": "code_arena_frontend_elo", "score": 1530, "reference_url": "https://seed.bytedance.com/en/blog/seed2-1-officially-released-advancing-ai-productivity", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "none", "sampling": "107,962-vote Arena snapshot", "judge": "human preference", "harness": "Arena Code leaderboard", "prompt_style": "official default", "temperature": "unknown" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Release blog EN: Code Arena: Frontend snapshot (https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/user-upload/4xfa4mqq4xsvz.jpg). Historical leaderboard snapshot reproduced by ByteDance." }, { "model_id": "longcat-flash-lite", "benchmark_id": "c_eval", "score": 86.55, "reference_url": "https://arxiv.org/html/2608.01662v2", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "benchmark-specified", "sampling": "accuracy", "judge": "benchmark-specified", "harness": "official LongCat evaluation; details not disclosed", "prompt_style": "official default", "temperature": "official model default", "context": "benchmark-specified" }, "matches_canonical": true, "source_type": "official_paper", "audit_status": "verified", "notes": "Table 16: C-Eval (acc); accuracy. " }, { "model_id": "longcat-flash-lite", "benchmark_id": "cmmlu", "score": 82.48, "reference_url": "https://arxiv.org/html/2608.01662v2", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "benchmark-specified", "sampling": "accuracy", "judge": "benchmark-specified", "harness": "official LongCat evaluation; details not disclosed", "prompt_style": "official default", "temperature": "official model default", "context": "benchmark-specified" }, "matches_canonical": true, "source_type": "official_paper", "audit_status": "verified", "notes": "Table 16: CMMLU (acc); accuracy. " }, { "model_id": "longcat-flash-lite", "benchmark_id": "gpqa_diamond", "score": 66.78, "reference_url": "https://arxiv.org/html/2608.01662v2", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "benchmark-specified", "sampling": "avg@16", "judge": "benchmark-specified", "harness": "official LongCat evaluation; details not disclosed", "prompt_style": "official default", "temperature": "official model default", "context": "benchmark-specified" }, "matches_canonical": false, "source_type": "official_paper", "audit_status": "verified", "notes": "Table 16: GPQA-Diamond (avg@16); avg@16. " }, { "model_id": "longcat-flash-lite", "benchmark_id": "math_500", "score": 96.8, "reference_url": "https://arxiv.org/html/2608.01662v2", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "benchmark-specified", "sampling": "accuracy", "judge": "benchmark-specified", "harness": "official LongCat evaluation; details not disclosed", "prompt_style": "official default", "temperature": "official model default", "context": "benchmark-specified" }, "matches_canonical": true, "source_type": "official_paper", "audit_status": "verified", "notes": "Table 16: MATH500 (acc); accuracy. " }, { "model_id": "longcat-flash-lite", "benchmark_id": "mmlu", "score": 85.52, "reference_url": "https://arxiv.org/html/2608.01662v2", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "benchmark-specified", "sampling": "accuracy", "judge": "benchmark-specified", "harness": "official LongCat evaluation; details not disclosed", "prompt_style": "official default", "temperature": "official model default", "context": "benchmark-specified" }, "matches_canonical": true, "source_type": "official_paper", "audit_status": "verified", "notes": "Table 16: MMLU (acc); accuracy. " }, { "model_id": "longcat-flash-lite", "benchmark_id": "mmlu_pro", "score": 78.29, "reference_url": "https://arxiv.org/html/2608.01662v2", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "benchmark-specified", "sampling": "accuracy", "judge": "benchmark-specified", "harness": "official LongCat evaluation; details not disclosed", "prompt_style": "official default", "temperature": "official model default", "context": "benchmark-specified" }, "matches_canonical": true, "source_type": "official_paper", "audit_status": "verified", "notes": "Table 16: MMLU-Pro (acc); accuracy. " }, { "model_id": "longcat-flash-lite", "benchmark_id": "swe_bench_multilingual", "score": 38.1, "reference_url": "https://arxiv.org/html/2608.01662v2", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "benchmark-specified", "sampling": "pass@1; source harness unspecified", "judge": "benchmark-specified", "harness": "official LongCat evaluation; details not disclosed", "prompt_style": "official default", "temperature": "official model default", "context": "benchmark-specified" }, "matches_canonical": false, "source_type": "official_paper", "audit_status": "verified", "notes": "Table 16: SWE-Bench Multilingual (acc); pass@1; source harness unspecified. " }, { "model_id": "longcat-flash-lite", "benchmark_id": "swe_bench_verified", "score": 54.4, "reference_url": "https://arxiv.org/html/2608.01662v2", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "benchmark-specified", "sampling": "pass@1; source harness unspecified", "judge": "benchmark-specified", "harness": "official LongCat evaluation; details not disclosed", "prompt_style": "official default", "temperature": "official model default", "context": "benchmark-specified" }, "matches_canonical": false, "source_type": "official_paper", "audit_status": "verified", "notes": "Table 16: SWE-Bench Verified (acc); pass@1; source harness unspecified. " }, { "model_id": "longcat-flash-lite", "benchmark_id": "tau2_bench_telecom", "score": 72.8, "reference_url": "https://arxiv.org/html/2608.01662v2", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "benchmark-specified", "sampling": "avg@4", "judge": "benchmark-specified", "harness": "official LongCat evaluation; details not disclosed", "prompt_style": "official default", "temperature": "official model default", "context": "benchmark-specified" }, "matches_canonical": false, "source_type": "official_paper", "audit_status": "verified", "notes": "Table 16: τ²-Telecom (avg@4); avg@4. " }, { "model_id": "longcat-flash-lite", "benchmark_id": "terminal_bench", "score": 33.75, "reference_url": "https://arxiv.org/html/2608.01662v2", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "benchmark-specified", "sampling": "Terminal-Bench 2.0 accuracy; harness unspecified", "judge": "benchmark-specified", "harness": "official LongCat evaluation; details not disclosed", "prompt_style": "official default", "temperature": "official model default", "context": "benchmark-specified" }, "matches_canonical": false, "source_type": "official_paper", "audit_status": "verified", "notes": "Table 16: TerminalBench 2.0 (acc); Terminal-Bench 2.0 accuracy; harness unspecified. " }, { "model_id": "longcat-flash-lite", "benchmark_id": "vitabench", "score": 7.0, "reference_url": "https://arxiv.org/html/2608.01662v2", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "benchmark-specified", "sampling": "avg@4", "judge": "benchmark-specified", "harness": "official LongCat evaluation; details not disclosed", "prompt_style": "official default", "temperature": "official model default", "context": "benchmark-specified" }, "matches_canonical": false, "source_type": "official_paper", "audit_status": "verified", "notes": "Table 16: VitaBench (avg@4); avg@4. " }, { "model_id": "longcat-flash-lite-sparse", "benchmark_id": "aa_lcr", "score": 48.0, "reference_url": "https://arxiv.org/html/2608.01662v2", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "one response per ATLAS instance", "judge": "deterministic benchmark-native evaluator", "harness": "Official source evaluation at benchmark-native instance lengths", "prompt_style": "ATLAS official", "temperature": "official model default", "context": "benchmark-native instance lengths", "hierarchical_indexing": "off" }, "matches_canonical": false, "source_type": "official_paper", "audit_status": "verified", "notes": "Table 15: AA-LCR; HI=off. ", "candidates": [ { "score": 47.33, "reference_url": "https://arxiv.org/html/2608.01662v2", "source_type": "official_paper", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "one response per ATLAS instance", "judge": "deterministic benchmark-native evaluator", "harness": "Official source evaluation at benchmark-native instance lengths", "prompt_style": "ATLAS official", "temperature": "official model default", "context": "benchmark-native instance lengths", "hierarchical_indexing": "on" }, "notes": "Table 15: AA-LCR; HI=on. " } ] }, { "model_id": "longcat-flash-lite-sparse", "benchmark_id": "aime_2026", "score": 65.73, "reference_url": "https://arxiv.org/html/2608.01662v2", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "benchmark-specified", "sampling": "avg@32", "judge": "benchmark-specified", "harness": "official LongCat evaluation; details not disclosed", "prompt_style": "official default", "temperature": "official model default", "context": "benchmark-specified", "hierarchical_indexing": "off" }, "matches_canonical": false, "source_type": "official_paper", "audit_status": "verified", "notes": "Table 16: AIME 2026 (avg@32); avg@32. ", "candidates": [ { "score": 64.9, "reference_url": "https://arxiv.org/html/2608.01662v2", "source_type": "official_paper", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "benchmark-specified", "sampling": "avg@32", "judge": "benchmark-specified", "harness": "official LongCat evaluation; details not disclosed", "prompt_style": "official default", "temperature": "official model default", "context": "benchmark-specified", "hierarchical_indexing": "on" }, "notes": "Table 16: AIME 2026 (avg@32); avg@32. " } ] }, { "model_id": "longcat-flash-lite-sparse", "benchmark_id": "atlas_amembench_acu_auc_1m", "score": 33.25, "reference_url": "https://arxiv.org/html/2608.01662v2", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "one response per ATLAS instance", "judge": "deterministic benchmark-native evaluator", "harness": "ATLAS full eight-slice evaluation through 1M", "prompt_style": "ATLAS official", "temperature": "official model default", "context": "through 1M", "hierarchical_indexing": "off" }, "matches_canonical": true, "source_type": "official_paper", "audit_status": "verified", "notes": "Table 15: AMemBench-ACU; HI=off. ", "candidates": [ { "score": 33.13, "reference_url": "https://arxiv.org/html/2608.01662v2", "source_type": "official_paper", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "one response per ATLAS instance", "judge": "deterministic benchmark-native evaluator", "harness": "ATLAS full eight-slice evaluation through 1M", "prompt_style": "ATLAS official", "temperature": "official model default", "context": "through 1M", "hierarchical_indexing": "on" }, "notes": "Table 15: AMemBench-ACU; HI=on. " } ] }, { "model_id": "longcat-flash-lite-sparse", "benchmark_id": "atlas_graphwalks_extend_auc_1m", "score": 66.27, "reference_url": "https://arxiv.org/html/2608.01662v2", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "one response per ATLAS instance", "judge": "deterministic benchmark-native evaluator", "harness": "ATLAS full eight-slice evaluation through 1M", "prompt_style": "ATLAS official", "temperature": "official model default", "context": "through 1M", "hierarchical_indexing": "off" }, "matches_canonical": true, "source_type": "official_paper", "audit_status": "verified", "notes": "Table 15: GraphWalks Extend; HI=off. ", "candidates": [ { "score": 65.63, "reference_url": "https://arxiv.org/html/2608.01662v2", "source_type": "official_paper", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "one response per ATLAS instance", "judge": "deterministic benchmark-native evaluator", "harness": "ATLAS full eight-slice evaluation through 1M", "prompt_style": "ATLAS official", "temperature": "official model default", "context": "through 1M", "hierarchical_indexing": "on" }, "notes": "Table 15: GraphWalks Extend; HI=on. " } ] }, { "model_id": "longcat-flash-lite-sparse", "benchmark_id": "atlas_helmet_icl_extend_auc_1m", "score": 91.63, "reference_url": "https://arxiv.org/html/2608.01662v2", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "one response per ATLAS instance", "judge": "deterministic benchmark-native evaluator", "harness": "ATLAS full eight-slice evaluation through 1M", "prompt_style": "ATLAS official", "temperature": "official model default", "context": "through 1M", "hierarchical_indexing": "off" }, "matches_canonical": true, "source_type": "official_paper", "audit_status": "verified", "notes": "Table 15: HELMET-ICL Extend; HI=off. ", "candidates": [ { "score": 90.5, "reference_url": "https://arxiv.org/html/2608.01662v2", "source_type": "official_paper", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "one response per ATLAS instance", "judge": "deterministic benchmark-native evaluator", "harness": "ATLAS full eight-slice evaluation through 1M", "prompt_style": "ATLAS official", "temperature": "official model default", "context": "through 1M", "hierarchical_indexing": "on" }, "notes": "Table 15: HELMET-ICL Extend; HI=on. " } ] }, { "model_id": "longcat-flash-lite-sparse", "benchmark_id": "atlas_loft_text_retrieval_extend_auc_1m", "score": 43.75, "reference_url": "https://arxiv.org/html/2608.01662v2", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "one response per ATLAS instance", "judge": "deterministic benchmark-native evaluator", "harness": "ATLAS full eight-slice evaluation through 1M", "prompt_style": "ATLAS official", "temperature": "official model default", "context": "through 1M", "hierarchical_indexing": "off" }, "matches_canonical": true, "source_type": "official_paper", "audit_status": "verified", "notes": "Table 15: LOFT Retrieval Extend; HI=off. ", "candidates": [ { "score": 44.38, "reference_url": "https://arxiv.org/html/2608.01662v2", "source_type": "official_paper", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "one response per ATLAS instance", "judge": "deterministic benchmark-native evaluator", "harness": "ATLAS full eight-slice evaluation through 1M", "prompt_style": "ATLAS official", "temperature": "official model default", "context": "through 1M", "hierarchical_indexing": "on" }, "notes": "Table 15: LOFT Retrieval Extend; HI=on. " } ] }, { "model_id": "longcat-flash-lite-sparse", "benchmark_id": "atlas_longcodeqa_auc_1m", "score": 62.3, "reference_url": "https://arxiv.org/html/2608.01662v2", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "one response per ATLAS instance", "judge": "deterministic benchmark-native evaluator", "harness": "ATLAS six-slice evaluation from 32K through 1M", "prompt_style": "ATLAS official", "temperature": "official model default", "context": "32K through 1M", "hierarchical_indexing": "off" }, "matches_canonical": true, "source_type": "official_paper", "audit_status": "verified", "notes": "Table 15: LongCodeQA; HI=off. ", "candidates": [ { "score": 59.37, "reference_url": "https://arxiv.org/html/2608.01662v2", "source_type": "official_paper", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "one response per ATLAS instance", "judge": "deterministic benchmark-native evaluator", "harness": "ATLAS six-slice evaluation from 32K through 1M", "prompt_style": "ATLAS official", "temperature": "official model default", "context": "32K through 1M", "hierarchical_indexing": "on" }, "notes": "Table 15: LongCodeQA; HI=on. " } ] }, { "model_id": "longcat-flash-lite-sparse", "benchmark_id": "atlas_mrcr_8needle_auc_1m", "score": 44.66, "reference_url": "https://arxiv.org/html/2608.01662v2", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "one response per ATLAS instance", "judge": "deterministic benchmark-native evaluator", "harness": "ATLAS full eight-slice evaluation through 1M", "prompt_style": "ATLAS official", "temperature": "official model default", "context": "through 1M", "hierarchical_indexing": "off" }, "matches_canonical": true, "source_type": "official_paper", "audit_status": "verified", "notes": "Table 15: MRCR (8-needle); HI=off. ", "candidates": [ { "score": 44.47, "reference_url": "https://arxiv.org/html/2608.01662v2", "source_type": "official_paper", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "one response per ATLAS instance", "judge": "deterministic benchmark-native evaluator", "harness": "ATLAS full eight-slice evaluation through 1M", "prompt_style": "ATLAS official", "temperature": "official model default", "context": "through 1M", "hierarchical_indexing": "on" }, "notes": "Table 15: MRCR (8-needle); HI=on. " } ] }, { "model_id": "longcat-flash-lite-sparse", "benchmark_id": "atlas_oolong_synth_auc_1m", "score": 38.42, "reference_url": "https://arxiv.org/html/2608.01662v2", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "one response per ATLAS instance", "judge": "deterministic benchmark-native evaluator", "harness": "ATLAS full eight-slice evaluation through 1M", "prompt_style": "ATLAS official", "temperature": "official model default", "context": "through 1M", "hierarchical_indexing": "off" }, "matches_canonical": true, "source_type": "official_paper", "audit_status": "verified", "notes": "Table 15: OOLong-Synth; HI=off. ", "candidates": [ { "score": 37.88, "reference_url": "https://arxiv.org/html/2608.01662v2", "source_type": "official_paper", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "one response per ATLAS instance", "judge": "deterministic benchmark-native evaluator", "harness": "ATLAS full eight-slice evaluation through 1M", "prompt_style": "ATLAS official", "temperature": "official model default", "context": "through 1M", "hierarchical_indexing": "on" }, "notes": "Table 15: OOLong-Synth; HI=on. " } ] }, { "model_id": "longcat-flash-lite-sparse", "benchmark_id": "beyond_aime", "score": 44.2, "reference_url": "https://arxiv.org/html/2608.01662v2", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "benchmark-specified", "sampling": "avg@10", "judge": "benchmark-specified", "harness": "official LongCat evaluation; details not disclosed", "prompt_style": "official default", "temperature": "official model default", "context": "benchmark-specified", "hierarchical_indexing": "off" }, "matches_canonical": false, "source_type": "official_paper", "audit_status": "verified", "notes": "Table 16: BeyondAIME (avg@10); avg@10. ", "candidates": [ { "score": 42.3, "reference_url": "https://arxiv.org/html/2608.01662v2", "source_type": "official_paper", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "benchmark-specified", "sampling": "avg@10", "judge": "benchmark-specified", "harness": "official LongCat evaluation; details not disclosed", "prompt_style": "official default", "temperature": "official model default", "context": "benchmark-specified", "hierarchical_indexing": "on" }, "notes": "Table 16: BeyondAIME (avg@10); avg@10. " } ] }, { "model_id": "longcat-flash-lite-sparse", "benchmark_id": "browsecomp", "score": 48.62, "reference_url": "https://arxiv.org/html/2608.01662v2", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "benchmark-specified", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official LongCat evaluation; details not disclosed", "prompt_style": "official default", "temperature": "official model default", "context": "benchmark-specified", "hierarchical_indexing": "off" }, "matches_canonical": true, "source_type": "official_paper", "audit_status": "verified", "notes": "Table 16: BrowseComp (pass@1); pass@1. ", "candidates": [ { "score": 48.18, "reference_url": "https://arxiv.org/html/2608.01662v2", "source_type": "official_paper", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "benchmark-specified", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official LongCat evaluation; details not disclosed", "prompt_style": "official default", "temperature": "official model default", "context": "benchmark-specified", "hierarchical_indexing": "on" }, "notes": "Table 16: BrowseComp (pass@1); pass@1. " } ] }, { "model_id": "longcat-flash-lite-sparse", "benchmark_id": "browsecomp_zh", "score": 61.94, "reference_url": "https://arxiv.org/html/2608.01662v2", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "benchmark-specified", "sampling": "pass@1 over 289 questions", "judge": "benchmark-specified", "harness": "official LongCat evaluation; details not disclosed", "prompt_style": "official default", "temperature": "official model default", "context": "benchmark-specified", "hierarchical_indexing": "off" }, "matches_canonical": false, "source_type": "official_paper", "audit_status": "verified", "notes": "Table 16: BrowseComp-zh (pass@1); pass@1 over 289 questions. ", "candidates": [ { "score": 61.59, "reference_url": "https://arxiv.org/html/2608.01662v2", "source_type": "official_paper", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "benchmark-specified", "sampling": "pass@1 over 289 questions", "judge": "benchmark-specified", "harness": "official LongCat evaluation; details not disclosed", "prompt_style": "official default", "temperature": "official model default", "context": "benchmark-specified", "hierarchical_indexing": "on" }, "notes": "Table 16: BrowseComp-zh (pass@1); pass@1 over 289 questions. " } ] }, { "model_id": "longcat-flash-lite-sparse", "benchmark_id": "c_eval", "score": 85.76, "reference_url": "https://arxiv.org/html/2608.01662v2", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "benchmark-specified", "sampling": "accuracy", "judge": "benchmark-specified", "harness": "official LongCat evaluation; details not disclosed", "prompt_style": "official default", "temperature": "official model default", "context": "benchmark-specified", "hierarchical_indexing": "off" }, "matches_canonical": true, "source_type": "official_paper", "audit_status": "verified", "notes": "Table 16: C-Eval (acc); accuracy. ", "candidates": [ { "score": 85.71, "reference_url": "https://arxiv.org/html/2608.01662v2", "source_type": "official_paper", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "benchmark-specified", "sampling": "accuracy", "judge": "benchmark-specified", "harness": "official LongCat evaluation; details not disclosed", "prompt_style": "official default", "temperature": "official model default", "context": "benchmark-specified", "hierarchical_indexing": "on" }, "notes": "Table 16: C-Eval (acc); accuracy. " } ] }, { "model_id": "longcat-flash-lite-sparse", "benchmark_id": "cmmlu", "score": 84.25, "reference_url": "https://arxiv.org/html/2608.01662v2", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "benchmark-specified", "sampling": "accuracy", "judge": "benchmark-specified", "harness": "official LongCat evaluation; details not disclosed", "prompt_style": "official default", "temperature": "official model default", "context": "benchmark-specified", "hierarchical_indexing": "off" }, "matches_canonical": true, "source_type": "official_paper", "audit_status": "verified", "notes": "Table 16: CMMLU (acc); accuracy. ", "candidates": [ { "score": 84.51, "reference_url": "https://arxiv.org/html/2608.01662v2", "source_type": "official_paper", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "benchmark-specified", "sampling": "accuracy", "judge": "benchmark-specified", "harness": "official LongCat evaluation; details not disclosed", "prompt_style": "official default", "temperature": "official model default", "context": "benchmark-specified", "hierarchical_indexing": "on" }, "notes": "Table 16: CMMLU (acc); accuracy. " } ] }, { "model_id": "longcat-flash-lite-sparse", "benchmark_id": "gpqa_diamond", "score": 69.49, "reference_url": "https://arxiv.org/html/2608.01662v2", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "benchmark-specified", "sampling": "avg@16", "judge": "benchmark-specified", "harness": "official LongCat evaluation; details not disclosed", "prompt_style": "official default", "temperature": "official model default", "context": "benchmark-specified", "hierarchical_indexing": "off" }, "matches_canonical": false, "source_type": "official_paper", "audit_status": "verified", "notes": "Table 16: GPQA-Diamond (avg@16); avg@16. ", "candidates": [ { "score": 69.03, "reference_url": "https://arxiv.org/html/2608.01662v2", "source_type": "official_paper", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "benchmark-specified", "sampling": "avg@16", "judge": "benchmark-specified", "harness": "official LongCat evaluation; details not disclosed", "prompt_style": "official default", "temperature": "official model default", "context": "benchmark-specified", "hierarchical_indexing": "on" }, "notes": "Table 16: GPQA-Diamond (avg@16); avg@16. " } ] }, { "model_id": "longcat-flash-lite-sparse", "benchmark_id": "hmmt_feb_2026", "score": 40.53, "reference_url": "https://arxiv.org/html/2608.01662v2", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "benchmark-specified", "sampling": "avg@32", "judge": "benchmark-specified", "harness": "official LongCat evaluation; details not disclosed", "prompt_style": "official default", "temperature": "official model default", "context": "benchmark-specified", "hierarchical_indexing": "off" }, "matches_canonical": false, "source_type": "official_paper", "audit_status": "verified", "notes": "Table 16: HMMT 2026 Feb (avg@32); avg@32. ", "candidates": [ { "score": 41.47, "reference_url": "https://arxiv.org/html/2608.01662v2", "source_type": "official_paper", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "benchmark-specified", "sampling": "avg@32", "judge": "benchmark-specified", "harness": "official LongCat evaluation; details not disclosed", "prompt_style": "official default", "temperature": "official model default", "context": "benchmark-specified", "hierarchical_indexing": "on" }, "notes": "Table 16: HMMT 2026 Feb (avg@32); avg@32. " } ] }, { "model_id": "longcat-flash-lite-sparse", "benchmark_id": "imo_answerbench", "score": 49.38, "reference_url": "https://arxiv.org/html/2608.01662v2", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "benchmark-specified", "sampling": "avg@4", "judge": "benchmark-specified", "harness": "official LongCat evaluation; details not disclosed", "prompt_style": "official default", "temperature": "official model default", "context": "benchmark-specified", "hierarchical_indexing": "off" }, "matches_canonical": false, "source_type": "official_paper", "audit_status": "verified", "notes": "Table 16: IMO AnswerBench (avg@4); avg@4. ", "candidates": [ { "score": 46.69, "reference_url": "https://arxiv.org/html/2608.01662v2", "source_type": "official_paper", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "benchmark-specified", "sampling": "avg@4", "judge": "benchmark-specified", "harness": "official LongCat evaluation; details not disclosed", "prompt_style": "official default", "temperature": "official model default", "context": "benchmark-specified", "hierarchical_indexing": "on" }, "notes": "Table 16: IMO AnswerBench (avg@4); avg@4. " } ] }, { "model_id": "longcat-flash-lite-sparse", "benchmark_id": "longbench_v2", "score": 52.5, "reference_url": "https://arxiv.org/html/2608.01662v2", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "one response per ATLAS instance", "judge": "deterministic benchmark-native evaluator", "harness": "Official source evaluation at benchmark-native instance lengths", "prompt_style": "ATLAS official", "temperature": "official model default", "context": "benchmark-native instance lengths", "hierarchical_indexing": "off" }, "matches_canonical": false, "source_type": "official_paper", "audit_status": "verified", "notes": "Table 15: LongBench-v2; HI=off. ", "candidates": [ { "score": 53.64, "reference_url": "https://arxiv.org/html/2608.01662v2", "source_type": "official_paper", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "none", "sampling": "one response per ATLAS instance", "judge": "deterministic benchmark-native evaluator", "harness": "Official source evaluation at benchmark-native instance lengths", "prompt_style": "ATLAS official", "temperature": "official model default", "context": "benchmark-native instance lengths", "hierarchical_indexing": "on" }, "notes": "Table 15: LongBench-v2; HI=on. " } ] }, { "model_id": "longcat-flash-lite-sparse", "benchmark_id": "math_500", "score": 95.8, "reference_url": "https://arxiv.org/html/2608.01662v2", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "benchmark-specified", "sampling": "accuracy", "judge": "benchmark-specified", "harness": "official LongCat evaluation; details not disclosed", "prompt_style": "official default", "temperature": "official model default", "context": "benchmark-specified", "hierarchical_indexing": "off" }, "matches_canonical": true, "source_type": "official_paper", "audit_status": "verified", "notes": "Table 16: MATH500 (acc); accuracy. ", "candidates": [ { "score": 96.8, "reference_url": "https://arxiv.org/html/2608.01662v2", "source_type": "official_paper", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "benchmark-specified", "sampling": "accuracy", "judge": "benchmark-specified", "harness": "official LongCat evaluation; details not disclosed", "prompt_style": "official default", "temperature": "official model default", "context": "benchmark-specified", "hierarchical_indexing": "on" }, "notes": "Table 16: MATH500 (acc); accuracy. " } ] }, { "model_id": "longcat-flash-lite-sparse", "benchmark_id": "mmlu", "score": 85.31, "reference_url": "https://arxiv.org/html/2608.01662v2", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "benchmark-specified", "sampling": "accuracy", "judge": "benchmark-specified", "harness": "official LongCat evaluation; details not disclosed", "prompt_style": "official default", "temperature": "official model default", "context": "benchmark-specified", "hierarchical_indexing": "off" }, "matches_canonical": true, "source_type": "official_paper", "audit_status": "verified", "notes": "Table 16: MMLU (acc); accuracy. ", "candidates": [ { "score": 85.14, "reference_url": "https://arxiv.org/html/2608.01662v2", "source_type": "official_paper", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "benchmark-specified", "sampling": "accuracy", "judge": "benchmark-specified", "harness": "official LongCat evaluation; details not disclosed", "prompt_style": "official default", "temperature": "official model default", "context": "benchmark-specified", "hierarchical_indexing": "on" }, "notes": "Table 16: MMLU (acc); accuracy. " } ] }, { "model_id": "longcat-flash-lite-sparse", "benchmark_id": "mmlu_pro", "score": 79.24, "reference_url": "https://arxiv.org/html/2608.01662v2", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "benchmark-specified", "sampling": "accuracy", "judge": "benchmark-specified", "harness": "official LongCat evaluation; details not disclosed", "prompt_style": "official default", "temperature": "official model default", "context": "benchmark-specified", "hierarchical_indexing": "off" }, "matches_canonical": true, "source_type": "official_paper", "audit_status": "verified", "notes": "Table 16: MMLU-Pro (acc); accuracy. ", "candidates": [ { "score": 78.68, "reference_url": "https://arxiv.org/html/2608.01662v2", "source_type": "official_paper", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "benchmark-specified", "sampling": "accuracy", "judge": "benchmark-specified", "harness": "official LongCat evaluation; details not disclosed", "prompt_style": "official default", "temperature": "official model default", "context": "benchmark-specified", "hierarchical_indexing": "on" }, "notes": "Table 16: MMLU-Pro (acc); accuracy. " } ] }, { "model_id": "longcat-flash-lite-sparse", "benchmark_id": "rwsearch", "score": 68.5, "reference_url": "https://arxiv.org/html/2608.01662v2", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "benchmark-specified", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official LongCat evaluation; details not disclosed", "prompt_style": "official default", "temperature": "official model default", "context": "benchmark-specified", "hierarchical_indexing": "off" }, "matches_canonical": true, "source_type": "official_paper", "audit_status": "verified", "notes": "Table 16: RWSearch (pass@1); pass@1. ", "candidates": [ { "score": 66.0, "reference_url": "https://arxiv.org/html/2608.01662v2", "source_type": "official_paper", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "benchmark-specified", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official LongCat evaluation; details not disclosed", "prompt_style": "official default", "temperature": "official model default", "context": "benchmark-specified", "hierarchical_indexing": "on" }, "notes": "Table 16: RWSearch (pass@1); pass@1. " } ] }, { "model_id": "longcat-flash-lite-sparse", "benchmark_id": "swe_bench_multilingual", "score": 59.33, "reference_url": "https://arxiv.org/html/2608.01662v2", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "benchmark-specified", "sampling": "pass@1; source harness unspecified", "judge": "benchmark-specified", "harness": "official LongCat evaluation; details not disclosed", "prompt_style": "official default", "temperature": "official model default", "context": "benchmark-specified", "hierarchical_indexing": "off" }, "matches_canonical": false, "source_type": "official_paper", "audit_status": "verified", "notes": "Table 16: SWE-Bench Multilingual (acc); pass@1; source harness unspecified. ", "candidates": [ { "score": 56.0, "reference_url": "https://arxiv.org/html/2608.01662v2", "source_type": "official_paper", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "benchmark-specified", "sampling": "pass@1; source harness unspecified", "judge": "benchmark-specified", "harness": "official LongCat evaluation; details not disclosed", "prompt_style": "official default", "temperature": "official model default", "context": "benchmark-specified", "hierarchical_indexing": "on" }, "notes": "Table 16: SWE-Bench Multilingual (acc); pass@1; source harness unspecified. " } ] }, { "model_id": "longcat-flash-lite-sparse", "benchmark_id": "swe_bench_pro", "score": 40.63, "reference_url": "https://arxiv.org/html/2608.01662v2", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "benchmark-specified", "sampling": "pass@1; source harness unspecified", "judge": "benchmark-specified", "harness": "official LongCat evaluation; details not disclosed", "prompt_style": "official default", "temperature": "official model default", "context": "benchmark-specified", "hierarchical_indexing": "off" }, "matches_canonical": false, "source_type": "official_paper", "audit_status": "verified", "notes": "Table 16: SWE-Bench Pro (acc); pass@1; source harness unspecified. ", "candidates": [ { "score": 39.4, "reference_url": "https://arxiv.org/html/2608.01662v2", "source_type": "official_paper", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "benchmark-specified", "sampling": "pass@1; source harness unspecified", "judge": "benchmark-specified", "harness": "official LongCat evaluation; details not disclosed", "prompt_style": "official default", "temperature": "official model default", "context": "benchmark-specified", "hierarchical_indexing": "on" }, "notes": "Table 16: SWE-Bench Pro (acc); pass@1; source harness unspecified. " } ] }, { "model_id": "longcat-flash-lite-sparse", "benchmark_id": "swe_bench_verified", "score": 68.2, "reference_url": "https://arxiv.org/html/2608.01662v2", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "benchmark-specified", "sampling": "pass@1; source harness unspecified", "judge": "benchmark-specified", "harness": "official LongCat evaluation; details not disclosed", "prompt_style": "official default", "temperature": "official model default", "context": "benchmark-specified", "hierarchical_indexing": "off" }, "matches_canonical": false, "source_type": "official_paper", "audit_status": "verified", "notes": "Table 16: SWE-Bench Verified (acc); pass@1; source harness unspecified. ", "candidates": [ { "score": 65.2, "reference_url": "https://arxiv.org/html/2608.01662v2", "source_type": "official_paper", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "benchmark-specified", "sampling": "pass@1; source harness unspecified", "judge": "benchmark-specified", "harness": "official LongCat evaluation; details not disclosed", "prompt_style": "official default", "temperature": "official model default", "context": "benchmark-specified", "hierarchical_indexing": "on" }, "notes": "Table 16: SWE-Bench Verified (acc); pass@1; source harness unspecified. " } ] }, { "model_id": "longcat-flash-lite-sparse", "benchmark_id": "tau2_bench_telecom", "score": 95.18, "reference_url": "https://arxiv.org/html/2608.01662v2", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "benchmark-specified", "sampling": "avg@4", "judge": "benchmark-specified", "harness": "official LongCat evaluation; details not disclosed", "prompt_style": "official default", "temperature": "official model default", "context": "benchmark-specified", "hierarchical_indexing": "off" }, "matches_canonical": false, "source_type": "official_paper", "audit_status": "verified", "notes": "Table 16: τ²-Telecom (avg@4); avg@4. ", "candidates": [ { "score": 96.05, "reference_url": "https://arxiv.org/html/2608.01662v2", "source_type": "official_paper", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "benchmark-specified", "sampling": "avg@4", "judge": "benchmark-specified", "harness": "official LongCat evaluation; details not disclosed", "prompt_style": "official default", "temperature": "official model default", "context": "benchmark-specified", "hierarchical_indexing": "on" }, "notes": "Table 16: τ²-Telecom (avg@4); avg@4. " } ] }, { "model_id": "longcat-flash-lite-sparse", "benchmark_id": "terminal_bench", "score": 33.7, "reference_url": "https://arxiv.org/html/2608.01662v2", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "benchmark-specified", "sampling": "Terminal-Bench 2.0 accuracy; harness unspecified", "judge": "benchmark-specified", "harness": "official LongCat evaluation; details not disclosed", "prompt_style": "official default", "temperature": "official model default", "context": "benchmark-specified", "hierarchical_indexing": "off" }, "matches_canonical": false, "source_type": "official_paper", "audit_status": "verified", "notes": "Table 16: TerminalBench 2.0 (acc); Terminal-Bench 2.0 accuracy; harness unspecified. ", "candidates": [ { "score": 32.58, "reference_url": "https://arxiv.org/html/2608.01662v2", "source_type": "official_paper", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "benchmark-specified", "sampling": "Terminal-Bench 2.0 accuracy; harness unspecified", "judge": "benchmark-specified", "harness": "official LongCat evaluation; details not disclosed", "prompt_style": "official default", "temperature": "official model default", "context": "benchmark-specified", "hierarchical_indexing": "on" }, "notes": "Table 16: TerminalBench 2.0 (acc); Terminal-Bench 2.0 accuracy; harness unspecified. " } ] }, { "model_id": "longcat-flash-lite-sparse", "benchmark_id": "vitabench", "score": 21.67, "reference_url": "https://arxiv.org/html/2608.01662v2", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "benchmark-specified", "sampling": "avg@4", "judge": "benchmark-specified", "harness": "official LongCat evaluation; details not disclosed", "prompt_style": "official default", "temperature": "official model default", "context": "benchmark-specified", "hierarchical_indexing": "off" }, "matches_canonical": false, "source_type": "official_paper", "audit_status": "verified", "notes": "Table 16: VitaBench (avg@4); avg@4. ", "candidates": [ { "score": 20.42, "reference_url": "https://arxiv.org/html/2608.01662v2", "source_type": "official_paper", "reported_setting": { "mode": "non-thinking", "effort": "n/a", "tools": "benchmark-specified", "sampling": "avg@4", "judge": "benchmark-specified", "harness": "official LongCat evaluation; details not disclosed", "prompt_style": "official default", "temperature": "official model default", "context": "benchmark-specified", "hierarchical_indexing": "on" }, "notes": "Table 16: VitaBench (avg@4); avg@4. " } ] }, { "model_id": "claude-haiku-4.5", "benchmark_id": "chartqapro", "score": 52.87, "reference_url": "https://huggingface.co/internlm/Intern-S2-Preview/resolve/3b53ea8d2f8ed46abefa507e3f6f56ee0c662c4a/figs/performance.png", "reported_setting": { "table_framework_claim": "OpenCompass and VLMEvalKit used to evaluate all models", "exact_per_cell_settings": "not stated" }, "matches_canonical": false, "source_type": "official_model_card", "audit_status": "verified", "notes": "Displayed exactly: 52.87" }, { "model_id": "claude-haiku-4.5", "benchmark_id": "frontier_science_research", "score": 17.78, "reference_url": "https://huggingface.co/internlm/Intern-S2-Preview/resolve/3b53ea8d2f8ed46abefa507e3f6f56ee0c662c4a/figs/performance.png", "reported_setting": { "table_framework_claim": "OpenCompass and VLMEvalKit used to evaluate all models", "exact_per_cell_settings": "not stated" }, "matches_canonical": false, "source_type": "official_model_card", "audit_status": "verified", "notes": "Displayed exactly: 17.78" }, { "model_id": "claude-haiku-4.5", "benchmark_id": "frontiersci_olympiad", "score": 31.67, "reference_url": "https://huggingface.co/internlm/Intern-S2-Preview/resolve/3b53ea8d2f8ed46abefa507e3f6f56ee0c662c4a/figs/performance.png", "reported_setting": { "table_framework_claim": "OpenCompass and VLMEvalKit used to evaluate all models", "exact_per_cell_settings": "not stated" }, "matches_canonical": false, "source_type": "official_model_card", "audit_status": "verified", "notes": "Displayed exactly: 31.67" }, { "model_id": "claude-haiku-4.5", "benchmark_id": "hle_multimodal", "score": 6.85, "reference_url": "https://huggingface.co/internlm/Intern-S2-Preview/resolve/3b53ea8d2f8ed46abefa507e3f6f56ee0c662c4a/figs/performance.png", "reported_setting": { "table_framework_claim": "OpenCompass and VLMEvalKit used to evaluate all models", "exact_per_cell_settings": "not stated" }, "matches_canonical": false, "source_type": "official_model_card", "audit_status": "verified", "notes": "Displayed exactly: 6.85" }, { "model_id": "claude-haiku-4.5", "benchmark_id": "hle_text", "score": 8.26, "reference_url": "https://huggingface.co/internlm/Intern-S2-Preview/resolve/3b53ea8d2f8ed46abefa507e3f6f56ee0c662c4a/figs/performance.png", "reported_setting": { "table_framework_claim": "OpenCompass and VLMEvalKit used to evaluate all models", "exact_per_cell_settings": "not stated" }, "matches_canonical": false, "source_type": "official_model_card", "audit_status": "verified", "notes": "Displayed exactly: 8.26", "candidates": [ { "score": 9.5, "reference_url": "https://microsoft.ai/pdf/MAI-Code-1-Flash-Model-Card.PDF", "source_type": "official_model_card", "reported_setting": { "mode": "adaptive reasoning", "effort": "adaptive solution length control", "tools": "none", "sampling": "source does not state trial count", "judge": "benchmark-specific", "harness": "Microsoft MAI-Code-1-Flash comparative evaluation", "prompt_style": "source does not state", "temperature": "source does not state", "context": "256K model context; per-task usage reported separately", "notes": "Accuracy column from the official model-card/release table; benchmark=hle_text." }, "notes": "Displayed exactly as 9.5. Official Microsoft-reported result." } ] }, { "model_id": "claude-haiku-4.5", "benchmark_id": "ifbench", "score": 49.92, "reference_url": "https://huggingface.co/internlm/Intern-S2-Preview/resolve/3b53ea8d2f8ed46abefa507e3f6f56ee0c662c4a/figs/performance.png", "reported_setting": { "table_framework_claim": "OpenCompass and VLMEvalKit used to evaluate all models", "exact_per_cell_settings": "not stated" }, "matches_canonical": false, "source_type": "official_model_card", "audit_status": "verified", "notes": "Displayed exactly: 49.92" }, { "model_id": "claude-haiku-4.5", "benchmark_id": "imo_answerbench", "score": 60.75, "reference_url": "https://huggingface.co/internlm/Intern-S2-Preview/resolve/3b53ea8d2f8ed46abefa507e3f6f56ee0c662c4a/figs/performance.png", "reported_setting": { "table_framework_claim": "OpenCompass and VLMEvalKit used to evaluate all models", "exact_per_cell_settings": "not stated" }, "matches_canonical": false, "source_type": "official_model_card", "audit_status": "verified", "notes": "Displayed exactly: 60.75" }, { "model_id": "claude-haiku-4.5", "benchmark_id": "mathvision", "score": 65.79, "reference_url": "https://huggingface.co/internlm/Intern-S2-Preview/resolve/3b53ea8d2f8ed46abefa507e3f6f56ee0c662c4a/figs/performance.png", "reported_setting": { "table_framework_claim": "OpenCompass and VLMEvalKit used to evaluate all models", "exact_per_cell_settings": "not stated" }, "matches_canonical": false, "source_type": "official_model_card", "audit_status": "verified", "notes": "Displayed exactly: 65.79" }, { "model_id": "claude-haiku-4.5", "benchmark_id": "microvqa", "score": 53.17, "reference_url": "https://huggingface.co/internlm/Intern-S2-Preview/resolve/3b53ea8d2f8ed46abefa507e3f6f56ee0c662c4a/figs/performance.png", "reported_setting": { "table_framework_claim": "OpenCompass and VLMEvalKit used to evaluate all models", "exact_per_cell_settings": "not stated" }, "matches_canonical": false, "source_type": "official_model_card", "audit_status": "verified", "notes": "Displayed exactly: 53.17" }, { "model_id": "claude-haiku-4.5", "benchmark_id": "mmmu_pro", "score": 63.12, "reference_url": "https://huggingface.co/internlm/Intern-S2-Preview/resolve/3b53ea8d2f8ed46abefa507e3f6f56ee0c662c4a/figs/performance.png", "reported_setting": { "table_framework_claim": "OpenCompass and VLMEvalKit used to evaluate all models", "exact_per_cell_settings": "not stated" }, "matches_canonical": false, "source_type": "official_model_card", "audit_status": "verified", "notes": "Displayed exactly: 63.12" }, { "model_id": "claude-haiku-4.5", "benchmark_id": "moleculariq", "score": 31.74, "reference_url": "https://huggingface.co/internlm/Intern-S2-Preview/resolve/3b53ea8d2f8ed46abefa507e3f6f56ee0c662c4a/figs/performance.png", "reported_setting": { "table_framework_claim": "OpenCompass and VLMEvalKit used to evaluate all models", "exact_per_cell_settings": "not stated" }, "matches_canonical": false, "source_type": "official_model_card", "audit_status": "verified", "notes": "Displayed exactly: 31.74" }, { "model_id": "claude-haiku-4.5", "benchmark_id": "pinchbench", "score": 85.77, "reference_url": "https://huggingface.co/internlm/Intern-S2-Preview/resolve/3b53ea8d2f8ed46abefa507e3f6f56ee0c662c4a/figs/performance.png", "reported_setting": { "table_framework_claim": "OpenCompass and VLMEvalKit used to evaluate all models", "exact_per_cell_settings": "not stated" }, "matches_canonical": false, "source_type": "official_model_card", "audit_status": "verified", "notes": "Displayed exactly: 85.77" }, { "model_id": "claude-haiku-4.5", "benchmark_id": "scicode", "score": 40.48, "reference_url": "https://huggingface.co/internlm/Intern-S2-Preview/resolve/3b53ea8d2f8ed46abefa507e3f6f56ee0c662c4a/figs/performance.png", "reported_setting": { "table_framework_claim": "OpenCompass and VLMEvalKit used to evaluate all models", "exact_per_cell_settings": "not stated" }, "matches_canonical": false, "source_type": "official_model_card", "audit_status": "verified", "notes": "Displayed exactly: 40.48" }, { "model_id": "claude-haiku-4.5", "benchmark_id": "sfe", "score": 48.51, "reference_url": "https://huggingface.co/internlm/Intern-S2-Preview/resolve/3b53ea8d2f8ed46abefa507e3f6f56ee0c662c4a/figs/performance.png", "reported_setting": { "table_framework_claim": "OpenCompass and VLMEvalKit used to evaluate all models", "exact_per_cell_settings": "not stated" }, "matches_canonical": false, "source_type": "official_model_card", "audit_status": "verified", "notes": "Displayed exactly: 48.51" }, { "model_id": "claude-haiku-4.5", "benchmark_id": "sgi_bench", "score": 37.11, "reference_url": "https://huggingface.co/internlm/Intern-S2-Preview/resolve/3b53ea8d2f8ed46abefa507e3f6f56ee0c662c4a/figs/performance.png", "reported_setting": { "table_framework_claim": "OpenCompass and VLMEvalKit used to evaluate all models", "exact_per_cell_settings": "not stated" }, "matches_canonical": false, "source_type": "official_model_card", "audit_status": "verified", "notes": "Displayed exactly: 37.11" }, { "model_id": "claude-haiku-4.5", "benchmark_id": "simpleqa_verified", "score": 8.0, "reference_url": "https://huggingface.co/internlm/Intern-S2-Preview/resolve/3b53ea8d2f8ed46abefa507e3f6f56ee0c662c4a/figs/performance.png", "reported_setting": { "table_framework_claim": "OpenCompass and VLMEvalKit used to evaluate all models", "exact_per_cell_settings": "not stated" }, "matches_canonical": false, "source_type": "official_model_card", "audit_status": "verified", "notes": "Current HF/GitHub image says SimpleQA-Verified; original release image said SimpleQA with visually unchanged values.; Displayed exactly: 8.00" }, { "model_id": "gemini-3.1-flash-lite", "benchmark_id": "chartqapro", "score": 67.77, "reference_url": "https://huggingface.co/internlm/Intern-S2-Preview/resolve/3b53ea8d2f8ed46abefa507e3f6f56ee0c662c4a/figs/performance.png", "reported_setting": { "table_framework_claim": "OpenCompass and VLMEvalKit used to evaluate all models", "exact_per_cell_settings": "not stated" }, "matches_canonical": false, "source_type": "official_model_card", "audit_status": "verified", "notes": "Displayed exactly: 67.77" }, { "model_id": "gemini-3.1-flash-lite", "benchmark_id": "frontier_science_research", "score": 3.89, "reference_url": "https://huggingface.co/internlm/Intern-S2-Preview/resolve/3b53ea8d2f8ed46abefa507e3f6f56ee0c662c4a/figs/performance.png", "reported_setting": { "table_framework_claim": "OpenCompass and VLMEvalKit used to evaluate all models", "exact_per_cell_settings": "not stated" }, "matches_canonical": false, "source_type": "official_model_card", "audit_status": "verified", "notes": "Displayed exactly: 3.89" }, { "model_id": "gemini-3.1-flash-lite", "benchmark_id": "frontiersci_olympiad", "score": 45.41, "reference_url": "https://huggingface.co/internlm/Intern-S2-Preview/resolve/3b53ea8d2f8ed46abefa507e3f6f56ee0c662c4a/figs/performance.png", "reported_setting": { "table_framework_claim": "OpenCompass and VLMEvalKit used to evaluate all models", "exact_per_cell_settings": "not stated" }, "matches_canonical": false, "source_type": "official_model_card", "audit_status": "verified", "notes": "Displayed exactly: 45.41" }, { "model_id": "gemini-3.1-flash-lite", "benchmark_id": "hle_multimodal", "score": 14.74, "reference_url": "https://huggingface.co/internlm/Intern-S2-Preview/resolve/3b53ea8d2f8ed46abefa507e3f6f56ee0c662c4a/figs/performance.png", "reported_setting": { "table_framework_claim": "OpenCompass and VLMEvalKit used to evaluate all models", "exact_per_cell_settings": "not stated" }, "matches_canonical": false, "source_type": "official_model_card", "audit_status": "verified", "notes": "Displayed exactly: 14.74" }, { "model_id": "gemini-3.1-flash-lite", "benchmark_id": "hle_text", "score": 19.41, "reference_url": "https://huggingface.co/internlm/Intern-S2-Preview/resolve/3b53ea8d2f8ed46abefa507e3f6f56ee0c662c4a/figs/performance.png", "reported_setting": { "table_framework_claim": "OpenCompass and VLMEvalKit used to evaluate all models", "exact_per_cell_settings": "not stated" }, "matches_canonical": false, "source_type": "official_model_card", "audit_status": "verified", "notes": "Displayed exactly: 19.41" }, { "model_id": "gemini-3.1-flash-lite", "benchmark_id": "ifbench", "score": 76.15, "reference_url": "https://huggingface.co/internlm/Intern-S2-Preview/resolve/3b53ea8d2f8ed46abefa507e3f6f56ee0c662c4a/figs/performance.png", "reported_setting": { "table_framework_claim": "OpenCompass and VLMEvalKit used to evaluate all models", "exact_per_cell_settings": "not stated" }, "matches_canonical": false, "source_type": "official_model_card", "audit_status": "verified", "notes": "Displayed exactly: 76.15" }, { "model_id": "gemini-3.1-flash-lite", "benchmark_id": "imo_answerbench", "score": 64.5, "reference_url": "https://huggingface.co/internlm/Intern-S2-Preview/resolve/3b53ea8d2f8ed46abefa507e3f6f56ee0c662c4a/figs/performance.png", "reported_setting": { "table_framework_claim": "OpenCompass and VLMEvalKit used to evaluate all models", "exact_per_cell_settings": "not stated" }, "matches_canonical": false, "source_type": "official_model_card", "audit_status": "verified", "notes": "Displayed exactly: 64.50" }, { "model_id": "gemini-3.1-flash-lite", "benchmark_id": "mathvision", "score": 70.43, "reference_url": "https://huggingface.co/internlm/Intern-S2-Preview/resolve/3b53ea8d2f8ed46abefa507e3f6f56ee0c662c4a/figs/performance.png", "reported_setting": { "table_framework_claim": "OpenCompass and VLMEvalKit used to evaluate all models", "exact_per_cell_settings": "not stated" }, "matches_canonical": false, "source_type": "official_model_card", "audit_status": "verified", "notes": "Displayed exactly: 70.43" }, { "model_id": "gemini-3.1-flash-lite", "benchmark_id": "microvqa", "score": 63.34, "reference_url": "https://huggingface.co/internlm/Intern-S2-Preview/resolve/3b53ea8d2f8ed46abefa507e3f6f56ee0c662c4a/figs/performance.png", "reported_setting": { "table_framework_claim": "OpenCompass and VLMEvalKit used to evaluate all models", "exact_per_cell_settings": "not stated" }, "matches_canonical": false, "source_type": "official_model_card", "audit_status": "verified", "notes": "Displayed exactly: 63.34" }, { "model_id": "gemini-3.1-flash-lite", "benchmark_id": "mmlu_pro", "score": 87.31, "reference_url": "https://huggingface.co/internlm/Intern-S2-Preview/resolve/3b53ea8d2f8ed46abefa507e3f6f56ee0c662c4a/figs/performance.png", "reported_setting": { "table_framework_claim": "OpenCompass and VLMEvalKit used to evaluate all models", "exact_per_cell_settings": "not stated" }, "matches_canonical": false, "source_type": "official_model_card", "audit_status": "verified", "notes": "Displayed exactly: 87.31" }, { "model_id": "gemini-3.1-flash-lite", "benchmark_id": "mmmu_pro", "score": 76.07, "reference_url": "https://huggingface.co/internlm/Intern-S2-Preview/resolve/3b53ea8d2f8ed46abefa507e3f6f56ee0c662c4a/figs/performance.png", "reported_setting": { "table_framework_claim": "OpenCompass and VLMEvalKit used to evaluate all models", "exact_per_cell_settings": "not stated" }, "matches_canonical": false, "source_type": "official_model_card", "audit_status": "verified", "notes": "Displayed exactly: 76.07" }, { "model_id": "gemini-3.1-flash-lite", "benchmark_id": "moleculariq", "score": 46.25, "reference_url": "https://huggingface.co/internlm/Intern-S2-Preview/resolve/3b53ea8d2f8ed46abefa507e3f6f56ee0c662c4a/figs/performance.png", "reported_setting": { "table_framework_claim": "OpenCompass and VLMEvalKit used to evaluate all models", "exact_per_cell_settings": "not stated" }, "matches_canonical": false, "source_type": "official_model_card", "audit_status": "verified", "notes": "Displayed exactly: 46.25" }, { "model_id": "gemini-3.1-flash-lite", "benchmark_id": "pinchbench", "score": 84.77, "reference_url": "https://huggingface.co/internlm/Intern-S2-Preview/resolve/3b53ea8d2f8ed46abefa507e3f6f56ee0c662c4a/figs/performance.png", "reported_setting": { "table_framework_claim": "OpenCompass and VLMEvalKit used to evaluate all models", "exact_per_cell_settings": "not stated" }, "matches_canonical": false, "source_type": "official_model_card", "audit_status": "verified", "notes": "Displayed exactly: 84.77" }, { "model_id": "gemini-3.1-flash-lite", "benchmark_id": "scicode", "score": 41.12, "reference_url": "https://huggingface.co/internlm/Intern-S2-Preview/resolve/3b53ea8d2f8ed46abefa507e3f6f56ee0c662c4a/figs/performance.png", "reported_setting": { "table_framework_claim": "OpenCompass and VLMEvalKit used to evaluate all models", "exact_per_cell_settings": "not stated" }, "matches_canonical": false, "source_type": "official_model_card", "audit_status": "verified", "notes": "Displayed exactly: 41.12" }, { "model_id": "gemini-3.1-flash-lite", "benchmark_id": "sfe", "score": 52.03, "reference_url": "https://huggingface.co/internlm/Intern-S2-Preview/resolve/3b53ea8d2f8ed46abefa507e3f6f56ee0c662c4a/figs/performance.png", "reported_setting": { "table_framework_claim": "OpenCompass and VLMEvalKit used to evaluate all models", "exact_per_cell_settings": "not stated" }, "matches_canonical": false, "source_type": "official_model_card", "audit_status": "verified", "notes": "Displayed exactly: 52.03" }, { "model_id": "gemini-3.1-flash-lite", "benchmark_id": "sgi_bench", "score": 16.35, "reference_url": "https://huggingface.co/internlm/Intern-S2-Preview/resolve/3b53ea8d2f8ed46abefa507e3f6f56ee0c662c4a/figs/performance.png", "reported_setting": { "table_framework_claim": "OpenCompass and VLMEvalKit used to evaluate all models", "exact_per_cell_settings": "not stated" }, "matches_canonical": false, "source_type": "official_model_card", "audit_status": "verified", "notes": "Displayed exactly: 16.35" }, { "model_id": "gemini-3.1-flash-lite", "benchmark_id": "simpleqa_verified", "score": 42.5, "reference_url": "https://huggingface.co/internlm/Intern-S2-Preview/resolve/3b53ea8d2f8ed46abefa507e3f6f56ee0c662c4a/figs/performance.png", "reported_setting": { "table_framework_claim": "OpenCompass and VLMEvalKit used to evaluate all models", "exact_per_cell_settings": "not stated" }, "matches_canonical": false, "source_type": "official_model_card", "audit_status": "verified", "notes": "Current HF/GitHub image says SimpleQA-Verified; original release image said SimpleQA with visually unchanged values.; Displayed exactly: 42.50" }, { "model_id": "gemini-3.1-flash-lite", "benchmark_id": "swe_bench_verified", "score": 59.6, "reference_url": "https://huggingface.co/internlm/Intern-S2-Preview/resolve/3b53ea8d2f8ed46abefa507e3f6f56ee0c662c4a/figs/performance.png", "reported_setting": { "table_framework_claim": "OpenCompass/VLMEvalKit for all models conflicts with SWE-specific footnote", "single_value_harness": "not explicitly identified; likely source-team unified evaluation for cells without slash, but not asserted", "exact_per_cell_settings": "not stated" }, "matches_canonical": false, "source_type": "official_model_card", "audit_status": "verified", "notes": "Displayed exactly: 59.60" }, { "model_id": "gemma-4-26b-a4b", "benchmark_id": "chartqapro", "score": 44.26, "reference_url": "https://huggingface.co/internlm/Intern-S2-Preview/resolve/3b53ea8d2f8ed46abefa507e3f6f56ee0c662c4a/figs/performance.png", "reported_setting": { "table_framework_claim": "OpenCompass and VLMEvalKit used to evaluate all models", "exact_per_cell_settings": "not stated" }, "matches_canonical": false, "source_type": "official_model_card", "audit_status": "verified", "notes": "Displayed exactly: 44.26" }, { "model_id": "gemma-4-26b-a4b", "benchmark_id": "frontier_science_research", "score": 3.33, "reference_url": "https://huggingface.co/internlm/Intern-S2-Preview/resolve/3b53ea8d2f8ed46abefa507e3f6f56ee0c662c4a/figs/performance.png", "reported_setting": { "table_framework_claim": "OpenCompass and VLMEvalKit used to evaluate all models", "exact_per_cell_settings": "not stated" }, "matches_canonical": false, "source_type": "official_model_card", "audit_status": "verified", "notes": "Displayed exactly: 3.33" }, { "model_id": "gemma-4-26b-a4b", "benchmark_id": "frontiersci_olympiad", "score": 58.0, "reference_url": "https://huggingface.co/internlm/Intern-S2-Preview/resolve/3b53ea8d2f8ed46abefa507e3f6f56ee0c662c4a/figs/performance.png", "reported_setting": { "table_framework_claim": "OpenCompass and VLMEvalKit used to evaluate all models", "exact_per_cell_settings": "not stated" }, "matches_canonical": false, "source_type": "official_model_card", "audit_status": "verified", "notes": "Displayed exactly: 58.00" }, { "model_id": "gemma-4-26b-a4b", "benchmark_id": "hle_multimodal", "score": 9.37, "reference_url": "https://huggingface.co/internlm/Intern-S2-Preview/resolve/3b53ea8d2f8ed46abefa507e3f6f56ee0c662c4a/figs/performance.png", "reported_setting": { "table_framework_claim": "OpenCompass and VLMEvalKit used to evaluate all models", "exact_per_cell_settings": "not stated" }, "matches_canonical": false, "source_type": "official_model_card", "audit_status": "verified", "notes": "Displayed exactly: 9.37" }, { "model_id": "gemma-4-26b-a4b", "benchmark_id": "hle_text", "score": 8.7, "reference_url": "https://huggingface.co/internlm/Intern-S2-Preview/resolve/3b53ea8d2f8ed46abefa507e3f6f56ee0c662c4a/figs/performance.png", "reported_setting": { "table_framework_claim": "OpenCompass and VLMEvalKit used to evaluate all models", "exact_per_cell_settings": "not stated" }, "matches_canonical": false, "source_type": "official_model_card", "audit_status": "verified", "notes": "Displayed exactly: 8.70" }, { "model_id": "gemma-4-26b-a4b", "benchmark_id": "imo_answerbench", "score": 64.25, "reference_url": "https://huggingface.co/internlm/Intern-S2-Preview/resolve/3b53ea8d2f8ed46abefa507e3f6f56ee0c662c4a/figs/performance.png", "reported_setting": { "table_framework_claim": "OpenCompass and VLMEvalKit used to evaluate all models", "exact_per_cell_settings": "not stated" }, "matches_canonical": false, "source_type": "official_model_card", "audit_status": "verified", "notes": "Displayed exactly: 64.25" }, { "model_id": "gemma-4-26b-a4b", "benchmark_id": "microvqa", "score": 63.92, "reference_url": "https://huggingface.co/internlm/Intern-S2-Preview/resolve/3b53ea8d2f8ed46abefa507e3f6f56ee0c662c4a/figs/performance.png", "reported_setting": { "table_framework_claim": "OpenCompass and VLMEvalKit used to evaluate all models", "exact_per_cell_settings": "not stated" }, "matches_canonical": false, "source_type": "official_model_card", "audit_status": "verified", "notes": "Displayed exactly: 63.92" }, { "model_id": "gemma-4-26b-a4b", "benchmark_id": "moleculariq", "score": 32.97, "reference_url": "https://huggingface.co/internlm/Intern-S2-Preview/resolve/3b53ea8d2f8ed46abefa507e3f6f56ee0c662c4a/figs/performance.png", "reported_setting": { "table_framework_claim": "OpenCompass and VLMEvalKit used to evaluate all models", "exact_per_cell_settings": "not stated" }, "matches_canonical": false, "source_type": "official_model_card", "audit_status": "verified", "notes": "Displayed exactly: 32.97" }, { "model_id": "gemma-4-26b-a4b", "benchmark_id": "pinchbench", "score": 84.13, "reference_url": "https://huggingface.co/internlm/Intern-S2-Preview/resolve/3b53ea8d2f8ed46abefa507e3f6f56ee0c662c4a/figs/performance.png", "reported_setting": { "table_framework_claim": "OpenCompass and VLMEvalKit used to evaluate all models", "exact_per_cell_settings": "not stated" }, "matches_canonical": false, "source_type": "official_model_card", "audit_status": "verified", "notes": "Displayed exactly: 84.13" }, { "model_id": "gemma-4-26b-a4b", "benchmark_id": "sfe", "score": 63.09, "reference_url": "https://huggingface.co/internlm/Intern-S2-Preview/resolve/3b53ea8d2f8ed46abefa507e3f6f56ee0c662c4a/figs/performance.png", "reported_setting": { "table_framework_claim": "OpenCompass and VLMEvalKit used to evaluate all models", "exact_per_cell_settings": "not stated" }, "matches_canonical": false, "source_type": "official_model_card", "audit_status": "verified", "notes": "Displayed exactly: 63.09" }, { "model_id": "gemma-4-26b-a4b", "benchmark_id": "sgi_bench", "score": 26.1, "reference_url": "https://huggingface.co/internlm/Intern-S2-Preview/resolve/3b53ea8d2f8ed46abefa507e3f6f56ee0c662c4a/figs/performance.png", "reported_setting": { "table_framework_claim": "OpenCompass and VLMEvalKit used to evaluate all models", "exact_per_cell_settings": "not stated" }, "matches_canonical": false, "source_type": "official_model_card", "audit_status": "verified", "notes": "Displayed exactly: 26.10" }, { "model_id": "gemma-4-26b-a4b", "benchmark_id": "simpleqa_verified", "score": 9.0, "reference_url": "https://huggingface.co/internlm/Intern-S2-Preview/resolve/3b53ea8d2f8ed46abefa507e3f6f56ee0c662c4a/figs/performance.png", "reported_setting": { "table_framework_claim": "OpenCompass and VLMEvalKit used to evaluate all models", "exact_per_cell_settings": "not stated" }, "matches_canonical": false, "source_type": "official_model_card", "audit_status": "verified", "notes": "Current HF/GitHub image says SimpleQA-Verified; original release image said SimpleQA with visually unchanged values.; Displayed exactly: 9.00" }, { "model_id": "gemma-4-26b-a4b", "benchmark_id": "swe_bench_verified", "score": 49.8, "reference_url": "https://huggingface.co/internlm/Intern-S2-Preview/resolve/3b53ea8d2f8ed46abefa507e3f6f56ee0c662c4a/figs/performance.png", "reported_setting": { "table_framework_claim": "OpenCompass/VLMEvalKit for all models conflicts with SWE-specific footnote", "single_value_harness": "not explicitly identified; likely source-team unified evaluation for cells without slash, but not asserted", "exact_per_cell_settings": "not stated" }, "matches_canonical": false, "source_type": "official_model_card", "audit_status": "verified", "notes": "Displayed exactly: 49.80" }, { "model_id": "gpt-5.4-nano", "benchmark_id": "chartqapro", "score": 57.5, "reference_url": "https://huggingface.co/internlm/Intern-S2-Preview/resolve/3b53ea8d2f8ed46abefa507e3f6f56ee0c662c4a/figs/performance.png", "reported_setting": { "table_framework_claim": "OpenCompass and VLMEvalKit used to evaluate all models", "exact_per_cell_settings": "not stated" }, "matches_canonical": false, "source_type": "official_model_card", "audit_status": "verified", "notes": "Displayed exactly: 57.50" }, { "model_id": "gpt-5.4-nano", "benchmark_id": "frontier_science_research", "score": 5.56, "reference_url": "https://huggingface.co/internlm/Intern-S2-Preview/resolve/3b53ea8d2f8ed46abefa507e3f6f56ee0c662c4a/figs/performance.png", "reported_setting": { "table_framework_claim": "OpenCompass and VLMEvalKit used to evaluate all models", "exact_per_cell_settings": "not stated" }, "matches_canonical": false, "source_type": "official_model_card", "audit_status": "verified", "notes": "Displayed exactly: 5.56" }, { "model_id": "gpt-5.4-nano", "benchmark_id": "frontiersci_olympiad", "score": 20.13, "reference_url": "https://huggingface.co/internlm/Intern-S2-Preview/resolve/3b53ea8d2f8ed46abefa507e3f6f56ee0c662c4a/figs/performance.png", "reported_setting": { "table_framework_claim": "OpenCompass and VLMEvalKit used to evaluate all models", "exact_per_cell_settings": "not stated" }, "matches_canonical": false, "source_type": "official_model_card", "audit_status": "verified", "notes": "Displayed exactly: 20.13" }, { "model_id": "gpt-5.4-nano", "benchmark_id": "hle_multimodal", "score": 15.15, "reference_url": "https://huggingface.co/internlm/Intern-S2-Preview/resolve/3b53ea8d2f8ed46abefa507e3f6f56ee0c662c4a/figs/performance.png", "reported_setting": { "table_framework_claim": "OpenCompass and VLMEvalKit used to evaluate all models", "exact_per_cell_settings": "not stated" }, "matches_canonical": false, "source_type": "official_model_card", "audit_status": "verified", "notes": "Displayed exactly: 15.15" }, { "model_id": "gpt-5.4-nano", "benchmark_id": "hle_text", "score": 19.87, "reference_url": "https://huggingface.co/internlm/Intern-S2-Preview/resolve/3b53ea8d2f8ed46abefa507e3f6f56ee0c662c4a/figs/performance.png", "reported_setting": { "table_framework_claim": "OpenCompass and VLMEvalKit used to evaluate all models", "exact_per_cell_settings": "not stated" }, "matches_canonical": false, "source_type": "official_model_card", "audit_status": "verified", "notes": "Displayed exactly: 19.87" }, { "model_id": "gpt-5.4-nano", "benchmark_id": "ifbench", "score": 66.21, "reference_url": "https://huggingface.co/internlm/Intern-S2-Preview/resolve/3b53ea8d2f8ed46abefa507e3f6f56ee0c662c4a/figs/performance.png", "reported_setting": { "table_framework_claim": "OpenCompass and VLMEvalKit used to evaluate all models", "exact_per_cell_settings": "not stated" }, "matches_canonical": false, "source_type": "official_model_card", "audit_status": "verified", "notes": "Displayed exactly: 66.21" }, { "model_id": "gpt-5.4-nano", "benchmark_id": "imo_answerbench", "score": 73.5, "reference_url": "https://huggingface.co/internlm/Intern-S2-Preview/resolve/3b53ea8d2f8ed46abefa507e3f6f56ee0c662c4a/figs/performance.png", "reported_setting": { "table_framework_claim": "OpenCompass and VLMEvalKit used to evaluate all models", "exact_per_cell_settings": "not stated" }, "matches_canonical": false, "source_type": "official_model_card", "audit_status": "verified", "notes": "Displayed exactly: 73.50" }, { "model_id": "gpt-5.4-nano", "benchmark_id": "mathvision", "score": 71.91, "reference_url": "https://huggingface.co/internlm/Intern-S2-Preview/resolve/3b53ea8d2f8ed46abefa507e3f6f56ee0c662c4a/figs/performance.png", "reported_setting": { "table_framework_claim": "OpenCompass and VLMEvalKit used to evaluate all models", "exact_per_cell_settings": "not stated" }, "matches_canonical": false, "source_type": "official_model_card", "audit_status": "verified", "notes": "Displayed exactly: 71.91" }, { "model_id": "gpt-5.4-nano", "benchmark_id": "microvqa", "score": 51.73, "reference_url": "https://huggingface.co/internlm/Intern-S2-Preview/resolve/3b53ea8d2f8ed46abefa507e3f6f56ee0c662c4a/figs/performance.png", "reported_setting": { "table_framework_claim": "OpenCompass and VLMEvalKit used to evaluate all models", "exact_per_cell_settings": "not stated" }, "matches_canonical": false, "source_type": "official_model_card", "audit_status": "verified", "notes": "Displayed exactly: 51.73" }, { "model_id": "gpt-5.4-nano", "benchmark_id": "mmlu_pro", "score": 78.91, "reference_url": "https://huggingface.co/internlm/Intern-S2-Preview/resolve/3b53ea8d2f8ed46abefa507e3f6f56ee0c662c4a/figs/performance.png", "reported_setting": { "table_framework_claim": "OpenCompass and VLMEvalKit used to evaluate all models", "exact_per_cell_settings": "not stated" }, "matches_canonical": false, "source_type": "official_model_card", "audit_status": "verified", "notes": "Displayed exactly: 78.91" }, { "model_id": "gpt-5.4-nano", "benchmark_id": "mmmu_pro", "score": 63.41, "reference_url": "https://huggingface.co/internlm/Intern-S2-Preview/resolve/3b53ea8d2f8ed46abefa507e3f6f56ee0c662c4a/figs/performance.png", "reported_setting": { "table_framework_claim": "OpenCompass and VLMEvalKit used to evaluate all models", "exact_per_cell_settings": "not stated" }, "matches_canonical": false, "source_type": "official_model_card", "audit_status": "verified", "notes": "Displayed exactly: 63.41" }, { "model_id": "gpt-5.4-nano", "benchmark_id": "moleculariq", "score": 45.72, "reference_url": "https://huggingface.co/internlm/Intern-S2-Preview/resolve/3b53ea8d2f8ed46abefa507e3f6f56ee0c662c4a/figs/performance.png", "reported_setting": { "table_framework_claim": "OpenCompass and VLMEvalKit used to evaluate all models", "exact_per_cell_settings": "not stated" }, "matches_canonical": false, "source_type": "official_model_card", "audit_status": "verified", "notes": "Displayed exactly: 45.72" }, { "model_id": "gpt-5.4-nano", "benchmark_id": "pinchbench", "score": 79.37, "reference_url": "https://huggingface.co/internlm/Intern-S2-Preview/resolve/3b53ea8d2f8ed46abefa507e3f6f56ee0c662c4a/figs/performance.png", "reported_setting": { "table_framework_claim": "OpenCompass and VLMEvalKit used to evaluate all models", "exact_per_cell_settings": "not stated" }, "matches_canonical": false, "source_type": "official_model_card", "audit_status": "verified", "notes": "Displayed exactly: 79.37" }, { "model_id": "gpt-5.4-nano", "benchmark_id": "scicode", "score": 35.5, "reference_url": "https://huggingface.co/internlm/Intern-S2-Preview/resolve/3b53ea8d2f8ed46abefa507e3f6f56ee0c662c4a/figs/performance.png", "reported_setting": { "table_framework_claim": "OpenCompass and VLMEvalKit used to evaluate all models", "exact_per_cell_settings": "not stated" }, "matches_canonical": false, "source_type": "official_model_card", "audit_status": "verified", "notes": "Displayed exactly: 35.50" }, { "model_id": "gpt-5.4-nano", "benchmark_id": "sfe", "score": 36.8, "reference_url": "https://huggingface.co/internlm/Intern-S2-Preview/resolve/3b53ea8d2f8ed46abefa507e3f6f56ee0c662c4a/figs/performance.png", "reported_setting": { "table_framework_claim": "OpenCompass and VLMEvalKit used to evaluate all models", "exact_per_cell_settings": "not stated" }, "matches_canonical": false, "source_type": "official_model_card", "audit_status": "verified", "notes": "Displayed exactly: 36.80" }, { "model_id": "gpt-5.4-nano", "benchmark_id": "sgi_bench", "score": 25.73, "reference_url": "https://huggingface.co/internlm/Intern-S2-Preview/resolve/3b53ea8d2f8ed46abefa507e3f6f56ee0c662c4a/figs/performance.png", "reported_setting": { "table_framework_claim": "OpenCompass and VLMEvalKit used to evaluate all models", "exact_per_cell_settings": "not stated" }, "matches_canonical": false, "source_type": "official_model_card", "audit_status": "verified", "notes": "Displayed exactly: 25.73" }, { "model_id": "gpt-5.4-nano", "benchmark_id": "simpleqa_verified", "score": 11.7, "reference_url": "https://huggingface.co/internlm/Intern-S2-Preview/resolve/3b53ea8d2f8ed46abefa507e3f6f56ee0c662c4a/figs/performance.png", "reported_setting": { "table_framework_claim": "OpenCompass and VLMEvalKit used to evaluate all models", "exact_per_cell_settings": "not stated" }, "matches_canonical": false, "source_type": "official_model_card", "audit_status": "verified", "notes": "Current HF/GitHub image says SimpleQA-Verified; original release image said SimpleQA with visually unchanged values.; Displayed exactly: 11.70" }, { "model_id": "gpt-5.4-nano", "benchmark_id": "swe_bench_verified", "score": 42.4, "reference_url": "https://huggingface.co/internlm/Intern-S2-Preview/resolve/3b53ea8d2f8ed46abefa507e3f6f56ee0c662c4a/figs/performance.png", "reported_setting": { "table_framework_claim": "OpenCompass/VLMEvalKit for all models conflicts with SWE-specific footnote", "single_value_harness": "not explicitly identified; likely source-team unified evaluation for cells without slash, but not asserted", "exact_per_cell_settings": "not stated" }, "matches_canonical": false, "source_type": "official_model_card", "audit_status": "verified", "notes": "Displayed exactly: 42.40" }, { "model_id": "intern-s1-pro", "benchmark_id": "chartqapro", "score": 64.66, "reference_url": "https://huggingface.co/internlm/Intern-S2-Preview/resolve/3b53ea8d2f8ed46abefa507e3f6f56ee0c662c4a/figs/performance.png", "reported_setting": { "table_framework_claim": "OpenCompass and VLMEvalKit used to evaluate all models", "exact_per_cell_settings": "not stated" }, "matches_canonical": false, "source_type": "official_model_card", "audit_status": "verified", "notes": "Displayed exactly: 64.66" }, { "model_id": "intern-s1-pro", "benchmark_id": "frontier_science_research", "score": 12.22, "reference_url": "https://huggingface.co/internlm/Intern-S2-Preview/resolve/3b53ea8d2f8ed46abefa507e3f6f56ee0c662c4a/figs/performance.png", "reported_setting": { "table_framework_claim": "OpenCompass and VLMEvalKit used to evaluate all models", "exact_per_cell_settings": "not stated" }, "matches_canonical": false, "source_type": "official_model_card", "audit_status": "verified", "notes": "Displayed exactly: 12.22" }, { "model_id": "intern-s1-pro", "benchmark_id": "frontiersci_olympiad", "score": 53.33, "reference_url": "https://huggingface.co/internlm/Intern-S2-Preview/resolve/3b53ea8d2f8ed46abefa507e3f6f56ee0c662c4a/figs/performance.png", "reported_setting": { "table_framework_claim": "OpenCompass and VLMEvalKit used to evaluate all models", "exact_per_cell_settings": "not stated" }, "matches_canonical": false, "source_type": "official_model_card", "audit_status": "verified", "notes": "Displayed exactly: 53.33" }, { "model_id": "intern-s1-pro", "benchmark_id": "hle_multimodal", "score": 10.63, "reference_url": "https://huggingface.co/internlm/Intern-S2-Preview/resolve/3b53ea8d2f8ed46abefa507e3f6f56ee0c662c4a/figs/performance.png", "reported_setting": { "table_framework_claim": "OpenCompass and VLMEvalKit used to evaluate all models", "exact_per_cell_settings": "not stated" }, "matches_canonical": false, "source_type": "official_model_card", "audit_status": "verified", "notes": "Displayed exactly: 10.63" }, { "model_id": "intern-s1-pro", "benchmark_id": "hle_text", "score": 19.2, "reference_url": "https://huggingface.co/internlm/Intern-S2-Preview/resolve/3b53ea8d2f8ed46abefa507e3f6f56ee0c662c4a/figs/performance.png", "reported_setting": { "table_framework_claim": "OpenCompass and VLMEvalKit used to evaluate all models", "exact_per_cell_settings": "not stated" }, "matches_canonical": false, "source_type": "official_model_card", "audit_status": "verified", "notes": "Displayed exactly: 19.20" }, { "model_id": "intern-s1-pro", "benchmark_id": "ifbench", "score": 69.0, "reference_url": "https://huggingface.co/internlm/Intern-S2-Preview/resolve/3b53ea8d2f8ed46abefa507e3f6f56ee0c662c4a/figs/performance.png", "reported_setting": { "table_framework_claim": "OpenCompass and VLMEvalKit used to evaluate all models", "exact_per_cell_settings": "not stated" }, "matches_canonical": false, "source_type": "official_model_card", "audit_status": "verified", "notes": "Displayed exactly: 69.00" }, { "model_id": "intern-s1-pro", "benchmark_id": "imo_answerbench", "score": 77.25, "reference_url": "https://huggingface.co/internlm/Intern-S2-Preview/resolve/3b53ea8d2f8ed46abefa507e3f6f56ee0c662c4a/figs/performance.png", "reported_setting": { "table_framework_claim": "OpenCompass and VLMEvalKit used to evaluate all models", "exact_per_cell_settings": "not stated" }, "matches_canonical": false, "source_type": "official_model_card", "audit_status": "verified", "notes": "Displayed exactly: 77.25" }, { "model_id": "intern-s1-pro", "benchmark_id": "mathvision", "score": 72.43, "reference_url": "https://huggingface.co/internlm/Intern-S2-Preview/resolve/3b53ea8d2f8ed46abefa507e3f6f56ee0c662c4a/figs/performance.png", "reported_setting": { "table_framework_claim": "OpenCompass and VLMEvalKit used to evaluate all models", "exact_per_cell_settings": "not stated" }, "matches_canonical": false, "source_type": "official_model_card", "audit_status": "verified", "notes": "Displayed exactly: 72.43" }, { "model_id": "intern-s1-pro", "benchmark_id": "microvqa", "score": 63.34, "reference_url": "https://huggingface.co/internlm/Intern-S2-Preview/resolve/3b53ea8d2f8ed46abefa507e3f6f56ee0c662c4a/figs/performance.png", "reported_setting": { "table_framework_claim": "OpenCompass and VLMEvalKit used to evaluate all models", "exact_per_cell_settings": "not stated" }, "matches_canonical": false, "source_type": "official_model_card", "audit_status": "verified", "notes": "Displayed exactly: 63.34" }, { "model_id": "intern-s1-pro", "benchmark_id": "mmlu_pro", "score": 86.59, "reference_url": "https://huggingface.co/internlm/Intern-S2-Preview/resolve/3b53ea8d2f8ed46abefa507e3f6f56ee0c662c4a/figs/performance.png", "reported_setting": { "table_framework_claim": "OpenCompass and VLMEvalKit used to evaluate all models", "exact_per_cell_settings": "not stated" }, "matches_canonical": false, "source_type": "official_model_card", "audit_status": "verified", "notes": "Displayed exactly: 86.59" }, { "model_id": "intern-s1-pro", "benchmark_id": "mmmu_pro", "score": 72.77, "reference_url": "https://huggingface.co/internlm/Intern-S2-Preview/resolve/3b53ea8d2f8ed46abefa507e3f6f56ee0c662c4a/figs/performance.png", "reported_setting": { "table_framework_claim": "OpenCompass and VLMEvalKit used to evaluate all models", "exact_per_cell_settings": "not stated" }, "matches_canonical": false, "source_type": "official_model_card", "audit_status": "verified", "notes": "Displayed exactly: 72.77" }, { "model_id": "intern-s1-pro", "benchmark_id": "moleculariq", "score": 25.31, "reference_url": "https://huggingface.co/internlm/Intern-S2-Preview/resolve/3b53ea8d2f8ed46abefa507e3f6f56ee0c662c4a/figs/performance.png", "reported_setting": { "table_framework_claim": "OpenCompass and VLMEvalKit used to evaluate all models", "exact_per_cell_settings": "not stated" }, "matches_canonical": false, "source_type": "official_model_card", "audit_status": "verified", "notes": "Displayed exactly: 25.31" }, { "model_id": "intern-s1-pro", "benchmark_id": "pinchbench", "score": 64.83, "reference_url": "https://huggingface.co/internlm/Intern-S2-Preview/resolve/3b53ea8d2f8ed46abefa507e3f6f56ee0c662c4a/figs/performance.png", "reported_setting": { "table_framework_claim": "OpenCompass and VLMEvalKit used to evaluate all models", "exact_per_cell_settings": "not stated" }, "matches_canonical": false, "source_type": "official_model_card", "audit_status": "verified", "notes": "Displayed exactly: 64.83" }, { "model_id": "intern-s1-pro", "benchmark_id": "scicode", "score": 42.74, "reference_url": "https://huggingface.co/internlm/Intern-S2-Preview/resolve/3b53ea8d2f8ed46abefa507e3f6f56ee0c662c4a/figs/performance.png", "reported_setting": { "table_framework_claim": "OpenCompass and VLMEvalKit used to evaluate all models", "exact_per_cell_settings": "not stated" }, "matches_canonical": false, "source_type": "official_model_card", "audit_status": "verified", "notes": "Displayed exactly: 42.74" }, { "model_id": "intern-s1-pro", "benchmark_id": "sfe", "score": 52.66, "reference_url": "https://huggingface.co/internlm/Intern-S2-Preview/resolve/3b53ea8d2f8ed46abefa507e3f6f56ee0c662c4a/figs/performance.png", "reported_setting": { "table_framework_claim": "OpenCompass and VLMEvalKit used to evaluate all models", "exact_per_cell_settings": "not stated" }, "matches_canonical": false, "source_type": "official_model_card", "audit_status": "verified", "notes": "Displayed exactly: 52.66" }, { "model_id": "intern-s1-pro", "benchmark_id": "sgi_bench", "score": 38.36, "reference_url": "https://huggingface.co/internlm/Intern-S2-Preview/resolve/3b53ea8d2f8ed46abefa507e3f6f56ee0c662c4a/figs/performance.png", "reported_setting": { "table_framework_claim": "OpenCompass and VLMEvalKit used to evaluate all models", "exact_per_cell_settings": "not stated" }, "matches_canonical": false, "source_type": "official_model_card", "audit_status": "verified", "notes": "Displayed exactly: 38.36" }, { "model_id": "intern-s1-pro", "benchmark_id": "simpleqa_verified", "score": 41.9, "reference_url": "https://huggingface.co/internlm/Intern-S2-Preview/resolve/3b53ea8d2f8ed46abefa507e3f6f56ee0c662c4a/figs/performance.png", "reported_setting": { "table_framework_claim": "OpenCompass and VLMEvalKit used to evaluate all models", "exact_per_cell_settings": "not stated" }, "matches_canonical": false, "source_type": "official_model_card", "audit_status": "verified", "notes": "Current HF/GitHub image says SimpleQA-Verified; original release image said SimpleQA with visually unchanged values.; Displayed exactly: 41.90" }, { "model_id": "intern-s1-pro", "benchmark_id": "swe_bench_verified", "score": 38.4, "reference_url": "https://huggingface.co/internlm/Intern-S2-Preview/resolve/3b53ea8d2f8ed46abefa507e3f6f56ee0c662c4a/figs/performance.png", "reported_setting": { "table_framework_claim": "OpenCompass/VLMEvalKit for all models conflicts with SWE-specific footnote", "single_value_harness": "not explicitly identified; likely source-team unified evaluation for cells without slash, but not asserted", "exact_per_cell_settings": "not stated" }, "matches_canonical": false, "source_type": "official_model_card", "audit_status": "verified", "notes": "Displayed exactly: 38.40" }, { "model_id": "intern-s2-preview-35b", "benchmark_id": "chartqapro", "score": 68.9, "reference_url": "https://huggingface.co/internlm/Intern-S2-Preview/resolve/3b53ea8d2f8ed46abefa507e3f6f56ee0c662c4a/figs/performance.png", "reported_setting": { "table_framework_claim": "OpenCompass and VLMEvalKit used to evaluate all models", "exact_per_cell_settings": "not stated", "mode": "thinking enabled by default", "maximum_inference_length_tokens": 64000, "recommended_generation": { "top_p": 0.95, "top_k": 50, "min_p": 0.0, "temperature": 0.8 }, "recommendation_is_exact_eval_setting": false }, "matches_canonical": false, "source_type": "official_model_card", "audit_status": "verified", "notes": "Displayed exactly: 68.90" }, { "model_id": "intern-s2-preview-35b", "benchmark_id": "frontier_science_research", "score": 19.44, "reference_url": "https://huggingface.co/internlm/Intern-S2-Preview/resolve/3b53ea8d2f8ed46abefa507e3f6f56ee0c662c4a/figs/performance.png", "reported_setting": { "table_framework_claim": "OpenCompass and VLMEvalKit used to evaluate all models", "exact_per_cell_settings": "not stated", "mode": "thinking enabled by default", "maximum_inference_length_tokens": 128000, "recommended_generation": { "top_p": 0.95, "top_k": 50, "min_p": 0.0, "temperature": 0.8 }, "recommendation_is_exact_eval_setting": false }, "matches_canonical": false, "source_type": "official_model_card", "audit_status": "verified", "notes": "Displayed exactly: 19.44" }, { "model_id": "intern-s2-preview-35b", "benchmark_id": "frontiersci_olympiad", "score": 60.33, "reference_url": "https://huggingface.co/internlm/Intern-S2-Preview/resolve/3b53ea8d2f8ed46abefa507e3f6f56ee0c662c4a/figs/performance.png", "reported_setting": { "table_framework_claim": "OpenCompass and VLMEvalKit used to evaluate all models", "exact_per_cell_settings": "not stated", "mode": "thinking enabled by default", "maximum_inference_length_tokens": 128000, "recommended_generation": { "top_p": 0.95, "top_k": 50, "min_p": 0.0, "temperature": 0.8 }, "recommendation_is_exact_eval_setting": false }, "matches_canonical": false, "source_type": "official_model_card", "audit_status": "verified", "notes": "Displayed exactly: 60.33" }, { "model_id": "intern-s2-preview-35b", "benchmark_id": "hle_multimodal", "score": 18.07, "reference_url": "https://huggingface.co/internlm/Intern-S2-Preview/resolve/3b53ea8d2f8ed46abefa507e3f6f56ee0c662c4a/figs/performance.png", "reported_setting": { "table_framework_claim": "OpenCompass and VLMEvalKit used to evaluate all models", "exact_per_cell_settings": "not stated", "mode": "thinking enabled by default", "maximum_inference_length_tokens": 64000, "recommended_generation": { "top_p": 0.95, "top_k": 50, "min_p": 0.0, "temperature": 0.8 }, "recommendation_is_exact_eval_setting": false }, "matches_canonical": false, "source_type": "official_model_card", "audit_status": "verified", "notes": "Displayed exactly: 18.07" }, { "model_id": "intern-s2-preview-35b", "benchmark_id": "hle_text", "score": 21.94, "reference_url": "https://huggingface.co/internlm/Intern-S2-Preview/resolve/3b53ea8d2f8ed46abefa507e3f6f56ee0c662c4a/figs/performance.png", "reported_setting": { "table_framework_claim": "OpenCompass and VLMEvalKit used to evaluate all models", "exact_per_cell_settings": "not stated", "mode": "thinking enabled by default", "maximum_inference_length_tokens": 128000, "recommended_generation": { "top_p": 0.95, "top_k": 50, "min_p": 0.0, "temperature": 0.8 }, "recommendation_is_exact_eval_setting": false }, "matches_canonical": false, "source_type": "official_model_card", "audit_status": "verified", "notes": "Displayed exactly: 21.94" }, { "model_id": "intern-s2-preview-35b", "benchmark_id": "ifbench", "score": 80.87, "reference_url": "https://huggingface.co/internlm/Intern-S2-Preview/resolve/3b53ea8d2f8ed46abefa507e3f6f56ee0c662c4a/figs/performance.png", "reported_setting": { "table_framework_claim": "OpenCompass and VLMEvalKit used to evaluate all models", "exact_per_cell_settings": "not stated", "mode": "thinking enabled by default", "maximum_inference_length_tokens": 128000, "recommended_generation": { "top_p": 0.95, "top_k": 50, "min_p": 0.0, "temperature": 0.8 }, "recommendation_is_exact_eval_setting": false }, "matches_canonical": false, "source_type": "official_model_card", "audit_status": "verified", "notes": "Displayed exactly: 80.87" }, { "model_id": "intern-s2-preview-35b", "benchmark_id": "imo_answerbench", "score": 84.0, "reference_url": "https://huggingface.co/internlm/Intern-S2-Preview/resolve/3b53ea8d2f8ed46abefa507e3f6f56ee0c662c4a/figs/performance.png", "reported_setting": { "table_framework_claim": "OpenCompass and VLMEvalKit used to evaluate all models", "exact_per_cell_settings": "not stated", "mode": "thinking enabled by default", "maximum_inference_length_tokens": 128000, "recommended_generation": { "top_p": 0.95, "top_k": 50, "min_p": 0.0, "temperature": 0.8 }, "recommendation_is_exact_eval_setting": false }, "matches_canonical": false, "source_type": "official_model_card", "audit_status": "verified", "notes": "Displayed exactly: 84.00" }, { "model_id": "intern-s2-preview-35b", "benchmark_id": "mathvision", "score": 83.36, "reference_url": "https://huggingface.co/internlm/Intern-S2-Preview/resolve/3b53ea8d2f8ed46abefa507e3f6f56ee0c662c4a/figs/performance.png", "reported_setting": { "table_framework_claim": "OpenCompass and VLMEvalKit used to evaluate all models", "exact_per_cell_settings": "not stated", "mode": "thinking enabled by default", "maximum_inference_length_tokens": 64000, "recommended_generation": { "top_p": 0.95, "top_k": 50, "min_p": 0.0, "temperature": 0.8 }, "recommendation_is_exact_eval_setting": false }, "matches_canonical": false, "source_type": "official_model_card", "audit_status": "verified", "notes": "Displayed exactly: 83.36" }, { "model_id": "intern-s2-preview-35b", "benchmark_id": "microvqa", "score": 66.22, "reference_url": "https://huggingface.co/internlm/Intern-S2-Preview/resolve/3b53ea8d2f8ed46abefa507e3f6f56ee0c662c4a/figs/performance.png", "reported_setting": { "table_framework_claim": "OpenCompass and VLMEvalKit used to evaluate all models", "exact_per_cell_settings": "not stated", "mode": "thinking enabled by default", "maximum_inference_length_tokens": 64000, "recommended_generation": { "top_p": 0.95, "top_k": 50, "min_p": 0.0, "temperature": 0.8 }, "recommendation_is_exact_eval_setting": false }, "matches_canonical": false, "source_type": "official_model_card", "audit_status": "verified", "notes": "Displayed exactly: 66.22" }, { "model_id": "intern-s2-preview-35b", "benchmark_id": "mmlu_pro", "score": 88.0, "reference_url": "https://huggingface.co/internlm/Intern-S2-Preview/resolve/3b53ea8d2f8ed46abefa507e3f6f56ee0c662c4a/figs/performance.png", "reported_setting": { "table_framework_claim": "OpenCompass and VLMEvalKit used to evaluate all models", "exact_per_cell_settings": "not stated", "mode": "thinking enabled by default", "maximum_inference_length_tokens": 128000, "recommended_generation": { "top_p": 0.95, "top_k": 50, "min_p": 0.0, "temperature": 0.8 }, "recommendation_is_exact_eval_setting": false }, "matches_canonical": false, "source_type": "official_model_card", "audit_status": "verified", "notes": "Displayed exactly: 88.00" }, { "model_id": "intern-s2-preview-35b", "benchmark_id": "mmmu_pro", "score": 76.88, "reference_url": "https://huggingface.co/internlm/Intern-S2-Preview/resolve/3b53ea8d2f8ed46abefa507e3f6f56ee0c662c4a/figs/performance.png", "reported_setting": { "table_framework_claim": "OpenCompass and VLMEvalKit used to evaluate all models", "exact_per_cell_settings": "not stated", "mode": "thinking enabled by default", "maximum_inference_length_tokens": 64000, "recommended_generation": { "top_p": 0.95, "top_k": 50, "min_p": 0.0, "temperature": 0.8 }, "recommendation_is_exact_eval_setting": false }, "matches_canonical": false, "source_type": "official_model_card", "audit_status": "verified", "notes": "Displayed exactly: 76.88" }, { "model_id": "intern-s2-preview-35b", "benchmark_id": "moleculariq", "score": 57.26, "reference_url": "https://huggingface.co/internlm/Intern-S2-Preview/resolve/3b53ea8d2f8ed46abefa507e3f6f56ee0c662c4a/figs/performance.png", "reported_setting": { "table_framework_claim": "OpenCompass and VLMEvalKit used to evaluate all models", "exact_per_cell_settings": "not stated", "mode": "thinking enabled by default", "maximum_inference_length_tokens": 128000, "recommended_generation": { "top_p": 0.95, "top_k": 50, "min_p": 0.0, "temperature": 0.8 }, "recommendation_is_exact_eval_setting": false }, "matches_canonical": false, "source_type": "official_model_card", "audit_status": "verified", "notes": "Displayed exactly: 57.26" }, { "model_id": "intern-s2-preview-35b", "benchmark_id": "pinchbench", "score": 88.22, "reference_url": "https://huggingface.co/internlm/Intern-S2-Preview/resolve/3b53ea8d2f8ed46abefa507e3f6f56ee0c662c4a/figs/performance.png", "reported_setting": { "table_framework_claim": "OpenCompass and VLMEvalKit used to evaluate all models", "exact_per_cell_settings": "not stated", "mode": "thinking enabled by default", "maximum_inference_length_tokens": 128000, "recommended_generation": { "top_p": 0.95, "top_k": 50, "min_p": 0.0, "temperature": 0.8 }, "recommendation_is_exact_eval_setting": false }, "matches_canonical": false, "source_type": "official_model_card", "audit_status": "verified", "notes": "Displayed exactly: 88.22" }, { "model_id": "intern-s2-preview-35b", "benchmark_id": "scicode", "score": 39.64, "reference_url": "https://huggingface.co/internlm/Intern-S2-Preview/resolve/3b53ea8d2f8ed46abefa507e3f6f56ee0c662c4a/figs/performance.png", "reported_setting": { "table_framework_claim": "OpenCompass and VLMEvalKit used to evaluate all models", "exact_per_cell_settings": "not stated", "mode": "thinking enabled by default", "maximum_inference_length_tokens": 128000, "recommended_generation": { "top_p": 0.95, "top_k": 50, "min_p": 0.0, "temperature": 0.8 }, "recommendation_is_exact_eval_setting": false }, "matches_canonical": false, "source_type": "official_model_card", "audit_status": "verified", "notes": "Displayed exactly: 39.64" }, { "model_id": "intern-s2-preview-35b", "benchmark_id": "sfe", "score": 58.43, "reference_url": "https://huggingface.co/internlm/Intern-S2-Preview/resolve/3b53ea8d2f8ed46abefa507e3f6f56ee0c662c4a/figs/performance.png", "reported_setting": { "table_framework_claim": "OpenCompass and VLMEvalKit used to evaluate all models", "exact_per_cell_settings": "not stated", "mode": "thinking enabled by default", "maximum_inference_length_tokens": 64000, "recommended_generation": { "top_p": 0.95, "top_k": 50, "min_p": 0.0, "temperature": 0.8 }, "recommendation_is_exact_eval_setting": false }, "matches_canonical": false, "source_type": "official_model_card", "audit_status": "verified", "notes": "Displayed exactly: 58.43" }, { "model_id": "intern-s2-preview-35b", "benchmark_id": "sgi_bench", "score": 52.52, "reference_url": "https://huggingface.co/internlm/Intern-S2-Preview/resolve/3b53ea8d2f8ed46abefa507e3f6f56ee0c662c4a/figs/performance.png", "reported_setting": { "table_framework_claim": "OpenCompass and VLMEvalKit used to evaluate all models", "exact_per_cell_settings": "not stated", "mode": "thinking enabled by default", "maximum_inference_length_tokens": 64000, "recommended_generation": { "top_p": 0.95, "top_k": 50, "min_p": 0.0, "temperature": 0.8 }, "recommendation_is_exact_eval_setting": false }, "matches_canonical": false, "source_type": "official_model_card", "audit_status": "verified", "notes": "Displayed exactly: 52.52" }, { "model_id": "intern-s2-preview-35b", "benchmark_id": "simpleqa_verified", "score": 27.8, "reference_url": "https://huggingface.co/internlm/Intern-S2-Preview/resolve/3b53ea8d2f8ed46abefa507e3f6f56ee0c662c4a/figs/performance.png", "reported_setting": { "table_framework_claim": "OpenCompass and VLMEvalKit used to evaluate all models", "exact_per_cell_settings": "not stated", "mode": "thinking enabled by default", "maximum_inference_length_tokens": 128000, "recommended_generation": { "top_p": 0.95, "top_k": 50, "min_p": 0.0, "temperature": 0.8 }, "recommendation_is_exact_eval_setting": false }, "matches_canonical": false, "source_type": "official_model_card", "audit_status": "verified", "notes": "Current HF/GitHub image says SimpleQA-Verified; original release image said SimpleQA with visually unchanged values.; Displayed exactly: 27.80" }, { "model_id": "intern-s2-preview-35b", "benchmark_id": "swe_bench_verified", "score": 64.0, "reference_url": "https://huggingface.co/internlm/Intern-S2-Preview/resolve/3b53ea8d2f8ed46abefa507e3f6f56ee0c662c4a/figs/performance.png", "reported_setting": { "table_framework_claim": "OpenCompass/VLMEvalKit for all models conflicts with SWE-specific footnote", "single_value_harness": "not explicitly identified; likely source-team unified evaluation for cells without slash, but not asserted", "exact_per_cell_settings": "not stated" }, "matches_canonical": false, "source_type": "official_model_card", "audit_status": "verified", "notes": "Displayed exactly: 64.00" }, { "model_id": "qwen3.6-35b-a3b", "benchmark_id": "chartqapro", "score": 64.54, "reference_url": "https://huggingface.co/internlm/Intern-S2-Preview/resolve/3b53ea8d2f8ed46abefa507e3f6f56ee0c662c4a/figs/performance.png", "reported_setting": { "table_framework_claim": "OpenCompass and VLMEvalKit used to evaluate all models", "exact_per_cell_settings": "not stated" }, "matches_canonical": false, "source_type": "official_model_card", "audit_status": "verified", "notes": "Displayed exactly: 64.54" }, { "model_id": "qwen3.6-35b-a3b", "benchmark_id": "frontier_science_research", "score": 10.0, "reference_url": "https://huggingface.co/internlm/Intern-S2-Preview/resolve/3b53ea8d2f8ed46abefa507e3f6f56ee0c662c4a/figs/performance.png", "reported_setting": { "table_framework_claim": "OpenCompass and VLMEvalKit used to evaluate all models", "exact_per_cell_settings": "not stated" }, "matches_canonical": false, "source_type": "official_model_card", "audit_status": "verified", "notes": "Displayed exactly: 10.00" }, { "model_id": "qwen3.6-35b-a3b", "benchmark_id": "frontiersci_olympiad", "score": 65.0, "reference_url": "https://huggingface.co/internlm/Intern-S2-Preview/resolve/3b53ea8d2f8ed46abefa507e3f6f56ee0c662c4a/figs/performance.png", "reported_setting": { "table_framework_claim": "OpenCompass and VLMEvalKit used to evaluate all models", "exact_per_cell_settings": "not stated" }, "matches_canonical": false, "source_type": "official_model_card", "audit_status": "verified", "notes": "Displayed exactly: 65.00" }, { "model_id": "qwen3.6-35b-a3b", "benchmark_id": "hle_multimodal", "score": 17.04, "reference_url": "https://huggingface.co/internlm/Intern-S2-Preview/resolve/3b53ea8d2f8ed46abefa507e3f6f56ee0c662c4a/figs/performance.png", "reported_setting": { "table_framework_claim": "OpenCompass and VLMEvalKit used to evaluate all models", "exact_per_cell_settings": "not stated" }, "matches_canonical": false, "source_type": "official_model_card", "audit_status": "verified", "notes": "Displayed exactly: 17.04" }, { "model_id": "qwen3.6-35b-a3b", "benchmark_id": "hle_text", "score": 21.42, "reference_url": "https://huggingface.co/internlm/Intern-S2-Preview/resolve/3b53ea8d2f8ed46abefa507e3f6f56ee0c662c4a/figs/performance.png", "reported_setting": { "table_framework_claim": "OpenCompass and VLMEvalKit used to evaluate all models", "exact_per_cell_settings": "not stated" }, "matches_canonical": false, "source_type": "official_model_card", "audit_status": "verified", "notes": "Displayed exactly: 21.42" }, { "model_id": "qwen3.6-35b-a3b", "benchmark_id": "ifbench", "score": 63.52, "reference_url": "https://huggingface.co/internlm/Intern-S2-Preview/resolve/3b53ea8d2f8ed46abefa507e3f6f56ee0c662c4a/figs/performance.png", "reported_setting": { "table_framework_claim": "OpenCompass and VLMEvalKit used to evaluate all models", "exact_per_cell_settings": "not stated" }, "matches_canonical": false, "source_type": "official_model_card", "audit_status": "verified", "notes": "Displayed exactly: 63.52" }, { "model_id": "qwen3.6-35b-a3b", "benchmark_id": "imo_answerbench", "score": 81.0, "reference_url": "https://huggingface.co/internlm/Intern-S2-Preview/resolve/3b53ea8d2f8ed46abefa507e3f6f56ee0c662c4a/figs/performance.png", "reported_setting": { "table_framework_claim": "OpenCompass and VLMEvalKit used to evaluate all models", "exact_per_cell_settings": "not stated" }, "matches_canonical": false, "source_type": "official_model_card", "audit_status": "verified", "notes": "Displayed exactly: 81.00" }, { "model_id": "qwen3.6-35b-a3b", "benchmark_id": "mathvision", "score": 80.59, "reference_url": "https://huggingface.co/internlm/Intern-S2-Preview/resolve/3b53ea8d2f8ed46abefa507e3f6f56ee0c662c4a/figs/performance.png", "reported_setting": { "table_framework_claim": "OpenCompass and VLMEvalKit used to evaluate all models", "exact_per_cell_settings": "not stated" }, "matches_canonical": false, "source_type": "official_model_card", "audit_status": "verified", "notes": "Displayed exactly: 80.59" }, { "model_id": "qwen3.6-35b-a3b", "benchmark_id": "microvqa", "score": 63.63, "reference_url": "https://huggingface.co/internlm/Intern-S2-Preview/resolve/3b53ea8d2f8ed46abefa507e3f6f56ee0c662c4a/figs/performance.png", "reported_setting": { "table_framework_claim": "OpenCompass and VLMEvalKit used to evaluate all models", "exact_per_cell_settings": "not stated" }, "matches_canonical": false, "source_type": "official_model_card", "audit_status": "verified", "notes": "Displayed exactly: 63.63" }, { "model_id": "qwen3.6-35b-a3b", "benchmark_id": "mmlu_pro", "score": 85.12, "reference_url": "https://huggingface.co/internlm/Intern-S2-Preview/resolve/3b53ea8d2f8ed46abefa507e3f6f56ee0c662c4a/figs/performance.png", "reported_setting": { "table_framework_claim": "OpenCompass and VLMEvalKit used to evaluate all models", "exact_per_cell_settings": "not stated" }, "matches_canonical": false, "source_type": "official_model_card", "audit_status": "verified", "notes": "Displayed exactly: 85.12" }, { "model_id": "qwen3.6-35b-a3b", "benchmark_id": "mmmu_pro", "score": 74.28, "reference_url": "https://huggingface.co/internlm/Intern-S2-Preview/resolve/3b53ea8d2f8ed46abefa507e3f6f56ee0c662c4a/figs/performance.png", "reported_setting": { "table_framework_claim": "OpenCompass and VLMEvalKit used to evaluate all models", "exact_per_cell_settings": "not stated" }, "matches_canonical": false, "source_type": "official_model_card", "audit_status": "verified", "notes": "Displayed exactly: 74.28" }, { "model_id": "qwen3.6-35b-a3b", "benchmark_id": "moleculariq", "score": 32.62, "reference_url": "https://huggingface.co/internlm/Intern-S2-Preview/resolve/3b53ea8d2f8ed46abefa507e3f6f56ee0c662c4a/figs/performance.png", "reported_setting": { "table_framework_claim": "OpenCompass and VLMEvalKit used to evaluate all models", "exact_per_cell_settings": "not stated" }, "matches_canonical": false, "source_type": "official_model_card", "audit_status": "verified", "notes": "Displayed exactly: 32.62" }, { "model_id": "qwen3.6-35b-a3b", "benchmark_id": "pinchbench", "score": 87.05, "reference_url": "https://huggingface.co/internlm/Intern-S2-Preview/resolve/3b53ea8d2f8ed46abefa507e3f6f56ee0c662c4a/figs/performance.png", "reported_setting": { "table_framework_claim": "OpenCompass and VLMEvalKit used to evaluate all models", "exact_per_cell_settings": "not stated" }, "matches_canonical": false, "source_type": "official_model_card", "audit_status": "verified", "notes": "Displayed exactly: 87.05" }, { "model_id": "qwen3.6-35b-a3b", "benchmark_id": "scicode", "score": 40.6, "reference_url": "https://huggingface.co/internlm/Intern-S2-Preview/resolve/3b53ea8d2f8ed46abefa507e3f6f56ee0c662c4a/figs/performance.png", "reported_setting": { "table_framework_claim": "OpenCompass and VLMEvalKit used to evaluate all models", "exact_per_cell_settings": "not stated" }, "matches_canonical": false, "source_type": "official_model_card", "audit_status": "verified", "notes": "Displayed exactly: 40.60" }, { "model_id": "qwen3.6-35b-a3b", "benchmark_id": "sfe", "score": 51.31, "reference_url": "https://huggingface.co/internlm/Intern-S2-Preview/resolve/3b53ea8d2f8ed46abefa507e3f6f56ee0c662c4a/figs/performance.png", "reported_setting": { "table_framework_claim": "OpenCompass and VLMEvalKit used to evaluate all models", "exact_per_cell_settings": "not stated" }, "matches_canonical": false, "source_type": "official_model_card", "audit_status": "verified", "notes": "Displayed exactly: 51.31" }, { "model_id": "qwen3.6-35b-a3b", "benchmark_id": "sgi_bench", "score": 37.3, "reference_url": "https://huggingface.co/internlm/Intern-S2-Preview/resolve/3b53ea8d2f8ed46abefa507e3f6f56ee0c662c4a/figs/performance.png", "reported_setting": { "table_framework_claim": "OpenCompass and VLMEvalKit used to evaluate all models", "exact_per_cell_settings": "not stated" }, "matches_canonical": false, "source_type": "official_model_card", "audit_status": "verified", "notes": "Displayed exactly: 37.30" }, { "model_id": "qwen3.6-35b-a3b", "benchmark_id": "simpleqa_verified", "score": 20.8, "reference_url": "https://huggingface.co/internlm/Intern-S2-Preview/resolve/3b53ea8d2f8ed46abefa507e3f6f56ee0c662c4a/figs/performance.png", "reported_setting": { "table_framework_claim": "OpenCompass and VLMEvalKit used to evaluate all models", "exact_per_cell_settings": "not stated" }, "matches_canonical": false, "source_type": "official_model_card", "audit_status": "verified", "notes": "Current HF/GitHub image says SimpleQA-Verified; original release image said SimpleQA with visually unchanged values.; Displayed exactly: 20.80" }, { "model_id": "qwen3.6-35b-a3b", "benchmark_id": "swe_bench_verified", "score": 73.4, "reference_url": "https://huggingface.co/internlm/Intern-S2-Preview/resolve/3b53ea8d2f8ed46abefa507e3f6f56ee0c662c4a/figs/performance.png", "reported_setting": { "before_slash": "officially reported in model technical report", "after_slash": "AgentCompass unified framework with Mini-SWE-Agent", "footnote": "[2]" }, "matches_canonical": false, "source_type": "official_model_card", "audit_status": "verified", "notes": "Displayed exactly: 73.40 / 62.80[2]; Alternatives: [{\"footnote\": \"[2]\", \"role\": \"official_report\", \"value\": \"73.40\"}, {\"footnote\": \"[2]\", \"harness\": \"AgentCompass with Mini-SWE-Agent\", \"role\": \"unified_re_evaluation\", \"value\": \"62.80\"}]", "candidates": [ { "score": 62.8, "reference_url": "https://huggingface.co/internlm/Intern-S2-Preview/resolve/3b53ea8d2f8ed46abefa507e3f6f56ee0c662c4a/figs/performance.png", "source_type": "official_model_card", "reported_setting": { "role": "unified_re_evaluation", "harness": "AgentCompass with Mini-SWE-Agent", "footnote": "[2]" }, "notes": "Alternative reported in compound source cell; displayed cell: 73.40 / 62.80[2]" } ] }, { "model_id": "step-3.5-flash", "benchmark_id": "frontier_science_research", "score": 10.0, "reference_url": "https://huggingface.co/internlm/Intern-S2-Preview/resolve/3b53ea8d2f8ed46abefa507e3f6f56ee0c662c4a/figs/performance.png", "reported_setting": { "table_framework_claim": "OpenCompass and VLMEvalKit used to evaluate all models", "exact_per_cell_settings": "not stated" }, "matches_canonical": false, "source_type": "official_model_card", "audit_status": "verified", "notes": "Displayed exactly: 10.00" }, { "model_id": "step-3.5-flash", "benchmark_id": "frontiersci_olympiad", "score": 70.0, "reference_url": "https://huggingface.co/internlm/Intern-S2-Preview/resolve/3b53ea8d2f8ed46abefa507e3f6f56ee0c662c4a/figs/performance.png", "reported_setting": { "table_framework_claim": "OpenCompass and VLMEvalKit used to evaluate all models", "exact_per_cell_settings": "not stated" }, "matches_canonical": false, "source_type": "official_model_card", "audit_status": "verified", "notes": "Displayed exactly: 70.00" }, { "model_id": "step-3.5-flash", "benchmark_id": "hle_text", "score": 23.04, "reference_url": "https://huggingface.co/internlm/Intern-S2-Preview/resolve/3b53ea8d2f8ed46abefa507e3f6f56ee0c662c4a/figs/performance.png", "reported_setting": { "table_framework_claim": "OpenCompass and VLMEvalKit used to evaluate all models", "exact_per_cell_settings": "not stated" }, "matches_canonical": false, "source_type": "official_model_card", "audit_status": "verified", "notes": "Displayed exactly: 23.04" }, { "model_id": "step-3.5-flash", "benchmark_id": "ifbench", "score": 65.72, "reference_url": "https://huggingface.co/internlm/Intern-S2-Preview/resolve/3b53ea8d2f8ed46abefa507e3f6f56ee0c662c4a/figs/performance.png", "reported_setting": { "table_framework_claim": "OpenCompass and VLMEvalKit used to evaluate all models", "exact_per_cell_settings": "not stated" }, "matches_canonical": false, "source_type": "official_model_card", "audit_status": "verified", "notes": "Displayed exactly: 65.72" }, { "model_id": "step-3.5-flash", "benchmark_id": "imo_answerbench", "score": 79.0, "reference_url": "https://huggingface.co/internlm/Intern-S2-Preview/resolve/3b53ea8d2f8ed46abefa507e3f6f56ee0c662c4a/figs/performance.png", "reported_setting": { "harness": "OpenCompass for displayed 79.00 per footnote [1]", "official_report_sampling": "pass@1 averaged over 8 independent generations/problem", "official_report_value": "85.4", "source": "arXiv:2602.10604v2" }, "matches_canonical": false, "source_type": "official_model_card", "audit_status": "verified", "notes": "Displayed exactly: 79.00[1]; Alternatives: [{\"footnote\": \"[1]\", \"role\": \"displayed_OpenCompass_evaluation\", \"value\": \"79.00\"}, {\"footnote\": \"[1]\", \"role\": \"official_technical_report\", \"sampling\": \"average of 8 generations/problem\", \"source_url\": \"https://arxiv.org/pdf/2602.10604v2\", \"value\": \"85.4\"}]", "candidates": [ { "score": 85.4, "reference_url": "https://arxiv.org/pdf/2602.10604v2", "source_type": "official_paper", "reported_setting": { "role": "official_technical_report", "sampling": "average of 8 generations/problem", "footnote": "[1]" }, "notes": "Alternative reported in compound source cell; displayed cell: 79.00[1]" } ] }, { "model_id": "step-3.5-flash", "benchmark_id": "mmlu_pro", "score": 83.44, "reference_url": "https://huggingface.co/internlm/Intern-S2-Preview/resolve/3b53ea8d2f8ed46abefa507e3f6f56ee0c662c4a/figs/performance.png", "reported_setting": { "table_framework_claim": "OpenCompass and VLMEvalKit used to evaluate all models", "exact_per_cell_settings": "not stated" }, "matches_canonical": false, "source_type": "official_model_card", "audit_status": "verified", "notes": "Displayed exactly: 83.44" }, { "model_id": "step-3.5-flash", "benchmark_id": "moleculariq", "score": 45.94, "reference_url": "https://huggingface.co/internlm/Intern-S2-Preview/resolve/3b53ea8d2f8ed46abefa507e3f6f56ee0c662c4a/figs/performance.png", "reported_setting": { "table_framework_claim": "OpenCompass and VLMEvalKit used to evaluate all models", "exact_per_cell_settings": "not stated" }, "matches_canonical": false, "source_type": "official_model_card", "audit_status": "verified", "notes": "Displayed exactly: 45.94" }, { "model_id": "step-3.5-flash", "benchmark_id": "pinchbench", "score": 85.0, "reference_url": "https://huggingface.co/internlm/Intern-S2-Preview/resolve/3b53ea8d2f8ed46abefa507e3f6f56ee0c662c4a/figs/performance.png", "reported_setting": { "table_framework_claim": "OpenCompass and VLMEvalKit used to evaluate all models", "exact_per_cell_settings": "not stated" }, "matches_canonical": false, "source_type": "official_model_card", "audit_status": "verified", "notes": "Displayed exactly: 85.00" }, { "model_id": "step-3.5-flash", "benchmark_id": "scicode", "score": 46.15, "reference_url": "https://huggingface.co/internlm/Intern-S2-Preview/resolve/3b53ea8d2f8ed46abefa507e3f6f56ee0c662c4a/figs/performance.png", "reported_setting": { "table_framework_claim": "OpenCompass and VLMEvalKit used to evaluate all models", "exact_per_cell_settings": "not stated" }, "matches_canonical": false, "source_type": "official_model_card", "audit_status": "verified", "notes": "Displayed exactly: 46.15" }, { "model_id": "step-3.5-flash", "benchmark_id": "sgi_bench", "score": 36.16, "reference_url": "https://huggingface.co/internlm/Intern-S2-Preview/resolve/3b53ea8d2f8ed46abefa507e3f6f56ee0c662c4a/figs/performance.png", "reported_setting": { "table_framework_claim": "OpenCompass and VLMEvalKit used to evaluate all models", "exact_per_cell_settings": "not stated" }, "matches_canonical": false, "source_type": "official_model_card", "audit_status": "verified", "notes": "Displayed exactly: 36.16" }, { "model_id": "step-3.5-flash", "benchmark_id": "simpleqa_verified", "score": 22.1, "reference_url": "https://huggingface.co/internlm/Intern-S2-Preview/resolve/3b53ea8d2f8ed46abefa507e3f6f56ee0c662c4a/figs/performance.png", "reported_setting": { "table_framework_claim": "OpenCompass and VLMEvalKit used to evaluate all models", "exact_per_cell_settings": "not stated" }, "matches_canonical": false, "source_type": "official_model_card", "audit_status": "verified", "notes": "Current HF/GitHub image says SimpleQA-Verified; original release image said SimpleQA with visually unchanged values.; Displayed exactly: 22.10" }, { "model_id": "step-3.5-flash", "benchmark_id": "swe_bench_verified", "score": 74.4, "reference_url": "https://static.stepfun.com/blog/step-3.7-flash/", "reported_setting": { "quantization_config": "common release evaluation; artifact precision not established", "configuration": "base Step evaluation configuration", "mode": "source does not state", "effort": "source does not state", "tools_search": "agentic coding tools", "shots": "source does not state", "trials_samples": "source does not state", "aggregation_pass_k": "% resolved", "context": "source does not state", "output_cap": "source does not state", "temperature": "source does not state", "top_p": "source does not state", "judge": "benchmark-specified; exact StepFun judge not stated", "harness_agent": "source does not state coding scaffold", "dataset_version_split": "source label only", "multimodal_input": false }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "Official StepFun launch comparison table.", "candidates": [ { "score": 69.0, "reference_url": "https://huggingface.co/internlm/Intern-S2-Preview/resolve/3b53ea8d2f8ed46abefa507e3f6f56ee0c662c4a/figs/performance.png", "source_type": "official_model_card", "reported_setting": { "role": "unified_re_evaluation", "harness": "AgentCompass with Mini-SWE-Agent", "footnote": "[2]" }, "notes": "Alternative reported in compound source cell; displayed cell: 74.40 / 69.00[2]" }, { "score": 74.4, "reference_url": "https://huggingface.co/internlm/Intern-S2-Preview/resolve/3b53ea8d2f8ed46abefa507e3f6f56ee0c662c4a/figs/performance.png", "source_type": "official_model_card", "reported_setting": { "before_slash": "officially reported in model technical report", "after_slash": "AgentCompass unified framework with Mini-SWE-Agent", "footnote": "[2]", "official_report_details": { "harness": "StepFun internal session-router agent infrastructure", "runs": 4, "temperature": 1, "top_p": 0.95, "tool_execution_timeout_seconds": 1200, "memory_limit_GB": 4 } }, "notes": "Displayed exactly: 74.40 / 69.00[2]; Alternatives: [{\"footnote\": \"[2]\", \"role\": \"official_report\", \"value\": \"74.40\"}, {\"footnote\": \"[2]\", \"harness\": \"AgentCompass with Mini-SWE-Agent\", \"role\": \"unified_re_evaluation\", \"value\": \"69.00\"}]" } ] }, { "model_id": "claude-opus-4.5", "benchmark_id": "lmarena_search_elo", "score": 1181, "reference_url": "https://ernie.baidu.com/blog/posts/ernie-5.1-0508-release/lmarena-search-ranking.png", "reported_setting": { "leaderboard": "Arena AI Search", "metric": "Arena score", "style_control": "off", "tools": "search arena environment", "snapshot_date": "2026-05-09" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "May 9, 2026 Search Arena snapshot; style control off. Research observation obs-047. Displayed rank 15. Baidu chart footer cites ARENA.AI/LEADERBOARD/SEARCH." }, { "model_id": "claude-opus-4.6", "benchmark_id": "lmarena_search_elo", "score": 1255, "reference_url": "https://ernie.baidu.com/blog/posts/ernie-5.1-0508-release/lmarena-search-ranking.png", "reported_setting": { "leaderboard": "Arena AI Search", "metric": "Arena score", "style_control": "off", "tools": "search arena environment", "snapshot_date": "2026-05-09" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "May 9, 2026 Search Arena snapshot; style control off. Research observation obs-033. Displayed rank 1. Baidu chart footer cites ARENA.AI/LEADERBOARD/SEARCH." }, { "model_id": "claude-opus-4.6", "benchmark_id": "spreadsheetbench_verified", "score": 82.4, "reference_url": "https://ernie.baidu.com/blog/posts/ernie-5.1-0508-release/benchmark.png", "reported_setting": { "tools": "code execution/spreadsheet agent", "sampling": "pass@1" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Verified split, pass@1, with spreadsheet/code execution. Research observation obs-015." }, { "model_id": "claude-opus-4.7", "benchmark_id": "chatbot_arena_elo", "score": 1503, "reference_url": "https://ernie.baidu.com/blog/posts/ernie-5.1-0508-release/creative-capability.png", "reported_setting": { "leaderboard": "Arena AI Text", "metric": "Arena score", "style_control": "on", "mode": "thinking", "snapshot_date": "2026-04-30" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "April 30, 2026 Text Arena snapshot; style control on. Research observation obs-048. Displayed rank 1. Identical asset is published in four page placements. The same PNG bytes are published under two filenames and four English/Chinese page placements.", "candidates": [ { "score": 1494, "reference_url": "https://ernie.baidu.com/blog/posts/ernie-5.1-0508-release/creative-capability.png", "source_type": "official_blog", "reported_setting": { "leaderboard": "Arena AI Text", "metric": "Arena score", "style_control": "on", "mode": "unspecified", "snapshot_date": "2026-04-30" }, "notes": "April 30, 2026 Text Arena snapshot; style control on. Research observation obs-051. Displayed rank 4. Identical asset is published in four page placements. The same PNG bytes are published under two filenames and four English/Chinese page placements." } ] }, { "model_id": "claude-opus-4.7", "benchmark_id": "lmarena_search_elo", "score": 1236, "reference_url": "https://ernie.baidu.com/blog/posts/ernie-5.1-0508-release/lmarena-search-ranking.png", "reported_setting": { "leaderboard": "Arena AI Search", "metric": "Arena score", "style_control": "off", "mode": "unspecified", "tools": "search arena environment", "snapshot_date": "2026-05-09" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "May 9, 2026 Search Arena snapshot; style control off. Research observation obs-035. Displayed rank 3. Baidu chart footer cites ARENA.AI/LEADERBOARD/SEARCH." }, { "model_id": "claude-sonnet-4.6", "benchmark_id": "lmarena_search_elo", "score": 1221, "reference_url": "https://ernie.baidu.com/blog/posts/ernie-5.1-0508-release/lmarena-search-ranking.png", "reported_setting": { "leaderboard": "Arena AI Search", "metric": "Arena score", "style_control": "off", "tools": "search arena environment", "snapshot_date": "2026-05-09" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "May 9, 2026 Search Arena snapshot; style control off. Research observation obs-037. Displayed rank 5. Baidu chart footer cites ARENA.AI/LEADERBOARD/SEARCH." }, { "model_id": "deepseek-v4-pro", "benchmark_id": "spreadsheetbench_verified", "score": 67.0, "reference_url": "https://ernie.baidu.com/blog/posts/ernie-5.1-0508-release/benchmark.png", "reported_setting": { "tools": "code execution/spreadsheet agent", "sampling": "pass@1" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Verified split, pass@1, with spreadsheet/code execution. Research observation obs-014." }, { "model_id": "ernie-5.1", "benchmark_id": "aime_2026", "score": 99.6, "reference_url": "https://ernie.baidu.com/blog/posts/ernie-5.1-0508-release/benchmark.png", "reported_setting": { "tools": "tools enabled (unspecified)", "sampling": "avg@32", "metric": "score" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "With-tool avg@32 value is noncanonical against pass@1/no-tools. Research observation obs-001.", "candidates": [ { "score": 95.0, "reference_url": "https://ernie.baidu.com/blog/posts/ernie-5.1-0508-release/benchmark.png", "source_type": "official_blog", "reported_setting": { "tools": "none implied by contrast with w/tools row", "sampling": "avg@32" }, "notes": "No-tool avg@32 value is noncanonical against pass@1. Research observation obs-017." } ] }, { "model_id": "ernie-5.1", "benchmark_id": "deepsearchqa_f1", "score": 77.3, "reference_url": "https://ernie.baidu.com/blog/posts/ernie-5.1-0508-release/benchmark.png", "reported_setting": { "tools": "search/agentic", "metric": "F1" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "F1 search-agent row matches the campaign benchmark identity. Research observation obs-009." }, { "model_id": "ernie-5.1", "benchmark_id": "lmarena_search_elo", "score": 1223, "reference_url": "https://ernie.baidu.com/blog/posts/ernie-5.1-0508-release/lmarena-search-ranking.png", "reported_setting": { "leaderboard": "Arena AI Search", "metric": "Arena score", "style_control": "off", "tools": "search arena environment", "snapshot_date": "2026-05-09" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "May 9, 2026 Search Arena snapshot; style control off. Research observation obs-036. Displayed rank 4. Baidu chart footer cites ARENA.AI/LEADERBOARD/SEARCH." }, { "model_id": "ernie-5.1", "benchmark_id": "mmlu_pro", "score": 84.3, "reference_url": "https://ernie.baidu.com/blog/posts/ernie-5.1-0508-release/benchmark.png", "reported_setting": { "metric": "accuracy" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Accuracy row matches MMLU-Pro; source does not disclose tools. Research observation obs-025." }, { "model_id": "ernie-5.1", "benchmark_id": "spreadsheetbench_verified", "score": 72.5, "reference_url": "https://ernie.baidu.com/blog/posts/ernie-5.1-0508-release/benchmark.png", "reported_setting": { "tools": "code execution/spreadsheet agent", "sampling": "pass@1" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Verified split, pass@1, with spreadsheet/code execution. Research observation obs-013." }, { "model_id": "ernie-5.1", "benchmark_id": "tau3_bench", "score": 67.9, "reference_url": "https://ernie.baidu.com/blog/posts/ernie-5.1-0508-release/benchmark.png", "reported_setting": { "tools": "agentic", "sampling": "avg@4" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "The source's avg@4 agentic setting matches the campaign row. Research observation obs-005." }, { "model_id": "ernie-5.1-preview", "benchmark_id": "chatbot_arena_elo", "score": 1476, "reference_url": "https://ernie.baidu.com/blog/posts/ernie-5.1-0508-release/creative-capability.png", "reported_setting": { "leaderboard": "Arena AI Text", "metric": "Arena score", "style_control": "on", "snapshot_date": "2026-04-30" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "April 30, 2026 Text Arena snapshot; style control on. Research observation obs-060. Displayed rank 13. Identical asset is published in four page placements. The same PNG bytes are published under two filenames and four English/Chinese page placements." }, { "model_id": "gemini-3-flash", "benchmark_id": "lmarena_search_elo", "score": 1209, "reference_url": "https://ernie.baidu.com/blog/posts/ernie-5.1-0508-release/lmarena-search-ranking.png", "reported_setting": { "leaderboard": "Arena AI Search", "metric": "Arena score", "style_control": "off", "tools": "grounding/search", "snapshot_date": "2026-05-09" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "May 9, 2026 Search Arena snapshot; style control off. Research observation obs-041. Displayed rank 9. Baidu chart footer cites ARENA.AI/LEADERBOARD/SEARCH." }, { "model_id": "gemini-3-pro", "benchmark_id": "lmarena_search_elo", "score": 1209, "reference_url": "https://ernie.baidu.com/blog/posts/ernie-5.1-0508-release/lmarena-search-ranking.png", "reported_setting": { "leaderboard": "Arena AI Search", "metric": "Arena score", "style_control": "off", "tools": "grounding/search", "snapshot_date": "2026-05-09" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "May 9, 2026 Search Arena snapshot; style control off. Research observation obs-040. Displayed rank 8. Baidu chart footer cites ARENA.AI/LEADERBOARD/SEARCH." }, { "model_id": "gemini-3.1-pro", "benchmark_id": "lmarena_search_elo", "score": 1217, "reference_url": "https://ernie.baidu.com/blog/posts/ernie-5.1-0508-release/lmarena-search-ranking.png", "reported_setting": { "leaderboard": "Arena AI Search", "metric": "Arena score", "style_control": "off", "tools": "grounding/search", "snapshot_date": "2026-05-09" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "May 9, 2026 Search Arena snapshot; style control off. Research observation obs-038. Displayed rank 6. Baidu chart footer cites ARENA.AI/LEADERBOARD/SEARCH." }, { "model_id": "gemini-3.1-pro", "benchmark_id": "spreadsheetbench_verified", "score": 80.8, "reference_url": "https://ernie.baidu.com/blog/posts/ernie-5.1-0508-release/benchmark.png", "reported_setting": { "tools": "code execution/spreadsheet agent", "sampling": "pass@1" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Verified split, pass@1, with spreadsheet/code execution. Research observation obs-016." }, { "model_id": "gpt-5.1", "benchmark_id": "lmarena_search_elo", "score": 1200, "reference_url": "https://ernie.baidu.com/blog/posts/ernie-5.1-0508-release/lmarena-search-ranking.png", "reported_setting": { "leaderboard": "Arena AI Search", "metric": "Arena score", "style_control": "off", "tools": "search arena environment", "snapshot_date": "2026-05-09" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "May 9, 2026 Search Arena snapshot; style control off. Research observation obs-044. Displayed rank 12. Baidu chart footer cites ARENA.AI/LEADERBOARD/SEARCH." }, { "model_id": "gpt-5.2", "benchmark_id": "lmarena_search_elo", "score": 1213, "reference_url": "https://ernie.baidu.com/blog/posts/ernie-5.1-0508-release/lmarena-search-ranking.png", "reported_setting": { "leaderboard": "Arena AI Search", "metric": "Arena score", "style_control": "off", "tools": "search arena environment", "snapshot_date": "2026-05-09" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "May 9, 2026 Search Arena snapshot; style control off. Research observation obs-039. Displayed rank 7. Baidu chart footer cites ARENA.AI/LEADERBOARD/SEARCH." }, { "model_id": "gpt-5.4", "benchmark_id": "chatbot_arena_elo", "score": 1478, "reference_url": "https://ernie.baidu.com/blog/posts/ernie-5.1-0508-release/creative-capability.png", "reported_setting": { "leaderboard": "Arena AI Text", "metric": "Arena score", "style_control": "on", "effort": "high", "snapshot_date": "2026-04-30" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "April 30, 2026 Text Arena snapshot; style control on. Research observation obs-057. Displayed rank 10. Identical asset is published in four page placements. The same PNG bytes are published under two filenames and four English/Chinese page placements." }, { "model_id": "gpt-5.4", "benchmark_id": "lmarena_search_elo", "score": 1201, "reference_url": "https://ernie.baidu.com/blog/posts/ernie-5.1-0508-release/lmarena-search-ranking.png", "reported_setting": { "leaderboard": "Arena AI Search", "metric": "Arena score", "style_control": "off", "tools": "search arena environment", "snapshot_date": "2026-05-09" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "May 9, 2026 Search Arena snapshot; style control off. Research observation obs-043. Displayed rank 11. Baidu chart footer cites ARENA.AI/LEADERBOARD/SEARCH." }, { "model_id": "gpt-5.5", "benchmark_id": "chatbot_arena_elo", "score": 1488, "reference_url": "https://ernie.baidu.com/blog/posts/ernie-5.1-0508-release/creative-capability.png", "reported_setting": { "leaderboard": "Arena AI Text", "metric": "Arena score", "style_control": "on", "effort": "high", "snapshot_date": "2026-04-30" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "April 30, 2026 Text Arena snapshot; style control on. Research observation obs-054. Displayed rank 7. Identical asset is published in four page placements. The same PNG bytes are published under two filenames and four English/Chinese page placements.", "candidates": [ { "score": 1473, "reference_url": "https://ernie.baidu.com/blog/posts/ernie-5.1-0508-release/creative-capability.png", "source_type": "official_blog", "reported_setting": { "leaderboard": "Arena AI Text", "metric": "Arena score", "style_control": "on", "snapshot_date": "2026-04-30" }, "notes": "April 30, 2026 Text Arena snapshot; style control on. Research observation obs-062. Displayed rank 15. Identical asset is published in four page placements. The same PNG bytes are published under two filenames and four English/Chinese page placements." } ] }, { "model_id": "gpt-5.5", "benchmark_id": "lmarena_search_elo", "score": 1242, "reference_url": "https://ernie.baidu.com/blog/posts/ernie-5.1-0508-release/lmarena-search-ranking.png", "reported_setting": { "leaderboard": "Arena AI Search", "metric": "Arena score", "style_control": "off", "tools": "search arena environment", "snapshot_date": "2026-05-09" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "May 9, 2026 Search Arena snapshot; style control off. Research observation obs-034. Displayed rank 2. Baidu chart footer cites ARENA.AI/LEADERBOARD/SEARCH." }, { "model_id": "grok-4.20", "benchmark_id": "chatbot_arena_elo", "score": 1479, "reference_url": "https://ernie.baidu.com/blog/posts/ernie-5.1-0508-release/creative-capability.png", "reported_setting": { "leaderboard": "Arena AI Text", "metric": "Arena score", "style_control": "on", "mode": "unspecified", "snapshot_date": "2026-04-30" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "April 30, 2026 Text Arena snapshot; style control on. Research observation obs-056. Displayed rank 9. Identical asset is published in four page placements. The same PNG bytes are published under two filenames and four English/Chinese page placements.", "candidates": [ { "score": 1477, "reference_url": "https://ernie.baidu.com/blog/posts/ernie-5.1-0508-release/creative-capability.png", "source_type": "official_blog", "reported_setting": { "leaderboard": "Arena AI Text", "metric": "Arena score", "style_control": "on", "mode": "reasoning", "snapshot_date": "2026-04-30" }, "notes": "April 30, 2026 Text Arena snapshot; style control on. Research observation obs-058. Displayed rank 11. Identical asset is published in four page placements. The same PNG bytes are published under two filenames and four English/Chinese page placements." }, { "score": 1475, "reference_url": "https://ernie.baidu.com/blog/posts/ernie-5.1-0508-release/creative-capability.png", "source_type": "official_blog", "reported_setting": { "leaderboard": "Arena AI Text", "metric": "Arena score", "style_control": "on", "mode": "multi-agent", "snapshot_date": "2026-04-30" }, "notes": "April 30, 2026 Text Arena snapshot; style control on. Research observation obs-061. Displayed rank 14. Identical asset is published in four page placements. The same PNG bytes are published under two filenames and four English/Chinese page placements." } ] }, { "model_id": "grok-4.20", "benchmark_id": "lmarena_search_elo", "score": 1209, "reference_url": "https://ernie.baidu.com/blog/posts/ernie-5.1-0508-release/lmarena-search-ranking.png", "reported_setting": { "leaderboard": "Arena AI Search", "metric": "Arena score", "style_control": "off", "mode": "multi-agent", "tools": "search arena environment", "snapshot_date": "2026-05-09" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "May 9, 2026 Search Arena snapshot; style control off. Research observation obs-042. Displayed rank 10. Baidu chart footer cites ARENA.AI/LEADERBOARD/SEARCH.", "candidates": [ { "score": 1197, "reference_url": "https://ernie.baidu.com/blog/posts/ernie-5.1-0508-release/lmarena-search-ranking.png", "source_type": "official_blog", "reported_setting": { "leaderboard": "Arena AI Search", "metric": "Arena score", "style_control": "off", "mode": "unspecified", "tools": "search arena environment", "snapshot_date": "2026-05-09" }, "notes": "May 9, 2026 Search Arena snapshot; style control off. Research observation obs-046. Displayed rank 14. Baidu chart footer cites ARENA.AI/LEADERBOARD/SEARCH." } ] }, { "model_id": "grok-4.3", "benchmark_id": "lmarena_search_elo", "score": 1200, "reference_url": "https://ernie.baidu.com/blog/posts/ernie-5.1-0508-release/lmarena-search-ranking.png", "reported_setting": { "leaderboard": "Arena AI Search", "metric": "Arena score", "style_control": "off", "tools": "search arena environment", "snapshot_date": "2026-05-09" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "May 9, 2026 Search Arena snapshot; style control off. Research observation obs-045. Displayed rank 13. Baidu chart footer cites ARENA.AI/LEADERBOARD/SEARCH." }, { "model_id": "muse-spark", "benchmark_id": "chatbot_arena_elo", "score": 1489, "reference_url": "https://ernie.baidu.com/blog/posts/ernie-5.1-0508-release/creative-capability.png", "reported_setting": { "leaderboard": "Arena AI Text", "metric": "Arena score", "style_control": "on", "snapshot_date": "2026-04-30" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "April 30, 2026 Text Arena snapshot; style control on. Research observation obs-053. Displayed rank 6. Identical asset is published in four page placements. The same PNG bytes are published under two filenames and four English/Chinese page placements." }, { "model_id": "claude-haiku-4.5", "benchmark_id": "aa_lcr_mistral_custom_gpt_4_1_mini_middle_out", "score": 80, "reference_url": "https://huggingface.co/mistralai/Mistral-Small-4-119B-2603/resolve/a11f36bebf709121056b1dbcc943d1c6afbe494d/images/aime.png?download=true", "reported_setting": { "mode": "thinking", "effort": "128K thinking budget", "tools": "none", "sampling": "pass@1; repeat count undisclosed by Mistral", "judge": "gpt-4.1-mini-2025-04-14 equality judge", "harness": "Mistral custom AA-LCR with middle-out context handling", "prompt_style": "default", "temperature": "default (Anthropic default)", "context": "source does not state" }, "matches_canonical": false, "source_type": "official_model_card", "audit_status": "verified", "notes": "Displayed exactly: '80'. Research observation: small-aime.png:1:3:score:reasoning. Chart footnotes: *Custom implementation of AA LCR using gpt-4.1-mini-2025-04-14 as a judge and a middle-out approach for models with shorter context lengths.", "candidates": [ { "score": 64, "reference_url": "https://huggingface.co/mistralai/Mistral-Small-4-119B-2603/resolve/a11f36bebf709121056b1dbcc943d1c6afbe494d/images/aime.png?download=true", "source_type": "official_model_card", "reported_setting": { "mode": "non-thinking", "effort": "none", "tools": "none", "sampling": "pass@1; repeat count undisclosed by Mistral", "judge": "gpt-4.1-mini-2025-04-14 equality judge", "harness": "Mistral custom AA-LCR with middle-out context handling", "prompt_style": "default", "temperature": "default (Anthropic default)", "context": "source does not state" }, "notes": "Displayed exactly: '64'. Research observation: small-aime.png:1:3:score:instruct. Chart footnotes: *Custom implementation of AA LCR using gpt-4.1-mini-2025-04-14 as a judge and a middle-out approach for models with shorter context lengths." } ] }, { "model_id": "claude-sonnet-4.5", "benchmark_id": "tau3_airline", "score": 72.0, "reference_url": "https://huggingface.co/mistralai/Mistral-Medium-3.5-128B/resolve/22b2b868a15677cfa6061277ed2f653d1349a9ab/images/image4.png?download=true", "reported_setting": { "mode": "thinking", "effort": "high (maximum)", "tools": "τ³ domain tools and retrieval environment", "sampling": "Sierra-reported; trial count not stated here", "judge": "benchmark-specified", "harness": "Sierra-reported τ³ result", "prompt_style": "interleaved scratchpads", "temperature": "default (top_p, temperature)", "context": "200K (default)" }, "matches_canonical": false, "source_type": "official_model_card", "audit_status": "verified", "notes": "Displayed exactly: '72.0'. Research observation: medium-image4.png:3:2:score:reported. Chart footnotes: *All model evaluations were run with maximum reasoning settings. | **Sonnet 4.6 encountered higher rates of reasoning truncation due to external API restrictions than other models, which affected its performance. | *** Self-reported | **** Self-reported, Mistral is using context management and a discard-all strategy at 100k tokens. | τ³ scores as reported by Sierra for Claude Sonnet 4.5 and Qwen3.5. Others with user simulator: gpt-5.2 with reasoning_effort: low. 4 trials. Banking domain evaluated with terminal- or embedding-based agentic search retrieval, only highest score is reported." }, { "model_id": "claude-sonnet-4.5", "benchmark_id": "tau3_banking", "score": 22.4, "reference_url": "https://huggingface.co/mistralai/Mistral-Medium-3.5-128B/resolve/22b2b868a15677cfa6061277ed2f653d1349a9ab/images/image4.png?download=true", "reported_setting": { "mode": "thinking", "effort": "high (maximum)", "tools": "τ³ Banking terminal or embedding retrieval; higher of the two reported", "sampling": "Sierra-reported; trial count not stated here", "judge": "benchmark-specified", "harness": "Sierra-reported τ³ result", "prompt_style": "interleaved scratchpads", "temperature": "default (top_p, temperature)", "context": "200K (default)" }, "matches_canonical": false, "source_type": "official_model_card", "audit_status": "verified", "notes": "Displayed exactly: '22.4'. Research observation: medium-image4.png:5:2:score:reported. Chart footnotes: *All model evaluations were run with maximum reasoning settings. | **Sonnet 4.6 encountered higher rates of reasoning truncation due to external API restrictions than other models, which affected its performance. | *** Self-reported | **** Self-reported, Mistral is using context management and a discard-all strategy at 100k tokens. | τ³ scores as reported by Sierra for Claude Sonnet 4.5 and Qwen3.5. Others with user simulator: gpt-5.2 with reasoning_effort: low. 4 trials. Banking domain evaluated with terminal- or embedding-based agentic search retrieval, only highest score is reported." }, { "model_id": "claude-sonnet-4.5", "benchmark_id": "tau3_retail", "score": 72.4, "reference_url": "https://huggingface.co/mistralai/Mistral-Medium-3.5-128B/resolve/22b2b868a15677cfa6061277ed2f653d1349a9ab/images/image4.png?download=true", "reported_setting": { "mode": "thinking", "effort": "high (maximum)", "tools": "τ³ domain tools and retrieval environment", "sampling": "Sierra-reported; trial count not stated here", "judge": "benchmark-specified", "harness": "Sierra-reported τ³ result", "prompt_style": "interleaved scratchpads", "temperature": "default (top_p, temperature)", "context": "200K (default)" }, "matches_canonical": false, "source_type": "official_model_card", "audit_status": "verified", "notes": "Displayed exactly: '72.4'. Research observation: medium-image4.png:4:2:score:reported. Chart footnotes: *All model evaluations were run with maximum reasoning settings. | **Sonnet 4.6 encountered higher rates of reasoning truncation due to external API restrictions than other models, which affected its performance. | *** Self-reported | **** Self-reported, Mistral is using context management and a discard-all strategy at 100k tokens. | τ³ scores as reported by Sierra for Claude Sonnet 4.5 and Qwen3.5. Others with user simulator: gpt-5.2 with reasoning_effort: low. 4 trials. Banking domain evaluated with terminal- or embedding-based agentic search retrieval, only highest score is reported." }, { "model_id": "claude-sonnet-4.5", "benchmark_id": "tau3_telecom", "score": 84.9, "reference_url": "https://huggingface.co/mistralai/Mistral-Medium-3.5-128B/resolve/22b2b868a15677cfa6061277ed2f653d1349a9ab/images/image4.png?download=true", "reported_setting": { "mode": "thinking", "effort": "high (maximum)", "tools": "τ³ domain tools and retrieval environment", "sampling": "Sierra-reported; trial count not stated here", "judge": "benchmark-specified", "harness": "Sierra-reported τ³ result", "prompt_style": "interleaved scratchpads", "temperature": "default (top_p, temperature)", "context": "200K (default)" }, "matches_canonical": false, "source_type": "official_model_card", "audit_status": "verified", "notes": "Displayed exactly: '84.9'. Research observation: medium-image4.png:2:2:score:reported. Chart footnotes: *All model evaluations were run with maximum reasoning settings. | **Sonnet 4.6 encountered higher rates of reasoning truncation due to external API restrictions than other models, which affected its performance. | *** Self-reported | **** Self-reported, Mistral is using context management and a discard-all strategy at 100k tokens. | τ³ scores as reported by Sierra for Claude Sonnet 4.5 and Qwen3.5. Others with user simulator: gpt-5.2 with reasoning_effort: low. 4 trials. Banking domain evaluated with terminal- or embedding-based agentic search retrieval, only highest score is reported." }, { "model_id": "claude-sonnet-4.6", "benchmark_id": "beyond_aime", "score": 47.3, "reference_url": "https://huggingface.co/mistralai/Mistral-Medium-3.5-128B/resolve/22b2b868a15677cfa6061277ed2f653d1349a9ab/images/image1.png?download=true", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "none", "sampling": "samples=16 (reported as avg@16)", "judge": "benchmark-specified", "harness": "Mistral official maximum-reasoning comparison", "prompt_style": "default", "temperature": "default", "context": "default" }, "matches_canonical": false, "source_type": "official_model_card", "audit_status": "verified", "notes": "Displayed exactly: '47.3'. Research observation: medium-image1.png:4:3:score:reported. Chart footnotes: *All model evaluations were run with maximum reasoning settings. | **Sonnet 4.6 encountered higher rates of reasoning truncation due to external API restrictions than other models, which affected its performance." }, { "model_id": "claude-sonnet-4.6", "benchmark_id": "collie", "score": 67.7, "reference_url": "https://huggingface.co/mistralai/Mistral-Medium-3.5-128B/resolve/22b2b868a15677cfa6061277ed2f653d1349a9ab/images/image1.png?download=true", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "none", "sampling": "pass@1; repeat count undisclosed", "judge": "benchmark-specified", "harness": "Mistral official maximum-reasoning comparison", "prompt_style": "default", "temperature": "default", "context": "default" }, "matches_canonical": false, "source_type": "official_model_card", "audit_status": "verified", "notes": "Displayed exactly: '67.7'. Research observation: medium-image1.png:3:3:score:reported. Chart footnotes: *All model evaluations were run with maximum reasoning settings. | **Sonnet 4.6 encountered higher rates of reasoning truncation due to external API restrictions than other models, which affected its performance." }, { "model_id": "claude-sonnet-4.6", "benchmark_id": "ifbench", "score": 57.1, "reference_url": "https://huggingface.co/mistralai/Mistral-Medium-3.5-128B/resolve/22b2b868a15677cfa6061277ed2f653d1349a9ab/images/image1.png?download=true", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "none", "sampling": "pass@1; repeat count undisclosed", "judge": "benchmark-specified", "harness": "Mistral official maximum-reasoning comparison", "prompt_style": "default", "temperature": "default", "context": "default" }, "matches_canonical": false, "source_type": "official_model_card", "audit_status": "verified", "notes": "Displayed exactly: '57.1'. Research observation: medium-image1.png:2:3:score:reported. Chart footnotes: *All model evaluations were run with maximum reasoning settings. | **Sonnet 4.6 encountered higher rates of reasoning truncation due to external API restrictions than other models, which affected its performance.", "candidates": [ { "score": 50.0, "reference_url": "https://microsoft.ai/wp-content/uploads/2026/06/main_20260602_2.pdf", "source_type": "official_paper", "reported_setting": { "mode": "thinking", "effort": "maximum reasoning effort", "tools": "none; benchmark-provided tools where required", "sampling": "source does not state trial count", "judge": "benchmark-specific Microsoft implementation", "harness": "Microsoft independent evaluation suite", "prompt_style": "benchmark-specific", "temperature": "source does not state", "context": "maximum sequence length", "notes": "Sonnet 4.6 result generated by Microsoft's own evaluation." }, "notes": "Displayed exactly as 50. Official Microsoft-reported result." } ] }, { "model_id": "claude-sonnet-4.6", "benchmark_id": "tau3_airline", "score": 83.0, "reference_url": "https://huggingface.co/mistralai/Mistral-Medium-3.5-128B/resolve/22b2b868a15677cfa6061277ed2f653d1349a9ab/images/image4.png?download=true", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "τ³ domain tools and retrieval environment", "sampling": "trials=4", "judge": "benchmark-specified", "harness": "Mistral τ³ evaluation with GPT-5.2 low-effort user simulator", "prompt_style": "default", "temperature": "default", "context": "default" }, "matches_canonical": false, "source_type": "official_model_card", "audit_status": "verified", "notes": "Displayed exactly: '83.0'. Research observation: medium-image4.png:3:3:score:reported. Chart footnotes: *All model evaluations were run with maximum reasoning settings. | **Sonnet 4.6 encountered higher rates of reasoning truncation due to external API restrictions than other models, which affected its performance. | *** Self-reported | **** Self-reported, Mistral is using context management and a discard-all strategy at 100k tokens. | τ³ scores as reported by Sierra for Claude Sonnet 4.5 and Qwen3.5. Others with user simulator: gpt-5.2 with reasoning_effort: low. 4 trials. Banking domain evaluated with terminal- or embedding-based agentic search retrieval, only highest score is reported." }, { "model_id": "claude-sonnet-4.6", "benchmark_id": "tau3_banking", "score": 28.4, "reference_url": "https://huggingface.co/mistralai/Mistral-Medium-3.5-128B/resolve/22b2b868a15677cfa6061277ed2f653d1349a9ab/images/image4.png?download=true", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "τ³ Banking terminal or embedding retrieval; higher of the two reported", "sampling": "trials=4", "judge": "benchmark-specified", "harness": "Mistral τ³ evaluation with GPT-5.2 low-effort user simulator", "prompt_style": "default", "temperature": "default", "context": "default" }, "matches_canonical": false, "source_type": "official_model_card", "audit_status": "verified", "notes": "Displayed exactly: '28.4'. Research observation: medium-image4.png:5:3:score:reported. Chart footnotes: *All model evaluations were run with maximum reasoning settings. | **Sonnet 4.6 encountered higher rates of reasoning truncation due to external API restrictions than other models, which affected its performance. | *** Self-reported | **** Self-reported, Mistral is using context management and a discard-all strategy at 100k tokens. | τ³ scores as reported by Sierra for Claude Sonnet 4.5 and Qwen3.5. Others with user simulator: gpt-5.2 with reasoning_effort: low. 4 trials. Banking domain evaluated with terminal- or embedding-based agentic search retrieval, only highest score is reported." }, { "model_id": "claude-sonnet-4.6", "benchmark_id": "tau3_retail", "score": 75.9, "reference_url": "https://huggingface.co/mistralai/Mistral-Medium-3.5-128B/resolve/22b2b868a15677cfa6061277ed2f653d1349a9ab/images/image4.png?download=true", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "τ³ domain tools and retrieval environment", "sampling": "trials=4", "judge": "benchmark-specified", "harness": "Mistral τ³ evaluation with GPT-5.2 low-effort user simulator", "prompt_style": "default", "temperature": "default", "context": "default" }, "matches_canonical": false, "source_type": "official_model_card", "audit_status": "verified", "notes": "Displayed exactly: '75.9'. Research observation: medium-image4.png:4:3:score:reported. Chart footnotes: *All model evaluations were run with maximum reasoning settings. | **Sonnet 4.6 encountered higher rates of reasoning truncation due to external API restrictions than other models, which affected its performance. | *** Self-reported | **** Self-reported, Mistral is using context management and a discard-all strategy at 100k tokens. | τ³ scores as reported by Sierra for Claude Sonnet 4.5 and Qwen3.5. Others with user simulator: gpt-5.2 with reasoning_effort: low. 4 trials. Banking domain evaluated with terminal- or embedding-based agentic search retrieval, only highest score is reported." }, { "model_id": "claude-sonnet-4.6", "benchmark_id": "tau3_telecom", "score": 70.4, "reference_url": "https://huggingface.co/mistralai/Mistral-Medium-3.5-128B/resolve/22b2b868a15677cfa6061277ed2f653d1349a9ab/images/image4.png?download=true", "reported_setting": { "mode": "thinking", "effort": "max", "tools": "τ³ domain tools and retrieval environment", "sampling": "trials=4", "judge": "benchmark-specified", "harness": "Mistral τ³ evaluation with GPT-5.2 low-effort user simulator", "prompt_style": "default", "temperature": "default", "context": "default" }, "matches_canonical": false, "source_type": "official_model_card", "audit_status": "verified", "notes": "Displayed exactly: '70.4'. Research observation: medium-image4.png:2:3:score:reported. Chart footnotes: *All model evaluations were run with maximum reasoning settings. | **Sonnet 4.6 encountered higher rates of reasoning truncation due to external API restrictions than other models, which affected its performance. | *** Self-reported | **** Self-reported, Mistral is using context management and a discard-all strategy at 100k tokens. | τ³ scores as reported by Sierra for Claude Sonnet 4.5 and Qwen3.5. Others with user simulator: gpt-5.2 with reasoning_effort: low. 4 trials. Banking domain evaluated with terminal- or embedding-based agentic search retrieval, only highest score is reported." }, { "model_id": "devstral-small-2", "benchmark_id": "swe_bench_verified", "score": 68.0, "reference_url": "https://huggingface.co/mistralai/Mistral-Medium-3.5-128B/resolve/22b2b868a15677cfa6061277ed2f653d1349a9ab/images/image3.png?download=true", "reported_setting": { "mode": "non-thinking", "effort": "none", "tools": "agentic coding scaffold and SWE-bench verifier", "sampling": "source does not state trial count", "judge": "benchmark-specified", "harness": "Mistral self-reported SWE-bench Verified evaluation", "prompt_style": "official CHAT_SYSTEM_PROMPT.txt", "temperature": "0.2 release recommendation; immutable HF examples use 0.15", "context": "256k" }, "matches_canonical": false, "source_type": "official_model_card", "audit_status": "verified", "notes": "Displayed exactly: '68.0'. Research observation: medium-image3.png:1:3:score:reported." }, { "model_id": "glm-5", "benchmark_id": "aime_2025", "score": 87.1, "reference_url": "https://huggingface.co/mistralai/Mistral-Medium-3.5-128B/resolve/22b2b868a15677cfa6061277ed2f653d1349a9ab/images/image1.png?download=true", "reported_setting": { "mode": "thinking", "effort": "default (maximum reasoning setting)", "tools": "none", "sampling": "samples=16 (reported as avg@16)", "judge": "benchmark-specified", "harness": "Mistral official maximum-reasoning comparison", "prompt_style": "default", "temperature": "default", "context": "default" }, "matches_canonical": false, "source_type": "official_model_card", "audit_status": "verified", "notes": "Displayed exactly: '87.1'. Research observation: medium-image1.png:1:6:score:reported. Chart footnotes: *All model evaluations were run with maximum reasoning settings. | **Sonnet 4.6 encountered higher rates of reasoning truncation due to external API restrictions than other models, which affected its performance." }, { "model_id": "glm-5", "benchmark_id": "collie", "score": 86.4, "reference_url": "https://huggingface.co/mistralai/Mistral-Medium-3.5-128B/resolve/22b2b868a15677cfa6061277ed2f653d1349a9ab/images/image1.png?download=true", "reported_setting": { "mode": "thinking", "effort": "default (maximum reasoning setting)", "tools": "none", "sampling": "pass@1; repeat count undisclosed", "judge": "benchmark-specified", "harness": "Mistral official maximum-reasoning comparison", "prompt_style": "default", "temperature": "default", "context": "default" }, "matches_canonical": false, "source_type": "official_model_card", "audit_status": "verified", "notes": "Displayed exactly: '86.4'. Research observation: medium-image1.png:3:6:score:reported. Chart footnotes: *All model evaluations were run with maximum reasoning settings. | **Sonnet 4.6 encountered higher rates of reasoning truncation due to external API restrictions than other models, which affected its performance." }, { "model_id": "glm-5", "benchmark_id": "ifbench", "score": 67.0, "reference_url": "https://huggingface.co/mistralai/Mistral-Medium-3.5-128B/resolve/22b2b868a15677cfa6061277ed2f653d1349a9ab/images/image1.png?download=true", "reported_setting": { "mode": "thinking", "effort": "default (maximum reasoning setting)", "tools": "none", "sampling": "pass@1; repeat count undisclosed", "judge": "benchmark-specified", "harness": "Mistral official maximum-reasoning comparison", "prompt_style": "default", "temperature": "default", "context": "default" }, "matches_canonical": false, "source_type": "official_model_card", "audit_status": "verified", "notes": "Displayed exactly: '67.0'. Research observation: medium-image1.png:2:6:score:reported. Chart footnotes: *All model evaluations were run with maximum reasoning settings. | **Sonnet 4.6 encountered higher rates of reasoning truncation due to external API restrictions than other models, which affected its performance." }, { "model_id": "gpt-oss-120b", "benchmark_id": "aa_lcr_mistral_custom_gpt_4_1_mini_middle_out", "score": 51, "reference_url": "https://huggingface.co/mistralai/Mistral-Small-4-119B-2603/resolve/a11f36bebf709121056b1dbcc943d1c6afbe494d/images/aime.png?download=true", "reported_setting": { "mode": "thinking", "effort": "chart reasoning mode; exact effort level undisclosed", "tools": "none", "sampling": "pass@1; repeat count undisclosed by Mistral", "judge": "gpt-4.1-mini-2025-04-14 equality judge", "harness": "Mistral custom AA-LCR with middle-out context handling", "prompt_style": "default", "temperature": "default", "context": "default (128K)" }, "matches_canonical": false, "source_type": "official_model_card", "audit_status": "verified", "notes": "Displayed exactly: '51'. Research observation: small-aime.png:1:2:score:reasoning. Chart footnotes: *Custom implementation of AA LCR using gpt-4.1-mini-2025-04-14 as a judge and a middle-out approach for models with shorter context lengths.", "candidates": [ { "score": 40, "reference_url": "https://huggingface.co/mistralai/Mistral-Small-4-119B-2603/resolve/a11f36bebf709121056b1dbcc943d1c6afbe494d/images/aime.png?download=true", "source_type": "official_model_card", "reported_setting": { "mode": "non-thinking", "effort": "none", "tools": "none", "sampling": "pass@1; repeat count undisclosed by Mistral", "judge": "gpt-4.1-mini-2025-04-14 equality judge", "harness": "Mistral custom AA-LCR with middle-out context handling", "prompt_style": "default", "temperature": "default", "context": "default (128K)" }, "notes": "Displayed exactly: '40'. Research observation: small-aime.png:1:2:score:instruct. Chart footnotes: *Custom implementation of AA LCR using gpt-4.1-mini-2025-04-14 as a judge and a middle-out approach for models with shorter context lengths." } ] }, { "model_id": "kimi-k2.5", "benchmark_id": "beyond_aime", "score": 60.3, "reference_url": "https://huggingface.co/mistralai/Mistral-Medium-3.5-128B/resolve/22b2b868a15677cfa6061277ed2f653d1349a9ab/images/image1.png?download=true", "reported_setting": { "mode": "thinking", "effort": "default (maximum reasoning setting)", "tools": "none", "sampling": "samples=16 (reported as avg@16)", "judge": "benchmark-specified", "harness": "Mistral official maximum-reasoning comparison", "prompt_style": "default", "temperature": "1.0", "context": "256K" }, "matches_canonical": false, "source_type": "official_model_card", "audit_status": "verified", "notes": "Displayed exactly: '60.3'. Research observation: medium-image1.png:4:4:score:reported. Chart footnotes: *All model evaluations were run with maximum reasoning settings. | **Sonnet 4.6 encountered higher rates of reasoning truncation due to external API restrictions than other models, which affected its performance." }, { "model_id": "kimi-k2.5", "benchmark_id": "collie", "score": 87.8, "reference_url": "https://huggingface.co/mistralai/Mistral-Medium-3.5-128B/resolve/22b2b868a15677cfa6061277ed2f653d1349a9ab/images/image1.png?download=true", "reported_setting": { "mode": "thinking", "effort": "default (maximum reasoning setting)", "tools": "none", "sampling": "pass@1; repeat count undisclosed", "judge": "benchmark-specified", "harness": "Mistral official maximum-reasoning comparison", "prompt_style": "default", "temperature": "1.0", "context": "256K" }, "matches_canonical": false, "source_type": "official_model_card", "audit_status": "verified", "notes": "Displayed exactly: '87.8'. Research observation: medium-image1.png:3:4:score:reported. Chart footnotes: *All model evaluations were run with maximum reasoning settings. | **Sonnet 4.6 encountered higher rates of reasoning truncation due to external API restrictions than other models, which affected its performance." }, { "model_id": "kimi-k2.5", "benchmark_id": "tau3_airline", "score": 76.5, "reference_url": "https://huggingface.co/mistralai/Mistral-Medium-3.5-128B/resolve/22b2b868a15677cfa6061277ed2f653d1349a9ab/images/image4.png?download=true", "reported_setting": { "mode": "thinking", "effort": "default (maximum reasoning setting)", "tools": "τ³ domain tools and retrieval environment", "sampling": "trials=4", "judge": "benchmark-specified", "harness": "Mistral τ³ evaluation with GPT-5.2 low-effort user simulator", "prompt_style": "default", "temperature": "1.0", "context": "256K" }, "matches_canonical": false, "source_type": "official_model_card", "audit_status": "verified", "notes": "Displayed exactly: '76.5'. Research observation: medium-image4.png:3:4:score:reported. Chart footnotes: *All model evaluations were run with maximum reasoning settings. | **Sonnet 4.6 encountered higher rates of reasoning truncation due to external API restrictions than other models, which affected its performance. | *** Self-reported | **** Self-reported, Mistral is using context management and a discard-all strategy at 100k tokens. | τ³ scores as reported by Sierra for Claude Sonnet 4.5 and Qwen3.5. Others with user simulator: gpt-5.2 with reasoning_effort: low. 4 trials. Banking domain evaluated with terminal- or embedding-based agentic search retrieval, only highest score is reported." }, { "model_id": "kimi-k2.5", "benchmark_id": "tau3_banking", "score": 14.9, "reference_url": "https://huggingface.co/mistralai/Mistral-Medium-3.5-128B/resolve/22b2b868a15677cfa6061277ed2f653d1349a9ab/images/image4.png?download=true", "reported_setting": { "mode": "thinking", "effort": "default (maximum reasoning setting)", "tools": "τ³ Banking terminal or embedding retrieval; higher of the two reported", "sampling": "trials=4", "judge": "benchmark-specified", "harness": "Mistral τ³ evaluation with GPT-5.2 low-effort user simulator", "prompt_style": "default", "temperature": "1.0", "context": "256K" }, "matches_canonical": false, "source_type": "official_model_card", "audit_status": "verified", "notes": "Displayed exactly: '14.9'. Research observation: medium-image4.png:5:4:score:reported. Chart footnotes: *All model evaluations were run with maximum reasoning settings. | **Sonnet 4.6 encountered higher rates of reasoning truncation due to external API restrictions than other models, which affected its performance. | *** Self-reported | **** Self-reported, Mistral is using context management and a discard-all strategy at 100k tokens. | τ³ scores as reported by Sierra for Claude Sonnet 4.5 and Qwen3.5. Others with user simulator: gpt-5.2 with reasoning_effort: low. 4 trials. Banking domain evaluated with terminal- or embedding-based agentic search retrieval, only highest score is reported." }, { "model_id": "kimi-k2.5", "benchmark_id": "tau3_retail", "score": 72.8, "reference_url": "https://huggingface.co/mistralai/Mistral-Medium-3.5-128B/resolve/22b2b868a15677cfa6061277ed2f653d1349a9ab/images/image4.png?download=true", "reported_setting": { "mode": "thinking", "effort": "default (maximum reasoning setting)", "tools": "τ³ domain tools and retrieval environment", "sampling": "trials=4", "judge": "benchmark-specified", "harness": "Mistral τ³ evaluation with GPT-5.2 low-effort user simulator", "prompt_style": "default", "temperature": "1.0", "context": "256K" }, "matches_canonical": false, "source_type": "official_model_card", "audit_status": "verified", "notes": "Displayed exactly: '72.8'. Research observation: medium-image4.png:4:4:score:reported. Chart footnotes: *All model evaluations were run with maximum reasoning settings. | **Sonnet 4.6 encountered higher rates of reasoning truncation due to external API restrictions than other models, which affected its performance. | *** Self-reported | **** Self-reported, Mistral is using context management and a discard-all strategy at 100k tokens. | τ³ scores as reported by Sierra for Claude Sonnet 4.5 and Qwen3.5. Others with user simulator: gpt-5.2 with reasoning_effort: low. 4 trials. Banking domain evaluated with terminal- or embedding-based agentic search retrieval, only highest score is reported." }, { "model_id": "kimi-k2.5", "benchmark_id": "tau3_telecom", "score": 86.8, "reference_url": "https://huggingface.co/mistralai/Mistral-Medium-3.5-128B/resolve/22b2b868a15677cfa6061277ed2f653d1349a9ab/images/image4.png?download=true", "reported_setting": { "mode": "thinking", "effort": "default (maximum reasoning setting)", "tools": "τ³ domain tools and retrieval environment", "sampling": "trials=4", "judge": "benchmark-specified", "harness": "Mistral τ³ evaluation with GPT-5.2 low-effort user simulator", "prompt_style": "default", "temperature": "1.0", "context": "256K" }, "matches_canonical": false, "source_type": "official_model_card", "audit_status": "verified", "notes": "Displayed exactly: '86.8'. Research observation: medium-image4.png:2:4:score:reported. Chart footnotes: *All model evaluations were run with maximum reasoning settings. | **Sonnet 4.6 encountered higher rates of reasoning truncation due to external API restrictions than other models, which affected its performance. | *** Self-reported | **** Self-reported, Mistral is using context management and a discard-all strategy at 100k tokens. | τ³ scores as reported by Sierra for Claude Sonnet 4.5 and Qwen3.5. Others with user simulator: gpt-5.2 with reasoning_effort: low. 4 trials. Banking domain evaluated with terminal- or embedding-based agentic search retrieval, only highest score is reported." }, { "model_id": "magistral-medium-1.2", "benchmark_id": "aa_lcr_mistral_custom_gpt_4_1_mini_middle_out", "score": 73, "reference_url": "https://huggingface.co/mistralai/Mistral-Small-4-119B-2603/resolve/a11f36bebf709121056b1dbcc943d1c6afbe494d/images/image2.png?download=true", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1; repeat count undisclosed by Mistral", "judge": "gpt-4.1-mini-2025-04-14 equality judge", "harness": "Mistral custom AA-LCR with middle-out context handling", "prompt_style": "official Magistral reasoning prompt", "temperature": "0.7; top_p=0.95 (family recommendation)", "context": "128k" }, "matches_canonical": false, "source_type": "official_model_card", "audit_status": "verified", "notes": "Displayed exactly: '73'. Research observation: small-image2.png:1:2:score:reported." }, { "model_id": "magistral-medium-1.2", "benchmark_id": "aime_2025", "score": 84.4, "reference_url": "https://huggingface.co/mistralai/Mistral-Small-4-119B-2603/resolve/a11f36bebf709121056b1dbcc943d1c6afbe494d/images/image2.png?download=true", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1; repeat count undisclosed", "judge": "benchmark-specified", "harness": "Mistral official model-card evaluation", "prompt_style": "official Magistral reasoning prompt", "temperature": "0.7; top_p=0.95 (family recommendation)", "context": "128k" }, "matches_canonical": false, "source_type": "official_model_card", "audit_status": "verified", "notes": "Displayed exactly: '84.4'. Research observation: small-image2.png:2:2:score:reported." }, { "model_id": "magistral-medium-1.2", "benchmark_id": "browsecomp", "score": 10.0, "reference_url": "https://huggingface.co/mistralai/Mistral-Medium-3.5-128B/resolve/22b2b868a15677cfa6061277ed2f653d1349a9ab/images/image2.png?download=true", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "web search/browser environment", "sampling": "source does not state trial count", "judge": "benchmark-specified", "harness": "Mistral official BrowseComp evaluation", "prompt_style": "official Magistral reasoning prompt", "temperature": "0.7; top_p=0.95 (family recommendation)", "context": "128k" }, "matches_canonical": false, "source_type": "official_model_card", "audit_status": "verified", "notes": "Displayed exactly: '10.0'. Research observation: medium-image2.png:5:2:score:reported." }, { "model_id": "magistral-medium-1.2", "benchmark_id": "collie", "score": 61.3, "reference_url": "https://huggingface.co/mistralai/Mistral-Small-4-119B-2603/resolve/a11f36bebf709121056b1dbcc943d1c6afbe494d/images/image2.png?download=true", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1; repeat count undisclosed", "judge": "benchmark-specified", "harness": "Mistral official model-card evaluation", "prompt_style": "official Magistral reasoning prompt", "temperature": "0.7; top_p=0.95 (family recommendation)", "context": "128k" }, "matches_canonical": false, "source_type": "official_model_card", "audit_status": "verified", "notes": "Displayed exactly: '61.3'. Research observation: small-image2.png:3:2:score:reported." }, { "model_id": "magistral-medium-1.2", "benchmark_id": "tau3_airline", "score": 53.5, "reference_url": "https://huggingface.co/mistralai/Mistral-Medium-3.5-128B/resolve/22b2b868a15677cfa6061277ed2f653d1349a9ab/images/image2.png?download=true", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "τ³ domain tools and retrieval environment", "sampling": "source does not state trial count", "judge": "benchmark-specified", "harness": "Mistral official τ³ evaluation", "prompt_style": "official Magistral reasoning prompt", "temperature": "0.7; top_p=0.95 (family recommendation)", "context": "128k" }, "matches_canonical": false, "source_type": "official_model_card", "audit_status": "verified", "notes": "Displayed exactly: '53.5'. Research observation: medium-image2.png:2:2:score:reported." }, { "model_id": "magistral-medium-1.2", "benchmark_id": "tau3_banking", "score": 7.7, "reference_url": "https://huggingface.co/mistralai/Mistral-Medium-3.5-128B/resolve/22b2b868a15677cfa6061277ed2f653d1349a9ab/images/image2.png?download=true", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "τ³ domain tools and retrieval environment", "sampling": "source does not state trial count", "judge": "benchmark-specified", "harness": "Mistral official τ³ evaluation", "prompt_style": "official Magistral reasoning prompt", "temperature": "0.7; top_p=0.95 (family recommendation)", "context": "128k" }, "matches_canonical": false, "source_type": "official_model_card", "audit_status": "verified", "notes": "Displayed exactly: '7.7'. Research observation: medium-image2.png:4:2:score:reported." }, { "model_id": "magistral-medium-1.2", "benchmark_id": "tau3_retail", "score": 70.2, "reference_url": "https://huggingface.co/mistralai/Mistral-Medium-3.5-128B/resolve/22b2b868a15677cfa6061277ed2f653d1349a9ab/images/image2.png?download=true", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "τ³ domain tools and retrieval environment", "sampling": "source does not state trial count", "judge": "benchmark-specified", "harness": "Mistral official τ³ evaluation", "prompt_style": "official Magistral reasoning prompt", "temperature": "0.7; top_p=0.95 (family recommendation)", "context": "128k" }, "matches_canonical": false, "source_type": "official_model_card", "audit_status": "verified", "notes": "Displayed exactly: '70.2'. Research observation: medium-image2.png:3:2:score:reported." }, { "model_id": "magistral-medium-1.2", "benchmark_id": "tau3_telecom", "score": 60.5, "reference_url": "https://huggingface.co/mistralai/Mistral-Medium-3.5-128B/resolve/22b2b868a15677cfa6061277ed2f653d1349a9ab/images/image2.png?download=true", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "τ³ domain tools and retrieval environment", "sampling": "source does not state trial count", "judge": "benchmark-specified", "harness": "Mistral official τ³ evaluation", "prompt_style": "official Magistral reasoning prompt", "temperature": "0.7; top_p=0.95 (family recommendation)", "context": "128k" }, "matches_canonical": false, "source_type": "official_model_card", "audit_status": "verified", "notes": "Displayed exactly: '60.5'. Research observation: medium-image2.png:1:2:score:reported." }, { "model_id": "magistral-small-1.2", "benchmark_id": "aa_lcr_mistral_custom_gpt_4_1_mini_middle_out", "score": 27, "reference_url": "https://huggingface.co/mistralai/Mistral-Small-4-119B-2603/resolve/a11f36bebf709121056b1dbcc943d1c6afbe494d/images/image2.png?download=true", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1; repeat count undisclosed by Mistral", "judge": "gpt-4.1-mini-2025-04-14 equality judge", "harness": "Mistral custom AA-LCR with middle-out context handling", "prompt_style": "official Magistral reasoning system prompt", "temperature": "0.7; top_p=0.95; max_tokens=131072", "context": "128k" }, "matches_canonical": false, "source_type": "official_model_card", "audit_status": "verified", "notes": "Displayed exactly: '27'. Research observation: small-image2.png:1:3:score:reported." }, { "model_id": "magistral-small-1.2", "benchmark_id": "aime_2025", "score": 80.2, "reference_url": "https://huggingface.co/mistralai/Mistral-Small-4-119B-2603/resolve/a11f36bebf709121056b1dbcc943d1c6afbe494d/images/image2.png?download=true", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1; repeat count undisclosed", "judge": "benchmark-specified", "harness": "Mistral official model-card evaluation", "prompt_style": "official Magistral reasoning system prompt", "temperature": "0.7; top_p=0.95; max_tokens=131072", "context": "128k" }, "matches_canonical": false, "source_type": "official_model_card", "audit_status": "verified", "notes": "Displayed exactly: '80.2'. Research observation: small-image2.png:2:3:score:reported." }, { "model_id": "magistral-small-1.2", "benchmark_id": "collie", "score": 60.3, "reference_url": "https://huggingface.co/mistralai/Mistral-Small-4-119B-2603/resolve/a11f36bebf709121056b1dbcc943d1c6afbe494d/images/image2.png?download=true", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1; repeat count undisclosed", "judge": "benchmark-specified", "harness": "Mistral official model-card evaluation", "prompt_style": "official Magistral reasoning system prompt", "temperature": "0.7; top_p=0.95; max_tokens=131072", "context": "128k" }, "matches_canonical": false, "source_type": "official_model_card", "audit_status": "verified", "notes": "Displayed exactly: '60.3'. Research observation: small-image2.png:3:3:score:reported." }, { "model_id": "mistral-large-3", "benchmark_id": "ifbench", "score": 37.8, "reference_url": "https://huggingface.co/mistralai/Mistral-Small-4-119B-2603/resolve/a11f36bebf709121056b1dbcc943d1c6afbe494d/images/image3.png?download=true", "reported_setting": { "mode": "non-thinking", "effort": "none", "tools": "none", "sampling": "pass@1; repeat count undisclosed", "judge": "benchmark-specified", "harness": "Mistral official model-card evaluation", "prompt_style": "official/default or source does not state", "temperature": "source does not state", "context": "source does not state" }, "matches_canonical": false, "source_type": "official_model_card", "audit_status": "verified", "notes": "Displayed exactly: '37.8'. Research observation: small-image3.png:3:4:score:reported." }, { "model_id": "mistral-large-3", "benchmark_id": "mmmu_pro", "score": 54, "reference_url": "https://huggingface.co/mistralai/Mistral-Small-4-119B-2603/resolve/a11f36bebf709121056b1dbcc943d1c6afbe494d/images/image3.png?download=true", "reported_setting": { "mode": "non-thinking", "effort": "none", "tools": "none", "sampling": "pass@1; repeat count undisclosed", "judge": "benchmark-specified", "harness": "Mistral official model-card evaluation", "prompt_style": "official/default or source does not state", "temperature": "source does not state", "context": "source does not state" }, "matches_canonical": false, "source_type": "official_model_card", "audit_status": "verified", "notes": "Displayed exactly: '54'. Research observation: small-image3.png:5:4:score:reported." }, { "model_id": "mistral-medium-3.1", "benchmark_id": "arena_hard", "score": 67.3, "reference_url": "https://huggingface.co/mistralai/Mistral-Small-4-119B-2603/resolve/a11f36bebf709121056b1dbcc943d1c6afbe494d/images/image3.png?download=true", "reported_setting": { "mode": "non-thinking", "effort": "none", "tools": "none", "sampling": "pass@1; repeat count undisclosed", "judge": "Arena-Hard automatic judge", "harness": "Mistral official model-card evaluation", "prompt_style": "official/default", "temperature": "undisclosed Mistral API default", "context": "128k" }, "matches_canonical": false, "source_type": "official_model_card", "audit_status": "verified", "notes": "Displayed exactly: '67.3'. Research observation: small-image3.png:4:3:score:reported." }, { "model_id": "mistral-medium-3.1", "benchmark_id": "browsecomp", "score": 7.8, "reference_url": "https://huggingface.co/mistralai/Mistral-Medium-3.5-128B/resolve/22b2b868a15677cfa6061277ed2f653d1349a9ab/images/image2.png?download=true", "reported_setting": { "mode": "non-thinking", "effort": "none", "tools": "web search/browser environment", "sampling": "source does not state trial count", "judge": "benchmark-specified", "harness": "Mistral official BrowseComp evaluation", "prompt_style": "official/default", "temperature": "undisclosed Mistral API default", "context": "128k" }, "matches_canonical": false, "source_type": "official_model_card", "audit_status": "verified", "notes": "Displayed exactly: '7.8'. Research observation: medium-image2.png:5:4:score:reported." }, { "model_id": "mistral-medium-3.1", "benchmark_id": "gpqa_diamond", "score": 65.7, "reference_url": "https://huggingface.co/mistralai/Mistral-Small-4-119B-2603/resolve/a11f36bebf709121056b1dbcc943d1c6afbe494d/images/image3.png?download=true", "reported_setting": { "mode": "non-thinking", "effort": "none", "tools": "none", "sampling": "pass@1; repeat count undisclosed", "judge": "benchmark-specified", "harness": "Mistral official model-card evaluation", "prompt_style": "official/default", "temperature": "undisclosed Mistral API default", "context": "128k" }, "matches_canonical": false, "source_type": "official_model_card", "audit_status": "verified", "notes": "Displayed exactly: '65.7'. Research observation: small-image3.png:1:3:score:reported." }, { "model_id": "mistral-medium-3.1", "benchmark_id": "ifbench", "score": 40.8, "reference_url": "https://huggingface.co/mistralai/Mistral-Small-4-119B-2603/resolve/a11f36bebf709121056b1dbcc943d1c6afbe494d/images/image3.png?download=true", "reported_setting": { "mode": "non-thinking", "effort": "none", "tools": "none", "sampling": "pass@1; repeat count undisclosed", "judge": "benchmark-specified", "harness": "Mistral official model-card evaluation", "prompt_style": "official/default", "temperature": "undisclosed Mistral API default", "context": "128k" }, "matches_canonical": false, "source_type": "official_model_card", "audit_status": "verified", "notes": "Displayed exactly: '40.8'. Research observation: small-image3.png:3:3:score:reported." }, { "model_id": "mistral-medium-3.1", "benchmark_id": "mmlu_pro", "score": 76.8, "reference_url": "https://huggingface.co/mistralai/Mistral-Small-4-119B-2603/resolve/a11f36bebf709121056b1dbcc943d1c6afbe494d/images/image3.png?download=true", "reported_setting": { "mode": "non-thinking", "effort": "none", "tools": "none", "sampling": "pass@1; repeat count undisclosed", "judge": "benchmark-specified", "harness": "Mistral official model-card evaluation", "prompt_style": "official/default", "temperature": "undisclosed Mistral API default", "context": "128k" }, "matches_canonical": false, "source_type": "official_model_card", "audit_status": "verified", "notes": "Displayed exactly: '76.8'. Research observation: small-image3.png:2:3:score:reported." }, { "model_id": "mistral-medium-3.1", "benchmark_id": "mmmu_pro", "score": 43.8, "reference_url": "https://huggingface.co/mistralai/Mistral-Small-4-119B-2603/resolve/a11f36bebf709121056b1dbcc943d1c6afbe494d/images/image3.png?download=true", "reported_setting": { "mode": "non-thinking", "effort": "none", "tools": "none", "sampling": "pass@1; repeat count undisclosed", "judge": "benchmark-specified", "harness": "Mistral official model-card evaluation", "prompt_style": "official/default", "temperature": "undisclosed Mistral API default", "context": "128k" }, "matches_canonical": false, "source_type": "official_model_card", "audit_status": "verified", "notes": "Displayed exactly: '43.8'. Research observation: small-image3.png:5:3:score:reported." }, { "model_id": "mistral-medium-3.1", "benchmark_id": "tau3_airline", "score": 41.5, "reference_url": "https://huggingface.co/mistralai/Mistral-Medium-3.5-128B/resolve/22b2b868a15677cfa6061277ed2f653d1349a9ab/images/image2.png?download=true", "reported_setting": { "mode": "non-thinking", "effort": "none", "tools": "τ³ domain tools and retrieval environment", "sampling": "source does not state trial count", "judge": "benchmark-specified", "harness": "Mistral official τ³ evaluation", "prompt_style": "official/default", "temperature": "undisclosed Mistral API default", "context": "128k" }, "matches_canonical": false, "source_type": "official_model_card", "audit_status": "verified", "notes": "Displayed exactly: '41.5'. Research observation: medium-image2.png:2:4:score:reported." }, { "model_id": "mistral-medium-3.1", "benchmark_id": "tau3_banking", "score": 5.7, "reference_url": "https://huggingface.co/mistralai/Mistral-Medium-3.5-128B/resolve/22b2b868a15677cfa6061277ed2f653d1349a9ab/images/image2.png?download=true", "reported_setting": { "mode": "non-thinking", "effort": "none", "tools": "τ³ domain tools and retrieval environment", "sampling": "source does not state trial count", "judge": "benchmark-specified", "harness": "Mistral official τ³ evaluation", "prompt_style": "official/default", "temperature": "undisclosed Mistral API default", "context": "128k" }, "matches_canonical": false, "source_type": "official_model_card", "audit_status": "verified", "notes": "Displayed exactly: '5.7'. Research observation: medium-image2.png:4:4:score:reported." }, { "model_id": "mistral-medium-3.1", "benchmark_id": "tau3_retail", "score": 64.3, "reference_url": "https://huggingface.co/mistralai/Mistral-Medium-3.5-128B/resolve/22b2b868a15677cfa6061277ed2f653d1349a9ab/images/image2.png?download=true", "reported_setting": { "mode": "non-thinking", "effort": "none", "tools": "τ³ domain tools and retrieval environment", "sampling": "source does not state trial count", "judge": "benchmark-specified", "harness": "Mistral official τ³ evaluation", "prompt_style": "official/default", "temperature": "undisclosed Mistral API default", "context": "128k" }, "matches_canonical": false, "source_type": "official_model_card", "audit_status": "verified", "notes": "Displayed exactly: '64.3'. Research observation: medium-image2.png:3:4:score:reported." }, { "model_id": "mistral-medium-3.1", "benchmark_id": "tau3_telecom", "score": 46.9, "reference_url": "https://huggingface.co/mistralai/Mistral-Medium-3.5-128B/resolve/22b2b868a15677cfa6061277ed2f653d1349a9ab/images/image2.png?download=true", "reported_setting": { "mode": "non-thinking", "effort": "none", "tools": "τ³ domain tools and retrieval environment", "sampling": "source does not state trial count", "judge": "benchmark-specified", "harness": "Mistral official τ³ evaluation", "prompt_style": "official/default", "temperature": "undisclosed Mistral API default", "context": "128k" }, "matches_canonical": false, "source_type": "official_model_card", "audit_status": "verified", "notes": "Displayed exactly: '46.9'. Research observation: medium-image2.png:1:4:score:reported." }, { "model_id": "mistral-medium-3.5-128b", "benchmark_id": "aime_2025", "score": 86.3, "reference_url": "https://huggingface.co/mistralai/Mistral-Medium-3.5-128B/resolve/22b2b868a15677cfa6061277ed2f653d1349a9ab/images/image1.png?download=true", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "samples=16 (reported as avg@16)", "judge": "benchmark-specified", "harness": "Mistral official maximum-reasoning comparison", "prompt_style": "official/default", "temperature": "0.7; top_p=0.95", "context": "256k" }, "matches_canonical": false, "source_type": "official_model_card", "audit_status": "verified", "notes": "Displayed exactly: '86.3'. Research observation: medium-image1.png:1:1:score:reported. Chart footnotes: *All model evaluations were run with maximum reasoning settings. | **Sonnet 4.6 encountered higher rates of reasoning truncation due to external API restrictions than other models, which affected its performance." }, { "model_id": "mistral-medium-3.5-128b", "benchmark_id": "beyond_aime", "score": 66.9, "reference_url": "https://huggingface.co/mistralai/Mistral-Medium-3.5-128B/resolve/22b2b868a15677cfa6061277ed2f653d1349a9ab/images/image1.png?download=true", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "samples=16 (reported as avg@16)", "judge": "benchmark-specified", "harness": "Mistral official maximum-reasoning comparison", "prompt_style": "official/default", "temperature": "0.7; top_p=0.95", "context": "256k" }, "matches_canonical": false, "source_type": "official_model_card", "audit_status": "verified", "notes": "Displayed exactly: '66.9'. Research observation: medium-image1.png:4:1:score:reported. Chart footnotes: *All model evaluations were run with maximum reasoning settings. | **Sonnet 4.6 encountered higher rates of reasoning truncation due to external API restrictions than other models, which affected its performance." }, { "model_id": "mistral-medium-3.5-128b", "benchmark_id": "browsecomp_cm", "score": 48.6, "reference_url": "https://huggingface.co/mistralai/Mistral-Medium-3.5-128B/resolve/22b2b868a15677cfa6061277ed2f653d1349a9ab/images/image2.png?download=true", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "web search/browser with context management and discard-all at 100k tokens", "sampling": "source does not state trial count", "judge": "benchmark-specified", "harness": "Mistral official BrowseComp evaluation", "prompt_style": "official/default", "temperature": "0.7; top_p=0.95", "context": "256k" }, "matches_canonical": false, "source_type": "official_model_card", "audit_status": "verified", "notes": "Displayed exactly: '48.6'. Research observation: medium-image2.png:5:1:score:reported. The companion chart footnote establishes that Medium 3.5 uses context management with discard-all at 100k tokens." }, { "model_id": "mistral-medium-3.5-128b", "benchmark_id": "collie", "score": 95.8, "reference_url": "https://huggingface.co/mistralai/Mistral-Medium-3.5-128B/resolve/22b2b868a15677cfa6061277ed2f653d1349a9ab/images/image1.png?download=true", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1; repeat count undisclosed", "judge": "benchmark-specified", "harness": "Mistral official maximum-reasoning comparison", "prompt_style": "official/default", "temperature": "0.7; top_p=0.95", "context": "256k" }, "matches_canonical": false, "source_type": "official_model_card", "audit_status": "verified", "notes": "Displayed exactly: '95.8'. Research observation: medium-image1.png:3:1:score:reported. Chart footnotes: *All model evaluations were run with maximum reasoning settings. | **Sonnet 4.6 encountered higher rates of reasoning truncation due to external API restrictions than other models, which affected its performance." }, { "model_id": "mistral-medium-3.5-128b", "benchmark_id": "ifbench", "score": 69.0, "reference_url": "https://huggingface.co/mistralai/Mistral-Medium-3.5-128B/resolve/22b2b868a15677cfa6061277ed2f653d1349a9ab/images/image1.png?download=true", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1; repeat count undisclosed", "judge": "benchmark-specified", "harness": "Mistral official maximum-reasoning comparison", "prompt_style": "official/default", "temperature": "0.7; top_p=0.95", "context": "256k" }, "matches_canonical": false, "source_type": "official_model_card", "audit_status": "verified", "notes": "Displayed exactly: '69.0'. Research observation: medium-image1.png:2:1:score:reported. Chart footnotes: *All model evaluations were run with maximum reasoning settings. | **Sonnet 4.6 encountered higher rates of reasoning truncation due to external API restrictions than other models, which affected its performance." }, { "model_id": "mistral-medium-3.5-128b", "benchmark_id": "swe_bench_verified", "score": 77.6, "reference_url": "https://huggingface.co/mistralai/Mistral-Medium-3.5-128B/resolve/22b2b868a15677cfa6061277ed2f653d1349a9ab/images/image3.png?download=true", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "agentic coding scaffold and SWE-bench verifier", "sampling": "source does not state trial count", "judge": "benchmark-specified", "harness": "Mistral self-reported SWE-bench Verified evaluation", "prompt_style": "official/default", "temperature": "0.7; top_p=0.95", "context": "256k" }, "matches_canonical": false, "source_type": "official_model_card", "audit_status": "verified", "notes": "Displayed exactly: '77.6'. Research observation: medium-image3.png:1:1:score:reported." }, { "model_id": "mistral-medium-3.5-128b", "benchmark_id": "tau3_airline", "score": 72.0, "reference_url": "https://huggingface.co/mistralai/Mistral-Medium-3.5-128B/resolve/22b2b868a15677cfa6061277ed2f653d1349a9ab/images/image2.png?download=true", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "τ³ domain tools and retrieval environment", "sampling": "trials=4", "judge": "benchmark-specified", "harness": "Mistral τ³ evaluation with GPT-5.2 low-effort user simulator", "prompt_style": "official/default", "temperature": "0.7; top_p=0.95", "context": "256k" }, "matches_canonical": false, "source_type": "official_model_card", "audit_status": "verified", "notes": "Displayed exactly: '72.0'. Research observation: medium-image2.png:2:1:score:reported." }, { "model_id": "mistral-medium-3.5-128b", "benchmark_id": "tau3_banking", "score": 13.4, "reference_url": "https://huggingface.co/mistralai/Mistral-Medium-3.5-128B/resolve/22b2b868a15677cfa6061277ed2f653d1349a9ab/images/image2.png?download=true", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "τ³ Banking terminal or embedding retrieval; higher of the two reported", "sampling": "trials=4", "judge": "benchmark-specified", "harness": "Mistral τ³ evaluation with GPT-5.2 low-effort user simulator", "prompt_style": "official/default", "temperature": "0.7; top_p=0.95", "context": "256k" }, "matches_canonical": false, "source_type": "official_model_card", "audit_status": "verified", "notes": "Displayed exactly: '13.4'. Research observation: medium-image2.png:4:1:score:reported." }, { "model_id": "mistral-medium-3.5-128b", "benchmark_id": "tau3_retail", "score": 76.1, "reference_url": "https://huggingface.co/mistralai/Mistral-Medium-3.5-128B/resolve/22b2b868a15677cfa6061277ed2f653d1349a9ab/images/image2.png?download=true", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "τ³ domain tools and retrieval environment", "sampling": "trials=4", "judge": "benchmark-specified", "harness": "Mistral τ³ evaluation with GPT-5.2 low-effort user simulator", "prompt_style": "official/default", "temperature": "0.7; top_p=0.95", "context": "256k" }, "matches_canonical": false, "source_type": "official_model_card", "audit_status": "verified", "notes": "Displayed exactly: '76.1'. Research observation: medium-image2.png:3:1:score:reported." }, { "model_id": "mistral-medium-3.5-128b", "benchmark_id": "tau3_telecom", "score": 91.4, "reference_url": "https://huggingface.co/mistralai/Mistral-Medium-3.5-128B/resolve/22b2b868a15677cfa6061277ed2f653d1349a9ab/images/image2.png?download=true", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "τ³ domain tools and retrieval environment", "sampling": "trials=4", "judge": "benchmark-specified", "harness": "Mistral τ³ evaluation with GPT-5.2 low-effort user simulator", "prompt_style": "official/default", "temperature": "0.7; top_p=0.95", "context": "256k" }, "matches_canonical": false, "source_type": "official_model_card", "audit_status": "verified", "notes": "Displayed exactly: '91.4'. Research observation: medium-image2.png:1:1:score:reported." }, { "model_id": "mistral-small-3.2", "benchmark_id": "arena_hard", "score": 43.1, "reference_url": "https://huggingface.co/mistralai/Mistral-Small-4-119B-2603/resolve/a11f36bebf709121056b1dbcc943d1c6afbe494d/images/image3.png?download=true", "reported_setting": { "mode": "non-thinking", "effort": "none", "tools": "none", "sampling": "pass@1; repeat count undisclosed", "judge": "Arena-Hard automatic judge", "harness": "Mistral official model-card evaluation", "prompt_style": "official SYSTEM_PROMPT.txt", "temperature": "0.15 recommended", "context": "128k" }, "matches_canonical": false, "source_type": "official_model_card", "audit_status": "verified", "notes": "Displayed exactly: '43.1'. Research observation: small-image3.png:4:2:score:reported." }, { "model_id": "mistral-small-3.2", "benchmark_id": "gpqa_diamond", "score": 50, "reference_url": "https://huggingface.co/mistralai/Mistral-Small-4-119B-2603/resolve/a11f36bebf709121056b1dbcc943d1c6afbe494d/images/image3.png?download=true", "reported_setting": { "mode": "non-thinking", "effort": "none", "tools": "none", "sampling": "pass@1; repeat count undisclosed", "judge": "benchmark-specified", "harness": "Mistral official model-card evaluation", "prompt_style": "official SYSTEM_PROMPT.txt", "temperature": "0.15 recommended", "context": "128k" }, "matches_canonical": false, "source_type": "official_model_card", "audit_status": "verified", "notes": "Displayed exactly: '50'. Research observation: small-image3.png:1:2:score:reported." }, { "model_id": "mistral-small-3.2", "benchmark_id": "ifbench", "score": 34, "reference_url": "https://huggingface.co/mistralai/Mistral-Small-4-119B-2603/resolve/a11f36bebf709121056b1dbcc943d1c6afbe494d/images/image3.png?download=true", "reported_setting": { "mode": "non-thinking", "effort": "none", "tools": "none", "sampling": "pass@1; repeat count undisclosed", "judge": "benchmark-specified", "harness": "Mistral official model-card evaluation", "prompt_style": "official SYSTEM_PROMPT.txt", "temperature": "0.15 recommended", "context": "128k" }, "matches_canonical": false, "source_type": "official_model_card", "audit_status": "verified", "notes": "Displayed exactly: '34'. Research observation: small-image3.png:3:2:score:reported." }, { "model_id": "mistral-small-3.2", "benchmark_id": "mmlu_pro", "score": 69, "reference_url": "https://huggingface.co/mistralai/Mistral-Small-4-119B-2603/resolve/a11f36bebf709121056b1dbcc943d1c6afbe494d/images/image3.png?download=true", "reported_setting": { "mode": "non-thinking", "effort": "none", "tools": "none", "sampling": "pass@1; repeat count undisclosed", "judge": "benchmark-specified", "harness": "Mistral official model-card evaluation", "prompt_style": "official SYSTEM_PROMPT.txt", "temperature": "0.15 recommended", "context": "128k" }, "matches_canonical": false, "source_type": "official_model_card", "audit_status": "verified", "notes": "Displayed exactly: '69'. Research observation: small-image3.png:2:2:score:reported." }, { "model_id": "mistral-small-3.2", "benchmark_id": "mmmu_pro", "score": 49.1, "reference_url": "https://huggingface.co/mistralai/Mistral-Small-4-119B-2603/resolve/a11f36bebf709121056b1dbcc943d1c6afbe494d/images/image3.png?download=true", "reported_setting": { "mode": "non-thinking", "effort": "none", "tools": "none", "sampling": "pass@1; repeat count undisclosed", "judge": "benchmark-specified", "harness": "Mistral official model-card evaluation", "prompt_style": "official SYSTEM_PROMPT.txt", "temperature": "0.15 recommended", "context": "128k" }, "matches_canonical": false, "source_type": "official_model_card", "audit_status": "verified", "notes": "Displayed exactly: '49.1'. Research observation: small-image3.png:5:2:score:reported." }, { "model_id": "mistral-small-4-119b-2603", "benchmark_id": "aa_lcr_mistral_custom_gpt_4_1_mini_middle_out", "score": 71.2, "reference_url": "https://huggingface.co/mistralai/Mistral-Small-4-119B-2603/resolve/a11f36bebf709121056b1dbcc943d1c6afbe494d/images/image2.png?download=true", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1; repeat count undisclosed by Mistral", "judge": "gpt-4.1-mini-2025-04-14 equality judge", "harness": "Mistral custom AA-LCR with middle-out context handling", "prompt_style": "official/default", "temperature": "0.7", "context": "256k" }, "matches_canonical": false, "source_type": "official_model_card", "audit_status": "verified", "notes": "Displayed exactly: '71.2'. Research observation: small-image2.png:1:1:score:high.", "candidates": [ { "score": 72, "reference_url": "https://huggingface.co/mistralai/Mistral-Small-4-119B-2603/resolve/a11f36bebf709121056b1dbcc943d1c6afbe494d/images/aime.png?download=true", "source_type": "official_model_card", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1; repeat count undisclosed by Mistral", "judge": "gpt-4.1-mini-2025-04-14 equality judge", "harness": "Mistral custom AA-LCR with middle-out context handling", "prompt_style": "official/default", "temperature": "0.7", "context": "256k" }, "notes": "Displayed exactly: '72'. Research observation: small-aime.png:1:1:score:reasoning. Chart footnotes: *Custom implementation of AA LCR using gpt-4.1-mini-2025-04-14 as a judge and a middle-out approach for models with shorter context lengths." }, { "score": 44, "reference_url": "https://huggingface.co/mistralai/Mistral-Small-4-119B-2603/resolve/a11f36bebf709121056b1dbcc943d1c6afbe494d/images/aime.png?download=true", "source_type": "official_model_card", "reported_setting": { "mode": "non-thinking", "effort": "none", "tools": "none", "sampling": "pass@1; repeat count undisclosed by Mistral", "judge": "gpt-4.1-mini-2025-04-14 equality judge", "harness": "Mistral custom AA-LCR with middle-out context handling", "prompt_style": "official/default", "temperature": "0.0-0.7 task-dependent", "context": "256k" }, "notes": "Displayed exactly: '44'. Research observation: small-aime.png:1:1:score:instruct. Chart footnotes: *Custom implementation of AA LCR using gpt-4.1-mini-2025-04-14 as a judge and a middle-out approach for models with shorter context lengths." } ] }, { "model_id": "mistral-small-4-119b-2603", "benchmark_id": "aime_2025", "score": 83.8, "reference_url": "https://huggingface.co/mistralai/Mistral-Small-4-119B-2603/resolve/a11f36bebf709121056b1dbcc943d1c6afbe494d/images/image2.png?download=true", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1; repeat count undisclosed", "judge": "benchmark-specified", "harness": "Mistral official model-card evaluation", "prompt_style": "official/default", "temperature": "0.7", "context": "256k" }, "matches_canonical": false, "source_type": "official_model_card", "audit_status": "verified", "notes": "Displayed exactly: '83.8'. Research observation: small-image2.png:2:1:score:high.", "candidates": [ { "score": 84, "reference_url": "https://huggingface.co/mistralai/Mistral-Small-4-119B-2603/resolve/a11f36bebf709121056b1dbcc943d1c6afbe494d/images/livecode.png?download=true", "source_type": "official_model_card", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1; repeat count undisclosed", "judge": "benchmark-specified", "harness": "Mistral official model-card evaluation", "prompt_style": "official/default", "temperature": "0.7", "context": "256k" }, "notes": "Displayed exactly: '84'. Research observation: small-livecode.png:1:1:score:reasoning." }, { "score": 36, "reference_url": "https://huggingface.co/mistralai/Mistral-Small-4-119B-2603/resolve/a11f36bebf709121056b1dbcc943d1c6afbe494d/images/livecode.png?download=true", "source_type": "official_model_card", "reported_setting": { "mode": "non-thinking", "effort": "none", "tools": "none", "sampling": "pass@1; repeat count undisclosed", "judge": "benchmark-specified", "harness": "Mistral official model-card evaluation", "prompt_style": "official/default", "temperature": "0.0-0.7 task-dependent", "context": "256k" }, "notes": "Displayed exactly: '36'. Research observation: small-livecode.png:1:1:score:instruct." } ] }, { "model_id": "mistral-small-4-119b-2603", "benchmark_id": "arena_hard", "score": 58.3, "reference_url": "https://huggingface.co/mistralai/Mistral-Small-4-119B-2603/resolve/a11f36bebf709121056b1dbcc943d1c6afbe494d/images/image3.png?download=true", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1; repeat count undisclosed", "judge": "Arena-Hard automatic judge", "harness": "Mistral official model-card evaluation", "prompt_style": "official/default", "temperature": "0.7", "context": "256k" }, "matches_canonical": false, "source_type": "official_model_card", "audit_status": "verified", "notes": "Displayed exactly: '58.3'. Research observation: small-image3.png:4:1:score:reasoning.", "candidates": [ { "score": 55.8, "reference_url": "https://huggingface.co/mistralai/Mistral-Small-4-119B-2603/resolve/a11f36bebf709121056b1dbcc943d1c6afbe494d/images/image3.png?download=true", "source_type": "official_model_card", "reported_setting": { "mode": "non-thinking", "effort": "none", "tools": "none", "sampling": "pass@1; repeat count undisclosed", "judge": "Arena-Hard automatic judge", "harness": "Mistral official model-card evaluation", "prompt_style": "official/default", "temperature": "0.0-0.7 task-dependent", "context": "256k" }, "notes": "Displayed exactly: '55.8'. Research observation: small-image3.png:4:1:score:instruct." } ] }, { "model_id": "mistral-small-4-119b-2603", "benchmark_id": "browsecomp", "score": 21.3, "reference_url": "https://huggingface.co/mistralai/Mistral-Medium-3.5-128B/resolve/22b2b868a15677cfa6061277ed2f653d1349a9ab/images/image2.png?download=true", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "web search/browser environment", "sampling": "source does not state trial count", "judge": "benchmark-specified", "harness": "Mistral official BrowseComp evaluation", "prompt_style": "official/default", "temperature": "0.7", "context": "256k" }, "matches_canonical": false, "source_type": "official_model_card", "audit_status": "verified", "notes": "Displayed exactly: '21.3'. Research observation: medium-image2.png:5:3:score:reported." }, { "model_id": "mistral-small-4-119b-2603", "benchmark_id": "collie", "score": 62.9, "reference_url": "https://huggingface.co/mistralai/Mistral-Small-4-119B-2603/resolve/a11f36bebf709121056b1dbcc943d1c6afbe494d/images/image2.png?download=true", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1; repeat count undisclosed", "judge": "benchmark-specified", "harness": "Mistral official model-card evaluation", "prompt_style": "official/default", "temperature": "0.7", "context": "256k" }, "matches_canonical": false, "source_type": "official_model_card", "audit_status": "verified", "notes": "Displayed exactly: '62.9'. Research observation: small-image2.png:3:1:score:high." }, { "model_id": "mistral-small-4-119b-2603", "benchmark_id": "gpqa_diamond", "score": 71.2, "reference_url": "https://huggingface.co/mistralai/Mistral-Small-4-119B-2603/resolve/a11f36bebf709121056b1dbcc943d1c6afbe494d/images/image3.png?download=true", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1; repeat count undisclosed", "judge": "benchmark-specified", "harness": "Mistral official model-card evaluation", "prompt_style": "official/default", "temperature": "0.7", "context": "256k" }, "matches_canonical": false, "source_type": "official_model_card", "audit_status": "verified", "notes": "Displayed exactly: '71.2'. Research observation: small-image3.png:1:1:score:reasoning.", "candidates": [ { "score": 59.1, "reference_url": "https://huggingface.co/mistralai/Mistral-Small-4-119B-2603/resolve/a11f36bebf709121056b1dbcc943d1c6afbe494d/images/image3.png?download=true", "source_type": "official_model_card", "reported_setting": { "mode": "non-thinking", "effort": "none", "tools": "none", "sampling": "pass@1; repeat count undisclosed", "judge": "benchmark-specified", "harness": "Mistral official model-card evaluation", "prompt_style": "official/default", "temperature": "0.0-0.7 task-dependent", "context": "256k" }, "notes": "Displayed exactly: '59.1'. Research observation: small-image3.png:1:1:score:instruct." } ] }, { "model_id": "mistral-small-4-119b-2603", "benchmark_id": "ifbench", "score": 48, "reference_url": "https://huggingface.co/mistralai/Mistral-Small-4-119B-2603/resolve/a11f36bebf709121056b1dbcc943d1c6afbe494d/images/image3.png?download=true", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1; repeat count undisclosed", "judge": "benchmark-specified", "harness": "Mistral official model-card evaluation", "prompt_style": "official/default", "temperature": "0.7", "context": "256k" }, "matches_canonical": false, "source_type": "official_model_card", "audit_status": "verified", "notes": "Displayed exactly: '48'. Research observation: small-image3.png:3:1:score:reasoning.", "candidates": [ { "score": 35.7, "reference_url": "https://huggingface.co/mistralai/Mistral-Small-4-119B-2603/resolve/a11f36bebf709121056b1dbcc943d1c6afbe494d/images/image3.png?download=true", "source_type": "official_model_card", "reported_setting": { "mode": "non-thinking", "effort": "none", "tools": "none", "sampling": "pass@1; repeat count undisclosed", "judge": "benchmark-specified", "harness": "Mistral official model-card evaluation", "prompt_style": "official/default", "temperature": "0.0-0.7 task-dependent", "context": "256k" }, "notes": "Displayed exactly: '35.7'. Research observation: small-image3.png:3:1:score:instruct." } ] }, { "model_id": "mistral-small-4-119b-2603", "benchmark_id": "mmlu_pro", "score": 78, "reference_url": "https://huggingface.co/mistralai/Mistral-Small-4-119B-2603/resolve/a11f36bebf709121056b1dbcc943d1c6afbe494d/images/image3.png?download=true", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1; repeat count undisclosed", "judge": "benchmark-specified", "harness": "Mistral official model-card evaluation", "prompt_style": "official/default", "temperature": "0.7", "context": "256k" }, "matches_canonical": false, "source_type": "official_model_card", "audit_status": "verified", "notes": "Displayed exactly: '78'. Research observation: small-image3.png:2:1:score:reasoning.", "candidates": [ { "score": 73.5, "reference_url": "https://huggingface.co/mistralai/Mistral-Small-4-119B-2603/resolve/a11f36bebf709121056b1dbcc943d1c6afbe494d/images/image3.png?download=true", "source_type": "official_model_card", "reported_setting": { "mode": "non-thinking", "effort": "none", "tools": "none", "sampling": "pass@1; repeat count undisclosed", "judge": "benchmark-specified", "harness": "Mistral official model-card evaluation", "prompt_style": "official/default", "temperature": "0.0-0.7 task-dependent", "context": "256k" }, "notes": "Displayed exactly: '73.5'. Research observation: small-image3.png:2:1:score:instruct." } ] }, { "model_id": "mistral-small-4-119b-2603", "benchmark_id": "mmmu_pro", "score": 60, "reference_url": "https://huggingface.co/mistralai/Mistral-Small-4-119B-2603/resolve/a11f36bebf709121056b1dbcc943d1c6afbe494d/images/image3.png?download=true", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "none", "sampling": "pass@1; repeat count undisclosed", "judge": "benchmark-specified", "harness": "Mistral official model-card evaluation", "prompt_style": "official/default", "temperature": "0.7", "context": "256k" }, "matches_canonical": false, "source_type": "official_model_card", "audit_status": "verified", "notes": "Displayed exactly: '60'. Research observation: small-image3.png:5:1:score:reasoning.", "candidates": [ { "score": 46.3, "reference_url": "https://huggingface.co/mistralai/Mistral-Small-4-119B-2603/resolve/a11f36bebf709121056b1dbcc943d1c6afbe494d/images/image3.png?download=true", "source_type": "official_model_card", "reported_setting": { "mode": "non-thinking", "effort": "none", "tools": "none", "sampling": "pass@1; repeat count undisclosed", "judge": "benchmark-specified", "harness": "Mistral official model-card evaluation", "prompt_style": "official/default", "temperature": "0.0-0.7 task-dependent", "context": "256k" }, "notes": "Displayed exactly: '46.3'. Research observation: small-image3.png:5:1:score:instruct." } ] }, { "model_id": "mistral-small-4-119b-2603", "benchmark_id": "tau3_airline", "score": 38.5, "reference_url": "https://huggingface.co/mistralai/Mistral-Medium-3.5-128B/resolve/22b2b868a15677cfa6061277ed2f653d1349a9ab/images/image2.png?download=true", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "τ³ domain tools and retrieval environment", "sampling": "source does not state trial count", "judge": "benchmark-specified", "harness": "Mistral official τ³ evaluation", "prompt_style": "official/default", "temperature": "0.7", "context": "256k" }, "matches_canonical": false, "source_type": "official_model_card", "audit_status": "verified", "notes": "Displayed exactly: '38.5'. Research observation: medium-image2.png:2:3:score:reported." }, { "model_id": "mistral-small-4-119b-2603", "benchmark_id": "tau3_banking", "score": 7.0, "reference_url": "https://huggingface.co/mistralai/Mistral-Medium-3.5-128B/resolve/22b2b868a15677cfa6061277ed2f653d1349a9ab/images/image2.png?download=true", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "τ³ domain tools and retrieval environment", "sampling": "source does not state trial count", "judge": "benchmark-specified", "harness": "Mistral official τ³ evaluation", "prompt_style": "official/default", "temperature": "0.7", "context": "256k" }, "matches_canonical": false, "source_type": "official_model_card", "audit_status": "verified", "notes": "Displayed exactly: '7.0'. Research observation: medium-image2.png:4:3:score:reported." }, { "model_id": "mistral-small-4-119b-2603", "benchmark_id": "tau3_retail", "score": 67.8, "reference_url": "https://huggingface.co/mistralai/Mistral-Medium-3.5-128B/resolve/22b2b868a15677cfa6061277ed2f653d1349a9ab/images/image2.png?download=true", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "τ³ domain tools and retrieval environment", "sampling": "source does not state trial count", "judge": "benchmark-specified", "harness": "Mistral official τ³ evaluation", "prompt_style": "official/default", "temperature": "0.7", "context": "256k" }, "matches_canonical": false, "source_type": "official_model_card", "audit_status": "verified", "notes": "Displayed exactly: '67.8'. Research observation: medium-image2.png:3:3:score:reported." }, { "model_id": "mistral-small-4-119b-2603", "benchmark_id": "tau3_telecom", "score": 47.1, "reference_url": "https://huggingface.co/mistralai/Mistral-Medium-3.5-128B/resolve/22b2b868a15677cfa6061277ed2f653d1349a9ab/images/image2.png?download=true", "reported_setting": { "mode": "thinking", "effort": "high", "tools": "τ³ domain tools and retrieval environment", "sampling": "source does not state trial count", "judge": "benchmark-specified", "harness": "Mistral official τ³ evaluation", "prompt_style": "official/default", "temperature": "0.7", "context": "256k" }, "matches_canonical": false, "source_type": "official_model_card", "audit_status": "verified", "notes": "Displayed exactly: '47.1'. Research observation: medium-image2.png:1:3:score:reported." }, { "model_id": "qwen3-next-80b-a3b-instruct", "benchmark_id": "aa_lcr_mistral_custom_gpt_4_1_mini_middle_out", "score": 62, "reference_url": "https://huggingface.co/mistralai/Mistral-Small-4-119B-2603/resolve/a11f36bebf709121056b1dbcc943d1c6afbe494d/images/aime.png?download=true", "reported_setting": { "mode": "non-thinking", "effort": "none", "tools": "none", "sampling": "pass@1; repeat count undisclosed by Mistral", "judge": "gpt-4.1-mini-2025-04-14 equality judge", "harness": "Mistral custom AA-LCR with middle-out context handling", "prompt_style": "official; benchmark output-format prompts", "temperature": "0.7; top_p=0.8; top_k=20; min_p=0", "context": "262,144 native; approximately 1,010,000 with official extension" }, "matches_canonical": false, "source_type": "official_model_card", "audit_status": "verified", "notes": "Displayed exactly: '62'. Research observation: small-aime.png:1:4:score:instruct. Chart footnotes: *Custom implementation of AA LCR using gpt-4.1-mini-2025-04-14 as a judge and a middle-out approach for models with shorter context lengths." }, { "model_id": "qwen3-next-80b-a3b-instruct", "benchmark_id": "aime_2025", "score": 65, "reference_url": "https://huggingface.co/mistralai/Mistral-Small-4-119B-2603/resolve/a11f36bebf709121056b1dbcc943d1c6afbe494d/images/livecode.png?download=true", "reported_setting": { "mode": "non-thinking", "effort": "none", "tools": "none", "sampling": "pass@1; repeat count undisclosed", "judge": "benchmark-specified", "harness": "Mistral official model-card evaluation", "prompt_style": "official; benchmark output-format prompts", "temperature": "0.7; top_p=0.8; top_k=20; min_p=0", "context": "262,144 native; approximately 1,010,000 with official extension" }, "matches_canonical": false, "source_type": "official_model_card", "audit_status": "verified", "notes": "Displayed exactly: '65'. Research observation: small-livecode.png:1:4:score:instruct." }, { "model_id": "qwen3-next-80b-a3b-thinking", "benchmark_id": "aa_lcr_mistral_custom_gpt_4_1_mini_middle_out", "score": 75, "reference_url": "https://huggingface.co/mistralai/Mistral-Small-4-119B-2603/resolve/a11f36bebf709121056b1dbcc943d1c6afbe494d/images/aime.png?download=true", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1; repeat count undisclosed by Mistral", "judge": "gpt-4.1-mini-2025-04-14 equality judge", "harness": "Mistral custom AA-LCR with middle-out context handling", "prompt_style": "official; benchmark output-format prompts", "temperature": "0.6; top_p=0.95; top_k=20; min_p=0", "context": "262,144 native; approximately 1,010,000 with official extension" }, "matches_canonical": false, "source_type": "official_model_card", "audit_status": "verified", "notes": "Displayed exactly: '75'. Research observation: small-aime.png:1:4:score:reasoning. Chart footnotes: *Custom implementation of AA LCR using gpt-4.1-mini-2025-04-14 as a judge and a middle-out approach for models with shorter context lengths." }, { "model_id": "qwen3-next-80b-a3b-thinking", "benchmark_id": "aime_2025", "score": 88, "reference_url": "https://huggingface.co/mistralai/Mistral-Small-4-119B-2603/resolve/a11f36bebf709121056b1dbcc943d1c6afbe494d/images/livecode.png?download=true", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1; repeat count undisclosed", "judge": "benchmark-specified", "harness": "Mistral official model-card evaluation", "prompt_style": "official; benchmark output-format prompts", "temperature": "0.6; top_p=0.95; top_k=20; min_p=0", "context": "262,144 native; approximately 1,010,000 with official extension" }, "matches_canonical": false, "source_type": "official_model_card", "audit_status": "verified", "notes": "Displayed exactly: '88'. Research observation: small-livecode.png:1:4:score:reasoning." }, { "model_id": "qwen3.5-122b-a10b", "benchmark_id": "aa_lcr_mistral_custom_gpt_4_1_mini_middle_out", "score": 84, "reference_url": "https://huggingface.co/mistralai/Mistral-Small-4-119B-2603/resolve/a11f36bebf709121056b1dbcc943d1c6afbe494d/images/aime.png?download=true", "reported_setting": { "mode": "non-thinking", "effort": "none", "tools": "none", "sampling": "pass@1; repeat count undisclosed by Mistral", "judge": "gpt-4.1-mini-2025-04-14 equality judge", "harness": "Mistral custom AA-LCR with middle-out context handling", "prompt_style": "official", "temperature": "1.0; top_p=0.95; top_k=20; presence_penalty=1.5", "context": "262,144 native" }, "matches_canonical": false, "source_type": "official_model_card", "audit_status": "verified", "notes": "Displayed exactly: '84'. Research observation: small-aime.png:1:5:score:instruct. Chart footnotes: *Custom implementation of AA LCR using gpt-4.1-mini-2025-04-14 as a judge and a middle-out approach for models with shorter context lengths." }, { "model_id": "qwen3.5-122b-a10b", "benchmark_id": "aime_2025", "score": 93, "reference_url": "https://huggingface.co/mistralai/Mistral-Small-4-119B-2603/resolve/a11f36bebf709121056b1dbcc943d1c6afbe494d/images/livecode.png?download=true", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1; repeat count undisclosed", "judge": "benchmark-specified", "harness": "Mistral official model-card evaluation", "prompt_style": "official", "temperature": "1.0; top_p=0.95; top_k=20; presence_penalty=1.5", "context": "262,144 native" }, "matches_canonical": false, "source_type": "official_model_card", "audit_status": "verified", "notes": "Displayed exactly: '93'. Research observation: small-livecode.png:1:5:score:reasoning.", "candidates": [ { "score": 80, "reference_url": "https://huggingface.co/mistralai/Mistral-Small-4-119B-2603/resolve/a11f36bebf709121056b1dbcc943d1c6afbe494d/images/livecode.png?download=true", "source_type": "official_model_card", "reported_setting": { "mode": "non-thinking", "effort": "none", "tools": "none", "sampling": "pass@1; repeat count undisclosed", "judge": "benchmark-specified", "harness": "Mistral official model-card evaluation", "prompt_style": "official", "temperature": "1.0; top_p=0.95; top_k=20; presence_penalty=1.5", "context": "262,144 native" }, "notes": "Displayed exactly: '80'. Research observation: small-livecode.png:1:5:score:instruct." } ] }, { "model_id": "qwen3.5-397b", "benchmark_id": "beyond_aime", "score": 72.3, "reference_url": "https://huggingface.co/mistralai/Mistral-Medium-3.5-128B/resolve/22b2b868a15677cfa6061277ed2f653d1349a9ab/images/image1.png?download=true", "reported_setting": { "mode": "thinking", "effort": "default (maximum reasoning setting)", "tools": "none", "sampling": "samples=16 (reported as avg@16)", "judge": "benchmark-specified", "harness": "Mistral official maximum-reasoning comparison", "prompt_style": "default (thinking by default)", "temperature": "default", "context": "default (262K)" }, "matches_canonical": false, "source_type": "official_model_card", "audit_status": "verified", "notes": "Displayed exactly: '72.3'. Research observation: medium-image1.png:4:5:score:reported. Chart footnotes: *All model evaluations were run with maximum reasoning settings. | **Sonnet 4.6 encountered higher rates of reasoning truncation due to external API restrictions than other models, which affected its performance." }, { "model_id": "qwen3.5-397b", "benchmark_id": "collie", "score": 88.9, "reference_url": "https://huggingface.co/mistralai/Mistral-Medium-3.5-128B/resolve/22b2b868a15677cfa6061277ed2f653d1349a9ab/images/image1.png?download=true", "reported_setting": { "mode": "thinking", "effort": "default (maximum reasoning setting)", "tools": "none", "sampling": "pass@1; repeat count undisclosed", "judge": "benchmark-specified", "harness": "Mistral official maximum-reasoning comparison", "prompt_style": "default (thinking by default)", "temperature": "default", "context": "default (262K)" }, "matches_canonical": false, "source_type": "official_model_card", "audit_status": "verified", "notes": "Displayed exactly: '88.9'. Research observation: medium-image1.png:3:5:score:reported. Chart footnotes: *All model evaluations were run with maximum reasoning settings. | **Sonnet 4.6 encountered higher rates of reasoning truncation due to external API restrictions than other models, which affected its performance." }, { "model_id": "gemma-3-1b-it", "benchmark_id": "bfcl_v4", "score": 7.17, "reference_url": "https://www.liquid.ai/blog/lfm2-5-230m", "reported_setting": { "mode": "non-thinking", "effort": "source does not state", "tools": "benchmark-provided tools", "sampling": "source does not state trial count", "judge": "benchmark-specified", "harness": "Liquid AI official comparative evaluation", "prompt_style": "source does not state", "temperature": "source does not state evaluation parameters", "context": "source does not state evaluation context" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Displayed exactly as 7.17. Official Liquid AI reported result." }, { "model_id": "gemma-3-1b-it", "benchmark_id": "tau2_bench_avg", "score": 20.6, "reference_url": "https://aypchzzf9pftwuto.public.blob.vercel-storage.com/lfm2_5_230m_benchmarks-EcJisF1zm7I7KiST75d7XsE2LwE0lD.png", "reported_setting": { "mode": "non-thinking", "effort": "source does not state", "tools": "benchmark-provided tools", "sampling": "source does not state trial count", "judge": "benchmark-specified", "harness": "Liquid AI official comparative evaluation", "prompt_style": "source does not state", "temperature": "source does not state evaluation parameters", "context": "source does not state evaluation context" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Displayed exactly as 20.6. Official Liquid AI reported result." }, { "model_id": "gemma-3-1b-it", "benchmark_id": "tau2_bench_retail", "score": 6.43, "reference_url": "https://www.liquid.ai/blog/lfm2-5-230m", "reported_setting": { "mode": "non-thinking", "effort": "source does not state", "tools": "benchmark-provided tools", "sampling": "source does not state trial count", "judge": "benchmark-specified", "harness": "Liquid AI official comparative evaluation", "prompt_style": "source does not state", "temperature": "source does not state evaluation parameters", "context": "source does not state evaluation context" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Displayed exactly as 6.43. Official Liquid AI reported result." }, { "model_id": "gemma-3-1b-it", "benchmark_id": "tau2_bench_telecom", "score": 9.36, "reference_url": "https://www.liquid.ai/blog/lfm2-5-230m", "reported_setting": { "mode": "non-thinking", "effort": "source does not state", "tools": "benchmark-provided tools", "sampling": "source does not state trial count", "judge": "benchmark-specified", "harness": "Liquid AI official comparative evaluation", "prompt_style": "source does not state", "temperature": "source does not state evaluation parameters", "context": "source does not state evaluation context" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Displayed exactly as 9.36. Official Liquid AI reported result." }, { "model_id": "gemma-4-26b-a4b", "benchmark_id": "aa_omniscience_accuracy", "score": 14.37, "reference_url": "https://www.liquid.ai/blog/lfm2-5-8b-a1b", "reported_setting": { "mode": "source/model default", "effort": "source does not state", "tools": "none", "sampling": "source does not state trial count", "judge": "official answer and abstention classification", "harness": "Artificial Analysis Omniscience evaluation reported by Liquid", "prompt_style": "source does not state", "temperature": "source does not state evaluation parameters", "context": "source does not state evaluation context" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Displayed exactly as 14.37. Official Liquid AI reported result." }, { "model_id": "gemma-4-26b-a4b", "benchmark_id": "aa_omniscience_index", "score": -62.07, "reference_url": "https://www.liquid.ai/blog/lfm2-5-8b-a1b", "reported_setting": { "mode": "source/model default", "effort": "source does not state", "tools": "none", "sampling": "source does not state trial count", "judge": "official answer and abstention classification", "harness": "Artificial Analysis Omniscience evaluation reported by Liquid", "prompt_style": "source does not state", "temperature": "source does not state evaluation parameters", "context": "source does not state evaluation context" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Displayed exactly as -62.07. Official Liquid AI reported result." }, { "model_id": "gemma-4-26b-a4b", "benchmark_id": "aa_omniscience_non_hallucination", "score": 10.75, "reference_url": "https://www.liquid.ai/blog/lfm2-5-8b-a1b", "reported_setting": { "mode": "source/model default", "effort": "source does not state", "tools": "none", "sampling": "source does not state trial count", "judge": "official answer and abstention classification", "harness": "Artificial Analysis Omniscience evaluation reported by Liquid", "prompt_style": "source does not state", "temperature": "source does not state evaluation parameters", "context": "source does not state evaluation context" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Displayed exactly as 10.75. Official Liquid AI reported result." }, { "model_id": "gemma-4-26b-a4b", "benchmark_id": "aime_2025", "score": 68.67, "reference_url": "https://www.liquid.ai/blog/lfm2-5-8b-a1b", "reported_setting": { "mode": "source/model default", "effort": "source does not state", "tools": "none", "sampling": "source does not state trial count", "judge": "benchmark-specified", "harness": "Liquid AI official comparative evaluation", "prompt_style": "source does not state", "temperature": "source does not state evaluation parameters", "context": "source does not state evaluation context" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Displayed exactly as 68.67. Official Liquid AI reported result." }, { "model_id": "gemma-4-26b-a4b", "benchmark_id": "bfcl_v3", "score": 68.87, "reference_url": "https://www.liquid.ai/blog/lfm2-5-8b-a1b", "reported_setting": { "mode": "source/model default", "effort": "source does not state", "tools": "benchmark-provided tools", "sampling": "source does not state trial count", "judge": "benchmark-specified", "harness": "Liquid AI official comparative evaluation", "prompt_style": "source does not state", "temperature": "source does not state evaluation parameters", "context": "source does not state evaluation context" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Displayed exactly as 68.87. Official Liquid AI reported result." }, { "model_id": "gemma-4-26b-a4b", "benchmark_id": "bfcl_v4", "score": 55.87, "reference_url": "https://www.liquid.ai/blog/lfm2-5-8b-a1b", "reported_setting": { "mode": "source/model default", "effort": "source does not state", "tools": "benchmark-provided tools", "sampling": "source does not state trial count", "judge": "benchmark-specified", "harness": "Liquid AI official comparative evaluation", "prompt_style": "source does not state", "temperature": "source does not state evaluation parameters", "context": "source does not state evaluation context" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Displayed exactly as 55.87. Official Liquid AI reported result." }, { "model_id": "gemma-4-26b-a4b", "benchmark_id": "math_500", "score": 94.2, "reference_url": "https://www.liquid.ai/blog/lfm2-5-8b-a1b", "reported_setting": { "mode": "source/model default", "effort": "source does not state", "tools": "none", "sampling": "source does not state trial count", "judge": "benchmark-specified", "harness": "Liquid AI official comparative evaluation", "prompt_style": "source does not state", "temperature": "source does not state evaluation parameters", "context": "source does not state evaluation context" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Displayed exactly as 94.20. Official Liquid AI reported result." }, { "model_id": "gemma-4-26b-a4b", "benchmark_id": "multi_if", "score": 82.06, "reference_url": "https://www.liquid.ai/blog/lfm2-5-8b-a1b", "reported_setting": { "mode": "source/model default", "effort": "source does not state", "tools": "none", "sampling": "source does not state trial count", "judge": "benchmark-specified", "harness": "Liquid AI official comparative evaluation", "prompt_style": "source does not state", "temperature": "source does not state evaluation parameters", "context": "source does not state evaluation context" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Displayed exactly as 82.06. Official Liquid AI reported result." }, { "model_id": "gemma-4-e2b", "benchmark_id": "aa_omniscience_accuracy", "score": 7.0, "reference_url": "https://www.liquid.ai/blog/lfm2-5-8b-a1b", "reported_setting": { "mode": "source/model default", "effort": "source does not state", "tools": "none", "sampling": "source does not state trial count", "judge": "official answer and abstention classification", "harness": "Artificial Analysis Omniscience evaluation reported by Liquid", "prompt_style": "source does not state", "temperature": "source does not state evaluation parameters", "context": "source does not state evaluation context" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Displayed exactly as 7.00. Official Liquid AI reported result." }, { "model_id": "gemma-4-e2b", "benchmark_id": "aa_omniscience_index", "score": -72.0, "reference_url": "https://www.liquid.ai/blog/lfm2-5-8b-a1b", "reported_setting": { "mode": "source/model default", "effort": "source does not state", "tools": "none", "sampling": "source does not state trial count", "judge": "official answer and abstention classification", "harness": "Artificial Analysis Omniscience evaluation reported by Liquid", "prompt_style": "source does not state", "temperature": "source does not state evaluation parameters", "context": "source does not state evaluation context" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Displayed exactly as -72. Official Liquid AI reported result." }, { "model_id": "gemma-4-e2b", "benchmark_id": "aa_omniscience_non_hallucination", "score": 15.05, "reference_url": "https://www.liquid.ai/blog/lfm2-5-8b-a1b", "reported_setting": { "mode": "source/model default", "effort": "source does not state", "tools": "none", "sampling": "source does not state trial count", "judge": "official answer and abstention classification", "harness": "Artificial Analysis Omniscience evaluation reported by Liquid", "prompt_style": "source does not state", "temperature": "source does not state evaluation parameters", "context": "source does not state evaluation context" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Displayed exactly as 15.05. Official Liquid AI reported result." }, { "model_id": "gemma-4-e2b", "benchmark_id": "aa_omniscience_public_accuracy", "score": 6.37, "reference_url": "https://huggingface.co/LiquidAI/LFM2.5-2.6B/resolve/ab00687315bc1298e9d54e9c4b611dde9867ccc2/README.md", "reported_setting": { "mode": "source/model default", "effort": "source does not state", "tools": "none", "sampling": "source does not state trial count", "judge": "official answer and abstention classification", "harness": "Artificial Analysis Omniscience evaluation reported by Liquid", "prompt_style": "source does not state", "temperature": "source does not state evaluation parameters", "context": "source does not state evaluation context" }, "matches_canonical": false, "source_type": "official_model_card", "audit_status": "verified", "notes": "Displayed exactly as 6.37. Official Liquid AI reported result." }, { "model_id": "gemma-4-e2b", "benchmark_id": "aa_omniscience_public_index", "score": -74.47, "reference_url": "https://www.liquid.ai/blog/lfm2-5-2-6b", "reported_setting": { "mode": "source/model default", "effort": "source does not state", "tools": "none", "sampling": "source does not state trial count", "judge": "official answer and abstention classification", "harness": "Artificial Analysis Omniscience evaluation reported by Liquid", "prompt_style": "source does not state", "temperature": "source does not state evaluation parameters", "context": "source does not state evaluation context" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Displayed exactly as -74.47. Official Liquid AI reported result." }, { "model_id": "gemma-4-e2b", "benchmark_id": "aa_omniscience_public_non_hallucination", "score": 13.67, "reference_url": "https://huggingface.co/LiquidAI/LFM2.5-2.6B/resolve/ab00687315bc1298e9d54e9c4b611dde9867ccc2/README.md", "reported_setting": { "mode": "source/model default", "effort": "source does not state", "tools": "none", "sampling": "source does not state trial count", "judge": "official answer and abstention classification", "harness": "Artificial Analysis Omniscience evaluation reported by Liquid", "prompt_style": "source does not state", "temperature": "source does not state evaluation parameters", "context": "source does not state evaluation context" }, "matches_canonical": false, "source_type": "official_model_card", "audit_status": "verified", "notes": "Displayed exactly as 13.67. Official Liquid AI reported result." }, { "model_id": "gemma-4-e2b", "benchmark_id": "aime_2025", "score": 26.0, "reference_url": "https://www.liquid.ai/blog/lfm2-5-8b-a1b", "reported_setting": { "mode": "source/model default", "effort": "source does not state", "tools": "none", "sampling": "source does not state trial count", "judge": "benchmark-specified", "harness": "Liquid AI official comparative evaluation", "prompt_style": "source does not state", "temperature": "source does not state evaluation parameters", "context": "source does not state evaluation context" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Displayed exactly as 26. Official Liquid AI reported result.", "candidates": [ { "score": 26.33, "reference_url": "https://www.liquid.ai/blog/lfm2-5-2-6b", "source_type": "official_blog", "reported_setting": { "mode": "source/model default", "effort": "source does not state", "tools": "none", "sampling": "source does not state trial count", "judge": "benchmark-specified", "harness": "Liquid AI official comparative evaluation", "prompt_style": "source does not state", "temperature": "source does not state evaluation parameters", "context": "source does not state evaluation context" }, "notes": "Displayed exactly as 26.33. Official Liquid AI reported result." } ] }, { "model_id": "gemma-4-e2b", "benchmark_id": "bfcl_v3", "score": 56.44, "reference_url": "https://www.liquid.ai/blog/lfm2-5-8b-a1b", "reported_setting": { "mode": "source/model default", "effort": "source does not state", "tools": "benchmark-provided tools", "sampling": "source does not state trial count", "judge": "benchmark-specified", "harness": "Liquid AI official comparative evaluation", "prompt_style": "source does not state", "temperature": "source does not state evaluation parameters", "context": "source does not state evaluation context" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Displayed exactly as 56.44. Official Liquid AI reported result." }, { "model_id": "gemma-4-e2b", "benchmark_id": "bfcl_v4", "score": 31.91, "reference_url": "https://www.liquid.ai/blog/lfm2-5-8b-a1b", "reported_setting": { "mode": "source/model default", "effort": "source does not state", "tools": "benchmark-provided tools", "sampling": "source does not state trial count", "judge": "benchmark-specified", "harness": "Liquid AI official comparative evaluation", "prompt_style": "source does not state", "temperature": "source does not state evaluation parameters", "context": "source does not state evaluation context" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Displayed exactly as 31.91. Official Liquid AI reported result.", "candidates": [ { "score": 36.98, "reference_url": "https://www.liquid.ai/blog/lfm2-5-2-6b", "source_type": "official_blog", "reported_setting": { "mode": "source/model default", "effort": "source does not state", "tools": "benchmark-provided tools", "sampling": "source does not state trial count", "judge": "benchmark-specified", "harness": "Liquid AI official comparative evaluation", "prompt_style": "source does not state", "temperature": "source does not state evaluation parameters", "context": "source does not state evaluation context" }, "notes": "Displayed exactly as 36.98. Official Liquid AI reported result." } ] }, { "model_id": "gemma-4-e2b", "benchmark_id": "browsecomp_plus_openclaw", "score": 8.31, "reference_url": "https://www.liquid.ai/blog/lfm2-5-2-6b", "reported_setting": { "mode": "source/model default", "effort": "source does not state", "tools": "OpenClaw search/retrieval agent over fixed corpus", "sampling": "source does not state trial count", "judge": "exact-answer benchmark evaluation", "harness": "BrowseComp-Plus fixed-corpus evaluation", "prompt_style": "source does not state", "temperature": "source does not state evaluation parameters", "context": "source does not state evaluation context" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Displayed exactly as 8.31. Official Liquid AI reported result." }, { "model_id": "gemma-4-e2b", "benchmark_id": "claw_eval_en_average", "score": 53.14, "reference_url": "https://www.liquid.ai/blog/lfm2-5-2-6b", "reported_setting": { "mode": "source/model default", "effort": "source does not state", "tools": "OpenClaw agent environment", "sampling": "Pass^3; N=3 independent trials", "judge": "benchmark-specified", "harness": "Claw-Eval English tasks across all three public splits", "prompt_style": "source does not state", "temperature": "source does not state evaluation parameters", "context": "source does not state evaluation context" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Displayed exactly as 53.14. Official Liquid AI reported result." }, { "model_id": "gemma-4-e2b", "benchmark_id": "ifstruct_v1", "score": 64.85, "reference_url": "https://www.liquid.ai/blog/lfm2-5-2-6b", "reported_setting": { "mode": "source/model default", "effort": "source does not state", "tools": "none", "sampling": "source does not state trial count", "judge": "binary schema and structural validator", "harness": "IFStruct v1.0 validator without constrained decoding", "prompt_style": "source does not state", "temperature": "source does not state evaluation parameters", "context": "source does not state evaluation context" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Displayed exactly as 64.85. Official Liquid AI reported result." }, { "model_id": "gemma-4-e2b", "benchmark_id": "math_500", "score": 64.0, "reference_url": "https://www.liquid.ai/blog/lfm2-5-8b-a1b", "reported_setting": { "mode": "source/model default", "effort": "source does not state", "tools": "none", "sampling": "source does not state trial count", "judge": "benchmark-specified", "harness": "Liquid AI official comparative evaluation", "prompt_style": "source does not state", "temperature": "source does not state evaluation parameters", "context": "source does not state evaluation context" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Displayed exactly as 64.00. Official Liquid AI reported result." }, { "model_id": "gemma-4-e2b", "benchmark_id": "multi_if", "score": 69.7, "reference_url": "https://www.liquid.ai/blog/lfm2-5-8b-a1b", "reported_setting": { "mode": "source/model default", "effort": "source does not state", "tools": "none", "sampling": "source does not state trial count", "judge": "benchmark-specified", "harness": "Liquid AI official comparative evaluation", "prompt_style": "source does not state", "temperature": "source does not state evaluation parameters", "context": "source does not state evaluation context" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Displayed exactly as 69.70. Official Liquid AI reported result.", "candidates": [ { "score": 69.44, "reference_url": "https://www.liquid.ai/blog/lfm2-5-2-6b", "source_type": "official_blog", "reported_setting": { "mode": "source/model default", "effort": "source does not state", "tools": "none", "sampling": "source does not state trial count", "judge": "benchmark-specified", "harness": "Liquid AI official comparative evaluation", "prompt_style": "source does not state", "temperature": "source does not state evaluation parameters", "context": "source does not state evaluation context" }, "notes": "Displayed exactly as 69.44. Official Liquid AI reported result." } ] }, { "model_id": "gemma-4-e2b", "benchmark_id": "pinchbench", "score": 44.24, "reference_url": "https://www.liquid.ai/blog/lfm2-5-2-6b", "reported_setting": { "mode": "source/model default", "effort": "source does not state", "tools": "OpenClaw coding environment", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "Liquid AI official comparative evaluation", "prompt_style": "source does not state", "temperature": "source does not state evaluation parameters", "context": "source does not state evaluation context" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Displayed exactly as 44.24. Official Liquid AI reported result." }, { "model_id": "gemma-4-e2b", "benchmark_id": "tau3_banking", "score": 3.35, "reference_url": "https://www.liquid.ai/blog/lfm2-5-2-6b", "reported_setting": { "mode": "source/model default", "effort": "source does not state", "tools": "benchmark-provided tools", "sampling": "source does not state trial count", "judge": "benchmark-specified", "harness": "Liquid AI official comparative evaluation", "prompt_style": "source does not state", "temperature": "source does not state evaluation parameters", "context": "source does not state evaluation context" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Displayed exactly as 3.35. Official Liquid AI reported result." }, { "model_id": "gemma-4-e2b", "benchmark_id": "tool_sandbox", "score": 52.4, "reference_url": "https://www.liquid.ai/blog/lfm2-5-2-6b", "reported_setting": { "mode": "source/model default", "effort": "source does not state", "tools": "benchmark-provided tools", "sampling": "source does not state trial count", "judge": "milestone and minefield similarity evaluation", "harness": "Apple ToolSandbox stateful conversational harness", "prompt_style": "source does not state", "temperature": "source does not state evaluation parameters", "context": "source does not state evaluation context" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Displayed exactly as 52.40. Official Liquid AI reported result." }, { "model_id": "gemma-4-e4b", "benchmark_id": "aa_omniscience_accuracy", "score": 8.1, "reference_url": "https://www.liquid.ai/blog/lfm2-5-8b-a1b", "reported_setting": { "mode": "source/model default", "effort": "source does not state", "tools": "none", "sampling": "source does not state trial count", "judge": "official answer and abstention classification", "harness": "Artificial Analysis Omniscience evaluation reported by Liquid", "prompt_style": "source does not state", "temperature": "source does not state evaluation parameters", "context": "source does not state evaluation context" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Displayed exactly as 8.10. Official Liquid AI reported result." }, { "model_id": "gemma-4-e4b", "benchmark_id": "aa_omniscience_index", "score": -50.67, "reference_url": "https://www.liquid.ai/blog/lfm2-5-8b-a1b", "reported_setting": { "mode": "source/model default", "effort": "source does not state", "tools": "none", "sampling": "source does not state trial count", "judge": "official answer and abstention classification", "harness": "Artificial Analysis Omniscience evaluation reported by Liquid", "prompt_style": "source does not state", "temperature": "source does not state evaluation parameters", "context": "source does not state evaluation context" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Displayed exactly as -50.67. Official Liquid AI reported result." }, { "model_id": "gemma-4-e4b", "benchmark_id": "aa_omniscience_non_hallucination", "score": 36.06, "reference_url": "https://www.liquid.ai/blog/lfm2-5-8b-a1b", "reported_setting": { "mode": "source/model default", "effort": "source does not state", "tools": "none", "sampling": "source does not state trial count", "judge": "official answer and abstention classification", "harness": "Artificial Analysis Omniscience evaluation reported by Liquid", "prompt_style": "source does not state", "temperature": "source does not state evaluation parameters", "context": "source does not state evaluation context" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Displayed exactly as 36.06. Official Liquid AI reported result." }, { "model_id": "gemma-4-e4b", "benchmark_id": "aa_omniscience_public_accuracy", "score": 8.33, "reference_url": "https://huggingface.co/LiquidAI/LFM2.5-2.6B/resolve/ab00687315bc1298e9d54e9c4b611dde9867ccc2/README.md", "reported_setting": { "mode": "source/model default", "effort": "source does not state", "tools": "none", "sampling": "source does not state trial count", "judge": "official answer and abstention classification", "harness": "Artificial Analysis Omniscience evaluation reported by Liquid", "prompt_style": "source does not state", "temperature": "source does not state evaluation parameters", "context": "source does not state evaluation context" }, "matches_canonical": false, "source_type": "official_model_card", "audit_status": "verified", "notes": "Displayed exactly as 8.33. Official Liquid AI reported result." }, { "model_id": "gemma-4-e4b", "benchmark_id": "aa_omniscience_public_index", "score": -49.03, "reference_url": "https://www.liquid.ai/blog/lfm2-5-2-6b", "reported_setting": { "mode": "source/model default", "effort": "source does not state", "tools": "none", "sampling": "source does not state trial count", "judge": "official answer and abstention classification", "harness": "Artificial Analysis Omniscience evaluation reported by Liquid", "prompt_style": "source does not state", "temperature": "source does not state evaluation parameters", "context": "source does not state evaluation context" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Displayed exactly as -49.03. Official Liquid AI reported result." }, { "model_id": "gemma-4-e4b", "benchmark_id": "aa_omniscience_public_non_hallucination", "score": 37.42, "reference_url": "https://huggingface.co/LiquidAI/LFM2.5-2.6B/resolve/ab00687315bc1298e9d54e9c4b611dde9867ccc2/README.md", "reported_setting": { "mode": "source/model default", "effort": "source does not state", "tools": "none", "sampling": "source does not state trial count", "judge": "official answer and abstention classification", "harness": "Artificial Analysis Omniscience evaluation reported by Liquid", "prompt_style": "source does not state", "temperature": "source does not state evaluation parameters", "context": "source does not state evaluation context" }, "matches_canonical": false, "source_type": "official_model_card", "audit_status": "verified", "notes": "Displayed exactly as 37.42. Official Liquid AI reported result." }, { "model_id": "gemma-4-e4b", "benchmark_id": "aime_2025", "score": 34.33, "reference_url": "https://www.liquid.ai/blog/lfm2-5-8b-a1b", "reported_setting": { "mode": "source/model default", "effort": "source does not state", "tools": "none", "sampling": "source does not state trial count", "judge": "benchmark-specified", "harness": "Liquid AI official comparative evaluation", "prompt_style": "source does not state", "temperature": "source does not state evaluation parameters", "context": "source does not state evaluation context" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Displayed exactly as 34.33. Official Liquid AI reported result.", "candidates": [ { "score": 34.27, "reference_url": "https://www.liquid.ai/blog/lfm2-5-2-6b", "source_type": "official_blog", "reported_setting": { "mode": "source/model default", "effort": "source does not state", "tools": "none", "sampling": "source does not state trial count", "judge": "benchmark-specified", "harness": "Liquid AI official comparative evaluation", "prompt_style": "source does not state", "temperature": "source does not state evaluation parameters", "context": "source does not state evaluation context" }, "notes": "Displayed exactly as 34.27. Official Liquid AI reported result." } ] }, { "model_id": "gemma-4-e4b", "benchmark_id": "bfcl_v3", "score": 57.31, "reference_url": "https://www.liquid.ai/blog/lfm2-5-8b-a1b", "reported_setting": { "mode": "source/model default", "effort": "source does not state", "tools": "benchmark-provided tools", "sampling": "source does not state trial count", "judge": "benchmark-specified", "harness": "Liquid AI official comparative evaluation", "prompt_style": "source does not state", "temperature": "source does not state evaluation parameters", "context": "source does not state evaluation context" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Displayed exactly as 57.31. Official Liquid AI reported result." }, { "model_id": "gemma-4-e4b", "benchmark_id": "bfcl_v4", "score": 33.92, "reference_url": "https://www.liquid.ai/blog/lfm2-5-8b-a1b", "reported_setting": { "mode": "source/model default", "effort": "source does not state", "tools": "benchmark-provided tools", "sampling": "source does not state trial count", "judge": "benchmark-specified", "harness": "Liquid AI official comparative evaluation", "prompt_style": "source does not state", "temperature": "source does not state evaluation parameters", "context": "source does not state evaluation context" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Displayed exactly as 33.92. Official Liquid AI reported result.", "candidates": [ { "score": 46.39, "reference_url": "https://www.liquid.ai/blog/lfm2-5-2-6b", "source_type": "official_blog", "reported_setting": { "mode": "source/model default", "effort": "source does not state", "tools": "benchmark-provided tools", "sampling": "source does not state trial count", "judge": "benchmark-specified", "harness": "Liquid AI official comparative evaluation", "prompt_style": "source does not state", "temperature": "source does not state evaluation parameters", "context": "source does not state evaluation context" }, "notes": "Displayed exactly as 46.39. Official Liquid AI reported result." } ] }, { "model_id": "gemma-4-e4b", "benchmark_id": "browsecomp_plus_openclaw", "score": 15.9, "reference_url": "https://www.liquid.ai/blog/lfm2-5-2-6b", "reported_setting": { "mode": "source/model default", "effort": "source does not state", "tools": "OpenClaw search/retrieval agent over fixed corpus", "sampling": "source does not state trial count", "judge": "exact-answer benchmark evaluation", "harness": "BrowseComp-Plus fixed-corpus evaluation", "prompt_style": "source does not state", "temperature": "source does not state evaluation parameters", "context": "source does not state evaluation context" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Displayed exactly as 15.90. Official Liquid AI reported result." }, { "model_id": "gemma-4-e4b", "benchmark_id": "claw_eval_en_average", "score": 58.02, "reference_url": "https://www.liquid.ai/blog/lfm2-5-2-6b", "reported_setting": { "mode": "source/model default", "effort": "source does not state", "tools": "OpenClaw agent environment", "sampling": "Pass^3; N=3 independent trials", "judge": "benchmark-specified", "harness": "Claw-Eval English tasks across all three public splits", "prompt_style": "source does not state", "temperature": "source does not state evaluation parameters", "context": "source does not state evaluation context" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Displayed exactly as 58.02. Official Liquid AI reported result." }, { "model_id": "gemma-4-e4b", "benchmark_id": "ifstruct_v1", "score": 76.65, "reference_url": "https://www.liquid.ai/blog/lfm2-5-2-6b", "reported_setting": { "mode": "source/model default", "effort": "source does not state", "tools": "none", "sampling": "source does not state trial count", "judge": "binary schema and structural validator", "harness": "IFStruct v1.0 validator without constrained decoding", "prompt_style": "source does not state", "temperature": "source does not state evaluation parameters", "context": "source does not state evaluation context" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Displayed exactly as 76.65. Official Liquid AI reported result." }, { "model_id": "gemma-4-e4b", "benchmark_id": "math_500", "score": 65.0, "reference_url": "https://www.liquid.ai/blog/lfm2-5-8b-a1b", "reported_setting": { "mode": "source/model default", "effort": "source does not state", "tools": "none", "sampling": "source does not state trial count", "judge": "benchmark-specified", "harness": "Liquid AI official comparative evaluation", "prompt_style": "source does not state", "temperature": "source does not state evaluation parameters", "context": "source does not state evaluation context" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Displayed exactly as 65.00. Official Liquid AI reported result." }, { "model_id": "gemma-4-e4b", "benchmark_id": "multi_if", "score": 77.58, "reference_url": "https://www.liquid.ai/blog/lfm2-5-8b-a1b", "reported_setting": { "mode": "source/model default", "effort": "source does not state", "tools": "none", "sampling": "source does not state trial count", "judge": "benchmark-specified", "harness": "Liquid AI official comparative evaluation", "prompt_style": "source does not state", "temperature": "source does not state evaluation parameters", "context": "source does not state evaluation context" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Displayed exactly as 77.58. Official Liquid AI reported result.", "candidates": [ { "score": 77.35, "reference_url": "https://www.liquid.ai/blog/lfm2-5-2-6b", "source_type": "official_blog", "reported_setting": { "mode": "source/model default", "effort": "source does not state", "tools": "none", "sampling": "source does not state trial count", "judge": "benchmark-specified", "harness": "Liquid AI official comparative evaluation", "prompt_style": "source does not state", "temperature": "source does not state evaluation parameters", "context": "source does not state evaluation context" }, "notes": "Displayed exactly as 77.35. Official Liquid AI reported result." } ] }, { "model_id": "gemma-4-e4b", "benchmark_id": "pinchbench", "score": 55.09, "reference_url": "https://www.liquid.ai/blog/lfm2-5-2-6b", "reported_setting": { "mode": "source/model default", "effort": "source does not state", "tools": "OpenClaw coding environment", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "Liquid AI official comparative evaluation", "prompt_style": "source does not state", "temperature": "source does not state evaluation parameters", "context": "source does not state evaluation context" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Displayed exactly as 55.09. Official Liquid AI reported result." }, { "model_id": "gemma-4-e4b", "benchmark_id": "tau3_banking", "score": 4.12, "reference_url": "https://www.liquid.ai/blog/lfm2-5-2-6b", "reported_setting": { "mode": "source/model default", "effort": "source does not state", "tools": "benchmark-provided tools", "sampling": "source does not state trial count", "judge": "benchmark-specified", "harness": "Liquid AI official comparative evaluation", "prompt_style": "source does not state", "temperature": "source does not state evaluation parameters", "context": "source does not state evaluation context" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Displayed exactly as 4.12. Official Liquid AI reported result." }, { "model_id": "gemma-4-e4b", "benchmark_id": "tool_sandbox", "score": 65.0, "reference_url": "https://www.liquid.ai/blog/lfm2-5-2-6b", "reported_setting": { "mode": "source/model default", "effort": "source does not state", "tools": "benchmark-provided tools", "sampling": "source does not state trial count", "judge": "milestone and minefield similarity evaluation", "harness": "Apple ToolSandbox stateful conversational harness", "prompt_style": "source does not state", "temperature": "source does not state evaluation parameters", "context": "source does not state evaluation context" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Displayed exactly as 65.00. Official Liquid AI reported result." }, { "model_id": "gpt-oss-20b", "benchmark_id": "aa_omniscience_accuracy", "score": 14.57, "reference_url": "https://www.liquid.ai/blog/lfm2-5-8b-a1b", "reported_setting": { "mode": "source/model default", "effort": "source does not state", "tools": "none", "sampling": "source does not state trial count", "judge": "official answer and abstention classification", "harness": "Artificial Analysis Omniscience evaluation reported by Liquid", "prompt_style": "source does not state", "temperature": "source does not state evaluation parameters", "context": "source does not state evaluation context" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Displayed exactly as 14.57. Official Liquid AI reported result." }, { "model_id": "gpt-oss-20b", "benchmark_id": "aa_omniscience_index", "score": -49.17, "reference_url": "https://www.liquid.ai/blog/lfm2-5-8b-a1b", "reported_setting": { "mode": "source/model default", "effort": "source does not state", "tools": "none", "sampling": "source does not state trial count", "judge": "official answer and abstention classification", "harness": "Artificial Analysis Omniscience evaluation reported by Liquid", "prompt_style": "source does not state", "temperature": "source does not state evaluation parameters", "context": "source does not state evaluation context" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Displayed exactly as -49.17. Official Liquid AI reported result." }, { "model_id": "gpt-oss-20b", "benchmark_id": "aa_omniscience_non_hallucination", "score": 24.5, "reference_url": "https://www.liquid.ai/blog/lfm2-5-8b-a1b", "reported_setting": { "mode": "source/model default", "effort": "source does not state", "tools": "none", "sampling": "source does not state trial count", "judge": "official answer and abstention classification", "harness": "Artificial Analysis Omniscience evaluation reported by Liquid", "prompt_style": "source does not state", "temperature": "source does not state evaluation parameters", "context": "source does not state evaluation context" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Displayed exactly as 24.50. Official Liquid AI reported result." }, { "model_id": "gpt-oss-20b", "benchmark_id": "aime_2026", "score": 68.67, "reference_url": "https://www.liquid.ai/blog/lfm2-5-8b-a1b", "reported_setting": { "mode": "source/model default", "effort": "source does not state", "tools": "none", "sampling": "source does not state trial count", "judge": "benchmark-specified", "harness": "Liquid AI official comparative evaluation", "prompt_style": "source does not state", "temperature": "source does not state evaluation parameters", "context": "source does not state evaluation context" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Displayed exactly as 68.67. Official Liquid AI reported result." }, { "model_id": "gpt-oss-20b", "benchmark_id": "bfcl_v3", "score": 62.52, "reference_url": "https://www.liquid.ai/blog/lfm2-5-8b-a1b", "reported_setting": { "mode": "source/model default", "effort": "source does not state", "tools": "benchmark-provided tools", "sampling": "source does not state trial count", "judge": "benchmark-specified", "harness": "Liquid AI official comparative evaluation", "prompt_style": "source does not state", "temperature": "source does not state evaluation parameters", "context": "source does not state evaluation context" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Displayed exactly as 62.52. Official Liquid AI reported result." }, { "model_id": "gpt-oss-20b", "benchmark_id": "bfcl_v4", "score": 49.88, "reference_url": "https://www.liquid.ai/blog/lfm2-5-8b-a1b", "reported_setting": { "mode": "source/model default", "effort": "source does not state", "tools": "benchmark-provided tools", "sampling": "source does not state trial count", "judge": "benchmark-specified", "harness": "Liquid AI official comparative evaluation", "prompt_style": "source does not state", "temperature": "source does not state evaluation parameters", "context": "source does not state evaluation context" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Displayed exactly as 49.88. Official Liquid AI reported result." }, { "model_id": "gpt-oss-20b", "benchmark_id": "ifbench", "score": 58.65, "reference_url": "https://www.liquid.ai/blog/lfm2-5-8b-a1b", "reported_setting": { "mode": "source/model default", "effort": "source does not state", "tools": "none", "sampling": "source does not state trial count", "judge": "benchmark-specified", "harness": "Liquid AI official comparative evaluation", "prompt_style": "source does not state", "temperature": "source does not state evaluation parameters", "context": "source does not state evaluation context" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Displayed exactly as 58.65. Official Liquid AI reported result." }, { "model_id": "gpt-oss-20b", "benchmark_id": "multi_if", "score": 76.64, "reference_url": "https://www.liquid.ai/blog/lfm2-5-8b-a1b", "reported_setting": { "mode": "source/model default", "effort": "source does not state", "tools": "none", "sampling": "source does not state trial count", "judge": "benchmark-specified", "harness": "Liquid AI official comparative evaluation", "prompt_style": "source does not state", "temperature": "source does not state evaluation parameters", "context": "source does not state evaluation context" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Displayed exactly as 76.64. Official Liquid AI reported result." }, { "model_id": "gpt-oss-20b", "benchmark_id": "tau2_bench_retail", "score": 53.51, "reference_url": "https://www.liquid.ai/blog/lfm2-5-8b-a1b", "reported_setting": { "mode": "source/model default", "effort": "source does not state", "tools": "benchmark-provided tools", "sampling": "source does not state trial count", "judge": "benchmark-specified", "harness": "Liquid AI official comparative evaluation", "prompt_style": "source does not state", "temperature": "source does not state evaluation parameters", "context": "source does not state evaluation context" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Displayed exactly as 53.51. Official Liquid AI reported result." }, { "model_id": "gpt-oss-20b", "benchmark_id": "tau2_bench_telecom", "score": 57.24, "reference_url": "https://www.liquid.ai/blog/lfm2-5-8b-a1b", "reported_setting": { "mode": "source/model default", "effort": "source does not state", "tools": "benchmark-provided tools", "sampling": "source does not state trial count", "judge": "benchmark-specified", "harness": "Liquid AI official comparative evaluation", "prompt_style": "source does not state", "temperature": "source does not state evaluation parameters", "context": "source does not state evaluation context" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Displayed exactly as 57.24. Official Liquid AI reported result." }, { "model_id": "granite-4.0-350m", "benchmark_id": "bfcl_v3", "score": 39.58, "reference_url": "https://www.liquid.ai/blog/lfm2-5-230m", "reported_setting": { "mode": "source/model default", "effort": "source does not state", "tools": "benchmark-provided tools", "sampling": "source does not state trial count", "judge": "benchmark-specified", "harness": "Liquid AI official comparative evaluation", "prompt_style": "source does not state", "temperature": "source does not state evaluation parameters", "context": "source does not state evaluation context" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Displayed exactly as 39.58. Official Liquid AI reported result." }, { "model_id": "granite-4.0-350m", "benchmark_id": "bfcl_v4", "score": 13.73, "reference_url": "https://www.liquid.ai/blog/lfm2-5-230m", "reported_setting": { "mode": "source/model default", "effort": "source does not state", "tools": "benchmark-provided tools", "sampling": "source does not state trial count", "judge": "benchmark-specified", "harness": "Liquid AI official comparative evaluation", "prompt_style": "source does not state", "temperature": "source does not state evaluation parameters", "context": "source does not state evaluation context" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Displayed exactly as 13.73. Official Liquid AI reported result." }, { "model_id": "granite-4.0-350m", "benchmark_id": "gpqa_diamond", "score": 25.91, "reference_url": "https://www.liquid.ai/blog/lfm2-5-230m", "reported_setting": { "mode": "source/model default", "effort": "source does not state", "tools": "none", "sampling": "source does not state trial count", "judge": "benchmark-specified", "harness": "Liquid AI official comparative evaluation", "prompt_style": "source does not state", "temperature": "source does not state evaluation parameters", "context": "source does not state evaluation context" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Displayed exactly as 25.91. Official Liquid AI reported result." }, { "model_id": "granite-4.0-350m", "benchmark_id": "ifbench", "score": 15.98, "reference_url": "https://www.liquid.ai/blog/lfm2-5-230m", "reported_setting": { "mode": "source/model default", "effort": "source does not state", "tools": "none", "sampling": "source does not state trial count", "judge": "benchmark-specified", "harness": "Liquid AI official comparative evaluation", "prompt_style": "source does not state", "temperature": "source does not state evaluation parameters", "context": "source does not state evaluation context" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Displayed exactly as 15.98. Official Liquid AI reported result." }, { "model_id": "granite-4.0-350m", "benchmark_id": "ifeval", "score": 53.48, "reference_url": "https://www.liquid.ai/blog/lfm2-5-230m", "reported_setting": { "mode": "source/model default", "effort": "source does not state", "tools": "none", "sampling": "source does not state trial count", "judge": "benchmark-specified", "harness": "Liquid AI official comparative evaluation", "prompt_style": "source does not state", "temperature": "source does not state evaluation parameters", "context": "source does not state evaluation context" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Displayed exactly as 53.48. Official Liquid AI reported result." }, { "model_id": "granite-4.0-350m", "benchmark_id": "mmlu_pro", "score": 12.84, "reference_url": "https://www.liquid.ai/blog/lfm2-5-230m", "reported_setting": { "mode": "source/model default", "effort": "source does not state", "tools": "none", "sampling": "source does not state trial count", "judge": "benchmark-specified", "harness": "Liquid AI official comparative evaluation", "prompt_style": "source does not state", "temperature": "source does not state evaluation parameters", "context": "source does not state evaluation context" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Displayed exactly as 12.84. Official Liquid AI reported result." }, { "model_id": "granite-4.0-350m", "benchmark_id": "multi_if", "score": 24.21, "reference_url": "https://www.liquid.ai/blog/lfm2-5-230m", "reported_setting": { "mode": "source/model default", "effort": "source does not state", "tools": "none", "sampling": "source does not state trial count", "judge": "benchmark-specified", "harness": "Liquid AI official comparative evaluation", "prompt_style": "source does not state", "temperature": "source does not state evaluation parameters", "context": "source does not state evaluation context" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Displayed exactly as 24.21. Official Liquid AI reported result." }, { "model_id": "granite-4.0-350m", "benchmark_id": "tau2_bench_retail", "score": 6.14, "reference_url": "https://www.liquid.ai/blog/lfm2-5-230m", "reported_setting": { "mode": "source/model default", "effort": "source does not state", "tools": "benchmark-provided tools", "sampling": "source does not state trial count", "judge": "benchmark-specified", "harness": "Liquid AI official comparative evaluation", "prompt_style": "source does not state", "temperature": "source does not state evaluation parameters", "context": "source does not state evaluation context" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Displayed exactly as 6.14. Official Liquid AI reported result." }, { "model_id": "granite-4.0-350m", "benchmark_id": "tau2_bench_telecom", "score": 2.92, "reference_url": "https://www.liquid.ai/blog/lfm2-5-230m", "reported_setting": { "mode": "source/model default", "effort": "source does not state", "tools": "benchmark-provided tools", "sampling": "source does not state trial count", "judge": "benchmark-specified", "harness": "Liquid AI official comparative evaluation", "prompt_style": "source does not state", "temperature": "source does not state evaluation parameters", "context": "source does not state evaluation context" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Displayed exactly as 2.92. Official Liquid AI reported result." }, { "model_id": "granite-4.0-h-350m", "benchmark_id": "bfcl_v3", "score": 43.07, "reference_url": "https://www.liquid.ai/blog/lfm2-5-230m", "reported_setting": { "mode": "source/model default", "effort": "source does not state", "tools": "benchmark-provided tools", "sampling": "source does not state trial count", "judge": "benchmark-specified", "harness": "Liquid AI official comparative evaluation", "prompt_style": "source does not state", "temperature": "source does not state evaluation parameters", "context": "source does not state evaluation context" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Displayed exactly as 43.07. Official Liquid AI reported result." }, { "model_id": "granite-4.0-h-350m", "benchmark_id": "bfcl_v4", "score": 13.28, "reference_url": "https://www.liquid.ai/blog/lfm2-5-230m", "reported_setting": { "mode": "source/model default", "effort": "source does not state", "tools": "benchmark-provided tools", "sampling": "source does not state trial count", "judge": "benchmark-specified", "harness": "Liquid AI official comparative evaluation", "prompt_style": "source does not state", "temperature": "source does not state evaluation parameters", "context": "source does not state evaluation context" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Displayed exactly as 13.28. Official Liquid AI reported result." }, { "model_id": "granite-4.0-h-350m", "benchmark_id": "gpqa_diamond", "score": 22.32, "reference_url": "https://www.liquid.ai/blog/lfm2-5-230m", "reported_setting": { "mode": "source/model default", "effort": "source does not state", "tools": "none", "sampling": "source does not state trial count", "judge": "benchmark-specified", "harness": "Liquid AI official comparative evaluation", "prompt_style": "source does not state", "temperature": "source does not state evaluation parameters", "context": "source does not state evaluation context" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Displayed exactly as 22.32. Official Liquid AI reported result." }, { "model_id": "granite-4.0-h-350m", "benchmark_id": "ifbench", "score": 17.22, "reference_url": "https://www.liquid.ai/blog/lfm2-5-230m", "reported_setting": { "mode": "source/model default", "effort": "source does not state", "tools": "none", "sampling": "source does not state trial count", "judge": "benchmark-specified", "harness": "Liquid AI official comparative evaluation", "prompt_style": "source does not state", "temperature": "source does not state evaluation parameters", "context": "source does not state evaluation context" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Displayed exactly as 17.22. Official Liquid AI reported result." }, { "model_id": "granite-4.0-h-350m", "benchmark_id": "ifeval", "score": 61.27, "reference_url": "https://www.liquid.ai/blog/lfm2-5-230m", "reported_setting": { "mode": "source/model default", "effort": "source does not state", "tools": "none", "sampling": "source does not state trial count", "judge": "benchmark-specified", "harness": "Liquid AI official comparative evaluation", "prompt_style": "source does not state", "temperature": "source does not state evaluation parameters", "context": "source does not state evaluation context" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Displayed exactly as 61.27. Official Liquid AI reported result." }, { "model_id": "granite-4.0-h-350m", "benchmark_id": "mmlu_pro", "score": 13.14, "reference_url": "https://www.liquid.ai/blog/lfm2-5-230m", "reported_setting": { "mode": "source/model default", "effort": "source does not state", "tools": "none", "sampling": "source does not state trial count", "judge": "benchmark-specified", "harness": "Liquid AI official comparative evaluation", "prompt_style": "source does not state", "temperature": "source does not state evaluation parameters", "context": "source does not state evaluation context" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Displayed exactly as 13.14. Official Liquid AI reported result." }, { "model_id": "granite-4.0-h-350m", "benchmark_id": "multi_if", "score": 28.7, "reference_url": "https://www.liquid.ai/blog/lfm2-5-230m", "reported_setting": { "mode": "source/model default", "effort": "source does not state", "tools": "none", "sampling": "source does not state trial count", "judge": "benchmark-specified", "harness": "Liquid AI official comparative evaluation", "prompt_style": "source does not state", "temperature": "source does not state evaluation parameters", "context": "source does not state evaluation context" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Displayed exactly as 28.70. Official Liquid AI reported result." }, { "model_id": "granite-4.0-h-350m", "benchmark_id": "tau2_bench_avg", "score": 18.85, "reference_url": "https://aypchzzf9pftwuto.public.blob.vercel-storage.com/lfm2_5_230m_benchmarks-EcJisF1zm7I7KiST75d7XsE2LwE0lD.png", "reported_setting": { "mode": "source/model default", "effort": "source does not state", "tools": "benchmark-provided tools", "sampling": "source does not state trial count", "judge": "benchmark-specified", "harness": "Liquid AI official comparative evaluation", "prompt_style": "source does not state", "temperature": "source does not state evaluation parameters", "context": "source does not state evaluation context" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Displayed exactly as 18.85. Official Liquid AI reported result." }, { "model_id": "granite-4.0-h-350m", "benchmark_id": "tau2_bench_retail", "score": 6.14, "reference_url": "https://www.liquid.ai/blog/lfm2-5-230m", "reported_setting": { "mode": "source/model default", "effort": "source does not state", "tools": "benchmark-provided tools", "sampling": "source does not state trial count", "judge": "benchmark-specified", "harness": "Liquid AI official comparative evaluation", "prompt_style": "source does not state", "temperature": "source does not state evaluation parameters", "context": "source does not state evaluation context" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Displayed exactly as 6.14. Official Liquid AI reported result." }, { "model_id": "granite-4.0-h-350m", "benchmark_id": "tau2_bench_telecom", "score": 13.74, "reference_url": "https://www.liquid.ai/blog/lfm2-5-230m", "reported_setting": { "mode": "source/model default", "effort": "source does not state", "tools": "benchmark-provided tools", "sampling": "source does not state trial count", "judge": "benchmark-specified", "harness": "Liquid AI official comparative evaluation", "prompt_style": "source does not state", "temperature": "source does not state evaluation parameters", "context": "source does not state evaluation context" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Displayed exactly as 13.74. Official Liquid AI reported result." }, { "model_id": "granite-4.0-h-tiny", "benchmark_id": "aa_omniscience_accuracy", "score": 9.37, "reference_url": "https://www.liquid.ai/blog/lfm2-5-8b-a1b", "reported_setting": { "mode": "source/model default", "effort": "source does not state", "tools": "none", "sampling": "source does not state trial count", "judge": "official answer and abstention classification", "harness": "Artificial Analysis Omniscience evaluation reported by Liquid", "prompt_style": "source does not state", "temperature": "source does not state evaluation parameters", "context": "source does not state evaluation context" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Displayed exactly as 9.37. Official Liquid AI reported result." }, { "model_id": "granite-4.0-h-tiny", "benchmark_id": "aa_omniscience_index", "score": -75.5, "reference_url": "https://www.liquid.ai/blog/lfm2-5-8b-a1b", "reported_setting": { "mode": "source/model default", "effort": "source does not state", "tools": "none", "sampling": "source does not state trial count", "judge": "official answer and abstention classification", "harness": "Artificial Analysis Omniscience evaluation reported by Liquid", "prompt_style": "source does not state", "temperature": "source does not state evaluation parameters", "context": "source does not state evaluation context" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Displayed exactly as -75.50. Official Liquid AI reported result." }, { "model_id": "granite-4.0-h-tiny", "benchmark_id": "aa_omniscience_non_hallucination", "score": 6.38, "reference_url": "https://www.liquid.ai/blog/lfm2-5-8b-a1b", "reported_setting": { "mode": "source/model default", "effort": "source does not state", "tools": "none", "sampling": "source does not state trial count", "judge": "official answer and abstention classification", "harness": "Artificial Analysis Omniscience evaluation reported by Liquid", "prompt_style": "source does not state", "temperature": "source does not state evaluation parameters", "context": "source does not state evaluation context" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Displayed exactly as 6.38. Official Liquid AI reported result." }, { "model_id": "granite-4.0-h-tiny", "benchmark_id": "aime_2025", "score": 4.93, "reference_url": "https://www.liquid.ai/blog/lfm2-5-8b-a1b", "reported_setting": { "mode": "source/model default", "effort": "source does not state", "tools": "none", "sampling": "source does not state trial count", "judge": "benchmark-specified", "harness": "Liquid AI official comparative evaluation", "prompt_style": "source does not state", "temperature": "source does not state evaluation parameters", "context": "source does not state evaluation context" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Displayed exactly as 4.93. Official Liquid AI reported result." }, { "model_id": "granite-4.0-h-tiny", "benchmark_id": "aime_2026", "score": 3.33, "reference_url": "https://www.liquid.ai/blog/lfm2-5-8b-a1b", "reported_setting": { "mode": "source/model default", "effort": "source does not state", "tools": "none", "sampling": "source does not state trial count", "judge": "benchmark-specified", "harness": "Liquid AI official comparative evaluation", "prompt_style": "source does not state", "temperature": "source does not state evaluation parameters", "context": "source does not state evaluation context" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Displayed exactly as 3.33. Official Liquid AI reported result." }, { "model_id": "granite-4.0-h-tiny", "benchmark_id": "bfcl_v3", "score": 56.89, "reference_url": "https://www.liquid.ai/blog/lfm2-5-8b-a1b", "reported_setting": { "mode": "source/model default", "effort": "source does not state", "tools": "benchmark-provided tools", "sampling": "source does not state trial count", "judge": "benchmark-specified", "harness": "Liquid AI official comparative evaluation", "prompt_style": "source does not state", "temperature": "source does not state evaluation parameters", "context": "source does not state evaluation context" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Displayed exactly as 56.89. Official Liquid AI reported result." }, { "model_id": "granite-4.0-h-tiny", "benchmark_id": "bfcl_v4", "score": 28.52, "reference_url": "https://www.liquid.ai/blog/lfm2-5-8b-a1b", "reported_setting": { "mode": "source/model default", "effort": "source does not state", "tools": "benchmark-provided tools", "sampling": "source does not state trial count", "judge": "benchmark-specified", "harness": "Liquid AI official comparative evaluation", "prompt_style": "source does not state", "temperature": "source does not state evaluation parameters", "context": "source does not state evaluation context" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Displayed exactly as 28.52. Official Liquid AI reported result." }, { "model_id": "granite-4.0-h-tiny", "benchmark_id": "ifbench", "score": 21.28, "reference_url": "https://www.liquid.ai/blog/lfm2-5-8b-a1b", "reported_setting": { "mode": "source/model default", "effort": "source does not state", "tools": "none", "sampling": "source does not state trial count", "judge": "benchmark-specified", "harness": "Liquid AI official comparative evaluation", "prompt_style": "source does not state", "temperature": "source does not state evaluation parameters", "context": "source does not state evaluation context" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Displayed exactly as 21.28. Official Liquid AI reported result." }, { "model_id": "granite-4.0-h-tiny", "benchmark_id": "ifeval", "score": 82.23, "reference_url": "https://www.liquid.ai/blog/lfm2-5-8b-a1b", "reported_setting": { "mode": "source/model default", "effort": "source does not state", "tools": "none", "sampling": "source does not state trial count", "judge": "benchmark-specified", "harness": "Liquid AI official comparative evaluation", "prompt_style": "source does not state", "temperature": "source does not state evaluation parameters", "context": "source does not state evaluation context" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Displayed exactly as 82.23. Official Liquid AI reported result." }, { "model_id": "granite-4.0-h-tiny", "benchmark_id": "math_500", "score": 59.2, "reference_url": "https://www.liquid.ai/blog/lfm2-5-8b-a1b", "reported_setting": { "mode": "source/model default", "effort": "source does not state", "tools": "none", "sampling": "source does not state trial count", "judge": "benchmark-specified", "harness": "Liquid AI official comparative evaluation", "prompt_style": "source does not state", "temperature": "source does not state evaluation parameters", "context": "source does not state evaluation context" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Displayed exactly as 59.20. Official Liquid AI reported result." }, { "model_id": "granite-4.0-h-tiny", "benchmark_id": "multi_if", "score": 59.0, "reference_url": "https://www.liquid.ai/blog/lfm2-5-8b-a1b", "reported_setting": { "mode": "source/model default", "effort": "source does not state", "tools": "none", "sampling": "source does not state trial count", "judge": "benchmark-specified", "harness": "Liquid AI official comparative evaluation", "prompt_style": "source does not state", "temperature": "source does not state evaluation parameters", "context": "source does not state evaluation context" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Displayed exactly as 59.00. Official Liquid AI reported result." }, { "model_id": "granite-4.0-h-tiny", "benchmark_id": "tau2_bench_retail", "score": 18.42, "reference_url": "https://www.liquid.ai/blog/lfm2-5-8b-a1b", "reported_setting": { "mode": "source/model default", "effort": "source does not state", "tools": "benchmark-provided tools", "sampling": "source does not state trial count", "judge": "benchmark-specified", "harness": "Liquid AI official comparative evaluation", "prompt_style": "source does not state", "temperature": "source does not state evaluation parameters", "context": "source does not state evaluation context" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Displayed exactly as 18.42. Official Liquid AI reported result." }, { "model_id": "granite-4.0-h-tiny", "benchmark_id": "tau2_bench_telecom", "score": 16.67, "reference_url": "https://www.liquid.ai/blog/lfm2-5-8b-a1b", "reported_setting": { "mode": "source/model default", "effort": "source does not state", "tools": "benchmark-provided tools", "sampling": "source does not state trial count", "judge": "benchmark-specified", "harness": "Liquid AI official comparative evaluation", "prompt_style": "source does not state", "temperature": "source does not state evaluation parameters", "context": "source does not state evaluation context" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Displayed exactly as 16.67. Official Liquid AI reported result." }, { "model_id": "lfm2-350m", "benchmark_id": "bfcl_v3", "score": 22.95, "reference_url": "https://www.liquid.ai/blog/lfm2-5-230m", "reported_setting": { "mode": "source/model default", "effort": "source does not state", "tools": "benchmark-provided tools", "sampling": "source does not state trial count", "judge": "benchmark-specified", "harness": "Liquid AI official comparative evaluation", "prompt_style": "source does not state", "temperature": "source does not state evaluation parameters", "context": "source does not state evaluation context" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Displayed exactly as 22.95. Official Liquid AI reported result." }, { "model_id": "lfm2-350m", "benchmark_id": "bfcl_v4", "score": 12.29, "reference_url": "https://www.liquid.ai/blog/lfm2-5-230m", "reported_setting": { "mode": "source/model default", "effort": "source does not state", "tools": "benchmark-provided tools", "sampling": "source does not state trial count", "judge": "benchmark-specified", "harness": "Liquid AI official comparative evaluation", "prompt_style": "source does not state", "temperature": "source does not state evaluation parameters", "context": "source does not state evaluation context" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Displayed exactly as 12.29. Official Liquid AI reported result." }, { "model_id": "lfm2-350m", "benchmark_id": "gpqa_diamond", "score": 27.58, "reference_url": "https://www.liquid.ai/blog/lfm2-5-230m", "reported_setting": { "mode": "source/model default", "effort": "source does not state", "tools": "none", "sampling": "source does not state trial count", "judge": "benchmark-specified", "harness": "Liquid AI official comparative evaluation", "prompt_style": "source does not state", "temperature": "source does not state evaluation parameters", "context": "source does not state evaluation context" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Displayed exactly as 27.58. Official Liquid AI reported result." }, { "model_id": "lfm2-350m", "benchmark_id": "ifbench", "score": 18.2, "reference_url": "https://www.liquid.ai/blog/lfm2-5-230m", "reported_setting": { "mode": "source/model default", "effort": "source does not state", "tools": "none", "sampling": "source does not state trial count", "judge": "benchmark-specified", "harness": "Liquid AI official comparative evaluation", "prompt_style": "source does not state", "temperature": "source does not state evaluation parameters", "context": "source does not state evaluation context" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Displayed exactly as 18.20. Official Liquid AI reported result." }, { "model_id": "lfm2-350m", "benchmark_id": "ifeval", "score": 64.96, "reference_url": "https://www.liquid.ai/blog/lfm2-5-230m", "reported_setting": { "mode": "source/model default", "effort": "source does not state", "tools": "none", "sampling": "source does not state trial count", "judge": "benchmark-specified", "harness": "Liquid AI official comparative evaluation", "prompt_style": "source does not state", "temperature": "source does not state evaluation parameters", "context": "source does not state evaluation context" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Displayed exactly as 64.96. Official Liquid AI reported result." }, { "model_id": "lfm2-350m", "benchmark_id": "mmlu_pro", "score": 19.29, "reference_url": "https://www.liquid.ai/blog/lfm2-5-230m", "reported_setting": { "mode": "source/model default", "effort": "source does not state", "tools": "none", "sampling": "source does not state trial count", "judge": "benchmark-specified", "harness": "Liquid AI official comparative evaluation", "prompt_style": "source does not state", "temperature": "source does not state evaluation parameters", "context": "source does not state evaluation context" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Displayed exactly as 19.29. Official Liquid AI reported result." }, { "model_id": "lfm2-350m", "benchmark_id": "multi_if", "score": 32.92, "reference_url": "https://www.liquid.ai/blog/lfm2-5-230m", "reported_setting": { "mode": "source/model default", "effort": "source does not state", "tools": "none", "sampling": "source does not state trial count", "judge": "benchmark-specified", "harness": "Liquid AI official comparative evaluation", "prompt_style": "source does not state", "temperature": "source does not state evaluation parameters", "context": "source does not state evaluation context" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Displayed exactly as 32.92. Official Liquid AI reported result." }, { "model_id": "lfm2-350m", "benchmark_id": "tau2_bench_avg", "score": 20.13, "reference_url": "https://aypchzzf9pftwuto.public.blob.vercel-storage.com/lfm2_5_230m_benchmarks-EcJisF1zm7I7KiST75d7XsE2LwE0lD.png", "reported_setting": { "mode": "source/model default", "effort": "source does not state", "tools": "benchmark-provided tools", "sampling": "source does not state trial count", "judge": "benchmark-specified", "harness": "Liquid AI official comparative evaluation", "prompt_style": "source does not state", "temperature": "source does not state evaluation parameters", "context": "source does not state evaluation context" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Displayed exactly as 20.13. Official Liquid AI reported result." }, { "model_id": "lfm2-350m", "benchmark_id": "tau2_bench_retail", "score": 5.56, "reference_url": "https://www.liquid.ai/blog/lfm2-5-230m", "reported_setting": { "mode": "source/model default", "effort": "source does not state", "tools": "benchmark-provided tools", "sampling": "source does not state trial count", "judge": "benchmark-specified", "harness": "Liquid AI official comparative evaluation", "prompt_style": "source does not state", "temperature": "source does not state evaluation parameters", "context": "source does not state evaluation context" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Displayed exactly as 5.56. Official Liquid AI reported result." }, { "model_id": "lfm2-350m", "benchmark_id": "tau2_bench_telecom", "score": 10.82, "reference_url": "https://www.liquid.ai/blog/lfm2-5-230m", "reported_setting": { "mode": "source/model default", "effort": "source does not state", "tools": "benchmark-provided tools", "sampling": "source does not state trial count", "judge": "benchmark-specified", "harness": "Liquid AI official comparative evaluation", "prompt_style": "source does not state", "temperature": "source does not state evaluation parameters", "context": "source does not state evaluation context" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Displayed exactly as 10.82. Official Liquid AI reported result." }, { "model_id": "lfm2-8b-a1b", "benchmark_id": "aa_omniscience_accuracy", "score": 7.33, "reference_url": "https://www.liquid.ai/blog/lfm2-5-8b-a1b", "reported_setting": { "mode": "non-thinking", "effort": "source does not state", "tools": "none", "sampling": "source does not state trial count", "judge": "official answer and abstention classification", "harness": "Artificial Analysis Omniscience evaluation reported by Liquid", "prompt_style": "source does not state", "temperature": "source does not state evaluation parameters", "context": "source does not state evaluation context" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Displayed exactly as 7.33. Official Liquid AI reported result." }, { "model_id": "lfm2-8b-a1b", "benchmark_id": "aa_omniscience_index", "score": -78.42, "reference_url": "https://www.liquid.ai/blog/lfm2-5-8b-a1b", "reported_setting": { "mode": "non-thinking", "effort": "source does not state", "tools": "none", "sampling": "source does not state trial count", "judge": "official answer and abstention classification", "harness": "Artificial Analysis Omniscience evaluation reported by Liquid", "prompt_style": "source does not state", "temperature": "source does not state evaluation parameters", "context": "source does not state evaluation context" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Displayed exactly as -78.42. Official Liquid AI reported result." }, { "model_id": "lfm2-8b-a1b", "benchmark_id": "aa_omniscience_non_hallucination", "score": 7.46, "reference_url": "https://www.liquid.ai/blog/lfm2-5-8b-a1b", "reported_setting": { "mode": "non-thinking", "effort": "source does not state", "tools": "none", "sampling": "source does not state trial count", "judge": "official answer and abstention classification", "harness": "Artificial Analysis Omniscience evaluation reported by Liquid", "prompt_style": "source does not state", "temperature": "source does not state evaluation parameters", "context": "source does not state evaluation context" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Displayed exactly as 7.46. Official Liquid AI reported result." }, { "model_id": "lfm2-8b-a1b", "benchmark_id": "aime_2025", "score": 20.0, "reference_url": "https://www.liquid.ai/blog/lfm2-5-8b-a1b", "reported_setting": { "mode": "non-thinking", "effort": "source does not state", "tools": "none", "sampling": "source does not state trial count", "judge": "benchmark-specified", "harness": "Liquid AI official comparative evaluation", "prompt_style": "source does not state", "temperature": "source does not state evaluation parameters", "context": "source does not state evaluation context" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Displayed exactly as 20.00. Official Liquid AI reported result." }, { "model_id": "lfm2-8b-a1b", "benchmark_id": "bfcl_v3", "score": 45.07, "reference_url": "https://www.liquid.ai/blog/lfm2-5-8b-a1b", "reported_setting": { "mode": "non-thinking", "effort": "source does not state", "tools": "benchmark-provided tools", "sampling": "source does not state trial count", "judge": "benchmark-specified", "harness": "Liquid AI official comparative evaluation", "prompt_style": "source does not state", "temperature": "source does not state evaluation parameters", "context": "source does not state evaluation context" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Displayed exactly as 45.07. Official Liquid AI reported result." }, { "model_id": "lfm2-8b-a1b", "benchmark_id": "bfcl_v4", "score": 25.52, "reference_url": "https://www.liquid.ai/blog/lfm2-5-8b-a1b", "reported_setting": { "mode": "non-thinking", "effort": "source does not state", "tools": "benchmark-provided tools", "sampling": "source does not state trial count", "judge": "benchmark-specified", "harness": "Liquid AI official comparative evaluation", "prompt_style": "source does not state", "temperature": "source does not state evaluation parameters", "context": "source does not state evaluation context" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Displayed exactly as 25.52. Official Liquid AI reported result." }, { "model_id": "lfm2-8b-a1b", "benchmark_id": "ifbench", "score": 26.0, "reference_url": "https://www.liquid.ai/blog/lfm2-5-8b-a1b", "reported_setting": { "mode": "non-thinking", "effort": "source does not state", "tools": "none", "sampling": "source does not state trial count", "judge": "benchmark-specified", "harness": "Liquid AI official comparative evaluation", "prompt_style": "source does not state", "temperature": "source does not state evaluation parameters", "context": "source does not state evaluation context" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Displayed exactly as 26.00. Official Liquid AI reported result." }, { "model_id": "lfm2-8b-a1b", "benchmark_id": "ifeval", "score": 79.44, "reference_url": "https://www.liquid.ai/blog/lfm2-5-8b-a1b", "reported_setting": { "mode": "non-thinking", "effort": "source does not state", "tools": "none", "sampling": "source does not state trial count", "judge": "benchmark-specified", "harness": "Liquid AI official comparative evaluation", "prompt_style": "source does not state", "temperature": "source does not state evaluation parameters", "context": "source does not state evaluation context" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Displayed exactly as 79.44. Official Liquid AI reported result." }, { "model_id": "lfm2-8b-a1b", "benchmark_id": "math_500", "score": 74.8, "reference_url": "https://www.liquid.ai/blog/lfm2-5-8b-a1b", "reported_setting": { "mode": "non-thinking", "effort": "source does not state", "tools": "none", "sampling": "source does not state trial count", "judge": "benchmark-specified", "harness": "Liquid AI official comparative evaluation", "prompt_style": "source does not state", "temperature": "source does not state evaluation parameters", "context": "source does not state evaluation context" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Displayed exactly as 74.80. Official Liquid AI reported result." }, { "model_id": "lfm2-8b-a1b", "benchmark_id": "multi_if", "score": 58.54, "reference_url": "https://www.liquid.ai/blog/lfm2-5-8b-a1b", "reported_setting": { "mode": "non-thinking", "effort": "source does not state", "tools": "none", "sampling": "source does not state trial count", "judge": "benchmark-specified", "harness": "Liquid AI official comparative evaluation", "prompt_style": "source does not state", "temperature": "source does not state evaluation parameters", "context": "source does not state evaluation context" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Displayed exactly as 58.54. Official Liquid AI reported result." }, { "model_id": "lfm2-8b-a1b", "benchmark_id": "tau2_bench_retail", "score": 7.02, "reference_url": "https://www.liquid.ai/blog/lfm2-5-8b-a1b", "reported_setting": { "mode": "non-thinking", "effort": "source does not state", "tools": "benchmark-provided tools", "sampling": "source does not state trial count", "judge": "benchmark-specified", "harness": "Liquid AI official comparative evaluation", "prompt_style": "source does not state", "temperature": "source does not state evaluation parameters", "context": "source does not state evaluation context" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Displayed exactly as 7.02. Official Liquid AI reported result." }, { "model_id": "lfm2-8b-a1b", "benchmark_id": "tau2_bench_telecom", "score": 13.6, "reference_url": "https://www.liquid.ai/blog/lfm2-5-8b-a1b", "reported_setting": { "mode": "non-thinking", "effort": "source does not state", "tools": "benchmark-provided tools", "sampling": "source does not state trial count", "judge": "benchmark-specified", "harness": "Liquid AI official comparative evaluation", "prompt_style": "source does not state", "temperature": "source does not state evaluation parameters", "context": "source does not state evaluation context" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Displayed exactly as 13.60. Official Liquid AI reported result." }, { "model_id": "lfm2.5-2.6b", "benchmark_id": "aa_omniscience_public_accuracy", "score": 8.13, "reference_url": "https://huggingface.co/LiquidAI/LFM2.5-2.6B/resolve/ab00687315bc1298e9d54e9c4b611dde9867ccc2/README.md", "reported_setting": { "mode": "thinking", "effort": "source does not state", "tools": "none", "sampling": "source does not state trial count", "judge": "official answer and abstention classification", "harness": "Artificial Analysis Omniscience evaluation reported by Liquid", "prompt_style": "source does not state", "temperature": "source does not state evaluation parameters", "context": "source does not state evaluation context" }, "matches_canonical": false, "source_type": "official_model_card", "audit_status": "verified", "notes": "Displayed exactly as 8.13. Official Liquid AI reported result." }, { "model_id": "lfm2.5-2.6b", "benchmark_id": "aa_omniscience_public_index", "score": -29.5, "reference_url": "https://www.liquid.ai/blog/lfm2-5-2-6b", "reported_setting": { "mode": "thinking", "effort": "source does not state", "tools": "none", "sampling": "source does not state trial count", "judge": "official answer and abstention classification", "harness": "Artificial Analysis Omniscience evaluation reported by Liquid", "prompt_style": "source does not state", "temperature": "source does not state evaluation parameters", "context": "source does not state evaluation context" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Displayed exactly as -29.50. Official Liquid AI reported result." }, { "model_id": "lfm2.5-2.6b", "benchmark_id": "aa_omniscience_public_non_hallucination", "score": 59.04, "reference_url": "https://huggingface.co/LiquidAI/LFM2.5-2.6B/resolve/ab00687315bc1298e9d54e9c4b611dde9867ccc2/README.md", "reported_setting": { "mode": "thinking", "effort": "source does not state", "tools": "none", "sampling": "source does not state trial count", "judge": "official answer and abstention classification", "harness": "Artificial Analysis Omniscience evaluation reported by Liquid", "prompt_style": "source does not state", "temperature": "source does not state evaluation parameters", "context": "source does not state evaluation context" }, "matches_canonical": false, "source_type": "official_model_card", "audit_status": "verified", "notes": "Displayed exactly as 59.04. Official Liquid AI reported result." }, { "model_id": "lfm2.5-2.6b", "benchmark_id": "aime_2025", "score": 51.87, "reference_url": "https://www.liquid.ai/blog/lfm2-5-2-6b", "reported_setting": { "mode": "thinking", "effort": "source does not state", "tools": "none", "sampling": "source does not state trial count", "judge": "benchmark-specified", "harness": "Liquid AI official comparative evaluation", "prompt_style": "source does not state", "temperature": "source does not state evaluation parameters", "context": "source does not state evaluation context" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Displayed exactly as 51.87. Official Liquid AI reported result." }, { "model_id": "lfm2.5-2.6b", "benchmark_id": "bfcl_v4", "score": 56.88, "reference_url": "https://www.liquid.ai/blog/lfm2-5-2-6b", "reported_setting": { "mode": "thinking", "effort": "source does not state", "tools": "benchmark-provided tools", "sampling": "source does not state trial count", "judge": "benchmark-specified", "harness": "Liquid AI official comparative evaluation", "prompt_style": "source does not state", "temperature": "source does not state evaluation parameters", "context": "source does not state evaluation context" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Displayed exactly as 56.88. Official Liquid AI reported result." }, { "model_id": "lfm2.5-2.6b", "benchmark_id": "browsecomp_plus_openclaw", "score": 26.89, "reference_url": "https://www.liquid.ai/blog/lfm2-5-2-6b", "reported_setting": { "mode": "thinking", "effort": "source does not state", "tools": "OpenClaw search/retrieval agent over fixed corpus", "sampling": "source does not state trial count", "judge": "exact-answer benchmark evaluation", "harness": "BrowseComp-Plus fixed-corpus evaluation", "prompt_style": "source does not state", "temperature": "source does not state evaluation parameters", "context": "source does not state evaluation context" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Displayed exactly as 26.89. Official Liquid AI reported result." }, { "model_id": "lfm2.5-2.6b", "benchmark_id": "claw_eval_en_average", "score": 62.85, "reference_url": "https://www.liquid.ai/blog/lfm2-5-2-6b", "reported_setting": { "mode": "thinking", "effort": "source does not state", "tools": "OpenClaw agent environment", "sampling": "Pass^3; N=3 independent trials", "judge": "benchmark-specified", "harness": "Claw-Eval English tasks across all three public splits", "prompt_style": "source does not state", "temperature": "source does not state evaluation parameters", "context": "source does not state evaluation context" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Displayed exactly as 62.85. Official Liquid AI reported result." }, { "model_id": "lfm2.5-2.6b", "benchmark_id": "ifbench", "score": 59.17, "reference_url": "https://www.liquid.ai/blog/lfm2-5-2-6b", "reported_setting": { "mode": "thinking", "effort": "source does not state", "tools": "none", "sampling": "source does not state trial count", "judge": "benchmark-specified", "harness": "Liquid AI official comparative evaluation", "prompt_style": "source does not state", "temperature": "source does not state evaluation parameters", "context": "source does not state evaluation context" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Displayed exactly as 59.17. Official Liquid AI reported result." }, { "model_id": "lfm2.5-2.6b", "benchmark_id": "ifstruct_v1", "score": 85.49, "reference_url": "https://www.liquid.ai/blog/lfm2-5-2-6b", "reported_setting": { "mode": "thinking", "effort": "source does not state", "tools": "none", "sampling": "source does not state trial count", "judge": "binary schema and structural validator", "harness": "IFStruct v1.0 validator without constrained decoding", "prompt_style": "source does not state", "temperature": "source does not state evaluation parameters", "context": "source does not state evaluation context" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Displayed exactly as 85.49. Official Liquid AI reported result." }, { "model_id": "lfm2.5-2.6b", "benchmark_id": "livecodebench_v6", "score": 59.41, "reference_url": "https://www.liquid.ai/blog/lfm2-5-2-6b", "reported_setting": { "mode": "thinking", "effort": "source does not state", "tools": "none", "sampling": "source does not state trial count", "judge": "benchmark-specified", "harness": "Liquid AI official comparative evaluation", "prompt_style": "source does not state", "temperature": "source does not state evaluation parameters", "context": "source does not state evaluation context" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Displayed exactly as 59.41. Official Liquid AI reported result." }, { "model_id": "lfm2.5-2.6b", "benchmark_id": "multi_if", "score": 80.07, "reference_url": "https://www.liquid.ai/blog/lfm2-5-2-6b", "reported_setting": { "mode": "thinking", "effort": "source does not state", "tools": "none", "sampling": "source does not state trial count", "judge": "benchmark-specified", "harness": "Liquid AI official comparative evaluation", "prompt_style": "source does not state", "temperature": "source does not state evaluation parameters", "context": "source does not state evaluation context" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Displayed exactly as 80.07. Official Liquid AI reported result." }, { "model_id": "lfm2.5-2.6b", "benchmark_id": "pinchbench", "score": 68.22, "reference_url": "https://www.liquid.ai/blog/lfm2-5-2-6b", "reported_setting": { "mode": "thinking", "effort": "source does not state", "tools": "OpenClaw coding environment", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "Liquid AI official comparative evaluation", "prompt_style": "source does not state", "temperature": "source does not state evaluation parameters", "context": "source does not state evaluation context" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Displayed exactly as 68.22. Official Liquid AI reported result." }, { "model_id": "lfm2.5-2.6b", "benchmark_id": "tau3_banking", "score": 5.67, "reference_url": "https://www.liquid.ai/blog/lfm2-5-2-6b", "reported_setting": { "mode": "thinking", "effort": "source does not state", "tools": "benchmark-provided tools", "sampling": "source does not state trial count", "judge": "benchmark-specified", "harness": "Liquid AI official comparative evaluation", "prompt_style": "source does not state", "temperature": "source does not state evaluation parameters", "context": "source does not state evaluation context" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Displayed exactly as 5.67. Official Liquid AI reported result." }, { "model_id": "lfm2.5-2.6b", "benchmark_id": "tool_sandbox", "score": 77.83, "reference_url": "https://www.liquid.ai/blog/lfm2-5-2-6b", "reported_setting": { "mode": "thinking", "effort": "source does not state", "tools": "benchmark-provided tools", "sampling": "source does not state trial count", "judge": "milestone and minefield similarity evaluation", "harness": "Apple ToolSandbox stateful conversational harness", "prompt_style": "source does not state", "temperature": "source does not state evaluation parameters", "context": "source does not state evaluation context" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Displayed exactly as 77.83. Official Liquid AI reported result." }, { "model_id": "lfm2.5-230m", "benchmark_id": "bfcl_v3", "score": 43.26, "reference_url": "https://www.liquid.ai/blog/lfm2-5-230m", "reported_setting": { "mode": "non-thinking", "effort": "source does not state", "tools": "benchmark-provided tools", "sampling": "source does not state trial count", "judge": "benchmark-specified", "harness": "Liquid AI official comparative evaluation", "prompt_style": "source does not state", "temperature": "source does not state evaluation parameters", "context": "source does not state evaluation context" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Displayed exactly as 43.26. Official Liquid AI reported result." }, { "model_id": "lfm2.5-230m", "benchmark_id": "bfcl_v4", "score": 21.03, "reference_url": "https://www.liquid.ai/blog/lfm2-5-230m", "reported_setting": { "mode": "non-thinking", "effort": "source does not state", "tools": "benchmark-provided tools", "sampling": "source does not state trial count", "judge": "benchmark-specified", "harness": "Liquid AI official comparative evaluation", "prompt_style": "source does not state", "temperature": "source does not state evaluation parameters", "context": "source does not state evaluation context" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Displayed exactly as 21.03. Official Liquid AI reported result." }, { "model_id": "lfm2.5-230m", "benchmark_id": "gpqa_diamond", "score": 25.41, "reference_url": "https://www.liquid.ai/blog/lfm2-5-230m", "reported_setting": { "mode": "non-thinking", "effort": "source does not state", "tools": "none", "sampling": "source does not state trial count", "judge": "benchmark-specified", "harness": "Liquid AI official comparative evaluation", "prompt_style": "source does not state", "temperature": "source does not state evaluation parameters", "context": "source does not state evaluation context" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Displayed exactly as 25.41. Official Liquid AI reported result." }, { "model_id": "lfm2.5-230m", "benchmark_id": "ifbench", "score": 38.4, "reference_url": "https://www.liquid.ai/blog/lfm2-5-230m", "reported_setting": { "mode": "non-thinking", "effort": "source does not state", "tools": "none", "sampling": "source does not state trial count", "judge": "benchmark-specified", "harness": "Liquid AI official comparative evaluation", "prompt_style": "source does not state", "temperature": "source does not state evaluation parameters", "context": "source does not state evaluation context" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Displayed exactly as 38.40. Official Liquid AI reported result." }, { "model_id": "lfm2.5-230m", "benchmark_id": "ifeval", "score": 71.71, "reference_url": "https://www.liquid.ai/blog/lfm2-5-230m", "reported_setting": { "mode": "non-thinking", "effort": "source does not state", "tools": "none", "sampling": "source does not state trial count", "judge": "benchmark-specified", "harness": "Liquid AI official comparative evaluation", "prompt_style": "source does not state", "temperature": "source does not state evaluation parameters", "context": "source does not state evaluation context" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Displayed exactly as 71.71. Official Liquid AI reported result." }, { "model_id": "lfm2.5-230m", "benchmark_id": "mmlu_pro", "score": 20.25, "reference_url": "https://www.liquid.ai/blog/lfm2-5-230m", "reported_setting": { "mode": "non-thinking", "effort": "source does not state", "tools": "none", "sampling": "source does not state trial count", "judge": "benchmark-specified", "harness": "Liquid AI official comparative evaluation", "prompt_style": "source does not state", "temperature": "source does not state evaluation parameters", "context": "source does not state evaluation context" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Displayed exactly as 20.25. Official Liquid AI reported result." }, { "model_id": "lfm2.5-230m", "benchmark_id": "multi_if", "score": 37.7, "reference_url": "https://www.liquid.ai/blog/lfm2-5-230m", "reported_setting": { "mode": "non-thinking", "effort": "source does not state", "tools": "none", "sampling": "source does not state trial count", "judge": "benchmark-specified", "harness": "Liquid AI official comparative evaluation", "prompt_style": "source does not state", "temperature": "source does not state evaluation parameters", "context": "source does not state evaluation context" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Displayed exactly as 37.70. Official Liquid AI reported result." }, { "model_id": "lfm2.5-230m", "benchmark_id": "tau2_bench_avg", "score": 16.98, "reference_url": "https://aypchzzf9pftwuto.public.blob.vercel-storage.com/lfm2_5_230m_benchmarks-EcJisF1zm7I7KiST75d7XsE2LwE0lD.png", "reported_setting": { "mode": "non-thinking", "effort": "source does not state", "tools": "benchmark-provided tools", "sampling": "source does not state trial count", "judge": "benchmark-specified", "harness": "Liquid AI official comparative evaluation", "prompt_style": "source does not state", "temperature": "source does not state evaluation parameters", "context": "source does not state evaluation context" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Displayed exactly as 16.98. Official Liquid AI reported result." }, { "model_id": "lfm2.5-230m", "benchmark_id": "tau2_bench_retail", "score": 13.68, "reference_url": "https://www.liquid.ai/blog/lfm2-5-230m", "reported_setting": { "mode": "non-thinking", "effort": "source does not state", "tools": "benchmark-provided tools", "sampling": "source does not state trial count", "judge": "benchmark-specified", "harness": "Liquid AI official comparative evaluation", "prompt_style": "source does not state", "temperature": "source does not state evaluation parameters", "context": "source does not state evaluation context" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Displayed exactly as 13.68. Official Liquid AI reported result." }, { "model_id": "lfm2.5-230m", "benchmark_id": "tau2_bench_telecom", "score": 5.26, "reference_url": "https://www.liquid.ai/blog/lfm2-5-230m", "reported_setting": { "mode": "non-thinking", "effort": "source does not state", "tools": "benchmark-provided tools", "sampling": "source does not state trial count", "judge": "benchmark-specified", "harness": "Liquid AI official comparative evaluation", "prompt_style": "source does not state", "temperature": "source does not state evaluation parameters", "context": "source does not state evaluation context" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Displayed exactly as 5.26. Official Liquid AI reported result." }, { "model_id": "lfm2.5-350m", "benchmark_id": "bfcl_v3", "score": 44.11, "reference_url": "https://www.liquid.ai/blog/lfm2-5-230m", "reported_setting": { "mode": "source/model default", "effort": "source does not state", "tools": "benchmark-provided tools", "sampling": "source does not state trial count", "judge": "benchmark-specified", "harness": "Liquid AI official comparative evaluation", "prompt_style": "source does not state", "temperature": "source does not state evaluation parameters", "context": "source does not state evaluation context" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Displayed exactly as 44.11. Official Liquid AI reported result." }, { "model_id": "lfm2.5-350m", "benchmark_id": "bfcl_v4", "score": 21.86, "reference_url": "https://www.liquid.ai/blog/lfm2-5-230m", "reported_setting": { "mode": "source/model default", "effort": "source does not state", "tools": "benchmark-provided tools", "sampling": "source does not state trial count", "judge": "benchmark-specified", "harness": "Liquid AI official comparative evaluation", "prompt_style": "source does not state", "temperature": "source does not state evaluation parameters", "context": "source does not state evaluation context" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Displayed exactly as 21.86. Official Liquid AI reported result." }, { "model_id": "lfm2.5-350m", "benchmark_id": "gpqa_diamond", "score": 30.64, "reference_url": "https://www.liquid.ai/blog/lfm2-5-230m", "reported_setting": { "mode": "source/model default", "effort": "source does not state", "tools": "none", "sampling": "source does not state trial count", "judge": "benchmark-specified", "harness": "Liquid AI official comparative evaluation", "prompt_style": "source does not state", "temperature": "source does not state evaluation parameters", "context": "source does not state evaluation context" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Displayed exactly as 30.64. Official Liquid AI reported result." }, { "model_id": "lfm2.5-350m", "benchmark_id": "ifbench", "score": 40.69, "reference_url": "https://www.liquid.ai/blog/lfm2-5-230m", "reported_setting": { "mode": "source/model default", "effort": "source does not state", "tools": "none", "sampling": "source does not state trial count", "judge": "benchmark-specified", "harness": "Liquid AI official comparative evaluation", "prompt_style": "source does not state", "temperature": "source does not state evaluation parameters", "context": "source does not state evaluation context" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Displayed exactly as 40.69. Official Liquid AI reported result." }, { "model_id": "lfm2.5-350m", "benchmark_id": "ifeval", "score": 76.96, "reference_url": "https://www.liquid.ai/blog/lfm2-5-230m", "reported_setting": { "mode": "source/model default", "effort": "source does not state", "tools": "none", "sampling": "source does not state trial count", "judge": "benchmark-specified", "harness": "Liquid AI official comparative evaluation", "prompt_style": "source does not state", "temperature": "source does not state evaluation parameters", "context": "source does not state evaluation context" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Displayed exactly as 76.96. Official Liquid AI reported result." }, { "model_id": "lfm2.5-350m", "benchmark_id": "mmlu_pro", "score": 20.01, "reference_url": "https://www.liquid.ai/blog/lfm2-5-230m", "reported_setting": { "mode": "source/model default", "effort": "source does not state", "tools": "none", "sampling": "source does not state trial count", "judge": "benchmark-specified", "harness": "Liquid AI official comparative evaluation", "prompt_style": "source does not state", "temperature": "source does not state evaluation parameters", "context": "source does not state evaluation context" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Displayed exactly as 20.01. Official Liquid AI reported result." }, { "model_id": "lfm2.5-350m", "benchmark_id": "multi_if", "score": 44.92, "reference_url": "https://www.liquid.ai/blog/lfm2-5-230m", "reported_setting": { "mode": "source/model default", "effort": "source does not state", "tools": "none", "sampling": "source does not state trial count", "judge": "benchmark-specified", "harness": "Liquid AI official comparative evaluation", "prompt_style": "source does not state", "temperature": "source does not state evaluation parameters", "context": "source does not state evaluation context" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Displayed exactly as 44.92. Official Liquid AI reported result." }, { "model_id": "lfm2.5-350m", "benchmark_id": "tau2_bench_avg", "score": 23.09, "reference_url": "https://aypchzzf9pftwuto.public.blob.vercel-storage.com/lfm2_5_230m_benchmarks-EcJisF1zm7I7KiST75d7XsE2LwE0lD.png", "reported_setting": { "mode": "source/model default", "effort": "source does not state", "tools": "benchmark-provided tools", "sampling": "source does not state trial count", "judge": "benchmark-specified", "harness": "Liquid AI official comparative evaluation", "prompt_style": "source does not state", "temperature": "source does not state evaluation parameters", "context": "source does not state evaluation context" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Displayed exactly as 23.09. Official Liquid AI reported result." }, { "model_id": "lfm2.5-350m", "benchmark_id": "tau2_bench_retail", "score": 17.84, "reference_url": "https://www.liquid.ai/blog/lfm2-5-230m", "reported_setting": { "mode": "source/model default", "effort": "source does not state", "tools": "benchmark-provided tools", "sampling": "source does not state trial count", "judge": "benchmark-specified", "harness": "Liquid AI official comparative evaluation", "prompt_style": "source does not state", "temperature": "source does not state evaluation parameters", "context": "source does not state evaluation context" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Displayed exactly as 17.84. Official Liquid AI reported result." }, { "model_id": "lfm2.5-350m", "benchmark_id": "tau2_bench_telecom", "score": 18.86, "reference_url": "https://www.liquid.ai/blog/lfm2-5-230m", "reported_setting": { "mode": "source/model default", "effort": "source does not state", "tools": "benchmark-provided tools", "sampling": "source does not state trial count", "judge": "benchmark-specified", "harness": "Liquid AI official comparative evaluation", "prompt_style": "source does not state", "temperature": "source does not state evaluation parameters", "context": "source does not state evaluation context" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Displayed exactly as 18.86. Official Liquid AI reported result." }, { "model_id": "lfm2.5-8b-a1b", "benchmark_id": "aa_omniscience_accuracy", "score": 8.67, "reference_url": "https://www.liquid.ai/blog/lfm2-5-8b-a1b", "reported_setting": { "mode": "thinking", "effort": "source does not state", "tools": "none", "sampling": "source does not state trial count", "judge": "official answer and abstention classification", "harness": "Artificial Analysis Omniscience evaluation reported by Liquid", "prompt_style": "source does not state", "temperature": "source does not state evaluation parameters", "context": "source does not state evaluation context" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Displayed exactly as 8.67. Official Liquid AI reported result." }, { "model_id": "lfm2.5-8b-a1b", "benchmark_id": "aa_omniscience_index", "score": -24.7, "reference_url": "https://www.liquid.ai/blog/lfm2-5-8b-a1b", "reported_setting": { "mode": "thinking", "effort": "source does not state", "tools": "none", "sampling": "source does not state trial count", "judge": "official answer and abstention classification", "harness": "Artificial Analysis Omniscience evaluation reported by Liquid", "prompt_style": "source does not state", "temperature": "source does not state evaluation parameters", "context": "source does not state evaluation context" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Displayed exactly as -24.70. Official Liquid AI reported result." }, { "model_id": "lfm2.5-8b-a1b", "benchmark_id": "aa_omniscience_non_hallucination", "score": 63.47, "reference_url": "https://www.liquid.ai/blog/lfm2-5-8b-a1b", "reported_setting": { "mode": "thinking", "effort": "source does not state", "tools": "none", "sampling": "source does not state trial count", "judge": "official answer and abstention classification", "harness": "Artificial Analysis Omniscience evaluation reported by Liquid", "prompt_style": "source does not state", "temperature": "source does not state evaluation parameters", "context": "source does not state evaluation context" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Displayed exactly as 63.47. Official Liquid AI reported result." }, { "model_id": "lfm2.5-8b-a1b", "benchmark_id": "aime_2025", "score": 42.53, "reference_url": "https://www.liquid.ai/blog/lfm2-5-8b-a1b", "reported_setting": { "mode": "thinking", "effort": "source does not state", "tools": "none", "sampling": "source does not state trial count", "judge": "benchmark-specified", "harness": "Liquid AI official comparative evaluation", "prompt_style": "source does not state", "temperature": "source does not state evaluation parameters", "context": "source does not state evaluation context" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Displayed exactly as 42.53. Official Liquid AI reported result." }, { "model_id": "lfm2.5-8b-a1b", "benchmark_id": "aime_2026", "score": 50.0, "reference_url": "https://www.liquid.ai/blog/lfm2-5-8b-a1b", "reported_setting": { "mode": "thinking", "effort": "source does not state", "tools": "none", "sampling": "source does not state trial count", "judge": "benchmark-specified", "harness": "Liquid AI official comparative evaluation", "prompt_style": "source does not state", "temperature": "source does not state evaluation parameters", "context": "source does not state evaluation context" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Displayed exactly as 50.00. Official Liquid AI reported result." }, { "model_id": "lfm2.5-8b-a1b", "benchmark_id": "bfcl_v3", "score": 64.79, "reference_url": "https://www.liquid.ai/blog/lfm2-5-8b-a1b", "reported_setting": { "mode": "thinking", "effort": "source does not state", "tools": "benchmark-provided tools", "sampling": "source does not state trial count", "judge": "benchmark-specified", "harness": "Liquid AI official comparative evaluation", "prompt_style": "source does not state", "temperature": "source does not state evaluation parameters", "context": "source does not state evaluation context" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Displayed exactly as 64.79. Official Liquid AI reported result.", "candidates": [ { "score": 64.36, "reference_url": "https://www.liquid.ai/blog/lfm2-5-8b-a1b", "source_type": "official_blog", "reported_setting": { "mode": "thinking", "effort": "source does not state", "tools": "benchmark-provided tools", "sampling": "source does not state trial count", "judge": "benchmark-specified", "harness": "Liquid AI official comparative evaluation", "prompt_style": "source does not state", "temperature": "source does not state evaluation parameters", "context": "source does not state evaluation context" }, "notes": "Displayed exactly as 64.36. Official Liquid AI reported result." } ] }, { "model_id": "lfm2.5-8b-a1b", "benchmark_id": "bfcl_v4", "score": 49.73, "reference_url": "https://www.liquid.ai/blog/lfm2-5-8b-a1b", "reported_setting": { "mode": "thinking", "effort": "source does not state", "tools": "benchmark-provided tools", "sampling": "source does not state trial count", "judge": "benchmark-specified", "harness": "Liquid AI official comparative evaluation", "prompt_style": "source does not state", "temperature": "source does not state evaluation parameters", "context": "source does not state evaluation context" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Displayed exactly as 49.73. Official Liquid AI reported result.", "candidates": [ { "score": 48.5, "reference_url": "https://www.liquid.ai/blog/lfm2-5-8b-a1b", "source_type": "official_blog", "reported_setting": { "mode": "thinking", "effort": "source does not state", "tools": "benchmark-provided tools", "sampling": "source does not state trial count", "judge": "benchmark-specified", "harness": "Liquid AI official comparative evaluation", "prompt_style": "source does not state", "temperature": "source does not state evaluation parameters", "context": "source does not state evaluation context" }, "notes": "Displayed exactly as 48.50. Official Liquid AI reported result." } ] }, { "model_id": "lfm2.5-8b-a1b", "benchmark_id": "ifbench", "score": 56.47, "reference_url": "https://www.liquid.ai/blog/lfm2-5-8b-a1b", "reported_setting": { "mode": "thinking", "effort": "source does not state", "tools": "none", "sampling": "source does not state trial count", "judge": "benchmark-specified", "harness": "Liquid AI official comparative evaluation", "prompt_style": "source does not state", "temperature": "source does not state evaluation parameters", "context": "source does not state evaluation context" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Displayed exactly as 56.47. Official Liquid AI reported result." }, { "model_id": "lfm2.5-8b-a1b", "benchmark_id": "ifeval", "score": 91.84, "reference_url": "https://www.liquid.ai/blog/lfm2-5-8b-a1b", "reported_setting": { "mode": "thinking", "effort": "source does not state", "tools": "none", "sampling": "source does not state trial count", "judge": "benchmark-specified", "harness": "Liquid AI official comparative evaluation", "prompt_style": "source does not state", "temperature": "source does not state evaluation parameters", "context": "source does not state evaluation context" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Displayed exactly as 91.84. Official Liquid AI reported result." }, { "model_id": "lfm2.5-8b-a1b", "benchmark_id": "math_500", "score": 88.76, "reference_url": "https://www.liquid.ai/blog/lfm2-5-8b-a1b", "reported_setting": { "mode": "thinking", "effort": "source does not state", "tools": "none", "sampling": "source does not state trial count", "judge": "benchmark-specified", "harness": "Liquid AI official comparative evaluation", "prompt_style": "source does not state", "temperature": "source does not state evaluation parameters", "context": "source does not state evaluation context" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Displayed exactly as 88.76. Official Liquid AI reported result." }, { "model_id": "lfm2.5-8b-a1b", "benchmark_id": "multi_if", "score": 79.93, "reference_url": "https://www.liquid.ai/blog/lfm2-5-8b-a1b", "reported_setting": { "mode": "thinking", "effort": "source does not state", "tools": "none", "sampling": "source does not state trial count", "judge": "benchmark-specified", "harness": "Liquid AI official comparative evaluation", "prompt_style": "source does not state", "temperature": "source does not state evaluation parameters", "context": "source does not state evaluation context" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Displayed exactly as 79.93. Official Liquid AI reported result." }, { "model_id": "lfm2.5-8b-a1b", "benchmark_id": "tau2_bench_retail", "score": 39.82, "reference_url": "https://www.liquid.ai/blog/lfm2-5-8b-a1b", "reported_setting": { "mode": "thinking", "effort": "source does not state", "tools": "benchmark-provided tools", "sampling": "source does not state trial count", "judge": "benchmark-specified", "harness": "Liquid AI official comparative evaluation", "prompt_style": "source does not state", "temperature": "source does not state evaluation parameters", "context": "source does not state evaluation context" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Displayed exactly as 39.82. Official Liquid AI reported result." }, { "model_id": "lfm2.5-8b-a1b", "benchmark_id": "tau2_bench_telecom", "score": 88.07, "reference_url": "https://www.liquid.ai/blog/lfm2-5-8b-a1b", "reported_setting": { "mode": "thinking", "effort": "source does not state", "tools": "benchmark-provided tools", "sampling": "source does not state trial count", "judge": "benchmark-specified", "harness": "Liquid AI official comparative evaluation", "prompt_style": "source does not state", "temperature": "source does not state evaluation parameters", "context": "source does not state evaluation context" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Displayed exactly as 88.07. Official Liquid AI reported result." }, { "model_id": "qwen3-30b-a3b-thinking-2507", "benchmark_id": "aa_omniscience_accuracy", "score": 18.8, "reference_url": "https://www.liquid.ai/blog/lfm2-5-8b-a1b", "reported_setting": { "mode": "thinking", "effort": "source does not state", "tools": "none", "sampling": "source does not state trial count", "judge": "official answer and abstention classification", "harness": "Artificial Analysis Omniscience evaluation reported by Liquid", "prompt_style": "source does not state", "temperature": "source does not state evaluation parameters", "context": "source does not state evaluation context" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Displayed exactly as 18.80. Official Liquid AI reported result." }, { "model_id": "qwen3-30b-a3b-thinking-2507", "benchmark_id": "aa_omniscience_index", "score": -51.31, "reference_url": "https://www.liquid.ai/blog/lfm2-5-8b-a1b", "reported_setting": { "mode": "thinking", "effort": "source does not state", "tools": "none", "sampling": "source does not state trial count", "judge": "official answer and abstention classification", "harness": "Artificial Analysis Omniscience evaluation reported by Liquid", "prompt_style": "source does not state", "temperature": "source does not state evaluation parameters", "context": "source does not state evaluation context" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Displayed exactly as -51.31. Official Liquid AI reported result." }, { "model_id": "qwen3-30b-a3b-thinking-2507", "benchmark_id": "aa_omniscience_non_hallucination", "score": 13.87, "reference_url": "https://www.liquid.ai/blog/lfm2-5-8b-a1b", "reported_setting": { "mode": "thinking", "effort": "source does not state", "tools": "none", "sampling": "source does not state trial count", "judge": "official answer and abstention classification", "harness": "Artificial Analysis Omniscience evaluation reported by Liquid", "prompt_style": "source does not state", "temperature": "source does not state evaluation parameters", "context": "source does not state evaluation context" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Displayed exactly as 13.87. Official Liquid AI reported result." }, { "model_id": "qwen3-30b-a3b-thinking-2507", "benchmark_id": "aime_2025", "score": 71.67, "reference_url": "https://www.liquid.ai/blog/lfm2-5-8b-a1b", "reported_setting": { "mode": "thinking", "effort": "source does not state", "tools": "none", "sampling": "source does not state trial count", "judge": "benchmark-specified", "harness": "Liquid AI official comparative evaluation", "prompt_style": "source does not state", "temperature": "source does not state evaluation parameters", "context": "source does not state evaluation context" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Displayed exactly as 71.67. Official Liquid AI reported result." }, { "model_id": "qwen3-30b-a3b-thinking-2507", "benchmark_id": "aime_2026", "score": 66.67, "reference_url": "https://www.liquid.ai/blog/lfm2-5-8b-a1b", "reported_setting": { "mode": "thinking", "effort": "source does not state", "tools": "none", "sampling": "source does not state trial count", "judge": "benchmark-specified", "harness": "Liquid AI official comparative evaluation", "prompt_style": "source does not state", "temperature": "source does not state evaluation parameters", "context": "source does not state evaluation context" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Displayed exactly as 66.67. Official Liquid AI reported result." }, { "model_id": "qwen3-30b-a3b-thinking-2507", "benchmark_id": "bfcl_v3", "score": 73.39, "reference_url": "https://www.liquid.ai/blog/lfm2-5-8b-a1b", "reported_setting": { "mode": "thinking", "effort": "source does not state", "tools": "benchmark-provided tools", "sampling": "source does not state trial count", "judge": "benchmark-specified", "harness": "Liquid AI official comparative evaluation", "prompt_style": "source does not state", "temperature": "source does not state evaluation parameters", "context": "source does not state evaluation context" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Displayed exactly as 73.39. Official Liquid AI reported result." }, { "model_id": "qwen3-30b-a3b-thinking-2507", "benchmark_id": "bfcl_v4", "score": 50.53, "reference_url": "https://www.liquid.ai/blog/lfm2-5-8b-a1b", "reported_setting": { "mode": "thinking", "effort": "source does not state", "tools": "benchmark-provided tools", "sampling": "source does not state trial count", "judge": "benchmark-specified", "harness": "Liquid AI official comparative evaluation", "prompt_style": "source does not state", "temperature": "source does not state evaluation parameters", "context": "source does not state evaluation context" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Displayed exactly as 50.53. Official Liquid AI reported result." }, { "model_id": "qwen3-30b-a3b-thinking-2507", "benchmark_id": "ifbench", "score": 51.11, "reference_url": "https://www.liquid.ai/blog/lfm2-5-8b-a1b", "reported_setting": { "mode": "thinking", "effort": "source does not state", "tools": "none", "sampling": "source does not state trial count", "judge": "benchmark-specified", "harness": "Liquid AI official comparative evaluation", "prompt_style": "source does not state", "temperature": "source does not state evaluation parameters", "context": "source does not state evaluation context" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Displayed exactly as 51.11. Official Liquid AI reported result." }, { "model_id": "qwen3-30b-a3b-thinking-2507", "benchmark_id": "ifeval", "score": 90.82, "reference_url": "https://www.liquid.ai/blog/lfm2-5-8b-a1b", "reported_setting": { "mode": "thinking", "effort": "source does not state", "tools": "none", "sampling": "source does not state trial count", "judge": "benchmark-specified", "harness": "Liquid AI official comparative evaluation", "prompt_style": "source does not state", "temperature": "source does not state evaluation parameters", "context": "source does not state evaluation context" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Displayed exactly as 90.82. Official Liquid AI reported result." }, { "model_id": "qwen3-30b-a3b-thinking-2507", "benchmark_id": "math_500", "score": 86.48, "reference_url": "https://www.liquid.ai/blog/lfm2-5-8b-a1b", "reported_setting": { "mode": "thinking", "effort": "source does not state", "tools": "none", "sampling": "source does not state trial count", "judge": "benchmark-specified", "harness": "Liquid AI official comparative evaluation", "prompt_style": "source does not state", "temperature": "source does not state evaluation parameters", "context": "source does not state evaluation context" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Displayed exactly as 86.48. Official Liquid AI reported result." }, { "model_id": "qwen3-30b-a3b-thinking-2507", "benchmark_id": "multi_if", "score": 79.04, "reference_url": "https://www.liquid.ai/blog/lfm2-5-8b-a1b", "reported_setting": { "mode": "thinking", "effort": "source does not state", "tools": "none", "sampling": "source does not state trial count", "judge": "benchmark-specified", "harness": "Liquid AI official comparative evaluation", "prompt_style": "source does not state", "temperature": "source does not state evaluation parameters", "context": "source does not state evaluation context" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Displayed exactly as 79.04. Official Liquid AI reported result." }, { "model_id": "qwen3-30b-a3b-thinking-2507", "benchmark_id": "tau2_bench_retail", "score": 56.14, "reference_url": "https://www.liquid.ai/blog/lfm2-5-8b-a1b", "reported_setting": { "mode": "thinking", "effort": "source does not state", "tools": "benchmark-provided tools", "sampling": "source does not state trial count", "judge": "benchmark-specified", "harness": "Liquid AI official comparative evaluation", "prompt_style": "source does not state", "temperature": "source does not state evaluation parameters", "context": "source does not state evaluation context" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Displayed exactly as 56.14. Official Liquid AI reported result." }, { "model_id": "qwen3-30b-a3b-thinking-2507", "benchmark_id": "tau2_bench_telecom", "score": 21.93, "reference_url": "https://www.liquid.ai/blog/lfm2-5-8b-a1b", "reported_setting": { "mode": "thinking", "effort": "source does not state", "tools": "benchmark-provided tools", "sampling": "source does not state trial count", "judge": "benchmark-specified", "harness": "Liquid AI official comparative evaluation", "prompt_style": "source does not state", "temperature": "source does not state evaluation parameters", "context": "source does not state evaluation context" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Displayed exactly as 21.93. Official Liquid AI reported result." }, { "model_id": "qwen3.5-0.8b-instruct", "benchmark_id": "bfcl_v3", "score": 35.08, "reference_url": "https://www.liquid.ai/blog/lfm2-5-230m", "reported_setting": { "mode": "non-thinking", "effort": "source does not state", "tools": "benchmark-provided tools", "sampling": "source does not state trial count", "judge": "benchmark-specified", "harness": "Liquid AI official comparative evaluation", "prompt_style": "source does not state", "temperature": "source does not state evaluation parameters", "context": "source does not state evaluation context" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Displayed exactly as 35.08. Official Liquid AI reported result." }, { "model_id": "qwen3.5-0.8b-instruct", "benchmark_id": "bfcl_v4", "score": 18.7, "reference_url": "https://www.liquid.ai/blog/lfm2-5-230m", "reported_setting": { "mode": "non-thinking", "effort": "source does not state", "tools": "benchmark-provided tools", "sampling": "source does not state trial count", "judge": "benchmark-specified", "harness": "Liquid AI official comparative evaluation", "prompt_style": "source does not state", "temperature": "source does not state evaluation parameters", "context": "source does not state evaluation context" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Displayed exactly as 18.70. Official Liquid AI reported result." }, { "model_id": "qwen3.5-0.8b-instruct", "benchmark_id": "gpqa_diamond", "score": 27.41, "reference_url": "https://www.liquid.ai/blog/lfm2-5-230m", "reported_setting": { "mode": "non-thinking", "effort": "source does not state", "tools": "none", "sampling": "source does not state trial count", "judge": "benchmark-specified", "harness": "Liquid AI official comparative evaluation", "prompt_style": "source does not state", "temperature": "source does not state evaluation parameters", "context": "source does not state evaluation context" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Displayed exactly as 27.41. Official Liquid AI reported result." }, { "model_id": "qwen3.5-0.8b-instruct", "benchmark_id": "ifbench", "score": 22.87, "reference_url": "https://www.liquid.ai/blog/lfm2-5-230m", "reported_setting": { "mode": "non-thinking", "effort": "source does not state", "tools": "none", "sampling": "source does not state trial count", "judge": "benchmark-specified", "harness": "Liquid AI official comparative evaluation", "prompt_style": "source does not state", "temperature": "source does not state evaluation parameters", "context": "source does not state evaluation context" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Displayed exactly as 22.87. Official Liquid AI reported result." }, { "model_id": "qwen3.5-0.8b-instruct", "benchmark_id": "ifeval", "score": 59.94, "reference_url": "https://www.liquid.ai/blog/lfm2-5-230m", "reported_setting": { "mode": "non-thinking", "effort": "source does not state", "tools": "none", "sampling": "source does not state trial count", "judge": "benchmark-specified", "harness": "Liquid AI official comparative evaluation", "prompt_style": "source does not state", "temperature": "source does not state evaluation parameters", "context": "source does not state evaluation context" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Displayed exactly as 59.94. Official Liquid AI reported result." }, { "model_id": "qwen3.5-0.8b-instruct", "benchmark_id": "mmlu_pro", "score": 37.42, "reference_url": "https://www.liquid.ai/blog/lfm2-5-230m", "reported_setting": { "mode": "non-thinking", "effort": "source does not state", "tools": "none", "sampling": "source does not state trial count", "judge": "benchmark-specified", "harness": "Liquid AI official comparative evaluation", "prompt_style": "source does not state", "temperature": "source does not state evaluation parameters", "context": "source does not state evaluation context" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Displayed exactly as 37.42. Official Liquid AI reported result." }, { "model_id": "qwen3.5-0.8b-instruct", "benchmark_id": "multi_if", "score": 41.68, "reference_url": "https://www.liquid.ai/blog/lfm2-5-230m", "reported_setting": { "mode": "non-thinking", "effort": "source does not state", "tools": "none", "sampling": "source does not state trial count", "judge": "benchmark-specified", "harness": "Liquid AI official comparative evaluation", "prompt_style": "source does not state", "temperature": "source does not state evaluation parameters", "context": "source does not state evaluation context" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Displayed exactly as 41.68. Official Liquid AI reported result." }, { "model_id": "qwen3.5-0.8b-instruct", "benchmark_id": "tau2_bench_avg", "score": 20.9, "reference_url": "https://aypchzzf9pftwuto.public.blob.vercel-storage.com/lfm2_5_230m_benchmarks-EcJisF1zm7I7KiST75d7XsE2LwE0lD.png", "reported_setting": { "mode": "non-thinking", "effort": "source does not state", "tools": "benchmark-provided tools", "sampling": "source does not state trial count", "judge": "benchmark-specified", "harness": "Liquid AI official comparative evaluation", "prompt_style": "source does not state", "temperature": "source does not state evaluation parameters", "context": "source does not state evaluation context" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Displayed exactly as 20.9. Official Liquid AI reported result." }, { "model_id": "qwen3.5-0.8b-instruct", "benchmark_id": "tau2_bench_retail", "score": 6.14, "reference_url": "https://www.liquid.ai/blog/lfm2-5-230m", "reported_setting": { "mode": "non-thinking", "effort": "source does not state", "tools": "benchmark-provided tools", "sampling": "source does not state trial count", "judge": "benchmark-specified", "harness": "Liquid AI official comparative evaluation", "prompt_style": "source does not state", "temperature": "source does not state evaluation parameters", "context": "source does not state evaluation context" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Displayed exactly as 6.14. Official Liquid AI reported result." }, { "model_id": "qwen3.5-0.8b-instruct", "benchmark_id": "tau2_bench_telecom", "score": 12.57, "reference_url": "https://www.liquid.ai/blog/lfm2-5-230m", "reported_setting": { "mode": "non-thinking", "effort": "source does not state", "tools": "benchmark-provided tools", "sampling": "source does not state trial count", "judge": "benchmark-specified", "harness": "Liquid AI official comparative evaluation", "prompt_style": "source does not state", "temperature": "source does not state evaluation parameters", "context": "source does not state evaluation context" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Displayed exactly as 12.57. Official Liquid AI reported result." }, { "model_id": "qwen3.5-4b", "benchmark_id": "aa_omniscience_accuracy", "score": 17.2, "reference_url": "https://www.liquid.ai/blog/lfm2-5-8b-a1b", "reported_setting": { "mode": "source/model default", "effort": "source does not state", "tools": "none", "sampling": "source does not state trial count", "judge": "official answer and abstention classification", "harness": "Artificial Analysis Omniscience evaluation reported by Liquid", "prompt_style": "source does not state", "temperature": "source does not state evaluation parameters", "context": "source does not state evaluation context" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Displayed exactly as 17.20. Official Liquid AI reported result." }, { "model_id": "qwen3.5-4b", "benchmark_id": "aa_omniscience_index", "score": -51.53, "reference_url": "https://www.liquid.ai/blog/lfm2-5-8b-a1b", "reported_setting": { "mode": "source/model default", "effort": "source does not state", "tools": "none", "sampling": "source does not state trial count", "judge": "official answer and abstention classification", "harness": "Artificial Analysis Omniscience evaluation reported by Liquid", "prompt_style": "source does not state", "temperature": "source does not state evaluation parameters", "context": "source does not state evaluation context" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Displayed exactly as -51.53. Official Liquid AI reported result." }, { "model_id": "qwen3.5-4b", "benchmark_id": "aa_omniscience_non_hallucination", "score": 16.99, "reference_url": "https://www.liquid.ai/blog/lfm2-5-8b-a1b", "reported_setting": { "mode": "source/model default", "effort": "source does not state", "tools": "none", "sampling": "source does not state trial count", "judge": "official answer and abstention classification", "harness": "Artificial Analysis Omniscience evaluation reported by Liquid", "prompt_style": "source does not state", "temperature": "source does not state evaluation parameters", "context": "source does not state evaluation context" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Displayed exactly as 16.99. Official Liquid AI reported result." }, { "model_id": "qwen3.5-4b", "benchmark_id": "aa_omniscience_public_accuracy", "score": 17.63, "reference_url": "https://huggingface.co/LiquidAI/LFM2.5-2.6B/resolve/ab00687315bc1298e9d54e9c4b611dde9867ccc2/README.md", "reported_setting": { "mode": "source/model default", "effort": "source does not state", "tools": "none", "sampling": "source does not state trial count", "judge": "official answer and abstention classification", "harness": "Artificial Analysis Omniscience evaluation reported by Liquid", "prompt_style": "source does not state", "temperature": "source does not state evaluation parameters", "context": "source does not state evaluation context" }, "matches_canonical": false, "source_type": "official_model_card", "audit_status": "verified", "notes": "Displayed exactly as 17.63. Official Liquid AI reported result." }, { "model_id": "qwen3.5-4b", "benchmark_id": "aa_omniscience_public_index", "score": -54.3, "reference_url": "https://www.liquid.ai/blog/lfm2-5-2-6b", "reported_setting": { "mode": "source/model default", "effort": "source does not state", "tools": "none", "sampling": "source does not state trial count", "judge": "official answer and abstention classification", "harness": "Artificial Analysis Omniscience evaluation reported by Liquid", "prompt_style": "source does not state", "temperature": "source does not state evaluation parameters", "context": "source does not state evaluation context" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Displayed exactly as -54.30. Official Liquid AI reported result." }, { "model_id": "qwen3.5-4b", "benchmark_id": "aa_omniscience_public_non_hallucination", "score": 12.66, "reference_url": "https://huggingface.co/LiquidAI/LFM2.5-2.6B/resolve/ab00687315bc1298e9d54e9c4b611dde9867ccc2/README.md", "reported_setting": { "mode": "source/model default", "effort": "source does not state", "tools": "none", "sampling": "source does not state trial count", "judge": "official answer and abstention classification", "harness": "Artificial Analysis Omniscience evaluation reported by Liquid", "prompt_style": "source does not state", "temperature": "source does not state evaluation parameters", "context": "source does not state evaluation context" }, "matches_canonical": false, "source_type": "official_model_card", "audit_status": "verified", "notes": "Displayed exactly as 12.66. Official Liquid AI reported result." }, { "model_id": "qwen3.5-4b", "benchmark_id": "aime_2025", "score": 54.28, "reference_url": "https://www.liquid.ai/blog/lfm2-5-8b-a1b", "reported_setting": { "mode": "source/model default", "effort": "source does not state", "tools": "none", "sampling": "source does not state trial count", "judge": "benchmark-specified", "harness": "Liquid AI official comparative evaluation", "prompt_style": "source does not state", "temperature": "source does not state evaluation parameters", "context": "source does not state evaluation context" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Displayed exactly as 54.28. Official Liquid AI reported result.", "candidates": [ { "score": 49.33, "reference_url": "https://www.liquid.ai/blog/lfm2-5-2-6b", "source_type": "official_blog", "reported_setting": { "mode": "source/model default", "effort": "source does not state", "tools": "none", "sampling": "source does not state trial count", "judge": "benchmark-specified", "harness": "Liquid AI official comparative evaluation", "prompt_style": "source does not state", "temperature": "source does not state evaluation parameters", "context": "source does not state evaluation context" }, "notes": "Displayed exactly as 49.33. Official Liquid AI reported result." } ] }, { "model_id": "qwen3.5-4b", "benchmark_id": "aime_2026", "score": 58.33, "reference_url": "https://www.liquid.ai/blog/lfm2-5-8b-a1b", "reported_setting": { "mode": "source/model default", "effort": "source does not state", "tools": "none", "sampling": "source does not state trial count", "judge": "benchmark-specified", "harness": "Liquid AI official comparative evaluation", "prompt_style": "source does not state", "temperature": "source does not state evaluation parameters", "context": "source does not state evaluation context" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Displayed exactly as 58.33. Official Liquid AI reported result." }, { "model_id": "qwen3.5-4b", "benchmark_id": "bfcl_v3", "score": 71.06, "reference_url": "https://www.liquid.ai/blog/lfm2-5-8b-a1b", "reported_setting": { "mode": "source/model default", "effort": "source does not state", "tools": "benchmark-provided tools", "sampling": "source does not state trial count", "judge": "benchmark-specified", "harness": "Liquid AI official comparative evaluation", "prompt_style": "source does not state", "temperature": "source does not state evaluation parameters", "context": "source does not state evaluation context" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Displayed exactly as 71.06. Official Liquid AI reported result." }, { "model_id": "qwen3.5-4b", "benchmark_id": "bfcl_v4", "score": 54.01, "reference_url": "https://www.liquid.ai/blog/lfm2-5-8b-a1b", "reported_setting": { "mode": "source/model default", "effort": "source does not state", "tools": "benchmark-provided tools", "sampling": "source does not state trial count", "judge": "benchmark-specified", "harness": "Liquid AI official comparative evaluation", "prompt_style": "source does not state", "temperature": "source does not state evaluation parameters", "context": "source does not state evaluation context" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Displayed exactly as 54.01. Official Liquid AI reported result.", "candidates": [ { "score": 50.56, "reference_url": "https://www.liquid.ai/blog/lfm2-5-2-6b", "source_type": "official_blog", "reported_setting": { "mode": "source/model default", "effort": "source does not state", "tools": "benchmark-provided tools", "sampling": "source does not state trial count", "judge": "benchmark-specified", "harness": "Liquid AI official comparative evaluation", "prompt_style": "source does not state", "temperature": "source does not state evaluation parameters", "context": "source does not state evaluation context" }, "notes": "Displayed exactly as 50.56. Official Liquid AI reported result." } ] }, { "model_id": "qwen3.5-4b", "benchmark_id": "browsecomp_plus_openclaw", "score": 24.46, "reference_url": "https://www.liquid.ai/blog/lfm2-5-2-6b", "reported_setting": { "mode": "source/model default", "effort": "source does not state", "tools": "OpenClaw search/retrieval agent over fixed corpus", "sampling": "source does not state trial count", "judge": "exact-answer benchmark evaluation", "harness": "BrowseComp-Plus fixed-corpus evaluation", "prompt_style": "source does not state", "temperature": "source does not state evaluation parameters", "context": "source does not state evaluation context" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Displayed exactly as 24.46. Official Liquid AI reported result." }, { "model_id": "qwen3.5-4b", "benchmark_id": "claw_eval_en_average", "score": 62.28, "reference_url": "https://www.liquid.ai/blog/lfm2-5-2-6b", "reported_setting": { "mode": "source/model default", "effort": "source does not state", "tools": "OpenClaw agent environment", "sampling": "Pass^3; N=3 independent trials", "judge": "benchmark-specified", "harness": "Claw-Eval English tasks across all three public splits", "prompt_style": "source does not state", "temperature": "source does not state evaluation parameters", "context": "source does not state evaluation context" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Displayed exactly as 62.28. Official Liquid AI reported result." }, { "model_id": "qwen3.5-4b", "benchmark_id": "ifbench", "score": 50.38, "reference_url": "https://www.liquid.ai/blog/lfm2-5-8b-a1b", "reported_setting": { "mode": "source/model default", "effort": "source does not state", "tools": "none", "sampling": "source does not state trial count", "judge": "benchmark-specified", "harness": "Liquid AI official comparative evaluation", "prompt_style": "source does not state", "temperature": "source does not state evaluation parameters", "context": "source does not state evaluation context" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Displayed exactly as 50.38. Official Liquid AI reported result.", "candidates": [ { "score": 48.4, "reference_url": "https://www.liquid.ai/blog/lfm2-5-2-6b", "source_type": "official_blog", "reported_setting": { "mode": "source/model default", "effort": "source does not state", "tools": "none", "sampling": "source does not state trial count", "judge": "benchmark-specified", "harness": "Liquid AI official comparative evaluation", "prompt_style": "source does not state", "temperature": "source does not state evaluation parameters", "context": "source does not state evaluation context" }, "notes": "Displayed exactly as 48.40. Official Liquid AI reported result." } ] }, { "model_id": "qwen3.5-4b", "benchmark_id": "ifeval", "score": 87.8, "reference_url": "https://www.liquid.ai/blog/lfm2-5-8b-a1b", "reported_setting": { "mode": "source/model default", "effort": "source does not state", "tools": "none", "sampling": "source does not state trial count", "judge": "benchmark-specified", "harness": "Liquid AI official comparative evaluation", "prompt_style": "source does not state", "temperature": "source does not state evaluation parameters", "context": "source does not state evaluation context" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Displayed exactly as 87.80. Official Liquid AI reported result." }, { "model_id": "qwen3.5-4b", "benchmark_id": "ifstruct_v1", "score": 36.25, "reference_url": "https://www.liquid.ai/blog/lfm2-5-2-6b", "reported_setting": { "mode": "source/model default", "effort": "source does not state", "tools": "none", "sampling": "source does not state trial count", "judge": "binary schema and structural validator", "harness": "IFStruct v1.0 validator without constrained decoding", "prompt_style": "source does not state", "temperature": "source does not state evaluation parameters", "context": "source does not state evaluation context" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Displayed exactly as 36.25. Official Liquid AI reported result." }, { "model_id": "qwen3.5-4b", "benchmark_id": "livecodebench_v6", "score": 60.85, "reference_url": "https://www.liquid.ai/blog/lfm2-5-2-6b", "reported_setting": { "mode": "source/model default", "effort": "source does not state", "tools": "none", "sampling": "source does not state trial count", "judge": "benchmark-specified", "harness": "Liquid AI official comparative evaluation", "prompt_style": "source does not state", "temperature": "source does not state evaluation parameters", "context": "source does not state evaluation context" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Displayed exactly as 60.85. Official Liquid AI reported result." }, { "model_id": "qwen3.5-4b", "benchmark_id": "math_500", "score": 80.76, "reference_url": "https://www.liquid.ai/blog/lfm2-5-8b-a1b", "reported_setting": { "mode": "source/model default", "effort": "source does not state", "tools": "none", "sampling": "source does not state trial count", "judge": "benchmark-specified", "harness": "Liquid AI official comparative evaluation", "prompt_style": "source does not state", "temperature": "source does not state evaluation parameters", "context": "source does not state evaluation context" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Displayed exactly as 80.76. Official Liquid AI reported result." }, { "model_id": "qwen3.5-4b", "benchmark_id": "multi_if", "score": 67.43, "reference_url": "https://www.liquid.ai/blog/lfm2-5-8b-a1b", "reported_setting": { "mode": "source/model default", "effort": "source does not state", "tools": "none", "sampling": "source does not state trial count", "judge": "benchmark-specified", "harness": "Liquid AI official comparative evaluation", "prompt_style": "source does not state", "temperature": "source does not state evaluation parameters", "context": "source does not state evaluation context" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Displayed exactly as 67.43. Official Liquid AI reported result.", "candidates": [ { "score": 55.67, "reference_url": "https://www.liquid.ai/blog/lfm2-5-2-6b", "source_type": "official_blog", "reported_setting": { "mode": "source/model default", "effort": "source does not state", "tools": "none", "sampling": "source does not state trial count", "judge": "benchmark-specified", "harness": "Liquid AI official comparative evaluation", "prompt_style": "source does not state", "temperature": "source does not state evaluation parameters", "context": "source does not state evaluation context" }, "notes": "Displayed exactly as 55.67. Official Liquid AI reported result." } ] }, { "model_id": "qwen3.5-4b", "benchmark_id": "pinchbench", "score": 71.26, "reference_url": "https://www.liquid.ai/blog/lfm2-5-2-6b", "reported_setting": { "mode": "source/model default", "effort": "source does not state", "tools": "OpenClaw coding environment", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "Liquid AI official comparative evaluation", "prompt_style": "source does not state", "temperature": "source does not state evaluation parameters", "context": "source does not state evaluation context" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Displayed exactly as 71.26. Official Liquid AI reported result." }, { "model_id": "qwen3.5-4b", "benchmark_id": "tau2_bench_retail", "score": 71.93, "reference_url": "https://www.liquid.ai/blog/lfm2-5-8b-a1b", "reported_setting": { "mode": "source/model default", "effort": "source does not state", "tools": "benchmark-provided tools", "sampling": "source does not state trial count", "judge": "benchmark-specified", "harness": "Liquid AI official comparative evaluation", "prompt_style": "source does not state", "temperature": "source does not state evaluation parameters", "context": "source does not state evaluation context" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Displayed exactly as 71.93. Official Liquid AI reported result." }, { "model_id": "qwen3.5-4b", "benchmark_id": "tau2_bench_telecom", "score": 87.72, "reference_url": "https://www.liquid.ai/blog/lfm2-5-8b-a1b", "reported_setting": { "mode": "source/model default", "effort": "source does not state", "tools": "benchmark-provided tools", "sampling": "source does not state trial count", "judge": "benchmark-specified", "harness": "Liquid AI official comparative evaluation", "prompt_style": "source does not state", "temperature": "source does not state evaluation parameters", "context": "source does not state evaluation context" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Displayed exactly as 87.72. Official Liquid AI reported result." }, { "model_id": "qwen3.5-4b", "benchmark_id": "tau3_banking", "score": 5.45, "reference_url": "https://www.liquid.ai/blog/lfm2-5-2-6b", "reported_setting": { "mode": "source/model default", "effort": "source does not state", "tools": "benchmark-provided tools", "sampling": "source does not state trial count", "judge": "benchmark-specified", "harness": "Liquid AI official comparative evaluation", "prompt_style": "source does not state", "temperature": "source does not state evaluation parameters", "context": "source does not state evaluation context" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Displayed exactly as 5.45. Official Liquid AI reported result." }, { "model_id": "qwen3.5-4b", "benchmark_id": "tool_sandbox", "score": 75.55, "reference_url": "https://www.liquid.ai/blog/lfm2-5-2-6b", "reported_setting": { "mode": "source/model default", "effort": "source does not state", "tools": "benchmark-provided tools", "sampling": "source does not state trial count", "judge": "milestone and minefield similarity evaluation", "harness": "Apple ToolSandbox stateful conversational harness", "prompt_style": "source does not state", "temperature": "source does not state evaluation parameters", "context": "source does not state evaluation context" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Displayed exactly as 75.55. Official Liquid AI reported result." }, { "model_id": "qwen3.5-9b", "benchmark_id": "aa_omniscience_public_accuracy", "score": 21.3, "reference_url": "https://huggingface.co/LiquidAI/LFM2.5-2.6B/resolve/ab00687315bc1298e9d54e9c4b611dde9867ccc2/README.md", "reported_setting": { "mode": "source/model default", "effort": "source does not state", "tools": "none", "sampling": "source does not state trial count", "judge": "official answer and abstention classification", "harness": "Artificial Analysis Omniscience evaluation reported by Liquid", "prompt_style": "source does not state", "temperature": "source does not state evaluation parameters", "context": "source does not state evaluation context" }, "matches_canonical": false, "source_type": "official_model_card", "audit_status": "verified", "notes": "Displayed exactly as 21.30. Official Liquid AI reported result." }, { "model_id": "qwen3.5-9b", "benchmark_id": "aa_omniscience_public_index", "score": -50.43, "reference_url": "https://www.liquid.ai/blog/lfm2-5-2-6b", "reported_setting": { "mode": "source/model default", "effort": "source does not state", "tools": "none", "sampling": "source does not state trial count", "judge": "official answer and abstention classification", "harness": "Artificial Analysis Omniscience evaluation reported by Liquid", "prompt_style": "source does not state", "temperature": "source does not state evaluation parameters", "context": "source does not state evaluation context" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Displayed exactly as -50.43. Official Liquid AI reported result." }, { "model_id": "qwen3.5-9b", "benchmark_id": "aa_omniscience_public_non_hallucination", "score": 8.84, "reference_url": "https://huggingface.co/LiquidAI/LFM2.5-2.6B/resolve/ab00687315bc1298e9d54e9c4b611dde9867ccc2/README.md", "reported_setting": { "mode": "source/model default", "effort": "source does not state", "tools": "none", "sampling": "source does not state trial count", "judge": "official answer and abstention classification", "harness": "Artificial Analysis Omniscience evaluation reported by Liquid", "prompt_style": "source does not state", "temperature": "source does not state evaluation parameters", "context": "source does not state evaluation context" }, "matches_canonical": false, "source_type": "official_model_card", "audit_status": "verified", "notes": "Displayed exactly as 8.84. Official Liquid AI reported result." }, { "model_id": "qwen3.5-9b", "benchmark_id": "aime_2025", "score": 56.07, "reference_url": "https://www.liquid.ai/blog/lfm2-5-2-6b", "reported_setting": { "mode": "source/model default", "effort": "source does not state", "tools": "none", "sampling": "source does not state trial count", "judge": "benchmark-specified", "harness": "Liquid AI official comparative evaluation", "prompt_style": "source does not state", "temperature": "source does not state evaluation parameters", "context": "source does not state evaluation context" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Displayed exactly as 56.07. Official Liquid AI reported result." }, { "model_id": "qwen3.5-9b", "benchmark_id": "bfcl_v4", "score": 60.13, "reference_url": "https://www.liquid.ai/blog/lfm2-5-2-6b", "reported_setting": { "mode": "source/model default", "effort": "source does not state", "tools": "benchmark-provided tools", "sampling": "source does not state trial count", "judge": "benchmark-specified", "harness": "Liquid AI official comparative evaluation", "prompt_style": "source does not state", "temperature": "source does not state evaluation parameters", "context": "source does not state evaluation context" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Displayed exactly as 60.13. Official Liquid AI reported result." }, { "model_id": "qwen3.5-9b", "benchmark_id": "browsecomp_plus_openclaw", "score": 27.23, "reference_url": "https://www.liquid.ai/blog/lfm2-5-2-6b", "reported_setting": { "mode": "source/model default", "effort": "source does not state", "tools": "OpenClaw search/retrieval agent over fixed corpus", "sampling": "source does not state trial count", "judge": "exact-answer benchmark evaluation", "harness": "BrowseComp-Plus fixed-corpus evaluation", "prompt_style": "source does not state", "temperature": "source does not state evaluation parameters", "context": "source does not state evaluation context" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Displayed exactly as 27.23. Official Liquid AI reported result." }, { "model_id": "qwen3.5-9b", "benchmark_id": "claw_eval_en_average", "score": 66.53, "reference_url": "https://www.liquid.ai/blog/lfm2-5-2-6b", "reported_setting": { "mode": "source/model default", "effort": "source does not state", "tools": "OpenClaw agent environment", "sampling": "Pass^3; N=3 independent trials", "judge": "benchmark-specified", "harness": "Claw-Eval English tasks across all three public splits", "prompt_style": "source does not state", "temperature": "source does not state evaluation parameters", "context": "source does not state evaluation context" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Displayed exactly as 66.53. Official Liquid AI reported result." }, { "model_id": "qwen3.5-9b", "benchmark_id": "ifbench", "score": 56.47, "reference_url": "https://www.liquid.ai/blog/lfm2-5-2-6b", "reported_setting": { "mode": "source/model default", "effort": "source does not state", "tools": "none", "sampling": "source does not state trial count", "judge": "benchmark-specified", "harness": "Liquid AI official comparative evaluation", "prompt_style": "source does not state", "temperature": "source does not state evaluation parameters", "context": "source does not state evaluation context" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Displayed exactly as 56.47. Official Liquid AI reported result." }, { "model_id": "qwen3.5-9b", "benchmark_id": "ifstruct_v1", "score": 78.5, "reference_url": "https://www.liquid.ai/blog/lfm2-5-2-6b", "reported_setting": { "mode": "source/model default", "effort": "source does not state", "tools": "none", "sampling": "source does not state trial count", "judge": "binary schema and structural validator", "harness": "IFStruct v1.0 validator without constrained decoding", "prompt_style": "source does not state", "temperature": "source does not state evaluation parameters", "context": "source does not state evaluation context" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Displayed exactly as 78.50. Official Liquid AI reported result." }, { "model_id": "qwen3.5-9b", "benchmark_id": "livecodebench_v6", "score": 69.86, "reference_url": "https://www.liquid.ai/blog/lfm2-5-2-6b", "reported_setting": { "mode": "source/model default", "effort": "source does not state", "tools": "none", "sampling": "source does not state trial count", "judge": "benchmark-specified", "harness": "Liquid AI official comparative evaluation", "prompt_style": "source does not state", "temperature": "source does not state evaluation parameters", "context": "source does not state evaluation context" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Displayed exactly as 69.86. Official Liquid AI reported result." }, { "model_id": "qwen3.5-9b", "benchmark_id": "multi_if", "score": 62.55, "reference_url": "https://www.liquid.ai/blog/lfm2-5-2-6b", "reported_setting": { "mode": "source/model default", "effort": "source does not state", "tools": "none", "sampling": "source does not state trial count", "judge": "benchmark-specified", "harness": "Liquid AI official comparative evaluation", "prompt_style": "source does not state", "temperature": "source does not state evaluation parameters", "context": "source does not state evaluation context" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Displayed exactly as 62.55. Official Liquid AI reported result." }, { "model_id": "qwen3.5-9b", "benchmark_id": "pinchbench", "score": 71.45, "reference_url": "https://www.liquid.ai/blog/lfm2-5-2-6b", "reported_setting": { "mode": "source/model default", "effort": "source does not state", "tools": "OpenClaw coding environment", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "Liquid AI official comparative evaluation", "prompt_style": "source does not state", "temperature": "source does not state evaluation parameters", "context": "source does not state evaluation context" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Displayed exactly as 71.45. Official Liquid AI reported result." }, { "model_id": "qwen3.5-9b", "benchmark_id": "tau3_banking", "score": 5.15, "reference_url": "https://www.liquid.ai/blog/lfm2-5-2-6b", "reported_setting": { "mode": "source/model default", "effort": "source does not state", "tools": "benchmark-provided tools", "sampling": "source does not state trial count", "judge": "benchmark-specified", "harness": "Liquid AI official comparative evaluation", "prompt_style": "source does not state", "temperature": "source does not state evaluation parameters", "context": "source does not state evaluation context" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Displayed exactly as 5.15. Official Liquid AI reported result." }, { "model_id": "qwen3.5-9b", "benchmark_id": "tool_sandbox", "score": 76.44, "reference_url": "https://www.liquid.ai/blog/lfm2-5-2-6b", "reported_setting": { "mode": "source/model default", "effort": "source does not state", "tools": "benchmark-provided tools", "sampling": "source does not state trial count", "judge": "milestone and minefield similarity evaluation", "harness": "Apple ToolSandbox stateful conversational harness", "prompt_style": "source does not state", "temperature": "source does not state evaluation parameters", "context": "source does not state evaluation context" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Displayed exactly as 76.44. Official Liquid AI reported result." }, { "model_id": "claude-haiku-4.5", "benchmark_id": "advancedif_average", "score": 56.9, "reference_url": "https://microsoft.ai/pdf/MAI-Code-1-Flash-Model-Card.PDF", "reported_setting": { "mode": "adaptive reasoning", "effort": "adaptive solution length control", "tools": "none", "sampling": "single/multi-turn average where applicable", "judge": "benchmark-specific", "harness": "Microsoft MAI-Code-1-Flash comparative evaluation", "prompt_style": "source does not state", "temperature": "source does not state", "context": "256K", "notes": "Official instruction-following or agentic-tool-use score; benchmark=advancedif_average." }, "matches_canonical": false, "source_type": "official_model_card", "audit_status": "verified", "notes": "Displayed exactly as 56.9. Official Microsoft-reported result." }, { "model_id": "claude-haiku-4.5", "benchmark_id": "aime_2026", "score": 83.3, "reference_url": "https://microsoft.ai/pdf/MAI-Code-1-Flash-Model-Card.PDF", "reported_setting": { "mode": "adaptive reasoning", "effort": "adaptive solution length control", "tools": "none", "sampling": "source does not state trial count", "judge": "benchmark-specific", "harness": "Microsoft MAI-Code-1-Flash comparative evaluation", "prompt_style": "source does not state", "temperature": "source does not state", "context": "256K model context; per-task usage reported separately", "notes": "Accuracy column from the official model-card/release table; benchmark=aime_2026." }, "matches_canonical": false, "source_type": "official_model_card", "audit_status": "verified", "notes": "Displayed exactly as 83.3. Official Microsoft-reported result." }, { "model_id": "claude-haiku-4.5", "benchmark_id": "amo_bench", "score": 16.0, "reference_url": "https://microsoft.ai/pdf/MAI-Code-1-Flash-Model-Card.PDF", "reported_setting": { "mode": "adaptive reasoning", "effort": "adaptive solution length control", "tools": "none", "sampling": "source does not state trial count", "judge": "benchmark-specific", "harness": "Microsoft MAI-Code-1-Flash comparative evaluation", "prompt_style": "source does not state", "temperature": "source does not state", "context": "256K model context; per-task usage reported separately", "notes": "Accuracy column from the official model-card/release table; benchmark=amo_bench." }, "matches_canonical": false, "source_type": "official_model_card", "audit_status": "verified", "notes": "Displayed exactly as 16.0. Official Microsoft-reported result." }, { "model_id": "claude-haiku-4.5", "benchmark_id": "artifactsbench", "score": 36.6, "reference_url": "https://microsoft.ai/pdf/MAI-Code-1-Flash-Model-Card.PDF", "reported_setting": { "mode": "adaptive reasoning", "effort": "adaptive solution length control", "tools": "none", "sampling": "source does not state trial count", "judge": "benchmark-specific", "harness": "Microsoft MAI-Code-1-Flash comparative evaluation", "prompt_style": "source does not state", "temperature": "source does not state", "context": "256K model context; per-task usage reported separately", "notes": "Accuracy column from the official model-card/release table; benchmark=artifactsbench." }, "matches_canonical": false, "source_type": "official_model_card", "audit_status": "verified", "notes": "Displayed exactly as 36.6. Official Microsoft-reported result." }, { "model_id": "claude-haiku-4.5", "benchmark_id": "frontiermath", "score": 2.8, "reference_url": "https://microsoft.ai/pdf/MAI-Code-1-Flash-Model-Card.PDF", "reported_setting": { "mode": "adaptive reasoning", "effort": "adaptive solution length control", "tools": "none", "sampling": "source does not state trial count", "judge": "benchmark-specific", "harness": "Microsoft MAI-Code-1-Flash comparative evaluation", "prompt_style": "source does not state", "temperature": "source does not state", "context": "256K model context; per-task usage reported separately", "notes": "Accuracy column from the official model-card/release table; benchmark=frontiermath." }, "matches_canonical": false, "source_type": "official_model_card", "audit_status": "verified", "notes": "Displayed exactly as 2.8. Official Microsoft-reported result." }, { "model_id": "claude-haiku-4.5", "benchmark_id": "ifbench_single_multiturn_average", "score": 46.1, "reference_url": "https://microsoft.ai/pdf/MAI-Code-1-Flash-Model-Card.PDF", "reported_setting": { "mode": "adaptive reasoning", "effort": "adaptive solution length control", "tools": "none", "sampling": "single/multi-turn average where applicable", "judge": "benchmark-specific", "harness": "Microsoft MAI-Code-1-Flash comparative evaluation", "prompt_style": "source does not state", "temperature": "source does not state", "context": "256K", "notes": "Official instruction-following or agentic-tool-use score; benchmark=ifbench_single_multiturn_average." }, "matches_canonical": false, "source_type": "official_model_card", "audit_status": "verified", "notes": "Displayed exactly as 46.1. Official Microsoft-reported result." }, { "model_id": "claude-haiku-4.5", "benchmark_id": "swe_bench_multilingual", "score": 62.7, "reference_url": "https://microsoft.ai/pdf/MAI-Code-1-Flash-Model-Card.PDF", "reported_setting": { "mode": "adaptive reasoning", "effort": "adaptive solution length control", "tools": "GitHub Copilot / VS Code production tools", "sampling": "pass@1; benchmark-specific where stated", "judge": "benchmark-specified", "harness": "GitHub Copilot VS Code production harness", "prompt_style": "official Microsoft evaluation prompts", "temperature": "source does not state globally", "context": "256K", "notes": "Coding-focused 137B-total / 5B-active model. Core coding scores use the same production harness for both models." }, "matches_canonical": false, "source_type": "official_model_card", "audit_status": "verified", "notes": "Displayed exactly as 62.7. Official Microsoft-reported result." }, { "model_id": "claude-sonnet-4.6", "benchmark_id": "advancedif_rubric_level", "score": 86.0, "reference_url": "https://microsoft.ai/wp-content/uploads/2026/06/main_20260602_2.pdf", "reported_setting": { "mode": "thinking", "effort": "maximum reasoning effort", "tools": "none; benchmark-provided tools where required", "sampling": "source does not state trial count", "judge": "benchmark-specific Microsoft implementation", "harness": "Microsoft independent evaluation suite", "prompt_style": "benchmark-specific", "temperature": "source does not state", "context": "maximum sequence length", "notes": "Sonnet 4.6 result generated by Microsoft's own evaluation." }, "matches_canonical": false, "source_type": "official_paper", "audit_status": "verified", "notes": "Displayed exactly as 86. Official Microsoft-reported result." }, { "model_id": "claude-sonnet-4.6", "benchmark_id": "air_bench_2024", "score": 88.0, "reference_url": "https://microsoft.ai/wp-content/uploads/2026/06/main_20260602_2.pdf", "reported_setting": { "mode": "thinking", "effort": "maximum reasoning effort", "tools": "none; benchmark-provided tools where required", "sampling": "source does not state trial count", "judge": "benchmark-specific Microsoft implementation", "harness": "Microsoft independent evaluation suite", "prompt_style": "benchmark-specific", "temperature": "source does not state", "context": "maximum sequence length", "notes": "Sonnet 4.6 result generated by Microsoft's own evaluation." }, "matches_canonical": false, "source_type": "official_paper", "audit_status": "verified", "notes": "Displayed exactly as 88. Official Microsoft-reported result." }, { "model_id": "claude-sonnet-4.6", "benchmark_id": "bfcl_v3", "score": 76.0, "reference_url": "https://microsoft.ai/wp-content/uploads/2026/06/main_20260602_2.pdf", "reported_setting": { "mode": "thinking", "effort": "maximum reasoning effort", "tools": "none; benchmark-provided tools where required", "sampling": "source does not state trial count", "judge": "benchmark-specific Microsoft implementation", "harness": "Microsoft independent evaluation suite", "prompt_style": "benchmark-specific", "temperature": "0.001; top_p=0.97", "context": "maximum sequence length", "notes": "Sonnet 4.6 result generated by Microsoft's own evaluation." }, "matches_canonical": false, "source_type": "official_paper", "audit_status": "verified", "notes": "Displayed exactly as 76. Official Microsoft-reported result." }, { "model_id": "claude-sonnet-4.6", "benchmark_id": "corpusqa_gpt54_judge", "score": 79.0, "reference_url": "https://microsoft.ai/wp-content/uploads/2026/06/main_20260602_2.pdf", "reported_setting": { "mode": "thinking", "effort": "maximum reasoning effort", "tools": "none; benchmark-provided tools where required", "sampling": "average of 4 independent evaluations", "judge": "benchmark-specific Microsoft implementation", "harness": "Microsoft independent evaluation suite", "prompt_style": "benchmark-specific", "temperature": "source does not state", "context": "128K maximum output tokens", "notes": "Sonnet 4.6 result generated by Microsoft's own evaluation." }, "matches_canonical": false, "source_type": "official_paper", "audit_status": "verified", "notes": "Displayed exactly as 79. Official Microsoft-reported result." }, { "model_id": "claude-sonnet-4.6", "benchmark_id": "cyberseceval4_autocomplete", "score": 56.0, "reference_url": "https://microsoft.ai/wp-content/uploads/2026/06/main_20260602_2.pdf", "reported_setting": { "mode": "thinking", "effort": "maximum reasoning effort", "tools": "none; benchmark-provided tools where required", "sampling": "source does not state trial count", "judge": "benchmark-specific Microsoft implementation", "harness": "Microsoft independent evaluation suite", "prompt_style": "benchmark-specific", "temperature": "source does not state", "context": "maximum sequence length", "notes": "Sonnet 4.6 result generated by Microsoft's own evaluation." }, "matches_canonical": false, "source_type": "official_paper", "audit_status": "verified", "notes": "Displayed exactly as 56. Official Microsoft-reported result." }, { "model_id": "claude-sonnet-4.6", "benchmark_id": "cyberseceval4_instruct", "score": 62.0, "reference_url": "https://microsoft.ai/wp-content/uploads/2026/06/main_20260602_2.pdf", "reported_setting": { "mode": "thinking", "effort": "maximum reasoning effort", "tools": "none; benchmark-provided tools where required", "sampling": "source does not state trial count", "judge": "benchmark-specific Microsoft implementation", "harness": "Microsoft independent evaluation suite", "prompt_style": "benchmark-specific", "temperature": "source does not state", "context": "maximum sequence length", "notes": "Sonnet 4.6 result generated by Microsoft's own evaluation." }, "matches_canonical": false, "source_type": "official_paper", "audit_status": "verified", "notes": "Displayed exactly as 62. Official Microsoft-reported result." }, { "model_id": "claude-sonnet-4.6", "benchmark_id": "graphwalks_lt128k_combined", "score": 96.0, "reference_url": "https://microsoft.ai/wp-content/uploads/2026/06/main_20260602_2.pdf", "reported_setting": { "mode": "thinking", "effort": "maximum reasoning effort", "tools": "none; benchmark-provided tools where required", "sampling": "source does not state trial count", "judge": "benchmark-specific Microsoft implementation", "harness": "Microsoft independent evaluation suite", "prompt_style": "benchmark-specific", "temperature": "source does not state", "context": "maximum sequence length", "notes": "Sonnet 4.6 result generated by Microsoft's own evaluation." }, "matches_canonical": false, "source_type": "official_paper", "audit_status": "verified", "notes": "Displayed exactly as 96. Official Microsoft-reported result." }, { "model_id": "claude-sonnet-4.6", "benchmark_id": "longbench_v2_256k", "score": 66.0, "reference_url": "https://microsoft.ai/wp-content/uploads/2026/06/main_20260602_2.pdf", "reported_setting": { "mode": "thinking", "effort": "maximum reasoning effort", "tools": "none; benchmark-provided tools where required", "sampling": "average of 4 independent evaluations", "judge": "benchmark-specific Microsoft implementation", "harness": "Microsoft independent evaluation suite", "prompt_style": "benchmark-specific", "temperature": "source does not state", "context": "128K maximum output tokens", "notes": "Sonnet 4.6 result generated by Microsoft's own evaluation." }, "matches_canonical": false, "source_type": "official_paper", "audit_status": "verified", "notes": "Displayed exactly as 66. Official Microsoft-reported result." }, { "model_id": "claude-sonnet-4.6", "benchmark_id": "longfact_claim_precision", "score": 98.0, "reference_url": "https://microsoft.ai/wp-content/uploads/2026/06/main_20260602_2.pdf", "reported_setting": { "mode": "thinking", "effort": "maximum reasoning effort", "tools": "none; benchmark-provided tools where required", "sampling": "source does not state trial count", "judge": "benchmark-specific Microsoft implementation", "harness": "Microsoft independent evaluation suite", "prompt_style": "benchmark-specific", "temperature": "source does not state", "context": "maximum sequence length", "notes": "Sonnet 4.6 result generated by Microsoft's own evaluation." }, "matches_canonical": false, "source_type": "official_paper", "audit_status": "verified", "notes": "Displayed exactly as 98. Official Microsoft-reported result." }, { "model_id": "claude-sonnet-4.6", "benchmark_id": "medxpertqa_text", "score": 49.0, "reference_url": "https://microsoft.ai/wp-content/uploads/2026/06/main_20260602_2.pdf", "reported_setting": { "mode": "thinking", "effort": "maximum reasoning effort", "tools": "none; benchmark-provided tools where required", "sampling": "source does not state trial count", "judge": "benchmark-specific Microsoft implementation", "harness": "Microsoft independent evaluation suite", "prompt_style": "benchmark-specific", "temperature": "source does not state", "context": "maximum sequence length", "notes": "Sonnet 4.6 result generated by Microsoft's own evaluation." }, "matches_canonical": false, "source_type": "official_paper", "audit_status": "verified", "notes": "Displayed exactly as 49. Official Microsoft-reported result." }, { "model_id": "claude-sonnet-4.6", "benchmark_id": "multichallenge", "score": 57.0, "reference_url": "https://microsoft.ai/wp-content/uploads/2026/06/main_20260602_2.pdf", "reported_setting": { "mode": "thinking", "effort": "maximum reasoning effort", "tools": "none; benchmark-provided tools where required", "sampling": "source does not state trial count", "judge": "benchmark-specific Microsoft implementation", "harness": "Microsoft independent evaluation suite", "prompt_style": "benchmark-specific", "temperature": "source does not state", "context": "maximum sequence length", "notes": "Sonnet 4.6 result generated by Microsoft's own evaluation." }, "matches_canonical": false, "source_type": "official_paper", "audit_status": "verified", "notes": "Displayed exactly as 57. Official Microsoft-reported result." }, { "model_id": "claude-sonnet-4.6", "benchmark_id": "simpleqa_verified", "score": 29.0, "reference_url": "https://microsoft.ai/wp-content/uploads/2026/06/main_20260602_2.pdf", "reported_setting": { "mode": "thinking", "effort": "maximum reasoning effort", "tools": "none; benchmark-provided tools where required", "sampling": "source does not state trial count", "judge": "benchmark-specific Microsoft implementation", "harness": "Microsoft independent evaluation suite", "prompt_style": "benchmark-specific", "temperature": "source does not state", "context": "maximum sequence length", "notes": "Sonnet 4.6 result generated by Microsoft's own evaluation." }, "matches_canonical": false, "source_type": "official_paper", "audit_status": "verified", "notes": "Displayed exactly as 29. Official Microsoft-reported result." }, { "model_id": "claude-sonnet-4.6", "benchmark_id": "truthfulqa_mc", "score": 88.0, "reference_url": "https://microsoft.ai/wp-content/uploads/2026/06/main_20260602_2.pdf", "reported_setting": { "mode": "thinking", "effort": "maximum reasoning effort", "tools": "none; benchmark-provided tools where required", "sampling": "source does not state trial count", "judge": "benchmark-specific Microsoft implementation", "harness": "Microsoft independent evaluation suite", "prompt_style": "benchmark-specific", "temperature": "source does not state", "context": "maximum sequence length", "notes": "Sonnet 4.6 result generated by Microsoft's own evaluation." }, "matches_canonical": false, "source_type": "official_paper", "audit_status": "verified", "notes": "Displayed exactly as 88. Official Microsoft-reported result." }, { "model_id": "mai-code-1-flash", "benchmark_id": "advancedif_average", "score": 71.4, "reference_url": "https://microsoft.ai/pdf/MAI-Code-1-Flash-Model-Card.PDF", "reported_setting": { "mode": "adaptive reasoning", "effort": "adaptive solution length control", "tools": "none", "sampling": "single/multi-turn average where applicable", "judge": "benchmark-specific", "harness": "Microsoft MAI-Code-1-Flash comparative evaluation", "prompt_style": "source does not state", "temperature": "source does not state", "context": "256K", "notes": "Official instruction-following or agentic-tool-use score; benchmark=advancedif_average." }, "matches_canonical": false, "source_type": "official_model_card", "audit_status": "verified", "notes": "Displayed exactly as 71.4. Official Microsoft-reported result." }, { "model_id": "mai-code-1-flash", "benchmark_id": "aime_2026", "score": 92.5, "reference_url": "https://microsoft.ai/pdf/MAI-Code-1-Flash-Model-Card.PDF", "reported_setting": { "mode": "adaptive reasoning", "effort": "adaptive solution length control", "tools": "none", "sampling": "source does not state trial count", "judge": "benchmark-specific", "harness": "Microsoft MAI-Code-1-Flash comparative evaluation", "prompt_style": "source does not state", "temperature": "source does not state", "context": "256K model context; per-task usage reported separately", "notes": "Accuracy column from the official model-card/release table; benchmark=aime_2026." }, "matches_canonical": false, "source_type": "official_model_card", "audit_status": "verified", "notes": "Displayed exactly as 92.5. Official Microsoft-reported result." }, { "model_id": "mai-code-1-flash", "benchmark_id": "amo_bench", "score": 40.0, "reference_url": "https://microsoft.ai/pdf/MAI-Code-1-Flash-Model-Card.PDF", "reported_setting": { "mode": "adaptive reasoning", "effort": "adaptive solution length control", "tools": "none", "sampling": "source does not state trial count", "judge": "benchmark-specific", "harness": "Microsoft MAI-Code-1-Flash comparative evaluation", "prompt_style": "source does not state", "temperature": "source does not state", "context": "256K model context; per-task usage reported separately", "notes": "Accuracy column from the official model-card/release table; benchmark=amo_bench." }, "matches_canonical": false, "source_type": "official_model_card", "audit_status": "verified", "notes": "Displayed exactly as 40.0. Official Microsoft-reported result." }, { "model_id": "mai-code-1-flash", "benchmark_id": "artifactsbench", "score": 36.4, "reference_url": "https://microsoft.ai/pdf/MAI-Code-1-Flash-Model-Card.PDF", "reported_setting": { "mode": "adaptive reasoning", "effort": "adaptive solution length control", "tools": "none", "sampling": "source does not state trial count", "judge": "benchmark-specific", "harness": "Microsoft MAI-Code-1-Flash comparative evaluation", "prompt_style": "source does not state", "temperature": "source does not state", "context": "256K model context; per-task usage reported separately", "notes": "Accuracy column from the official model-card/release table; benchmark=artifactsbench." }, "matches_canonical": false, "source_type": "official_model_card", "audit_status": "verified", "notes": "Displayed exactly as 36.4. Official Microsoft-reported result." }, { "model_id": "mai-code-1-flash", "benchmark_id": "frontiermath", "score": 6.3, "reference_url": "https://microsoft.ai/pdf/MAI-Code-1-Flash-Model-Card.PDF", "reported_setting": { "mode": "adaptive reasoning", "effort": "adaptive solution length control", "tools": "none", "sampling": "source does not state trial count", "judge": "benchmark-specific", "harness": "Microsoft MAI-Code-1-Flash comparative evaluation", "prompt_style": "source does not state", "temperature": "source does not state", "context": "256K model context; per-task usage reported separately", "notes": "Accuracy column from the official model-card/release table; benchmark=frontiermath." }, "matches_canonical": false, "source_type": "official_model_card", "audit_status": "verified", "notes": "Displayed exactly as 6.3. Official Microsoft-reported result." }, { "model_id": "mai-code-1-flash", "benchmark_id": "gpqa_diamond", "score": 84.6, "reference_url": "https://microsoft.ai/pdf/MAI-Code-1-Flash-Model-Card.PDF", "reported_setting": { "mode": "adaptive reasoning", "effort": "adaptive solution length control", "tools": "none", "sampling": "source does not state trial count", "judge": "benchmark-specific", "harness": "Microsoft MAI-Code-1-Flash comparative evaluation", "prompt_style": "source does not state", "temperature": "source does not state", "context": "256K model context; per-task usage reported separately", "notes": "Accuracy column from the official model-card/release table; benchmark=gpqa_diamond." }, "matches_canonical": false, "source_type": "official_model_card", "audit_status": "verified", "notes": "Displayed exactly as 84.6. Official Microsoft-reported result." }, { "model_id": "mai-code-1-flash", "benchmark_id": "hle_text", "score": 18.0, "reference_url": "https://microsoft.ai/pdf/MAI-Code-1-Flash-Model-Card.PDF", "reported_setting": { "mode": "adaptive reasoning", "effort": "adaptive solution length control", "tools": "none", "sampling": "source does not state trial count", "judge": "benchmark-specific", "harness": "Microsoft MAI-Code-1-Flash comparative evaluation", "prompt_style": "source does not state", "temperature": "source does not state", "context": "256K model context; per-task usage reported separately", "notes": "Accuracy column from the official model-card/release table; benchmark=hle_text." }, "matches_canonical": false, "source_type": "official_model_card", "audit_status": "verified", "notes": "Displayed exactly as 18.0. Official Microsoft-reported result." }, { "model_id": "mai-code-1-flash", "benchmark_id": "ifbench_single_multiturn_average", "score": 75.0, "reference_url": "https://microsoft.ai/pdf/MAI-Code-1-Flash-Model-Card.PDF", "reported_setting": { "mode": "adaptive reasoning", "effort": "adaptive solution length control", "tools": "none", "sampling": "single/multi-turn average where applicable", "judge": "benchmark-specific", "harness": "Microsoft MAI-Code-1-Flash comparative evaluation", "prompt_style": "source does not state", "temperature": "source does not state", "context": "256K", "notes": "Official instruction-following or agentic-tool-use score; benchmark=ifbench_single_multiturn_average." }, "matches_canonical": false, "source_type": "official_model_card", "audit_status": "verified", "notes": "Displayed exactly as 75.0. Official Microsoft-reported result." }, { "model_id": "mai-code-1-flash", "benchmark_id": "swe_bench_multilingual", "score": 65.5, "reference_url": "https://microsoft.ai/pdf/MAI-Code-1-Flash-Model-Card.PDF", "reported_setting": { "mode": "adaptive reasoning", "effort": "adaptive solution length control", "tools": "GitHub Copilot / VS Code production tools", "sampling": "pass@1; benchmark-specific where stated", "judge": "benchmark-specified", "harness": "GitHub Copilot VS Code production harness", "prompt_style": "official Microsoft evaluation prompts", "temperature": "source does not state globally", "context": "256K", "notes": "Coding-focused 137B-total / 5B-active model. Core coding scores use the same production harness for both models." }, "matches_canonical": true, "source_type": "official_model_card", "audit_status": "verified", "notes": "Displayed exactly as 65.5. Official Microsoft-reported result." }, { "model_id": "mai-code-1-flash", "benchmark_id": "swe_bench_pro", "score": 51.2, "reference_url": "https://microsoft.ai/pdf/MAI-Code-1-Flash-Model-Card.PDF", "reported_setting": { "mode": "adaptive reasoning", "effort": "adaptive solution length control", "tools": "GitHub Copilot / VS Code production tools", "sampling": "pass@1; benchmark-specific where stated", "judge": "benchmark-specified", "harness": "GitHub Copilot VS Code production harness", "prompt_style": "official Microsoft evaluation prompts", "temperature": "source does not state globally", "context": "256K", "notes": "Coding-focused 137B-total / 5B-active model. Core coding scores use the same production harness for both models." }, "matches_canonical": true, "source_type": "official_model_card", "audit_status": "verified", "notes": "Displayed exactly as 51.2. Official Microsoft-reported result." }, { "model_id": "mai-code-1-flash", "benchmark_id": "swe_bench_verified", "score": 71.6, "reference_url": "https://microsoft.ai/pdf/MAI-Code-1-Flash-Model-Card.PDF", "reported_setting": { "mode": "adaptive reasoning", "effort": "adaptive solution length control", "tools": "GitHub Copilot / VS Code production tools", "sampling": "pass@1; benchmark-specific where stated", "judge": "benchmark-specified", "harness": "GitHub Copilot VS Code production harness", "prompt_style": "official Microsoft evaluation prompts", "temperature": "source does not state globally", "context": "256K", "notes": "Coding-focused 137B-total / 5B-active model. Core coding scores use the same production harness for both models." }, "matches_canonical": true, "source_type": "official_model_card", "audit_status": "verified", "notes": "Displayed exactly as 71.6. Official Microsoft-reported result." }, { "model_id": "mai-code-1-flash", "benchmark_id": "tau2_bench_telecom", "score": 71.7, "reference_url": "https://microsoft.ai/pdf/MAI-Code-1-Flash-Model-Card.PDF", "reported_setting": { "mode": "adaptive reasoning", "effort": "adaptive solution length control", "tools": "Tau2 telecom tools", "sampling": "single/multi-turn average where applicable", "judge": "benchmark-specific", "harness": "Microsoft MAI-Code-1-Flash comparative evaluation", "prompt_style": "source does not state", "temperature": "source does not state", "context": "256K", "notes": "Official instruction-following or agentic-tool-use score; benchmark=tau2_bench_telecom." }, "matches_canonical": false, "source_type": "official_model_card", "audit_status": "verified", "notes": "Displayed exactly as 71.7. Official Microsoft-reported result." }, { "model_id": "mai-code-1-flash", "benchmark_id": "terminal_bench", "score": 54.8, "reference_url": "https://microsoft.ai/pdf/MAI-Code-1-Flash-Model-Card.PDF", "reported_setting": { "mode": "adaptive reasoning", "effort": "adaptive solution length control", "tools": "GitHub Copilot / VS Code production tools", "sampling": "pass@1; benchmark-specific where stated", "judge": "benchmark-specified", "harness": "GitHub Copilot VS Code production harness", "prompt_style": "official Microsoft evaluation prompts", "temperature": "source does not state globally", "context": "256K", "notes": "Coding-focused 137B-total / 5B-active model. Core coding scores use the same production harness for both models." }, "matches_canonical": true, "source_type": "official_model_card", "audit_status": "verified", "notes": "Displayed exactly as 54.8. Official Microsoft-reported result." }, { "model_id": "mai-thinking-1", "benchmark_id": "advancedif_rubric_level", "score": 85.0, "reference_url": "https://microsoft.ai/wp-content/uploads/2026/06/main_20260602_2.pdf", "reported_setting": { "mode": "thinking", "effort": "maximum reasoning effort", "tools": "none; benchmark-provided tools for BFCL v3", "sampling": "average of 4 runs", "judge": "calibrated LLM judge", "harness": "Microsoft official evaluation suite", "prompt_style": "benchmark-specific Microsoft implementation", "temperature": "1.0; top_p=0.97", "context": "maximum sequence length", "notes": "AdvancedIF rubric-level score, not all-rubrics pass rate." }, "matches_canonical": false, "source_type": "official_paper", "audit_status": "verified", "notes": "Displayed exactly as 85. Official Microsoft-reported result." }, { "model_id": "mai-thinking-1", "benchmark_id": "aime_2025", "score": 97.0, "reference_url": "https://microsoft.ai/wp-content/uploads/2026/06/main_20260602_2.pdf", "reported_setting": { "mode": "thinking", "effort": "maximum available reasoning", "tools": "none", "sampling": "average of 4 runs; temperature=1; top_p=0.97", "judge": "regex/SymPy with GPT-5-mini fallback", "harness": "Microsoft STEM evaluation", "prompt_style": "boxed final answer for STEM; ReAct for agentic coding", "temperature": "1.0; top_p=0.97", "context": "256K maximum output tokens", "notes": "Technical report headline evaluation configuration." }, "matches_canonical": false, "source_type": "official_paper", "audit_status": "verified", "notes": "Displayed exactly as 97.0. Official Microsoft-reported result.", "candidates": [ { "score": 95.0, "reference_url": "https://microsoft.ai/wp-content/uploads/2026/06/main_20260602_2.pdf", "source_type": "official_paper", "reported_setting": { "mode": "thinking", "effort": "maximum available reasoning", "tools": "none", "sampling": "average of 4 runs; temperature=1; top_p=0.97", "judge": "regex/SymPy with GPT-5-mini fallback", "harness": "Microsoft STEM evaluation", "prompt_style": "boxed final answer for STEM; ReAct for agentic coding", "temperature": "1.0; top_p=0.97", "context": "128K maximum output tokens", "notes": "Technical report headline evaluation configuration." }, "notes": "Displayed exactly as 95.0. Official Microsoft-reported result." } ] }, { "model_id": "mai-thinking-1", "benchmark_id": "aime_2026", "score": 94.5, "reference_url": "https://microsoft.ai/wp-content/uploads/2026/06/main_20260602_2.pdf", "reported_setting": { "mode": "thinking", "effort": "maximum available reasoning", "tools": "none", "sampling": "average of 4 runs; temperature=1; top_p=0.97", "judge": "regex/SymPy with GPT-5-mini fallback", "harness": "Microsoft STEM evaluation", "prompt_style": "boxed final answer for STEM; ReAct for agentic coding", "temperature": "1.0; top_p=0.97", "context": "256K maximum output tokens", "notes": "Technical report headline evaluation configuration." }, "matches_canonical": false, "source_type": "official_paper", "audit_status": "verified", "notes": "Displayed exactly as 94.5. Official Microsoft-reported result.", "candidates": [ { "score": 93.6, "reference_url": "https://microsoft.ai/wp-content/uploads/2026/06/main_20260602_2.pdf", "source_type": "official_paper", "reported_setting": { "mode": "thinking", "effort": "maximum available reasoning", "tools": "none", "sampling": "average of 4 runs; temperature=1; top_p=0.97", "judge": "regex/SymPy with GPT-5-mini fallback", "harness": "Microsoft STEM evaluation", "prompt_style": "boxed final answer for STEM; ReAct for agentic coding", "temperature": "1.0; top_p=0.97", "context": "128K maximum output tokens", "notes": "Technical report headline evaluation configuration." }, "notes": "Displayed exactly as 93.6. Official Microsoft-reported result." } ] }, { "model_id": "mai-thinking-1", "benchmark_id": "air_bench_2024", "score": 88.0, "reference_url": "https://microsoft.ai/wp-content/uploads/2026/06/main_20260602_2.pdf", "reported_setting": { "mode": "thinking", "effort": "maximum reasoning effort", "tools": "none; benchmark-provided tools for BFCL v3", "sampling": "average of 4 runs", "judge": "benchmark-specific", "harness": "Microsoft official evaluation suite", "prompt_style": "benchmark-specific Microsoft implementation", "temperature": "1.0; top_p=0.97", "context": "maximum sequence length", "notes": "Table 12 configuration from the technical report." }, "matches_canonical": false, "source_type": "official_paper", "audit_status": "verified", "notes": "Displayed exactly as 88. Official Microsoft-reported result." }, { "model_id": "mai-thinking-1", "benchmark_id": "bfcl_v3", "score": 72.0, "reference_url": "https://microsoft.ai/wp-content/uploads/2026/06/main_20260602_2.pdf", "reported_setting": { "mode": "thinking", "effort": "maximum reasoning effort", "tools": "BFCL-provided functions", "sampling": "average of 4 runs", "judge": "benchmark-specific", "harness": "Microsoft official evaluation suite", "prompt_style": "benchmark-specific Microsoft implementation", "temperature": "0.001; top_p=0.97", "context": "maximum sequence length", "notes": "Table 12 configuration from the technical report." }, "matches_canonical": false, "source_type": "official_paper", "audit_status": "verified", "notes": "Displayed exactly as 72. Official Microsoft-reported result." }, { "model_id": "mai-thinking-1", "benchmark_id": "corpusqa_gpt54_judge", "score": 82.0, "reference_url": "https://microsoft.ai/wp-content/uploads/2026/06/main_20260602_2.pdf", "reported_setting": { "mode": "thinking", "effort": "maximum reasoning effort", "tools": "none", "sampling": "average of 4 independent evaluations", "judge": "GPT-5.4 high", "harness": "Microsoft long-context evaluation", "prompt_style": "source does not state", "context": "128K maximum output; input limit not stated", "temperature": "1.0; top_p=0.97", "notes": "Technical report Table 19 long-context evaluation." }, "matches_canonical": false, "source_type": "official_paper", "audit_status": "verified", "notes": "Displayed exactly as 82. Official Microsoft-reported result." }, { "model_id": "mai-thinking-1", "benchmark_id": "cyberseceval4_autocomplete", "score": 63.0, "reference_url": "https://microsoft.ai/wp-content/uploads/2026/06/main_20260602_2.pdf", "reported_setting": { "mode": "thinking", "effort": "maximum reasoning effort", "tools": "none; benchmark-provided tools for BFCL v3", "sampling": "average of 4 runs", "judge": "benchmark-specific", "harness": "Microsoft official evaluation suite", "prompt_style": "benchmark-specific Microsoft implementation", "temperature": "1.0; top_p=0.97", "context": "maximum sequence length", "notes": "Table 12 configuration from the technical report." }, "matches_canonical": false, "source_type": "official_paper", "audit_status": "verified", "notes": "Displayed exactly as 63. Official Microsoft-reported result." }, { "model_id": "mai-thinking-1", "benchmark_id": "cyberseceval4_instruct", "score": 63.0, "reference_url": "https://microsoft.ai/wp-content/uploads/2026/06/main_20260602_2.pdf", "reported_setting": { "mode": "thinking", "effort": "maximum reasoning effort", "tools": "none; benchmark-provided tools for BFCL v3", "sampling": "average of 4 runs", "judge": "benchmark-specific", "harness": "Microsoft official evaluation suite", "prompt_style": "benchmark-specific Microsoft implementation", "temperature": "1.0; top_p=0.97", "context": "maximum sequence length", "notes": "Table 12 configuration from the technical report." }, "matches_canonical": false, "source_type": "official_paper", "audit_status": "verified", "notes": "Displayed exactly as 63. Official Microsoft-reported result." }, { "model_id": "mai-thinking-1", "benchmark_id": "gpqa_diamond", "score": 84.2, "reference_url": "https://microsoft.ai/wp-content/uploads/2026/06/main_20260602_2.pdf", "reported_setting": { "mode": "thinking", "effort": "maximum available reasoning", "tools": "none", "sampling": "pass@1 averaged over 16 rollouts; temperature=1; top_p=0.97", "judge": "regex/SymPy with GPT-5-mini fallback", "harness": "Microsoft STEM evaluation", "prompt_style": "boxed final answer for STEM; ReAct for agentic coding", "temperature": "1.0; top_p=0.97", "context": "256K maximum output tokens", "notes": "Technical report headline evaluation configuration." }, "matches_canonical": false, "source_type": "official_paper", "audit_status": "verified", "notes": "Displayed exactly as 84.2. Official Microsoft-reported result.", "candidates": [ { "score": 84.2, "reference_url": "https://microsoft.ai/wp-content/uploads/2026/06/main_20260602_2.pdf", "source_type": "official_paper", "reported_setting": { "mode": "thinking", "effort": "maximum available reasoning", "tools": "none", "sampling": "pass@1 averaged over 16 rollouts; temperature=1; top_p=0.97", "judge": "regex/SymPy with GPT-5-mini fallback", "harness": "Microsoft STEM evaluation", "prompt_style": "boxed final answer for STEM; ReAct for agentic coding", "temperature": "1.0; top_p=0.97", "context": "128K maximum output tokens", "notes": "Technical report headline evaluation configuration." }, "notes": "Displayed exactly as 84.2. Official Microsoft-reported result." } ] }, { "model_id": "mai-thinking-1", "benchmark_id": "graphwalks_lt128k_combined", "score": 90.0, "reference_url": "https://microsoft.ai/wp-content/uploads/2026/06/main_20260602_2.pdf", "reported_setting": { "mode": "thinking", "effort": "maximum reasoning effort", "tools": "none; benchmark-provided tools for BFCL v3", "sampling": "average of 4 runs", "judge": "answer-set F1", "harness": "Microsoft official evaluation suite", "prompt_style": "benchmark-specific Microsoft implementation", "temperature": "1.0; top_p=0.97", "context": "<=128K measured with o200k_base tokenizer", "notes": "Table 12 configuration from the technical report." }, "matches_canonical": false, "source_type": "official_paper", "audit_status": "verified", "notes": "Displayed exactly as 90. Official Microsoft-reported result." }, { "model_id": "mai-thinking-1", "benchmark_id": "healthbench_professional_length_adjusted", "score": 35.0, "reference_url": "https://microsoft.ai/wp-content/uploads/2026/06/main_20260602_2.pdf", "reported_setting": { "mode": "thinking", "effort": "maximum reasoning effort", "tools": "none; benchmark-provided tools for BFCL v3", "sampling": "average of 4 runs", "judge": "GPT-5.4 grader with official rubrics", "harness": "Microsoft official evaluation suite", "prompt_style": "benchmark-specific Microsoft implementation", "temperature": "1.0; top_p=0.97", "context": "maximum sequence length", "notes": "Length-penalized HealthBench Professional primary metric." }, "matches_canonical": false, "source_type": "official_paper", "audit_status": "verified", "notes": "Displayed exactly as 35. Official Microsoft-reported result." }, { "model_id": "mai-thinking-1", "benchmark_id": "hmmt_feb_2026", "score": 84.9, "reference_url": "https://microsoft.ai/wp-content/uploads/2026/06/main_20260602_2.pdf", "reported_setting": { "mode": "thinking", "effort": "maximum available reasoning", "tools": "none", "sampling": "average of 4 runs; temperature=1; top_p=0.97", "judge": "regex/SymPy with GPT-5-mini fallback", "harness": "Microsoft STEM evaluation", "prompt_style": "boxed final answer for STEM; ReAct for agentic coding", "temperature": "1.0; top_p=0.97", "context": "256K maximum output tokens", "notes": "Technical report headline evaluation configuration." }, "matches_canonical": false, "source_type": "official_paper", "audit_status": "verified", "notes": "Displayed exactly as 84.9. Official Microsoft-reported result.", "candidates": [ { "score": 84.3, "reference_url": "https://microsoft.ai/wp-content/uploads/2026/06/main_20260602_2.pdf", "source_type": "official_paper", "reported_setting": { "mode": "thinking", "effort": "maximum available reasoning", "tools": "none", "sampling": "average of 4 runs; temperature=1; top_p=0.97", "judge": "regex/SymPy with GPT-5-mini fallback", "harness": "Microsoft STEM evaluation", "prompt_style": "boxed final answer for STEM; ReAct for agentic coding", "temperature": "1.0; top_p=0.97", "context": "128K maximum output tokens", "notes": "Technical report headline evaluation configuration." }, "notes": "Displayed exactly as 84.3. Official Microsoft-reported result." } ] }, { "model_id": "mai-thinking-1", "benchmark_id": "ifbench", "score": 69.0, "reference_url": "https://microsoft.ai/wp-content/uploads/2026/06/main_20260602_2.pdf", "reported_setting": { "mode": "thinking", "effort": "maximum reasoning effort", "tools": "none; benchmark-provided tools for BFCL v3", "sampling": "average of 4 runs", "judge": "benchmark-specific", "harness": "Microsoft official evaluation suite", "prompt_style": "benchmark-specific Microsoft implementation", "temperature": "1.0; top_p=0.97", "context": "maximum sequence length", "notes": "Table 12 configuration from the technical report." }, "matches_canonical": false, "source_type": "official_paper", "audit_status": "verified", "notes": "Displayed exactly as 69. Official Microsoft-reported result." }, { "model_id": "mai-thinking-1", "benchmark_id": "livecodebench_v6", "score": 87.7, "reference_url": "https://microsoft.ai/wp-content/uploads/2026/06/main_20260602_2.pdf", "reported_setting": { "mode": "thinking", "effort": "maximum available reasoning", "tools": "none", "sampling": "average of 4 runs; temperature=1; top_p=0.97", "judge": "regex/SymPy with GPT-5-mini fallback", "harness": "Microsoft STEM evaluation", "prompt_style": "boxed final answer for STEM; ReAct for agentic coding", "temperature": "1.0; top_p=0.97", "context": "256K maximum output tokens", "notes": "Technical report headline evaluation configuration." }, "matches_canonical": false, "source_type": "official_paper", "audit_status": "verified", "notes": "Displayed exactly as 87.7. Official Microsoft-reported result.", "candidates": [ { "score": 87.3, "reference_url": "https://microsoft.ai/wp-content/uploads/2026/06/main_20260602_2.pdf", "source_type": "official_paper", "reported_setting": { "mode": "thinking", "effort": "maximum available reasoning", "tools": "none", "sampling": "average of 4 runs; temperature=1; top_p=0.97", "judge": "regex/SymPy with GPT-5-mini fallback", "harness": "Microsoft STEM evaluation", "prompt_style": "boxed final answer for STEM; ReAct for agentic coding", "temperature": "1.0; top_p=0.97", "context": "128K maximum output tokens", "notes": "Technical report headline evaluation configuration." }, "notes": "Displayed exactly as 87.3. Official Microsoft-reported result." } ] }, { "model_id": "mai-thinking-1", "benchmark_id": "longbench_v2_256k", "score": 61.0, "reference_url": "https://microsoft.ai/wp-content/uploads/2026/06/main_20260602_2.pdf", "reported_setting": { "mode": "thinking", "effort": "maximum reasoning effort", "tools": "none", "sampling": "average of 4 independent evaluations", "judge": "exact multiple-choice answer matching", "harness": "Microsoft long-context evaluation", "prompt_style": "source does not state", "context": "256K maximum input; 128K maximum output", "temperature": "1.0; top_p=0.97", "notes": "Technical report Table 19 long-context evaluation." }, "matches_canonical": false, "source_type": "official_paper", "audit_status": "verified", "notes": "Displayed exactly as 61. Official Microsoft-reported result." }, { "model_id": "mai-thinking-1", "benchmark_id": "longfact_claim_precision", "score": 98.0, "reference_url": "https://microsoft.ai/wp-content/uploads/2026/06/main_20260602_2.pdf", "reported_setting": { "mode": "thinking", "effort": "maximum reasoning effort", "tools": "none; benchmark-provided tools for BFCL v3", "sampling": "average of 4 runs", "judge": "simplified claim extraction plus LLM judge", "harness": "Microsoft official evaluation suite", "prompt_style": "benchmark-specific Microsoft implementation", "temperature": "1.0; top_p=0.97", "context": "maximum sequence length", "notes": "Table 12 configuration from the technical report." }, "matches_canonical": false, "source_type": "official_paper", "audit_status": "verified", "notes": "Displayed exactly as 98. Official Microsoft-reported result." }, { "model_id": "mai-thinking-1", "benchmark_id": "medxpertqa_text", "score": 43.0, "reference_url": "https://microsoft.ai/wp-content/uploads/2026/06/main_20260602_2.pdf", "reported_setting": { "mode": "thinking", "effort": "maximum reasoning effort", "tools": "none; benchmark-provided tools for BFCL v3", "sampling": "average of 4 runs", "judge": "GPT-5.4 parses the answer letter", "harness": "Microsoft official evaluation suite", "prompt_style": "benchmark-specific Microsoft implementation", "temperature": "1.0; top_p=0.97", "context": "maximum sequence length", "notes": "2,450 expert-curated 10-choice text questions." }, "matches_canonical": false, "source_type": "official_paper", "audit_status": "verified", "notes": "Displayed exactly as 43. Official Microsoft-reported result." }, { "model_id": "mai-thinking-1", "benchmark_id": "mmlu_pro", "score": 85.0, "reference_url": "https://microsoft.ai/wp-content/uploads/2026/06/main_20260602_2.pdf", "reported_setting": { "mode": "thinking", "effort": "maximum reasoning effort", "tools": "none; benchmark-provided tools for BFCL v3", "sampling": "average of 4 runs", "judge": "benchmark-specific", "harness": "Microsoft official evaluation suite", "prompt_style": "benchmark-specific Microsoft implementation", "temperature": "1.0; top_p=0.97", "context": "maximum sequence length", "notes": "Table 12 configuration from the technical report." }, "matches_canonical": false, "source_type": "official_paper", "audit_status": "verified", "notes": "Displayed exactly as 85. Official Microsoft-reported result." }, { "model_id": "mai-thinking-1", "benchmark_id": "multichallenge", "score": 53.0, "reference_url": "https://microsoft.ai/wp-content/uploads/2026/06/main_20260602_2.pdf", "reported_setting": { "mode": "thinking", "effort": "maximum reasoning effort", "tools": "none; benchmark-provided tools for BFCL v3", "sampling": "average of 4 runs", "judge": "benchmark-specific", "harness": "Microsoft official evaluation suite", "prompt_style": "benchmark-specific Microsoft implementation", "temperature": "1.0; top_p=0.97", "context": "maximum sequence length", "notes": "Table 12 configuration from the technical report." }, "matches_canonical": false, "source_type": "official_paper", "audit_status": "verified", "notes": "Displayed exactly as 53. Official Microsoft-reported result." }, { "model_id": "mai-thinking-1", "benchmark_id": "simpleqa_verified", "score": 31.0, "reference_url": "https://microsoft.ai/wp-content/uploads/2026/06/main_20260602_2.pdf", "reported_setting": { "mode": "thinking", "effort": "maximum reasoning effort", "tools": "none; benchmark-provided tools for BFCL v3", "sampling": "average of 4 runs", "judge": "benchmark-specific", "harness": "Microsoft official evaluation suite", "prompt_style": "benchmark-specific Microsoft implementation", "temperature": "1.0; top_p=0.97", "context": "maximum sequence length", "notes": "Table 12 configuration from the technical report." }, "matches_canonical": false, "source_type": "official_paper", "audit_status": "verified", "notes": "Displayed exactly as 31. Official Microsoft-reported result." }, { "model_id": "mai-thinking-1", "benchmark_id": "swe_bench_pro", "score": 52.8, "reference_url": "https://microsoft.ai/wp-content/uploads/2026/06/main_20260602_2.pdf", "reported_setting": { "mode": "thinking", "effort": "maximum available reasoning", "tools": "ReAct loop with bash and string-replacement tools", "sampling": "average of 4 runs; temperature=1; top_p=0.97", "judge": "benchmark verifier", "harness": "Microsoft ReAct agent evaluation", "prompt_style": "boxed final answer for STEM; ReAct for agentic coding", "temperature": "1.0; top_p=0.97", "context": "256K total context; 8K max output; 1,000 max steps", "notes": "Technical report headline evaluation configuration." }, "matches_canonical": false, "source_type": "official_paper", "audit_status": "verified", "notes": "Displayed exactly as 52.8. Official Microsoft-reported result." }, { "model_id": "mai-thinking-1", "benchmark_id": "swe_bench_verified", "score": 73.5, "reference_url": "https://microsoft.ai/wp-content/uploads/2026/06/main_20260602_2.pdf", "reported_setting": { "mode": "thinking", "effort": "maximum available reasoning", "tools": "ReAct loop with bash and string-replacement tools", "sampling": "average of 4 runs; temperature=1; top_p=0.97", "judge": "benchmark verifier", "harness": "Microsoft ReAct agent evaluation", "prompt_style": "boxed final answer for STEM; ReAct for agentic coding", "temperature": "1.0; top_p=0.97", "context": "256K total context; 8K max output; 1,000 max steps", "notes": "Technical report headline evaluation configuration." }, "matches_canonical": false, "source_type": "official_paper", "audit_status": "verified", "notes": "Displayed exactly as 73.5. Official Microsoft-reported result." }, { "model_id": "mai-thinking-1", "benchmark_id": "terminal_bench", "score": 46.0, "reference_url": "https://microsoft.ai/wp-content/uploads/2026/06/main_20260602_2.pdf", "reported_setting": { "mode": "thinking", "effort": "maximum available reasoning", "tools": "ReAct loop with bash only", "sampling": "average of 4 runs; temperature=1; top_p=0.97", "judge": "benchmark verifier", "harness": "Microsoft ReAct agent evaluation", "prompt_style": "boxed final answer for STEM; ReAct for agentic coding", "temperature": "1.0; top_p=0.97", "context": "256K total context; 32K max output; 1,000 max steps", "notes": "Technical report headline evaluation configuration." }, "matches_canonical": false, "source_type": "official_paper", "audit_status": "verified", "notes": "Displayed exactly as 46.0. Official Microsoft-reported result." }, { "model_id": "mai-thinking-1", "benchmark_id": "truthfulqa_mc", "score": 88.0, "reference_url": "https://microsoft.ai/wp-content/uploads/2026/06/main_20260602_2.pdf", "reported_setting": { "mode": "thinking", "effort": "maximum reasoning effort", "tools": "none; benchmark-provided tools for BFCL v3", "sampling": "average of 4 runs", "judge": "benchmark-specific", "harness": "Microsoft official evaluation suite", "prompt_style": "benchmark-specific Microsoft implementation", "temperature": "1.0; top_p=0.97", "context": "maximum sequence length", "notes": "Table 12 configuration from the technical report." }, "matches_canonical": false, "source_type": "official_paper", "audit_status": "verified", "notes": "Displayed exactly as 88. Official Microsoft-reported result." }, { "model_id": "claude-opus-4.6", "benchmark_id": "claw_eval_multimodal_pass3", "score": 24.8, "reference_url": "https://raw.githubusercontent.com/claw-eval/claw-eval.github.io/19d378a646d892ca6c8148691aea02da1b184eb8/assets/benchmark-0fIVfCD7.js", "reported_setting": { "mode": "source does not state", "effort": "source does not state", "tools": "official Claw-Eval agent environment", "sampling": "Pass^3; all three trajectories must pass", "judge": "full-trajectory benchmark grading", "harness": "official Claw-Eval v1.1 leaderboard", "prompt_style": "official Claw-Eval task prompts", "temperature": "source does not state", "context": "source does not state", "input_modalities": "image and text", "trials": "3 per task; 303 generations", "task_subset": "multimodal 101-task subset", "task_count": 101 }, "matches_canonical": false, "source_type": "official_model_card", "audit_status": "verified", "notes": "Exact official leaderboard field for the multimodal 101-task subset. Model mode/effort is not inferred." }, { "model_id": "claude-opus-4.6", "benchmark_id": "claw_eval_pass3", "score": 70.4, "reference_url": "https://raw.githubusercontent.com/claw-eval/claw-eval.github.io/19d378a646d892ca6c8148691aea02da1b184eb8/assets/benchmark-0fIVfCD7.js", "reported_setting": { "mode": "source does not state", "effort": "source does not state", "tools": "official Claw-Eval agent environment", "sampling": "Pass^3; all three trajectories must pass", "judge": "full-trajectory benchmark grading", "harness": "official Claw-Eval v1.1 leaderboard", "prompt_style": "official Claw-Eval task prompts", "temperature": "source does not state", "context": "source does not state", "input_modalities": "text", "trials": "3 per task; 597 generations", "task_subset": "non-multimodal 199-task subset", "task_count": 199 }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Exact official leaderboard field for the non-multimodal 199-task subset. Model mode/effort is not inferred.", "candidates": [ { "score": 70.8, "reference_url": "https://raw.githubusercontent.com/claw-eval/claw-eval.github.io/19d378a646d892ca6c8148691aea02da1b184eb8/assets/benchmark-0fIVfCD7.js", "source_type": "official_model_card", "reported_setting": { "mode": "source does not state", "effort": "source does not state", "tools": "official Claw-Eval agent environment", "sampling": "Pass^3; all three trajectories must pass", "judge": "full-trajectory benchmark grading", "harness": "official Claw-Eval v1.1 leaderboard", "prompt_style": "official Claw-Eval task prompts", "temperature": "source does not state", "context": "source does not state", "input_modalities": "text", "trials": "3 per task; 483 generations", "task_subset": "core-general 161-task subset", "task_count": 161 }, "notes": "Exact official leaderboard field for the core-general 161-task subset. Model mode/effort is not inferred." } ] }, { "model_id": "claude-sonnet-4.6", "benchmark_id": "claw_eval_multimodal_pass3", "score": 23.8, "reference_url": "https://raw.githubusercontent.com/claw-eval/claw-eval.github.io/19d378a646d892ca6c8148691aea02da1b184eb8/assets/benchmark-0fIVfCD7.js", "reported_setting": { "mode": "source does not state", "effort": "source does not state", "tools": "official Claw-Eval agent environment", "sampling": "Pass^3; all three trajectories must pass", "judge": "full-trajectory benchmark grading", "harness": "official Claw-Eval v1.1 leaderboard", "prompt_style": "official Claw-Eval task prompts", "temperature": "source does not state", "context": "source does not state", "input_modalities": "image and text", "trials": "3 per task; 303 generations", "task_subset": "multimodal 101-task subset", "task_count": 101 }, "matches_canonical": false, "source_type": "official_model_card", "audit_status": "verified", "notes": "Exact official leaderboard field for the multimodal 101-task subset. Model mode/effort is not inferred." }, { "model_id": "claude-sonnet-4.6", "benchmark_id": "claw_eval_pass3", "score": 68.3, "reference_url": "https://raw.githubusercontent.com/claw-eval/claw-eval.github.io/19d378a646d892ca6c8148691aea02da1b184eb8/assets/benchmark-0fIVfCD7.js", "reported_setting": { "mode": "source does not state", "effort": "source does not state", "tools": "official Claw-Eval agent environment", "sampling": "Pass^3; all three trajectories must pass", "judge": "full-trajectory benchmark grading", "harness": "official Claw-Eval v1.1 leaderboard", "prompt_style": "official Claw-Eval task prompts", "temperature": "source does not state", "context": "source does not state", "input_modalities": "text", "trials": "3 per task; 483 generations", "task_subset": "core-general 161-task subset", "task_count": 161 }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Average tokens per trajectory=142233 is a physical efficiency annotation, not a benchmark score or score setting. Exact official leaderboard field for the core-general 161-task subset. Model mode/effort is not inferred." }, { "model_id": "deepseek-v3.2", "benchmark_id": "claw_eval_pass3", "score": 42.2, "reference_url": "https://raw.githubusercontent.com/claw-eval/claw-eval.github.io/19d378a646d892ca6c8148691aea02da1b184eb8/assets/benchmark-0fIVfCD7.js", "reported_setting": { "mode": "source does not state", "effort": "source does not state", "tools": "official Claw-Eval agent environment", "sampling": "Pass^3; all three trajectories must pass", "judge": "full-trajectory benchmark grading", "harness": "official Claw-Eval v1.1 leaderboard", "prompt_style": "official Claw-Eval task prompts", "temperature": "source does not state", "context": "source does not state", "input_modalities": "text", "trials": "3 per task; 483 generations", "task_subset": "core-general 161-task subset", "task_count": 161 }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Average tokens per trajectory=237262 is a physical efficiency annotation, not a benchmark score or score setting. Exact official leaderboard field for the core-general 161-task subset. Model mode/effort is not inferred." }, { "model_id": "deepseek-v4-flash", "benchmark_id": "bigbench_hard", "score": 86.9, "reference_url": "https://huggingface.co/XiaomiMiMo/MiMo-V2.5-Pro/resolve/21d1ecfecd7bd70f31be25ca49d7edd21f003659/README.md", "reported_setting": { "mode": "base pretrained checkpoint", "effort": "not applicable", "tools": "none", "sampling": "3-shot; single completion per prompt", "judge": "benchmark-native rule-based evaluator", "harness": "Xiaomi MiMo base-checkpoint evaluation", "prompt_style": "3-shot few-shot continuation", "temperature": "source does not state", "context": "source does not state", "input_modalities": "text", "trials": "one completion per item", "checkpoint": "base" }, "matches_canonical": false, "source_type": "official_model_card_base_checkpoint", "audit_status": "verified", "notes": "Base checkpoint observation attached to the canonical post-trained model ID. Base result is attached to the canonical post-trained model ID. The Xiaomi table does not state this comparator Base checkpoint's context length, and no comparator Base config is locked; no Pro context value or conflict is copied onto this observation." }, { "model_id": "deepseek-v4-flash", "benchmark_id": "c_eval", "score": 92.1, "reference_url": "https://huggingface.co/XiaomiMiMo/MiMo-V2.5-Pro/resolve/21d1ecfecd7bd70f31be25ca49d7edd21f003659/README.md", "reported_setting": { "mode": "base pretrained checkpoint", "effort": "not applicable", "tools": "none", "sampling": "5-shot; single completion per prompt", "judge": "benchmark-native rule-based evaluator", "harness": "Xiaomi MiMo base-checkpoint evaluation", "prompt_style": "5-shot few-shot continuation", "temperature": "source does not state", "context": "source does not state", "input_modalities": "text", "trials": "one completion per item", "checkpoint": "base" }, "matches_canonical": false, "source_type": "official_model_card_base_checkpoint", "audit_status": "verified", "notes": "Base checkpoint observation attached to the canonical post-trained model ID. Base result is attached to the canonical post-trained model ID. The Xiaomi table does not state this comparator Base checkpoint's context length, and no comparator Base config is locked; no Pro context value or conflict is copied onto this observation." }, { "model_id": "deepseek-v4-flash", "benchmark_id": "cmmlu", "score": 90.4, "reference_url": "https://huggingface.co/XiaomiMiMo/MiMo-V2.5-Pro/resolve/21d1ecfecd7bd70f31be25ca49d7edd21f003659/README.md", "reported_setting": { "mode": "base pretrained checkpoint", "effort": "not applicable", "tools": "none", "sampling": "5-shot; single completion per prompt", "judge": "benchmark-native rule-based evaluator", "harness": "Xiaomi MiMo base-checkpoint evaluation", "prompt_style": "5-shot few-shot continuation", "temperature": "source does not state", "context": "source does not state", "input_modalities": "text", "trials": "one completion per item", "checkpoint": "base" }, "matches_canonical": false, "source_type": "official_model_card_base_checkpoint", "audit_status": "verified", "notes": "Base checkpoint observation attached to the canonical post-trained model ID. Base result is attached to the canonical post-trained model ID. The Xiaomi table does not state this comparator Base checkpoint's context length, and no comparator Base config is locked; no Pro context value or conflict is copied onto this observation." }, { "model_id": "deepseek-v4-flash", "benchmark_id": "drop", "score": 88.6, "reference_url": "https://huggingface.co/XiaomiMiMo/MiMo-V2.5-Pro/resolve/21d1ecfecd7bd70f31be25ca49d7edd21f003659/README.md", "reported_setting": { "mode": "base pretrained checkpoint", "effort": "not applicable", "tools": "none", "sampling": "3-shot; single completion per prompt", "judge": "benchmark-native rule-based evaluator", "harness": "Xiaomi MiMo base-checkpoint evaluation", "prompt_style": "3-shot few-shot continuation", "temperature": "source does not state", "context": "source does not state", "input_modalities": "text", "trials": "one completion per item", "checkpoint": "base" }, "matches_canonical": false, "source_type": "official_model_card_base_checkpoint", "audit_status": "verified", "notes": "Base checkpoint observation attached to the canonical post-trained model ID. Base result is attached to the canonical post-trained model ID. The Xiaomi table does not state this comparator Base checkpoint's context length, and no comparator Base config is locked; no Pro context value or conflict is copied onto this observation." }, { "model_id": "deepseek-v4-flash", "benchmark_id": "gsm8k", "score": 90.8, "reference_url": "https://huggingface.co/XiaomiMiMo/MiMo-V2.5-Pro/resolve/21d1ecfecd7bd70f31be25ca49d7edd21f003659/README.md", "reported_setting": { "mode": "base pretrained checkpoint", "effort": "not applicable", "tools": "none", "sampling": "8-shot; single completion per prompt", "judge": "benchmark-native rule-based evaluator", "harness": "Xiaomi MiMo base-checkpoint evaluation", "prompt_style": "8-shot few-shot continuation", "temperature": "source does not state", "context": "source does not state", "input_modalities": "text", "trials": "one completion per item", "checkpoint": "base" }, "matches_canonical": false, "source_type": "official_model_card_base_checkpoint", "audit_status": "verified", "notes": "Base checkpoint observation attached to the canonical post-trained model ID. Base result is attached to the canonical post-trained model ID. The Xiaomi table does not state this comparator Base checkpoint's context length, and no comparator Base config is locked; no Pro context value or conflict is copied onto this observation." }, { "model_id": "deepseek-v4-flash", "benchmark_id": "hellaswag", "score": 85.7, "reference_url": "https://huggingface.co/XiaomiMiMo/MiMo-V2.5-Pro/resolve/21d1ecfecd7bd70f31be25ca49d7edd21f003659/README.md", "reported_setting": { "mode": "base pretrained checkpoint", "effort": "not applicable", "tools": "none", "sampling": "10-shot; single completion per prompt", "judge": "benchmark-native rule-based evaluator", "harness": "Xiaomi MiMo base-checkpoint evaluation", "prompt_style": "10-shot few-shot continuation", "temperature": "source does not state", "context": "source does not state", "input_modalities": "text", "trials": "one completion per item", "checkpoint": "base" }, "matches_canonical": false, "source_type": "official_model_card_base_checkpoint", "audit_status": "verified", "notes": "Base checkpoint observation attached to the canonical post-trained model ID. Base result is attached to the canonical post-trained model ID. The Xiaomi table does not state this comparator Base checkpoint's context length, and no comparator Base config is locked; no Pro context value or conflict is copied onto this observation." }, { "model_id": "deepseek-v4-flash", "benchmark_id": "math", "score": 57.4, "reference_url": "https://huggingface.co/XiaomiMiMo/MiMo-V2.5-Pro/resolve/21d1ecfecd7bd70f31be25ca49d7edd21f003659/README.md", "reported_setting": { "mode": "base pretrained checkpoint", "effort": "not applicable", "tools": "none", "sampling": "4-shot; single completion per prompt", "judge": "benchmark-native rule-based evaluator", "harness": "Xiaomi MiMo base-checkpoint evaluation", "prompt_style": "4-shot few-shot continuation", "temperature": "source does not state", "context": "source does not state", "input_modalities": "text", "trials": "one completion per item", "checkpoint": "base" }, "matches_canonical": false, "source_type": "official_model_card_base_checkpoint", "audit_status": "verified", "notes": "Base checkpoint observation attached to the canonical post-trained model ID. Base result is attached to the canonical post-trained model ID. The Xiaomi table does not state this comparator Base checkpoint's context length, and no comparator Base config is locked; no Pro context value or conflict is copied onto this observation." }, { "model_id": "deepseek-v4-flash", "benchmark_id": "mmlu", "score": 88.7, "reference_url": "https://huggingface.co/XiaomiMiMo/MiMo-V2.5-Pro/resolve/21d1ecfecd7bd70f31be25ca49d7edd21f003659/README.md", "reported_setting": { "mode": "base pretrained checkpoint", "effort": "not applicable", "tools": "none", "sampling": "5-shot; single completion per prompt", "judge": "benchmark-native rule-based evaluator", "harness": "Xiaomi MiMo base-checkpoint evaluation", "prompt_style": "5-shot few-shot continuation", "temperature": "source does not state", "context": "source does not state", "input_modalities": "text", "trials": "one completion per item", "checkpoint": "base" }, "matches_canonical": false, "source_type": "official_model_card_base_checkpoint", "audit_status": "verified", "notes": "Base checkpoint observation attached to the canonical post-trained model ID. Base result is attached to the canonical post-trained model ID. The Xiaomi table does not state this comparator Base checkpoint's context length, and no comparator Base config is locked; no Pro context value or conflict is copied onto this observation." }, { "model_id": "deepseek-v4-flash", "benchmark_id": "mmlu_redux", "score": 89.4, "reference_url": "https://huggingface.co/XiaomiMiMo/MiMo-V2.5-Pro/resolve/21d1ecfecd7bd70f31be25ca49d7edd21f003659/README.md", "reported_setting": { "mode": "base pretrained checkpoint", "effort": "not applicable", "tools": "none", "sampling": "5-shot; single completion per prompt", "judge": "benchmark-native rule-based evaluator", "harness": "Xiaomi MiMo base-checkpoint evaluation", "prompt_style": "5-shot few-shot continuation", "temperature": "source does not state", "context": "source does not state", "input_modalities": "text", "trials": "one completion per item", "checkpoint": "base" }, "matches_canonical": false, "source_type": "official_model_card_base_checkpoint", "audit_status": "verified", "notes": "Base checkpoint observation attached to the canonical post-trained model ID. Base result is attached to the canonical post-trained model ID. The Xiaomi table does not state this comparator Base checkpoint's context length, and no comparator Base config is locked; no Pro context value or conflict is copied onto this observation." }, { "model_id": "deepseek-v4-pro", "benchmark_id": "bigbench_hard", "score": 87.5, "reference_url": "https://huggingface.co/XiaomiMiMo/MiMo-V2.5-Pro/resolve/21d1ecfecd7bd70f31be25ca49d7edd21f003659/README.md", "reported_setting": { "mode": "base pretrained checkpoint", "effort": "not applicable", "tools": "none", "sampling": "3-shot; single completion per prompt", "judge": "benchmark-native rule-based evaluator", "harness": "Xiaomi MiMo base-checkpoint evaluation", "prompt_style": "3-shot few-shot continuation", "temperature": "source does not state", "context": "source does not state", "input_modalities": "text", "trials": "one completion per item", "checkpoint": "base" }, "matches_canonical": false, "source_type": "official_model_card_base_checkpoint", "audit_status": "verified", "notes": "Base checkpoint observation attached to the canonical post-trained model ID. Base result is attached to the canonical post-trained model ID. The Xiaomi table does not state this comparator Base checkpoint's context length, and no comparator Base config is locked; no Pro context value or conflict is copied onto this observation." }, { "model_id": "deepseek-v4-pro", "benchmark_id": "c_eval", "score": 93.1, "reference_url": "https://huggingface.co/XiaomiMiMo/MiMo-V2.5-Pro/resolve/21d1ecfecd7bd70f31be25ca49d7edd21f003659/README.md", "reported_setting": { "mode": "base pretrained checkpoint", "effort": "not applicable", "tools": "none", "sampling": "5-shot; single completion per prompt", "judge": "benchmark-native rule-based evaluator", "harness": "Xiaomi MiMo base-checkpoint evaluation", "prompt_style": "5-shot few-shot continuation", "temperature": "source does not state", "context": "source does not state", "input_modalities": "text", "trials": "one completion per item", "checkpoint": "base" }, "matches_canonical": false, "source_type": "official_model_card_base_checkpoint", "audit_status": "verified", "notes": "Base checkpoint observation attached to the canonical post-trained model ID. Base result is attached to the canonical post-trained model ID. The Xiaomi table does not state this comparator Base checkpoint's context length, and no comparator Base config is locked; no Pro context value or conflict is copied onto this observation." }, { "model_id": "deepseek-v4-pro", "benchmark_id": "cmmlu", "score": 90.8, "reference_url": "https://huggingface.co/XiaomiMiMo/MiMo-V2.5-Pro/resolve/21d1ecfecd7bd70f31be25ca49d7edd21f003659/README.md", "reported_setting": { "mode": "base pretrained checkpoint", "effort": "not applicable", "tools": "none", "sampling": "5-shot; single completion per prompt", "judge": "benchmark-native rule-based evaluator", "harness": "Xiaomi MiMo base-checkpoint evaluation", "prompt_style": "5-shot few-shot continuation", "temperature": "source does not state", "context": "source does not state", "input_modalities": "text", "trials": "one completion per item", "checkpoint": "base" }, "matches_canonical": false, "source_type": "official_model_card_base_checkpoint", "audit_status": "verified", "notes": "Base checkpoint observation attached to the canonical post-trained model ID. Base result is attached to the canonical post-trained model ID. The Xiaomi table does not state this comparator Base checkpoint's context length, and no comparator Base config is locked; no Pro context value or conflict is copied onto this observation." }, { "model_id": "deepseek-v4-pro", "benchmark_id": "drop", "score": 88.7, "reference_url": "https://huggingface.co/XiaomiMiMo/MiMo-V2.5-Pro/resolve/21d1ecfecd7bd70f31be25ca49d7edd21f003659/README.md", "reported_setting": { "mode": "base pretrained checkpoint", "effort": "not applicable", "tools": "none", "sampling": "3-shot; single completion per prompt", "judge": "benchmark-native rule-based evaluator", "harness": "Xiaomi MiMo base-checkpoint evaluation", "prompt_style": "3-shot few-shot continuation", "temperature": "source does not state", "context": "source does not state", "input_modalities": "text", "trials": "one completion per item", "checkpoint": "base" }, "matches_canonical": false, "source_type": "official_model_card_base_checkpoint", "audit_status": "verified", "notes": "Base checkpoint observation attached to the canonical post-trained model ID. Base result is attached to the canonical post-trained model ID. The Xiaomi table does not state this comparator Base checkpoint's context length, and no comparator Base config is locked; no Pro context value or conflict is copied onto this observation." }, { "model_id": "deepseek-v4-pro", "benchmark_id": "gsm8k", "score": 92.6, "reference_url": "https://huggingface.co/XiaomiMiMo/MiMo-V2.5-Pro/resolve/21d1ecfecd7bd70f31be25ca49d7edd21f003659/README.md", "reported_setting": { "mode": "base pretrained checkpoint", "effort": "not applicable", "tools": "none", "sampling": "8-shot; single completion per prompt", "judge": "benchmark-native rule-based evaluator", "harness": "Xiaomi MiMo base-checkpoint evaluation", "prompt_style": "8-shot few-shot continuation", "temperature": "source does not state", "context": "source does not state", "input_modalities": "text", "trials": "one completion per item", "checkpoint": "base" }, "matches_canonical": false, "source_type": "official_model_card_base_checkpoint", "audit_status": "verified", "notes": "Base checkpoint observation attached to the canonical post-trained model ID. Base result is attached to the canonical post-trained model ID. The Xiaomi table does not state this comparator Base checkpoint's context length, and no comparator Base config is locked; no Pro context value or conflict is copied onto this observation." }, { "model_id": "deepseek-v4-pro", "benchmark_id": "hellaswag", "score": 88.0, "reference_url": "https://huggingface.co/XiaomiMiMo/MiMo-V2.5-Pro/resolve/21d1ecfecd7bd70f31be25ca49d7edd21f003659/README.md", "reported_setting": { "mode": "base pretrained checkpoint", "effort": "not applicable", "tools": "none", "sampling": "10-shot; single completion per prompt", "judge": "benchmark-native rule-based evaluator", "harness": "Xiaomi MiMo base-checkpoint evaluation", "prompt_style": "10-shot few-shot continuation", "temperature": "source does not state", "context": "source does not state", "input_modalities": "text", "trials": "one completion per item", "checkpoint": "base" }, "matches_canonical": false, "source_type": "official_model_card_base_checkpoint", "audit_status": "verified", "notes": "Base checkpoint observation attached to the canonical post-trained model ID. Base result is attached to the canonical post-trained model ID. The Xiaomi table does not state this comparator Base checkpoint's context length, and no comparator Base config is locked; no Pro context value or conflict is copied onto this observation." }, { "model_id": "deepseek-v4-pro", "benchmark_id": "hle", "score": 37.7, "reference_url": "https://huggingface.co/deepseek-ai/DeepSeek-V4-Pro/resolve/b5968e9190ef611bbf34a7229255be88a0e937c1/README.md", "reported_setting": { "mode": "thinking (Think Max)", "effort": "Max (DeepSeek-V4-Pro-Max)", "tools": "source does not state", "sampling": "Pass@1", "judge": "source does not state", "harness": "source does not state", "prompt_style": "source does not state", "temperature": "source does not state", "context": "source does not state", "input_modalities": "source does not state", "trials": "source does not state" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Semantic alternative printed inside the same physical HLE cell; it does not add a physical position. Exact DeepSeek official model-card score from the DeepSeek-V4-Pro-Max table. Only the source-backed Think Max mode and Max effort are retained; benchmark-specific harness and unstated sampling details remain unknown." }, { "model_id": "deepseek-v4-pro", "benchmark_id": "hle_tools", "score": 48.2, "reference_url": "https://huggingface.co/deepseek-ai/DeepSeek-V4-Pro/resolve/b5968e9190ef611bbf34a7229255be88a0e937c1/README.md", "reported_setting": { "mode": "thinking (Think Max)", "effort": "Max (DeepSeek-V4-Pro-Max)", "tools": "with tools; exact tool suite source does not state", "sampling": "Pass@1", "judge": "source does not state", "harness": "source does not state", "prompt_style": "source does not state", "temperature": "source does not state", "context": "source does not state", "input_modalities": "source does not state", "trials": "source does not state" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "DeepSeek V4 Pro is explicitly reported at max effort. Exact DeepSeek official model-card score from the DeepSeek-V4-Pro-Max table. Only the source-backed Think Max mode and Max effort are retained; benchmark-specific harness and unstated sampling details remain unknown.", "candidates": [ { "score": 48.2, "reference_url": "https://static.stepfun.com/blog/step-3.7-flash/", "source_type": "official_blog", "reported_setting": { "mode": "thinking (Think Max)", "effort": "Max (DeepSeek-V4-Pro-Max)", "tools": "with tools; exact tool suite source does not state", "sampling": "Pass@1", "judge": "source does not state", "harness": "source does not state", "prompt_style": "source does not state", "temperature": "source does not state", "context": "source does not state", "input_modalities": "source does not state", "trials": "source does not state", "multimodal_input": true }, "notes": "Official StepFun launch comparison table. Provider-quoted value is independently locked by the campaign's primary official-source record." } ] }, { "model_id": "deepseek-v4-pro", "benchmark_id": "math", "score": 64.5, "reference_url": "https://huggingface.co/XiaomiMiMo/MiMo-V2.5-Pro/resolve/21d1ecfecd7bd70f31be25ca49d7edd21f003659/README.md", "reported_setting": { "mode": "base pretrained checkpoint", "effort": "not applicable", "tools": "none", "sampling": "4-shot; single completion per prompt", "judge": "benchmark-native rule-based evaluator", "harness": "Xiaomi MiMo base-checkpoint evaluation", "prompt_style": "4-shot few-shot continuation", "temperature": "source does not state", "context": "source does not state", "input_modalities": "text", "trials": "one completion per item", "checkpoint": "base" }, "matches_canonical": false, "source_type": "official_model_card_base_checkpoint", "audit_status": "verified", "notes": "Base checkpoint observation attached to the canonical post-trained model ID. Base result is attached to the canonical post-trained model ID. The Xiaomi table does not state this comparator Base checkpoint's context length, and no comparator Base config is locked; no Pro context value or conflict is copied onto this observation." }, { "model_id": "deepseek-v4-pro", "benchmark_id": "mmlu", "score": 90.1, "reference_url": "https://huggingface.co/XiaomiMiMo/MiMo-V2.5-Pro/resolve/21d1ecfecd7bd70f31be25ca49d7edd21f003659/README.md", "reported_setting": { "mode": "base pretrained checkpoint", "effort": "not applicable", "tools": "none", "sampling": "5-shot; single completion per prompt", "judge": "benchmark-native rule-based evaluator", "harness": "Xiaomi MiMo base-checkpoint evaluation", "prompt_style": "5-shot few-shot continuation", "temperature": "source does not state", "context": "source does not state", "input_modalities": "text", "trials": "one completion per item", "checkpoint": "base" }, "matches_canonical": false, "source_type": "official_model_card_base_checkpoint", "audit_status": "verified", "notes": "Base checkpoint observation attached to the canonical post-trained model ID. Base result is attached to the canonical post-trained model ID. The Xiaomi table does not state this comparator Base checkpoint's context length, and no comparator Base config is locked; no Pro context value or conflict is copied onto this observation." }, { "model_id": "deepseek-v4-pro", "benchmark_id": "mmlu_redux", "score": 90.8, "reference_url": "https://huggingface.co/XiaomiMiMo/MiMo-V2.5-Pro/resolve/21d1ecfecd7bd70f31be25ca49d7edd21f003659/README.md", "reported_setting": { "mode": "base pretrained checkpoint", "effort": "not applicable", "tools": "none", "sampling": "5-shot; single completion per prompt", "judge": "benchmark-native rule-based evaluator", "harness": "Xiaomi MiMo base-checkpoint evaluation", "prompt_style": "5-shot few-shot continuation", "temperature": "source does not state", "context": "source does not state", "input_modalities": "text", "trials": "one completion per item", "checkpoint": "base" }, "matches_canonical": false, "source_type": "official_model_card_base_checkpoint", "audit_status": "verified", "notes": "Base checkpoint observation attached to the canonical post-trained model ID. Base result is attached to the canonical post-trained model ID. The Xiaomi table does not state this comparator Base checkpoint's context length, and no comparator Base config is locked; no Pro context value or conflict is copied onto this observation." }, { "model_id": "gemini-3-flash", "benchmark_id": "claw_eval_pass3", "score": 48.4, "reference_url": "https://raw.githubusercontent.com/claw-eval/claw-eval.github.io/19d378a646d892ca6c8148691aea02da1b184eb8/assets/benchmark-0fIVfCD7.js", "reported_setting": { "mode": "source does not state", "effort": "source does not state", "tools": "official Claw-Eval agent environment", "sampling": "Pass^3; all three trajectories must pass", "judge": "full-trajectory benchmark grading", "harness": "official Claw-Eval v1.1 leaderboard", "prompt_style": "official Claw-Eval task prompts", "temperature": "source does not state", "context": "source does not state", "input_modalities": "text", "trials": "3 per task; 483 generations", "task_subset": "core-general 161-task subset", "task_count": 161 }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Average tokens per trajectory=82126 is a physical efficiency annotation, not a benchmark score or score setting. Exact official leaderboard field for the core-general 161-task subset. Model mode/effort is not inferred." }, { "model_id": "glm-5.1", "benchmark_id": "hle", "score": 31.0, "reference_url": "https://z.ai/blog/assets/glm-5.1-B3CLmgrT.js", "reported_setting": { "mode": "thinking", "effort": "official reported setting", "tools": "none", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official Z.ai GLM-5.1 evaluation", "prompt_style": "provider official evaluation prompt", "temperature": "source/provider setting", "context": "source/provider setting", "input_modalities": "text", "trials": "source-reported" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Semantic alternative printed inside the same physical HLE cell; it does not add a physical position. Exact provider-official score and reported setting." }, { "model_id": "glm-5.1", "benchmark_id": "hle_tools", "score": 52.3, "reference_url": "https://z.ai/blog/assets/glm-5.1-B3CLmgrT.js", "reported_setting": { "mode": "thinking", "effort": "official reported setting", "tools": "source-reported search/code/web tools", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official Z.ai GLM-5.1 evaluation", "prompt_style": "provider official evaluation prompt", "temperature": "source/provider setting", "context": "source/provider setting", "input_modalities": "text", "trials": "source-reported" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Exact provider-official score and reported setting.", "candidates": [ { "score": 52.3, "reference_url": "https://static.stepfun.com/blog/step-3.7-flash/", "source_type": "official_blog", "reported_setting": { "mode": "thinking", "effort": "official reported setting", "tools": "source-reported search/code/web tools", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official Z.ai GLM-5.1 evaluation", "prompt_style": "provider official evaluation prompt", "temperature": "source/provider setting", "context": "source/provider setting", "input_modalities": "text", "trials": "source-reported", "multimodal_input": true }, "notes": "Official StepFun launch comparison table. Provider-quoted value is independently locked by the campaign's primary official-source record." } ] }, { "model_id": "gpt-5.4", "benchmark_id": "graphwalks_bfs_bucket_0_4k", "score": 100, "reference_url": "https://mimo.xiaomi.com/mimo-v2-5-pro", "reported_setting": { "mode": "source does not state", "effort": "source does not state", "tools": "none", "sampling": "one response per graph prompt", "judge": "deterministic set-overlap F1 against answer node list", "harness": "Xiaomi MiMo GraphWalks evaluation with Anthropic fixes", "prompt_style": "3-shot GraphWalks prompt", "temperature": "source does not state", "context": "0–4K", "input_modalities": "text", "trials": "n=58", "context_window_variant": "standard" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Exact heatmap bucket. Counts and the 1M variant are preserved; no aggregation into campaign range rows." }, { "model_id": "gpt-5.4", "benchmark_id": "graphwalks_bfs_bucket_128k_256k", "score": 57, "reference_url": "https://mimo.xiaomi.com/mimo-v2-5-pro", "reported_setting": { "mode": "source does not state", "effort": "source does not state", "tools": "none", "sampling": "one response per graph prompt", "judge": "deterministic set-overlap F1 against answer node list", "harness": "Xiaomi MiMo GraphWalks evaluation with Anthropic fixes", "prompt_style": "3-shot GraphWalks prompt", "temperature": "source does not state", "context": "128K–256K", "input_modalities": "text", "trials": "n=100", "context_window_variant": "standard" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Exact heatmap bucket. Counts and the 1M variant are preserved; no aggregation into campaign range rows." }, { "model_id": "gpt-5.4", "benchmark_id": "graphwalks_bfs_bucket_16k_32k", "score": 92, "reference_url": "https://mimo.xiaomi.com/mimo-v2-5-pro", "reported_setting": { "mode": "source does not state", "effort": "source does not state", "tools": "none", "sampling": "one response per graph prompt", "judge": "deterministic set-overlap F1 against answer node list", "harness": "Xiaomi MiMo GraphWalks evaluation with Anthropic fixes", "prompt_style": "3-shot GraphWalks prompt", "temperature": "source does not state", "context": "16K–32K", "input_modalities": "text", "trials": "n=49", "context_window_variant": "standard" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Exact heatmap bucket. Counts and the 1M variant are preserved; no aggregation into campaign range rows." }, { "model_id": "gpt-5.4", "benchmark_id": "graphwalks_bfs_bucket_32k_64k", "score": 90, "reference_url": "https://mimo.xiaomi.com/mimo-v2-5-pro", "reported_setting": { "mode": "source does not state", "effort": "source does not state", "tools": "none", "sampling": "one response per graph prompt", "judge": "deterministic set-overlap F1 against answer node list", "harness": "Xiaomi MiMo GraphWalks evaluation with Anthropic fixes", "prompt_style": "3-shot GraphWalks prompt", "temperature": "source does not state", "context": "32K–64K", "input_modalities": "text", "trials": "n=50", "context_window_variant": "standard" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Exact heatmap bucket. Counts and the 1M variant are preserved; no aggregation into campaign range rows." }, { "model_id": "gpt-5.4", "benchmark_id": "graphwalks_bfs_bucket_4k_8k", "score": 100, "reference_url": "https://mimo.xiaomi.com/mimo-v2-5-pro", "reported_setting": { "mode": "source does not state", "effort": "source does not state", "tools": "none", "sampling": "one response per graph prompt", "judge": "deterministic set-overlap F1 against answer node list", "harness": "Xiaomi MiMo GraphWalks evaluation with Anthropic fixes", "prompt_style": "3-shot GraphWalks prompt", "temperature": "source does not state", "context": "4K–8K", "input_modalities": "text", "trials": "n=77", "context_window_variant": "standard" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Exact heatmap bucket. Counts and the 1M variant are preserved; no aggregation into campaign range rows." }, { "model_id": "gpt-5.4", "benchmark_id": "graphwalks_bfs_bucket_64k_128k", "score": 78, "reference_url": "https://mimo.xiaomi.com/mimo-v2-5-pro", "reported_setting": { "mode": "source does not state", "effort": "source does not state", "tools": "none", "sampling": "one response per graph prompt", "judge": "deterministic set-overlap F1 against answer node list", "harness": "Xiaomi MiMo GraphWalks evaluation with Anthropic fixes", "prompt_style": "3-shot GraphWalks prompt", "temperature": "source does not state", "context": "64K–128K", "input_modalities": "text", "trials": "n=51", "context_window_variant": "standard" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Exact heatmap bucket. Counts and the 1M variant are preserved; no aggregation into campaign range rows." }, { "model_id": "gpt-5.4", "benchmark_id": "graphwalks_bfs_bucket_8k_16k", "score": 100, "reference_url": "https://mimo.xiaomi.com/mimo-v2-5-pro", "reported_setting": { "mode": "source does not state", "effort": "source does not state", "tools": "none", "sampling": "one response per graph prompt", "judge": "deterministic set-overlap F1 against answer node list", "harness": "Xiaomi MiMo GraphWalks evaluation with Anthropic fixes", "prompt_style": "3-shot GraphWalks prompt", "temperature": "source does not state", "context": "8K–16K", "input_modalities": "text", "trials": "n=84", "context_window_variant": "standard" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Exact heatmap bucket. Counts and the 1M variant are preserved; no aggregation into campaign range rows." }, { "model_id": "gpt-5.4", "benchmark_id": "graphwalks_parents_bucket_0_4k", "score": 100, "reference_url": "https://mimo.xiaomi.com/mimo-v2-5-pro", "reported_setting": { "mode": "source does not state", "effort": "source does not state", "tools": "none", "sampling": "one response per graph prompt", "judge": "deterministic set-overlap F1 against answer node list", "harness": "Xiaomi MiMo GraphWalks evaluation with Anthropic fixes", "prompt_style": "3-shot GraphWalks prompt", "temperature": "source does not state", "context": "0–4K", "input_modalities": "text", "trials": "n=120", "context_window_variant": "standard" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Exact heatmap bucket. Counts and the 1M variant are preserved; no aggregation into campaign range rows." }, { "model_id": "gpt-5.4", "benchmark_id": "graphwalks_parents_bucket_128k_256k", "score": 85, "reference_url": "https://mimo.xiaomi.com/mimo-v2-5-pro", "reported_setting": { "mode": "source does not state", "effort": "source does not state", "tools": "none", "sampling": "one response per graph prompt", "judge": "deterministic set-overlap F1 against answer node list", "harness": "Xiaomi MiMo GraphWalks evaluation with Anthropic fixes", "prompt_style": "3-shot GraphWalks prompt", "temperature": "source does not state", "context": "128K–256K", "input_modalities": "text", "trials": "n=100", "context_window_variant": "standard" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Exact heatmap bucket. Counts and the 1M variant are preserved; no aggregation into campaign range rows." }, { "model_id": "gpt-5.4", "benchmark_id": "graphwalks_parents_bucket_16k_32k", "score": 100, "reference_url": "https://mimo.xiaomi.com/mimo-v2-5-pro", "reported_setting": { "mode": "source does not state", "effort": "source does not state", "tools": "none", "sampling": "one response per graph prompt", "judge": "deterministic set-overlap F1 against answer node list", "harness": "Xiaomi MiMo GraphWalks evaluation with Anthropic fixes", "prompt_style": "3-shot GraphWalks prompt", "temperature": "source does not state", "context": "16K–32K", "input_modalities": "text", "trials": "n=50", "context_window_variant": "standard" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Exact heatmap bucket. Counts and the 1M variant are preserved; no aggregation into campaign range rows." }, { "model_id": "gpt-5.4", "benchmark_id": "graphwalks_parents_bucket_32k_64k", "score": 99, "reference_url": "https://mimo.xiaomi.com/mimo-v2-5-pro", "reported_setting": { "mode": "source does not state", "effort": "source does not state", "tools": "none", "sampling": "one response per graph prompt", "judge": "deterministic set-overlap F1 against answer node list", "harness": "Xiaomi MiMo GraphWalks evaluation with Anthropic fixes", "prompt_style": "3-shot GraphWalks prompt", "temperature": "source does not state", "context": "32K–64K", "input_modalities": "text", "trials": "n=50", "context_window_variant": "standard" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Exact heatmap bucket. Counts and the 1M variant are preserved; no aggregation into campaign range rows." }, { "model_id": "gpt-5.4", "benchmark_id": "graphwalks_parents_bucket_4k_8k", "score": 100, "reference_url": "https://mimo.xiaomi.com/mimo-v2-5-pro", "reported_setting": { "mode": "source does not state", "effort": "source does not state", "tools": "none", "sampling": "one response per graph prompt", "judge": "deterministic set-overlap F1 against answer node list", "harness": "Xiaomi MiMo GraphWalks evaluation with Anthropic fixes", "prompt_style": "3-shot GraphWalks prompt", "temperature": "source does not state", "context": "4K–8K", "input_modalities": "text", "trials": "n=76", "context_window_variant": "standard" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Exact heatmap bucket. Counts and the 1M variant are preserved; no aggregation into campaign range rows." }, { "model_id": "gpt-5.4", "benchmark_id": "graphwalks_parents_bucket_64k_128k", "score": 89, "reference_url": "https://mimo.xiaomi.com/mimo-v2-5-pro", "reported_setting": { "mode": "source does not state", "effort": "source does not state", "tools": "none", "sampling": "one response per graph prompt", "judge": "deterministic set-overlap F1 against answer node list", "harness": "Xiaomi MiMo GraphWalks evaluation with Anthropic fixes", "prompt_style": "3-shot GraphWalks prompt", "temperature": "source does not state", "context": "64K–128K", "input_modalities": "text", "trials": "n=50", "context_window_variant": "standard" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Exact heatmap bucket. Counts and the 1M variant are preserved; no aggregation into campaign range rows." }, { "model_id": "gpt-5.4", "benchmark_id": "graphwalks_parents_bucket_8k_16k", "score": 100, "reference_url": "https://mimo.xiaomi.com/mimo-v2-5-pro", "reported_setting": { "mode": "source does not state", "effort": "source does not state", "tools": "none", "sampling": "one response per graph prompt", "judge": "deterministic set-overlap F1 against answer node list", "harness": "Xiaomi MiMo GraphWalks evaluation with Anthropic fixes", "prompt_style": "3-shot GraphWalks prompt", "temperature": "source does not state", "context": "8K–16K", "input_modalities": "text", "trials": "n=54", "context_window_variant": "standard" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Exact heatmap bucket. Counts and the 1M variant are preserved; no aggregation into campaign range rows." }, { "model_id": "kimi-k2", "benchmark_id": "bigbench_hard", "score": 88.7, "reference_url": "https://huggingface.co/XiaomiMiMo/MiMo-V2.5-Pro/resolve/21d1ecfecd7bd70f31be25ca49d7edd21f003659/README.md", "reported_setting": { "mode": "base pretrained checkpoint", "effort": "not applicable", "tools": "none", "sampling": "3-shot; single completion per prompt", "judge": "benchmark-native rule-based evaluator", "harness": "Xiaomi MiMo base-checkpoint evaluation", "prompt_style": "3-shot few-shot continuation", "temperature": "source does not state", "context": "source does not state", "input_modalities": "text", "trials": "one completion per item", "checkpoint": "base" }, "matches_canonical": false, "source_type": "official_model_card_base_checkpoint", "audit_status": "verified", "notes": "Base checkpoint observation attached to the canonical post-trained model ID. Base result is attached to the canonical post-trained model ID. The Xiaomi table does not state this comparator Base checkpoint's context length, and no comparator Base config is locked; no Pro context value or conflict is copied onto this observation." }, { "model_id": "kimi-k2", "benchmark_id": "c_eval", "score": 92.5, "reference_url": "https://huggingface.co/XiaomiMiMo/MiMo-V2.5-Pro/resolve/21d1ecfecd7bd70f31be25ca49d7edd21f003659/README.md", "reported_setting": { "mode": "base pretrained checkpoint", "effort": "not applicable", "tools": "none", "sampling": "5-shot; single completion per prompt", "judge": "benchmark-native rule-based evaluator", "harness": "Xiaomi MiMo base-checkpoint evaluation", "prompt_style": "5-shot few-shot continuation", "temperature": "source does not state", "context": "source does not state", "input_modalities": "text", "trials": "one completion per item", "checkpoint": "base" }, "matches_canonical": false, "source_type": "official_model_card_base_checkpoint", "audit_status": "verified", "notes": "Base checkpoint observation attached to the canonical post-trained model ID. Base result is attached to the canonical post-trained model ID. The Xiaomi table does not state this comparator Base checkpoint's context length, and no comparator Base config is locked; no Pro context value or conflict is copied onto this observation." }, { "model_id": "kimi-k2", "benchmark_id": "cmmlu", "score": 90.9, "reference_url": "https://huggingface.co/XiaomiMiMo/MiMo-V2.5-Pro/resolve/21d1ecfecd7bd70f31be25ca49d7edd21f003659/README.md", "reported_setting": { "mode": "base pretrained checkpoint", "effort": "not applicable", "tools": "none", "sampling": "5-shot; single completion per prompt", "judge": "benchmark-native rule-based evaluator", "harness": "Xiaomi MiMo base-checkpoint evaluation", "prompt_style": "5-shot few-shot continuation", "temperature": "source does not state", "context": "source does not state", "input_modalities": "text", "trials": "one completion per item", "checkpoint": "base" }, "matches_canonical": false, "source_type": "official_model_card_base_checkpoint", "audit_status": "verified", "notes": "Base checkpoint observation attached to the canonical post-trained model ID. Base result is attached to the canonical post-trained model ID. The Xiaomi table does not state this comparator Base checkpoint's context length, and no comparator Base config is locked; no Pro context value or conflict is copied onto this observation." }, { "model_id": "kimi-k2", "benchmark_id": "drop", "score": 83.6, "reference_url": "https://huggingface.co/XiaomiMiMo/MiMo-V2.5-Pro/resolve/21d1ecfecd7bd70f31be25ca49d7edd21f003659/README.md", "reported_setting": { "mode": "base pretrained checkpoint", "effort": "not applicable", "tools": "none", "sampling": "3-shot; single completion per prompt", "judge": "benchmark-native rule-based evaluator", "harness": "Xiaomi MiMo base-checkpoint evaluation", "prompt_style": "3-shot few-shot continuation", "temperature": "source does not state", "context": "source does not state", "input_modalities": "text", "trials": "one completion per item", "checkpoint": "base" }, "matches_canonical": false, "source_type": "official_model_card_base_checkpoint", "audit_status": "verified", "notes": "Base checkpoint observation attached to the canonical post-trained model ID. Base result is attached to the canonical post-trained model ID. The Xiaomi table does not state this comparator Base checkpoint's context length, and no comparator Base config is locked; no Pro context value or conflict is copied onto this observation." }, { "model_id": "kimi-k2", "benchmark_id": "humaneval_plus", "score": 84.8, "reference_url": "https://huggingface.co/XiaomiMiMo/MiMo-V2.5-Pro/resolve/21d1ecfecd7bd70f31be25ca49d7edd21f003659/README.md", "reported_setting": { "mode": "base pretrained checkpoint", "effort": "not applicable", "tools": "none", "sampling": "1-shot; single completion per prompt", "judge": "benchmark-native rule-based evaluator", "harness": "Xiaomi MiMo base-checkpoint evaluation", "prompt_style": "1-shot few-shot continuation", "temperature": "source does not state", "context": "source does not state", "input_modalities": "text", "trials": "one completion per item", "checkpoint": "base" }, "matches_canonical": false, "source_type": "official_model_card_base_checkpoint", "audit_status": "verified", "notes": "Base checkpoint observation attached to the canonical post-trained model ID. Base result is attached to the canonical post-trained model ID. The Xiaomi table does not state this comparator Base checkpoint's context length, and no comparator Base config is locked; no Pro context value or conflict is copied onto this observation." }, { "model_id": "kimi-k2", "benchmark_id": "mbpp_plus", "score": 73.8, "reference_url": "https://huggingface.co/XiaomiMiMo/MiMo-V2.5-Pro/resolve/21d1ecfecd7bd70f31be25ca49d7edd21f003659/README.md", "reported_setting": { "mode": "base pretrained checkpoint", "effort": "not applicable", "tools": "none", "sampling": "3-shot; single completion per prompt", "judge": "benchmark-native rule-based evaluator", "harness": "Xiaomi MiMo base-checkpoint evaluation", "prompt_style": "3-shot few-shot continuation", "temperature": "source does not state", "context": "source does not state", "input_modalities": "text", "trials": "one completion per item", "checkpoint": "base" }, "matches_canonical": false, "source_type": "official_model_card_base_checkpoint", "audit_status": "verified", "notes": "Base checkpoint observation attached to the canonical post-trained model ID. Base result is attached to the canonical post-trained model ID. The Xiaomi table does not state this comparator Base checkpoint's context length, and no comparator Base config is locked; no Pro context value or conflict is copied onto this observation." }, { "model_id": "kimi-k2.5", "benchmark_id": "claw_eval_pass3", "score": 52.8, "reference_url": "https://raw.githubusercontent.com/claw-eval/claw-eval.github.io/19d378a646d892ca6c8148691aea02da1b184eb8/assets/benchmark-0fIVfCD7.js", "reported_setting": { "mode": "source does not state", "effort": "source does not state", "tools": "official Claw-Eval agent environment", "sampling": "Pass^3; all three trajectories must pass", "judge": "full-trajectory benchmark grading", "harness": "official Claw-Eval v1.1 leaderboard", "prompt_style": "official Claw-Eval task prompts", "temperature": "source does not state", "context": "source does not state", "input_modalities": "text", "trials": "3 per task; 483 generations", "task_subset": "core-general 161-task subset", "task_count": 161 }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Average tokens per trajectory=97512 is a physical efficiency annotation, not a benchmark score or score setting. Exact official leaderboard field for the core-general 161-task subset. Model mode/effort is not inferred." }, { "model_id": "kimi-k2.6", "benchmark_id": "hle_tools", "score": 54.0, "reference_url": "https://huggingface.co/moonshotai/Kimi-K2.6/resolve/d9cb81bc88b9bd2dc89877599f23c614ded72b9c/README.md", "reported_setting": { "mode": "thinking", "effort": "max; up to 98,304 generation tokens", "tools": "source-reported search/code/web tools", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official Kimi K2.6 evaluation", "prompt_style": "provider official evaluation prompt", "temperature": "1.0; top_p=1.0", "context": "262,144 (256K)", "input_modalities": "text", "trials": "source-reported" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Exact provider-official score and reported setting.", "candidates": [ { "score": 54.0, "reference_url": "https://static.stepfun.com/blog/step-3.7-flash/", "source_type": "official_blog", "reported_setting": { "mode": "thinking", "effort": "max; up to 98,304 generation tokens", "tools": "source-reported search/code/web tools", "sampling": "pass@1", "judge": "benchmark-specified", "harness": "official Kimi K2.6 evaluation", "prompt_style": "provider official evaluation prompt", "temperature": "1.0; top_p=1.0", "context": "262,144 (256K)", "input_modalities": "text", "trials": "source-reported", "multimodal_input": true }, "notes": "Official StepFun launch comparison table. Provider-quoted value is independently locked by the campaign's primary official-source record." } ] }, { "model_id": "mimo-v2-omni", "benchmark_id": "claw_eval_multimodal_pass3", "score": 15.8, "reference_url": "https://raw.githubusercontent.com/claw-eval/claw-eval.github.io/19d378a646d892ca6c8148691aea02da1b184eb8/assets/benchmark-0fIVfCD7.js", "reported_setting": { "mode": "source does not state", "effort": "source does not state", "tools": "official Claw-Eval agent environment", "sampling": "Pass^3; all three trajectories must pass", "judge": "full-trajectory benchmark grading", "harness": "official Claw-Eval v1.1 leaderboard", "prompt_style": "official Claw-Eval task prompts", "temperature": "source does not state", "context": "source does not state", "input_modalities": "image and text", "trials": "3 per task; 303 generations", "task_subset": "multimodal 101-task subset", "task_count": 101 }, "matches_canonical": false, "source_type": "official_model_card", "audit_status": "verified", "notes": "Exact official leaderboard field for the multimodal 101-task subset. Model mode/effort is not inferred." }, { "model_id": "mimo-v2-omni", "benchmark_id": "claw_eval_pass3", "score": 52.2, "reference_url": "https://raw.githubusercontent.com/claw-eval/claw-eval.github.io/19d378a646d892ca6c8148691aea02da1b184eb8/assets/benchmark-0fIVfCD7.js", "reported_setting": { "mode": "source does not state", "effort": "source does not state", "tools": "official Claw-Eval agent environment", "sampling": "Pass^3; all three trajectories must pass", "judge": "full-trajectory benchmark grading", "harness": "official Claw-Eval v1.1 leaderboard", "prompt_style": "official Claw-Eval task prompts", "temperature": "source does not state", "context": "source does not state", "input_modalities": "text", "trials": "3 per task; 483 generations", "task_subset": "core-general 161-task subset", "task_count": 161 }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Average tokens per trajectory=141781 is a physical efficiency annotation, not a benchmark score or score setting. Exact official leaderboard field for the core-general 161-task subset. Model mode/effort is not inferred." }, { "model_id": "mimo-v2-omni", "benchmark_id": "graphwalks_bfs_128k", "score": 68.0, "reference_url": "https://huggingface.co/XiaomiMiMo/MiMo-V2.5/resolve/63651580ca774f8504f676040460aed3e1244ac1/assets/mimo-v2.5-graphwalks.jpeg", "reported_setting": { "mode": "source does not state", "effort": "source does not state", "tools": "none", "sampling": "one response per graph prompt", "judge": "deterministic set-overlap F1 against answer node list", "harness": "Xiaomi MiMo-V2.5 GraphWalks exact-length evaluation", "prompt_style": "3-shot GraphWalks prompt", "temperature": "source does not state", "context": "128K", "input_modalities": "text", "trials": "source does not state", "context_window_variant": "exact-length line chart" }, "matches_canonical": false, "source_type": "official_model_card", "audit_status": "verified", "notes": "Literal chart label 0.68; normalized to 68.0 percent without changing meaning. Source prints 0.68 on [0,1]; campaign stores the equivalent 68.0%." }, { "model_id": "mimo-v2-omni", "benchmark_id": "graphwalks_bfs_256k", "score": 54.0, "reference_url": "https://huggingface.co/XiaomiMiMo/MiMo-V2.5/resolve/63651580ca774f8504f676040460aed3e1244ac1/assets/mimo-v2.5-graphwalks.jpeg", "reported_setting": { "mode": "source does not state", "effort": "source does not state", "tools": "none", "sampling": "one response per graph prompt", "judge": "deterministic set-overlap F1 against answer node list", "harness": "Xiaomi MiMo-V2.5 GraphWalks exact-length evaluation", "prompt_style": "3-shot GraphWalks prompt", "temperature": "source does not state", "context": "256K", "input_modalities": "text", "trials": "source does not state", "context_window_variant": "exact-length line chart" }, "matches_canonical": false, "source_type": "official_model_card", "audit_status": "verified", "notes": "Literal chart label 0.54; normalized to 54.0 percent without changing meaning. Source prints 0.54 on [0,1]; campaign stores the equivalent 54.0%." }, { "model_id": "mimo-v2-omni", "benchmark_id": "graphwalks_bfs_32k", "score": 87.0, "reference_url": "https://huggingface.co/XiaomiMiMo/MiMo-V2.5/resolve/63651580ca774f8504f676040460aed3e1244ac1/assets/mimo-v2.5-graphwalks.jpeg", "reported_setting": { "mode": "source does not state", "effort": "source does not state", "tools": "none", "sampling": "one response per graph prompt", "judge": "deterministic set-overlap F1 against answer node list", "harness": "Xiaomi MiMo-V2.5 GraphWalks exact-length evaluation", "prompt_style": "3-shot GraphWalks prompt", "temperature": "source does not state", "context": "32K", "input_modalities": "text", "trials": "source does not state", "context_window_variant": "exact-length line chart" }, "matches_canonical": false, "source_type": "official_model_card", "audit_status": "verified", "notes": "Literal chart label 0.87; normalized to 87.0 percent without changing meaning. Source prints 0.87 on [0,1]; campaign stores the equivalent 87.0%." }, { "model_id": "mimo-v2-omni", "benchmark_id": "graphwalks_bfs_64k", "score": 79.0, "reference_url": "https://huggingface.co/XiaomiMiMo/MiMo-V2.5/resolve/63651580ca774f8504f676040460aed3e1244ac1/assets/mimo-v2.5-graphwalks.jpeg", "reported_setting": { "mode": "source does not state", "effort": "source does not state", "tools": "none", "sampling": "one response per graph prompt", "judge": "deterministic set-overlap F1 against answer node list", "harness": "Xiaomi MiMo-V2.5 GraphWalks exact-length evaluation", "prompt_style": "3-shot GraphWalks prompt", "temperature": "source does not state", "context": "64K", "input_modalities": "text", "trials": "source does not state", "context_window_variant": "exact-length line chart" }, "matches_canonical": false, "source_type": "official_model_card", "audit_status": "verified", "notes": "Literal chart label 0.79; normalized to 79.0 percent without changing meaning. Source prints 0.79 on [0,1]; campaign stores the equivalent 79.0%." }, { "model_id": "mimo-v2-omni", "benchmark_id": "graphwalks_parents_128k", "score": 56.99999999999999, "reference_url": "https://huggingface.co/XiaomiMiMo/MiMo-V2.5/resolve/63651580ca774f8504f676040460aed3e1244ac1/assets/mimo-v2.5-graphwalks.jpeg", "reported_setting": { "mode": "source does not state", "effort": "source does not state", "tools": "none", "sampling": "one response per graph prompt", "judge": "deterministic set-overlap F1 against answer node list", "harness": "Xiaomi MiMo-V2.5 GraphWalks exact-length evaluation", "prompt_style": "3-shot GraphWalks prompt", "temperature": "source does not state", "context": "128K", "input_modalities": "text", "trials": "source does not state", "context_window_variant": "exact-length line chart" }, "matches_canonical": false, "source_type": "official_model_card", "audit_status": "verified", "notes": "Literal chart label 0.57; normalized to 57.0 percent without changing meaning. Source prints 0.57 on [0,1]; campaign stores the equivalent 57.0%." }, { "model_id": "mimo-v2-omni", "benchmark_id": "graphwalks_parents_256k", "score": 17.0, "reference_url": "https://huggingface.co/XiaomiMiMo/MiMo-V2.5/resolve/63651580ca774f8504f676040460aed3e1244ac1/assets/mimo-v2.5-graphwalks.jpeg", "reported_setting": { "mode": "source does not state", "effort": "source does not state", "tools": "none", "sampling": "one response per graph prompt", "judge": "deterministic set-overlap F1 against answer node list", "harness": "Xiaomi MiMo-V2.5 GraphWalks exact-length evaluation", "prompt_style": "3-shot GraphWalks prompt", "temperature": "source does not state", "context": "256K", "input_modalities": "text", "trials": "source does not state", "context_window_variant": "exact-length line chart" }, "matches_canonical": false, "source_type": "official_model_card", "audit_status": "verified", "notes": "Literal chart label 0.17; normalized to 17.0 percent without changing meaning. Source prints 0.17 on [0,1]; campaign stores the equivalent 17.0%." }, { "model_id": "mimo-v2-omni", "benchmark_id": "graphwalks_parents_32k", "score": 85.0, "reference_url": "https://huggingface.co/XiaomiMiMo/MiMo-V2.5/resolve/63651580ca774f8504f676040460aed3e1244ac1/assets/mimo-v2.5-graphwalks.jpeg", "reported_setting": { "mode": "source does not state", "effort": "source does not state", "tools": "none", "sampling": "one response per graph prompt", "judge": "deterministic set-overlap F1 against answer node list", "harness": "Xiaomi MiMo-V2.5 GraphWalks exact-length evaluation", "prompt_style": "3-shot GraphWalks prompt", "temperature": "source does not state", "context": "32K", "input_modalities": "text", "trials": "source does not state", "context_window_variant": "exact-length line chart" }, "matches_canonical": false, "source_type": "official_model_card", "audit_status": "verified", "notes": "Literal chart label 0.85; normalized to 85.0 percent without changing meaning. Source prints 0.85 on [0,1]; campaign stores the equivalent 85.0%." }, { "model_id": "mimo-v2-omni", "benchmark_id": "graphwalks_parents_64k", "score": 70.0, "reference_url": "https://huggingface.co/XiaomiMiMo/MiMo-V2.5/resolve/63651580ca774f8504f676040460aed3e1244ac1/assets/mimo-v2.5-graphwalks.jpeg", "reported_setting": { "mode": "source does not state", "effort": "source does not state", "tools": "none", "sampling": "one response per graph prompt", "judge": "deterministic set-overlap F1 against answer node list", "harness": "Xiaomi MiMo-V2.5 GraphWalks exact-length evaluation", "prompt_style": "3-shot GraphWalks prompt", "temperature": "source does not state", "context": "64K", "input_modalities": "text", "trials": "source does not state", "context_window_variant": "exact-length line chart" }, "matches_canonical": false, "source_type": "official_model_card", "audit_status": "verified", "notes": "Literal chart label 0.70; normalized to 70.0 percent without changing meaning. Source prints 0.70 on [0,1]; campaign stores the equivalent 70.0%." }, { "model_id": "mimo-v2-pro", "benchmark_id": "claw_eval_pass3", "score": 57.8, "reference_url": "https://raw.githubusercontent.com/claw-eval/claw-eval.github.io/19d378a646d892ca6c8148691aea02da1b184eb8/assets/benchmark-0fIVfCD7.js", "reported_setting": { "mode": "source does not state", "effort": "source does not state", "tools": "official Claw-Eval agent environment", "sampling": "Pass^3; all three trajectories must pass", "judge": "full-trajectory benchmark grading", "harness": "official Claw-Eval v1.1 leaderboard", "prompt_style": "official Claw-Eval task prompts", "temperature": "source does not state", "context": "source does not state", "input_modalities": "text", "trials": "3 per task; 597 generations", "task_subset": "non-multimodal 199-task subset", "task_count": 199 }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Exact official leaderboard field for the non-multimodal 199-task subset. Model mode/effort is not inferred.", "candidates": [ { "score": 57.1, "reference_url": "https://raw.githubusercontent.com/claw-eval/claw-eval.github.io/19d378a646d892ca6c8148691aea02da1b184eb8/assets/benchmark-0fIVfCD7.js", "source_type": "official_blog", "reported_setting": { "mode": "source does not state", "effort": "source does not state", "tools": "official Claw-Eval agent environment", "sampling": "Pass^3; all three trajectories must pass", "judge": "full-trajectory benchmark grading", "harness": "official Claw-Eval v1.1 leaderboard", "prompt_style": "official Claw-Eval task prompts", "temperature": "source does not state", "context": "source does not state", "input_modalities": "text", "trials": "3 per task; 483 generations", "task_subset": "core-general 161-task subset", "task_count": 161 }, "notes": "Average tokens per trajectory=101238 is a physical efficiency annotation, not a benchmark score or score setting. Exact official leaderboard field for the core-general 161-task subset. Model mode/effort is not inferred." } ] }, { "model_id": "mimo-v2-pro", "benchmark_id": "graphwalks_bfs_128k", "score": 68.0, "reference_url": "https://huggingface.co/XiaomiMiMo/MiMo-V2.5-Pro/resolve/21d1ecfecd7bd70f31be25ca49d7edd21f003659/assets/post_training_evaluation.jpg", "reported_setting": { "mode": "source does not state", "effort": "source does not state", "tools": "none", "sampling": "one response per graph prompt", "judge": "deterministic set-overlap F1 against answer node list", "harness": "Xiaomi MiMo-V2.5-Pro GraphWalks exact-length evaluation", "prompt_style": "3-shot GraphWalks prompt", "temperature": "source does not state", "context": "128K", "input_modalities": "text", "trials": "source does not state", "context_window_variant": "exact-length line chart" }, "matches_canonical": false, "source_type": "official_model_card", "audit_status": "verified", "notes": "Literal chart label 0.68; normalized to 68.0 percent without changing meaning. Source prints 0.68 on [0,1]; campaign stores the equivalent 68.0%." }, { "model_id": "mimo-v2-pro", "benchmark_id": "graphwalks_bfs_1m", "score": 0.0, "reference_url": "https://huggingface.co/XiaomiMiMo/MiMo-V2.5-Pro/resolve/21d1ecfecd7bd70f31be25ca49d7edd21f003659/assets/post_training_evaluation.jpg", "reported_setting": { "mode": "source does not state", "effort": "source does not state", "tools": "none", "sampling": "one response per graph prompt", "judge": "deterministic set-overlap F1 against answer node list", "harness": "Xiaomi MiMo-V2.5-Pro GraphWalks exact-length evaluation", "prompt_style": "3-shot GraphWalks prompt", "temperature": "source does not state", "context": "1M", "input_modalities": "text", "trials": "source does not state", "context_window_variant": "exact-length line chart" }, "matches_canonical": false, "source_type": "official_model_card", "audit_status": "verified", "notes": "Literal chart label 0.00; normalized to 0.0 percent without changing meaning. Source prints 0.00 on [0,1]; campaign stores the equivalent 0.0%." }, { "model_id": "mimo-v2-pro", "benchmark_id": "graphwalks_bfs_256k", "score": 44.0, "reference_url": "https://huggingface.co/XiaomiMiMo/MiMo-V2.5-Pro/resolve/21d1ecfecd7bd70f31be25ca49d7edd21f003659/assets/post_training_evaluation.jpg", "reported_setting": { "mode": "source does not state", "effort": "source does not state", "tools": "none", "sampling": "one response per graph prompt", "judge": "deterministic set-overlap F1 against answer node list", "harness": "Xiaomi MiMo-V2.5-Pro GraphWalks exact-length evaluation", "prompt_style": "3-shot GraphWalks prompt", "temperature": "source does not state", "context": "256K", "input_modalities": "text", "trials": "source does not state", "context_window_variant": "exact-length line chart" }, "matches_canonical": false, "source_type": "official_model_card", "audit_status": "verified", "notes": "Literal chart label 0.44; normalized to 44.0 percent without changing meaning. Source prints 0.44 on [0,1]; campaign stores the equivalent 44.0%." }, { "model_id": "mimo-v2-pro", "benchmark_id": "graphwalks_bfs_32k", "score": 81.0, "reference_url": "https://huggingface.co/XiaomiMiMo/MiMo-V2.5-Pro/resolve/21d1ecfecd7bd70f31be25ca49d7edd21f003659/assets/post_training_evaluation.jpg", "reported_setting": { "mode": "source does not state", "effort": "source does not state", "tools": "none", "sampling": "one response per graph prompt", "judge": "deterministic set-overlap F1 against answer node list", "harness": "Xiaomi MiMo-V2.5-Pro GraphWalks exact-length evaluation", "prompt_style": "3-shot GraphWalks prompt", "temperature": "source does not state", "context": "32K", "input_modalities": "text", "trials": "source does not state", "context_window_variant": "exact-length line chart" }, "matches_canonical": false, "source_type": "official_model_card", "audit_status": "verified", "notes": "Literal chart label 0.81; normalized to 81.0 percent without changing meaning. Source prints 0.81 on [0,1]; campaign stores the equivalent 81.0%." }, { "model_id": "mimo-v2-pro", "benchmark_id": "graphwalks_bfs_512k", "score": 12.0, "reference_url": "https://huggingface.co/XiaomiMiMo/MiMo-V2.5-Pro/resolve/21d1ecfecd7bd70f31be25ca49d7edd21f003659/assets/post_training_evaluation.jpg", "reported_setting": { "mode": "source does not state", "effort": "source does not state", "tools": "none", "sampling": "one response per graph prompt", "judge": "deterministic set-overlap F1 against answer node list", "harness": "Xiaomi MiMo-V2.5-Pro GraphWalks exact-length evaluation", "prompt_style": "3-shot GraphWalks prompt", "temperature": "source does not state", "context": "512K", "input_modalities": "text", "trials": "source does not state", "context_window_variant": "exact-length line chart" }, "matches_canonical": false, "source_type": "official_model_card", "audit_status": "verified", "notes": "Literal chart label 0.12; normalized to 12.0 percent without changing meaning. Source prints 0.12 on [0,1]; campaign stores the equivalent 12.0%." }, { "model_id": "mimo-v2-pro", "benchmark_id": "graphwalks_bfs_64k", "score": 75.0, "reference_url": "https://huggingface.co/XiaomiMiMo/MiMo-V2.5-Pro/resolve/21d1ecfecd7bd70f31be25ca49d7edd21f003659/assets/post_training_evaluation.jpg", "reported_setting": { "mode": "source does not state", "effort": "source does not state", "tools": "none", "sampling": "one response per graph prompt", "judge": "deterministic set-overlap F1 against answer node list", "harness": "Xiaomi MiMo-V2.5-Pro GraphWalks exact-length evaluation", "prompt_style": "3-shot GraphWalks prompt", "temperature": "source does not state", "context": "64K", "input_modalities": "text", "trials": "source does not state", "context_window_variant": "exact-length line chart" }, "matches_canonical": false, "source_type": "official_model_card", "audit_status": "verified", "notes": "Literal chart label 0.75; normalized to 75.0 percent without changing meaning. Source prints 0.75 on [0,1]; campaign stores the equivalent 75.0%." }, { "model_id": "mimo-v2-pro", "benchmark_id": "graphwalks_bfs_bucket_0_4k", "score": 100, "reference_url": "https://mimo.xiaomi.com/mimo-v2-5-pro", "reported_setting": { "mode": "source does not state", "effort": "source does not state", "tools": "none", "sampling": "one response per graph prompt", "judge": "deterministic set-overlap F1 against answer node list", "harness": "Xiaomi MiMo GraphWalks evaluation with Anthropic fixes", "prompt_style": "3-shot GraphWalks prompt", "temperature": "source does not state", "context": "0–4K", "input_modalities": "text", "trials": "n=50", "context_window_variant": "standard" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Exact heatmap bucket. Counts and the 1M variant are preserved; no aggregation into campaign range rows." }, { "model_id": "mimo-v2-pro", "benchmark_id": "graphwalks_bfs_bucket_128k_256k", "score": 46, "reference_url": "https://mimo.xiaomi.com/mimo-v2-5-pro", "reported_setting": { "mode": "source does not state", "effort": "source does not state", "tools": "none", "sampling": "one response per graph prompt", "judge": "deterministic set-overlap F1 against answer node list", "harness": "Xiaomi MiMo GraphWalks evaluation with Anthropic fixes", "prompt_style": "3-shot GraphWalks prompt", "temperature": "source does not state", "context": "128K–256K", "input_modalities": "text", "trials": "n=50", "context_window_variant": "standard" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Exact heatmap bucket. Counts and the 1M variant are preserved; no aggregation into campaign range rows." }, { "model_id": "mimo-v2-pro", "benchmark_id": "graphwalks_bfs_bucket_16k_32k", "score": 66, "reference_url": "https://mimo.xiaomi.com/mimo-v2-5-pro", "reported_setting": { "mode": "source does not state", "effort": "source does not state", "tools": "none", "sampling": "one response per graph prompt", "judge": "deterministic set-overlap F1 against answer node list", "harness": "Xiaomi MiMo GraphWalks evaluation with Anthropic fixes", "prompt_style": "3-shot GraphWalks prompt", "temperature": "source does not state", "context": "16K–32K", "input_modalities": "text", "trials": "n=50", "context_window_variant": "standard" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Exact heatmap bucket. Counts and the 1M variant are preserved; no aggregation into campaign range rows." }, { "model_id": "mimo-v2-pro", "benchmark_id": "graphwalks_bfs_bucket_32k_64k", "score": 62, "reference_url": "https://mimo.xiaomi.com/mimo-v2-5-pro", "reported_setting": { "mode": "source does not state", "effort": "source does not state", "tools": "none", "sampling": "one response per graph prompt", "judge": "deterministic set-overlap F1 against answer node list", "harness": "Xiaomi MiMo GraphWalks evaluation with Anthropic fixes", "prompt_style": "3-shot GraphWalks prompt", "temperature": "source does not state", "context": "32K–64K", "input_modalities": "text", "trials": "n=50", "context_window_variant": "standard" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Exact heatmap bucket. Counts and the 1M variant are preserved; no aggregation into campaign range rows." }, { "model_id": "mimo-v2-pro", "benchmark_id": "graphwalks_bfs_bucket_4k_8k", "score": 98, "reference_url": "https://mimo.xiaomi.com/mimo-v2-5-pro", "reported_setting": { "mode": "source does not state", "effort": "source does not state", "tools": "none", "sampling": "one response per graph prompt", "judge": "deterministic set-overlap F1 against answer node list", "harness": "Xiaomi MiMo GraphWalks evaluation with Anthropic fixes", "prompt_style": "3-shot GraphWalks prompt", "temperature": "source does not state", "context": "4K–8K", "input_modalities": "text", "trials": "n=50", "context_window_variant": "standard" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Exact heatmap bucket. Counts and the 1M variant are preserved; no aggregation into campaign range rows." }, { "model_id": "mimo-v2-pro", "benchmark_id": "graphwalks_bfs_bucket_64k_128k", "score": 56, "reference_url": "https://mimo.xiaomi.com/mimo-v2-5-pro", "reported_setting": { "mode": "source does not state", "effort": "source does not state", "tools": "none", "sampling": "one response per graph prompt", "judge": "deterministic set-overlap F1 against answer node list", "harness": "Xiaomi MiMo GraphWalks evaluation with Anthropic fixes", "prompt_style": "3-shot GraphWalks prompt", "temperature": "source does not state", "context": "64K–128K", "input_modalities": "text", "trials": "n=50", "context_window_variant": "standard" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Exact heatmap bucket. Counts and the 1M variant are preserved; no aggregation into campaign range rows." }, { "model_id": "mimo-v2-pro", "benchmark_id": "graphwalks_bfs_bucket_8k_16k", "score": 76, "reference_url": "https://mimo.xiaomi.com/mimo-v2-5-pro", "reported_setting": { "mode": "source does not state", "effort": "source does not state", "tools": "none", "sampling": "one response per graph prompt", "judge": "deterministic set-overlap F1 against answer node list", "harness": "Xiaomi MiMo GraphWalks evaluation with Anthropic fixes", "prompt_style": "3-shot GraphWalks prompt", "temperature": "source does not state", "context": "8K–16K", "input_modalities": "text", "trials": "n=50", "context_window_variant": "standard" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Exact heatmap bucket. Counts and the 1M variant are preserved; no aggregation into campaign range rows." }, { "model_id": "mimo-v2-pro", "benchmark_id": "graphwalks_parents_128k", "score": 44.0, "reference_url": "https://huggingface.co/XiaomiMiMo/MiMo-V2.5-Pro/resolve/21d1ecfecd7bd70f31be25ca49d7edd21f003659/assets/post_training_evaluation.jpg", "reported_setting": { "mode": "source does not state", "effort": "source does not state", "tools": "none", "sampling": "one response per graph prompt", "judge": "deterministic set-overlap F1 against answer node list", "harness": "Xiaomi MiMo-V2.5-Pro GraphWalks exact-length evaluation", "prompt_style": "3-shot GraphWalks prompt", "temperature": "source does not state", "context": "128K", "input_modalities": "text", "trials": "source does not state", "context_window_variant": "exact-length line chart" }, "matches_canonical": false, "source_type": "official_model_card", "audit_status": "verified", "notes": "Literal chart label 0.44; normalized to 44.0 percent without changing meaning. Source prints 0.44 on [0,1]; campaign stores the equivalent 44.0%." }, { "model_id": "mimo-v2-pro", "benchmark_id": "graphwalks_parents_1m", "score": 0.0, "reference_url": "https://huggingface.co/XiaomiMiMo/MiMo-V2.5-Pro/resolve/21d1ecfecd7bd70f31be25ca49d7edd21f003659/assets/post_training_evaluation.jpg", "reported_setting": { "mode": "source does not state", "effort": "source does not state", "tools": "none", "sampling": "one response per graph prompt", "judge": "deterministic set-overlap F1 against answer node list", "harness": "Xiaomi MiMo-V2.5-Pro GraphWalks exact-length evaluation", "prompt_style": "3-shot GraphWalks prompt", "temperature": "source does not state", "context": "1M", "input_modalities": "text", "trials": "source does not state", "context_window_variant": "exact-length line chart" }, "matches_canonical": false, "source_type": "official_model_card", "audit_status": "verified", "notes": "Literal chart label 0.00; normalized to 0.0 percent without changing meaning. Source prints 0.00 on [0,1]; campaign stores the equivalent 0.0%." }, { "model_id": "mimo-v2-pro", "benchmark_id": "graphwalks_parents_256k", "score": 35.0, "reference_url": "https://huggingface.co/XiaomiMiMo/MiMo-V2.5-Pro/resolve/21d1ecfecd7bd70f31be25ca49d7edd21f003659/assets/post_training_evaluation.jpg", "reported_setting": { "mode": "source does not state", "effort": "source does not state", "tools": "none", "sampling": "one response per graph prompt", "judge": "deterministic set-overlap F1 against answer node list", "harness": "Xiaomi MiMo-V2.5-Pro GraphWalks exact-length evaluation", "prompt_style": "3-shot GraphWalks prompt", "temperature": "source does not state", "context": "256K", "input_modalities": "text", "trials": "source does not state", "context_window_variant": "exact-length line chart" }, "matches_canonical": false, "source_type": "official_model_card", "audit_status": "verified", "notes": "Literal chart label 0.35; normalized to 35.0 percent without changing meaning. Source prints 0.35 on [0,1]; campaign stores the equivalent 35.0%." }, { "model_id": "mimo-v2-pro", "benchmark_id": "graphwalks_parents_32k", "score": 79.0, "reference_url": "https://huggingface.co/XiaomiMiMo/MiMo-V2.5-Pro/resolve/21d1ecfecd7bd70f31be25ca49d7edd21f003659/assets/post_training_evaluation.jpg", "reported_setting": { "mode": "source does not state", "effort": "source does not state", "tools": "none", "sampling": "one response per graph prompt", "judge": "deterministic set-overlap F1 against answer node list", "harness": "Xiaomi MiMo-V2.5-Pro GraphWalks exact-length evaluation", "prompt_style": "3-shot GraphWalks prompt", "temperature": "source does not state", "context": "32K", "input_modalities": "text", "trials": "source does not state", "context_window_variant": "exact-length line chart" }, "matches_canonical": false, "source_type": "official_model_card", "audit_status": "verified", "notes": "Literal chart label 0.79; normalized to 79.0 percent without changing meaning. Source prints 0.79 on [0,1]; campaign stores the equivalent 79.0%." }, { "model_id": "mimo-v2-pro", "benchmark_id": "graphwalks_parents_512k", "score": 14.000000000000002, "reference_url": "https://huggingface.co/XiaomiMiMo/MiMo-V2.5-Pro/resolve/21d1ecfecd7bd70f31be25ca49d7edd21f003659/assets/post_training_evaluation.jpg", "reported_setting": { "mode": "source does not state", "effort": "source does not state", "tools": "none", "sampling": "one response per graph prompt", "judge": "deterministic set-overlap F1 against answer node list", "harness": "Xiaomi MiMo-V2.5-Pro GraphWalks exact-length evaluation", "prompt_style": "3-shot GraphWalks prompt", "temperature": "source does not state", "context": "512K", "input_modalities": "text", "trials": "source does not state", "context_window_variant": "exact-length line chart" }, "matches_canonical": false, "source_type": "official_model_card", "audit_status": "verified", "notes": "Literal chart label 0.14; normalized to 14.0 percent without changing meaning. Source prints 0.14 on [0,1]; campaign stores the equivalent 14.0%." }, { "model_id": "mimo-v2-pro", "benchmark_id": "graphwalks_parents_64k", "score": 63.0, "reference_url": "https://huggingface.co/XiaomiMiMo/MiMo-V2.5-Pro/resolve/21d1ecfecd7bd70f31be25ca49d7edd21f003659/assets/post_training_evaluation.jpg", "reported_setting": { "mode": "source does not state", "effort": "source does not state", "tools": "none", "sampling": "one response per graph prompt", "judge": "deterministic set-overlap F1 against answer node list", "harness": "Xiaomi MiMo-V2.5-Pro GraphWalks exact-length evaluation", "prompt_style": "3-shot GraphWalks prompt", "temperature": "source does not state", "context": "64K", "input_modalities": "text", "trials": "source does not state", "context_window_variant": "exact-length line chart" }, "matches_canonical": false, "source_type": "official_model_card", "audit_status": "verified", "notes": "Literal chart label 0.63; normalized to 63.0 percent without changing meaning. Source prints 0.63 on [0,1]; campaign stores the equivalent 63.0%." }, { "model_id": "mimo-v2-pro", "benchmark_id": "graphwalks_parents_bucket_0_4k", "score": 100, "reference_url": "https://mimo.xiaomi.com/mimo-v2-5-pro", "reported_setting": { "mode": "source does not state", "effort": "source does not state", "tools": "none", "sampling": "one response per graph prompt", "judge": "deterministic set-overlap F1 against answer node list", "harness": "Xiaomi MiMo GraphWalks evaluation with Anthropic fixes", "prompt_style": "3-shot GraphWalks prompt", "temperature": "source does not state", "context": "0–4K", "input_modalities": "text", "trials": "n=100", "context_window_variant": "standard" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Exact heatmap bucket. Counts and the 1M variant are preserved; no aggregation into campaign range rows." }, { "model_id": "mimo-v2-pro", "benchmark_id": "graphwalks_parents_bucket_128k_256k", "score": 14, "reference_url": "https://mimo.xiaomi.com/mimo-v2-5-pro", "reported_setting": { "mode": "source does not state", "effort": "source does not state", "tools": "none", "sampling": "one response per graph prompt", "judge": "deterministic set-overlap F1 against answer node list", "harness": "Xiaomi MiMo GraphWalks evaluation with Anthropic fixes", "prompt_style": "3-shot GraphWalks prompt", "temperature": "source does not state", "context": "128K–256K", "input_modalities": "text", "trials": "n=50", "context_window_variant": "standard" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Exact heatmap bucket. Counts and the 1M variant are preserved; no aggregation into campaign range rows." }, { "model_id": "mimo-v2-pro", "benchmark_id": "graphwalks_parents_bucket_16k_32k", "score": 80, "reference_url": "https://mimo.xiaomi.com/mimo-v2-5-pro", "reported_setting": { "mode": "source does not state", "effort": "source does not state", "tools": "none", "sampling": "one response per graph prompt", "judge": "deterministic set-overlap F1 against answer node list", "harness": "Xiaomi MiMo GraphWalks evaluation with Anthropic fixes", "prompt_style": "3-shot GraphWalks prompt", "temperature": "source does not state", "context": "16K–32K", "input_modalities": "text", "trials": "n=50", "context_window_variant": "standard" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Exact heatmap bucket. Counts and the 1M variant are preserved; no aggregation into campaign range rows." }, { "model_id": "mimo-v2-pro", "benchmark_id": "graphwalks_parents_bucket_32k_64k", "score": 56, "reference_url": "https://mimo.xiaomi.com/mimo-v2-5-pro", "reported_setting": { "mode": "source does not state", "effort": "source does not state", "tools": "none", "sampling": "one response per graph prompt", "judge": "deterministic set-overlap F1 against answer node list", "harness": "Xiaomi MiMo GraphWalks evaluation with Anthropic fixes", "prompt_style": "3-shot GraphWalks prompt", "temperature": "source does not state", "context": "32K–64K", "input_modalities": "text", "trials": "n=50", "context_window_variant": "standard" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Exact heatmap bucket. Counts and the 1M variant are preserved; no aggregation into campaign range rows." }, { "model_id": "mimo-v2-pro", "benchmark_id": "graphwalks_parents_bucket_4k_8k", "score": 100, "reference_url": "https://mimo.xiaomi.com/mimo-v2-5-pro", "reported_setting": { "mode": "source does not state", "effort": "source does not state", "tools": "none", "sampling": "one response per graph prompt", "judge": "deterministic set-overlap F1 against answer node list", "harness": "Xiaomi MiMo GraphWalks evaluation with Anthropic fixes", "prompt_style": "3-shot GraphWalks prompt", "temperature": "source does not state", "context": "4K–8K", "input_modalities": "text", "trials": "n=100", "context_window_variant": "standard" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Exact heatmap bucket. Counts and the 1M variant are preserved; no aggregation into campaign range rows." }, { "model_id": "mimo-v2-pro", "benchmark_id": "graphwalks_parents_bucket_64k_128k", "score": 47, "reference_url": "https://mimo.xiaomi.com/mimo-v2-5-pro", "reported_setting": { "mode": "source does not state", "effort": "source does not state", "tools": "none", "sampling": "one response per graph prompt", "judge": "deterministic set-overlap F1 against answer node list", "harness": "Xiaomi MiMo GraphWalks evaluation with Anthropic fixes", "prompt_style": "3-shot GraphWalks prompt", "temperature": "source does not state", "context": "64K–128K", "input_modalities": "text", "trials": "n=50", "context_window_variant": "standard" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Exact heatmap bucket. Counts and the 1M variant are preserved; no aggregation into campaign range rows." }, { "model_id": "mimo-v2-pro", "benchmark_id": "graphwalks_parents_bucket_8k_16k", "score": 98, "reference_url": "https://mimo.xiaomi.com/mimo-v2-5-pro", "reported_setting": { "mode": "source does not state", "effort": "source does not state", "tools": "none", "sampling": "one response per graph prompt", "judge": "deterministic set-overlap F1 against answer node list", "harness": "Xiaomi MiMo GraphWalks evaluation with Anthropic fixes", "prompt_style": "3-shot GraphWalks prompt", "temperature": "source does not state", "context": "8K–16K", "input_modalities": "text", "trials": "n=100", "context_window_variant": "standard" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Exact heatmap bucket. Counts and the 1M variant are preserved; no aggregation into campaign range rows." }, { "model_id": "mimo-v2.5", "benchmark_id": "arc_challenge", "score": 96.5, "reference_url": "https://huggingface.co/XiaomiMiMo/MiMo-V2.5-Pro/resolve/21d1ecfecd7bd70f31be25ca49d7edd21f003659/README.md", "reported_setting": { "mode": "base pretrained checkpoint", "effort": "not applicable", "tools": "none", "sampling": "25-shot; single completion per prompt", "judge": "benchmark-native rule-based evaluator", "harness": "Xiaomi MiMo base-checkpoint evaluation", "prompt_style": "25-shot few-shot continuation", "temperature": "source does not state", "context": "262,144 (256K)", "input_modalities": "text", "trials": "one completion per item", "checkpoint": "base" }, "matches_canonical": false, "source_type": "official_model_card_base_checkpoint", "audit_status": "verified", "notes": "Base checkpoint observation attached to the canonical post-trained model ID. Base result is attached to the canonical post-trained model ID. The locked MiMo-V2.5-Base config exposes max_position_embeddings=262,144, consistent with the 256K download/table description; there is no context conflict." }, { "model_id": "mimo-v2.5", "benchmark_id": "bigbench_hard", "score": 87.2, "reference_url": "https://huggingface.co/XiaomiMiMo/MiMo-V2.5-Pro/resolve/21d1ecfecd7bd70f31be25ca49d7edd21f003659/README.md", "reported_setting": { "mode": "base pretrained checkpoint", "effort": "not applicable", "tools": "none", "sampling": "3-shot; single completion per prompt", "judge": "benchmark-native rule-based evaluator", "harness": "Xiaomi MiMo base-checkpoint evaluation", "prompt_style": "3-shot few-shot continuation", "temperature": "source does not state", "context": "262,144 (256K)", "input_modalities": "text", "trials": "one completion per item", "checkpoint": "base" }, "matches_canonical": false, "source_type": "official_model_card_base_checkpoint", "audit_status": "verified", "notes": "Base checkpoint observation attached to the canonical post-trained model ID. Base result is attached to the canonical post-trained model ID. The locked MiMo-V2.5-Base config exposes max_position_embeddings=262,144, consistent with the 256K download/table description; there is no context conflict." }, { "model_id": "mimo-v2.5", "benchmark_id": "c_eval", "score": 88.6, "reference_url": "https://huggingface.co/XiaomiMiMo/MiMo-V2.5-Pro/resolve/21d1ecfecd7bd70f31be25ca49d7edd21f003659/README.md", "reported_setting": { "mode": "base pretrained checkpoint", "effort": "not applicable", "tools": "none", "sampling": "5-shot; single completion per prompt", "judge": "benchmark-native rule-based evaluator", "harness": "Xiaomi MiMo base-checkpoint evaluation", "prompt_style": "5-shot few-shot continuation", "temperature": "source does not state", "context": "262,144 (256K)", "input_modalities": "text", "trials": "one completion per item", "checkpoint": "base" }, "matches_canonical": false, "source_type": "official_model_card_base_checkpoint", "audit_status": "verified", "notes": "Base checkpoint observation attached to the canonical post-trained model ID. Base result is attached to the canonical post-trained model ID. The locked MiMo-V2.5-Base config exposes max_position_embeddings=262,144, consistent with the 256K download/table description; there is no context conflict." }, { "model_id": "mimo-v2.5", "benchmark_id": "charxiv_reasoning", "score": 81.0, "reference_url": "https://huggingface.co/XiaomiMiMo/MiMo-V2.5/resolve/63651580ca774f8504f676040460aed3e1244ac1/assets/mimo-v2.5-multimodal-bench.png", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1; source does not state repeated-trial count", "judge": "benchmark-specified", "harness": "official Xiaomi MiMo family benchmark evaluation", "prompt_style": "Xiaomi MiMo official evaluation prompt", "temperature": "1.0; top_p=0.95", "context": "1,048,576", "input_modalities": "image and text", "trials": "source does not state" }, "matches_canonical": false, "source_type": "official_model_card", "audit_status": "verified", "notes": "Locked Xiaomi observation for CharXiv RQ." }, { "model_id": "mimo-v2.5", "benchmark_id": "claw_eval_multimodal_pass3", "score": 23.8, "reference_url": "https://raw.githubusercontent.com/claw-eval/claw-eval.github.io/19d378a646d892ca6c8148691aea02da1b184eb8/assets/benchmark-0fIVfCD7.js", "reported_setting": { "mode": "source does not state", "effort": "source does not state", "tools": "official Claw-Eval agent environment", "sampling": "Pass^3; all three trajectories must pass", "judge": "full-trajectory benchmark grading", "harness": "official Claw-Eval v1.1 leaderboard", "prompt_style": "official Claw-Eval task prompts", "temperature": "source does not state", "context": "source does not state", "input_modalities": "image and text", "trials": "3 per task; 303 generations", "task_subset": "multimodal 101-task subset", "task_count": 101 }, "matches_canonical": false, "source_type": "official_model_card", "audit_status": "verified", "notes": "Exact official leaderboard field for the multimodal 101-task subset. Model mode/effort is not inferred." }, { "model_id": "mimo-v2.5", "benchmark_id": "claw_eval_pass3", "score": 62.3, "reference_url": "https://raw.githubusercontent.com/claw-eval/claw-eval.github.io/19d378a646d892ca6c8148691aea02da1b184eb8/assets/benchmark-0fIVfCD7.js", "reported_setting": { "mode": "source does not state", "effort": "source does not state", "tools": "official Claw-Eval agent environment", "sampling": "Pass^3; all three trajectories must pass", "judge": "full-trajectory benchmark grading", "harness": "official Claw-Eval v1.1 leaderboard", "prompt_style": "official Claw-Eval task prompts", "temperature": "source does not state", "context": "source does not state", "input_modalities": "text", "trials": "3 per task; 597 generations", "task_subset": "non-multimodal 199-task subset", "task_count": 199 }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Exact official leaderboard field for the non-multimodal 199-task subset. Model mode/effort is not inferred.", "candidates": [ { "score": 62.1, "reference_url": "https://raw.githubusercontent.com/claw-eval/claw-eval.github.io/19d378a646d892ca6c8148691aea02da1b184eb8/assets/benchmark-0fIVfCD7.js", "source_type": "official_blog", "reported_setting": { "mode": "source does not state", "effort": "source does not state", "tools": "official Claw-Eval agent environment", "sampling": "Pass^3; all three trajectories must pass", "judge": "full-trajectory benchmark grading", "harness": "official Claw-Eval v1.1 leaderboard", "prompt_style": "official Claw-Eval task prompts", "temperature": "source does not state", "context": "source does not state", "input_modalities": "text", "trials": "3 per task; 483 generations", "task_subset": "core-general 161-task subset", "task_count": 161 }, "notes": "Average tokens per trajectory=81927 is a physical efficiency annotation, not a benchmark score or score setting. Exact official leaderboard field for the core-general 161-task subset. Model mode/effort is not inferred." } ] }, { "model_id": "mimo-v2.5", "benchmark_id": "cmmlu", "score": 88.2, "reference_url": "https://huggingface.co/XiaomiMiMo/MiMo-V2.5-Pro/resolve/21d1ecfecd7bd70f31be25ca49d7edd21f003659/README.md", "reported_setting": { "mode": "base pretrained checkpoint", "effort": "not applicable", "tools": "none", "sampling": "5-shot; single completion per prompt", "judge": "benchmark-native rule-based evaluator", "harness": "Xiaomi MiMo base-checkpoint evaluation", "prompt_style": "5-shot few-shot continuation", "temperature": "source does not state", "context": "262,144 (256K)", "input_modalities": "text", "trials": "one completion per item", "checkpoint": "base" }, "matches_canonical": false, "source_type": "official_model_card_base_checkpoint", "audit_status": "verified", "notes": "Base checkpoint observation attached to the canonical post-trained model ID. Base result is attached to the canonical post-trained model ID. The locked MiMo-V2.5-Base config exposes max_position_embeddings=262,144, consistent with the 256K download/table description; there is no context conflict." }, { "model_id": "mimo-v2.5", "benchmark_id": "drop", "score": 83.7, "reference_url": "https://huggingface.co/XiaomiMiMo/MiMo-V2.5-Pro/resolve/21d1ecfecd7bd70f31be25ca49d7edd21f003659/README.md", "reported_setting": { "mode": "base pretrained checkpoint", "effort": "not applicable", "tools": "none", "sampling": "3-shot; single completion per prompt", "judge": "benchmark-native rule-based evaluator", "harness": "Xiaomi MiMo base-checkpoint evaluation", "prompt_style": "3-shot few-shot continuation", "temperature": "source does not state", "context": "262,144 (256K)", "input_modalities": "text", "trials": "one completion per item", "checkpoint": "base" }, "matches_canonical": false, "source_type": "official_model_card_base_checkpoint", "audit_status": "verified", "notes": "Base checkpoint observation attached to the canonical post-trained model ID. Base result is attached to the canonical post-trained model ID. The locked MiMo-V2.5-Base config exposes max_position_embeddings=262,144, consistent with the 256K download/table description; there is no context conflict." }, { "model_id": "mimo-v2.5", "benchmark_id": "gpqa_diamond", "score": 58.1, "reference_url": "https://huggingface.co/XiaomiMiMo/MiMo-V2.5-Pro/resolve/21d1ecfecd7bd70f31be25ca49d7edd21f003659/README.md", "reported_setting": { "mode": "base pretrained checkpoint", "effort": "not applicable", "tools": "none", "sampling": "5-shot; single completion per prompt", "judge": "benchmark-native rule-based evaluator", "harness": "Xiaomi MiMo base-checkpoint evaluation", "prompt_style": "5-shot few-shot continuation", "temperature": "source does not state", "context": "262,144 (256K)", "input_modalities": "text", "trials": "one completion per item", "checkpoint": "base" }, "matches_canonical": false, "source_type": "official_model_card_base_checkpoint", "audit_status": "verified", "notes": "Base checkpoint observation attached to the canonical post-trained model ID. Base result is attached to the canonical post-trained model ID. The locked MiMo-V2.5-Base config exposes max_position_embeddings=262,144, consistent with the 256K download/table description; there is no context conflict." }, { "model_id": "mimo-v2.5", "benchmark_id": "graphwalks_bfs_128k", "score": 75.0, "reference_url": "https://huggingface.co/XiaomiMiMo/MiMo-V2.5/resolve/63651580ca774f8504f676040460aed3e1244ac1/assets/mimo-v2.5-graphwalks.jpeg", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "one response per graph prompt", "judge": "deterministic set-overlap F1 against answer node list", "harness": "Xiaomi MiMo-V2.5 GraphWalks exact-length evaluation", "prompt_style": "3-shot GraphWalks prompt", "temperature": "1.0; top_p=0.95", "context": "128K", "input_modalities": "text", "trials": "source does not state", "context_window_variant": "exact-length line chart" }, "matches_canonical": false, "source_type": "official_model_card", "audit_status": "verified", "notes": "Literal chart label 0.75; normalized to 75.0 percent without changing meaning. Source prints 0.75 on [0,1]; campaign stores the equivalent 75.0%." }, { "model_id": "mimo-v2.5", "benchmark_id": "graphwalks_bfs_1m", "score": 54.0, "reference_url": "https://huggingface.co/XiaomiMiMo/MiMo-V2.5/resolve/63651580ca774f8504f676040460aed3e1244ac1/assets/mimo-v2.5-graphwalks.jpeg", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "one response per graph prompt", "judge": "deterministic set-overlap F1 against answer node list", "harness": "Xiaomi MiMo-V2.5 GraphWalks exact-length evaluation", "prompt_style": "3-shot GraphWalks prompt", "temperature": "1.0; top_p=0.95", "context": "1M", "input_modalities": "text", "trials": "source does not state", "context_window_variant": "exact-length line chart" }, "matches_canonical": false, "source_type": "official_model_card", "audit_status": "verified", "notes": "Literal chart label 0.54; normalized to 54.0 percent without changing meaning. Source prints 0.54 on [0,1]; campaign stores the equivalent 54.0%." }, { "model_id": "mimo-v2.5", "benchmark_id": "graphwalks_bfs_256k", "score": 78.0, "reference_url": "https://huggingface.co/XiaomiMiMo/MiMo-V2.5/resolve/63651580ca774f8504f676040460aed3e1244ac1/assets/mimo-v2.5-graphwalks.jpeg", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "one response per graph prompt", "judge": "deterministic set-overlap F1 against answer node list", "harness": "Xiaomi MiMo-V2.5 GraphWalks exact-length evaluation", "prompt_style": "3-shot GraphWalks prompt", "temperature": "1.0; top_p=0.95", "context": "256K", "input_modalities": "text", "trials": "source does not state", "context_window_variant": "exact-length line chart" }, "matches_canonical": false, "source_type": "official_model_card", "audit_status": "verified", "notes": "Literal chart label 0.78; normalized to 78.0 percent without changing meaning. Source prints 0.78 on [0,1]; campaign stores the equivalent 78.0%." }, { "model_id": "mimo-v2.5", "benchmark_id": "graphwalks_bfs_32k", "score": 86.0, "reference_url": "https://huggingface.co/XiaomiMiMo/MiMo-V2.5/resolve/63651580ca774f8504f676040460aed3e1244ac1/assets/mimo-v2.5-graphwalks.jpeg", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "one response per graph prompt", "judge": "deterministic set-overlap F1 against answer node list", "harness": "Xiaomi MiMo-V2.5 GraphWalks exact-length evaluation", "prompt_style": "3-shot GraphWalks prompt", "temperature": "1.0; top_p=0.95", "context": "32K", "input_modalities": "text", "trials": "source does not state", "context_window_variant": "exact-length line chart" }, "matches_canonical": false, "source_type": "official_model_card", "audit_status": "verified", "notes": "Literal chart label 0.86; normalized to 86.0 percent without changing meaning. Source prints 0.86 on [0,1]; campaign stores the equivalent 86.0%." }, { "model_id": "mimo-v2.5", "benchmark_id": "graphwalks_bfs_512k", "score": 57.99999999999999, "reference_url": "https://huggingface.co/XiaomiMiMo/MiMo-V2.5/resolve/63651580ca774f8504f676040460aed3e1244ac1/assets/mimo-v2.5-graphwalks.jpeg", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "one response per graph prompt", "judge": "deterministic set-overlap F1 against answer node list", "harness": "Xiaomi MiMo-V2.5 GraphWalks exact-length evaluation", "prompt_style": "3-shot GraphWalks prompt", "temperature": "1.0; top_p=0.95", "context": "512K", "input_modalities": "text", "trials": "source does not state", "context_window_variant": "exact-length line chart" }, "matches_canonical": false, "source_type": "official_model_card", "audit_status": "verified", "notes": "Literal chart label 0.58; normalized to 58.0 percent without changing meaning. Source prints 0.58 on [0,1]; campaign stores the equivalent 58.0%." }, { "model_id": "mimo-v2.5", "benchmark_id": "graphwalks_bfs_64k", "score": 85.0, "reference_url": "https://huggingface.co/XiaomiMiMo/MiMo-V2.5/resolve/63651580ca774f8504f676040460aed3e1244ac1/assets/mimo-v2.5-graphwalks.jpeg", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "one response per graph prompt", "judge": "deterministic set-overlap F1 against answer node list", "harness": "Xiaomi MiMo-V2.5 GraphWalks exact-length evaluation", "prompt_style": "3-shot GraphWalks prompt", "temperature": "1.0; top_p=0.95", "context": "64K", "input_modalities": "text", "trials": "source does not state", "context_window_variant": "exact-length line chart" }, "matches_canonical": false, "source_type": "official_model_card", "audit_status": "verified", "notes": "Literal chart label 0.85; normalized to 85.0 percent without changing meaning. Source prints 0.85 on [0,1]; campaign stores the equivalent 85.0%." }, { "model_id": "mimo-v2.5", "benchmark_id": "graphwalks_parents_128k", "score": 100.0, "reference_url": "https://huggingface.co/XiaomiMiMo/MiMo-V2.5/resolve/63651580ca774f8504f676040460aed3e1244ac1/assets/mimo-v2.5-graphwalks.jpeg", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "one response per graph prompt", "judge": "deterministic set-overlap F1 against answer node list", "harness": "Xiaomi MiMo-V2.5 GraphWalks exact-length evaluation", "prompt_style": "3-shot GraphWalks prompt", "temperature": "1.0; top_p=0.95", "context": "128K", "input_modalities": "text", "trials": "source does not state", "context_window_variant": "exact-length line chart" }, "matches_canonical": false, "source_type": "official_model_card", "audit_status": "verified", "notes": "Literal chart label 1.00; normalized to 100.0 percent without changing meaning. Source prints 1.00 on [0,1]; campaign stores the equivalent 100.0%." }, { "model_id": "mimo-v2.5", "benchmark_id": "graphwalks_parents_1m", "score": 87.0, "reference_url": "https://huggingface.co/XiaomiMiMo/MiMo-V2.5/resolve/63651580ca774f8504f676040460aed3e1244ac1/assets/mimo-v2.5-graphwalks.jpeg", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "one response per graph prompt", "judge": "deterministic set-overlap F1 against answer node list", "harness": "Xiaomi MiMo-V2.5 GraphWalks exact-length evaluation", "prompt_style": "3-shot GraphWalks prompt", "temperature": "1.0; top_p=0.95", "context": "1M", "input_modalities": "text", "trials": "source does not state", "context_window_variant": "exact-length line chart" }, "matches_canonical": false, "source_type": "official_model_card", "audit_status": "verified", "notes": "Literal chart label 0.87; normalized to 87.0 percent without changing meaning. Source prints 0.87 on [0,1]; campaign stores the equivalent 87.0%." }, { "model_id": "mimo-v2.5", "benchmark_id": "graphwalks_parents_256k", "score": 98.0, "reference_url": "https://huggingface.co/XiaomiMiMo/MiMo-V2.5/resolve/63651580ca774f8504f676040460aed3e1244ac1/assets/mimo-v2.5-graphwalks.jpeg", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "one response per graph prompt", "judge": "deterministic set-overlap F1 against answer node list", "harness": "Xiaomi MiMo-V2.5 GraphWalks exact-length evaluation", "prompt_style": "3-shot GraphWalks prompt", "temperature": "1.0; top_p=0.95", "context": "256K", "input_modalities": "text", "trials": "source does not state", "context_window_variant": "exact-length line chart" }, "matches_canonical": false, "source_type": "official_model_card", "audit_status": "verified", "notes": "Literal chart label 0.98; normalized to 98.0 percent without changing meaning. Source prints 0.98 on [0,1]; campaign stores the equivalent 98.0%." }, { "model_id": "mimo-v2.5", "benchmark_id": "graphwalks_parents_32k", "score": 100.0, "reference_url": "https://huggingface.co/XiaomiMiMo/MiMo-V2.5/resolve/63651580ca774f8504f676040460aed3e1244ac1/assets/mimo-v2.5-graphwalks.jpeg", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "one response per graph prompt", "judge": "deterministic set-overlap F1 against answer node list", "harness": "Xiaomi MiMo-V2.5 GraphWalks exact-length evaluation", "prompt_style": "3-shot GraphWalks prompt", "temperature": "1.0; top_p=0.95", "context": "32K", "input_modalities": "text", "trials": "source does not state", "context_window_variant": "exact-length line chart" }, "matches_canonical": false, "source_type": "official_model_card", "audit_status": "verified", "notes": "Literal chart label 1.00; normalized to 100.0 percent without changing meaning. Source prints 1.00 on [0,1]; campaign stores the equivalent 100.0%." }, { "model_id": "mimo-v2.5", "benchmark_id": "graphwalks_parents_512k", "score": 97.0, "reference_url": "https://huggingface.co/XiaomiMiMo/MiMo-V2.5/resolve/63651580ca774f8504f676040460aed3e1244ac1/assets/mimo-v2.5-graphwalks.jpeg", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "one response per graph prompt", "judge": "deterministic set-overlap F1 against answer node list", "harness": "Xiaomi MiMo-V2.5 GraphWalks exact-length evaluation", "prompt_style": "3-shot GraphWalks prompt", "temperature": "1.0; top_p=0.95", "context": "512K", "input_modalities": "text", "trials": "source does not state", "context_window_variant": "exact-length line chart" }, "matches_canonical": false, "source_type": "official_model_card", "audit_status": "verified", "notes": "Literal chart label 0.97; normalized to 97.0 percent without changing meaning. Source prints 0.97 on [0,1]; campaign stores the equivalent 97.0%." }, { "model_id": "mimo-v2.5", "benchmark_id": "graphwalks_parents_64k", "score": 100.0, "reference_url": "https://huggingface.co/XiaomiMiMo/MiMo-V2.5/resolve/63651580ca774f8504f676040460aed3e1244ac1/assets/mimo-v2.5-graphwalks.jpeg", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "one response per graph prompt", "judge": "deterministic set-overlap F1 against answer node list", "harness": "Xiaomi MiMo-V2.5 GraphWalks exact-length evaluation", "prompt_style": "3-shot GraphWalks prompt", "temperature": "1.0; top_p=0.95", "context": "64K", "input_modalities": "text", "trials": "source does not state", "context_window_variant": "exact-length line chart" }, "matches_canonical": false, "source_type": "official_model_card", "audit_status": "verified", "notes": "Literal chart label 1.00; normalized to 100.0 percent without changing meaning. Source prints 1.00 on [0,1]; campaign stores the equivalent 100.0%." }, { "model_id": "mimo-v2.5", "benchmark_id": "gsm8k", "score": 83.3, "reference_url": "https://huggingface.co/XiaomiMiMo/MiMo-V2.5-Pro/resolve/21d1ecfecd7bd70f31be25ca49d7edd21f003659/README.md", "reported_setting": { "mode": "base pretrained checkpoint", "effort": "not applicable", "tools": "none", "sampling": "8-shot; single completion per prompt", "judge": "benchmark-native rule-based evaluator", "harness": "Xiaomi MiMo base-checkpoint evaluation", "prompt_style": "8-shot few-shot continuation", "temperature": "source does not state", "context": "262,144 (256K)", "input_modalities": "text", "trials": "one completion per item", "checkpoint": "base" }, "matches_canonical": false, "source_type": "official_model_card_base_checkpoint", "audit_status": "verified", "notes": "Base checkpoint observation attached to the canonical post-trained model ID. Base result is attached to the canonical post-trained model ID. The locked MiMo-V2.5-Base config exposes max_position_embeddings=262,144, consistent with the 256K download/table description; there is no context conflict." }, { "model_id": "mimo-v2.5", "benchmark_id": "hellaswag", "score": 88.6, "reference_url": "https://huggingface.co/XiaomiMiMo/MiMo-V2.5-Pro/resolve/21d1ecfecd7bd70f31be25ca49d7edd21f003659/README.md", "reported_setting": { "mode": "base pretrained checkpoint", "effort": "not applicable", "tools": "none", "sampling": "10-shot; single completion per prompt", "judge": "benchmark-native rule-based evaluator", "harness": "Xiaomi MiMo base-checkpoint evaluation", "prompt_style": "10-shot few-shot continuation", "temperature": "source does not state", "context": "262,144 (256K)", "input_modalities": "text", "trials": "one completion per item", "checkpoint": "base" }, "matches_canonical": false, "source_type": "official_model_card_base_checkpoint", "audit_status": "verified", "notes": "Base checkpoint observation attached to the canonical post-trained model ID. Base result is attached to the canonical post-trained model ID. The locked MiMo-V2.5-Base config exposes max_position_embeddings=262,144, consistent with the 256K download/table description; there is no context conflict." }, { "model_id": "mimo-v2.5", "benchmark_id": "humaneval_plus", "score": 71.3, "reference_url": "https://huggingface.co/XiaomiMiMo/MiMo-V2.5-Pro/resolve/21d1ecfecd7bd70f31be25ca49d7edd21f003659/README.md", "reported_setting": { "mode": "base pretrained checkpoint", "effort": "not applicable", "tools": "none", "sampling": "1-shot; single completion per prompt", "judge": "benchmark-native rule-based evaluator", "harness": "Xiaomi MiMo base-checkpoint evaluation", "prompt_style": "1-shot few-shot continuation", "temperature": "source does not state", "context": "262,144 (256K)", "input_modalities": "text", "trials": "one completion per item", "checkpoint": "base" }, "matches_canonical": false, "source_type": "official_model_card_base_checkpoint", "audit_status": "verified", "notes": "Base checkpoint observation attached to the canonical post-trained model ID. Base result is attached to the canonical post-trained model ID. The locked MiMo-V2.5-Base config exposes max_position_embeddings=262,144, consistent with the 256K download/table description; there is no context conflict." }, { "model_id": "mimo-v2.5", "benchmark_id": "livecodebench_v6", "score": 35.5, "reference_url": "https://huggingface.co/XiaomiMiMo/MiMo-V2.5-Pro/resolve/21d1ecfecd7bd70f31be25ca49d7edd21f003659/README.md", "reported_setting": { "mode": "base pretrained checkpoint", "effort": "not applicable", "tools": "none", "sampling": "1-shot; single completion per prompt", "judge": "benchmark-native rule-based evaluator", "harness": "Xiaomi MiMo base-checkpoint evaluation", "prompt_style": "1-shot few-shot continuation", "temperature": "source does not state", "context": "262,144 (256K)", "input_modalities": "text", "trials": "one completion per item", "checkpoint": "base" }, "matches_canonical": false, "source_type": "official_model_card_base_checkpoint", "audit_status": "verified", "notes": "Base checkpoint observation attached to the canonical post-trained model ID. Base result is attached to the canonical post-trained model ID. The locked MiMo-V2.5-Base config exposes max_position_embeddings=262,144, consistent with the 256K download/table description; there is no context conflict." }, { "model_id": "mimo-v2.5", "benchmark_id": "math", "score": 67.7, "reference_url": "https://huggingface.co/XiaomiMiMo/MiMo-V2.5-Pro/resolve/21d1ecfecd7bd70f31be25ca49d7edd21f003659/README.md", "reported_setting": { "mode": "base pretrained checkpoint", "effort": "not applicable", "tools": "none", "sampling": "4-shot; single completion per prompt", "judge": "benchmark-native rule-based evaluator", "harness": "Xiaomi MiMo base-checkpoint evaluation", "prompt_style": "4-shot few-shot continuation", "temperature": "source does not state", "context": "262,144 (256K)", "input_modalities": "text", "trials": "one completion per item", "checkpoint": "base" }, "matches_canonical": false, "source_type": "official_model_card_base_checkpoint", "audit_status": "verified", "notes": "Base checkpoint observation attached to the canonical post-trained model ID. Base result is attached to the canonical post-trained model ID. The locked MiMo-V2.5-Base config exposes max_position_embeddings=262,144, consistent with the 256K download/table description; there is no context conflict." }, { "model_id": "mimo-v2.5", "benchmark_id": "mbpp_plus", "score": 70.9, "reference_url": "https://huggingface.co/XiaomiMiMo/MiMo-V2.5-Pro/resolve/21d1ecfecd7bd70f31be25ca49d7edd21f003659/README.md", "reported_setting": { "mode": "base pretrained checkpoint", "effort": "not applicable", "tools": "none", "sampling": "3-shot; single completion per prompt", "judge": "benchmark-native rule-based evaluator", "harness": "Xiaomi MiMo base-checkpoint evaluation", "prompt_style": "3-shot few-shot continuation", "temperature": "source does not state", "context": "262,144 (256K)", "input_modalities": "text", "trials": "one completion per item", "checkpoint": "base" }, "matches_canonical": false, "source_type": "official_model_card_base_checkpoint", "audit_status": "verified", "notes": "Base checkpoint observation attached to the canonical post-trained model ID. Base result is attached to the canonical post-trained model ID. The locked MiMo-V2.5-Base config exposes max_position_embeddings=262,144, consistent with the 256K download/table description; there is no context conflict." }, { "model_id": "mimo-v2.5", "benchmark_id": "mmlu", "score": 86.3, "reference_url": "https://huggingface.co/XiaomiMiMo/MiMo-V2.5-Pro/resolve/21d1ecfecd7bd70f31be25ca49d7edd21f003659/README.md", "reported_setting": { "mode": "base pretrained checkpoint", "effort": "not applicable", "tools": "none", "sampling": "5-shot; single completion per prompt", "judge": "benchmark-native rule-based evaluator", "harness": "Xiaomi MiMo base-checkpoint evaluation", "prompt_style": "5-shot few-shot continuation", "temperature": "source does not state", "context": "262,144 (256K)", "input_modalities": "text", "trials": "one completion per item", "checkpoint": "base" }, "matches_canonical": false, "source_type": "official_model_card_base_checkpoint", "audit_status": "verified", "notes": "Base checkpoint observation attached to the canonical post-trained model ID. Base result is attached to the canonical post-trained model ID. The locked MiMo-V2.5-Base config exposes max_position_embeddings=262,144, consistent with the 256K download/table description; there is no context conflict." }, { "model_id": "mimo-v2.5", "benchmark_id": "mmlu_pro", "score": 65.8, "reference_url": "https://huggingface.co/XiaomiMiMo/MiMo-V2.5-Pro/resolve/21d1ecfecd7bd70f31be25ca49d7edd21f003659/README.md", "reported_setting": { "mode": "base pretrained checkpoint", "effort": "not applicable", "tools": "none", "sampling": "5-shot; single completion per prompt", "judge": "benchmark-native rule-based evaluator", "harness": "Xiaomi MiMo base-checkpoint evaluation", "prompt_style": "5-shot few-shot continuation", "temperature": "source does not state", "context": "262,144 (256K)", "input_modalities": "text", "trials": "one completion per item", "checkpoint": "base" }, "matches_canonical": false, "source_type": "official_model_card_base_checkpoint", "audit_status": "verified", "notes": "Base checkpoint observation attached to the canonical post-trained model ID. Base result is attached to the canonical post-trained model ID. The locked MiMo-V2.5-Base config exposes max_position_embeddings=262,144, consistent with the 256K download/table description; there is no context conflict." }, { "model_id": "mimo-v2.5", "benchmark_id": "mmlu_redux", "score": 89.8, "reference_url": "https://huggingface.co/XiaomiMiMo/MiMo-V2.5-Pro/resolve/21d1ecfecd7bd70f31be25ca49d7edd21f003659/README.md", "reported_setting": { "mode": "base pretrained checkpoint", "effort": "not applicable", "tools": "none", "sampling": "5-shot; single completion per prompt", "judge": "benchmark-native rule-based evaluator", "harness": "Xiaomi MiMo base-checkpoint evaluation", "prompt_style": "5-shot few-shot continuation", "temperature": "source does not state", "context": "262,144 (256K)", "input_modalities": "text", "trials": "one completion per item", "checkpoint": "base" }, "matches_canonical": false, "source_type": "official_model_card_base_checkpoint", "audit_status": "verified", "notes": "Base checkpoint observation attached to the canonical post-trained model ID. Base result is attached to the canonical post-trained model ID. The locked MiMo-V2.5-Base config exposes max_position_embeddings=262,144, consistent with the 256K download/table description; there is no context conflict." }, { "model_id": "mimo-v2.5", "benchmark_id": "mmmu_pro", "score": 77.9, "reference_url": "https://huggingface.co/XiaomiMiMo/MiMo-V2.5/resolve/63651580ca774f8504f676040460aed3e1244ac1/assets/mimo-v2.5-multimodal-bench.png", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1; source does not state repeated-trial count", "judge": "benchmark-specified", "harness": "official Xiaomi MiMo family benchmark evaluation", "prompt_style": "Xiaomi MiMo official evaluation prompt", "temperature": "1.0; top_p=0.95", "context": "1,048,576", "input_modalities": "image and text", "trials": "source does not state" }, "matches_canonical": false, "source_type": "official_model_card", "audit_status": "verified", "notes": "Locked Xiaomi observation for MMMU-Pro." }, { "model_id": "mimo-v2.5", "benchmark_id": "swe_bench_pro", "score": 56.1, "reference_url": "https://mimo.xiaomi.com/mimo-v2-5-pro", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "agentic code/terminal tools", "sampling": "source does not state repeated-trial count", "judge": "benchmark-specified", "harness": "official Xiaomi MiMo family benchmark evaluation", "prompt_style": "Xiaomi MiMo official evaluation prompt", "temperature": "1.0; top_p=0.95", "context": "1,048,576", "input_modalities": "text", "trials": "source does not state" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Locked Xiaomi observation for SWE-Bench Pro." }, { "model_id": "mimo-v2.5", "benchmark_id": "tau3_bench", "score": 69.5, "reference_url": "https://mimo.xiaomi.com/mimo-v2-5-pro", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1; source does not state repeated-trial count", "judge": "benchmark-specified", "harness": "official Xiaomi MiMo family benchmark evaluation", "prompt_style": "Xiaomi MiMo official evaluation prompt", "temperature": "1.0; top_p=0.95", "context": "1,048,576", "input_modalities": "text", "trials": "source does not state" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Locked Xiaomi observation for τ³-bench." }, { "model_id": "mimo-v2.5", "benchmark_id": "terminal_bench", "score": 65.8, "reference_url": "https://mimo.xiaomi.com/mimo-v2-5-pro", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "agentic code/terminal tools", "sampling": "source does not state repeated-trial count", "judge": "benchmark-specified", "harness": "official Xiaomi MiMo family benchmark evaluation", "prompt_style": "Xiaomi MiMo official evaluation prompt", "temperature": "1.0; top_p=0.95", "context": "1,048,576", "input_modalities": "text", "trials": "source does not state" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Locked Xiaomi observation for Terminal-Bench 2.0." }, { "model_id": "mimo-v2.5", "benchmark_id": "video_mme", "score": 87.7, "reference_url": "https://huggingface.co/XiaomiMiMo/MiMo-V2.5/resolve/63651580ca774f8504f676040460aed3e1244ac1/assets/mimo-v2.5-multimodal-bench.png", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1; source does not state repeated-trial count", "judge": "benchmark-specified", "harness": "official Xiaomi MiMo family benchmark evaluation", "prompt_style": "Xiaomi MiMo official evaluation prompt", "temperature": "1.0; top_p=0.95", "context": "1,048,576", "input_modalities": "video, audio, image, and text", "trials": "source does not state" }, "matches_canonical": false, "source_type": "official_model_card", "audit_status": "verified", "notes": "Locked Xiaomi observation for Video-MME." }, { "model_id": "mimo-v2.5", "benchmark_id": "videoholmes", "score": 64.0, "reference_url": "https://huggingface.co/XiaomiMiMo/MiMo-V2.5/resolve/63651580ca774f8504f676040460aed3e1244ac1/assets/mimo-v2.5-multimodal-bench.png", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1; source does not state repeated-trial count", "judge": "benchmark-specified", "harness": "official Xiaomi MiMo family benchmark evaluation", "prompt_style": "Xiaomi MiMo official evaluation prompt", "temperature": "1.0; top_p=0.95", "context": "1,048,576", "input_modalities": "video, audio, image, and text", "trials": "source does not state" }, "matches_canonical": false, "source_type": "official_model_card", "audit_status": "verified", "notes": "Locked Xiaomi observation for VideoHolmes." }, { "model_id": "mimo-v2.5-pro", "benchmark_id": "arc_challenge", "score": 97.2, "reference_url": "https://huggingface.co/XiaomiMiMo/MiMo-V2.5-Pro/resolve/21d1ecfecd7bd70f31be25ca49d7edd21f003659/README.md", "reported_setting": { "mode": "base pretrained checkpoint", "effort": "not applicable", "tools": "none", "sampling": "25-shot; single completion per prompt", "judge": "benchmark-native rule-based evaluator", "harness": "Xiaomi MiMo base-checkpoint evaluation", "prompt_style": "25-shot few-shot continuation", "temperature": "source does not state", "context": "256K (reported score-table/model-download setting)", "input_modalities": "text", "trials": "one completion per item", "checkpoint": "base" }, "matches_canonical": false, "source_type": "official_model_card_base_checkpoint", "audit_status": "verified", "notes": "Base checkpoint observation attached to the canonical post-trained model ID. Base result is attached to the canonical post-trained model ID. The locked MiMo-V2.5-Pro-Base config exposes max_position_embeddings=1,048,576 while the score-table/download surface reports 256K; both exact values and the conflict are preserved." }, { "model_id": "mimo-v2.5-pro", "benchmark_id": "bigbench_hard", "score": 88.4, "reference_url": "https://huggingface.co/XiaomiMiMo/MiMo-V2.5-Pro/resolve/21d1ecfecd7bd70f31be25ca49d7edd21f003659/README.md", "reported_setting": { "mode": "base pretrained checkpoint", "effort": "not applicable", "tools": "none", "sampling": "3-shot; single completion per prompt", "judge": "benchmark-native rule-based evaluator", "harness": "Xiaomi MiMo base-checkpoint evaluation", "prompt_style": "3-shot few-shot continuation", "temperature": "source does not state", "context": "256K (reported score-table/model-download setting)", "input_modalities": "text", "trials": "one completion per item", "checkpoint": "base" }, "matches_canonical": false, "source_type": "official_model_card_base_checkpoint", "audit_status": "verified", "notes": "Base checkpoint observation attached to the canonical post-trained model ID. Base result is attached to the canonical post-trained model ID. The locked MiMo-V2.5-Pro-Base config exposes max_position_embeddings=1,048,576 while the score-table/download surface reports 256K; both exact values and the conflict are preserved." }, { "model_id": "mimo-v2.5-pro", "benchmark_id": "c_eval", "score": 91.5, "reference_url": "https://huggingface.co/XiaomiMiMo/MiMo-V2.5-Pro/resolve/21d1ecfecd7bd70f31be25ca49d7edd21f003659/README.md", "reported_setting": { "mode": "base pretrained checkpoint", "effort": "not applicable", "tools": "none", "sampling": "5-shot; single completion per prompt", "judge": "benchmark-native rule-based evaluator", "harness": "Xiaomi MiMo base-checkpoint evaluation", "prompt_style": "5-shot few-shot continuation", "temperature": "source does not state", "context": "256K (reported score-table/model-download setting)", "input_modalities": "text", "trials": "one completion per item", "checkpoint": "base" }, "matches_canonical": false, "source_type": "official_model_card_base_checkpoint", "audit_status": "verified", "notes": "Base checkpoint observation attached to the canonical post-trained model ID. Base result is attached to the canonical post-trained model ID. The locked MiMo-V2.5-Pro-Base config exposes max_position_embeddings=1,048,576 while the score-table/download surface reports 256K; both exact values and the conflict are preserved." }, { "model_id": "mimo-v2.5-pro", "benchmark_id": "claw_eval_pass3", "score": 63.8, "reference_url": "https://raw.githubusercontent.com/claw-eval/claw-eval.github.io/19d378a646d892ca6c8148691aea02da1b184eb8/assets/benchmark-0fIVfCD7.js", "reported_setting": { "mode": "source does not state", "effort": "source does not state", "tools": "official Claw-Eval agent environment", "sampling": "Pass^3; all three trajectories must pass", "judge": "full-trajectory benchmark grading", "harness": "official Claw-Eval v1.1 leaderboard", "prompt_style": "official Claw-Eval task prompts", "temperature": "source does not state", "context": "source does not state", "input_modalities": "text", "trials": "3 per task; 597 generations", "task_subset": "non-multimodal 199-task subset", "task_count": 199 }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Exact official leaderboard field for the non-multimodal 199-task subset. Model mode/effort is not inferred.", "candidates": [ { "score": 64.0, "reference_url": "https://raw.githubusercontent.com/claw-eval/claw-eval.github.io/19d378a646d892ca6c8148691aea02da1b184eb8/assets/benchmark-0fIVfCD7.js", "source_type": "official_blog", "reported_setting": { "mode": "source does not state", "effort": "source does not state", "tools": "official Claw-Eval agent environment", "sampling": "Pass^3; all three trajectories must pass", "judge": "full-trajectory benchmark grading", "harness": "official Claw-Eval v1.1 leaderboard", "prompt_style": "official Claw-Eval task prompts", "temperature": "source does not state", "context": "source does not state", "input_modalities": "text", "trials": "3 per task; 483 generations", "task_subset": "core-general 161-task subset", "task_count": 161 }, "notes": "Average tokens per trajectory=69827 is a physical efficiency annotation, not a benchmark score or score setting. Exact official leaderboard field for the core-general 161-task subset. Model mode/effort is not inferred." } ] }, { "model_id": "mimo-v2.5-pro", "benchmark_id": "cmmlu", "score": 90.2, "reference_url": "https://huggingface.co/XiaomiMiMo/MiMo-V2.5-Pro/resolve/21d1ecfecd7bd70f31be25ca49d7edd21f003659/README.md", "reported_setting": { "mode": "base pretrained checkpoint", "effort": "not applicable", "tools": "none", "sampling": "5-shot; single completion per prompt", "judge": "benchmark-native rule-based evaluator", "harness": "Xiaomi MiMo base-checkpoint evaluation", "prompt_style": "5-shot few-shot continuation", "temperature": "source does not state", "context": "256K (reported score-table/model-download setting)", "input_modalities": "text", "trials": "one completion per item", "checkpoint": "base" }, "matches_canonical": false, "source_type": "official_model_card_base_checkpoint", "audit_status": "verified", "notes": "Base checkpoint observation attached to the canonical post-trained model ID. Base result is attached to the canonical post-trained model ID. The locked MiMo-V2.5-Pro-Base config exposes max_position_embeddings=1,048,576 while the score-table/download surface reports 256K; both exact values and the conflict are preserved." }, { "model_id": "mimo-v2.5-pro", "benchmark_id": "drop", "score": 86.3, "reference_url": "https://huggingface.co/XiaomiMiMo/MiMo-V2.5-Pro/resolve/21d1ecfecd7bd70f31be25ca49d7edd21f003659/README.md", "reported_setting": { "mode": "base pretrained checkpoint", "effort": "not applicable", "tools": "none", "sampling": "3-shot; single completion per prompt", "judge": "benchmark-native rule-based evaluator", "harness": "Xiaomi MiMo base-checkpoint evaluation", "prompt_style": "3-shot few-shot continuation", "temperature": "source does not state", "context": "256K (reported score-table/model-download setting)", "input_modalities": "text", "trials": "one completion per item", "checkpoint": "base" }, "matches_canonical": false, "source_type": "official_model_card_base_checkpoint", "audit_status": "verified", "notes": "Base checkpoint observation attached to the canonical post-trained model ID. Base result is attached to the canonical post-trained model ID. The locked MiMo-V2.5-Pro-Base config exposes max_position_embeddings=1,048,576 while the score-table/download surface reports 256K; both exact values and the conflict are preserved." }, { "model_id": "mimo-v2.5-pro", "benchmark_id": "gdpval_aa_elo", "score": 1581, "reference_url": "https://mimo.xiaomi.com/mimo-v2-5-pro", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1; source does not state repeated-trial count", "judge": "benchmark-specified", "harness": "official Xiaomi MiMo family benchmark evaluation", "prompt_style": "Xiaomi MiMo official evaluation prompt", "temperature": "1.0; top_p=0.95", "context": "1,048,576", "input_modalities": "text", "trials": "source does not state" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Locked Xiaomi observation for GDPVal-AA (Elo)." }, { "model_id": "mimo-v2.5-pro", "benchmark_id": "gpqa_diamond", "score": 66.7, "reference_url": "https://huggingface.co/XiaomiMiMo/MiMo-V2.5-Pro/resolve/21d1ecfecd7bd70f31be25ca49d7edd21f003659/README.md", "reported_setting": { "mode": "base pretrained checkpoint", "effort": "not applicable", "tools": "none", "sampling": "5-shot; single completion per prompt", "judge": "benchmark-native rule-based evaluator", "harness": "Xiaomi MiMo base-checkpoint evaluation", "prompt_style": "5-shot few-shot continuation", "temperature": "source does not state", "context": "256K (reported score-table/model-download setting)", "input_modalities": "text", "trials": "one completion per item", "checkpoint": "base" }, "matches_canonical": false, "source_type": "official_model_card_base_checkpoint", "audit_status": "verified", "notes": "Base checkpoint observation attached to the canonical post-trained model ID. Base result is attached to the canonical post-trained model ID. The locked MiMo-V2.5-Pro-Base config exposes max_position_embeddings=1,048,576 while the score-table/download surface reports 256K; both exact values and the conflict are preserved." }, { "model_id": "mimo-v2.5-pro", "benchmark_id": "graphwalks_bfs_128k", "score": 82.0, "reference_url": "https://huggingface.co/XiaomiMiMo/MiMo-V2.5-Pro/resolve/21d1ecfecd7bd70f31be25ca49d7edd21f003659/assets/post_training_evaluation.jpg", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "one response per graph prompt", "judge": "deterministic set-overlap F1 against answer node list", "harness": "Xiaomi MiMo-V2.5-Pro GraphWalks exact-length evaluation", "prompt_style": "3-shot GraphWalks prompt", "temperature": "1.0; top_p=0.95", "context": "128K", "input_modalities": "text", "trials": "source does not state", "context_window_variant": "exact-length line chart" }, "matches_canonical": false, "source_type": "official_model_card", "audit_status": "verified", "notes": "Literal chart label 0.82; normalized to 82.0 percent without changing meaning. Source prints 0.82 on [0,1]; campaign stores the equivalent 82.0%." }, { "model_id": "mimo-v2.5-pro", "benchmark_id": "graphwalks_bfs_1m", "score": 37.0, "reference_url": "https://huggingface.co/XiaomiMiMo/MiMo-V2.5-Pro/resolve/21d1ecfecd7bd70f31be25ca49d7edd21f003659/assets/post_training_evaluation.jpg", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "one response per graph prompt", "judge": "deterministic set-overlap F1 against answer node list", "harness": "Xiaomi MiMo-V2.5-Pro GraphWalks exact-length evaluation", "prompt_style": "3-shot GraphWalks prompt", "temperature": "1.0; top_p=0.95", "context": "1M", "input_modalities": "text", "trials": "source does not state", "context_window_variant": "exact-length line chart" }, "matches_canonical": false, "source_type": "official_model_card", "audit_status": "verified", "notes": "Literal chart label 0.37; normalized to 37.0 percent without changing meaning. Source prints 0.37 on [0,1]; campaign stores the equivalent 37.0%." }, { "model_id": "mimo-v2.5-pro", "benchmark_id": "graphwalks_bfs_256k", "score": 77.0, "reference_url": "https://huggingface.co/XiaomiMiMo/MiMo-V2.5-Pro/resolve/21d1ecfecd7bd70f31be25ca49d7edd21f003659/assets/post_training_evaluation.jpg", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "one response per graph prompt", "judge": "deterministic set-overlap F1 against answer node list", "harness": "Xiaomi MiMo-V2.5-Pro GraphWalks exact-length evaluation", "prompt_style": "3-shot GraphWalks prompt", "temperature": "1.0; top_p=0.95", "context": "256K", "input_modalities": "text", "trials": "source does not state", "context_window_variant": "exact-length line chart" }, "matches_canonical": false, "source_type": "official_model_card", "audit_status": "verified", "notes": "Literal chart label 0.77; normalized to 77.0 percent without changing meaning. Source prints 0.77 on [0,1]; campaign stores the equivalent 77.0%." }, { "model_id": "mimo-v2.5-pro", "benchmark_id": "graphwalks_bfs_32k", "score": 81.0, "reference_url": "https://huggingface.co/XiaomiMiMo/MiMo-V2.5-Pro/resolve/21d1ecfecd7bd70f31be25ca49d7edd21f003659/assets/post_training_evaluation.jpg", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "one response per graph prompt", "judge": "deterministic set-overlap F1 against answer node list", "harness": "Xiaomi MiMo-V2.5-Pro GraphWalks exact-length evaluation", "prompt_style": "3-shot GraphWalks prompt", "temperature": "1.0; top_p=0.95", "context": "32K", "input_modalities": "text", "trials": "source does not state", "context_window_variant": "exact-length line chart" }, "matches_canonical": false, "source_type": "official_model_card", "audit_status": "verified", "notes": "Literal chart label 0.81; normalized to 81.0 percent without changing meaning. Source prints 0.81 on [0,1]; campaign stores the equivalent 81.0%." }, { "model_id": "mimo-v2.5-pro", "benchmark_id": "graphwalks_bfs_512k", "score": 56.00000000000001, "reference_url": "https://huggingface.co/XiaomiMiMo/MiMo-V2.5-Pro/resolve/21d1ecfecd7bd70f31be25ca49d7edd21f003659/assets/post_training_evaluation.jpg", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "one response per graph prompt", "judge": "deterministic set-overlap F1 against answer node list", "harness": "Xiaomi MiMo-V2.5-Pro GraphWalks exact-length evaluation", "prompt_style": "3-shot GraphWalks prompt", "temperature": "1.0; top_p=0.95", "context": "512K", "input_modalities": "text", "trials": "source does not state", "context_window_variant": "exact-length line chart" }, "matches_canonical": false, "source_type": "official_model_card", "audit_status": "verified", "notes": "Literal chart label 0.56; normalized to 56.0 percent without changing meaning. Source prints 0.56 on [0,1]; campaign stores the equivalent 56.0%." }, { "model_id": "mimo-v2.5-pro", "benchmark_id": "graphwalks_bfs_64k", "score": 79.0, "reference_url": "https://huggingface.co/XiaomiMiMo/MiMo-V2.5-Pro/resolve/21d1ecfecd7bd70f31be25ca49d7edd21f003659/assets/post_training_evaluation.jpg", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "one response per graph prompt", "judge": "deterministic set-overlap F1 against answer node list", "harness": "Xiaomi MiMo-V2.5-Pro GraphWalks exact-length evaluation", "prompt_style": "3-shot GraphWalks prompt", "temperature": "1.0; top_p=0.95", "context": "64K", "input_modalities": "text", "trials": "source does not state", "context_window_variant": "exact-length line chart" }, "matches_canonical": false, "source_type": "official_model_card", "audit_status": "verified", "notes": "Literal chart label 0.79; normalized to 79.0 percent without changing meaning. Source prints 0.79 on [0,1]; campaign stores the equivalent 79.0%." }, { "model_id": "mimo-v2.5-pro", "benchmark_id": "graphwalks_bfs_bucket_0_4k", "score": 94, "reference_url": "https://mimo.xiaomi.com/mimo-v2-5-pro", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "one response per graph prompt", "judge": "deterministic set-overlap F1 against answer node list", "harness": "Xiaomi MiMo GraphWalks evaluation with Anthropic fixes", "prompt_style": "3-shot GraphWalks prompt", "temperature": "1.0; top_p=0.95", "context": "0–4K", "input_modalities": "text", "trials": "n=50", "context_window_variant": "standard" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Exact heatmap bucket. Counts and the 1M variant are preserved; no aggregation into campaign range rows.", "candidates": [ { "score": 94, "reference_url": "https://mimo.xiaomi.com/mimo-v2-5-pro", "source_type": "official_blog", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "one response per graph prompt", "judge": "deterministic set-overlap F1 against answer node list", "harness": "Xiaomi MiMo GraphWalks evaluation with Anthropic fixes", "prompt_style": "3-shot GraphWalks prompt", "temperature": "1.0; top_p=0.95", "context": "0–4K", "input_modalities": "text", "trials": "n=50", "context_window_variant": "1M" }, "notes": "Exact heatmap bucket. Counts and the 1M variant are preserved; no aggregation into campaign range rows." } ] }, { "model_id": "mimo-v2.5-pro", "benchmark_id": "graphwalks_bfs_bucket_128k_256k", "score": 79, "reference_url": "https://mimo.xiaomi.com/mimo-v2-5-pro", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "one response per graph prompt", "judge": "deterministic set-overlap F1 against answer node list", "harness": "Xiaomi MiMo GraphWalks evaluation with Anthropic fixes", "prompt_style": "3-shot GraphWalks prompt", "temperature": "1.0; top_p=0.95", "context": "128K–256K", "input_modalities": "text", "trials": "n=50", "context_window_variant": "standard" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Exact heatmap bucket. Counts and the 1M variant are preserved; no aggregation into campaign range rows.", "candidates": [ { "score": 85, "reference_url": "https://mimo.xiaomi.com/mimo-v2-5-pro", "source_type": "official_blog", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "one response per graph prompt", "judge": "deterministic set-overlap F1 against answer node list", "harness": "Xiaomi MiMo GraphWalks evaluation with Anthropic fixes", "prompt_style": "3-shot GraphWalks prompt", "temperature": "1.0; top_p=0.95", "context": "128K–256K", "input_modalities": "text", "trials": "n=50", "context_window_variant": "1M" }, "notes": "Exact heatmap bucket. Counts and the 1M variant are preserved; no aggregation into campaign range rows." } ] }, { "model_id": "mimo-v2.5-pro", "benchmark_id": "graphwalks_bfs_bucket_16k_32k", "score": 89, "reference_url": "https://mimo.xiaomi.com/mimo-v2-5-pro", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "one response per graph prompt", "judge": "deterministic set-overlap F1 against answer node list", "harness": "Xiaomi MiMo GraphWalks evaluation with Anthropic fixes", "prompt_style": "3-shot GraphWalks prompt", "temperature": "1.0; top_p=0.95", "context": "16K–32K", "input_modalities": "text", "trials": "n=50", "context_window_variant": "standard" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Exact heatmap bucket. Counts and the 1M variant are preserved; no aggregation into campaign range rows.", "candidates": [ { "score": 86, "reference_url": "https://mimo.xiaomi.com/mimo-v2-5-pro", "source_type": "official_blog", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "one response per graph prompt", "judge": "deterministic set-overlap F1 against answer node list", "harness": "Xiaomi MiMo GraphWalks evaluation with Anthropic fixes", "prompt_style": "3-shot GraphWalks prompt", "temperature": "1.0; top_p=0.95", "context": "16K–32K", "input_modalities": "text", "trials": "n=50", "context_window_variant": "1M" }, "notes": "Exact heatmap bucket. Counts and the 1M variant are preserved; no aggregation into campaign range rows." } ] }, { "model_id": "mimo-v2.5-pro", "benchmark_id": "graphwalks_bfs_bucket_256k_512k", "score": 62, "reference_url": "https://mimo.xiaomi.com/mimo-v2-5-pro", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "one response per graph prompt", "judge": "deterministic set-overlap F1 against answer node list", "harness": "Xiaomi MiMo GraphWalks evaluation with Anthropic fixes", "prompt_style": "3-shot GraphWalks prompt", "temperature": "1.0; top_p=0.95", "context": "256K–512K", "input_modalities": "text", "trials": "n=8", "context_window_variant": "1M" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Exact heatmap bucket. Counts and the 1M variant are preserved; no aggregation into campaign range rows." }, { "model_id": "mimo-v2.5-pro", "benchmark_id": "graphwalks_bfs_bucket_32k_64k", "score": 81, "reference_url": "https://mimo.xiaomi.com/mimo-v2-5-pro", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "one response per graph prompt", "judge": "deterministic set-overlap F1 against answer node list", "harness": "Xiaomi MiMo GraphWalks evaluation with Anthropic fixes", "prompt_style": "3-shot GraphWalks prompt", "temperature": "1.0; top_p=0.95", "context": "32K–64K", "input_modalities": "text", "trials": "n=50", "context_window_variant": "standard" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Exact heatmap bucket. Counts and the 1M variant are preserved; no aggregation into campaign range rows.", "candidates": [ { "score": 82, "reference_url": "https://mimo.xiaomi.com/mimo-v2-5-pro", "source_type": "official_blog", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "one response per graph prompt", "judge": "deterministic set-overlap F1 against answer node list", "harness": "Xiaomi MiMo GraphWalks evaluation with Anthropic fixes", "prompt_style": "3-shot GraphWalks prompt", "temperature": "1.0; top_p=0.95", "context": "32K–64K", "input_modalities": "text", "trials": "n=50", "context_window_variant": "1M" }, "notes": "Exact heatmap bucket. Counts and the 1M variant are preserved; no aggregation into campaign range rows." } ] }, { "model_id": "mimo-v2.5-pro", "benchmark_id": "graphwalks_bfs_bucket_4k_8k", "score": 91, "reference_url": "https://mimo.xiaomi.com/mimo-v2-5-pro", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "one response per graph prompt", "judge": "deterministic set-overlap F1 against answer node list", "harness": "Xiaomi MiMo GraphWalks evaluation with Anthropic fixes", "prompt_style": "3-shot GraphWalks prompt", "temperature": "1.0; top_p=0.95", "context": "4K–8K", "input_modalities": "text", "trials": "n=50", "context_window_variant": "standard" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Exact heatmap bucket. Counts and the 1M variant are preserved; no aggregation into campaign range rows.", "candidates": [ { "score": 85, "reference_url": "https://mimo.xiaomi.com/mimo-v2-5-pro", "source_type": "official_blog", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "one response per graph prompt", "judge": "deterministic set-overlap F1 against answer node list", "harness": "Xiaomi MiMo GraphWalks evaluation with Anthropic fixes", "prompt_style": "3-shot GraphWalks prompt", "temperature": "1.0; top_p=0.95", "context": "4K–8K", "input_modalities": "text", "trials": "n=50", "context_window_variant": "1M" }, "notes": "Exact heatmap bucket. Counts and the 1M variant are preserved; no aggregation into campaign range rows." } ] }, { "model_id": "mimo-v2.5-pro", "benchmark_id": "graphwalks_bfs_bucket_64k_128k", "score": 77, "reference_url": "https://mimo.xiaomi.com/mimo-v2-5-pro", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "one response per graph prompt", "judge": "deterministic set-overlap F1 against answer node list", "harness": "Xiaomi MiMo GraphWalks evaluation with Anthropic fixes", "prompt_style": "3-shot GraphWalks prompt", "temperature": "1.0; top_p=0.95", "context": "64K–128K", "input_modalities": "text", "trials": "n=50", "context_window_variant": "standard" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Exact heatmap bucket. Counts and the 1M variant are preserved; no aggregation into campaign range rows.", "candidates": [ { "score": 78, "reference_url": "https://mimo.xiaomi.com/mimo-v2-5-pro", "source_type": "official_blog", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "one response per graph prompt", "judge": "deterministic set-overlap F1 against answer node list", "harness": "Xiaomi MiMo GraphWalks evaluation with Anthropic fixes", "prompt_style": "3-shot GraphWalks prompt", "temperature": "1.0; top_p=0.95", "context": "64K–128K", "input_modalities": "text", "trials": "n=50", "context_window_variant": "1M" }, "notes": "Exact heatmap bucket. Counts and the 1M variant are preserved; no aggregation into campaign range rows." } ] }, { "model_id": "mimo-v2.5-pro", "benchmark_id": "graphwalks_bfs_bucket_8k_16k", "score": 79, "reference_url": "https://mimo.xiaomi.com/mimo-v2-5-pro", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "one response per graph prompt", "judge": "deterministic set-overlap F1 against answer node list", "harness": "Xiaomi MiMo GraphWalks evaluation with Anthropic fixes", "prompt_style": "3-shot GraphWalks prompt", "temperature": "1.0; top_p=0.95", "context": "8K–16K", "input_modalities": "text", "trials": "n=50", "context_window_variant": "standard" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Exact heatmap bucket. Counts and the 1M variant are preserved; no aggregation into campaign range rows.", "candidates": [ { "score": 82, "reference_url": "https://mimo.xiaomi.com/mimo-v2-5-pro", "source_type": "official_blog", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "one response per graph prompt", "judge": "deterministic set-overlap F1 against answer node list", "harness": "Xiaomi MiMo GraphWalks evaluation with Anthropic fixes", "prompt_style": "3-shot GraphWalks prompt", "temperature": "1.0; top_p=0.95", "context": "8K–16K", "input_modalities": "text", "trials": "n=50", "context_window_variant": "1M" }, "notes": "Exact heatmap bucket. Counts and the 1M variant are preserved; no aggregation into campaign range rows." } ] }, { "model_id": "mimo-v2.5-pro", "benchmark_id": "graphwalks_parents_128k", "score": 99.0, "reference_url": "https://huggingface.co/XiaomiMiMo/MiMo-V2.5-Pro/resolve/21d1ecfecd7bd70f31be25ca49d7edd21f003659/assets/post_training_evaluation.jpg", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "one response per graph prompt", "judge": "deterministic set-overlap F1 against answer node list", "harness": "Xiaomi MiMo-V2.5-Pro GraphWalks exact-length evaluation", "prompt_style": "3-shot GraphWalks prompt", "temperature": "1.0; top_p=0.95", "context": "128K", "input_modalities": "text", "trials": "source does not state", "context_window_variant": "exact-length line chart" }, "matches_canonical": false, "source_type": "official_model_card", "audit_status": "verified", "notes": "Literal chart label 0.99; normalized to 99.0 percent without changing meaning. Source prints 0.99 on [0,1]; campaign stores the equivalent 99.0%." }, { "model_id": "mimo-v2.5-pro", "benchmark_id": "graphwalks_parents_1m", "score": 62.0, "reference_url": "https://huggingface.co/XiaomiMiMo/MiMo-V2.5-Pro/resolve/21d1ecfecd7bd70f31be25ca49d7edd21f003659/assets/post_training_evaluation.jpg", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "one response per graph prompt", "judge": "deterministic set-overlap F1 against answer node list", "harness": "Xiaomi MiMo-V2.5-Pro GraphWalks exact-length evaluation", "prompt_style": "3-shot GraphWalks prompt", "temperature": "1.0; top_p=0.95", "context": "1M", "input_modalities": "text", "trials": "source does not state", "context_window_variant": "exact-length line chart" }, "matches_canonical": false, "source_type": "official_model_card", "audit_status": "verified", "notes": "Literal chart label 0.62; normalized to 62.0 percent without changing meaning. Source prints 0.62 on [0,1]; campaign stores the equivalent 62.0%." }, { "model_id": "mimo-v2.5-pro", "benchmark_id": "graphwalks_parents_256k", "score": 97.0, "reference_url": "https://huggingface.co/XiaomiMiMo/MiMo-V2.5-Pro/resolve/21d1ecfecd7bd70f31be25ca49d7edd21f003659/assets/post_training_evaluation.jpg", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "one response per graph prompt", "judge": "deterministic set-overlap F1 against answer node list", "harness": "Xiaomi MiMo-V2.5-Pro GraphWalks exact-length evaluation", "prompt_style": "3-shot GraphWalks prompt", "temperature": "1.0; top_p=0.95", "context": "256K", "input_modalities": "text", "trials": "source does not state", "context_window_variant": "exact-length line chart" }, "matches_canonical": false, "source_type": "official_model_card", "audit_status": "verified", "notes": "Literal chart label 0.97; normalized to 97.0 percent without changing meaning. Source prints 0.97 on [0,1]; campaign stores the equivalent 97.0%." }, { "model_id": "mimo-v2.5-pro", "benchmark_id": "graphwalks_parents_32k", "score": 98.0, "reference_url": "https://huggingface.co/XiaomiMiMo/MiMo-V2.5-Pro/resolve/21d1ecfecd7bd70f31be25ca49d7edd21f003659/assets/post_training_evaluation.jpg", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "one response per graph prompt", "judge": "deterministic set-overlap F1 against answer node list", "harness": "Xiaomi MiMo-V2.5-Pro GraphWalks exact-length evaluation", "prompt_style": "3-shot GraphWalks prompt", "temperature": "1.0; top_p=0.95", "context": "32K", "input_modalities": "text", "trials": "source does not state", "context_window_variant": "exact-length line chart" }, "matches_canonical": false, "source_type": "official_model_card", "audit_status": "verified", "notes": "Literal chart label 0.98; normalized to 98.0 percent without changing meaning. Source prints 0.98 on [0,1]; campaign stores the equivalent 98.0%." }, { "model_id": "mimo-v2.5-pro", "benchmark_id": "graphwalks_parents_512k", "score": 92.0, "reference_url": "https://huggingface.co/XiaomiMiMo/MiMo-V2.5-Pro/resolve/21d1ecfecd7bd70f31be25ca49d7edd21f003659/assets/post_training_evaluation.jpg", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "one response per graph prompt", "judge": "deterministic set-overlap F1 against answer node list", "harness": "Xiaomi MiMo-V2.5-Pro GraphWalks exact-length evaluation", "prompt_style": "3-shot GraphWalks prompt", "temperature": "1.0; top_p=0.95", "context": "512K", "input_modalities": "text", "trials": "source does not state", "context_window_variant": "exact-length line chart" }, "matches_canonical": false, "source_type": "official_model_card", "audit_status": "verified", "notes": "Literal chart label 0.92; normalized to 92.0 percent without changing meaning. Source prints 0.92 on [0,1]; campaign stores the equivalent 92.0%." }, { "model_id": "mimo-v2.5-pro", "benchmark_id": "graphwalks_parents_64k", "score": 98.0, "reference_url": "https://huggingface.co/XiaomiMiMo/MiMo-V2.5-Pro/resolve/21d1ecfecd7bd70f31be25ca49d7edd21f003659/assets/post_training_evaluation.jpg", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "one response per graph prompt", "judge": "deterministic set-overlap F1 against answer node list", "harness": "Xiaomi MiMo-V2.5-Pro GraphWalks exact-length evaluation", "prompt_style": "3-shot GraphWalks prompt", "temperature": "1.0; top_p=0.95", "context": "64K", "input_modalities": "text", "trials": "source does not state", "context_window_variant": "exact-length line chart" }, "matches_canonical": false, "source_type": "official_model_card", "audit_status": "verified", "notes": "Literal chart label 0.98; normalized to 98.0 percent without changing meaning. Source prints 0.98 on [0,1]; campaign stores the equivalent 98.0%." }, { "model_id": "mimo-v2.5-pro", "benchmark_id": "graphwalks_parents_bucket_0_4k", "score": 99, "reference_url": "https://mimo.xiaomi.com/mimo-v2-5-pro", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "one response per graph prompt", "judge": "deterministic set-overlap F1 against answer node list", "harness": "Xiaomi MiMo GraphWalks evaluation with Anthropic fixes", "prompt_style": "3-shot GraphWalks prompt", "temperature": "1.0; top_p=0.95", "context": "0–4K", "input_modalities": "text", "trials": "n=100", "context_window_variant": "standard" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Exact heatmap bucket. Counts and the 1M variant are preserved; no aggregation into campaign range rows.", "candidates": [ { "score": 98, "reference_url": "https://mimo.xiaomi.com/mimo-v2-5-pro", "source_type": "official_blog", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "one response per graph prompt", "judge": "deterministic set-overlap F1 against answer node list", "harness": "Xiaomi MiMo GraphWalks evaluation with Anthropic fixes", "prompt_style": "3-shot GraphWalks prompt", "temperature": "1.0; top_p=0.95", "context": "0–4K", "input_modalities": "text", "trials": "n=100", "context_window_variant": "1M" }, "notes": "Exact heatmap bucket. Counts and the 1M variant are preserved; no aggregation into campaign range rows." } ] }, { "model_id": "mimo-v2.5-pro", "benchmark_id": "graphwalks_parents_bucket_128k_256k", "score": 95, "reference_url": "https://mimo.xiaomi.com/mimo-v2-5-pro", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "one response per graph prompt", "judge": "deterministic set-overlap F1 against answer node list", "harness": "Xiaomi MiMo GraphWalks evaluation with Anthropic fixes", "prompt_style": "3-shot GraphWalks prompt", "temperature": "1.0; top_p=0.95", "context": "128K–256K", "input_modalities": "text", "trials": "n=50", "context_window_variant": "standard" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Exact heatmap bucket. Counts and the 1M variant are preserved; no aggregation into campaign range rows.", "candidates": [ { "score": 98, "reference_url": "https://mimo.xiaomi.com/mimo-v2-5-pro", "source_type": "official_blog", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "one response per graph prompt", "judge": "deterministic set-overlap F1 against answer node list", "harness": "Xiaomi MiMo GraphWalks evaluation with Anthropic fixes", "prompt_style": "3-shot GraphWalks prompt", "temperature": "1.0; top_p=0.95", "context": "128K–256K", "input_modalities": "text", "trials": "n=48", "context_window_variant": "1M" }, "notes": "Exact heatmap bucket. Counts and the 1M variant are preserved; no aggregation into campaign range rows." } ] }, { "model_id": "mimo-v2.5-pro", "benchmark_id": "graphwalks_parents_bucket_16k_32k", "score": 100, "reference_url": "https://mimo.xiaomi.com/mimo-v2-5-pro", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "one response per graph prompt", "judge": "deterministic set-overlap F1 against answer node list", "harness": "Xiaomi MiMo GraphWalks evaluation with Anthropic fixes", "prompt_style": "3-shot GraphWalks prompt", "temperature": "1.0; top_p=0.95", "context": "16K–32K", "input_modalities": "text", "trials": "n=50", "context_window_variant": "standard" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Exact heatmap bucket. Counts and the 1M variant are preserved; no aggregation into campaign range rows.", "candidates": [ { "score": 98, "reference_url": "https://mimo.xiaomi.com/mimo-v2-5-pro", "source_type": "official_blog", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "one response per graph prompt", "judge": "deterministic set-overlap F1 against answer node list", "harness": "Xiaomi MiMo GraphWalks evaluation with Anthropic fixes", "prompt_style": "3-shot GraphWalks prompt", "temperature": "1.0; top_p=0.95", "context": "16K–32K", "input_modalities": "text", "trials": "n=50", "context_window_variant": "1M" }, "notes": "Exact heatmap bucket. Counts and the 1M variant are preserved; no aggregation into campaign range rows." } ] }, { "model_id": "mimo-v2.5-pro", "benchmark_id": "graphwalks_parents_bucket_256k_512k", "score": 97, "reference_url": "https://mimo.xiaomi.com/mimo-v2-5-pro", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "one response per graph prompt", "judge": "deterministic set-overlap F1 against answer node list", "harness": "Xiaomi MiMo GraphWalks evaluation with Anthropic fixes", "prompt_style": "3-shot GraphWalks prompt", "temperature": "1.0; top_p=0.95", "context": "256K–512K", "input_modalities": "text", "trials": "n=3", "context_window_variant": "1M" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Exact heatmap bucket. Counts and the 1M variant are preserved; no aggregation into campaign range rows." }, { "model_id": "mimo-v2.5-pro", "benchmark_id": "graphwalks_parents_bucket_32k_64k", "score": 100, "reference_url": "https://mimo.xiaomi.com/mimo-v2-5-pro", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "one response per graph prompt", "judge": "deterministic set-overlap F1 against answer node list", "harness": "Xiaomi MiMo GraphWalks evaluation with Anthropic fixes", "prompt_style": "3-shot GraphWalks prompt", "temperature": "1.0; top_p=0.95", "context": "32K–64K", "input_modalities": "text", "trials": "n=50", "context_window_variant": "standard" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Exact heatmap bucket. Counts and the 1M variant are preserved; no aggregation into campaign range rows.", "candidates": [ { "score": 100, "reference_url": "https://mimo.xiaomi.com/mimo-v2-5-pro", "source_type": "official_blog", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "one response per graph prompt", "judge": "deterministic set-overlap F1 against answer node list", "harness": "Xiaomi MiMo GraphWalks evaluation with Anthropic fixes", "prompt_style": "3-shot GraphWalks prompt", "temperature": "1.0; top_p=0.95", "context": "32K–64K", "input_modalities": "text", "trials": "n=50", "context_window_variant": "1M" }, "notes": "Exact heatmap bucket. Counts and the 1M variant are preserved; no aggregation into campaign range rows." } ] }, { "model_id": "mimo-v2.5-pro", "benchmark_id": "graphwalks_parents_bucket_4k_8k", "score": 100, "reference_url": "https://mimo.xiaomi.com/mimo-v2-5-pro", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "one response per graph prompt", "judge": "deterministic set-overlap F1 against answer node list", "harness": "Xiaomi MiMo GraphWalks evaluation with Anthropic fixes", "prompt_style": "3-shot GraphWalks prompt", "temperature": "1.0; top_p=0.95", "context": "4K–8K", "input_modalities": "text", "trials": "n=50", "context_window_variant": "standard" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Exact heatmap bucket. Counts and the 1M variant are preserved; no aggregation into campaign range rows.", "candidates": [ { "score": 100, "reference_url": "https://mimo.xiaomi.com/mimo-v2-5-pro", "source_type": "official_blog", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "one response per graph prompt", "judge": "deterministic set-overlap F1 against answer node list", "harness": "Xiaomi MiMo GraphWalks evaluation with Anthropic fixes", "prompt_style": "3-shot GraphWalks prompt", "temperature": "1.0; top_p=0.95", "context": "4K–8K", "input_modalities": "text", "trials": "n=50", "context_window_variant": "1M" }, "notes": "Exact heatmap bucket. Counts and the 1M variant are preserved; no aggregation into campaign range rows." } ] }, { "model_id": "mimo-v2.5-pro", "benchmark_id": "graphwalks_parents_bucket_64k_128k", "score": 99, "reference_url": "https://mimo.xiaomi.com/mimo-v2-5-pro", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "one response per graph prompt", "judge": "deterministic set-overlap F1 against answer node list", "harness": "Xiaomi MiMo GraphWalks evaluation with Anthropic fixes", "prompt_style": "3-shot GraphWalks prompt", "temperature": "1.0; top_p=0.95", "context": "64K–128K", "input_modalities": "text", "trials": "n=50", "context_window_variant": "standard" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Exact heatmap bucket. Counts and the 1M variant are preserved; no aggregation into campaign range rows.", "candidates": [ { "score": 99, "reference_url": "https://mimo.xiaomi.com/mimo-v2-5-pro", "source_type": "official_blog", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "one response per graph prompt", "judge": "deterministic set-overlap F1 against answer node list", "harness": "Xiaomi MiMo GraphWalks evaluation with Anthropic fixes", "prompt_style": "3-shot GraphWalks prompt", "temperature": "1.0; top_p=0.95", "context": "64K–128K", "input_modalities": "text", "trials": "n=50", "context_window_variant": "1M" }, "notes": "Exact heatmap bucket. Counts and the 1M variant are preserved; no aggregation into campaign range rows." } ] }, { "model_id": "mimo-v2.5-pro", "benchmark_id": "graphwalks_parents_bucket_8k_16k", "score": 100, "reference_url": "https://mimo.xiaomi.com/mimo-v2-5-pro", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "one response per graph prompt", "judge": "deterministic set-overlap F1 against answer node list", "harness": "Xiaomi MiMo GraphWalks evaluation with Anthropic fixes", "prompt_style": "3-shot GraphWalks prompt", "temperature": "1.0; top_p=0.95", "context": "8K–16K", "input_modalities": "text", "trials": "n=50", "context_window_variant": "standard" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Exact heatmap bucket. Counts and the 1M variant are preserved; no aggregation into campaign range rows.", "candidates": [ { "score": 100, "reference_url": "https://mimo.xiaomi.com/mimo-v2-5-pro", "source_type": "official_blog", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "one response per graph prompt", "judge": "deterministic set-overlap F1 against answer node list", "harness": "Xiaomi MiMo GraphWalks evaluation with Anthropic fixes", "prompt_style": "3-shot GraphWalks prompt", "temperature": "1.0; top_p=0.95", "context": "8K–16K", "input_modalities": "text", "trials": "n=50", "context_window_variant": "1M" }, "notes": "Exact heatmap bucket. Counts and the 1M variant are preserved; no aggregation into campaign range rows." } ] }, { "model_id": "mimo-v2.5-pro", "benchmark_id": "gsm8k", "score": 99.6, "reference_url": "https://huggingface.co/XiaomiMiMo/MiMo-V2.5-Pro/resolve/21d1ecfecd7bd70f31be25ca49d7edd21f003659/README.md", "reported_setting": { "mode": "base pretrained checkpoint", "effort": "not applicable", "tools": "none", "sampling": "8-shot; single completion per prompt", "judge": "benchmark-native rule-based evaluator", "harness": "Xiaomi MiMo base-checkpoint evaluation", "prompt_style": "8-shot few-shot continuation", "temperature": "source does not state", "context": "256K (reported score-table/model-download setting)", "input_modalities": "text", "trials": "one completion per item", "checkpoint": "base" }, "matches_canonical": false, "source_type": "official_model_card_base_checkpoint", "audit_status": "verified", "notes": "Base checkpoint observation attached to the canonical post-trained model ID. Base result is attached to the canonical post-trained model ID. The locked MiMo-V2.5-Pro-Base config exposes max_position_embeddings=1,048,576 while the score-table/download surface reports 256K; both exact values and the conflict are preserved." }, { "model_id": "mimo-v2.5-pro", "benchmark_id": "hellaswag", "score": 89.8, "reference_url": "https://huggingface.co/XiaomiMiMo/MiMo-V2.5-Pro/resolve/21d1ecfecd7bd70f31be25ca49d7edd21f003659/README.md", "reported_setting": { "mode": "base pretrained checkpoint", "effort": "not applicable", "tools": "none", "sampling": "10-shot; single completion per prompt", "judge": "benchmark-native rule-based evaluator", "harness": "Xiaomi MiMo base-checkpoint evaluation", "prompt_style": "10-shot few-shot continuation", "temperature": "source does not state", "context": "256K (reported score-table/model-download setting)", "input_modalities": "text", "trials": "one completion per item", "checkpoint": "base" }, "matches_canonical": false, "source_type": "official_model_card_base_checkpoint", "audit_status": "verified", "notes": "Base checkpoint observation attached to the canonical post-trained model ID. Base result is attached to the canonical post-trained model ID. The locked MiMo-V2.5-Pro-Base config exposes max_position_embeddings=1,048,576 while the score-table/download surface reports 256K; both exact values and the conflict are preserved." }, { "model_id": "mimo-v2.5-pro", "benchmark_id": "hle", "score": 34.0, "reference_url": "https://mimo.xiaomi.com/mimo-v2-5-pro", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1; source does not state trial count", "judge": "benchmark-specified", "harness": "official Xiaomi MiMo family benchmark evaluation", "prompt_style": "Xiaomi MiMo official evaluation prompt", "temperature": "1.0; top_p=0.95", "context": "1,048,576", "input_modalities": "text and image", "trials": "source does not state" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Semantic alternative printed inside the same physical HLE cell; it does not add a physical position. Locked Xiaomi observation for Humanity's Last Exam." }, { "model_id": "mimo-v2.5-pro", "benchmark_id": "hle_tools", "score": 48.0, "reference_url": "https://mimo.xiaomi.com/mimo-v2-5-pro", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "source-reported tools; exact tool set undisclosed", "sampling": "pass@1; source does not state trial count", "judge": "benchmark-specified", "harness": "official Xiaomi MiMo family benchmark evaluation", "prompt_style": "Xiaomi MiMo official evaluation prompt", "temperature": "1.0; top_p=0.95", "context": "1,048,576", "input_modalities": "text and image", "trials": "source does not state" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Locked Xiaomi observation for Humanity's Last Exam." }, { "model_id": "mimo-v2.5-pro", "benchmark_id": "humaneval_plus", "score": 75.6, "reference_url": "https://huggingface.co/XiaomiMiMo/MiMo-V2.5-Pro/resolve/21d1ecfecd7bd70f31be25ca49d7edd21f003659/README.md", "reported_setting": { "mode": "base pretrained checkpoint", "effort": "not applicable", "tools": "none", "sampling": "1-shot; single completion per prompt", "judge": "benchmark-native rule-based evaluator", "harness": "Xiaomi MiMo base-checkpoint evaluation", "prompt_style": "1-shot few-shot continuation", "temperature": "source does not state", "context": "256K (reported score-table/model-download setting)", "input_modalities": "text", "trials": "one completion per item", "checkpoint": "base" }, "matches_canonical": false, "source_type": "official_model_card_base_checkpoint", "audit_status": "verified", "notes": "Base checkpoint observation attached to the canonical post-trained model ID. Base result is attached to the canonical post-trained model ID. The locked MiMo-V2.5-Pro-Base config exposes max_position_embeddings=1,048,576 while the score-table/download surface reports 256K; both exact values and the conflict are preserved." }, { "model_id": "mimo-v2.5-pro", "benchmark_id": "livecodebench_v6", "score": 39.6, "reference_url": "https://huggingface.co/XiaomiMiMo/MiMo-V2.5-Pro/resolve/21d1ecfecd7bd70f31be25ca49d7edd21f003659/README.md", "reported_setting": { "mode": "base pretrained checkpoint", "effort": "not applicable", "tools": "none", "sampling": "1-shot; single completion per prompt", "judge": "benchmark-native rule-based evaluator", "harness": "Xiaomi MiMo base-checkpoint evaluation", "prompt_style": "1-shot few-shot continuation", "temperature": "source does not state", "context": "256K (reported score-table/model-download setting)", "input_modalities": "text", "trials": "one completion per item", "checkpoint": "base" }, "matches_canonical": false, "source_type": "official_model_card_base_checkpoint", "audit_status": "verified", "notes": "Base checkpoint observation attached to the canonical post-trained model ID. Base result is attached to the canonical post-trained model ID. The locked MiMo-V2.5-Pro-Base config exposes max_position_embeddings=1,048,576 while the score-table/download surface reports 256K; both exact values and the conflict are preserved." }, { "model_id": "mimo-v2.5-pro", "benchmark_id": "math", "score": 86.2, "reference_url": "https://huggingface.co/XiaomiMiMo/MiMo-V2.5-Pro/resolve/21d1ecfecd7bd70f31be25ca49d7edd21f003659/README.md", "reported_setting": { "mode": "base pretrained checkpoint", "effort": "not applicable", "tools": "none", "sampling": "4-shot; single completion per prompt", "judge": "benchmark-native rule-based evaluator", "harness": "Xiaomi MiMo base-checkpoint evaluation", "prompt_style": "4-shot few-shot continuation", "temperature": "source does not state", "context": "256K (reported score-table/model-download setting)", "input_modalities": "text", "trials": "one completion per item", "checkpoint": "base" }, "matches_canonical": false, "source_type": "official_model_card_base_checkpoint", "audit_status": "verified", "notes": "Base checkpoint observation attached to the canonical post-trained model ID. Base result is attached to the canonical post-trained model ID. The locked MiMo-V2.5-Pro-Base config exposes max_position_embeddings=1,048,576 while the score-table/download surface reports 256K; both exact values and the conflict are preserved." }, { "model_id": "mimo-v2.5-pro", "benchmark_id": "mbpp_plus", "score": 74.1, "reference_url": "https://huggingface.co/XiaomiMiMo/MiMo-V2.5-Pro/resolve/21d1ecfecd7bd70f31be25ca49d7edd21f003659/README.md", "reported_setting": { "mode": "base pretrained checkpoint", "effort": "not applicable", "tools": "none", "sampling": "3-shot; single completion per prompt", "judge": "benchmark-native rule-based evaluator", "harness": "Xiaomi MiMo base-checkpoint evaluation", "prompt_style": "3-shot few-shot continuation", "temperature": "source does not state", "context": "256K (reported score-table/model-download setting)", "input_modalities": "text", "trials": "one completion per item", "checkpoint": "base" }, "matches_canonical": false, "source_type": "official_model_card_base_checkpoint", "audit_status": "verified", "notes": "Base checkpoint observation attached to the canonical post-trained model ID. Base result is attached to the canonical post-trained model ID. The locked MiMo-V2.5-Pro-Base config exposes max_position_embeddings=1,048,576 while the score-table/download surface reports 256K; both exact values and the conflict are preserved." }, { "model_id": "mimo-v2.5-pro", "benchmark_id": "mmlu", "score": 89.4, "reference_url": "https://huggingface.co/XiaomiMiMo/MiMo-V2.5-Pro/resolve/21d1ecfecd7bd70f31be25ca49d7edd21f003659/README.md", "reported_setting": { "mode": "base pretrained checkpoint", "effort": "not applicable", "tools": "none", "sampling": "5-shot; single completion per prompt", "judge": "benchmark-native rule-based evaluator", "harness": "Xiaomi MiMo base-checkpoint evaluation", "prompt_style": "5-shot few-shot continuation", "temperature": "source does not state", "context": "256K (reported score-table/model-download setting)", "input_modalities": "text", "trials": "one completion per item", "checkpoint": "base" }, "matches_canonical": false, "source_type": "official_model_card_base_checkpoint", "audit_status": "verified", "notes": "Base checkpoint observation attached to the canonical post-trained model ID. Base result is attached to the canonical post-trained model ID. The locked MiMo-V2.5-Pro-Base config exposes max_position_embeddings=1,048,576 while the score-table/download surface reports 256K; both exact values and the conflict are preserved." }, { "model_id": "mimo-v2.5-pro", "benchmark_id": "mmlu_pro", "score": 68.5, "reference_url": "https://huggingface.co/XiaomiMiMo/MiMo-V2.5-Pro/resolve/21d1ecfecd7bd70f31be25ca49d7edd21f003659/README.md", "reported_setting": { "mode": "base pretrained checkpoint", "effort": "not applicable", "tools": "none", "sampling": "5-shot; single completion per prompt", "judge": "benchmark-native rule-based evaluator", "harness": "Xiaomi MiMo base-checkpoint evaluation", "prompt_style": "5-shot few-shot continuation", "temperature": "source does not state", "context": "256K (reported score-table/model-download setting)", "input_modalities": "text", "trials": "one completion per item", "checkpoint": "base" }, "matches_canonical": false, "source_type": "official_model_card_base_checkpoint", "audit_status": "verified", "notes": "Base checkpoint observation attached to the canonical post-trained model ID. Base result is attached to the canonical post-trained model ID. The locked MiMo-V2.5-Pro-Base config exposes max_position_embeddings=1,048,576 while the score-table/download surface reports 256K; both exact values and the conflict are preserved." }, { "model_id": "mimo-v2.5-pro", "benchmark_id": "mmlu_redux", "score": 92.8, "reference_url": "https://huggingface.co/XiaomiMiMo/MiMo-V2.5-Pro/resolve/21d1ecfecd7bd70f31be25ca49d7edd21f003659/README.md", "reported_setting": { "mode": "base pretrained checkpoint", "effort": "not applicable", "tools": "none", "sampling": "5-shot; single completion per prompt", "judge": "benchmark-native rule-based evaluator", "harness": "Xiaomi MiMo base-checkpoint evaluation", "prompt_style": "5-shot few-shot continuation", "temperature": "source does not state", "context": "256K (reported score-table/model-download setting)", "input_modalities": "text", "trials": "one completion per item", "checkpoint": "base" }, "matches_canonical": false, "source_type": "official_model_card_base_checkpoint", "audit_status": "verified", "notes": "Base checkpoint observation attached to the canonical post-trained model ID. Base result is attached to the canonical post-trained model ID. The locked MiMo-V2.5-Pro-Base config exposes max_position_embeddings=1,048,576 while the score-table/download surface reports 256K; both exact values and the conflict are preserved." }, { "model_id": "mimo-v2.5-pro", "benchmark_id": "swe_bench_pro", "score": 57.2, "reference_url": "https://mimo.xiaomi.com/mimo-v2-5-pro", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "agentic code/terminal tools", "sampling": "source does not state repeated-trial count", "judge": "benchmark-specified", "harness": "official Xiaomi MiMo family benchmark evaluation", "prompt_style": "Xiaomi MiMo official evaluation prompt", "temperature": "1.0; top_p=0.95", "context": "1,048,576", "input_modalities": "text", "trials": "source does not state" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Locked Xiaomi observation for SWE-Bench Pro." }, { "model_id": "mimo-v2.5-pro", "benchmark_id": "swe_bench_verified", "score": 78.9, "reference_url": "https://mimo.xiaomi.com/mimo-v2-5-pro", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "agentic code/terminal tools", "sampling": "source does not state repeated-trial count", "judge": "benchmark-specified", "harness": "official Xiaomi MiMo family benchmark evaluation", "prompt_style": "Xiaomi MiMo official evaluation prompt", "temperature": "1.0; top_p=0.95", "context": "1,048,576", "input_modalities": "text", "trials": "source does not state" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Locked Xiaomi observation for SWE-bench Verified." }, { "model_id": "mimo-v2.5-pro", "benchmark_id": "tau3_bench", "score": 72.9, "reference_url": "https://mimo.xiaomi.com/mimo-v2-5-pro", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "none", "sampling": "pass@1; source does not state repeated-trial count", "judge": "benchmark-specified", "harness": "official Xiaomi MiMo family benchmark evaluation", "prompt_style": "Xiaomi MiMo official evaluation prompt", "temperature": "1.0; top_p=0.95", "context": "1,048,576", "input_modalities": "text", "trials": "source does not state" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Locked Xiaomi observation for τ³-bench." }, { "model_id": "mimo-v2.5-pro", "benchmark_id": "terminal_bench", "score": 68.4, "reference_url": "https://mimo.xiaomi.com/mimo-v2-5-pro", "reported_setting": { "mode": "thinking", "effort": "default", "tools": "agentic code/terminal tools", "sampling": "source does not state repeated-trial count", "judge": "benchmark-specified", "harness": "official Xiaomi MiMo family benchmark evaluation", "prompt_style": "Xiaomi MiMo official evaluation prompt", "temperature": "1.0; top_p=0.95", "context": "1,048,576", "input_modalities": "text", "trials": "source does not state" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Locked Xiaomi observation for Terminal-Bench 2.0." }, { "model_id": "minimax-m2.7", "benchmark_id": "claw_eval_pass3", "score": 49.7, "reference_url": "https://raw.githubusercontent.com/claw-eval/claw-eval.github.io/19d378a646d892ca6c8148691aea02da1b184eb8/assets/benchmark-0fIVfCD7.js", "reported_setting": { "mode": "source does not state", "effort": "source does not state", "tools": "official Claw-Eval agent environment", "sampling": "Pass^3; all three trajectories must pass", "judge": "full-trajectory benchmark grading", "harness": "official Claw-Eval v1.1 leaderboard", "prompt_style": "official Claw-Eval task prompts", "temperature": "source does not state", "context": "source does not state", "input_modalities": "text", "trials": "3 per task; 483 generations", "task_subset": "core-general 161-task subset", "task_count": 161 }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Average tokens per trajectory=91135 is a physical efficiency annotation, not a benchmark score or score setting. Exact official leaderboard field for the core-general 161-task subset. Model mode/effort is not inferred." }, { "model_id": "muse-spark", "benchmark_id": "claw_eval_pass3", "score": 62.7, "reference_url": "https://raw.githubusercontent.com/claw-eval/claw-eval.github.io/19d378a646d892ca6c8148691aea02da1b184eb8/assets/benchmark-0fIVfCD7.js", "reported_setting": { "mode": "source does not state", "effort": "source does not state", "tools": "official Claw-Eval agent environment", "sampling": "Pass^3; all three trajectories must pass", "judge": "full-trajectory benchmark grading", "harness": "official Claw-Eval v1.1 leaderboard", "prompt_style": "official Claw-Eval task prompts", "temperature": "source does not state", "context": "source does not state", "input_modalities": "text", "trials": "3 per task; 483 generations", "task_subset": "core-general 161-task subset", "task_count": 161 }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Average tokens per trajectory=162688 is a physical efficiency annotation, not a benchmark score or score setting. Exact official leaderboard field for the core-general 161-task subset. Model mode/effort is not inferred." }, { "model_id": "qwen3.5-397b", "benchmark_id": "claw_eval_pass3", "score": 57.8, "reference_url": "https://raw.githubusercontent.com/claw-eval/claw-eval.github.io/19d378a646d892ca6c8148691aea02da1b184eb8/assets/benchmark-0fIVfCD7.js", "reported_setting": { "mode": "source does not state", "effort": "source does not state", "tools": "official Claw-Eval agent environment", "sampling": "Pass^3; all three trajectories must pass", "judge": "full-trajectory benchmark grading", "harness": "official Claw-Eval v1.1 leaderboard", "prompt_style": "official Claw-Eval task prompts", "temperature": "source does not state", "context": "source does not state", "input_modalities": "text", "trials": "3 per task; 483 generations", "task_subset": "core-general 161-task subset", "task_count": 161 }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Average tokens per trajectory=147963 is a physical efficiency annotation, not a benchmark score or score setting. Exact official leaderboard field for the core-general 161-task subset. Model mode/effort is not inferred." }, { "model_id": "ling-2.6-1t", "benchmark_id": "aime_2026", "score": 87.4, "reference_url": "https://arxiv.org/pdf/2606.15079v1", "reported_setting": { "checkpoint": "post-trained", "mode": "instant/fast-thinking", "effort": "default", "shots": "source does not state", "samples": 64, "trials": 64, "aggregation": "mean@64", "tools": "none", "search": "none", "context_management": "none stated", "judge": "benchmark-specified", "harness": "official InclusionAI / benchmark-specific", "dataset_version": "AIME 2026", "split": "source does not state", "context_length": "262,144 native config", "multimodal_inputs": false, "temperature": "source does not state", "top_p": "source does not state", "top_k": "source does not state", "prompt_style": "source does not state", "max_new_tokens": "source does not state", "timeout": "source does not state", "output_parser": "source does not state", "evaluation_provenance": "official InclusionAI evaluation or locked provider rendering" }, "matches_canonical": true, "source_type": "official_paper", "audit_status": "verified", "notes": "Displayed exactly as 87.40. Locked official InclusionAI observation." }, { "model_id": "ling-2.6-1t", "benchmark_id": "bfcl_v4", "score": 70.64, "reference_url": "https://arxiv.org/pdf/2606.15079v1", "reported_setting": { "checkpoint": "post-trained", "mode": "instant/fast-thinking", "effort": "default", "shots": "source does not state", "samples": 1, "trials": 1, "aggregation": "Accuracy", "tools": "benchmark-provided functions", "search": "none", "context_management": "none stated", "judge": "official function-call match", "harness": "Berkeley Function Calling Leaderboard", "dataset_version": "BFCL v4", "split": "source does not state", "context_length": "262,144 native config", "multimodal_inputs": false, "temperature": "source does not state", "top_p": "source does not state", "top_k": "source does not state", "prompt_style": "source does not state", "max_new_tokens": "source does not state", "timeout": "source does not state", "output_parser": "source does not state", "evaluation_provenance": "official InclusionAI evaluation or locked provider rendering" }, "matches_canonical": true, "source_type": "official_paper", "audit_status": "verified", "notes": "Displayed exactly as 70.64. Locked official InclusionAI observation." }, { "model_id": "ling-2.6-1t", "benchmark_id": "bigbench_extra_hard", "score": 52.37, "reference_url": "https://arxiv.org/pdf/2606.15079v1", "reported_setting": { "checkpoint": "post-trained", "mode": "instant/fast-thinking", "effort": "default", "shots": "source does not state", "samples": 1, "trials": 1, "aggregation": "pass@1", "tools": "none", "search": "none", "context_management": "none stated", "judge": "official BBEH exact-match/micro-accuracy evaluator", "harness": "official InclusionAI / benchmark-specific", "dataset_version": "bbeh", "split": "full 4520 examples", "context_length": "262,144 native config", "multimodal_inputs": false, "temperature": "source does not state", "top_p": "source does not state", "top_k": "source does not state", "prompt_style": "source does not state", "max_new_tokens": "source does not state", "timeout": "source does not state", "output_parser": "source does not state", "evaluation_provenance": "official InclusionAI evaluation or locked provider rendering" }, "matches_canonical": true, "source_type": "official_paper", "audit_status": "verified", "notes": "Displayed exactly as 52.37. Locked official InclusionAI observation." }, { "model_id": "ling-2.6-1t", "benchmark_id": "chinese_simpleqa", "score": 76.53, "reference_url": "https://arxiv.org/pdf/2606.15079v1", "reported_setting": { "checkpoint": "post-trained", "mode": "instant/fast-thinking", "effort": "default", "shots": "source does not state", "samples": 1, "trials": 1, "aggregation": "Correct", "tools": "none", "search": "none", "context_management": "none stated", "judge": "benchmark-specified", "harness": "official InclusionAI / benchmark-specific", "dataset_version": "C-SimpleQA", "split": "source does not state", "context_length": "262,144 native config", "multimodal_inputs": false, "temperature": "source does not state", "top_p": "source does not state", "top_k": "source does not state", "prompt_style": "source does not state", "max_new_tokens": "source does not state", "timeout": "source does not state", "output_parser": "source does not state", "evaluation_provenance": "official InclusionAI evaluation or locked provider rendering" }, "matches_canonical": true, "source_type": "official_paper", "audit_status": "verified", "notes": "Displayed exactly as 76.53. Locked official InclusionAI observation.", "candidates": [ { "score": 76.83, "reference_url": "https://arxiv.org/pdf/2606.15079v1", "source_type": "official_paper", "reported_setting": { "checkpoint": "base", "mode": "base pretraining checkpoint", "effort": "not applicable", "shots": "5-shot", "samples": 1, "trials": 1, "aggregation": "EM, 5-shot", "tools": "none", "search": "none", "context_management": "none stated", "judge": "benchmark-specified", "harness": "official InclusionAI / benchmark-specific", "dataset_version": "C-SimpleQA", "split": "source does not state", "context_length": "262,144 in locked base config", "multimodal_inputs": false, "temperature": "source does not state", "top_p": "source does not state", "top_k": "source does not state", "prompt_style": "source does not state", "max_new_tokens": "source does not state", "timeout": "source does not state", "output_parser": "source does not state", "evaluation_provenance": "official InclusionAI evaluation or locked provider rendering" }, "notes": "Displayed exactly as 76.83. Locked official InclusionAI observation. Base checkpoint evidence remains attached to the canonical post-trained model identity." } ] }, { "model_id": "ling-2.6-1t", "benchmark_id": "gpqa_diamond", "score": 76.17, "reference_url": "https://arxiv.org/pdf/2606.15079v1", "reported_setting": { "checkpoint": "post-trained", "mode": "instant/fast-thinking", "effort": "default", "shots": "source does not state", "samples": 4, "trials": 4, "aggregation": "mean@4", "tools": "none", "search": "none", "context_management": "none stated", "judge": "benchmark-specified", "harness": "official InclusionAI / benchmark-specific", "dataset_version": "GPQA-Diamond", "split": "source does not state", "context_length": "262,144 native config", "multimodal_inputs": false, "temperature": "source does not state", "top_p": "source does not state", "top_k": "source does not state", "prompt_style": "source does not state", "max_new_tokens": "source does not state", "timeout": "source does not state", "output_parser": "source does not state", "evaluation_provenance": "official InclusionAI evaluation or locked provider rendering" }, "matches_canonical": true, "source_type": "official_paper", "audit_status": "verified", "notes": "Displayed exactly as 76.17. Locked official InclusionAI observation." }, { "model_id": "ling-2.6-1t", "benchmark_id": "hmmt_nov_2025", "score": 81.93, "reference_url": "https://arxiv.org/pdf/2606.15079v1", "reported_setting": { "checkpoint": "post-trained", "mode": "instant/fast-thinking", "effort": "default", "shots": "source does not state", "samples": 64, "trials": 64, "aggregation": "mean@64", "tools": "none", "search": "none", "context_management": "none stated", "judge": "benchmark-specified", "harness": "official InclusionAI / benchmark-specific", "dataset_version": "HMMT November 2025", "split": "source does not state", "context_length": "262,144 native config", "multimodal_inputs": false, "temperature": "source does not state", "top_p": "source does not state", "top_k": "source does not state", "prompt_style": "source does not state", "max_new_tokens": "source does not state", "timeout": "source does not state", "output_parser": "source does not state", "evaluation_provenance": "official InclusionAI evaluation or locked provider rendering" }, "matches_canonical": true, "source_type": "official_paper", "audit_status": "verified", "notes": "Displayed exactly as 81.93. Locked official InclusionAI observation." }, { "model_id": "ling-2.6-1t", "benchmark_id": "ifbench", "score": 57.62, "reference_url": "https://arxiv.org/pdf/2606.15079v1", "reported_setting": { "checkpoint": "post-trained", "mode": "instant/fast-thinking", "effort": "default", "shots": "source does not state", "samples": "source does not state", "trials": "source does not state", "aggregation": "source does not state", "tools": "none", "search": "none", "context_management": "none stated", "judge": "benchmark-specified", "harness": "official InclusionAI / benchmark-specific", "dataset_version": "IFBench", "split": "source does not state", "context_length": "262,144 native config", "multimodal_inputs": false, "temperature": "source does not state", "top_p": "source does not state", "top_k": "source does not state", "prompt_style": "source does not state", "max_new_tokens": "source does not state", "timeout": "source does not state", "output_parser": "source does not state", "evaluation_provenance": "official InclusionAI evaluation or locked provider rendering" }, "matches_canonical": true, "source_type": "official_paper", "audit_status": "verified", "notes": "Displayed exactly as 57.62. Locked official InclusionAI observation." }, { "model_id": "ling-2.6-1t", "benchmark_id": "imo_answerbench", "score": 65.81, "reference_url": "https://arxiv.org/pdf/2606.15079v1", "reported_setting": { "checkpoint": "post-trained", "mode": "instant/fast-thinking", "effort": "default", "shots": "source does not state", "samples": 8, "trials": 8, "aggregation": "mean@8", "tools": "none", "search": "none", "context_management": "none stated", "judge": "benchmark-specified", "harness": "official InclusionAI / benchmark-specific", "dataset_version": "IMO-AnswerBench", "split": "source does not state", "context_length": "262,144 native config", "multimodal_inputs": false, "temperature": "source does not state", "top_p": "source does not state", "top_k": "source does not state", "prompt_style": "source does not state", "max_new_tokens": "source does not state", "timeout": "source does not state", "output_parser": "source does not state", "evaluation_provenance": "official InclusionAI evaluation or locked provider rendering" }, "matches_canonical": true, "source_type": "official_paper", "audit_status": "verified", "notes": "Displayed exactly as 65.81. Locked official InclusionAI observation." }, { "model_id": "ling-2.6-1t", "benchmark_id": "livecodebench_v6_2408_2505", "score": 65.58, "reference_url": "https://arxiv.org/pdf/2606.15079v1", "reported_setting": { "checkpoint": "post-trained", "mode": "instant/fast-thinking", "effort": "default", "shots": "source does not state", "samples": 4, "trials": 4, "aggregation": "mean@4", "tools": "code execution for generated solutions", "search": "none", "context_management": "none stated", "judge": "official test execution", "harness": "official InclusionAI / benchmark-specific", "dataset_version": "LiveCodeBench v6, 2024-08 to 2025-05", "split": "454 code-generation problems", "context_length": "262,144 native config", "multimodal_inputs": false, "temperature": "source does not state", "top_p": "source does not state", "top_k": "source does not state", "prompt_style": "source does not state", "max_new_tokens": "source does not state", "timeout": "source does not state", "output_parser": "source does not state", "evaluation_provenance": "official InclusionAI evaluation or locked provider rendering" }, "matches_canonical": true, "source_type": "official_paper", "audit_status": "verified", "notes": "Displayed exactly as 65.58. Locked official InclusionAI observation.", "candidates": [ { "score": 44.27, "reference_url": "https://arxiv.org/pdf/2606.15079v1", "source_type": "official_paper", "reported_setting": { "checkpoint": "base", "mode": "base pretraining checkpoint", "effort": "not applicable", "shots": "source does not state", "samples": 1, "trials": 1, "aggregation": "pass@1", "tools": "code execution for generated solutions", "search": "none", "context_management": "none stated", "judge": "official test execution", "harness": "official InclusionAI / benchmark-specific", "dataset_version": "LiveCodeBench v6, 2024-08 to 2025-05", "split": "454 code-generation problems", "context_length": "262,144 in locked base config", "multimodal_inputs": false, "temperature": "source does not state", "top_p": "source does not state", "top_k": "source does not state", "prompt_style": "source does not state", "max_new_tokens": "source does not state", "timeout": "source does not state", "output_parser": "source does not state", "evaluation_provenance": "official InclusionAI evaluation or locked provider rendering" }, "notes": "Displayed exactly as 44.27. Locked official InclusionAI observation. Base checkpoint evidence remains attached to the canonical post-trained model identity." } ] }, { "model_id": "ling-2.6-1t", "benchmark_id": "longbench_v2", "score": 48.31, "reference_url": "https://arxiv.org/pdf/2606.15079v1", "reported_setting": { "checkpoint": "post-trained", "mode": "instant/fast-thinking", "effort": "default", "shots": "source does not state", "samples": "source does not state", "trials": "source does not state", "aggregation": "source does not state", "tools": "none", "search": "none", "context_management": "none stated", "judge": "benchmark-specified", "harness": "official InclusionAI / benchmark-specific", "dataset_version": "LongBenchV2", "split": "source does not state", "context_length": "262,144 native config", "multimodal_inputs": false, "temperature": "source does not state", "top_p": "source does not state", "top_k": "source does not state", "prompt_style": "source does not state", "max_new_tokens": "source does not state", "timeout": "source does not state", "output_parser": "source does not state", "evaluation_provenance": "official InclusionAI evaluation or locked provider rendering" }, "matches_canonical": true, "source_type": "official_paper", "audit_status": "verified", "notes": "Displayed exactly as 48.31. Locked official InclusionAI observation.", "candidates": [ { "score": 43.54, "reference_url": "https://arxiv.org/pdf/2606.15079v1", "source_type": "official_paper", "reported_setting": { "checkpoint": "base", "mode": "base pretraining checkpoint", "effort": "not applicable", "shots": "0-shot", "samples": 1, "trials": 1, "aggregation": "Accuracy, 0-shot", "tools": "none", "search": "none", "context_management": "none stated", "judge": "benchmark-specified", "harness": "official InclusionAI / benchmark-specific", "dataset_version": "LongBenchv2", "split": "source does not state", "context_length": "262,144 in locked base config", "multimodal_inputs": false, "temperature": "source does not state", "top_p": "source does not state", "top_k": "source does not state", "prompt_style": "source does not state", "max_new_tokens": "source does not state", "timeout": "source does not state", "output_parser": "source does not state", "evaluation_provenance": "official InclusionAI evaluation or locked provider rendering" }, "notes": "Displayed exactly as 43.54. Locked official InclusionAI observation. Base checkpoint evidence remains attached to the canonical post-trained model identity." } ] }, { "model_id": "ling-2.6-1t", "benchmark_id": "simpleqa_verified", "score": 31.5, "reference_url": "https://arxiv.org/pdf/2606.15079v1", "reported_setting": { "checkpoint": "post-trained", "mode": "instant/fast-thinking", "effort": "default", "shots": "source does not state", "samples": 1, "trials": 1, "aggregation": "Correct", "tools": "none", "search": "none", "context_management": "none stated", "judge": "benchmark-specified", "harness": "official InclusionAI / benchmark-specific", "dataset_version": "SimpleQA-Verified", "split": "source does not state", "context_length": "262,144 native config", "multimodal_inputs": false, "temperature": "source does not state", "top_p": "source does not state", "top_k": "source does not state", "prompt_style": "source does not state", "max_new_tokens": "source does not state", "timeout": "source does not state", "output_parser": "source does not state", "evaluation_provenance": "official InclusionAI evaluation or locked provider rendering" }, "matches_canonical": true, "source_type": "official_paper", "audit_status": "verified", "notes": "Displayed exactly as 31.50. Locked official InclusionAI observation." }, { "model_id": "ling-2.6-1t", "benchmark_id": "supergpqa", "score": 58.32, "reference_url": "https://arxiv.org/pdf/2606.15079v1", "reported_setting": { "checkpoint": "post-trained", "mode": "instant/fast-thinking", "effort": "default", "shots": "source does not state", "samples": 1, "trials": 1, "aggregation": "EM, CoT", "tools": "none", "search": "none", "context_management": "none stated", "judge": "benchmark-specified", "harness": "official InclusionAI / benchmark-specific", "dataset_version": "SuperGPQA", "split": "source does not state", "context_length": "262,144 native config", "multimodal_inputs": false, "temperature": "source does not state", "top_p": "source does not state", "top_k": "source does not state", "prompt_style": "source does not state", "max_new_tokens": "source does not state", "timeout": "source does not state", "output_parser": "source does not state", "evaluation_provenance": "official InclusionAI evaluation or locked provider rendering" }, "matches_canonical": true, "source_type": "official_paper", "audit_status": "verified", "notes": "Displayed exactly as 58.32. Locked official InclusionAI observation.", "candidates": [ { "score": 44.72, "reference_url": "https://arxiv.org/pdf/2606.15079v1", "source_type": "official_paper", "reported_setting": { "checkpoint": "base", "mode": "base pretraining checkpoint", "effort": "not applicable", "shots": "5-shot", "samples": 1, "trials": 1, "aggregation": "EM, 5-shot", "tools": "none", "search": "none", "context_management": "none stated", "judge": "benchmark-specified", "harness": "official InclusionAI / benchmark-specific", "dataset_version": "SuperGPQA", "split": "source does not state", "context_length": "262,144 in locked base config", "multimodal_inputs": false, "temperature": "source does not state", "top_p": "source does not state", "top_k": "source does not state", "prompt_style": "source does not state", "max_new_tokens": "source does not state", "timeout": "source does not state", "output_parser": "source does not state", "evaluation_provenance": "official InclusionAI evaluation or locked provider rendering" }, "notes": "Displayed exactly as 44.72. Locked official InclusionAI observation. Base checkpoint evidence remains attached to the canonical post-trained model identity." } ] }, { "model_id": "ling-2.6-1t", "benchmark_id": "swe_bench_verified", "score": 72.2, "reference_url": "https://huggingface.co/inclusionAI/Ling-2.6-1T/blob/07b9e26bb1f97494b4fd62eed83f22722ffd560a/.eval_results/swe-bench_verified.yaml", "reported_setting": { "checkpoint": "post-trained", "mode": "instant/fast-thinking", "effort": "default", "shots": "source does not state", "samples": 1, "trials": 1, "aggregation": "% resolved (pass@1)", "tools": "coding-agent shell/editor tools", "search": "none", "context_management": "none stated", "judge": "official repository test suite", "harness": "source does not state", "dataset_version": "SWE-bench Verified", "split": "Verified 500", "context_length": "262,144 native config", "multimodal_inputs": false, "temperature": "source does not state", "top_p": "source does not state", "top_k": "source does not state", "prompt_style": "source does not state", "max_new_tokens": "source does not state", "timeout": "source does not state", "output_parser": "source does not state", "evaluation_provenance": "official InclusionAI evaluation or locked provider rendering" }, "matches_canonical": true, "source_type": "official_model_card", "audit_status": "verified", "notes": "Displayed exactly as 72.2. Locked official InclusionAI observation." }, { "model_id": "ling-2.6-1t", "benchmark_id": "tau2_bench_avg", "score": 78.36, "reference_url": "https://arxiv.org/pdf/2606.15079v1", "reported_setting": { "checkpoint": "post-trained", "mode": "instant/fast-thinking", "effort": "default", "shots": "source does not state", "samples": "source does not state", "trials": "source does not state", "aggregation": "source-reported average", "tools": "official tau2 domain tools", "search": "none", "context_management": "none stated", "judge": "task success", "harness": "official tau2 harness with source-specific user simulator", "dataset_version": "tau2 v1.0.0 average", "split": "source does not state", "context_length": "262,144 native config", "multimodal_inputs": false, "temperature": "source does not state", "top_p": "source does not state", "top_k": "source does not state", "prompt_style": "source does not state", "max_new_tokens": "source does not state", "timeout": "source does not state", "output_parser": "source does not state", "evaluation_provenance": "official InclusionAI evaluation or locked provider rendering" }, "matches_canonical": true, "source_type": "official_paper", "audit_status": "verified", "notes": "Displayed exactly as 78.36. Locked official InclusionAI observation." }, { "model_id": "ling-2.6-1t", "benchmark_id": "terminal_bench", "score": 40.45, "reference_url": "https://arxiv.org/pdf/2606.15079v1", "reported_setting": { "checkpoint": "post-trained", "mode": "instant/fast-thinking", "effort": "default", "shots": "source does not state", "samples": 1, "trials": 1, "aggregation": "Accuracy", "tools": "terminal", "search": "none", "context_management": "none stated", "judge": "programmatic task tests", "harness": "source-specified terminal agent", "dataset_version": "Terminal-Bench 2.0, 74 tasks", "split": "source does not state", "context_length": "262,144 native config", "multimodal_inputs": false, "temperature": "source does not state", "top_p": "source does not state", "top_k": "source does not state", "prompt_style": "source does not state", "max_new_tokens": "source does not state", "timeout": "source does not state", "output_parser": "source does not state", "evaluation_provenance": "official InclusionAI evaluation or locked provider rendering" }, "matches_canonical": true, "source_type": "official_paper", "audit_status": "verified", "notes": "Displayed exactly as 40.45. Locked official InclusionAI observation." }, { "model_id": "ling-2.6-flash", "benchmark_id": "aime_2026", "score": 73.85, "reference_url": "https://huggingface.co/inclusionAI/Ling-2.6-flash/blob/11236968749136a78d4b7cbfb786c4460f2f353e/.eval_results/aime_2026.yaml", "reported_setting": { "checkpoint": "post-trained", "mode": "instant/non-reasoning", "effort": "default", "shots": "source does not state", "samples": 64, "trials": 64, "aggregation": "mean@64", "tools": "none", "search": "none", "context_management": "none stated", "judge": "benchmark-specified", "harness": "official InclusionAI / benchmark-specific", "dataset_version": "AIME 2026", "split": "source does not state", "context_length": "131,072 native config", "multimodal_inputs": false, "temperature": "source does not state", "top_p": "source does not state", "top_k": "source does not state", "prompt_style": "source does not state", "max_new_tokens": "source does not state", "timeout": "source does not state", "output_parser": "source does not state", "evaluation_provenance": "official InclusionAI evaluation or locked provider rendering" }, "matches_canonical": true, "source_type": "official_model_card", "audit_status": "verified", "notes": "Displayed exactly as 73.85. Locked official InclusionAI observation.", "candidates": [ { "score": 73.65, "reference_url": "https://raw.githubusercontent.com/inclusionAI/LLaDA2.X/f402f0b52817e1c3586a024ebf89e69ad8ca5523/LLaDA2_2_tech_report.pdf", "source_type": "official_paper", "reported_setting": { "checkpoint": "post-trained", "mode": "instant/non-reasoning", "effort": "default", "shots": "source does not state", "samples": "source does not state", "trials": "source does not state", "aggregation": "source does not state", "tools": "none", "search": "none", "context_management": "none stated", "judge": "benchmark-specified", "harness": "official InclusionAI / benchmark-specific", "dataset_version": "AIME 2026", "split": "source does not state", "context_length": "131,072 native config", "multimodal_inputs": false, "temperature": "source does not state", "top_p": "source does not state", "top_k": "source does not state", "prompt_style": "source does not state", "max_new_tokens": "source does not state", "timeout": "source does not state", "output_parser": "source does not state", "evaluation_provenance": "official InclusionAI evaluation or locked provider rendering" }, "notes": "Displayed exactly as 73.65. Locked official InclusionAI observation." } ] }, { "model_id": "ling-2.6-flash", "benchmark_id": "bfcl_v3", "score": 76.05, "reference_url": "https://raw.githubusercontent.com/inclusionAI/LLaDA2.X/f402f0b52817e1c3586a024ebf89e69ad8ca5523/LLaDA2_2_tech_report.pdf", "reported_setting": { "checkpoint": "post-trained", "mode": "instant/non-reasoning", "effort": "default", "shots": "source does not state", "samples": "source does not state", "trials": "source does not state", "aggregation": "source does not state", "tools": "benchmark-provided functions", "search": "none", "context_management": "none stated", "judge": "official function-call match", "harness": "Berkeley Function Calling Leaderboard", "dataset_version": "BFCL v3", "split": "source does not state", "context_length": "131,072 native config", "multimodal_inputs": false, "temperature": "source does not state", "top_p": "source does not state", "top_k": "source does not state", "prompt_style": "source does not state", "max_new_tokens": "source does not state", "timeout": "source does not state", "output_parser": "source does not state", "evaluation_provenance": "official InclusionAI evaluation or locked provider rendering" }, "matches_canonical": true, "source_type": "official_paper", "audit_status": "verified", "notes": "Displayed exactly as 76.05. Locked official InclusionAI observation." }, { "model_id": "ling-2.6-flash", "benchmark_id": "bfcl_v4", "score": 66.81, "reference_url": "https://arxiv.org/pdf/2606.15079v1", "reported_setting": { "checkpoint": "post-trained", "mode": "instant/non-reasoning", "effort": "default", "shots": "source does not state", "samples": 1, "trials": 1, "aggregation": "Accuracy", "tools": "benchmark-provided functions", "search": "none", "context_management": "none stated", "judge": "official function-call match", "harness": "Berkeley Function Calling Leaderboard", "dataset_version": "BFCL v4", "split": "source does not state", "context_length": "131,072 native config", "multimodal_inputs": false, "temperature": "source does not state", "top_p": "source does not state", "top_k": "source does not state", "prompt_style": "source does not state", "max_new_tokens": "source does not state", "timeout": "source does not state", "output_parser": "source does not state", "evaluation_provenance": "official InclusionAI evaluation or locked provider rendering" }, "matches_canonical": true, "source_type": "official_paper", "audit_status": "verified", "notes": "Displayed exactly as 66.81. Locked official InclusionAI observation." }, { "model_id": "ling-2.6-flash", "benchmark_id": "chinese_simpleqa", "score": 60.23, "reference_url": "https://arxiv.org/pdf/2606.15079v1", "reported_setting": { "checkpoint": "post-trained", "mode": "instant/non-reasoning", "effort": "default", "shots": "source does not state", "samples": 1, "trials": 1, "aggregation": "Correct", "tools": "none", "search": "none", "context_management": "none stated", "judge": "benchmark-specified", "harness": "official InclusionAI / benchmark-specific", "dataset_version": "C-SimpleQA", "split": "source does not state", "context_length": "131,072 native config", "multimodal_inputs": false, "temperature": "source does not state", "top_p": "source does not state", "top_k": "source does not state", "prompt_style": "source does not state", "max_new_tokens": "source does not state", "timeout": "source does not state", "output_parser": "source does not state", "evaluation_provenance": "official InclusionAI evaluation or locked provider rendering" }, "matches_canonical": true, "source_type": "official_paper", "audit_status": "verified", "notes": "Displayed exactly as 60.23. Locked official InclusionAI observation.", "candidates": [ { "score": 63.53, "reference_url": "https://arxiv.org/pdf/2606.15079v1", "source_type": "official_paper", "reported_setting": { "checkpoint": "base", "mode": "base pretraining checkpoint", "effort": "not applicable", "shots": "5-shot", "samples": 1, "trials": 1, "aggregation": "EM, 5-shot", "tools": "none", "search": "none", "context_management": "none stated", "judge": "benchmark-specified", "harness": "official InclusionAI / benchmark-specific", "dataset_version": "C-SimpleQA", "split": "source does not state", "context_length": "262,144 in locked base config", "multimodal_inputs": false, "temperature": "source does not state", "top_p": "source does not state", "top_k": "source does not state", "prompt_style": "source does not state", "max_new_tokens": "source does not state", "timeout": "source does not state", "output_parser": "source does not state", "evaluation_provenance": "official InclusionAI evaluation or locked provider rendering" }, "notes": "Displayed exactly as 63.53. Locked official InclusionAI observation. Base checkpoint evidence remains attached to the canonical post-trained model identity." } ] }, { "model_id": "ling-2.6-flash", "benchmark_id": "gpqa_diamond", "score": 60.35, "reference_url": "https://raw.githubusercontent.com/inclusionAI/LLaDA2.X/f402f0b52817e1c3586a024ebf89e69ad8ca5523/LLaDA2_2_tech_report.pdf", "reported_setting": { "checkpoint": "post-trained", "mode": "instant/non-reasoning", "effort": "default", "shots": "source does not state", "samples": "source does not state", "trials": "source does not state", "aggregation": "source does not state", "tools": "none", "search": "none", "context_management": "none stated", "judge": "benchmark-specified", "harness": "official InclusionAI / benchmark-specific", "dataset_version": "GPQA-Diamond", "split": "source does not state", "context_length": "131,072 native config", "multimodal_inputs": false, "temperature": "source does not state", "top_p": "source does not state", "top_k": "source does not state", "prompt_style": "source does not state", "max_new_tokens": "source does not state", "timeout": "source does not state", "output_parser": "source does not state", "evaluation_provenance": "official InclusionAI evaluation or locked provider rendering" }, "matches_canonical": true, "source_type": "official_paper", "audit_status": "verified", "notes": "Displayed exactly as 60.35. Locked official InclusionAI observation." }, { "model_id": "ling-2.6-flash", "benchmark_id": "hmmt_feb_2026", "score": 49.29, "reference_url": "https://huggingface.co/inclusionAI/Ling-2.6-flash/blob/11236968749136a78d4b7cbfb786c4460f2f353e/.eval_results/hmmt_feb_2026.yaml", "reported_setting": { "checkpoint": "post-trained", "mode": "instant/non-reasoning", "effort": "default", "shots": "source does not state", "samples": 64, "trials": 64, "aggregation": "mean@64", "tools": "none", "search": "none", "context_management": "none stated", "judge": "benchmark-specified", "harness": "official InclusionAI / benchmark-specific", "dataset_version": "HMMT February 2026", "split": "source does not state", "context_length": "131,072 native config", "multimodal_inputs": false, "temperature": "source does not state", "top_p": "source does not state", "top_k": "source does not state", "prompt_style": "source does not state", "max_new_tokens": "source does not state", "timeout": "source does not state", "output_parser": "source does not state", "evaluation_provenance": "official InclusionAI evaluation or locked provider rendering" }, "matches_canonical": true, "source_type": "official_model_card", "audit_status": "verified", "notes": "Displayed exactly as 49.29. Locked official InclusionAI observation." }, { "model_id": "ling-2.6-flash", "benchmark_id": "ifbench", "score": 57.4, "reference_url": "https://arxiv.org/pdf/2606.15079v1", "reported_setting": { "checkpoint": "post-trained", "mode": "instant/non-reasoning", "effort": "default", "shots": "source does not state", "samples": 5, "trials": 5, "aggregation": "mean@5", "tools": "none", "search": "none", "context_management": "none stated", "judge": "benchmark-specified", "harness": "official InclusionAI / benchmark-specific", "dataset_version": "IFBench", "split": "source does not state", "context_length": "131,072 native config", "multimodal_inputs": false, "temperature": "source does not state", "top_p": "source does not state", "top_k": "source does not state", "prompt_style": "source does not state", "max_new_tokens": "source does not state", "timeout": "source does not state", "output_parser": "source does not state", "evaluation_provenance": "official InclusionAI evaluation or locked provider rendering" }, "matches_canonical": true, "source_type": "official_paper", "audit_status": "verified", "notes": "Displayed exactly as 57.40. Locked official InclusionAI observation." }, { "model_id": "ling-2.6-flash", "benchmark_id": "imo_answerbench", "score": 54.28, "reference_url": "https://arxiv.org/pdf/2606.15079v1", "reported_setting": { "checkpoint": "post-trained", "mode": "instant/non-reasoning", "effort": "default", "shots": "source does not state", "samples": 8, "trials": 8, "aggregation": "mean@8", "tools": "none", "search": "none", "context_management": "none stated", "judge": "benchmark-specified", "harness": "official InclusionAI / benchmark-specific", "dataset_version": "IMO-AnswerBench", "split": "source does not state", "context_length": "131,072 native config", "multimodal_inputs": false, "temperature": "source does not state", "top_p": "source does not state", "top_k": "source does not state", "prompt_style": "source does not state", "max_new_tokens": "source does not state", "timeout": "source does not state", "output_parser": "source does not state", "evaluation_provenance": "official InclusionAI evaluation or locked provider rendering" }, "matches_canonical": true, "source_type": "official_paper", "audit_status": "verified", "notes": "Displayed exactly as 54.28. Locked official InclusionAI observation." }, { "model_id": "ling-2.6-flash", "benchmark_id": "korbench", "score": 66.48, "reference_url": "https://raw.githubusercontent.com/inclusionAI/LLaDA2.X/f402f0b52817e1c3586a024ebf89e69ad8ca5523/LLaDA2_2_tech_report.pdf", "reported_setting": { "checkpoint": "post-trained", "mode": "instant/non-reasoning", "effort": "default", "shots": "source does not state", "samples": "source does not state", "trials": "source does not state", "aggregation": "source does not state", "tools": "none", "search": "none", "context_management": "none stated", "judge": "benchmark-specified", "harness": "official InclusionAI / benchmark-specific", "dataset_version": "KORBench", "split": "source does not state", "context_length": "131,072 native config", "multimodal_inputs": false, "temperature": "source does not state", "top_p": "source does not state", "top_k": "source does not state", "prompt_style": "source does not state", "max_new_tokens": "source does not state", "timeout": "source does not state", "output_parser": "source does not state", "evaluation_provenance": "official InclusionAI evaluation or locked provider rendering" }, "matches_canonical": true, "source_type": "official_paper", "audit_status": "verified", "notes": "Displayed exactly as 66.48. Locked official InclusionAI observation.", "candidates": [ { "score": 44.96, "reference_url": "https://arxiv.org/pdf/2606.15079v1", "source_type": "official_paper", "reported_setting": { "checkpoint": "base", "mode": "base pretraining checkpoint", "effort": "not applicable", "shots": "source does not state", "samples": 1, "trials": 1, "aggregation": "Accuracy", "tools": "none", "search": "none", "context_management": "none stated", "judge": "benchmark-specified", "harness": "official InclusionAI / benchmark-specific", "dataset_version": "KOR-Bench", "split": "source does not state", "context_length": "262,144 in locked base config", "multimodal_inputs": false, "temperature": "source does not state", "top_p": "source does not state", "top_k": "source does not state", "prompt_style": "source does not state", "max_new_tokens": "source does not state", "timeout": "source does not state", "output_parser": "source does not state", "evaluation_provenance": "official InclusionAI evaluation or locked provider rendering" }, "notes": "Displayed exactly as 44.96. Locked official InclusionAI observation. Base checkpoint evidence remains attached to the canonical post-trained model identity." } ] }, { "model_id": "ling-2.6-flash", "benchmark_id": "livecodebench_v6_2408_2505", "score": 62.28, "reference_url": "https://arxiv.org/pdf/2606.15079v1", "reported_setting": { "checkpoint": "post-trained", "mode": "instant/non-reasoning", "effort": "default", "shots": "source does not state", "samples": 4, "trials": 4, "aggregation": "mean@4", "tools": "code execution for generated solutions", "search": "none", "context_management": "none stated", "judge": "official test execution", "harness": "official InclusionAI / benchmark-specific", "dataset_version": "LiveCodeBench v6, 2024-08 to 2025-05", "split": "454 code-generation problems", "context_length": "131,072 native config", "multimodal_inputs": false, "temperature": "source does not state", "top_p": "source does not state", "top_k": "source does not state", "prompt_style": "source does not state", "max_new_tokens": "source does not state", "timeout": "source does not state", "output_parser": "source does not state", "evaluation_provenance": "official InclusionAI evaluation or locked provider rendering" }, "matches_canonical": true, "source_type": "official_paper", "audit_status": "verified", "notes": "Displayed exactly as 62.28. Locked official InclusionAI observation.", "candidates": [ { "score": 33.48, "reference_url": "https://arxiv.org/pdf/2606.15079v1", "source_type": "official_paper", "reported_setting": { "checkpoint": "base", "mode": "base pretraining checkpoint", "effort": "not applicable", "shots": "source does not state", "samples": 1, "trials": 1, "aggregation": "pass@1", "tools": "code execution for generated solutions", "search": "none", "context_management": "none stated", "judge": "official test execution", "harness": "official InclusionAI / benchmark-specific", "dataset_version": "LiveCodeBench v6, 2024-08 to 2025-05", "split": "454 code-generation problems", "context_length": "262,144 in locked base config", "multimodal_inputs": false, "temperature": "source does not state", "top_p": "source does not state", "top_k": "source does not state", "prompt_style": "source does not state", "max_new_tokens": "source does not state", "timeout": "source does not state", "output_parser": "source does not state", "evaluation_provenance": "official InclusionAI evaluation or locked provider rendering" }, "notes": "Displayed exactly as 33.48. Locked official InclusionAI observation. Base checkpoint evidence remains attached to the canonical post-trained model identity." } ] }, { "model_id": "ling-2.6-flash", "benchmark_id": "longbench_v2", "score": 42.94, "reference_url": "https://raw.githubusercontent.com/inclusionAI/LLaDA2.X/f402f0b52817e1c3586a024ebf89e69ad8ca5523/LLaDA2_2_tech_report.pdf", "reported_setting": { "checkpoint": "post-trained", "mode": "instant/non-reasoning", "effort": "default", "shots": "source does not state", "samples": "source does not state", "trials": "source does not state", "aggregation": "source does not state", "tools": "none", "search": "none", "context_management": "none stated", "judge": "benchmark-specified", "harness": "official InclusionAI / benchmark-specific", "dataset_version": "LongBench v2", "split": "source does not state", "context_length": "131,072 native config", "multimodal_inputs": false, "temperature": "source does not state", "top_p": "source does not state", "top_k": "source does not state", "prompt_style": "source does not state", "max_new_tokens": "source does not state", "timeout": "source does not state", "output_parser": "source does not state", "evaluation_provenance": "official InclusionAI evaluation or locked provider rendering" }, "matches_canonical": true, "source_type": "official_paper", "audit_status": "verified", "notes": "Displayed exactly as 42.94. Locked official InclusionAI observation.", "candidates": [ { "score": 34.19, "reference_url": "https://arxiv.org/pdf/2606.15079v1", "source_type": "official_paper", "reported_setting": { "checkpoint": "base", "mode": "base pretraining checkpoint", "effort": "not applicable", "shots": "0-shot", "samples": 1, "trials": 1, "aggregation": "Accuracy, 0-shot", "tools": "none", "search": "none", "context_management": "none stated", "judge": "benchmark-specified", "harness": "official InclusionAI / benchmark-specific", "dataset_version": "LongBenchv2", "split": "source does not state", "context_length": "262,144 in locked base config", "multimodal_inputs": false, "temperature": "source does not state", "top_p": "source does not state", "top_k": "source does not state", "prompt_style": "source does not state", "max_new_tokens": "source does not state", "timeout": "source does not state", "output_parser": "source does not state", "evaluation_provenance": "official InclusionAI evaluation or locked provider rendering" }, "notes": "Displayed exactly as 34.19. Locked official InclusionAI observation. Base checkpoint evidence remains attached to the canonical post-trained model identity." } ] }, { "model_id": "ling-2.6-flash", "benchmark_id": "mcpatlas", "score": 41.12, "reference_url": "https://raw.githubusercontent.com/inclusionAI/LLaDA2.X/f402f0b52817e1c3586a024ebf89e69ad8ca5523/LLaDA2_2_tech_report.pdf", "reported_setting": { "checkpoint": "post-trained", "mode": "instant/non-reasoning", "effort": "default", "shots": "source does not state", "samples": 1, "trials": 5, "aggregation": "arithmetic mean of 5 runs", "tools": "36 MCP servers / public tool suite", "search": "none", "context_management": "none stated", "judge": "Gemini-2.5-Pro claim coverage; pass at >=0.75", "harness": "official MCP-Atlas v1, 20-turn limit", "dataset_version": "MCP-Atlas", "split": "public 500 tasks", "context_length": "131,072 native config", "multimodal_inputs": false, "temperature": "source does not state", "top_p": "source does not state", "top_k": "source does not state", "prompt_style": "source does not state", "max_new_tokens": "source does not state", "timeout": "source does not state", "output_parser": "source does not state", "evaluation_provenance": "official InclusionAI evaluation or locked provider rendering" }, "matches_canonical": true, "source_type": "official_paper", "audit_status": "verified", "notes": "Displayed exactly as 41.12. Locked official InclusionAI observation." }, { "model_id": "ling-2.6-flash", "benchmark_id": "multi_if", "score": 74.8, "reference_url": "https://arxiv.org/pdf/2606.15079v1", "reported_setting": { "checkpoint": "post-trained", "mode": "instant/non-reasoning", "effort": "default", "shots": "source does not state", "samples": 1, "trials": 1, "aggregation": "turn-3", "tools": "none", "search": "none", "context_management": "none stated", "judge": "benchmark-specified", "harness": "official InclusionAI / benchmark-specific", "dataset_version": "Multi-IF turn-3", "split": "source does not state", "context_length": "131,072 native config", "multimodal_inputs": false, "temperature": "source does not state", "top_p": "source does not state", "top_k": "source does not state", "prompt_style": "source does not state", "max_new_tokens": "source does not state", "timeout": "source does not state", "output_parser": "source does not state", "evaluation_provenance": "official InclusionAI evaluation or locked provider rendering" }, "matches_canonical": true, "source_type": "official_paper", "audit_status": "verified", "notes": "Displayed exactly as 74.80. Locked official InclusionAI observation." }, { "model_id": "ling-2.6-flash", "benchmark_id": "multichallenge", "score": 39.71, "reference_url": "https://arxiv.org/pdf/2606.15079v1", "reported_setting": { "checkpoint": "post-trained", "mode": "instant/non-reasoning", "effort": "default", "shots": "source does not state", "samples": 1, "trials": 1, "aggregation": "Accuracy", "tools": "none", "search": "none", "context_management": "none stated", "judge": "benchmark-specified", "harness": "official InclusionAI / benchmark-specific", "dataset_version": "Multichallenge", "split": "source does not state", "context_length": "131,072 native config", "multimodal_inputs": false, "temperature": "source does not state", "top_p": "source does not state", "top_k": "source does not state", "prompt_style": "source does not state", "max_new_tokens": "source does not state", "timeout": "source does not state", "output_parser": "source does not state", "evaluation_provenance": "official InclusionAI evaluation or locked provider rendering" }, "matches_canonical": true, "source_type": "official_paper", "audit_status": "verified", "notes": "Displayed exactly as 39.71. Locked official InclusionAI observation." }, { "model_id": "ling-2.6-flash", "benchmark_id": "simpleqa_verified", "score": 15.1, "reference_url": "https://arxiv.org/pdf/2606.15079v1", "reported_setting": { "checkpoint": "post-trained", "mode": "instant/non-reasoning", "effort": "default", "shots": "source does not state", "samples": 1, "trials": 1, "aggregation": "Correct", "tools": "none", "search": "none", "context_management": "none stated", "judge": "benchmark-specified", "harness": "official InclusionAI / benchmark-specific", "dataset_version": "SimpleQA-Verified", "split": "source does not state", "context_length": "131,072 native config", "multimodal_inputs": false, "temperature": "source does not state", "top_p": "source does not state", "top_k": "source does not state", "prompt_style": "source does not state", "max_new_tokens": "source does not state", "timeout": "source does not state", "output_parser": "source does not state", "evaluation_provenance": "official InclusionAI evaluation or locked provider rendering" }, "matches_canonical": true, "source_type": "official_paper", "audit_status": "verified", "notes": "Displayed exactly as 15.10. Locked official InclusionAI observation." }, { "model_id": "ling-2.6-flash", "benchmark_id": "swe_bench_multilingual", "score": 33.73, "reference_url": "https://raw.githubusercontent.com/inclusionAI/LLaDA2.X/f402f0b52817e1c3586a024ebf89e69ad8ca5523/LLaDA2_2_tech_report.pdf", "reported_setting": { "checkpoint": "post-trained", "mode": "instant/non-reasoning", "effort": "default", "shots": "source does not state", "samples": "source does not state", "trials": "source does not state", "aggregation": "source does not state", "tools": "coding-agent shell/editor tools", "search": "none", "context_management": "none stated", "judge": "official repository test suite", "harness": "Claude Code", "dataset_version": "SWE-bench Multilingual", "split": "Multilingual 300", "context_length": "131,072 native config", "multimodal_inputs": false, "temperature": "source does not state", "top_p": "source does not state", "top_k": "source does not state", "prompt_style": "source does not state", "max_new_tokens": "source does not state", "timeout": "source does not state", "output_parser": "source does not state", "evaluation_provenance": "official InclusionAI evaluation or locked provider rendering" }, "matches_canonical": true, "source_type": "official_paper", "audit_status": "verified", "notes": "Displayed exactly as 33.73. Locked official InclusionAI observation." }, { "model_id": "ling-2.6-flash", "benchmark_id": "swe_bench_pro", "score": 31.88, "reference_url": "https://raw.githubusercontent.com/inclusionAI/LLaDA2.X/f402f0b52817e1c3586a024ebf89e69ad8ca5523/LLaDA2_2_tech_report.pdf", "reported_setting": { "checkpoint": "post-trained", "mode": "instant/non-reasoning", "effort": "default", "shots": "source does not state", "samples": "source does not state", "trials": "source does not state", "aggregation": "source does not state", "tools": "coding-agent shell/editor tools", "search": "none", "context_management": "none stated", "judge": "official repository test suite", "harness": "Claude Code", "dataset_version": "SWE-bench Pro", "split": "public 731", "context_length": "131,072 native config", "multimodal_inputs": false, "temperature": "source does not state", "top_p": "source does not state", "top_k": "source does not state", "prompt_style": "source does not state", "max_new_tokens": "source does not state", "timeout": "source does not state", "output_parser": "source does not state", "evaluation_provenance": "official InclusionAI evaluation or locked provider rendering" }, "matches_canonical": true, "source_type": "official_paper", "audit_status": "verified", "notes": "Displayed exactly as 31.88. Locked official InclusionAI observation." }, { "model_id": "ling-2.6-flash", "benchmark_id": "swe_bench_verified", "score": 61.2, "reference_url": "https://huggingface.co/inclusionAI/Ling-2.6-flash/blob/11236968749136a78d4b7cbfb786c4460f2f353e/.eval_results/swe-bench_verified.yaml", "reported_setting": { "checkpoint": "post-trained", "mode": "instant/non-reasoning", "effort": "default", "shots": "source does not state", "samples": 1, "trials": 1, "aggregation": "% resolved (pass@1)", "tools": "coding-agent shell/editor tools", "search": "none", "context_management": "none stated", "judge": "official repository test suite", "harness": "OpenHands", "dataset_version": "SWE-bench Verified", "split": "Verified 500", "context_length": "131,072 native config", "multimodal_inputs": false, "temperature": "source does not state", "top_p": "source does not state", "top_k": "source does not state", "prompt_style": "source does not state", "max_new_tokens": "source does not state", "timeout": "source does not state", "output_parser": "source does not state", "evaluation_provenance": "official InclusionAI evaluation or locked provider rendering" }, "matches_canonical": true, "source_type": "official_model_card", "audit_status": "verified", "notes": "Displayed exactly as 61.2. Locked official InclusionAI observation." }, { "model_id": "ling-2.6-flash", "benchmark_id": "tau2_bench_airline", "score": 63.0, "reference_url": "https://mdn.alipayobjects.com/huamei_3p6pd0/afts/img/4bI1SK8pNM8AAAAAgBAAAAgADryCAQFr/original", "reported_setting": { "checkpoint": "post-trained", "mode": "instant/non-reasoning", "effort": "default", "shots": "source does not state", "samples": "source does not state", "trials": "source does not state", "aggregation": "source does not state", "tools": "official tau2 domain tools", "search": "none", "context_management": "none stated", "judge": "task success", "harness": "official tau2 harness with source-specific user simulator", "dataset_version": "tau2 v1.0.0 airline", "split": "source does not state", "context_length": "131,072 native config", "multimodal_inputs": false, "temperature": "source does not state", "top_p": "source does not state", "top_k": "source does not state", "prompt_style": "source does not state", "max_new_tokens": "source does not state", "timeout": "source does not state", "output_parser": "source does not state", "evaluation_provenance": "official InclusionAI evaluation or locked provider rendering" }, "matches_canonical": true, "source_type": "official_model_card", "audit_status": "verified", "notes": "Displayed exactly as 63.00. Locked official InclusionAI observation." }, { "model_id": "ling-2.6-flash", "benchmark_id": "tau2_bench_avg", "score": 76.36, "reference_url": "https://arxiv.org/pdf/2606.15079v1", "reported_setting": { "checkpoint": "post-trained", "mode": "instant/non-reasoning", "effort": "default", "shots": "source does not state", "samples": 4, "trials": 4, "aggregation": "mean@4", "tools": "official tau2 domain tools", "search": "none", "context_management": "none stated", "judge": "GPT-5.2 user simulator; task-success judge", "harness": "official tau2 harness with source-specific user simulator", "dataset_version": "tau2 v1.0.0 average", "split": "source does not state", "context_length": "131,072 native config", "multimodal_inputs": false, "temperature": "source does not state", "top_p": "source does not state", "top_k": "source does not state", "prompt_style": "source does not state", "max_new_tokens": "source does not state", "timeout": "source does not state", "output_parser": "source does not state", "evaluation_provenance": "official InclusionAI evaluation or locked provider rendering" }, "matches_canonical": true, "source_type": "official_paper", "audit_status": "verified", "notes": "Displayed exactly as 76.36. Locked official InclusionAI observation." }, { "model_id": "ling-2.6-flash", "benchmark_id": "tau2_bench_retail", "score": 71.49, "reference_url": "https://mdn.alipayobjects.com/huamei_3p6pd0/afts/img/4bI1SK8pNM8AAAAAgBAAAAgADryCAQFr/original", "reported_setting": { "checkpoint": "post-trained", "mode": "instant/non-reasoning", "effort": "default", "shots": "source does not state", "samples": "source does not state", "trials": "source does not state", "aggregation": "source does not state", "tools": "official tau2 domain tools", "search": "none", "context_management": "none stated", "judge": "task success", "harness": "official tau2 harness with source-specific user simulator", "dataset_version": "tau2 v1.0.0 retail", "split": "source does not state", "context_length": "131,072 native config", "multimodal_inputs": false, "temperature": "source does not state", "top_p": "source does not state", "top_k": "source does not state", "prompt_style": "source does not state", "max_new_tokens": "source does not state", "timeout": "source does not state", "output_parser": "source does not state", "evaluation_provenance": "official InclusionAI evaluation or locked provider rendering" }, "matches_canonical": true, "source_type": "official_model_card", "audit_status": "verified", "notes": "Displayed exactly as 71.49. Locked official InclusionAI observation." }, { "model_id": "ling-2.6-flash", "benchmark_id": "tau2_bench_telecom", "score": 94.96, "reference_url": "https://mdn.alipayobjects.com/huamei_3p6pd0/afts/img/4bI1SK8pNM8AAAAAgBAAAAgADryCAQFr/original", "reported_setting": { "checkpoint": "post-trained", "mode": "instant/non-reasoning", "effort": "default", "shots": "source does not state", "samples": "source does not state", "trials": "source does not state", "aggregation": "source does not state", "tools": "official tau2 domain tools", "search": "none", "context_management": "none stated", "judge": "task success", "harness": "official tau2 harness with source-specific user simulator", "dataset_version": "tau2 v1.0.0 telecom", "split": "source does not state", "context_length": "131,072 native config", "multimodal_inputs": false, "temperature": "source does not state", "top_p": "source does not state", "top_k": "source does not state", "prompt_style": "source does not state", "max_new_tokens": "source does not state", "timeout": "source does not state", "output_parser": "source does not state", "evaluation_provenance": "official InclusionAI evaluation or locked provider rendering" }, "matches_canonical": true, "source_type": "official_model_card", "audit_status": "verified", "notes": "Displayed exactly as 94.96. Locked official InclusionAI observation." }, { "model_id": "ling-3.0-flash", "benchmark_id": "aa_lcr", "score": 65.1, "reference_url": "https://intranetproxy.alipay.com/skylark/lark/0/2026/png/23157180/1785831264180-d6ca4404-acef-4424-84db-fbc5a4c6db5f.png", "reported_setting": { "checkpoint": "post-trained", "mode": "thinking enabled by default", "effort": "default", "shots": "source does not state", "samples": "source does not state", "trials": "source does not state", "aggregation": "source does not state", "tools": "none", "search": "none", "context_management": "none stated", "judge": "Artificial Analysis equality-checker LLM", "harness": "official InclusionAI / benchmark-specific", "dataset_version": "AA-LCR Intelligence Index methodology", "split": "source does not state", "context_length": "262,144 native evaluation/config", "multimodal_inputs": false, "temperature": "0.6", "top_p": "0.95", "top_k": "20", "prompt_style": "source does not state", "max_new_tokens": "source does not state", "timeout": "source does not state", "output_parser": "source does not state", "evaluation_provenance": "official InclusionAI evaluation or locked provider rendering" }, "matches_canonical": true, "source_type": "official_model_card", "audit_status": "verified", "notes": "Displayed exactly as 65.1. Locked official InclusionAI observation." }, { "model_id": "ling-3.0-flash", "benchmark_id": "aime_2026", "score": 93.2, "reference_url": "https://huggingface.co/inclusionAI/Ling-3.0-flash/blob/ecde16176a497adaff7419ff4de59da603c4edaa/.eval_results/aime_2026.yaml", "reported_setting": { "checkpoint": "post-trained", "mode": "thinking enabled by default", "effort": "default", "shots": "source does not state", "samples": "source does not state", "trials": "source does not state", "aggregation": "source does not state", "tools": "none", "search": "none", "context_management": "none stated", "judge": "benchmark-specified", "harness": "official InclusionAI / benchmark-specific", "dataset_version": "AIME 2026", "split": "source does not state", "context_length": "262,144 native evaluation/config", "multimodal_inputs": false, "temperature": "0.6", "top_p": "0.95", "top_k": "20", "prompt_style": "source does not state", "max_new_tokens": "source does not state", "timeout": "source does not state", "output_parser": "source does not state", "evaluation_provenance": "official InclusionAI evaluation or locked provider rendering" }, "matches_canonical": true, "source_type": "official_model_card", "audit_status": "verified", "notes": "Displayed exactly as 93.2. Locked official InclusionAI observation." }, { "model_id": "ling-3.0-flash", "benchmark_id": "bfcl_v4", "score": 73.0, "reference_url": "https://intranetproxy.alipay.com/skylark/lark/0/2026/png/23157180/1785831264180-d6ca4404-acef-4424-84db-fbc5a4c6db5f.png", "reported_setting": { "checkpoint": "post-trained", "mode": "thinking enabled by default", "effort": "default", "shots": "source does not state", "samples": "source does not state", "trials": "source does not state", "aggregation": "source does not state", "tools": "benchmark-provided functions", "search": "none", "context_management": "none stated", "judge": "official function-call match", "harness": "Berkeley Function Calling Leaderboard", "dataset_version": "BFCL v4", "split": "source does not state", "context_length": "262,144 native evaluation/config", "multimodal_inputs": false, "temperature": "0.6", "top_p": "0.95", "top_k": "20", "prompt_style": "source does not state", "max_new_tokens": "source does not state", "timeout": "source does not state", "output_parser": "source does not state", "evaluation_provenance": "official InclusionAI evaluation or locked provider rendering" }, "matches_canonical": true, "source_type": "official_model_card", "audit_status": "verified", "notes": "Displayed exactly as 73.0. Locked official InclusionAI observation." }, { "model_id": "ling-3.0-flash", "benchmark_id": "browsecomp", "score": 72.2, "reference_url": "https://intranetproxy.alipay.com/skylark/lark/0/2026/png/23157180/1785831264180-d6ca4404-acef-4424-84db-fbc5a4c6db5f.png", "reported_setting": { "checkpoint": "post-trained", "mode": "thinking enabled by default", "effort": "default", "shots": "source does not state", "samples": "source does not state", "trials": "source does not state", "aggregation": "source does not state", "tools": "browser/search and code execution", "search": "enabled", "context_management": "single-agent resume from trajectory summary at 64K", "judge": "BrowseComp exact-answer judge", "harness": "InclusionAI internal search-agent harness", "dataset_version": "BrowseComp", "split": "source does not state", "context_length": "262,144 native evaluation/config", "multimodal_inputs": false, "temperature": "0.6", "top_p": "0.95", "top_k": "20", "prompt_style": "source does not state", "max_new_tokens": "source does not state", "timeout": "source does not state", "output_parser": "source does not state", "evaluation_provenance": "official InclusionAI evaluation or locked provider rendering" }, "matches_canonical": true, "source_type": "official_model_card", "audit_status": "verified", "notes": "Displayed exactly as 72.2 (w/ ctx); 82.0 (MA). Locked official InclusionAI observation.", "candidates": [ { "score": 82.0, "reference_url": "https://intranetproxy.alipay.com/skylark/lark/0/2026/png/23157180/1785831264180-d6ca4404-acef-4424-84db-fbc5a4c6db5f.png", "source_type": "official_model_card", "reported_setting": { "checkpoint": "post-trained", "mode": "thinking enabled by default", "effort": "default", "shots": "source does not state", "samples": "source does not state", "trials": "source does not state", "aggregation": "source does not state", "tools": "browser/search and code execution", "search": "enabled", "context_management": "multi-agent SearchSwarm/Tongyi DeepResearch", "judge": "BrowseComp exact-answer judge", "harness": "InclusionAI internal search-agent harness", "dataset_version": "BrowseComp", "split": "source does not state", "context_length": "128K main agent / 64K sub-agent", "multimodal_inputs": false, "temperature": "0.85", "top_p": "0.95", "top_k": "source does not state", "prompt_style": "source does not state", "max_new_tokens": "source does not state", "timeout": "source does not state", "output_parser": "source does not state", "evaluation_provenance": "official InclusionAI evaluation or locked provider rendering" }, "notes": "Displayed exactly as 72.2 (w/ ctx); 82.0 (MA). Locked official InclusionAI observation." } ] }, { "model_id": "ling-3.0-flash", "benchmark_id": "draco", "score": 70.4, "reference_url": "https://intranetproxy.alipay.com/skylark/lark/0/2026/png/23157180/1785831264180-d6ca4404-acef-4424-84db-fbc5a4c6db5f.png", "reported_setting": { "checkpoint": "post-trained", "mode": "thinking enabled by default", "effort": "default", "shots": "source does not state", "samples": "source does not state", "trials": "source does not state", "aggregation": "source does not state", "tools": "browser/search and code execution", "search": "enabled", "context_management": "none stated", "judge": "official rubrics; Claude Opus 4.6 scoring model", "harness": "InclusionAI internal search-agent harness", "dataset_version": "Draco", "split": "source does not state", "context_length": "262,144 native evaluation/config", "multimodal_inputs": false, "temperature": "0.6", "top_p": "0.95", "top_k": "20", "prompt_style": "source does not state", "max_new_tokens": "source does not state", "timeout": "source does not state", "output_parser": "source does not state", "evaluation_provenance": "official InclusionAI evaluation or locked provider rendering" }, "matches_canonical": true, "source_type": "official_model_card", "audit_status": "verified", "notes": "Displayed exactly as 70.4. Locked official InclusionAI observation." }, { "model_id": "ling-3.0-flash", "benchmark_id": "gdpval_aa_v2_elo", "score": 1107.0, "reference_url": "https://intranetproxy.alipay.com/skylark/lark/0/2026/png/23157180/1785831264180-d6ca4404-acef-4424-84db-fbc5a4c6db5f.png", "reported_setting": { "checkpoint": "post-trained", "mode": "thinking enabled by default", "effort": "default", "shots": "source does not state", "samples": "source does not state", "trials": "source does not state", "aggregation": "source does not state", "tools": "Stirrup agent tools", "search": "none", "context_management": "none stated", "judge": "pairwise judge panel / Elo", "harness": "GDPval-AA v2; 250 turns; 5-hour timeout", "dataset_version": "GDPval v2-AA", "split": "220 public tasks", "context_length": "262,144 native evaluation/config", "multimodal_inputs": false, "temperature": "0.6", "top_p": "0.95", "top_k": "20", "prompt_style": "source does not state", "max_new_tokens": "source does not state", "timeout": "source does not state", "output_parser": "source does not state", "evaluation_provenance": "official InclusionAI evaluation or locked provider rendering" }, "matches_canonical": true, "source_type": "official_model_card", "audit_status": "verified", "notes": "Displayed exactly as 1107. Locked official InclusionAI observation." }, { "model_id": "ling-3.0-flash", "benchmark_id": "hmmt_feb_2026", "score": 87.0, "reference_url": "https://huggingface.co/inclusionAI/Ling-3.0-flash/blob/ecde16176a497adaff7419ff4de59da603c4edaa/.eval_results/hmmt_feb_2026.yaml", "reported_setting": { "checkpoint": "post-trained", "mode": "thinking enabled by default", "effort": "default", "shots": "source does not state", "samples": "source does not state", "trials": "source does not state", "aggregation": "source does not state", "tools": "none", "search": "none", "context_management": "none stated", "judge": "benchmark-specified", "harness": "official InclusionAI / benchmark-specific", "dataset_version": "HMMT February 2026", "split": "source does not state", "context_length": "262,144 native evaluation/config", "multimodal_inputs": false, "temperature": "0.6", "top_p": "0.95", "top_k": "20", "prompt_style": "source does not state", "max_new_tokens": "source does not state", "timeout": "source does not state", "output_parser": "source does not state", "evaluation_provenance": "official InclusionAI evaluation or locked provider rendering" }, "matches_canonical": true, "source_type": "official_model_card", "audit_status": "verified", "notes": "Displayed exactly as 87.0. Locked official InclusionAI observation." }, { "model_id": "ling-3.0-flash", "benchmark_id": "ifbench", "score": 74.5, "reference_url": "https://intranetproxy.alipay.com/skylark/lark/0/2026/png/23157180/1785831264180-d6ca4404-acef-4424-84db-fbc5a4c6db5f.png", "reported_setting": { "checkpoint": "post-trained", "mode": "thinking enabled by default", "effort": "default", "shots": "source does not state", "samples": "source does not state", "trials": "source does not state", "aggregation": "source does not state", "tools": "none", "search": "none", "context_management": "none stated", "judge": "benchmark-specified", "harness": "official InclusionAI / benchmark-specific", "dataset_version": "IFBench", "split": "source does not state", "context_length": "262,144 native evaluation/config", "multimodal_inputs": false, "temperature": "0.6", "top_p": "0.95", "top_k": "20", "prompt_style": "source does not state", "max_new_tokens": "source does not state", "timeout": "source does not state", "output_parser": "source does not state", "evaluation_provenance": "official InclusionAI evaluation or locked provider rendering" }, "matches_canonical": true, "source_type": "official_model_card", "audit_status": "verified", "notes": "Displayed exactly as 74.5. Locked official InclusionAI observation." }, { "model_id": "ling-3.0-flash", "benchmark_id": "imo_answerbench", "score": 83.7, "reference_url": "https://intranetproxy.alipay.com/skylark/lark/0/2026/png/23157180/1785831264180-d6ca4404-acef-4424-84db-fbc5a4c6db5f.png", "reported_setting": { "checkpoint": "post-trained", "mode": "thinking enabled by default", "effort": "default", "shots": "source does not state", "samples": "source does not state", "trials": "source does not state", "aggregation": "source does not state", "tools": "none", "search": "none", "context_management": "none stated", "judge": "benchmark-specified", "harness": "official InclusionAI / benchmark-specific", "dataset_version": "IMO-AnswerBench", "split": "source does not state", "context_length": "262,144 native evaluation/config", "multimodal_inputs": false, "temperature": "0.6", "top_p": "0.95", "top_k": "20", "prompt_style": "source does not state", "max_new_tokens": "source does not state", "timeout": "source does not state", "output_parser": "source does not state", "evaluation_provenance": "official InclusionAI evaluation or locked provider rendering" }, "matches_canonical": true, "source_type": "official_model_card", "audit_status": "verified", "notes": "Displayed exactly as 83.7. Locked official InclusionAI observation." }, { "model_id": "ling-3.0-flash", "benchmark_id": "livecodebench_v6_2408_2505", "score": 82.8, "reference_url": "https://intranetproxy.alipay.com/skylark/lark/0/2026/png/23157180/1785831264180-d6ca4404-acef-4424-84db-fbc5a4c6db5f.png", "reported_setting": { "checkpoint": "post-trained", "mode": "thinking enabled by default", "effort": "default", "shots": "source does not state", "samples": "source does not state", "trials": "source does not state", "aggregation": "source does not state", "tools": "code execution for generated solutions", "search": "none", "context_management": "none stated", "judge": "official test execution", "harness": "official InclusionAI / benchmark-specific", "dataset_version": "LiveCodeBench v6, 2024-08 to 2025-05", "split": "454 code-generation problems", "context_length": "262,144 native evaluation/config", "multimodal_inputs": false, "temperature": "0.6", "top_p": "0.95", "top_k": "20", "prompt_style": "source does not state", "max_new_tokens": "source does not state", "timeout": "source does not state", "output_parser": "source does not state", "evaluation_provenance": "official InclusionAI evaluation or locked provider rendering" }, "matches_canonical": true, "source_type": "official_model_card", "audit_status": "verified", "notes": "Displayed exactly as 82.8. Locked official InclusionAI observation." }, { "model_id": "ling-3.0-flash", "benchmark_id": "mcpatlas", "score": 65.5, "reference_url": "https://intranetproxy.alipay.com/skylark/lark/0/2026/png/23157180/1785831264180-d6ca4404-acef-4424-84db-fbc5a4c6db5f.png", "reported_setting": { "checkpoint": "post-trained", "mode": "thinking enabled by default", "effort": "default", "shots": "source does not state", "samples": "source does not state", "trials": "source does not state", "aggregation": "source does not state", "tools": "36 MCP servers / public tool suite", "search": "none", "context_management": "none stated", "judge": "Gemini-2.5-Pro claim coverage; pass at >=0.75", "harness": "official MCP-Atlas v1, 20-turn limit", "dataset_version": "MCP-Atlas", "split": "public 500 tasks", "context_length": "262,144 native evaluation/config", "multimodal_inputs": false, "temperature": "0.6", "top_p": "0.95", "top_k": "20", "prompt_style": "source does not state", "max_new_tokens": "source does not state", "timeout": "source does not state", "output_parser": "source does not state", "evaluation_provenance": "official InclusionAI evaluation or locked provider rendering" }, "matches_canonical": true, "source_type": "official_model_card", "audit_status": "verified", "notes": "Displayed exactly as 65.5. Locked official InclusionAI observation." }, { "model_id": "ling-3.0-flash", "benchmark_id": "multi_if", "score": 87.7, "reference_url": "https://intranetproxy.alipay.com/skylark/lark/0/2026/png/23157180/1785831264180-d6ca4404-acef-4424-84db-fbc5a4c6db5f.png", "reported_setting": { "checkpoint": "post-trained", "mode": "thinking enabled by default", "effort": "default", "shots": "source does not state", "samples": "source does not state", "trials": "source does not state", "aggregation": "source does not state", "tools": "none", "search": "none", "context_management": "none stated", "judge": "benchmark-specified", "harness": "official InclusionAI / benchmark-specific", "dataset_version": "Multi-IF turn-3", "split": "source does not state", "context_length": "262,144 native evaluation/config", "multimodal_inputs": false, "temperature": "0.6", "top_p": "0.95", "top_k": "20", "prompt_style": "source does not state", "max_new_tokens": "source does not state", "timeout": "source does not state", "output_parser": "source does not state", "evaluation_provenance": "official InclusionAI evaluation or locked provider rendering" }, "matches_canonical": true, "source_type": "official_model_card", "audit_status": "verified", "notes": "Displayed exactly as 87.7. Locked official InclusionAI observation." }, { "model_id": "ling-3.0-flash", "benchmark_id": "swe_bench_multilingual", "score": 72.4, "reference_url": "https://huggingface.co/inclusionAI/Ling-3.0-flash/blob/ecde16176a497adaff7419ff4de59da603c4edaa/.eval_results/swe-bench_multilingual.yaml", "reported_setting": { "checkpoint": "post-trained", "mode": "thinking enabled by default", "effort": "default", "shots": "source does not state", "samples": "source does not state", "trials": "source does not state", "aggregation": "source does not state", "tools": "coding-agent shell/editor tools", "search": "none", "context_management": "none stated", "judge": "official repository test suite", "harness": "OpenHands with tailored prompts", "dataset_version": "SWE-bench Multilingual", "split": "Multilingual 300", "context_length": "256K", "multimodal_inputs": false, "temperature": "0.6", "top_p": "0.95", "top_k": "20", "prompt_style": "OpenHands tailored prompts", "max_new_tokens": "32K", "timeout": "source does not state", "output_parser": "source does not state", "evaluation_provenance": "official InclusionAI evaluation or locked provider rendering" }, "matches_canonical": true, "source_type": "official_model_card", "audit_status": "verified", "notes": "Displayed exactly as 72.4. Locked official InclusionAI observation." }, { "model_id": "ling-3.0-flash", "benchmark_id": "swe_bench_pro", "score": 56.6, "reference_url": "https://huggingface.co/inclusionAI/Ling-3.0-flash/blob/ecde16176a497adaff7419ff4de59da603c4edaa/.eval_results/swe-bench_pro.yaml", "reported_setting": { "checkpoint": "post-trained", "mode": "thinking enabled by default", "effort": "default", "shots": "source does not state", "samples": "source does not state", "trials": "source does not state", "aggregation": "source does not state", "tools": "coding-agent shell/editor tools", "search": "none", "context_management": "none stated", "judge": "official repository test suite", "harness": "OpenHands with tailored prompts", "dataset_version": "SWE-bench Pro", "split": "public 731", "context_length": "256K", "multimodal_inputs": false, "temperature": "0.6", "top_p": "0.95", "top_k": "20", "prompt_style": "OpenHands tailored prompts", "max_new_tokens": "32K", "timeout": "source does not state", "output_parser": "source does not state", "evaluation_provenance": "official InclusionAI evaluation or locked provider rendering" }, "matches_canonical": true, "source_type": "official_model_card", "audit_status": "verified", "notes": "Displayed exactly as 56.6. Locked official InclusionAI observation." }, { "model_id": "ling-3.0-flash", "benchmark_id": "terminal_bench_2_1", "score": 57.0, "reference_url": "https://intranetproxy.alipay.com/skylark/lark/0/2026/png/23157180/1785831264180-d6ca4404-acef-4424-84db-fbc5a4c6db5f.png", "reported_setting": { "checkpoint": "post-trained", "mode": "thinking enabled by default", "effort": "default", "shots": "source does not state", "samples": 1, "trials": 3, "aggregation": "arithmetic mean of 3 runs", "tools": "terminal", "search": "none", "context_management": "none stated", "judge": "programmatic task tests", "harness": "Terminus 2 under Artificial Analysis protocol", "dataset_version": "Terminal-Bench 2.1, 89 tasks", "split": "source does not state", "context_length": "256K", "multimodal_inputs": false, "temperature": "0.6", "top_p": "1.0", "top_k": "20", "prompt_style": "source does not state", "max_new_tokens": "32K", "timeout": "2 hours per task", "output_parser": "provided JSON parser in preserve-thinking mode", "evaluation_provenance": "official InclusionAI evaluation or locked provider rendering" }, "matches_canonical": true, "source_type": "official_model_card", "audit_status": "verified", "notes": "Displayed exactly as 57.0. Locked official InclusionAI observation. The adjacent benchmark-labeled methodology bullet is scoped row-wide across every displayed model column." }, { "model_id": "llada2.2-flash", "benchmark_id": "aime_2026", "score": 62.24, "reference_url": "https://raw.githubusercontent.com/inclusionAI/LLaDA2.X/f402f0b52817e1c3586a024ebf89e69ad8ca5523/LLaDA2_2_tech_report.pdf", "reported_setting": { "checkpoint": "post-trained", "mode": "block-diffusion with Levenshtein editing", "effort": "default", "shots": "source does not state", "samples": 5, "trials": 5, "aggregation": "arithmetic mean of five runs", "tools": "none", "search": "none", "context_management": "none stated", "judge": "benchmark-specified", "harness": "official InclusionAI / benchmark-specific", "dataset_version": "AIME 2026", "split": "source does not state", "context_length": "131,072 / 128K", "multimodal_inputs": false, "temperature": "1.0", "top_p": "source does not state", "top_k": "source does not state", "prompt_style": "block_length=32; threshold=0.5; editing_threshold=0.0", "max_new_tokens": "source does not state", "timeout": "source does not state", "output_parser": "source does not state", "evaluation_provenance": "official InclusionAI evaluation or locked provider rendering" }, "matches_canonical": true, "source_type": "official_paper", "audit_status": "verified", "notes": "Displayed exactly as 62.24. Locked official InclusionAI observation." }, { "model_id": "llada2.2-flash", "benchmark_id": "bfcl_v3", "score": 67.17, "reference_url": "https://raw.githubusercontent.com/inclusionAI/LLaDA2.X/f402f0b52817e1c3586a024ebf89e69ad8ca5523/LLaDA2_2_tech_report.pdf", "reported_setting": { "checkpoint": "post-trained", "mode": "block-diffusion with Levenshtein editing", "effort": "default", "shots": "source does not state", "samples": 5, "trials": 5, "aggregation": "arithmetic mean of five runs", "tools": "benchmark-provided functions", "search": "none", "context_management": "none stated", "judge": "official function-call match", "harness": "Berkeley Function Calling Leaderboard", "dataset_version": "BFCL v3", "split": "source does not state", "context_length": "131,072 / 128K", "multimodal_inputs": false, "temperature": "1.0", "top_p": "source does not state", "top_k": "source does not state", "prompt_style": "block_length=32; threshold=0.5; editing_threshold=0.0", "max_new_tokens": "source does not state", "timeout": "source does not state", "output_parser": "source does not state", "evaluation_provenance": "official InclusionAI evaluation or locked provider rendering" }, "matches_canonical": true, "source_type": "official_paper", "audit_status": "verified", "notes": "Displayed exactly as 67.17. Locked official InclusionAI observation." }, { "model_id": "llada2.2-flash", "benchmark_id": "bfcl_v4", "score": 60.78, "reference_url": "https://raw.githubusercontent.com/inclusionAI/LLaDA2.X/f402f0b52817e1c3586a024ebf89e69ad8ca5523/LLaDA2_2_tech_report.pdf", "reported_setting": { "checkpoint": "post-trained", "mode": "block-diffusion with Levenshtein editing", "effort": "default", "shots": "source does not state", "samples": 5, "trials": 5, "aggregation": "arithmetic mean of five runs", "tools": "benchmark-provided functions", "search": "none", "context_management": "none stated", "judge": "official function-call match", "harness": "Berkeley Function Calling Leaderboard", "dataset_version": "BFCL v4", "split": "source does not state", "context_length": "131,072 / 128K", "multimodal_inputs": false, "temperature": "1.0", "top_p": "source does not state", "top_k": "source does not state", "prompt_style": "block_length=32; threshold=0.5; editing_threshold=0.0", "max_new_tokens": "source does not state", "timeout": "source does not state", "output_parser": "source does not state", "evaluation_provenance": "official InclusionAI evaluation or locked provider rendering" }, "matches_canonical": true, "source_type": "official_paper", "audit_status": "verified", "notes": "Displayed exactly as 60.78. Locked official InclusionAI observation." }, { "model_id": "llada2.2-flash", "benchmark_id": "gpqa_diamond", "score": 48.67, "reference_url": "https://raw.githubusercontent.com/inclusionAI/LLaDA2.X/f402f0b52817e1c3586a024ebf89e69ad8ca5523/LLaDA2_2_tech_report.pdf", "reported_setting": { "checkpoint": "post-trained", "mode": "block-diffusion with Levenshtein editing", "effort": "default", "shots": "source does not state", "samples": 5, "trials": 5, "aggregation": "arithmetic mean of five runs", "tools": "none", "search": "none", "context_management": "none stated", "judge": "benchmark-specified", "harness": "official InclusionAI / benchmark-specific", "dataset_version": "GPQA-Diamond", "split": "source does not state", "context_length": "131,072 / 128K", "multimodal_inputs": false, "temperature": "1.0", "top_p": "source does not state", "top_k": "source does not state", "prompt_style": "block_length=32; threshold=0.5; editing_threshold=0.0", "max_new_tokens": "source does not state", "timeout": "source does not state", "output_parser": "source does not state", "evaluation_provenance": "official InclusionAI evaluation or locked provider rendering" }, "matches_canonical": true, "source_type": "official_paper", "audit_status": "verified", "notes": "Displayed exactly as 48.67. Locked official InclusionAI observation." }, { "model_id": "llada2.2-flash", "benchmark_id": "ifbench", "score": 30.2, "reference_url": "https://raw.githubusercontent.com/inclusionAI/LLaDA2.X/f402f0b52817e1c3586a024ebf89e69ad8ca5523/LLaDA2_2_tech_report.pdf", "reported_setting": { "checkpoint": "post-trained", "mode": "block-diffusion with Levenshtein editing", "effort": "default", "shots": "source does not state", "samples": 5, "trials": 5, "aggregation": "arithmetic mean of five runs", "tools": "none", "search": "none", "context_management": "none stated", "judge": "benchmark-specified", "harness": "official InclusionAI / benchmark-specific", "dataset_version": "IFBench", "split": "source does not state", "context_length": "131,072 / 128K", "multimodal_inputs": false, "temperature": "1.0", "top_p": "source does not state", "top_k": "source does not state", "prompt_style": "block_length=32; threshold=0.5; editing_threshold=0.0", "max_new_tokens": "source does not state", "timeout": "source does not state", "output_parser": "source does not state", "evaluation_provenance": "official InclusionAI evaluation or locked provider rendering" }, "matches_canonical": true, "source_type": "official_paper", "audit_status": "verified", "notes": "Displayed exactly as 30.20. Locked official InclusionAI observation." }, { "model_id": "llada2.2-flash", "benchmark_id": "korbench", "score": 60.96, "reference_url": "https://raw.githubusercontent.com/inclusionAI/LLaDA2.X/f402f0b52817e1c3586a024ebf89e69ad8ca5523/LLaDA2_2_tech_report.pdf", "reported_setting": { "checkpoint": "post-trained", "mode": "block-diffusion with Levenshtein editing", "effort": "default", "shots": "source does not state", "samples": 5, "trials": 5, "aggregation": "arithmetic mean of five runs", "tools": "none", "search": "none", "context_management": "none stated", "judge": "benchmark-specified", "harness": "official InclusionAI / benchmark-specific", "dataset_version": "KORBench", "split": "source does not state", "context_length": "131,072 / 128K", "multimodal_inputs": false, "temperature": "1.0", "top_p": "source does not state", "top_k": "source does not state", "prompt_style": "block_length=32; threshold=0.5; editing_threshold=0.0", "max_new_tokens": "source does not state", "timeout": "source does not state", "output_parser": "source does not state", "evaluation_provenance": "official InclusionAI evaluation or locked provider rendering" }, "matches_canonical": true, "source_type": "official_paper", "audit_status": "verified", "notes": "Displayed exactly as 60.96. Locked official InclusionAI observation." }, { "model_id": "llada2.2-flash", "benchmark_id": "longbench_v2", "score": 45.13, "reference_url": "https://raw.githubusercontent.com/inclusionAI/LLaDA2.X/f402f0b52817e1c3586a024ebf89e69ad8ca5523/LLaDA2_2_tech_report.pdf", "reported_setting": { "checkpoint": "post-trained", "mode": "block-diffusion with Levenshtein editing", "effort": "default", "shots": "source does not state", "samples": 5, "trials": 5, "aggregation": "arithmetic mean of five runs", "tools": "none", "search": "none", "context_management": "none stated", "judge": "benchmark-specified", "harness": "official InclusionAI / benchmark-specific", "dataset_version": "LongBench v2", "split": "source does not state", "context_length": "131,072 / 128K", "multimodal_inputs": false, "temperature": "1.0", "top_p": "source does not state", "top_k": "source does not state", "prompt_style": "block_length=32; threshold=0.5; editing_threshold=0.0", "max_new_tokens": "source does not state", "timeout": "source does not state", "output_parser": "source does not state", "evaluation_provenance": "official InclusionAI evaluation or locked provider rendering" }, "matches_canonical": true, "source_type": "official_paper", "audit_status": "verified", "notes": "Displayed exactly as 45.13. Locked official InclusionAI observation." }, { "model_id": "llada2.2-flash", "benchmark_id": "mcpatlas", "score": 46.21, "reference_url": "https://raw.githubusercontent.com/inclusionAI/LLaDA2.X/f402f0b52817e1c3586a024ebf89e69ad8ca5523/LLaDA2_2_tech_report.pdf", "reported_setting": { "checkpoint": "post-trained", "mode": "block-diffusion with Levenshtein editing", "effort": "default", "shots": "source does not state", "samples": 5, "trials": 5, "aggregation": "arithmetic mean of five runs", "tools": "36 MCP servers / public tool suite", "search": "none", "context_management": "none stated", "judge": "Gemini-2.5-Pro claim coverage; pass at >=0.75", "harness": "official MCP-Atlas v1, 20-turn limit", "dataset_version": "MCP-Atlas", "split": "public 500 tasks", "context_length": "131,072 / 128K", "multimodal_inputs": false, "temperature": "1.0", "top_p": "source does not state", "top_k": "source does not state", "prompt_style": "block_length=32; threshold=0.5; editing_threshold=0.0", "max_new_tokens": "source does not state", "timeout": "source does not state", "output_parser": "source does not state", "evaluation_provenance": "official InclusionAI evaluation or locked provider rendering" }, "matches_canonical": true, "source_type": "official_paper", "audit_status": "verified", "notes": "Displayed exactly as 46.21. Locked official InclusionAI observation." }, { "model_id": "llada2.2-flash", "benchmark_id": "multi_if", "score": 73.67, "reference_url": "https://raw.githubusercontent.com/inclusionAI/LLaDA2.X/f402f0b52817e1c3586a024ebf89e69ad8ca5523/LLaDA2_2_tech_report.pdf", "reported_setting": { "checkpoint": "post-trained", "mode": "block-diffusion with Levenshtein editing", "effort": "default", "shots": "source does not state", "samples": 5, "trials": 5, "aggregation": "arithmetic mean of five runs", "tools": "none", "search": "none", "context_management": "none stated", "judge": "benchmark-specified", "harness": "official InclusionAI / benchmark-specific", "dataset_version": "Multi-IF turn-3", "split": "source does not state", "context_length": "131,072 / 128K", "multimodal_inputs": false, "temperature": "1.0", "top_p": "source does not state", "top_k": "source does not state", "prompt_style": "block_length=32; threshold=0.5; editing_threshold=0.0", "max_new_tokens": "source does not state", "timeout": "source does not state", "output_parser": "source does not state", "evaluation_provenance": "official InclusionAI evaluation or locked provider rendering" }, "matches_canonical": true, "source_type": "official_paper", "audit_status": "verified", "notes": "Displayed exactly as 73.67. Locked official InclusionAI observation." }, { "model_id": "llada2.2-flash", "benchmark_id": "swe_bench_multilingual", "score": 25.0, "reference_url": "https://raw.githubusercontent.com/inclusionAI/LLaDA2.X/f402f0b52817e1c3586a024ebf89e69ad8ca5523/LLaDA2_2_tech_report.pdf", "reported_setting": { "checkpoint": "post-trained", "mode": "block-diffusion with Levenshtein editing", "effort": "default", "shots": "source does not state", "samples": 5, "trials": 5, "aggregation": "arithmetic mean of five runs", "tools": "coding-agent shell/editor tools", "search": "none", "context_management": "none stated", "judge": "official repository test suite", "harness": "Claude Code", "dataset_version": "SWE-bench Multilingual", "split": "Multilingual 300", "context_length": "131,072 / 128K", "multimodal_inputs": false, "temperature": "1.0", "top_p": "source does not state", "top_k": "source does not state", "prompt_style": "block_length=32; threshold=0.5; editing_threshold=0.0", "max_new_tokens": "source does not state", "timeout": "source does not state", "output_parser": "source does not state", "evaluation_provenance": "official InclusionAI evaluation or locked provider rendering" }, "matches_canonical": true, "source_type": "official_paper", "audit_status": "verified", "notes": "Displayed exactly as 25.00. Locked official InclusionAI observation." }, { "model_id": "llada2.2-flash", "benchmark_id": "swe_bench_pro", "score": 30.1, "reference_url": "https://raw.githubusercontent.com/inclusionAI/LLaDA2.X/f402f0b52817e1c3586a024ebf89e69ad8ca5523/LLaDA2_2_tech_report.pdf", "reported_setting": { "checkpoint": "post-trained", "mode": "block-diffusion with Levenshtein editing", "effort": "default", "shots": "source does not state", "samples": 5, "trials": 5, "aggregation": "arithmetic mean of five runs", "tools": "coding-agent shell/editor tools", "search": "none", "context_management": "none stated", "judge": "official repository test suite", "harness": "Claude Code", "dataset_version": "SWE-bench Pro", "split": "public 731", "context_length": "131,072 / 128K", "multimodal_inputs": false, "temperature": "1.0", "top_p": "source does not state", "top_k": "source does not state", "prompt_style": "block_length=32; threshold=0.5; editing_threshold=0.0", "max_new_tokens": "source does not state", "timeout": "source does not state", "output_parser": "source does not state", "evaluation_provenance": "official InclusionAI evaluation or locked provider rendering" }, "matches_canonical": true, "source_type": "official_paper", "audit_status": "verified", "notes": "Displayed exactly as 30.10. Locked official InclusionAI observation." }, { "model_id": "llada2.2-flash", "benchmark_id": "swe_bench_verified", "score": 49.28, "reference_url": "https://raw.githubusercontent.com/inclusionAI/LLaDA2.X/f402f0b52817e1c3586a024ebf89e69ad8ca5523/LLaDA2_2_tech_report.pdf", "reported_setting": { "checkpoint": "post-trained", "mode": "block-diffusion with Levenshtein editing", "effort": "default", "shots": "source does not state", "samples": 5, "trials": 5, "aggregation": "arithmetic mean of five runs", "tools": "coding-agent shell/editor tools", "search": "none", "context_management": "none stated", "judge": "official repository test suite", "harness": "OpenHands", "dataset_version": "SWE-bench Verified", "split": "Verified 500", "context_length": "131,072 / 128K", "multimodal_inputs": false, "temperature": "1.0", "top_p": "source does not state", "top_k": "source does not state", "prompt_style": "block_length=32; threshold=0.5; editing_threshold=0.0", "max_new_tokens": "source does not state", "timeout": "source does not state", "output_parser": "source does not state", "evaluation_provenance": "official InclusionAI evaluation or locked provider rendering" }, "matches_canonical": true, "source_type": "official_paper", "audit_status": "verified", "notes": "Displayed exactly as 49.28. Locked official InclusionAI observation." }, { "model_id": "llada2.2-flash", "benchmark_id": "tau2_bench_avg", "score": 80.33, "reference_url": "https://raw.githubusercontent.com/inclusionAI/LLaDA2.X/f402f0b52817e1c3586a024ebf89e69ad8ca5523/LLaDA2_2_tech_report.pdf", "reported_setting": { "checkpoint": "post-trained", "mode": "block-diffusion with Levenshtein editing", "effort": "default", "shots": "source does not state", "samples": 5, "trials": 5, "aggregation": "arithmetic mean of five runs", "tools": "official tau2 domain tools", "search": "none", "context_management": "none stated", "judge": "task success", "harness": "official tau2 harness with source-specific user simulator", "dataset_version": "tau2 v1.0.0 average", "split": "source does not state", "context_length": "131,072 / 128K", "multimodal_inputs": false, "temperature": "1.0", "top_p": "source does not state", "top_k": "source does not state", "prompt_style": "block_length=32; threshold=0.5; editing_threshold=0.0", "max_new_tokens": "source does not state", "timeout": "source does not state", "output_parser": "source does not state", "evaluation_provenance": "official InclusionAI evaluation or locked provider rendering" }, "matches_canonical": true, "source_type": "official_paper", "audit_status": "verified", "notes": "Displayed exactly as 80.33. Locked official InclusionAI observation." }, { "model_id": "ring-2.6-1t", "benchmark_id": "aa_lcr", "score": 64.3, "reference_url": "https://intranetproxy.alipay.com/skylark/lark/0/2026/png/23157180/1785831264180-d6ca4404-acef-4424-84db-fbc5a4c6db5f.png", "reported_setting": { "checkpoint": "post-trained", "mode": "thinking", "effort": "xhigh", "shots": "source does not state", "samples": "source does not state", "trials": "source does not state", "aggregation": "source does not state", "tools": "none", "search": "none", "context_management": "none stated", "judge": "Artificial Analysis equality-checker LLM", "harness": "official InclusionAI / benchmark-specific", "dataset_version": "AA-LCR Intelligence Index methodology", "split": "source does not state", "context_length": "131,072 native; card advertises 256K YaRN", "multimodal_inputs": false, "temperature": "source does not state", "top_p": "source does not state", "top_k": "source does not state", "prompt_style": "source does not state", "max_new_tokens": "source does not state", "timeout": "source does not state", "output_parser": "source does not state", "evaluation_provenance": "official InclusionAI evaluation or locked provider rendering" }, "matches_canonical": false, "source_type": "official_model_card", "audit_status": "verified", "notes": "Displayed exactly as 64.3. Locked official InclusionAI observation." }, { "model_id": "ring-2.6-1t", "benchmark_id": "aime_2026", "score": 95.83, "reference_url": "https://huggingface.co/inclusionAI/Ring-2.6-1T/blob/1e58be9318352541575130d4dbbdcc735fca7a03/.eval_results/aime_2026.yaml", "reported_setting": { "checkpoint": "post-trained", "mode": "thinking", "effort": "xhigh", "shots": "source does not state", "samples": 64, "trials": 64, "aggregation": "mean@64", "tools": "none", "search": "none", "context_management": "none stated", "judge": "benchmark-specified", "harness": "official InclusionAI / benchmark-specific", "dataset_version": "AIME 2026", "split": "source does not state", "context_length": "131,072 native; card advertises 256K YaRN", "multimodal_inputs": false, "temperature": "source does not state", "top_p": "source does not state", "top_k": "source does not state", "prompt_style": "source does not state", "max_new_tokens": "source does not state", "timeout": "source does not state", "output_parser": "source does not state", "evaluation_provenance": "official InclusionAI evaluation or locked provider rendering" }, "matches_canonical": true, "source_type": "official_model_card", "audit_status": "verified", "notes": "Displayed exactly as 95.83. Locked official InclusionAI observation.", "candidates": [ { "score": 95.78, "reference_url": "https://arxiv.org/pdf/2606.15079v1", "source_type": "official_paper", "reported_setting": { "checkpoint": "post-trained", "mode": "thinking", "effort": "xhigh", "shots": "source does not state", "samples": 64, "trials": 64, "aggregation": "mean@64", "tools": "none", "search": "none", "context_management": "none stated", "judge": "benchmark-specified", "harness": "official InclusionAI / benchmark-specific", "dataset_version": "AIME 2026", "split": "source does not state", "context_length": "131,072 native; card advertises 256K YaRN", "multimodal_inputs": false, "temperature": "source does not state", "top_p": "source does not state", "top_k": "source does not state", "prompt_style": "source does not state", "max_new_tokens": "source does not state", "timeout": "source does not state", "output_parser": "source does not state", "evaluation_provenance": "official InclusionAI evaluation or locked provider rendering" }, "notes": "Displayed exactly as 95.78. Locked official InclusionAI observation." }, { "score": 95.8, "reference_url": "https://intranetproxy.alipay.com/skylark/lark/0/2026/png/23157180/1785831264180-d6ca4404-acef-4424-84db-fbc5a4c6db5f.png", "source_type": "official_model_card", "reported_setting": { "checkpoint": "post-trained", "mode": "thinking", "effort": "xhigh", "shots": "source does not state", "samples": "source does not state", "trials": "source does not state", "aggregation": "source does not state", "tools": "none", "search": "none", "context_management": "none stated", "judge": "benchmark-specified", "harness": "official InclusionAI / benchmark-specific", "dataset_version": "AIME 2026", "split": "source does not state", "context_length": "131,072 native; card advertises 256K YaRN", "multimodal_inputs": false, "temperature": "0.6", "top_p": "0.95", "top_k": "20", "prompt_style": "source does not state", "max_new_tokens": "source does not state", "timeout": "source does not state", "output_parser": "source does not state", "evaluation_provenance": "official InclusionAI evaluation or locked provider rendering" }, "notes": "Displayed exactly as 95.8. Locked official InclusionAI observation." }, { "score": 87.86, "reference_url": "https://arxiv.org/pdf/2606.15079v1", "source_type": "official_paper", "reported_setting": { "checkpoint": "post-trained", "mode": "thinking", "effort": "high", "shots": "source does not state", "samples": 64, "trials": 64, "aggregation": "mean@64", "tools": "none", "search": "none", "context_management": "none stated", "judge": "benchmark-specified", "harness": "official InclusionAI / benchmark-specific", "dataset_version": "AIME 2026", "split": "source does not state", "context_length": "131,072 native; card advertises 256K YaRN", "multimodal_inputs": false, "temperature": "source does not state", "top_p": "source does not state", "top_k": "source does not state", "prompt_style": "source does not state", "max_new_tokens": "source does not state", "timeout": "source does not state", "output_parser": "source does not state", "evaluation_provenance": "official InclusionAI evaluation or locked provider rendering" }, "notes": "Displayed exactly as 87.86. Locked official InclusionAI observation." } ] }, { "model_id": "ring-2.6-1t", "benchmark_id": "arc_agi_2", "score": 66.18, "reference_url": "https://arxiv.org/pdf/2606.15079v1", "reported_setting": { "checkpoint": "post-trained", "mode": "thinking", "effort": "xhigh", "shots": "source does not state", "samples": 2, "trials": 2, "aggregation": "pass@2", "tools": "none", "search": "none", "context_management": "none stated", "judge": "ARC exact task solver", "harness": "official InclusionAI / benchmark-specific", "dataset_version": "ARC-AGI-2", "split": "ARC-AGI-2 public evaluation set", "context_length": "131,072 native; card advertises 256K YaRN", "multimodal_inputs": false, "temperature": "source does not state", "top_p": "source does not state", "top_k": "source does not state", "prompt_style": "source does not state", "max_new_tokens": "source does not state", "timeout": "source does not state", "output_parser": "source does not state", "evaluation_provenance": "official InclusionAI evaluation or locked provider rendering" }, "matches_canonical": true, "source_type": "official_paper", "audit_status": "verified", "notes": "Displayed exactly as 66.18. Locked official InclusionAI observation." }, { "model_id": "ring-2.6-1t", "benchmark_id": "bfcl_v4", "score": 64.8, "reference_url": "https://intranetproxy.alipay.com/skylark/lark/0/2026/png/23157180/1785831264180-d6ca4404-acef-4424-84db-fbc5a4c6db5f.png", "reported_setting": { "checkpoint": "post-trained", "mode": "thinking", "effort": "xhigh", "shots": "source does not state", "samples": "source does not state", "trials": "source does not state", "aggregation": "source does not state", "tools": "benchmark-provided functions", "search": "none", "context_management": "none stated", "judge": "official function-call match", "harness": "Berkeley Function Calling Leaderboard", "dataset_version": "BFCL v4", "split": "source does not state", "context_length": "131,072 native; card advertises 256K YaRN", "multimodal_inputs": false, "temperature": "source does not state", "top_p": "source does not state", "top_k": "source does not state", "prompt_style": "source does not state", "max_new_tokens": "source does not state", "timeout": "source does not state", "output_parser": "source does not state", "evaluation_provenance": "official InclusionAI evaluation or locked provider rendering" }, "matches_canonical": false, "source_type": "official_model_card", "audit_status": "verified", "notes": "Displayed exactly as 64.8. Locked official InclusionAI observation." }, { "model_id": "ring-2.6-1t", "benchmark_id": "browsecomp", "score": 71.7, "reference_url": "https://intranetproxy.alipay.com/skylark/lark/0/2026/png/23157180/1785831264180-d6ca4404-acef-4424-84db-fbc5a4c6db5f.png", "reported_setting": { "checkpoint": "post-trained", "mode": "thinking", "effort": "xhigh", "shots": "source does not state", "samples": "source does not state", "trials": "source does not state", "aggregation": "source does not state", "tools": "browser/search and code execution", "search": "enabled", "context_management": "none stated", "judge": "BrowseComp exact-answer judge", "harness": "InclusionAI internal search-agent harness", "dataset_version": "BrowseComp", "split": "source does not state", "context_length": "131,072 native; card advertises 256K YaRN", "multimodal_inputs": false, "temperature": "source does not state", "top_p": "source does not state", "top_k": "source does not state", "prompt_style": "source does not state", "max_new_tokens": "source does not state", "timeout": "source does not state", "output_parser": "source does not state", "evaluation_provenance": "official InclusionAI evaluation or locked provider rendering" }, "matches_canonical": false, "source_type": "official_model_card", "audit_status": "verified", "notes": "Displayed exactly as 71.7. Locked official InclusionAI observation." }, { "model_id": "ring-2.6-1t", "benchmark_id": "claw_eval_general_pass3", "score": 63.82, "reference_url": "https://huggingface.co/inclusionAI/Ring-2.6-1T/blob/1e58be9318352541575130d4dbbdcc735fca7a03/.eval_results/claw_eval.yaml", "reported_setting": { "checkpoint": "post-trained", "mode": "thinking", "effort": "high", "shots": "source does not state", "samples": 3, "trials": 3, "aggregation": "Pass^3: all three trials must pass", "tools": "OpenClaw agent environment", "search": "none", "context_management": "none stated", "judge": "full-trajectory completion/safety/robustness grading", "harness": "official Claw-Eval v1.1", "dataset_version": "ClawEval", "split": "general 161 tasks", "context_length": "131,072 native; card advertises 256K YaRN", "multimodal_inputs": false, "temperature": "source does not state", "top_p": "source does not state", "top_k": "source does not state", "prompt_style": "source does not state", "max_new_tokens": "source does not state", "timeout": "source does not state", "output_parser": "source does not state", "evaluation_provenance": "official InclusionAI evaluation or locked provider rendering" }, "matches_canonical": true, "source_type": "official_model_card", "audit_status": "verified", "notes": "Displayed exactly as 63.82. Locked official InclusionAI observation." }, { "model_id": "ring-2.6-1t", "benchmark_id": "gdpval_aa_v2_elo", "score": 920.0, "reference_url": "https://intranetproxy.alipay.com/skylark/lark/0/2026/png/23157180/1785831264180-d6ca4404-acef-4424-84db-fbc5a4c6db5f.png", "reported_setting": { "checkpoint": "post-trained", "mode": "thinking", "effort": "xhigh", "shots": "source does not state", "samples": "source does not state", "trials": "source does not state", "aggregation": "source does not state", "tools": "Stirrup agent tools", "search": "none", "context_management": "none stated", "judge": "pairwise judge panel / Elo", "harness": "GDPval-AA v2; 250 turns; 5-hour timeout", "dataset_version": "GDPval v2-AA", "split": "220 public tasks", "context_length": "131,072 native; card advertises 256K YaRN", "multimodal_inputs": false, "temperature": "source does not state", "top_p": "source does not state", "top_k": "source does not state", "prompt_style": "source does not state", "max_new_tokens": "source does not state", "timeout": "source does not state", "output_parser": "source does not state", "evaluation_provenance": "official InclusionAI evaluation or locked provider rendering" }, "matches_canonical": false, "source_type": "official_model_card", "audit_status": "verified", "notes": "Displayed exactly as 920. Locked official InclusionAI observation." }, { "model_id": "ring-2.6-1t", "benchmark_id": "gpqa_diamond", "score": 88.27, "reference_url": "https://huggingface.co/inclusionAI/Ring-2.6-1T/blob/1e58be9318352541575130d4dbbdcc735fca7a03/.eval_results/gpqa_diamond.yaml", "reported_setting": { "checkpoint": "post-trained", "mode": "thinking", "effort": "xhigh", "shots": "source does not state", "samples": 1, "trials": 1, "aggregation": "pass@1", "tools": "none", "search": "none", "context_management": "none stated", "judge": "benchmark-specified", "harness": "official InclusionAI / benchmark-specific", "dataset_version": "GPQA-Diamond", "split": "source does not state", "context_length": "131,072 native; card advertises 256K YaRN", "multimodal_inputs": false, "temperature": "source does not state", "top_p": "source does not state", "top_k": "source does not state", "prompt_style": "source does not state", "max_new_tokens": "source does not state", "timeout": "source does not state", "output_parser": "source does not state", "evaluation_provenance": "official InclusionAI evaluation or locked provider rendering" }, "matches_canonical": true, "source_type": "official_model_card", "audit_status": "verified", "notes": "Displayed exactly as 88.27. Locked official InclusionAI observation.", "candidates": [ { "score": 85.89, "reference_url": "https://arxiv.org/pdf/2606.15079v1", "source_type": "official_paper", "reported_setting": { "checkpoint": "post-trained", "mode": "thinking", "effort": "xhigh", "shots": "source does not state", "samples": 16, "trials": 16, "aggregation": "mean@16", "tools": "none", "search": "none", "context_management": "none stated", "judge": "benchmark-specified", "harness": "official InclusionAI / benchmark-specific", "dataset_version": "GPQA-Diamond", "split": "source does not state", "context_length": "131,072 native; card advertises 256K YaRN", "multimodal_inputs": false, "temperature": "source does not state", "top_p": "source does not state", "top_k": "source does not state", "prompt_style": "source does not state", "max_new_tokens": "source does not state", "timeout": "source does not state", "output_parser": "source does not state", "evaluation_provenance": "official InclusionAI evaluation or locked provider rendering" }, "notes": "Displayed exactly as 85.89. Locked official InclusionAI observation." }, { "score": 76.1, "reference_url": "https://arxiv.org/pdf/2606.15079v1", "source_type": "official_paper", "reported_setting": { "checkpoint": "post-trained", "mode": "thinking", "effort": "high", "shots": "source does not state", "samples": 16, "trials": 16, "aggregation": "mean@16", "tools": "none", "search": "none", "context_management": "none stated", "judge": "benchmark-specified", "harness": "official InclusionAI / benchmark-specific", "dataset_version": "GPQA-Diamond", "split": "source does not state", "context_length": "131,072 native; card advertises 256K YaRN", "multimodal_inputs": false, "temperature": "source does not state", "top_p": "source does not state", "top_k": "source does not state", "prompt_style": "source does not state", "max_new_tokens": "source does not state", "timeout": "source does not state", "output_parser": "source does not state", "evaluation_provenance": "official InclusionAI evaluation or locked provider rendering" }, "notes": "Displayed exactly as 76.10. Locked official InclusionAI observation." } ] }, { "model_id": "ring-2.6-1t", "benchmark_id": "hmmt_feb_2026", "score": 93.47, "reference_url": "https://arxiv.org/pdf/2606.15079v1", "reported_setting": { "checkpoint": "post-trained", "mode": "thinking", "effort": "xhigh", "shots": "source does not state", "samples": 64, "trials": 64, "aggregation": "mean@64", "tools": "none", "search": "none", "context_management": "none stated", "judge": "benchmark-specified", "harness": "official InclusionAI / benchmark-specific", "dataset_version": "HMMT February 2026", "split": "source does not state", "context_length": "131,072 native; card advertises 256K YaRN", "multimodal_inputs": false, "temperature": "source does not state", "top_p": "source does not state", "top_k": "source does not state", "prompt_style": "source does not state", "max_new_tokens": "source does not state", "timeout": "source does not state", "output_parser": "source does not state", "evaluation_provenance": "official InclusionAI evaluation or locked provider rendering" }, "matches_canonical": true, "source_type": "official_paper", "audit_status": "verified", "notes": "Displayed exactly as 93.47. Locked official InclusionAI observation.", "candidates": [ { "score": 93.5, "reference_url": "https://intranetproxy.alipay.com/skylark/lark/0/2026/png/23157180/1785831264180-d6ca4404-acef-4424-84db-fbc5a4c6db5f.png", "source_type": "official_model_card", "reported_setting": { "checkpoint": "post-trained", "mode": "thinking", "effort": "xhigh", "shots": "source does not state", "samples": "source does not state", "trials": "source does not state", "aggregation": "source does not state", "tools": "none", "search": "none", "context_management": "none stated", "judge": "benchmark-specified", "harness": "official InclusionAI / benchmark-specific", "dataset_version": "HMMT February 2026", "split": "source does not state", "context_length": "131,072 native; card advertises 256K YaRN", "multimodal_inputs": false, "temperature": "0.6", "top_p": "0.95", "top_k": "20", "prompt_style": "source does not state", "max_new_tokens": "source does not state", "timeout": "source does not state", "output_parser": "source does not state", "evaluation_provenance": "official InclusionAI evaluation or locked provider rendering" }, "notes": "Displayed exactly as 93.5. Locked official InclusionAI observation." }, { "score": 67.8, "reference_url": "https://arxiv.org/pdf/2606.15079v1", "source_type": "official_paper", "reported_setting": { "checkpoint": "post-trained", "mode": "thinking", "effort": "high", "shots": "source does not state", "samples": 64, "trials": 64, "aggregation": "mean@64", "tools": "none", "search": "none", "context_management": "none stated", "judge": "benchmark-specified", "harness": "official InclusionAI / benchmark-specific", "dataset_version": "HMMT February 2026", "split": "source does not state", "context_length": "131,072 native; card advertises 256K YaRN", "multimodal_inputs": false, "temperature": "source does not state", "top_p": "source does not state", "top_k": "source does not state", "prompt_style": "source does not state", "max_new_tokens": "source does not state", "timeout": "source does not state", "output_parser": "source does not state", "evaluation_provenance": "official InclusionAI evaluation or locked provider rendering" }, "notes": "Displayed exactly as 67.80. Locked official InclusionAI observation." } ] }, { "model_id": "ring-2.6-1t", "benchmark_id": "ifbench", "score": 44.6, "reference_url": "https://intranetproxy.alipay.com/skylark/lark/0/2026/png/23157180/1785831264180-d6ca4404-acef-4424-84db-fbc5a4c6db5f.png", "reported_setting": { "checkpoint": "post-trained", "mode": "thinking", "effort": "xhigh", "shots": "source does not state", "samples": "source does not state", "trials": "source does not state", "aggregation": "source does not state", "tools": "none", "search": "none", "context_management": "none stated", "judge": "benchmark-specified", "harness": "official InclusionAI / benchmark-specific", "dataset_version": "IFBench", "split": "source does not state", "context_length": "131,072 native; card advertises 256K YaRN", "multimodal_inputs": false, "temperature": "source does not state", "top_p": "source does not state", "top_k": "source does not state", "prompt_style": "source does not state", "max_new_tokens": "source does not state", "timeout": "source does not state", "output_parser": "source does not state", "evaluation_provenance": "official InclusionAI evaluation or locked provider rendering" }, "matches_canonical": false, "source_type": "official_model_card", "audit_status": "verified", "notes": "Displayed exactly as 44.6. Locked official InclusionAI observation." }, { "model_id": "ring-2.6-1t", "benchmark_id": "imo_answerbench", "score": 86.12, "reference_url": "https://arxiv.org/pdf/2606.15079v1", "reported_setting": { "checkpoint": "post-trained", "mode": "thinking", "effort": "xhigh", "shots": "source does not state", "samples": 8, "trials": 8, "aggregation": "mean@8", "tools": "none", "search": "none", "context_management": "none stated", "judge": "benchmark-specified", "harness": "official InclusionAI / benchmark-specific", "dataset_version": "IMO-AnswerBench", "split": "source does not state", "context_length": "131,072 native; card advertises 256K YaRN", "multimodal_inputs": false, "temperature": "source does not state", "top_p": "source does not state", "top_k": "source does not state", "prompt_style": "source does not state", "max_new_tokens": "source does not state", "timeout": "source does not state", "output_parser": "source does not state", "evaluation_provenance": "official InclusionAI evaluation or locked provider rendering" }, "matches_canonical": true, "source_type": "official_paper", "audit_status": "verified", "notes": "Displayed exactly as 86.12. Locked official InclusionAI observation.", "candidates": [ { "score": 86.1, "reference_url": "https://intranetproxy.alipay.com/skylark/lark/0/2026/png/23157180/1785831264180-d6ca4404-acef-4424-84db-fbc5a4c6db5f.png", "source_type": "official_model_card", "reported_setting": { "checkpoint": "post-trained", "mode": "thinking", "effort": "xhigh", "shots": "source does not state", "samples": "source does not state", "trials": "source does not state", "aggregation": "source does not state", "tools": "none", "search": "none", "context_management": "none stated", "judge": "benchmark-specified", "harness": "official InclusionAI / benchmark-specific", "dataset_version": "IMO-AnswerBench", "split": "source does not state", "context_length": "131,072 native; card advertises 256K YaRN", "multimodal_inputs": false, "temperature": "source does not state", "top_p": "source does not state", "top_k": "source does not state", "prompt_style": "source does not state", "max_new_tokens": "source does not state", "timeout": "source does not state", "output_parser": "source does not state", "evaluation_provenance": "official InclusionAI evaluation or locked provider rendering" }, "notes": "Displayed exactly as 86.1. Locked official InclusionAI observation." }, { "score": 66.44, "reference_url": "https://arxiv.org/pdf/2606.15079v1", "source_type": "official_paper", "reported_setting": { "checkpoint": "post-trained", "mode": "thinking", "effort": "high", "shots": "source does not state", "samples": 8, "trials": 8, "aggregation": "mean@8", "tools": "none", "search": "none", "context_management": "none stated", "judge": "benchmark-specified", "harness": "official InclusionAI / benchmark-specific", "dataset_version": "IMO-AnswerBench", "split": "source does not state", "context_length": "131,072 native; card advertises 256K YaRN", "multimodal_inputs": false, "temperature": "source does not state", "top_p": "source does not state", "top_k": "source does not state", "prompt_style": "source does not state", "max_new_tokens": "source does not state", "timeout": "source does not state", "output_parser": "source does not state", "evaluation_provenance": "official InclusionAI evaluation or locked provider rendering" }, "notes": "Displayed exactly as 66.44. Locked official InclusionAI observation." } ] }, { "model_id": "ring-2.6-1t", "benchmark_id": "livecodebench_v6_2408_2505", "score": 86.95, "reference_url": "https://arxiv.org/pdf/2606.15079v1", "reported_setting": { "checkpoint": "post-trained", "mode": "thinking", "effort": "xhigh", "shots": "source does not state", "samples": 4, "trials": 4, "aggregation": "mean@4", "tools": "code execution for generated solutions", "search": "none", "context_management": "none stated", "judge": "official test execution", "harness": "official InclusionAI / benchmark-specific", "dataset_version": "LiveCodeBench v6, 2024-08 to 2025-05", "split": "454 code-generation problems", "context_length": "131,072 native; card advertises 256K YaRN", "multimodal_inputs": false, "temperature": "source does not state", "top_p": "source does not state", "top_k": "source does not state", "prompt_style": "source does not state", "max_new_tokens": "source does not state", "timeout": "source does not state", "output_parser": "source does not state", "evaluation_provenance": "official InclusionAI evaluation or locked provider rendering" }, "matches_canonical": true, "source_type": "official_paper", "audit_status": "verified", "notes": "Displayed exactly as 86.95. Locked official InclusionAI observation.", "candidates": [ { "score": 87.0, "reference_url": "https://intranetproxy.alipay.com/skylark/lark/0/2026/png/23157180/1785831264180-d6ca4404-acef-4424-84db-fbc5a4c6db5f.png", "source_type": "official_model_card", "reported_setting": { "checkpoint": "post-trained", "mode": "thinking", "effort": "xhigh", "shots": "source does not state", "samples": "source does not state", "trials": "source does not state", "aggregation": "source does not state", "tools": "code execution for generated solutions", "search": "none", "context_management": "none stated", "judge": "official test execution", "harness": "official InclusionAI / benchmark-specific", "dataset_version": "LiveCodeBench v6, 2024-08 to 2025-05", "split": "454 code-generation problems", "context_length": "131,072 native; card advertises 256K YaRN", "multimodal_inputs": false, "temperature": "source does not state", "top_p": "source does not state", "top_k": "source does not state", "prompt_style": "source does not state", "max_new_tokens": "source does not state", "timeout": "source does not state", "output_parser": "source does not state", "evaluation_provenance": "official InclusionAI evaluation or locked provider rendering" }, "notes": "Displayed exactly as 87.0. Locked official InclusionAI observation." }, { "score": 76.71, "reference_url": "https://arxiv.org/pdf/2606.15079v1", "source_type": "official_paper", "reported_setting": { "checkpoint": "post-trained", "mode": "thinking", "effort": "high", "shots": "source does not state", "samples": 4, "trials": 4, "aggregation": "mean@4", "tools": "code execution for generated solutions", "search": "none", "context_management": "none stated", "judge": "official test execution", "harness": "official InclusionAI / benchmark-specific", "dataset_version": "LiveCodeBench v6, 2024-08 to 2025-05", "split": "454 code-generation problems", "context_length": "131,072 native; card advertises 256K YaRN", "multimodal_inputs": false, "temperature": "source does not state", "top_p": "source does not state", "top_k": "source does not state", "prompt_style": "source does not state", "max_new_tokens": "source does not state", "timeout": "source does not state", "output_parser": "source does not state", "evaluation_provenance": "official InclusionAI evaluation or locked provider rendering" }, "notes": "Displayed exactly as 76.71. Locked official InclusionAI observation." } ] }, { "model_id": "ring-2.6-1t", "benchmark_id": "mcpatlas", "score": 61.2, "reference_url": "https://intranetproxy.alipay.com/skylark/lark/0/2026/png/23157180/1785831264180-d6ca4404-acef-4424-84db-fbc5a4c6db5f.png", "reported_setting": { "checkpoint": "post-trained", "mode": "thinking", "effort": "xhigh", "shots": "source does not state", "samples": "source does not state", "trials": "source does not state", "aggregation": "source does not state", "tools": "36 MCP servers / public tool suite", "search": "none", "context_management": "none stated", "judge": "Gemini-2.5-Pro claim coverage; pass at >=0.75", "harness": "official MCP-Atlas v1, 20-turn limit", "dataset_version": "MCP-Atlas", "split": "public 500 tasks", "context_length": "131,072 native; card advertises 256K YaRN", "multimodal_inputs": false, "temperature": "source does not state", "top_p": "source does not state", "top_k": "source does not state", "prompt_style": "source does not state", "max_new_tokens": "source does not state", "timeout": "source does not state", "output_parser": "source does not state", "evaluation_provenance": "official InclusionAI evaluation or locked provider rendering" }, "matches_canonical": false, "source_type": "official_model_card", "audit_status": "verified", "notes": "Displayed exactly as 61.2. Locked official InclusionAI observation." }, { "model_id": "ring-2.6-1t", "benchmark_id": "multi_if", "score": 89.3, "reference_url": "https://intranetproxy.alipay.com/skylark/lark/0/2026/png/23157180/1785831264180-d6ca4404-acef-4424-84db-fbc5a4c6db5f.png", "reported_setting": { "checkpoint": "post-trained", "mode": "thinking", "effort": "xhigh", "shots": "source does not state", "samples": "source does not state", "trials": "source does not state", "aggregation": "source does not state", "tools": "none", "search": "none", "context_management": "none stated", "judge": "benchmark-specified", "harness": "official InclusionAI / benchmark-specific", "dataset_version": "Multi-IF turn-3", "split": "source does not state", "context_length": "131,072 native; card advertises 256K YaRN", "multimodal_inputs": false, "temperature": "source does not state", "top_p": "source does not state", "top_k": "source does not state", "prompt_style": "source does not state", "max_new_tokens": "source does not state", "timeout": "source does not state", "output_parser": "source does not state", "evaluation_provenance": "official InclusionAI evaluation or locked provider rendering" }, "matches_canonical": false, "source_type": "official_model_card", "audit_status": "verified", "notes": "Displayed exactly as 89.3. Locked official InclusionAI observation." }, { "model_id": "ring-2.6-1t", "benchmark_id": "pinchbench_123", "score": 87.6, "reference_url": "https://arxiv.org/pdf/2606.15079v1", "reported_setting": { "checkpoint": "post-trained", "mode": "thinking", "effort": "high", "shots": "source does not state", "samples": 3, "trials": 3, "aggregation": "mean@3", "tools": "OpenClaw agent environment", "search": "none", "context_management": "none stated", "judge": "task-specific automatic and/or LLM judge", "harness": "official PinchBench/OpenClaw suite", "dataset_version": "PinchBench 123-task manifest at commit 27afe091b6ae04ec4b6aa9f5459bd280da0fd61d (2026-04-24)", "split": "123-task manifest at commit 27afe091b6ae04ec4b6aa9f5459bd280da0fd61d (2026-04-24)", "context_length": "131,072 native; card advertises 256K YaRN", "multimodal_inputs": false, "temperature": "source does not state", "top_p": "source does not state", "top_k": "source does not state", "prompt_style": "source does not state", "max_new_tokens": "source does not state", "timeout": "source does not state", "output_parser": "source does not state", "evaluation_provenance": "official InclusionAI evaluation or locked provider rendering" }, "matches_canonical": true, "source_type": "official_paper", "audit_status": "verified", "notes": "Displayed exactly as 87.60. Locked official InclusionAI observation." }, { "model_id": "ring-2.6-1t", "benchmark_id": "swe_bench_multilingual", "score": 56.7, "reference_url": "https://intranetproxy.alipay.com/skylark/lark/0/2026/png/23157180/1785831264180-d6ca4404-acef-4424-84db-fbc5a4c6db5f.png", "reported_setting": { "checkpoint": "post-trained", "mode": "thinking", "effort": "xhigh", "shots": "source does not state", "samples": "source does not state", "trials": "source does not state", "aggregation": "source does not state", "tools": "coding-agent shell/editor tools", "search": "none", "context_management": "none stated", "judge": "official repository test suite", "harness": "OpenHands with tailored prompts", "dataset_version": "SWE-bench Multilingual", "split": "Multilingual 300", "context_length": "256K", "multimodal_inputs": false, "temperature": "0.6", "top_p": "0.95", "top_k": "source does not state", "prompt_style": "OpenHands tailored prompts", "max_new_tokens": "32K", "timeout": "source does not state", "output_parser": "source does not state", "evaluation_provenance": "official InclusionAI evaluation or locked provider rendering" }, "matches_canonical": false, "source_type": "official_model_card", "audit_status": "verified", "notes": "Displayed exactly as 56.7. Locked official InclusionAI observation. The adjacent benchmark-labeled methodology bullet is scoped row-wide across every displayed model column." }, { "model_id": "ring-2.6-1t", "benchmark_id": "swe_bench_pro", "score": 53.76, "reference_url": "https://arxiv.org/pdf/2606.15079v1", "reported_setting": { "checkpoint": "post-trained", "mode": "thinking", "effort": "high", "shots": "source does not state", "samples": 1, "trials": 1, "aggregation": "% resolved (pass@1)", "tools": "coding-agent shell/editor tools", "search": "none", "context_management": "none stated", "judge": "official repository test suite", "harness": "Claude Code", "dataset_version": "SWE-bench Pro", "split": "public 731", "context_length": "131,072 native; card advertises 256K YaRN", "multimodal_inputs": false, "temperature": "source does not state", "top_p": "source does not state", "top_k": "source does not state", "prompt_style": "source does not state", "max_new_tokens": "source does not state", "timeout": "source does not state", "output_parser": "source does not state", "evaluation_provenance": "official InclusionAI evaluation or locked provider rendering" }, "matches_canonical": true, "source_type": "official_paper", "audit_status": "verified", "notes": "Displayed exactly as 53.76. Locked official InclusionAI observation.", "candidates": [ { "score": 53.9, "reference_url": "https://intranetproxy.alipay.com/skylark/lark/0/2026/png/23157180/1785831264180-d6ca4404-acef-4424-84db-fbc5a4c6db5f.png", "source_type": "official_model_card", "reported_setting": { "checkpoint": "post-trained", "mode": "thinking", "effort": "xhigh", "shots": "source does not state", "samples": "source does not state", "trials": "source does not state", "aggregation": "source does not state", "tools": "coding-agent shell/editor tools", "search": "none", "context_management": "none stated", "judge": "official repository test suite", "harness": "OpenHands with tailored prompts", "dataset_version": "SWE-bench Pro", "split": "public 731", "context_length": "256K", "multimodal_inputs": false, "temperature": "0.6", "top_p": "0.95", "top_k": "source does not state", "prompt_style": "OpenHands tailored prompts", "max_new_tokens": "32K", "timeout": "source does not state", "output_parser": "source does not state", "evaluation_provenance": "official InclusionAI evaluation or locked provider rendering" }, "notes": "Displayed exactly as 53.9. Locked official InclusionAI observation. The adjacent benchmark-labeled methodology bullet is scoped row-wide across every displayed model column." } ] }, { "model_id": "ring-2.6-1t", "benchmark_id": "swe_bench_verified", "score": 74.0, "reference_url": "https://huggingface.co/inclusionAI/Ring-2.6-1T/blob/1e58be9318352541575130d4dbbdcc735fca7a03/.eval_results/swe_bench_verified.yaml", "reported_setting": { "checkpoint": "post-trained", "mode": "thinking", "effort": "high", "shots": "source does not state", "samples": 1, "trials": 1, "aggregation": "% resolved (pass@1)", "tools": "coding-agent shell/editor tools", "search": "none", "context_management": "none stated", "judge": "official repository test suite", "harness": "source does not state", "dataset_version": "SWE-bench Verified", "split": "Verified 500", "context_length": "131,072 native; card advertises 256K YaRN", "multimodal_inputs": false, "temperature": "source does not state", "top_p": "source does not state", "top_k": "source does not state", "prompt_style": "source does not state", "max_new_tokens": "source does not state", "timeout": "source does not state", "output_parser": "source does not state", "evaluation_provenance": "official InclusionAI evaluation or locked provider rendering" }, "matches_canonical": true, "source_type": "official_model_card", "audit_status": "verified", "notes": "Displayed exactly as 74.0. Locked official InclusionAI observation." }, { "model_id": "ring-2.6-1t", "benchmark_id": "tau2_bench_airline", "score": 78.0, "reference_url": "https://arxiv.org/pdf/2606.15079v1", "reported_setting": { "checkpoint": "post-trained", "mode": "thinking", "effort": "high", "shots": "source does not state", "samples": 1, "trials": 1, "aggregation": "Accuracy", "tools": "official tau2 domain tools", "search": "none", "context_management": "none stated", "judge": "task success", "harness": "official tau2 harness with source-specific user simulator", "dataset_version": "tau2 v1.0.0 airline", "split": "source does not state", "context_length": "131,072 native; card advertises 256K YaRN", "multimodal_inputs": false, "temperature": "source does not state", "top_p": "source does not state", "top_k": "source does not state", "prompt_style": "source does not state", "max_new_tokens": "source does not state", "timeout": "source does not state", "output_parser": "source does not state", "evaluation_provenance": "official InclusionAI evaluation or locked provider rendering" }, "matches_canonical": true, "source_type": "official_paper", "audit_status": "verified", "notes": "Displayed exactly as 78.00. Locked official InclusionAI observation.", "candidates": [ { "score": 77.5, "reference_url": "https://arxiv.org/pdf/2606.15079v1", "source_type": "official_paper", "reported_setting": { "checkpoint": "post-trained", "mode": "thinking", "effort": "xhigh", "shots": "source does not state", "samples": 1, "trials": 1, "aggregation": "Accuracy", "tools": "official tau2 domain tools", "search": "none", "context_management": "none stated", "judge": "task success", "harness": "official tau2 harness with source-specific user simulator", "dataset_version": "tau2 v1.0.0 airline", "split": "source does not state", "context_length": "131,072 native; card advertises 256K YaRN", "multimodal_inputs": false, "temperature": "source does not state", "top_p": "source does not state", "top_k": "source does not state", "prompt_style": "source does not state", "max_new_tokens": "source does not state", "timeout": "source does not state", "output_parser": "source does not state", "evaluation_provenance": "official InclusionAI evaluation or locked provider rendering" }, "notes": "Displayed exactly as 77.50. Locked official InclusionAI observation." } ] }, { "model_id": "ring-2.6-1t", "benchmark_id": "tau2_bench_avg", "score": 84.26, "reference_url": "https://arxiv.org/pdf/2606.15079v1", "reported_setting": { "checkpoint": "post-trained", "mode": "thinking", "effort": "high", "shots": "source does not state", "samples": 1, "trials": 1, "aggregation": "Accuracy", "tools": "official tau2 domain tools", "search": "none", "context_management": "none stated", "judge": "task success", "harness": "official tau2 harness with source-specific user simulator", "dataset_version": "tau2 v1.0.0 average", "split": "source does not state", "context_length": "131,072 native; card advertises 256K YaRN", "multimodal_inputs": false, "temperature": "source does not state", "top_p": "source does not state", "top_k": "source does not state", "prompt_style": "source does not state", "max_new_tokens": "source does not state", "timeout": "source does not state", "output_parser": "source does not state", "evaluation_provenance": "official InclusionAI evaluation or locked provider rendering" }, "matches_canonical": true, "source_type": "official_paper", "audit_status": "verified", "notes": "Displayed exactly as 84.26. Locked official InclusionAI observation.", "candidates": [ { "score": 83.44, "reference_url": "https://arxiv.org/pdf/2606.15079v1", "source_type": "official_paper", "reported_setting": { "checkpoint": "post-trained", "mode": "thinking", "effort": "xhigh", "shots": "source does not state", "samples": 1, "trials": 1, "aggregation": "Accuracy", "tools": "official tau2 domain tools", "search": "none", "context_management": "none stated", "judge": "task success", "harness": "official tau2 harness with source-specific user simulator", "dataset_version": "tau2 v1.0.0 average", "split": "source does not state", "context_length": "131,072 native; card advertises 256K YaRN", "multimodal_inputs": false, "temperature": "source does not state", "top_p": "source does not state", "top_k": "source does not state", "prompt_style": "source does not state", "max_new_tokens": "source does not state", "timeout": "source does not state", "output_parser": "source does not state", "evaluation_provenance": "official InclusionAI evaluation or locked provider rendering" }, "notes": "Displayed exactly as 83.44. Locked official InclusionAI observation." } ] }, { "model_id": "ring-2.6-1t", "benchmark_id": "tau2_bench_retail", "score": 78.07, "reference_url": "https://arxiv.org/pdf/2606.15079v1", "reported_setting": { "checkpoint": "post-trained", "mode": "thinking", "effort": "high", "shots": "source does not state", "samples": 1, "trials": 1, "aggregation": "Accuracy", "tools": "official tau2 domain tools", "search": "none", "context_management": "none stated", "judge": "task success", "harness": "official tau2 harness with source-specific user simulator", "dataset_version": "tau2 v1.0.0 retail", "split": "source does not state", "context_length": "131,072 native; card advertises 256K YaRN", "multimodal_inputs": false, "temperature": "source does not state", "top_p": "source does not state", "top_k": "source does not state", "prompt_style": "source does not state", "max_new_tokens": "source does not state", "timeout": "source does not state", "output_parser": "source does not state", "evaluation_provenance": "official InclusionAI evaluation or locked provider rendering" }, "matches_canonical": true, "source_type": "official_paper", "audit_status": "verified", "notes": "Displayed exactly as 78.07. Locked official InclusionAI observation.", "candidates": [ { "score": 77.85, "reference_url": "https://arxiv.org/pdf/2606.15079v1", "source_type": "official_paper", "reported_setting": { "checkpoint": "post-trained", "mode": "thinking", "effort": "xhigh", "shots": "source does not state", "samples": 1, "trials": 1, "aggregation": "Accuracy", "tools": "official tau2 domain tools", "search": "none", "context_management": "none stated", "judge": "task success", "harness": "official tau2 harness with source-specific user simulator", "dataset_version": "tau2 v1.0.0 retail", "split": "source does not state", "context_length": "131,072 native; card advertises 256K YaRN", "multimodal_inputs": false, "temperature": "source does not state", "top_p": "source does not state", "top_k": "source does not state", "prompt_style": "source does not state", "max_new_tokens": "source does not state", "timeout": "source does not state", "output_parser": "source does not state", "evaluation_provenance": "official InclusionAI evaluation or locked provider rendering" }, "notes": "Displayed exactly as 77.85. Locked official InclusionAI observation." } ] }, { "model_id": "ring-2.6-1t", "benchmark_id": "tau2_bench_telecom", "score": 96.71, "reference_url": "https://arxiv.org/pdf/2606.15079v1", "reported_setting": { "checkpoint": "post-trained", "mode": "thinking", "effort": "high", "shots": "source does not state", "samples": 1, "trials": 1, "aggregation": "Accuracy", "tools": "official tau2 domain tools", "search": "none", "context_management": "none stated", "judge": "task success", "harness": "official tau2 harness with source-specific user simulator", "dataset_version": "tau2 v1.0.0 telecom", "split": "source does not state", "context_length": "131,072 native; card advertises 256K YaRN", "multimodal_inputs": false, "temperature": "source does not state", "top_p": "source does not state", "top_k": "source does not state", "prompt_style": "source does not state", "max_new_tokens": "source does not state", "timeout": "source does not state", "output_parser": "source does not state", "evaluation_provenance": "official InclusionAI evaluation or locked provider rendering" }, "matches_canonical": true, "source_type": "official_paper", "audit_status": "verified", "notes": "Displayed exactly as 96.71. Locked official InclusionAI observation.", "candidates": [ { "score": 95.32, "reference_url": "https://mdn.alipayobjects.com/huamei_d2byvp/afts/img/SSFgSJxTFPgAAAAAgDAAAAgADod9AQFr/original", "source_type": "official_model_card", "reported_setting": { "checkpoint": "post-trained", "mode": "thinking", "effort": "high", "shots": "source does not state", "samples": 3, "trials": 3, "aggregation": "mean@3", "tools": "official tau2 domain tools", "search": "none", "context_management": "none stated", "judge": "task success", "harness": "official tau2 harness with source-specific user simulator", "dataset_version": "tau2 v1.0.0 telecom", "split": "source does not state", "context_length": "131,072 native; card advertises 256K YaRN", "multimodal_inputs": false, "temperature": "source does not state", "top_p": "source does not state", "top_k": "source does not state", "prompt_style": "source does not state", "max_new_tokens": "source does not state", "timeout": "source does not state", "output_parser": "source does not state", "evaluation_provenance": "official InclusionAI evaluation or locked provider rendering" }, "notes": "Displayed exactly as 95.32. Locked official InclusionAI observation." }, { "score": 94.96, "reference_url": "https://arxiv.org/pdf/2606.15079v1", "source_type": "official_paper", "reported_setting": { "checkpoint": "post-trained", "mode": "thinking", "effort": "xhigh", "shots": "source does not state", "samples": 1, "trials": 1, "aggregation": "Accuracy", "tools": "official tau2 domain tools", "search": "none", "context_management": "none stated", "judge": "task success", "harness": "official tau2 harness with source-specific user simulator", "dataset_version": "tau2 v1.0.0 telecom", "split": "source does not state", "context_length": "131,072 native; card advertises 256K YaRN", "multimodal_inputs": false, "temperature": "source does not state", "top_p": "source does not state", "top_k": "source does not state", "prompt_style": "source does not state", "max_new_tokens": "source does not state", "timeout": "source does not state", "output_parser": "source does not state", "evaluation_provenance": "official InclusionAI evaluation or locked provider rendering" }, "notes": "Displayed exactly as 94.96. Locked official InclusionAI observation." } ] }, { "model_id": "ring-2.6-1t", "benchmark_id": "terminal_bench_2_1", "score": 43.1, "reference_url": "https://intranetproxy.alipay.com/skylark/lark/0/2026/png/23157180/1785831264180-d6ca4404-acef-4424-84db-fbc5a4c6db5f.png", "reported_setting": { "checkpoint": "post-trained", "mode": "thinking", "effort": "xhigh", "shots": "source does not state", "samples": 1, "trials": 3, "aggregation": "arithmetic mean of 3 runs", "tools": "terminal", "search": "none", "context_management": "none stated", "judge": "programmatic task tests", "harness": "Terminus 2 under Artificial Analysis protocol", "dataset_version": "Terminal-Bench 2.1, 89 tasks", "split": "source does not state", "context_length": "256K", "multimodal_inputs": false, "temperature": "0.6", "top_p": "1.0", "top_k": "source does not state", "prompt_style": "source does not state", "max_new_tokens": "32K", "timeout": "2 hours per task", "output_parser": "provided JSON parser in preserve-thinking mode", "evaluation_provenance": "official InclusionAI evaluation or locked provider rendering" }, "matches_canonical": false, "source_type": "official_model_card", "audit_status": "verified", "notes": "Displayed exactly as 43.1. Locked official InclusionAI observation. The adjacent benchmark-labeled methodology bullet is scoped row-wide across every displayed model column." }, { "model_id": "claude-opus-4.7", "benchmark_id": "deepsearchqa_acc", "score": 82.3, "reference_url": "https://static.stepfun.com/blog/step-3.7-flash/", "reported_setting": { "quantization_config": "common release evaluation; artifact precision not established", "configuration": "base Step evaluation configuration", "mode": "source does not state", "effort": "source does not state", "tools_search": "web search/research agent", "shots": "source does not state", "trials_samples": "source does not state", "aggregation_pass_k": "source-reported score", "context": "source does not state", "output_cap": "source does not state", "temperature": "source does not state", "top_p": "source does not state", "judge": "Gemini 2.5 Flash with official Kaggle grading prompt", "harness_agent": "official StepFun evaluation; exact harness not stated", "dataset_version_split": "DeepSearchQA 900-prompt eval split", "multimodal_input": false }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Official StepFun launch comparison table." }, { "model_id": "claude-opus-4.7", "benchmark_id": "deepsearchqa_f1", "score": 91.7, "reference_url": "https://static.stepfun.com/blog/step-3.7-flash/", "reported_setting": { "quantization_config": "common release evaluation; artifact precision not established", "configuration": "base Step evaluation configuration", "mode": "source does not state", "effort": "source does not state", "tools_search": "web search/research agent", "shots": "source does not state", "trials_samples": "source does not state", "aggregation_pass_k": "source-reported score", "context": "source does not state", "output_cap": "source does not state", "temperature": "source does not state", "top_p": "source does not state", "judge": "Gemini 2.5 Flash with official Kaggle grading prompt", "harness_agent": "official StepFun evaluation; exact harness not stated", "dataset_version_split": "DeepSearchQA 900-prompt eval split", "multimodal_input": false }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Official StepFun launch comparison table." }, { "model_id": "claude-opus-4.7", "benchmark_id": "gdpval_aa_elo", "score": 1753.0, "reference_url": "https://static.stepfun.com/blog/step-3.7-flash/", "reported_setting": { "quantization_config": "common release evaluation; artifact precision not established", "configuration": "base Step evaluation configuration", "mode": "source does not state", "effort": "source does not state", "tools_search": "official Artificial Analysis Stirrup harness", "shots": "source does not state", "trials_samples": "source does not state", "aggregation_pass_k": "source-reported score", "context": "source does not state", "output_cap": "source does not state", "temperature": "source does not state", "top_p": "source does not state", "judge": "pairwise Elo evaluation anchored to human experts", "harness_agent": "official Artificial Analysis Stirrup leaderboard", "dataset_version_split": "GDPval-AA historical 220-task Stirrup leaderboard quoted 2026-05-29", "multimodal_input": false }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Official StepFun launch comparison table." }, { "model_id": "claude-opus-4.7", "benchmark_id": "hle_tools", "score": 54.7, "reference_url": "https://www-cdn.anthropic.com/images/4zrzovbb/website/d434d15757c6abac1122af483617741776d5a114-2600x2638.png", "reported_setting": { "quantization_config": "common release evaluation; artifact precision not established", "configuration": "base Step evaluation configuration", "mode": "adaptive reasoning", "effort": "official release evaluation; exact effort not shown in chart", "tools_search": "tools enabled; exact suite undisclosed", "shots": "source does not state", "trials_samples": "source does not state", "aggregation_pass_k": "source-reported score", "context": "source does not state", "output_cap": "source does not state", "temperature": "source does not state", "top_p": "source does not state", "judge": "benchmark-specified; exact StepFun judge not stated", "harness_agent": "official Anthropic HLE with-tools evaluation", "dataset_version_split": "HLE finalized full set", "multimodal_input": true }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Official StepFun launch comparison table. Direct locked official score authority reports HLE w. tool=54.7; the StepFun surface is retained as physical corroboration." }, { "model_id": "claude-opus-4.7", "benchmark_id": "researchrubrics", "score": 73.9, "reference_url": "https://static.stepfun.com/blog/step-3.7-flash/", "reported_setting": { "quantization_config": "common release evaluation; artifact precision not established", "configuration": "base Step evaluation configuration", "mode": "source does not state", "effort": "source does not state", "tools_search": "research tools", "shots": "source does not state", "trials_samples": "source does not state", "aggregation_pass_k": "source-reported score", "context": "source does not state", "output_cap": "source does not state", "temperature": "source does not state", "top_p": "source does not state", "judge": "weighted binary rubric compliance evaluation", "harness_agent": "official StepFun evaluation; exact harness not stated", "dataset_version_split": "ResearchRubrics 101-task official release", "multimodal_input": false }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Official StepFun launch comparison table." }, { "model_id": "deepseek-v4-flash", "benchmark_id": "deepsearchqa_acc", "score": 79.8, "reference_url": "https://static.stepfun.com/blog/step-3.7-flash/", "reported_setting": { "quantization_config": "common release evaluation; artifact precision not established", "configuration": "base Step evaluation configuration", "mode": "source does not state", "effort": "source does not state", "tools_search": "web search/research agent", "shots": "source does not state", "trials_samples": "source does not state", "aggregation_pass_k": "source-reported score", "context": "source does not state", "output_cap": "source does not state", "temperature": "source does not state", "top_p": "source does not state", "judge": "Gemini 2.5 Flash with official Kaggle grading prompt", "harness_agent": "official StepFun evaluation; exact harness not stated", "dataset_version_split": "DeepSearchQA 900-prompt eval split", "multimodal_input": false }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Official StepFun launch comparison table." }, { "model_id": "deepseek-v4-flash", "benchmark_id": "hle_tools", "score": 45.1, "reference_url": "https://huggingface.co/deepseek-ai/DeepSeek-V4-Pro/resolve/b5968e9190ef611bbf34a7229255be88a0e937c1/README.md?download=true", "reported_setting": { "quantization_config": "common release evaluation; artifact precision not established", "configuration": "base Step evaluation configuration", "mode": "thinking", "effort": "Max", "tools_search": "tools enabled; exact suite undisclosed", "shots": "source does not state", "trials_samples": "source does not state", "aggregation_pass_k": "source-reported score", "context": "source does not state", "output_cap": "source does not state", "temperature": "source does not state", "top_p": "source does not state", "judge": "benchmark-specified; exact StepFun judge not stated", "harness_agent": "official DeepSeek HLE tool evaluation", "dataset_version_split": "HLE finalized full set", "multimodal_input": true }, "matches_canonical": false, "source_type": "official_model_card", "audit_status": "verified", "notes": "Official StepFun launch comparison table. Direct locked official score authority reports HLE w. tool=45.1; the StepFun surface is retained as physical corroboration." }, { "model_id": "deepseek-v4-flash", "benchmark_id": "researchrubrics", "score": 66.2, "reference_url": "https://static.stepfun.com/blog/step-3.7-flash/", "reported_setting": { "quantization_config": "common release evaluation; artifact precision not established", "configuration": "base Step evaluation configuration", "mode": "source does not state", "effort": "source does not state", "tools_search": "research tools", "shots": "source does not state", "trials_samples": "source does not state", "aggregation_pass_k": "source-reported score", "context": "source does not state", "output_cap": "source does not state", "temperature": "source does not state", "top_p": "source does not state", "judge": "weighted binary rubric compliance evaluation", "harness_agent": "official StepFun evaluation; exact harness not stated", "dataset_version_split": "ResearchRubrics 101-task official release", "multimodal_input": false }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Official StepFun launch comparison table." }, { "model_id": "deepseek-v4-pro", "benchmark_id": "researchrubrics", "score": 68.3, "reference_url": "https://static.stepfun.com/blog/step-3.7-flash/", "reported_setting": { "quantization_config": "common release evaluation; artifact precision not established", "configuration": "base Step evaluation configuration", "mode": "source does not state", "effort": "source does not state", "tools_search": "research tools", "shots": "source does not state", "trials_samples": "source does not state", "aggregation_pass_k": "source-reported score", "context": "source does not state", "output_cap": "source does not state", "temperature": "source does not state", "top_p": "source does not state", "judge": "weighted binary rubric compliance evaluation", "harness_agent": "official StepFun evaluation; exact harness not stated", "dataset_version_split": "ResearchRubrics 101-task official release", "multimodal_input": false }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Official StepFun launch comparison table." }, { "model_id": "gemini-3-flash", "benchmark_id": "android_daily", "score": 63.21, "reference_url": "https://static.stepfun.com/blog/step-3.7-flash/", "reported_setting": { "quantization_config": "common release evaluation; artifact precision not established", "configuration": "base Step evaluation configuration", "mode": "source does not state", "effort": "source does not state", "tools_search": "physical Android phone-use action environment", "shots": "source does not state", "trials_samples": "source does not state", "aggregation_pass_k": "pass@1 task success rate", "context": "source does not state", "output_cap": "source does not state", "temperature": "source does not state", "top_p": "source does not state", "judge": "benchmark-specified; exact StepFun judge not stated", "harness_agent": "StepFun phone-use stack; exact launch harness undisclosed", "dataset_version_split": "AndroidDaily v1, 350 tasks / 94 apps", "multimodal_input": true }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Corrected directly from the current locked official DOM. The prior research's 48.2/43.7/43.3/37.6 transcription was stale and is not retained." }, { "model_id": "gemini-3.5-flash", "benchmark_id": "researchrubrics", "score": 63.6, "reference_url": "https://static.stepfun.com/blog/step-3.7-flash/", "reported_setting": { "quantization_config": "common release evaluation; artifact precision not established", "configuration": "base Step evaluation configuration", "mode": "source does not state", "effort": "source does not state", "tools_search": "research tools", "shots": "source does not state", "trials_samples": "source does not state", "aggregation_pass_k": "source-reported score", "context": "source does not state", "output_cap": "source does not state", "temperature": "source does not state", "top_p": "source does not state", "judge": "weighted binary rubric compliance evaluation", "harness_agent": "official StepFun evaluation; exact harness not stated", "dataset_version_split": "ResearchRubrics 101-task official release", "multimodal_input": false }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Official StepFun launch comparison table." }, { "model_id": "glm-5.1", "benchmark_id": "deepsearchqa_acc", "score": 81.3, "reference_url": "https://static.stepfun.com/blog/step-3.7-flash/", "reported_setting": { "quantization_config": "common release evaluation; artifact precision not established", "configuration": "base Step evaluation configuration", "mode": "source does not state", "effort": "source does not state", "tools_search": "web search/research agent", "shots": "source does not state", "trials_samples": "source does not state", "aggregation_pass_k": "source-reported score", "context": "source does not state", "output_cap": "source does not state", "temperature": "source does not state", "top_p": "source does not state", "judge": "Gemini 2.5 Flash with official Kaggle grading prompt", "harness_agent": "official StepFun evaluation; exact harness not stated", "dataset_version_split": "DeepSearchQA 900-prompt eval split", "multimodal_input": false }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Official StepFun launch comparison table." }, { "model_id": "glm-5.1", "benchmark_id": "researchrubrics", "score": 67.9, "reference_url": "https://static.stepfun.com/blog/step-3.7-flash/", "reported_setting": { "quantization_config": "common release evaluation; artifact precision not established", "configuration": "base Step evaluation configuration", "mode": "source does not state", "effort": "source does not state", "tools_search": "research tools", "shots": "source does not state", "trials_samples": "source does not state", "aggregation_pass_k": "source-reported score", "context": "source does not state", "output_cap": "source does not state", "temperature": "source does not state", "top_p": "source does not state", "judge": "weighted binary rubric compliance evaluation", "harness_agent": "official StepFun evaluation; exact harness not stated", "dataset_version_split": "ResearchRubrics 101-task official release", "multimodal_input": false }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Official StepFun launch comparison table." }, { "model_id": "glm-5v-turbo", "benchmark_id": "android_daily", "score": 51.68, "reference_url": "https://static.stepfun.com/blog/step-3.7-flash/", "reported_setting": { "quantization_config": "common release evaluation; artifact precision not established", "configuration": "base Step evaluation configuration", "mode": "source does not state", "effort": "source does not state", "tools_search": "physical Android phone-use action environment", "shots": "source does not state", "trials_samples": "source does not state", "aggregation_pass_k": "pass@1 task success rate", "context": "source does not state", "output_cap": "source does not state", "temperature": "source does not state", "top_p": "source does not state", "judge": "benchmark-specified; exact StepFun judge not stated", "harness_agent": "StepFun phone-use stack; exact launch harness undisclosed", "dataset_version_split": "AndroidDaily v1, 350 tasks / 94 apps", "multimodal_input": true }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Corrected directly from the current locked official DOM. The prior research's 48.2/43.7/43.3/37.6 transcription was stale and is not retained." }, { "model_id": "glm-5v-turbo", "benchmark_id": "hr_bench_4k", "score": 84.62, "reference_url": "https://static.stepfun.com/blog/step-3.7-flash/", "reported_setting": { "quantization_config": "common release evaluation; artifact precision not established", "configuration": "base Step evaluation configuration", "mode": "source does not state", "effort": "source does not state", "tools_search": "Python visual tool: crop/zoom/draw/search actions", "shots": "source does not state", "trials_samples": "source does not state", "aggregation_pass_k": "accuracy", "context": "source does not state", "output_cap": "source does not state", "temperature": "source does not state", "top_p": "source does not state", "judge": "benchmark-specified; exact StepFun judge not stated", "harness_agent": "provider-aligned crop + search and other visual tools; exact tool composition not disclosed", "dataset_version_split": "HR-Bench 4K official 800-row release", "multimodal_input": true }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Official StepFun benchmark-specific tool table." }, { "model_id": "glm-5v-turbo", "benchmark_id": "hr_bench_8k", "score": 83.12, "reference_url": "https://static.stepfun.com/blog/step-3.7-flash/", "reported_setting": { "quantization_config": "common release evaluation; artifact precision not established", "configuration": "base Step evaluation configuration", "mode": "source does not state", "effort": "source does not state", "tools_search": "Python visual tool: crop/zoom/draw/search actions", "shots": "source does not state", "trials_samples": "source does not state", "aggregation_pass_k": "accuracy", "context": "source does not state", "output_cap": "source does not state", "temperature": "source does not state", "top_p": "source does not state", "judge": "benchmark-specified; exact StepFun judge not stated", "harness_agent": "provider-aligned crop + search and other visual tools; exact tool composition not disclosed", "dataset_version_split": "HR-Bench 8K official 800-row release", "multimodal_input": true }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Official StepFun benchmark-specific tool table." }, { "model_id": "glm-5v-turbo", "benchmark_id": "simplevqa", "score": 78.2, "reference_url": "https://cdn.bigmodel.cn/markdown/1775044860800img_v3_0210b_c42d0355-22af-4be7-8582-16fc6e7d0bfg.png?attname=img_v3_0210b_c42d0355-22af-4be7-8582-16fc6e7d0bfg.png", "reported_setting": { "quantization_config": "common release evaluation; artifact precision not established", "configuration": "base Step evaluation configuration", "mode": "source does not state", "effort": "source does not state", "tools_search": "Visual Search", "shots": "source does not state", "trials_samples": "source does not state", "aggregation_pass_k": "source-reported score", "context": "source does not state", "output_cap": "source does not state", "temperature": "source does not state", "top_p": "source does not state", "judge": "LLM-as-judge; exact StepFun judge not stated", "harness_agent": "official GLM multimodal ToolUse evaluation; exact tool composition not disclosed", "dataset_version_split": "SimpleVQA official 2,025-item test release", "multimodal_input": true }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Official StepFun benchmark-specific tool table. Direct locked official score authority reports SimpleVQA=78.2; the StepFun surface is retained as physical corroboration." }, { "model_id": "glm-5v-turbo", "benchmark_id": "vstar", "score": 89.0, "reference_url": "https://cdn.bigmodel.cn/markdown/1775044860800img_v3_0210b_c42d0355-22af-4be7-8582-16fc6e7d0bfg.png?attname=img_v3_0210b_c42d0355-22af-4be7-8582-16fc6e7d0bfg.png", "reported_setting": { "quantization_config": "common release evaluation; artifact precision not established", "configuration": "base Step evaluation configuration", "mode": "source does not state", "effort": "source does not state", "tools_search": "Python visual tool: crop/zoom/draw/search actions", "shots": "source does not state", "trials_samples": "source does not state", "aggregation_pass_k": "accuracy", "context": "source does not state", "output_cap": "source does not state", "temperature": "source does not state", "top_p": "source does not state", "judge": "benchmark-specified; exact StepFun judge not stated", "harness_agent": "official GLM multimodal ToolUse evaluation; exact tool composition not disclosed", "dataset_version_split": "V* Bench official 191-question release", "multimodal_input": true }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Official StepFun benchmark-specific tool table. Direct locked official score authority reports V*=89.0; the StepFun surface is retained as physical corroboration." }, { "model_id": "glm-5v-turbo", "benchmark_id": "worldvqa", "score": 47.81, "reference_url": "https://static.stepfun.com/blog/step-3.7-flash/", "reported_setting": { "quantization_config": "common release evaluation; artifact precision not established", "configuration": "base Step evaluation configuration", "mode": "source does not state", "effort": "source does not state", "tools_search": "Visual Search", "shots": "source does not state", "trials_samples": "source does not state", "aggregation_pass_k": "source-reported score", "context": "source does not state", "output_cap": "source does not state", "temperature": "source does not state", "top_p": "source does not state", "judge": "LLM-as-judge; exact StepFun judge not stated", "harness_agent": "official StepFun evaluation; exact harness not stated", "dataset_version_split": "WorldVQA first-eight-category 3,000-item release", "multimodal_input": true }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Official StepFun benchmark-specific tool table." }, { "model_id": "gpt-5.4", "benchmark_id": "claw_eval_pass3", "score": 60.3, "reference_url": "https://raw.githubusercontent.com/claw-eval/claw-eval.github.io/e10516b5c2a6dfc139be35dacb54cdf5e034801d/assets/benchmark-BgnHRA8t.js", "reported_setting": { "quantization_config": "common release evaluation; artifact precision not established", "configuration": "base Step evaluation configuration", "mode": "source does not state", "effort": "source does not state", "tools_search": "official Claw-Eval agent environment", "shots": "source does not state", "trials_samples": 3, "aggregation_pass_k": "Pass^3: all three trials must pass", "context": "source does not state", "output_cap": "source does not state", "temperature": "source does not state", "top_p": "source does not state", "judge": "full-trajectory completion/safety/robustness grading", "harness_agent": "official Claw-Eval v1.1 non-multimodal leaderboard", "dataset_version_split": "Claw-Eval v1.1 non-multimodal: 161 general + 38 multi-turn tasks", "multimodal_input": false }, "matches_canonical": true, "source_type": "official_repository", "audit_status": "verified", "notes": "Official launch hero chart label. Direct locked official score authority reports Claweval-v1.1=60.3; the StepFun surface is retained as physical corroboration." }, { "model_id": "kimi-k2.6", "benchmark_id": "android_daily", "score": 53.36, "reference_url": "https://static.stepfun.com/blog/step-3.7-flash/", "reported_setting": { "quantization_config": "common release evaluation; artifact precision not established", "configuration": "base Step evaluation configuration", "mode": "source does not state", "effort": "source does not state", "tools_search": "physical Android phone-use action environment", "shots": "source does not state", "trials_samples": "source does not state", "aggregation_pass_k": "pass@1 task success rate", "context": "source does not state", "output_cap": "source does not state", "temperature": "source does not state", "top_p": "source does not state", "judge": "benchmark-specified; exact StepFun judge not stated", "harness_agent": "StepFun phone-use stack; exact launch harness undisclosed", "dataset_version_split": "AndroidDaily v1, 350 tasks / 94 apps", "multimodal_input": true }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Corrected directly from the current locked official DOM. The prior research's 48.2/43.7/43.3/37.6 transcription was stale and is not retained." }, { "model_id": "kimi-k2.6", "benchmark_id": "hr_bench_4k", "score": 91.25, "reference_url": "https://static.stepfun.com/blog/step-3.7-flash/", "reported_setting": { "quantization_config": "common release evaluation; artifact precision not established", "configuration": "base Step evaluation configuration", "mode": "source does not state", "effort": "source does not state", "tools_search": "Python visual tool: crop/zoom/draw/search actions", "shots": "source does not state", "trials_samples": "source does not state", "aggregation_pass_k": "accuracy", "context": "source does not state", "output_cap": "source does not state", "temperature": "source does not state", "top_p": "source does not state", "judge": "benchmark-specified; exact StepFun judge not stated", "harness_agent": "official StepFun evaluation; exact harness not stated", "dataset_version_split": "HR-Bench 4K official 800-row release", "multimodal_input": true }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Official StepFun benchmark-specific tool table." }, { "model_id": "kimi-k2.6", "benchmark_id": "hr_bench_8k", "score": 90.13, "reference_url": "https://static.stepfun.com/blog/step-3.7-flash/", "reported_setting": { "quantization_config": "common release evaluation; artifact precision not established", "configuration": "base Step evaluation configuration", "mode": "source does not state", "effort": "source does not state", "tools_search": "Python visual tool: crop/zoom/draw/search actions", "shots": "source does not state", "trials_samples": "source does not state", "aggregation_pass_k": "accuracy", "context": "source does not state", "output_cap": "source does not state", "temperature": "source does not state", "top_p": "source does not state", "judge": "benchmark-specified; exact StepFun judge not stated", "harness_agent": "official StepFun evaluation; exact harness not stated", "dataset_version_split": "HR-Bench 8K official 800-row release", "multimodal_input": true }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Official StepFun benchmark-specific tool table." }, { "model_id": "kimi-k2.6", "benchmark_id": "researchrubrics", "score": 63.0, "reference_url": "https://static.stepfun.com/blog/step-3.7-flash/", "reported_setting": { "quantization_config": "common release evaluation; artifact precision not established", "configuration": "base Step evaluation configuration", "mode": "source does not state", "effort": "source does not state", "tools_search": "research tools", "shots": "source does not state", "trials_samples": "source does not state", "aggregation_pass_k": "source-reported score", "context": "source does not state", "output_cap": "source does not state", "temperature": "source does not state", "top_p": "source does not state", "judge": "weighted binary rubric compliance evaluation", "harness_agent": "official StepFun evaluation; exact harness not stated", "dataset_version_split": "ResearchRubrics 101-task official release", "multimodal_input": false }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Official StepFun launch comparison table." }, { "model_id": "kimi-k2.6", "benchmark_id": "simplevqa", "score": 78.24, "reference_url": "https://static.stepfun.com/blog/step-3.7-flash/", "reported_setting": { "quantization_config": "common release evaluation; artifact precision not established", "configuration": "base Step evaluation configuration", "mode": "source does not state", "effort": "source does not state", "tools_search": "Visual Search", "shots": "source does not state", "trials_samples": "source does not state", "aggregation_pass_k": "source-reported score", "context": "source does not state", "output_cap": "source does not state", "temperature": "source does not state", "top_p": "source does not state", "judge": "LLM-as-judge; exact StepFun judge not stated", "harness_agent": "official StepFun evaluation; exact harness not stated", "dataset_version_split": "SimpleVQA official 2,025-item test release", "multimodal_input": true }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Official StepFun benchmark-specific tool table." }, { "model_id": "kimi-k2.6", "benchmark_id": "vstar", "score": 96.9, "reference_url": "https://huggingface.co/moonshotai/Kimi-K2.6/resolve/d9cb81bc88b9bd2dc89877599f23c614ded72b9c/README.md?download=true", "reported_setting": { "quantization_config": "common release evaluation; artifact precision not established", "configuration": "base Step evaluation configuration", "mode": "source does not state", "effort": "source does not state", "tools_search": "Python visual tool: crop/zoom/draw/search actions", "shots": "source does not state", "trials_samples": "source does not state", "aggregation_pass_k": "accuracy", "context": "source does not state", "output_cap": "source does not state", "temperature": "source does not state", "top_p": "source does not state", "judge": "benchmark-specified; exact StepFun judge not stated", "harness_agent": "official Kimi V* evaluation with Python visual tool", "dataset_version_split": "V* Bench official 191-question release", "multimodal_input": true }, "matches_canonical": false, "source_type": "official_model_card", "audit_status": "verified", "notes": "Official StepFun benchmark-specific tool table. Direct locked official score authority reports V*=96.9; the StepFun surface is retained as physical corroboration." }, { "model_id": "kimi-k2.6", "benchmark_id": "worldvqa", "score": 55.98, "reference_url": "https://static.stepfun.com/blog/step-3.7-flash/", "reported_setting": { "quantization_config": "common release evaluation; artifact precision not established", "configuration": "base Step evaluation configuration", "mode": "source does not state", "effort": "source does not state", "tools_search": "Visual Search", "shots": "source does not state", "trials_samples": "source does not state", "aggregation_pass_k": "source-reported score", "context": "source does not state", "output_cap": "source does not state", "temperature": "source does not state", "top_p": "source does not state", "judge": "LLM-as-judge; exact StepFun judge not stated", "harness_agent": "official StepFun evaluation; exact harness not stated", "dataset_version_split": "WorldVQA first-eight-category 3,000-item release", "multimodal_input": true }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Official StepFun benchmark-specific tool table." }, { "model_id": "step-3.5-flash", "benchmark_id": "aa_lcr", "score": 45.5, "reference_url": "https://static.stepfun.com/blog/step-3.7-flash/", "reported_setting": { "quantization_config": "common release evaluation; artifact precision not established", "configuration": "base Step evaluation configuration", "mode": "source does not state", "effort": "source does not state", "tools_search": "none", "shots": "source does not state", "trials_samples": 16, "aggregation_pass_k": "average accuracy over 16 samples", "context": "source does not state", "output_cap": "source does not state", "temperature": "source does not state", "top_p": "source does not state", "judge": "AA equality checker; StepFun exact judge revision undisclosed", "harness_agent": "official StepFun evaluation; exact harness not stated", "dataset_version_split": "AA-LCR 100-question release", "multimodal_input": false }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Official StepFun launch comparison table." }, { "model_id": "step-3.5-flash", "benchmark_id": "browsecomp", "score": 69.0, "reference_url": "https://static.stepfun.com/blog/step-3.7-flash/", "reported_setting": { "quantization_config": "common release evaluation; artifact precision not established", "configuration": "base Step evaluation configuration", "mode": "source does not state", "effort": "source does not state", "tools_search": "web browsing agent", "shots": "source does not state", "trials_samples": "source does not state", "aggregation_pass_k": "source-reported score", "context": "source does not state", "output_cap": "source does not state", "temperature": "source does not state", "top_p": "source does not state", "judge": "benchmark-specified; exact StepFun judge not stated", "harness_agent": "official StepFun evaluation; exact harness not stated", "dataset_version_split": "BrowseComp 1,266-question release", "multimodal_input": false }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "Official StepFun launch comparison table." }, { "model_id": "step-3.5-flash", "benchmark_id": "claw_eval_pass3", "score": 43.6, "reference_url": "https://static.stepfun.com/blog/step-3.7-flash/", "reported_setting": { "quantization_config": "common release evaluation; artifact precision not established", "configuration": "base Step evaluation configuration", "mode": "source does not state", "effort": "source does not state", "tools_search": "official Claw-Eval agent environment", "shots": "source does not state", "trials_samples": 3, "aggregation_pass_k": "Pass^3: all three trials must pass", "context": "source does not state", "output_cap": "source does not state", "temperature": "source does not state", "top_p": "source does not state", "judge": "full-trajectory completion/safety/robustness grading", "harness_agent": "official StepFun evaluation; exact harness not stated", "dataset_version_split": "Claw-Eval v1.1 non-multimodal: 161 general + 38 multi-turn tasks", "multimodal_input": false }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "Official StepFun launch comparison table." }, { "model_id": "step-3.5-flash", "benchmark_id": "deepsearchqa_acc", "score": 73.4, "reference_url": "https://static.stepfun.com/blog/step-3.7-flash/", "reported_setting": { "quantization_config": "common release evaluation; artifact precision not established", "configuration": "base Step evaluation configuration", "mode": "source does not state", "effort": "source does not state", "tools_search": "web search/research agent", "shots": "source does not state", "trials_samples": "source does not state", "aggregation_pass_k": "source-reported score", "context": "source does not state", "output_cap": "source does not state", "temperature": "source does not state", "top_p": "source does not state", "judge": "Gemini 2.5 Flash with official Kaggle grading prompt", "harness_agent": "official StepFun evaluation; exact harness not stated", "dataset_version_split": "DeepSearchQA 900-prompt eval split", "multimodal_input": false }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "Official StepFun launch comparison table." }, { "model_id": "step-3.5-flash", "benchmark_id": "deepsearchqa_f1", "score": 85.5, "reference_url": "https://static.stepfun.com/blog/step-3.7-flash/", "reported_setting": { "quantization_config": "common release evaluation; artifact precision not established", "configuration": "base Step evaluation configuration", "mode": "source does not state", "effort": "source does not state", "tools_search": "web search/research agent", "shots": "source does not state", "trials_samples": "source does not state", "aggregation_pass_k": "source-reported score", "context": "source does not state", "output_cap": "source does not state", "temperature": "source does not state", "top_p": "source does not state", "judge": "Gemini 2.5 Flash with official Kaggle grading prompt", "harness_agent": "official StepFun evaluation; exact harness not stated", "dataset_version_split": "DeepSearchQA 900-prompt eval split", "multimodal_input": false }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "Official StepFun launch comparison table." }, { "model_id": "step-3.5-flash", "benchmark_id": "gdpval_aa_elo", "score": 1055.0, "reference_url": "https://static.stepfun.com/blog/step-3.7-flash/", "reported_setting": { "quantization_config": "common release evaluation; artifact precision not established", "configuration": "base Step evaluation configuration", "mode": "source does not state", "effort": "source does not state", "tools_search": "official Artificial Analysis Stirrup harness", "shots": "source does not state", "trials_samples": "source does not state", "aggregation_pass_k": "source-reported score", "context": "source does not state", "output_cap": "source does not state", "temperature": "source does not state", "top_p": "source does not state", "judge": "pairwise Elo evaluation anchored to human experts", "harness_agent": "official Artificial Analysis Stirrup leaderboard", "dataset_version_split": "GDPval-AA historical 220-task Stirrup leaderboard quoted 2026-05-29", "multimodal_input": false }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Official StepFun launch comparison table." }, { "model_id": "step-3.5-flash", "benchmark_id": "hle_tools", "score": 35.7, "reference_url": "https://static.stepfun.com/blog/step-3.7-flash/", "reported_setting": { "quantization_config": "common release evaluation; artifact precision not established", "configuration": "base Step evaluation configuration", "mode": "source does not state", "effort": "source does not state", "tools_search": "tools enabled; exact suite undisclosed", "shots": "source does not state", "trials_samples": "source does not state", "aggregation_pass_k": "source-reported score", "context": "source does not state", "output_cap": "source does not state", "temperature": "source does not state", "top_p": "source does not state", "judge": "benchmark-specified; exact StepFun judge not stated", "harness_agent": "official StepFun evaluation; exact harness not stated", "dataset_version_split": "HLE finalized full set", "multimodal_input": true }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "Official StepFun launch comparison table." }, { "model_id": "step-3.5-flash", "benchmark_id": "researchrubrics", "score": 65.3, "reference_url": "https://static.stepfun.com/blog/step-3.7-flash/", "reported_setting": { "quantization_config": "common release evaluation; artifact precision not established", "configuration": "base Step evaluation configuration", "mode": "source does not state", "effort": "source does not state", "tools_search": "research tools", "shots": "source does not state", "trials_samples": "source does not state", "aggregation_pass_k": "source-reported score", "context": "source does not state", "output_cap": "source does not state", "temperature": "source does not state", "top_p": "source does not state", "judge": "weighted binary rubric compliance evaluation", "harness_agent": "official StepFun evaluation; exact harness not stated", "dataset_version_split": "ResearchRubrics 101-task official release", "multimodal_input": false }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "Official StepFun launch comparison table." }, { "model_id": "step-3.5-flash", "benchmark_id": "swe_bench_multilingual", "score": 67.4, "reference_url": "https://static.stepfun.com/blog/step-3.7-flash/", "reported_setting": { "quantization_config": "common release evaluation; artifact precision not established", "configuration": "base Step evaluation configuration", "mode": "source does not state", "effort": "source does not state", "tools_search": "agentic coding tools", "shots": "source does not state", "trials_samples": "source does not state", "aggregation_pass_k": "% resolved", "context": "source does not state", "output_cap": "source does not state", "temperature": "source does not state", "top_p": "source does not state", "judge": "benchmark-specified; exact StepFun judge not stated", "harness_agent": "source does not state coding scaffold", "dataset_version_split": "SWE-bench Multilingual official 300-instance test set", "multimodal_input": false }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "Official StepFun launch comparison table." }, { "model_id": "step-3.5-flash", "benchmark_id": "swe_bench_pro", "score": 51.3, "reference_url": "https://static.stepfun.com/blog/step-3.7-flash/", "reported_setting": { "quantization_config": "common release evaluation; artifact precision not established", "configuration": "base Step evaluation configuration", "mode": "source does not state", "effort": "source does not state", "tools_search": "agentic coding tools", "shots": "source does not state", "trials_samples": "source does not state", "aggregation_pass_k": "% resolved", "context": "source does not state", "output_cap": "source does not state", "temperature": "source does not state", "top_p": "source does not state", "judge": "benchmark-specified; exact StepFun judge not stated", "harness_agent": "source does not state coding scaffold", "dataset_version_split": "source label only", "multimodal_input": false }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "Official StepFun launch comparison table." }, { "model_id": "step-3.5-flash", "benchmark_id": "terminal_bench_2_1", "score": 53.4, "reference_url": "https://static.stepfun.com/blog/step-3.7-flash/", "reported_setting": { "quantization_config": "common release evaluation; artifact precision not established", "configuration": "base Step evaluation configuration", "mode": "source does not state", "effort": "source does not state", "tools_search": "terminal tools", "shots": "source does not state", "trials_samples": "source does not state", "aggregation_pass_k": "% tasks resolved", "context": "source does not state", "output_cap": "source does not state", "temperature": "source does not state", "top_p": "source does not state", "judge": "benchmark-specified; exact StepFun judge not stated", "harness_agent": "source does not state terminal scaffold", "dataset_version_split": "Terminal-Bench 2.1", "multimodal_input": false }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Official StepFun launch comparison table." }, { "model_id": "step-3.7-flash", "benchmark_id": "aa_lcr", "score": 63.9, "reference_url": "https://static.stepfun.com/blog/step-3.7-flash/", "reported_setting": { "quantization_config": "common release evaluation; artifact precision not established", "configuration": "base Step evaluation configuration", "mode": "source does not state", "effort": "source does not state", "tools_search": "none", "shots": "source does not state", "trials_samples": 16, "aggregation_pass_k": "average accuracy over 16 samples", "context": "source does not state", "output_cap": "source does not state", "temperature": "source does not state", "top_p": "source does not state", "judge": "AA equality checker; StepFun exact judge revision undisclosed", "harness_agent": "official StepFun evaluation; exact harness not stated", "dataset_version_split": "AA-LCR 100-question release", "multimodal_input": false }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Official StepFun launch comparison table." }, { "model_id": "step-3.7-flash", "benchmark_id": "android_daily", "score": 61.87, "reference_url": "https://static.stepfun.com/blog/step-3.7-flash/", "reported_setting": { "quantization_config": "common release evaluation; artifact precision not established", "configuration": "base Step evaluation configuration", "mode": "source does not state", "effort": "source does not state", "tools_search": "physical Android phone-use action environment", "shots": "source does not state", "trials_samples": "source does not state", "aggregation_pass_k": "pass@1 task success rate", "context": "source does not state", "output_cap": "source does not state", "temperature": "source does not state", "top_p": "source does not state", "judge": "benchmark-specified; exact StepFun judge not stated", "harness_agent": "StepFun phone-use stack; exact launch harness undisclosed", "dataset_version_split": "AndroidDaily v1, 350 tasks / 94 apps", "multimodal_input": true }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "Corrected directly from the current locked official DOM. The prior research's 48.2/43.7/43.3/37.6 transcription was stale and is not retained." }, { "model_id": "step-3.7-flash", "benchmark_id": "browsecomp", "score": 75.8, "reference_url": "https://static.stepfun.com/blog/step-3.7-flash/", "reported_setting": { "quantization_config": "common release evaluation; artifact precision not established", "configuration": "base Step evaluation configuration", "mode": "source does not state", "effort": "source does not state", "tools_search": "web browsing agent", "shots": "source does not state", "trials_samples": "source does not state", "aggregation_pass_k": "source-reported score", "context": "source does not state", "output_cap": "source does not state", "temperature": "source does not state", "top_p": "source does not state", "judge": "benchmark-specified; exact StepFun judge not stated", "harness_agent": "official StepFun evaluation; exact harness not stated", "dataset_version_split": "BrowseComp 1,266-question release", "multimodal_input": false }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "Official StepFun launch comparison table." }, { "model_id": "step-3.7-flash", "benchmark_id": "claw_eval_pass3", "score": 67.1, "reference_url": "https://static.stepfun.com/blog/step-3.7-flash/", "reported_setting": { "quantization_config": "common release evaluation; artifact precision not established", "configuration": "base Step evaluation configuration", "mode": "source does not state", "effort": "source does not state", "tools_search": "official Claw-Eval agent environment", "shots": "source does not state", "trials_samples": 3, "aggregation_pass_k": "Pass^3: all three trials must pass", "context": "source does not state", "output_cap": "source does not state", "temperature": "source does not state", "top_p": "source does not state", "judge": "full-trajectory completion/safety/robustness grading", "harness_agent": "official StepFun evaluation; exact harness not stated", "dataset_version_split": "Claw-Eval v1.1 non-multimodal: 161 general + 38 multi-turn tasks", "multimodal_input": false }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "Official StepFun launch comparison table." }, { "model_id": "step-3.7-flash", "benchmark_id": "deepsearchqa_acc", "score": 81.7, "reference_url": "https://static.stepfun.com/blog/step-3.7-flash/", "reported_setting": { "quantization_config": "common release evaluation; artifact precision not established", "configuration": "base Step evaluation configuration", "mode": "source does not state", "effort": "source does not state", "tools_search": "web search/research agent", "shots": "source does not state", "trials_samples": "source does not state", "aggregation_pass_k": "source-reported score", "context": "source does not state", "output_cap": "source does not state", "temperature": "source does not state", "top_p": "source does not state", "judge": "Gemini 2.5 Flash with official Kaggle grading prompt", "harness_agent": "official StepFun evaluation; exact harness not stated", "dataset_version_split": "DeepSearchQA 900-prompt eval split", "multimodal_input": false }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "Official StepFun launch comparison table." }, { "model_id": "step-3.7-flash", "benchmark_id": "deepsearchqa_f1", "score": 92.8, "reference_url": "https://static.stepfun.com/blog/step-3.7-flash/", "reported_setting": { "quantization_config": "common release evaluation; artifact precision not established", "configuration": "base Step evaluation configuration", "mode": "source does not state", "effort": "source does not state", "tools_search": "web search/research agent", "shots": "source does not state", "trials_samples": "source does not state", "aggregation_pass_k": "source-reported score", "context": "source does not state", "output_cap": "source does not state", "temperature": "source does not state", "top_p": "source does not state", "judge": "Gemini 2.5 Flash with official Kaggle grading prompt", "harness_agent": "official StepFun evaluation; exact harness not stated", "dataset_version_split": "DeepSearchQA 900-prompt eval split", "multimodal_input": false }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "Official StepFun launch comparison table." }, { "model_id": "step-3.7-flash", "benchmark_id": "gpqa_diamond", "score": 78.41, "reference_url": "https://raw.githubusercontent.com/stepfun-ai/Step-3.7-Flash/e2132efea2836f5d1638f8145e2178b2f6c9c747/README.md", "reported_setting": { "quantization_config": "NVFP4 weights/activations with FP8 KV cache", "configuration": "NVFP4 base decoding without MTP speculative decoding", "mode": "source does not state", "effort": "source does not state", "tools_search": "none", "shots": "source does not state", "trials_samples": 16, "aggregation_pass_k": "item accuracy over 3,168 records", "context": "source does not state", "output_cap": "source does not state", "temperature": "source does not state", "top_p": "source does not state", "judge": "benchmark-specified; exact StepFun judge not stated", "harness_agent": "official StepFun evaluation; exact harness not stated", "dataset_version_split": "GPQA Diamond 198 items", "multimodal_input": false }, "matches_canonical": false, "source_type": "official_repository", "audit_status": "verified", "notes": "Configuration-specific quality result over 3,168 records; kept distinct from non-quantized/common evaluation.", "candidates": [ { "score": 77.81, "reference_url": "https://raw.githubusercontent.com/stepfun-ai/Step-3.7-Flash/e2132efea2836f5d1638f8145e2178b2f6c9c747/README.md", "source_type": "official_repository", "reported_setting": { "quantization_config": "NVFP4 weights/activations with FP8 KV cache", "configuration": "NVFP4 with MTP speculative decoding, 3 speculative tokens", "mode": "source does not state", "effort": "source does not state", "tools_search": "none", "shots": "source does not state", "trials_samples": 16, "aggregation_pass_k": "item accuracy over 3,168 records", "context": "source does not state", "output_cap": "source does not state", "temperature": "source does not state", "top_p": "source does not state", "judge": "benchmark-specified; exact StepFun judge not stated", "harness_agent": "official StepFun evaluation; exact harness not stated", "dataset_version_split": "GPQA Diamond 198 items", "multimodal_input": false }, "notes": "Configuration-specific quality result over 3,168 records; kept distinct from non-quantized/common evaluation." } ] }, { "model_id": "step-3.7-flash", "benchmark_id": "hle_text", "score": 49.7, "reference_url": "https://static.stepfun.com/blog/step-3.7-flash/", "reported_setting": { "quantization_config": "common release evaluation; artifact precision not established", "configuration": "base Step evaluation configuration", "mode": "source does not state", "effort": "source does not state", "tools_search": "none", "shots": "source does not state", "trials_samples": "source does not state", "aggregation_pass_k": "source-reported score", "context": "source does not state", "output_cap": "source does not state", "temperature": "source does not state", "top_p": "source does not state", "judge": "benchmark-specified; exact StepFun judge not stated", "harness_agent": "official StepFun evaluation; exact harness not stated", "dataset_version_split": "HLE finalized 2,158-question text-only subset", "multimodal_input": false }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "Physically embedded alternative in the HLE tools cell; mapped to the distinct text-only identity." }, { "model_id": "step-3.7-flash", "benchmark_id": "hle_tools", "score": 48.1, "reference_url": "https://raw.githubusercontent.com/stepfun-ai/Step-3.7-Flash/e2132efea2836f5d1638f8145e2178b2f6c9c747/README.md", "reported_setting": { "quantization_config": "common release evaluation; artifact precision not established", "configuration": "base Step evaluation configuration", "mode": "source does not state", "effort": "source does not state", "tools_search": "tools enabled; exact suite undisclosed", "shots": "source does not state", "trials_samples": "source does not state", "aggregation_pass_k": "source-reported score", "context": "source does not state", "output_cap": "source does not state", "temperature": "source does not state", "top_p": "source does not state", "judge": "benchmark-specified; exact StepFun judge not stated", "harness_agent": "official StepFun evaluation; exact harness not stated", "dataset_version_split": "HLE finalized full set", "multimodal_input": true }, "matches_canonical": true, "source_type": "official_repository", "audit_status": "verified", "notes": "The post-launch repository prose reports 48.1 tools, distinct from the 47.2 launch table/chart.", "candidates": [ { "score": 47.2, "reference_url": "https://static.stepfun.com/blog/step-3.7-flash/", "source_type": "official_blog", "reported_setting": { "quantization_config": "common release evaluation; artifact precision not established", "configuration": "base Step evaluation configuration", "mode": "source does not state", "effort": "source does not state", "tools_search": "tools enabled; exact suite undisclosed", "shots": "source does not state", "trials_samples": "source does not state", "aggregation_pass_k": "source-reported score", "context": "source does not state", "output_cap": "source does not state", "temperature": "source does not state", "top_p": "source does not state", "judge": "benchmark-specified; exact StepFun judge not stated", "harness_agent": "official StepFun evaluation; exact harness not stated", "dataset_version_split": "HLE finalized full set", "multimodal_input": true }, "notes": "Official StepFun launch comparison table." } ] }, { "model_id": "step-3.7-flash", "benchmark_id": "hr_bench_4k", "score": 89.13, "reference_url": "https://static.stepfun.com/blog/step-3.7-flash/", "reported_setting": { "quantization_config": "common release evaluation; artifact precision not established", "configuration": "base Step evaluation configuration", "mode": "source does not state", "effort": "source does not state", "tools_search": "Python visual tool: crop/zoom/draw/search actions", "shots": "source does not state", "trials_samples": "source does not state", "aggregation_pass_k": "accuracy", "context": "source does not state", "output_cap": "source does not state", "temperature": "source does not state", "top_p": "source does not state", "judge": "benchmark-specified; exact StepFun judge not stated", "harness_agent": "official StepFun evaluation; exact harness not stated", "dataset_version_split": "HR-Bench 4K official 800-row release", "multimodal_input": true }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Official StepFun benchmark-specific tool table." }, { "model_id": "step-3.7-flash", "benchmark_id": "hr_bench_8k", "score": 86.34, "reference_url": "https://static.stepfun.com/blog/step-3.7-flash/", "reported_setting": { "quantization_config": "common release evaluation; artifact precision not established", "configuration": "base Step evaluation configuration", "mode": "source does not state", "effort": "source does not state", "tools_search": "Python visual tool: crop/zoom/draw/search actions", "shots": "source does not state", "trials_samples": "source does not state", "aggregation_pass_k": "accuracy", "context": "source does not state", "output_cap": "source does not state", "temperature": "source does not state", "top_p": "source does not state", "judge": "benchmark-specified; exact StepFun judge not stated", "harness_agent": "official StepFun evaluation; exact harness not stated", "dataset_version_split": "HR-Bench 8K official 800-row release", "multimodal_input": true }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Official StepFun benchmark-specific tool table." }, { "model_id": "step-3.7-flash", "benchmark_id": "researchrubrics", "score": 71.7, "reference_url": "https://static.stepfun.com/blog/step-3.7-flash/", "reported_setting": { "quantization_config": "common release evaluation; artifact precision not established", "configuration": "base Step evaluation configuration", "mode": "source does not state", "effort": "source does not state", "tools_search": "research tools", "shots": "source does not state", "trials_samples": "source does not state", "aggregation_pass_k": "source-reported score", "context": "source does not state", "output_cap": "source does not state", "temperature": "source does not state", "top_p": "source does not state", "judge": "weighted binary rubric compliance evaluation", "harness_agent": "official StepFun evaluation; exact harness not stated", "dataset_version_split": "ResearchRubrics 101-task official release", "multimodal_input": false }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "Official StepFun launch comparison table." }, { "model_id": "step-3.7-flash", "benchmark_id": "simplevqa", "score": 79.16, "reference_url": "https://static.stepfun.com/blog/step-3.7-flash/", "reported_setting": { "quantization_config": "common release evaluation; artifact precision not established", "configuration": "base Step evaluation configuration", "mode": "source does not state", "effort": "source does not state", "tools_search": "Visual Search", "shots": "source does not state", "trials_samples": "source does not state", "aggregation_pass_k": "source-reported score", "context": "source does not state", "output_cap": "source does not state", "temperature": "source does not state", "top_p": "source does not state", "judge": "LLM-as-judge; exact StepFun judge not stated", "harness_agent": "official StepFun evaluation; exact harness not stated", "dataset_version_split": "SimpleVQA official 2,025-item test release", "multimodal_input": true }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Official StepFun benchmark-specific tool table." }, { "model_id": "step-3.7-flash", "benchmark_id": "swe_bench_multilingual", "score": 72.4, "reference_url": "https://static.stepfun.com/blog/step-3.7-flash/", "reported_setting": { "quantization_config": "common release evaluation; artifact precision not established", "configuration": "base Step evaluation configuration", "mode": "source does not state", "effort": "source does not state", "tools_search": "agentic coding tools", "shots": "source does not state", "trials_samples": "source does not state", "aggregation_pass_k": "% resolved", "context": "source does not state", "output_cap": "source does not state", "temperature": "source does not state", "top_p": "source does not state", "judge": "benchmark-specified; exact StepFun judge not stated", "harness_agent": "source does not state coding scaffold", "dataset_version_split": "SWE-bench Multilingual official 300-instance test set", "multimodal_input": false }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "Official StepFun launch comparison table." }, { "model_id": "step-3.7-flash", "benchmark_id": "swe_bench_pro", "score": 56.3, "reference_url": "https://static.stepfun.com/blog/step-3.7-flash/", "reported_setting": { "quantization_config": "common release evaluation; artifact precision not established", "configuration": "base Step evaluation configuration", "mode": "source does not state", "effort": "source does not state", "tools_search": "agentic coding tools", "shots": "source does not state", "trials_samples": "source does not state", "aggregation_pass_k": "% resolved", "context": "source does not state", "output_cap": "source does not state", "temperature": "source does not state", "top_p": "source does not state", "judge": "benchmark-specified; exact StepFun judge not stated", "harness_agent": "source does not state coding scaffold", "dataset_version_split": "source label only", "multimodal_input": false }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "Official StepFun launch comparison table." }, { "model_id": "step-3.7-flash", "benchmark_id": "swe_bench_verified", "score": 76.5, "reference_url": "https://static.stepfun.com/blog/step-3.7-flash/", "reported_setting": { "quantization_config": "common release evaluation; artifact precision not established", "configuration": "base Step evaluation configuration", "mode": "source does not state", "effort": "source does not state", "tools_search": "agentic coding tools", "shots": "source does not state", "trials_samples": "source does not state", "aggregation_pass_k": "% resolved", "context": "source does not state", "output_cap": "source does not state", "temperature": "source does not state", "top_p": "source does not state", "judge": "benchmark-specified; exact StepFun judge not stated", "harness_agent": "source does not state coding scaffold", "dataset_version_split": "source label only", "multimodal_input": false }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "Official StepFun launch comparison table.", "candidates": [ { "score": 73.7, "reference_url": "https://static.stepfun.com/blog/step-3.7-flash/", "source_type": "official_blog", "reported_setting": { "quantization_config": "common release evaluation; artifact precision not established", "configuration": "base executor", "mode": "source does not state", "effort": "source does not state", "tools_search": "agentic coding tools", "shots": "source does not state", "trials_samples": "source does not state", "aggregation_pass_k": "% resolved", "context": "source does not state", "output_cap": "source does not state", "temperature": "source does not state", "top_p": "source does not state", "judge": "benchmark-specified; exact StepFun judge not stated", "harness_agent": "Claude Opus 4.6 internal reproduce harness", "dataset_version_split": "SWE-bench Verified 500", "multimodal_input": false }, "notes": "Advisor is an inference configuration of Step 3.7 Flash, not a separate model. Cost is excluded separately." }, { "score": 76.3, "reference_url": "https://static.stepfun.com/blog/step-3.7-flash/", "source_type": "official_blog", "reported_setting": { "quantization_config": "common release evaluation; artifact precision not established", "configuration": "Advisor Mode", "mode": "source does not state", "effort": "source does not state", "tools_search": "agentic coding tools", "shots": "source does not state", "trials_samples": "source does not state", "aggregation_pass_k": "% resolved", "context": "source does not state", "output_cap": "source does not state", "temperature": "source does not state", "top_p": "source does not state", "judge": "benchmark-specified; exact StepFun judge not stated", "harness_agent": "Claude Opus 4.6 internal reproduce harness", "dataset_version_split": "SWE-bench Verified 500", "multimodal_input": false }, "notes": "Advisor is an inference configuration of Step 3.7 Flash, not a separate model. Cost is excluded separately." } ] }, { "model_id": "step-3.7-flash", "benchmark_id": "terminal_bench_2_1", "score": 59.55, "reference_url": "https://static.stepfun.com/blog/step-3.7-flash/script.js", "reported_setting": { "quantization_config": "common release evaluation; artifact precision not established", "configuration": "base Step evaluation configuration", "mode": "source does not state", "effort": "source does not state", "tools_search": "terminal tools", "shots": "source does not state", "trials_samples": "source does not state", "aggregation_pass_k": "% tasks resolved", "context": "source does not state", "output_cap": "source does not state", "temperature": "source does not state", "top_p": "source does not state", "judge": "benchmark-specified; exact StepFun judge not stated", "harness_agent": "source does not state terminal scaffold", "dataset_version_split": "Terminal-Bench 2.1", "multimodal_input": false }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Official StepFun launch comparison table. Physical rendering 59.6 is a rounded display of the locked official script value 59.55; semantic mutation uses the precise value and this rendering is a duplicate. Direct locked official score authority reports Terminal-Bench 2.1=59.55; the StepFun surface is retained as physical corroboration." }, { "model_id": "step-3.7-flash", "benchmark_id": "vstar", "score": 95.29, "reference_url": "https://static.stepfun.com/blog/step-3.7-flash/", "reported_setting": { "quantization_config": "common release evaluation; artifact precision not established", "configuration": "base Step evaluation configuration", "mode": "source does not state", "effort": "source does not state", "tools_search": "Python visual tool: crop/zoom/draw/search actions", "shots": "source does not state", "trials_samples": "source does not state", "aggregation_pass_k": "accuracy", "context": "source does not state", "output_cap": "source does not state", "temperature": "source does not state", "top_p": "source does not state", "judge": "benchmark-specified; exact StepFun judge not stated", "harness_agent": "official StepFun evaluation; exact harness not stated", "dataset_version_split": "V* Bench official 191-question release", "multimodal_input": true }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Official StepFun benchmark-specific tool table." }, { "model_id": "step-3.7-flash", "benchmark_id": "worldvqa", "score": 58.1, "reference_url": "https://static.stepfun.com/blog/step-3.7-flash/", "reported_setting": { "quantization_config": "common release evaluation; artifact precision not established", "configuration": "base Step evaluation configuration", "mode": "source does not state", "effort": "source does not state", "tools_search": "Visual Search", "shots": "source does not state", "trials_samples": "source does not state", "aggregation_pass_k": "source-reported score", "context": "source does not state", "output_cap": "source does not state", "temperature": "source does not state", "top_p": "source does not state", "judge": "LLM-as-judge; exact StepFun judge not stated", "harness_agent": "official StepFun evaluation; exact harness not stated", "dataset_version_split": "WorldVQA first-eight-category 3,000-item release", "multimodal_input": true }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Official StepFun benchmark-specific tool table." }, { "model_id": "claude-sonnet-5", "benchmark_id": "code_arena_frontend_elo", "score": 1540, "reference_url": "https://arena.ai/leaderboard/code", "reported_setting": { "mode": "thinking", "effort": "high", "prompt_style": "not disclosed", "temperature": "not disclosed", "context": "not disclosed", "output_limit": "not disclosed", "trials": "not disclosed", "tools": "web application generation", "sampling": "not disclosed", "aggregation": "Arena Elo from pairwise human votes", "judge": "human pairwise preference votes", "harness": "Code Arena WebDev", "dataset_version": "Code Arena WebDev live leaderboard", "dataset_split": "live WebDev Arena" }, "matches_canonical": true, "source_type": "leaderboard", "audit_status": "verified", "notes": "Current official leaderboard update; the Google launch table remains preserved as the August release snapshot.", "candidates": [ { "score": 1541, "reference_url": "https://storage.googleapis.com/deepmind-media/gemini/gemini_3-7_flash_model_evaluation.pdf", "source_type": "tech_report", "reported_setting": { "mode": "thinking", "effort": "max", "prompt_style": "not disclosed", "temperature": "not disclosed", "context": "not disclosed", "output_limit": "not disclosed", "trials": "not disclosed", "tools": "web application generation", "sampling": "not disclosed", "aggregation": "Arena Elo from pairwise human votes", "judge": "human pairwise preference votes", "harness": "Code Arena WebDev", "dataset_version": "Code Arena WebDev live leaderboard", "dataset_split": "live WebDev Arena" }, "notes": "Elo. Physical rendering retained even when semantically identical to another official rendering." } ] }, { "model_id": "claude-sonnet-5", "benchmark_id": "frontiercode_main_v1_1", "score": 42.7, "reference_url": "https://storage.googleapis.com/deepmind-media/gemini/gemini_3-7_flash_model_evaluation.pdf", "reported_setting": { "mode": "thinking", "effort": "max", "prompt_style": "not disclosed", "temperature": "not disclosed", "context": "not disclosed", "output_limit": "not disclosed", "trials": "not disclosed", "tools": "coding agent with guarded internet access", "sampling": "not disclosed", "judge": "blocking held-out functional criteria plus weighted maintainer-authored rubric criteria", "harness": "FrontierCode 1.1 official leaderboard", "dataset_version": "FrontierCode 1.1", "dataset_split": "100-task Main split" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Score; production code quality. Physical rendering retained even when semantically identical to another official rendering." }, { "model_id": "claude-sonnet-5", "benchmark_id": "gdp_pdf", "score": 28.0, "reference_url": "https://storage.googleapis.com/deepmind-media/gemini/gemini_3-7_flash_model_evaluation.pdf", "reported_setting": { "mode": "thinking", "effort": "max", "prompt_style": "not disclosed", "temperature": "not disclosed", "context": "not disclosed", "output_limit": "not disclosed", "trials": "not disclosed", "dataset_version": "GDP.pdf 100-task held-out test set", "dataset_split": "test", "multimodal_input": true, "tools": "none", "sampling": "not disclosed", "aggregation": "strict all_pass/mean", "judge": "Gemini 3.5 Flash rubric judge", "harness": "Google self-computed GDP.pdf evaluation" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Strict task pass rate; expert PDF comprehension. Physical rendering retained even when semantically identical to another official rendering." }, { "model_id": "claude-sonnet-5", "benchmark_id": "harvey_lab_aa", "score": 90.1, "reference_url": "https://storage.googleapis.com/deepmind-media/gemini/gemini_3-7_flash_model_evaluation.pdf", "reported_setting": { "mode": "thinking", "effort": "max", "prompt_style": "not disclosed", "temperature": "not disclosed", "context": "not disclosed", "output_limit": "not disclosed", "trials": "not disclosed", "tools": "Stirrup shell and document workflow tools", "sampling": "not disclosed", "aggregation": "rubric-criterion pass rate", "judge": "rubric criteria over extracted deliverables", "harness": "Artificial Analysis Harvey LAB-AA", "dataset_version": "Harvey LAB-AA", "dataset_split": "120 tasks" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Complex legal workflows. Physical rendering retained even when semantically identical to another official rendering." }, { "model_id": "claude-sonnet-5", "benchmark_id": "hle_verified", "score": 31.0, "reference_url": "https://storage.googleapis.com/deepmind-media/gemini/gemini_3-7_flash_model_evaluation.pdf", "reported_setting": { "mode": "thinking", "effort": "max", "prompt_style": "not disclosed", "temperature": "not disclosed", "context": "not disclosed", "output_limit": "not disclosed", "trials": "not disclosed", "tools": "none", "sampling": "not disclosed", "judge": "HLE-Verified answer evaluator", "harness": "Google self-computed HLE-Verified evaluation", "multimodal_input": true, "dataset_version": "HLE-Verified full verified set", "dataset_split": "Gold 668 plus Revision 1,143" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "1,811 items; Gold + Revision; Uncertain excluded. Physical rendering retained even when semantically identical to another official rendering." }, { "model_id": "claude-sonnet-5", "benchmark_id": "labbench2", "score": 80.1, "reference_url": "https://storage.googleapis.com/deepmind-media/gemini/gemini_3-7_flash_model_evaluation.pdf", "reported_setting": { "mode": "thinking", "effort": "max", "prompt_style": "not disclosed", "temperature": "not disclosed", "context": "not disclosed", "output_limit": "not disclosed", "trials": "not disclosed", "tools": "not disclosed", "sampling": "not disclosed", "judge": "not disclosed", "harness": "provider self-report; exact harness not disclosed", "multimodal_input": true, "dataset_version": "LABBench2", "dataset_split": "full 1,912-task suite" }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Biology real-world research tasks. Physical rendering retained even when semantically identical to another official rendering." }, { "model_id": "claude-sonnet-5", "benchmark_id": "lvbench", "score": 68.5, "reference_url": "https://storage.googleapis.com/deepmind-media/gemini/gemini_3-7_flash_model_evaluation.pdf", "reported_setting": { "mode": "thinking", "effort": "max", "prompt_style": "not disclosed", "temperature": "not disclosed", "context": "not disclosed", "output_limit": "not disclosed", "trials": "not disclosed", "tools": "none", "sampling": "not disclosed", "judge": "exact multiple-choice answer", "harness": "Google self-computed LVBench evaluation", "multimodal_input": true, "dataset_version": "LVBench", "dataset_split": "103 videos; 1,549 questions", "video_frames": 300 }, "matches_canonical": false, "source_type": "tech_report", "audit_status": "verified", "notes": "Long video understanding. Physical rendering retained even when semantically identical to another official rendering." }, { "model_id": "gemini-3.6-flash", "benchmark_id": "code_arena_frontend_elo", "score": 1538, "reference_url": "https://storage.googleapis.com/deepmind-media/gemini/gemini_3-7_flash_model_evaluation.pdf", "reported_setting": { "mode": "thinking", "effort": "not disclosed", "prompt_style": "not disclosed", "temperature": "not disclosed", "context": "not disclosed", "output_limit": "not disclosed", "trials": "not disclosed", "tools": "web application generation", "sampling": "not disclosed", "aggregation": "Arena Elo from pairwise human votes", "judge": "human pairwise preference votes", "harness": "Code Arena WebDev", "dataset_version": "Code Arena WebDev live leaderboard", "dataset_split": "live WebDev Arena" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Elo. Physical rendering retained even when semantically identical to another official rendering.", "candidates": [ { "score": 1537, "reference_url": "https://arena.ai/leaderboard/code", "source_type": "leaderboard", "reported_setting": { "mode": "thinking", "effort": "high", "prompt_style": "not disclosed", "temperature": "not disclosed", "context": "not disclosed", "output_limit": "not disclosed", "trials": "not disclosed", "tools": "web application generation", "sampling": "not disclosed", "aggregation": "Arena Elo from pairwise human votes", "judge": "human pairwise preference votes", "harness": "Code Arena WebDev", "dataset_version": "Code Arena WebDev live leaderboard", "dataset_split": "live WebDev Arena" }, "notes": "Current official leaderboard update; the Google launch table remains preserved as the August release snapshot." } ] }, { "model_id": "gemini-3.6-flash", "benchmark_id": "frontiercode_main_v1_1", "score": 34.4, "reference_url": "https://storage.googleapis.com/deepmind-media/gemini/gemini_3-7_flash_model_evaluation.pdf", "reported_setting": { "mode": "thinking", "effort": "not disclosed", "prompt_style": "not disclosed", "temperature": "not disclosed", "context": "not disclosed", "output_limit": "not disclosed", "trials": "not disclosed", "tools": "coding agent with guarded internet access", "sampling": "not disclosed", "judge": "blocking held-out functional criteria plus weighted maintainer-authored rubric criteria", "harness": "FrontierCode 1.1 official leaderboard", "dataset_version": "FrontierCode 1.1", "dataset_split": "100-task Main split" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Score; production code quality. Physical rendering retained even when semantically identical to another official rendering." }, { "model_id": "gemini-3.6-flash", "benchmark_id": "gdp_pdf", "score": 22.0, "reference_url": "https://storage.googleapis.com/deepmind-media/gemini/gemini_3-7_flash_model_evaluation.pdf", "reported_setting": { "mode": "thinking", "effort": "not disclosed", "prompt_style": "not disclosed", "temperature": "not disclosed", "context": "not disclosed", "output_limit": "not disclosed", "trials": "not disclosed", "dataset_version": "GDP.pdf 100-task held-out test set", "dataset_split": "test", "multimodal_input": true, "tools": "none", "sampling": "pass@1", "aggregation": "strict all_pass/mean", "judge": "Gemini 3.5 Flash rubric judge", "harness": "Google self-computed GDP.pdf evaluation" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Strict task pass rate; expert PDF comprehension. Physical rendering retained even when semantically identical to another official rendering." }, { "model_id": "gemini-3.6-flash", "benchmark_id": "harvey_lab_aa", "score": 85.1, "reference_url": "https://storage.googleapis.com/deepmind-media/gemini/gemini_3-7_flash_model_evaluation.pdf", "reported_setting": { "mode": "thinking", "effort": "not disclosed", "prompt_style": "not disclosed", "temperature": "not disclosed", "context": "not disclosed", "output_limit": "not disclosed", "trials": "not disclosed", "tools": "Stirrup shell and document workflow tools", "sampling": "not disclosed", "aggregation": "rubric-criterion pass rate", "judge": "rubric criteria over extracted deliverables", "harness": "Artificial Analysis Harvey LAB-AA", "dataset_version": "Harvey LAB-AA", "dataset_split": "120 tasks" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Complex legal workflows. Physical rendering retained even when semantically identical to another official rendering." }, { "model_id": "gemini-3.6-flash", "benchmark_id": "hle_verified", "score": 51.2, "reference_url": "https://storage.googleapis.com/deepmind-media/gemini/gemini_3-7_flash_model_evaluation.pdf", "reported_setting": { "mode": "thinking", "effort": "not disclosed", "prompt_style": "not disclosed", "temperature": "not disclosed", "context": "not disclosed", "output_limit": "not disclosed", "trials": "not disclosed", "tools": "none", "sampling": "pass@1", "judge": "HLE-Verified answer evaluator", "harness": "Google self-computed HLE-Verified evaluation", "multimodal_input": true, "dataset_version": "HLE-Verified full verified set", "dataset_split": "Gold 668 plus Revision 1,143" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "1,811 items; Gold + Revision; Uncertain excluded. Physical rendering retained even when semantically identical to another official rendering." }, { "model_id": "gemini-3.6-flash", "benchmark_id": "labbench2", "score": 76.1, "reference_url": "https://storage.googleapis.com/deepmind-media/gemini/gemini_3-7_flash_model_evaluation.pdf", "reported_setting": { "mode": "thinking", "effort": "not disclosed", "prompt_style": "not disclosed", "temperature": "not disclosed", "context": "not disclosed", "output_limit": "not disclosed", "trials": "not disclosed", "tools": "Linux terminal, bioinformatics tools, Python, R, and internet", "sampling": "pass@1", "judge": "task-specific exact/programmatic evaluator", "harness": "Google self-computed LABBench2 evaluation", "multimodal_input": true, "dataset_version": "LABBench2", "dataset_split": "full 1,912-task suite" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Biology real-world research tasks. Physical rendering retained even when semantically identical to another official rendering." }, { "model_id": "gemini-3.6-flash", "benchmark_id": "lvbench", "score": 84.2, "reference_url": "https://storage.googleapis.com/deepmind-media/gemini/gemini_3-7_flash_model_evaluation.pdf", "reported_setting": { "mode": "thinking", "effort": "not disclosed", "prompt_style": "not disclosed", "temperature": "not disclosed", "context": "not disclosed", "output_limit": "not disclosed", "trials": "not disclosed", "tools": "none", "sampling": "pass@1", "judge": "exact multiple-choice answer", "harness": "Google self-computed LVBench evaluation", "multimodal_input": true, "dataset_version": "LVBench", "dataset_split": "103 videos; 1,549 questions", "video_frames": 1024 }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Long video understanding. Physical rendering retained even when semantically identical to another official rendering." }, { "model_id": "gemini-3.6-flash", "benchmark_id": "osworld_2_0_2026_06_24_partial", "score": 33.8, "reference_url": "https://storage.googleapis.com/deepmind-media/gemini/gemini_3-7_flash_model_evaluation.pdf", "reported_setting": { "mode": "thinking", "effort": "not disclosed", "prompt_style": "not disclosed", "temperature": "not disclosed", "context": "not disclosed", "output_limit": "not disclosed", "trials": 3, "dataset_version": "osworld-v2-2026.06.24", "dataset_split": "108 tasks; before 2026-08-08 patch", "multimodal_input": true, "tools": "pyautogui actuation plus UI-specific function declarations", "sampling": "pass@1", "aggregation": "maximum partial score over three single-attempt runs", "judge": "official weighted checkpoint partial scorer", "harness": "OSWorld 2.0 Docker with official Gemini/Anthropic CUA harness", "screen": "1080p screenshot-only observation", "max_steps": 500 }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Weighted checkpoint partial score. Physical rendering retained even when semantically identical to another official rendering." }, { "model_id": "gemini-3.6-flash", "benchmark_id": "terminal_bench_3_0", "score": 5.4, "reference_url": "https://storage.googleapis.com/deepmind-media/gemini/gemini_3-7_flash_model_evaluation.pdf", "reported_setting": { "mode": "thinking", "effort": "not disclosed", "prompt_style": "not disclosed", "temperature": "not disclosed", "context": "not disclosed", "output_limit": "not disclosed", "trials": "not disclosed", "tools": "container terminal", "sampling": "pass@1", "judge": "Terminal-Bench 3 task-success evaluator", "harness": "Google self-computed mini-swe-agent with LiteLLM 1.96; two documented task compatibility modifications", "dataset_version": "Terminal-Bench 3.0 / FrontierBench", "dataset_split": "74 tasks" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "General agent capabilities. Physical rendering retained even when semantically identical to another official rendering." }, { "model_id": "gemini-3.7-flash", "benchmark_id": "charxiv_reasoning", "score": 84.5, "reference_url": "https://storage.googleapis.com/deepmind-media/gemini/gemini_3-7_flash_model_evaluation.pdf", "reported_setting": { "mode": "thinking", "effort": "medium", "prompt_style": "not disclosed", "temperature": "not disclosed", "context": "not disclosed", "output_limit": "not disclosed", "trials": "not disclosed", "tools": "none", "sampling": "pass@1", "judge": "not disclosed", "harness": "Google self-computed CharXiv evaluation", "multimodal_input": true, "dataset_version": "CharXiv Reasoning", "dataset_split": "1,000 validation questions" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Information synthesis from complex charts. Physical rendering retained even when semantically identical to another official rendering.", "candidates": [ { "score": 88.7, "reference_url": "https://storage.googleapis.com/deepmind-media/gemini/gemini_3-7_flash_model_evaluation.pdf", "source_type": "tech_report", "reported_setting": { "mode": "thinking", "effort": "medium", "prompt_style": "not disclosed", "temperature": "not disclosed", "context": "not disclosed", "output_limit": "not disclosed", "trials": "not disclosed", "tools": "search and code execution", "sampling": "pass@1", "judge": "not disclosed", "harness": "Google self-computed CharXiv evaluation", "multimodal_input": true, "dataset_version": "CharXiv Reasoning", "dataset_split": "1,000 validation questions" }, "notes": "Gemini uses search and code execution. Physical rendering retained even when semantically identical to another official rendering." } ] }, { "model_id": "gemini-3.7-flash", "benchmark_id": "code_arena_frontend_elo", "score": 1588, "reference_url": "https://storage.googleapis.com/deepmind-media/gemini/gemini_3-7_flash_model_evaluation.pdf", "reported_setting": { "mode": "thinking", "effort": "medium", "prompt_style": "not disclosed", "temperature": "not disclosed", "context": "not disclosed", "output_limit": "not disclosed", "trials": "not disclosed", "tools": "web application generation", "sampling": "not disclosed", "aggregation": "Arena Elo from pairwise human votes", "judge": "human pairwise preference votes", "harness": "Code Arena WebDev", "dataset_version": "Code Arena WebDev live leaderboard", "dataset_split": "live WebDev Arena" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Elo. Physical rendering retained even when semantically identical to another official rendering.", "candidates": [ { "score": 1587, "reference_url": "https://arena.ai/leaderboard/code", "source_type": "leaderboard", "reported_setting": { "mode": "thinking", "effort": "high", "prompt_style": "not disclosed", "temperature": "not disclosed", "context": "not disclosed", "output_limit": "not disclosed", "trials": "not disclosed", "tools": "web application generation", "sampling": "not disclosed", "aggregation": "Arena Elo from pairwise human votes", "judge": "human pairwise preference votes", "harness": "Code Arena WebDev", "dataset_version": "Code Arena WebDev live leaderboard", "dataset_split": "live WebDev Arena" }, "notes": "Current official leaderboard update; the Google launch table remains preserved as the August release snapshot." } ] }, { "model_id": "gemini-3.7-flash", "benchmark_id": "deep_swe_v1_1", "score": 65.3, "reference_url": "https://storage.googleapis.com/deepmind-media/gemini/gemini_3-7_flash_model_evaluation.pdf", "reported_setting": { "mode": "thinking", "effort": "high", "prompt_style": "not disclosed", "temperature": "not disclosed", "context": "not disclosed", "output_limit": "not disclosed", "trials": "not disclosed", "dataset_version": "DeepSWE v1.1", "dataset_split": "113 tasks across 91 repositories", "tools": "mini-swe-agent repository shell/editor toolset", "sampling": "pass@1", "judge": "DeepSWE programmatic verifier", "harness": "Google self-computed mini-swe-agent with LiteLLM 1.96" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Long-horizon software engineering. Physical rendering retained even when semantically identical to another official rendering.", "candidates": [ { "score": 65.0, "reference_url": "https://deepswe.datacurve.ai/", "source_type": "leaderboard", "reported_setting": { "mode": "thinking", "effort": "high", "prompt_style": "not disclosed", "temperature": "not disclosed", "context": "not disclosed", "output_limit": "not disclosed", "trials": "not disclosed", "dataset_version": "DeepSWE v1.1", "dataset_split": "113 tasks across 91 repositories", "tools": "mini-swe-agent repository shell/editor toolset", "sampling": "pass@1", "judge": "functional verifier plus regression checks", "harness": "official DeepSWE v1.1 mini-swe-agent leaderboard" }, "notes": "Current official leaderboard update; distinct harness or rounding/revision is preserved as an alternative." } ] }, { "model_id": "gemini-3.7-flash", "benchmark_id": "frontiercode_main_v1_1", "score": 43.6, "reference_url": "https://storage.googleapis.com/deepmind-media/gemini/gemini_3-7_flash_model_evaluation.pdf", "reported_setting": { "mode": "thinking", "effort": "medium", "prompt_style": "not disclosed", "temperature": "not disclosed", "context": "not disclosed", "output_limit": "not disclosed", "trials": "not disclosed", "tools": "coding agent with guarded internet access", "sampling": "not disclosed", "judge": "blocking held-out functional criteria plus weighted maintainer-authored rubric criteria", "harness": "FrontierCode 1.1 official leaderboard", "dataset_version": "FrontierCode 1.1", "dataset_split": "100-task Main split" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Score; production code quality. Physical rendering retained even when semantically identical to another official rendering." }, { "model_id": "gemini-3.7-flash", "benchmark_id": "gdm_mrcr_v2_8needle_upto_128k", "score": 97.0, "reference_url": "https://storage.googleapis.com/deepmind-media/gemini/gemini_3-7_flash_model_evaluation.pdf", "reported_setting": { "mode": "thinking", "effort": "medium", "prompt_style": "not disclosed", "temperature": "not disclosed", "context": "not disclosed", "output_limit": "not disclosed", "trials": "not disclosed", "tools": "none", "sampling": "pass@1", "judge": "strict MRCR scorer", "harness": "Google self-computed GDM eval_hub", "dataset_version": "GDM MRCR v2.1, 8 needles", "dataset_split": "cumulative through 128K; 484 examples", "sequence_length": "cumulative through 128K" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Cumulative through 128K. Physical rendering retained even when semantically identical to another official rendering." }, { "model_id": "gemini-3.7-flash", "benchmark_id": "gdp_pdf", "score": 34.0, "reference_url": "https://storage.googleapis.com/deepmind-media/gemini/gemini_3-7_flash_model_evaluation.pdf", "reported_setting": { "mode": "thinking", "effort": "medium", "prompt_style": "not disclosed", "temperature": "not disclosed", "context": "not disclosed", "output_limit": "not disclosed", "trials": "not disclosed", "dataset_version": "GDP.pdf 100-task held-out test set", "dataset_split": "test", "multimodal_input": true, "tools": "none", "sampling": "pass@1", "aggregation": "strict all_pass/mean", "judge": "Gemini 3.5 Flash rubric judge", "harness": "Google self-computed GDP.pdf evaluation" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Strict task pass rate; expert PDF comprehension. Physical rendering retained even when semantically identical to another official rendering." }, { "model_id": "gemini-3.7-flash", "benchmark_id": "gdpval_aa_v2_elo", "score": 1525, "reference_url": "https://storage.googleapis.com/deepmind-media/gemini/gemini_3-7_flash_model_evaluation.pdf", "reported_setting": { "mode": "thinking", "effort": "medium", "prompt_style": "not disclosed", "temperature": "not disclosed", "context": "not disclosed", "output_limit": "not disclosed", "trials": 1, "tools": "Web Fetch, Web Search, View Image, Code Exec, Finish, Abandon Task", "sampling": "pass@1", "aggregation": "blind pairwise Bradley-Terry Elo over one agentic submission per task", "judge": "panel of three frontier LLM judges sampled per comparison", "harness": "Artificial Analysis Stirrup; fresh E2B sandbox; 250-turn limit", "dataset_version": "GDPval-AA v2", "dataset_split": "220 tasks" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Elo; knowledge work. Physical rendering retained even when semantically identical to another official rendering." }, { "model_id": "gemini-3.7-flash", "benchmark_id": "harvey_lab_aa", "score": 90.7, "reference_url": "https://storage.googleapis.com/deepmind-media/gemini/gemini_3-7_flash_model_evaluation.pdf", "reported_setting": { "mode": "thinking", "effort": "medium", "prompt_style": "not disclosed", "temperature": "not disclosed", "context": "not disclosed", "output_limit": "not disclosed", "trials": "not disclosed", "tools": "Stirrup shell and document workflow tools", "sampling": "not disclosed", "aggregation": "rubric-criterion pass rate", "judge": "rubric criteria over extracted deliverables", "harness": "Artificial Analysis Harvey LAB-AA", "dataset_version": "Harvey LAB-AA", "dataset_split": "120 tasks" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Complex legal workflows. Physical rendering retained even when semantically identical to another official rendering." }, { "model_id": "gemini-3.7-flash", "benchmark_id": "hle_verified", "score": 53.6, "reference_url": "https://storage.googleapis.com/deepmind-media/gemini/gemini_3-7_flash_model_evaluation.pdf", "reported_setting": { "mode": "thinking", "effort": "medium", "prompt_style": "not disclosed", "temperature": "not disclosed", "context": "not disclosed", "output_limit": "not disclosed", "trials": "not disclosed", "tools": "none", "sampling": "pass@1", "judge": "HLE-Verified answer evaluator", "harness": "Google self-computed HLE-Verified evaluation", "multimodal_input": true, "dataset_version": "HLE-Verified full verified set", "dataset_split": "Gold 668 plus Revision 1,143" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "1,811 items; Gold + Revision; Uncertain excluded. Physical rendering retained even when semantically identical to another official rendering." }, { "model_id": "gemini-3.7-flash", "benchmark_id": "labbench2", "score": 82.1, "reference_url": "https://storage.googleapis.com/deepmind-media/gemini/gemini_3-7_flash_model_evaluation.pdf", "reported_setting": { "mode": "thinking", "effort": "medium", "prompt_style": "not disclosed", "temperature": "not disclosed", "context": "not disclosed", "output_limit": "not disclosed", "trials": "not disclosed", "tools": "Linux terminal, bioinformatics tools, Python, R, and internet", "sampling": "pass@1", "judge": "task-specific exact/programmatic evaluator", "harness": "Google self-computed LABBench2 evaluation", "multimodal_input": true, "dataset_version": "LABBench2", "dataset_split": "full 1,912-task suite" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Biology real-world research tasks. Physical rendering retained even when semantically identical to another official rendering." }, { "model_id": "gemini-3.7-flash", "benchmark_id": "lvbench", "score": 85.4, "reference_url": "https://storage.googleapis.com/deepmind-media/gemini/gemini_3-7_flash_model_evaluation.pdf", "reported_setting": { "mode": "thinking", "effort": "medium", "prompt_style": "not disclosed", "temperature": "not disclosed", "context": "not disclosed", "output_limit": "not disclosed", "trials": "not disclosed", "tools": "none", "sampling": "pass@1", "judge": "exact multiple-choice answer", "harness": "Google self-computed LVBench evaluation", "multimodal_input": true, "dataset_version": "LVBench", "dataset_split": "103 videos; 1,549 questions", "video_frames": 1024 }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Long video understanding. Physical rendering retained even when semantically identical to another official rendering." }, { "model_id": "gemini-3.7-flash", "benchmark_id": "osworld_2_0_2026_06_24_partial", "score": 47.9, "reference_url": "https://storage.googleapis.com/deepmind-media/gemini/gemini_3-7_flash_model_evaluation.pdf", "reported_setting": { "mode": "thinking", "effort": "medium", "prompt_style": "not disclosed", "temperature": "not disclosed", "context": "not disclosed", "output_limit": "not disclosed", "trials": 3, "dataset_version": "osworld-v2-2026.06.24", "dataset_split": "108 tasks; before 2026-08-08 patch", "multimodal_input": true, "tools": "pyautogui actuation plus UI-specific function declarations", "sampling": "pass@1", "aggregation": "maximum partial score over three single-attempt runs", "judge": "official weighted checkpoint partial scorer", "harness": "OSWorld 2.0 Docker with official Gemini/Anthropic CUA harness", "screen": "1080p screenshot-only observation", "max_steps": 500 }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Weighted checkpoint partial score. Physical rendering retained even when semantically identical to another official rendering." }, { "model_id": "gemini-3.7-flash", "benchmark_id": "terminal_bench_2_1", "score": 85.8, "reference_url": "https://storage.googleapis.com/deepmind-media/gemini/gemini_3-7_flash_model_evaluation.pdf", "reported_setting": { "mode": "thinking", "effort": "medium", "prompt_style": "not disclosed", "temperature": "not disclosed", "context": "not disclosed", "output_limit": "not disclosed", "trials": "not disclosed", "dataset_version": "Terminal-Bench 2.1", "dataset_split": "89 tasks", "tools": "Terminus 2 terminal toolset", "sampling": "pass@1", "judge": "Terminal-Bench 2.1 executable verifier", "harness": "Terminus 2; Google self-computed" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Agentic terminal coding. Physical rendering retained even when semantically identical to another official rendering." }, { "model_id": "gemini-3.7-flash", "benchmark_id": "terminal_bench_3_0", "score": 14.9, "reference_url": "https://storage.googleapis.com/deepmind-media/gemini/gemini_3-7_flash_model_evaluation.pdf", "reported_setting": { "mode": "thinking", "effort": "medium", "prompt_style": "not disclosed", "temperature": "not disclosed", "context": "not disclosed", "output_limit": "not disclosed", "trials": "not disclosed", "tools": "container terminal", "sampling": "pass@1", "judge": "Terminal-Bench 3 task-success evaluator", "harness": "Google self-computed mini-swe-agent with LiteLLM 1.96; two documented task compatibility modifications", "dataset_version": "Terminal-Bench 3.0 / FrontierBench", "dataset_split": "74 tasks" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "General agent capabilities. Physical rendering retained even when semantically identical to another official rendering." }, { "model_id": "gpt-5.6-terra", "benchmark_id": "charxiv_reasoning", "score": 85.9, "reference_url": "https://storage.googleapis.com/deepmind-media/gemini/gemini_3-7_flash_model_evaluation.pdf", "reported_setting": { "mode": "thinking", "effort": "not disclosed", "prompt_style": "not disclosed", "temperature": "not disclosed", "context": "not disclosed", "output_limit": "not disclosed", "trials": "not disclosed", "tools": "none", "sampling": "not disclosed", "judge": "not disclosed", "harness": "Google self-computed CharXiv evaluation", "multimodal_input": true, "dataset_version": "CharXiv Reasoning", "dataset_split": "1,000 validation questions" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Information synthesis from complex charts. Physical rendering retained even when semantically identical to another official rendering." }, { "model_id": "gpt-5.6-terra", "benchmark_id": "code_arena_frontend_elo", "score": 1523, "reference_url": "https://storage.googleapis.com/deepmind-media/gemini/gemini_3-7_flash_model_evaluation.pdf", "reported_setting": { "mode": "thinking", "effort": "not disclosed", "prompt_style": "not disclosed", "temperature": "not disclosed", "context": "not disclosed", "output_limit": "not disclosed", "trials": "not disclosed", "tools": "web application generation", "sampling": "not disclosed", "aggregation": "Arena Elo from pairwise human votes", "judge": "human pairwise preference votes", "harness": "Code Arena WebDev", "dataset_version": "Code Arena WebDev live leaderboard", "dataset_split": "live WebDev Arena" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Elo. Physical rendering retained even when semantically identical to another official rendering.", "candidates": [ { "score": 1521, "reference_url": "https://arena.ai/leaderboard/code", "source_type": "leaderboard", "reported_setting": { "mode": "thinking", "effort": "xhigh", "prompt_style": "not disclosed", "temperature": "not disclosed", "context": "not disclosed", "output_limit": "not disclosed", "trials": "not disclosed", "tools": "web application generation", "sampling": "not disclosed", "aggregation": "Arena Elo from pairwise human votes", "judge": "human pairwise preference votes", "harness": "Code Arena WebDev", "dataset_version": "Code Arena WebDev live leaderboard", "dataset_split": "live WebDev Arena" }, "notes": "Current official leaderboard update; the Google launch table remains preserved as the August release snapshot." } ] }, { "model_id": "gpt-5.6-terra", "benchmark_id": "frontiercode_main_v1_1", "score": 41.3, "reference_url": "https://storage.googleapis.com/deepmind-media/gemini/gemini_3-7_flash_model_evaluation.pdf", "reported_setting": { "mode": "thinking", "effort": "not disclosed", "prompt_style": "not disclosed", "temperature": "not disclosed", "context": "not disclosed", "output_limit": "not disclosed", "trials": "not disclosed", "tools": "coding agent with guarded internet access", "sampling": "not disclosed", "judge": "blocking held-out functional criteria plus weighted maintainer-authored rubric criteria", "harness": "FrontierCode 1.1 official leaderboard", "dataset_version": "FrontierCode 1.1", "dataset_split": "100-task Main split" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Score; production code quality. Physical rendering retained even when semantically identical to another official rendering." }, { "model_id": "gpt-5.6-terra", "benchmark_id": "gdm_mrcr_v2_8needle_upto_128k", "score": 93.5, "reference_url": "https://storage.googleapis.com/deepmind-media/gemini/gemini_3-7_flash_model_evaluation.pdf", "reported_setting": { "mode": "thinking", "effort": "not disclosed", "prompt_style": "not disclosed", "temperature": "not disclosed", "context": "not disclosed", "output_limit": "not disclosed", "trials": "not disclosed", "tools": "none", "sampling": "not disclosed", "judge": "strict MRCR scorer", "harness": "Google self-computed GDM eval_hub", "dataset_version": "GDM MRCR v2.1, 8 needles", "dataset_split": "cumulative through 128K; 484 examples", "sequence_length": "cumulative through 128K" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Cumulative through 128K. Physical rendering retained even when semantically identical to another official rendering." }, { "model_id": "gpt-5.6-terra", "benchmark_id": "harvey_lab_aa", "score": 85.2, "reference_url": "https://storage.googleapis.com/deepmind-media/gemini/gemini_3-7_flash_model_evaluation.pdf", "reported_setting": { "mode": "thinking", "effort": "not disclosed", "prompt_style": "not disclosed", "temperature": "not disclosed", "context": "not disclosed", "output_limit": "not disclosed", "trials": "not disclosed", "tools": "Stirrup shell and document workflow tools", "sampling": "not disclosed", "aggregation": "rubric-criterion pass rate", "judge": "rubric criteria over extracted deliverables", "harness": "Artificial Analysis Harvey LAB-AA", "dataset_version": "Harvey LAB-AA", "dataset_split": "120 tasks" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Complex legal workflows. Physical rendering retained even when semantically identical to another official rendering." }, { "model_id": "gpt-5.6-terra", "benchmark_id": "hle_verified", "score": 51.1, "reference_url": "https://storage.googleapis.com/deepmind-media/gemini/gemini_3-7_flash_model_evaluation.pdf", "reported_setting": { "mode": "thinking", "effort": "not disclosed", "prompt_style": "not disclosed", "temperature": "not disclosed", "context": "not disclosed", "output_limit": "not disclosed", "trials": "not disclosed", "tools": "none", "sampling": "not disclosed", "judge": "HLE-Verified answer evaluator", "harness": "Google self-computed HLE-Verified evaluation", "multimodal_input": true, "dataset_version": "HLE-Verified full verified set", "dataset_split": "Gold 668 plus Revision 1,143" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "1,811 items; Gold + Revision; Uncertain excluded. Physical rendering retained even when semantically identical to another official rendering." }, { "model_id": "gpt-5.6-terra", "benchmark_id": "labbench2", "score": 81.2, "reference_url": "https://storage.googleapis.com/deepmind-media/gemini/gemini_3-7_flash_model_evaluation.pdf", "reported_setting": { "mode": "thinking", "effort": "not disclosed", "prompt_style": "not disclosed", "temperature": "not disclosed", "context": "not disclosed", "output_limit": "not disclosed", "trials": "not disclosed", "tools": "Linux terminal, bioinformatics tools, Python, R, and internet", "sampling": "not disclosed", "judge": "task-specific exact/programmatic evaluator", "harness": "Google self-computed LABBench2 evaluation", "multimodal_input": true, "dataset_version": "LABBench2", "dataset_split": "full 1,912-task suite" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Biology real-world research tasks. Physical rendering retained even when semantically identical to another official rendering." }, { "model_id": "gpt-5.6-terra", "benchmark_id": "lvbench", "score": 78.9, "reference_url": "https://storage.googleapis.com/deepmind-media/gemini/gemini_3-7_flash_model_evaluation.pdf", "reported_setting": { "mode": "thinking", "effort": "not disclosed", "prompt_style": "not disclosed", "temperature": "not disclosed", "context": "not disclosed", "output_limit": "not disclosed", "trials": "not disclosed", "tools": "none", "sampling": "not disclosed", "judge": "exact multiple-choice answer", "harness": "Google self-computed LVBench evaluation", "multimodal_input": true, "dataset_version": "LVBench", "dataset_split": "103 videos; 1,549 questions", "video_frames": 1024 }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "Long video understanding. Physical rendering retained even when semantically identical to another official rendering." }, { "model_id": "gpt-5.6-terra", "benchmark_id": "osworld_2_0_2026_06_24_partial", "score": 50.2, "reference_url": "https://openai.com/index/gpt-5-6/", "reported_setting": { "mode": "not disclosed", "effort": "not disclosed", "prompt_style": "not disclosed", "temperature": "not disclosed", "context": "not disclosed", "output_limit": "not disclosed", "trials": "not disclosed", "dataset_version": "osworld-v2-2026.06.24", "dataset_split": "108 tasks; before 2026-08-08 patch", "multimodal_input": true, "tools": "not disclosed", "sampling": "not disclosed", "judge": "not disclosed", "harness": "not disclosed" }, "matches_canonical": true, "source_type": "official_blog", "audit_status": "verified", "notes": "Exact provider table captured from the rendered official page." }, { "model_id": "gpt-5.6-terra", "benchmark_id": "terminal_bench_3_0", "score": 20.8, "reference_url": "https://storage.googleapis.com/deepmind-media/gemini/gemini_3-7_flash_model_evaluation.pdf", "reported_setting": { "mode": "thinking", "effort": "not disclosed", "prompt_style": "not disclosed", "temperature": "not disclosed", "context": "not disclosed", "output_limit": "not disclosed", "trials": "not disclosed", "tools": "container terminal", "sampling": "not disclosed", "judge": "Terminal-Bench 3 task-success evaluator", "harness": "historic official Terminal-Bench 3 leaderboard", "dataset_version": "Terminal-Bench 3.0 / FrontierBench", "dataset_split": "74 tasks" }, "matches_canonical": true, "source_type": "tech_report", "audit_status": "verified", "notes": "General agent capabilities. Physical rendering retained even when semantically identical to another official rendering." }, { "model_id": "claude-opus-5", "benchmark_id": "mcpatlas_full_1000", "score": 85.8, "reference_url": "https://research.meta.ai/articles/introducing-muse-code-and-muse-spark-1-2/evaluations/mcp-atlas-v1.png", "reported_setting": { "mode": "thinking", "effort": "xhigh", "prompt_style": "not disclosed", "temperature": "not disclosed", "context": "not disclosed", "agent": "Scale AI MCP-Atlas agent harness", "attempts": 1, "tools": "controlled target and distractor tools across 36 MCP servers and 220 tools", "sampling": "pass@1 over all 1,000 tasks", "judge": "claim-level 1/0.5/0; pass at coverage >=0.75; Gemini 3.1 Pro Preview primary", "harness": "Scale AI MCP-Atlas agent harness and scoring pipeline", "internet": "containerized real MCP servers under benchmark controls", "configuration": "500 public plus 500 held-out private tasks" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Exact printed value from the locked Meta release chart; evaluation settings come from the locked official methodology." }, { "model_id": "claude-opus-5", "benchmark_id": "meta_internal_coding_bench", "score": 79.4, "reference_url": "https://research.meta.ai/articles/introducing-muse-code-and-muse-spark-1-2/evaluations/meta-internal-coding-bench-v1.png", "reported_setting": { "mode": "thinking", "effort": "max", "prompt_style": "not disclosed", "temperature": "not disclosed", "context": "not disclosed", "agent": "not disclosed", "attempts": 2, "tools": "Meta internal agentic coding environment", "sampling": "two attempts per task; average task-level success rate", "judge": "compile and unit tests in dedicated grading containers", "harness": "Meta internal agentic coding harness and isolated task sandbox", "internet": "disabled", "configuration": "440 internal pull-request tasks" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Exact printed value from the locked Meta release chart; evaluation settings come from the locked official methodology." }, { "model_id": "claude-opus-5", "benchmark_id": "terminal_bench_2_1", "score": 86.7, "reference_url": "https://research.meta.ai/articles/introducing-muse-code-and-muse-spark-1-2/evaluations/terminal-bench-2-1-v1.png", "reported_setting": { "mode": "thinking", "effort": "max", "prompt_style": "not disclosed", "temperature": "not disclosed", "context": "not disclosed", "agent": "Claude Code", "attempts": 5, "tools": "Claude Code terminal coding toolset", "sampling": "pass@1 averaged across five attempts per task", "judge": "Terminal-Bench 2.1 executable task verifier", "harness": "Meta internal agent evaluation framework; Claude Code; isolated Daytona sandbox", "internet": "not disclosed", "configuration": "selected official agent product in isolated Daytona" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Exact printed value from the locked Meta release chart; evaluation settings come from the locked official methodology." }, { "model_id": "gemini-3.5-flash", "benchmark_id": "mcpatlas_full_1000", "score": 83.6, "reference_url": "https://research.meta.ai/articles/introducing-muse-code-and-muse-spark-1-2/evaluations/mcp-atlas-v1.png", "reported_setting": { "mode": "thinking", "effort": "high", "prompt_style": "not disclosed", "temperature": "not disclosed", "context": "not disclosed", "agent": "Scale AI MCP-Atlas agent harness", "attempts": 1, "tools": "controlled target and distractor tools across 36 MCP servers and 220 tools", "sampling": "pass@1 over all 1,000 tasks", "judge": "claim-level 1/0.5/0; pass at coverage >=0.75; Gemini 3.1 Pro Preview primary", "harness": "Scale AI MCP-Atlas agent harness and scoring pipeline", "internet": "containerized real MCP servers under benchmark controls", "configuration": "500 public plus 500 held-out private tasks" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Exact printed value from the locked Meta release chart; evaluation settings come from the locked official methodology." }, { "model_id": "gemini-3.6-flash", "benchmark_id": "meta_internal_coding_bench", "score": 63.9, "reference_url": "https://research.meta.ai/articles/introducing-muse-code-and-muse-spark-1-2/evaluations/meta-internal-coding-bench-v1.png", "reported_setting": { "mode": "thinking", "effort": "high", "prompt_style": "not disclosed", "temperature": "not disclosed", "context": "not disclosed", "agent": "not disclosed", "attempts": 2, "tools": "Meta internal agentic coding environment", "sampling": "two attempts per task; average task-level success rate", "judge": "compile and unit tests in dedicated grading containers", "harness": "Meta internal agentic coding harness and isolated task sandbox", "internet": "disabled", "configuration": "440 internal pull-request tasks" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Exact printed value from the locked Meta release chart; evaluation settings come from the locked official methodology." }, { "model_id": "gpt-5.6-terra", "benchmark_id": "meta_internal_coding_bench", "score": 65.4, "reference_url": "https://research.meta.ai/articles/introducing-muse-code-and-muse-spark-1-2/evaluations/meta-internal-coding-bench-v1.png", "reported_setting": { "mode": "thinking", "effort": "max", "prompt_style": "not disclosed", "temperature": "not disclosed", "context": "not disclosed", "agent": "not disclosed", "attempts": 2, "tools": "Meta internal agentic coding environment", "sampling": "two attempts per task; average task-level success rate", "judge": "compile and unit tests in dedicated grading containers", "harness": "Meta internal agentic coding harness and isolated task sandbox", "internet": "disabled", "configuration": "440 internal pull-request tasks" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Exact printed value from the locked Meta release chart; evaluation settings come from the locked official methodology." }, { "model_id": "kimi-k3", "benchmark_id": "mcpatlas_full_1000", "score": 82.3, "reference_url": "https://research.meta.ai/articles/introducing-muse-code-and-muse-spark-1-2/evaluations/mcp-atlas-v1.png", "reported_setting": { "mode": "thinking", "effort": "max", "prompt_style": "not disclosed", "temperature": "not disclosed", "context": "not disclosed", "agent": "Scale AI MCP-Atlas agent harness", "attempts": 1, "tools": "controlled target and distractor tools across 36 MCP servers and 220 tools", "sampling": "pass@1 over all 1,000 tasks", "judge": "claim-level 1/0.5/0; pass at coverage >=0.75; Gemini 3.1 Pro Preview primary", "harness": "Scale AI MCP-Atlas agent harness and scoring pipeline", "internet": "containerized real MCP servers under benchmark controls", "configuration": "500 public plus 500 held-out private tasks" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Exact printed value from the locked Meta release chart; evaluation settings come from the locked official methodology." }, { "model_id": "muse-spark-1.2", "benchmark_id": "deep_swe_v1_1", "score": 59.3, "reference_url": "https://research.meta.ai/articles/introducing-muse-code-and-muse-spark-1-2/evaluations/deepswe-1-1-v1.png", "reported_setting": { "mode": "thinking", "effort": "xhigh", "prompt_style": "not disclosed", "temperature": "not disclosed", "context": "not disclosed", "agent": "Muse Code", "attempts": 5, "tools": "Muse Code repository shell/editor toolset", "sampling": "pass@1 averaged across five attempts per task", "judge": "functional verifier plus regression checks in a pristine verifier container", "harness": "Meta internal agent evaluation framework; Muse Code; isolated Daytona sandbox; Harbor conversion", "internet": "disabled during rollout and grading; model endpoint only", "configuration": "five-language Harbor conversion with pristine verifier" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Exact printed value from the locked Meta release chart; evaluation settings come from the locked official methodology." }, { "model_id": "muse-spark-1.2", "benchmark_id": "gdpval_aa_v2_elo", "score": 1631, "reference_url": "https://research.meta.ai/articles/introducing-muse-code-and-muse-spark-1-2/evaluations/gdpval-aa-v2-v1.png", "reported_setting": { "mode": "thinking", "effort": "xhigh", "prompt_style": "not disclosed", "temperature": "not disclosed", "context": "not disclosed", "agent": "Artificial Analysis Stirrup", "attempts": 1, "tools": "Web Fetch, Web Search, View Image, Code Exec, Finish, Abandon Task", "sampling": "one agentic submission per each of 220 tasks", "judge": "blind pairwise panel of three frontier LLM judges; Bradley-Terry Elo", "harness": "Artificial Analysis Stirrup; fresh E2B sandbox; 250-turn limit", "internet": "web fetch and Brave-backed web search available", "configuration": "fresh E2B sandbox per task; 250-turn limit" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Exact printed value from the locked Meta release chart; evaluation settings come from the locked official methodology." }, { "model_id": "muse-spark-1.2", "benchmark_id": "mcpatlas_full_1000", "score": 90.3, "reference_url": "https://research.meta.ai/articles/introducing-muse-code-and-muse-spark-1-2/evaluations/mcp-atlas-v1.png", "reported_setting": { "mode": "thinking", "effort": "xhigh", "prompt_style": "not disclosed", "temperature": "not disclosed", "context": "not disclosed", "agent": "Scale AI MCP-Atlas agent harness", "attempts": 1, "tools": "controlled target and distractor tools across 36 MCP servers and 220 tools", "sampling": "pass@1 over all 1,000 tasks", "judge": "claim-level 1/0.5/0; pass at coverage >=0.75; Gemini 3.1 Pro Preview primary", "harness": "Scale AI MCP-Atlas agent harness and scoring pipeline", "internet": "containerized real MCP servers under benchmark controls", "configuration": "500 public plus 500 held-out private tasks" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Exact printed value from the locked Meta release chart; evaluation settings come from the locked official methodology." }, { "model_id": "muse-spark-1.2", "benchmark_id": "meta_internal_coding_bench", "score": 70.6, "reference_url": "https://research.meta.ai/articles/introducing-muse-code-and-muse-spark-1-2/evaluations/meta-internal-coding-bench-v1.png", "reported_setting": { "mode": "thinking", "effort": "xhigh", "prompt_style": "not disclosed", "temperature": "not disclosed", "context": "not disclosed", "agent": "not disclosed", "attempts": 2, "tools": "Meta internal agentic coding environment", "sampling": "two attempts per task; average task-level success rate", "judge": "compile and unit tests in dedicated grading containers", "harness": "Meta internal agentic coding harness and isolated task sandbox", "internet": "disabled", "configuration": "440 internal pull-request tasks" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Exact printed value from the locked Meta release chart; evaluation settings come from the locked official methodology." }, { "model_id": "muse-spark-1.2", "benchmark_id": "terminal_bench_2_1", "score": 82.9, "reference_url": "https://research.meta.ai/articles/introducing-muse-code-and-muse-spark-1-2/evaluations/terminal-bench-2-1-v1.png", "reported_setting": { "mode": "thinking", "effort": "xhigh", "prompt_style": "not disclosed", "temperature": "not disclosed", "context": "not disclosed", "agent": "Muse Code", "attempts": 5, "tools": "Muse Code terminal coding toolset", "sampling": "pass@1 averaged across five attempts per task", "judge": "Terminal-Bench 2.1 executable task verifier", "harness": "Meta internal agent evaluation framework; Muse Code; isolated Daytona sandbox", "internet": "not disclosed", "configuration": "selected official agent product in isolated Daytona" }, "matches_canonical": false, "source_type": "official_blog", "audit_status": "verified", "notes": "Exact printed value from the locked Meta release chart; evaluation settings come from the locked official methodology." } ], "generated": "2026-08-18T17:22:24Z" }